Search results for: population-based algorithms

Authors: Lakshmi N. Reddi, Akanksha Varma Sagi

Abstract:

Studies on biomimetics are largely developed, deriving inspiration from natural processes in our objective world to develop novel technologies. Recent studies are diverse in nature, making their categorization quite challenging. Based on an exhaustive survey, we developed categorizations based on either the essential elements of nature - air, water, land, fire, and space, or on form/shape, functionality, and process. Such diverse studies as aircraft wings inspired by bird wings, a self-cleaning coating inspired by a lotus petal, wetsuits inspired by beaver fur, and search algorithms inspired by arboreal ant path networks lend themselves to these categorizations. Our categorizations of biomimetic studies allowed us to define a different dimension of biomimetics. This new dimension is not restricted to inspiration from the objective world. It is based on the premise that the biological processes observed in the objective world find their reflections in our human bodies in a variety of ways. For example, the lungs provide the most efficient example for liquid-gas phase exchange, the heart exemplifies a very efficient pumping and circulatory system, and the kidneys epitomize the most effective cleaning system. The main focus of this paper is to bring out the magnificence of the cerebro-spinal system (CSS) insofar as it relates to our current computer architecture. In particular, the paper uses four key measures to analyze the differences between CSS and human- engineered computational systems. These are adaptability, sustainability, energy efficiency, and resilience. We found that the cerebrospinal system reveals some important challenges in the development and evolution of our current computer architectures. In particular, the myriad ways in which the CSS is integrated with other systems/processes (circulatory, respiration, etc) offer useful insights on how the human-engineered computational systems could be made more sustainable, energy-efficient, resilient, and adaptable. In our paper, we highlight the energy consumption differences between CSS and our current computational designs. Apart from the obvious differences in materials used between the two, the systemic nature of how CSS functions provides clues to enhance life-cycles of our current computational systems. The rapid formation and changes in the physiology of dendritic spines and their synaptic plasticity causing memory changes (ex., long-term potentiation and long-term depression) allowed us to formulate differences in the adaptability and resilience of CSS. In addition, the CSS is sustained by integrative functions of various organs, and its robustness comes from its interdependence with the circulatory system. The paper documents and analyzes quantifiable differences between the two in terms of the four measures. Our analyses point out the possibilities in the development of computational systems that are more adaptable, sustainable, energy efficient, and resilient. It concludes with the potential approaches for technological advancement through creation of more interconnected and interdependent systems to replicate the effective operation of cerebro-spinal system.

Keywords: cerebrospinal system, computer architecture, adaptability, sustainability, resilience, energy efficiency

Procedia PDF Downloads 97

Authors: Ali Kazemi

Abstract:

The digital evolution of financial services, while offering unprecedented convenience and accessibility, has also escalated the vulnerabilities to fraudulent activities. In this study, we introduce a distinct approach to real-time anomaly detection in financial transactions, aiming to fortify the defenses of banking and financial institutions against such threats. Utilizing unsupervised machine learning algorithms, specifically autoencoders and isolation forests, our research focuses on identifying irregular patterns indicative of fraud within transactional data, thus enabling immediate action to prevent financial loss. The data we used in this study included the monetary value of each transaction. This is a crucial feature as fraudulent transactions may have distributions of different amounts than legitimate ones, such as timestamps indicating when transactions occurred. Analyzing transactions' temporal patterns can reveal anomalies (e.g., unusual activity in the middle of the night). Also, the sector or category of the merchant where the transaction occurred, such as retail, groceries, online services, etc. Specific categories may be more prone to fraud. Moreover, the type of payment used (e.g., credit, debit, online payment systems). Different payment methods have varying risk levels associated with fraud. This dataset, anonymized to ensure privacy, reflects a wide array of transactions typical of a global banking institution, ranging from small-scale retail purchases to large wire transfers, embodying the diverse nature of potentially fraudulent activities. By engineering features that capture the essence of transactions, including normalized amounts and encoded categorical variables, we tailor our data to enhance model sensitivity to anomalies. The autoencoder model leverages its reconstruction error mechanism to flag transactions that deviate significantly from the learned normal pattern, while the isolation forest identifies anomalies based on their susceptibility to isolation from the dataset's majority. Our experimental results, validated through techniques such as k-fold cross-validation, are evaluated using precision, recall, and the F1 score alongside the area under the receiver operating characteristic (ROC) curve. Our models achieved an F1 score of 0.85 and a ROC AUC of 0.93, indicating high accuracy in detecting fraudulent transactions without excessive false positives. This study contributes to the academic discourse on financial fraud detection and provides a practical framework for banking institutions seeking to implement real-time anomaly detection systems. By demonstrating the effectiveness of unsupervised learning techniques in a real-world context, our research offers a pathway to significantly reduce the incidence of financial fraud, thereby enhancing the security and trustworthiness of digital financial services.

Keywords: anomaly detection, financial fraud, machine learning, autoencoders, isolation forest, transactional data analysis

Procedia PDF Downloads 57

Authors: Binder Hans

Abstract:

Cancer is no longer seen as solely a genetic disease where genetic defects such as mutations and copy number variations affect gene regulation and eventually lead to aberrant cell functioning which can be monitored by transcriptome analysis. It has become obvious that epigenetic alterations represent a further important layer of (de-)regulation of gene activity. For example, aberrant DNA methylation is a hallmark of many cancer types, and methylation patterns were successfully used to subtype cancer heterogeneity. Hence, unraveling the interplay between different omics levels such as genome, transcriptome and epigenome is inevitable for a mechanistic understanding of molecular deregulation causing complex diseases such as cancer. This objective requires powerful downstream integrative bioinformatics methods as an essential prerequisite to discover the whole genome mutational, transcriptome and epigenome landscapes of cancer specimen and to discover cancer genesis, progression and heterogeneity. Basic challenges and tasks arise ‘beyond sequencing’ because of the big size of the data, their complexity, the need to search for hidden structures in the data, for knowledge mining to discover biological function and also systems biology conceptual models to deduce developmental interrelations between different cancer states. These tasks are tightly related to cancer biology as an (epi-)genetic disease giving rise to aberrant genomic regulation under micro-environmental control and clonal evolution which leads to heterogeneous cellular states. Machine learning algorithms such as self organizing maps (SOM) represent one interesting option to tackle these bioinformatics tasks. The SOMmethod enables recognizing complex patterns in large-scale data generated by highthroughput omics technologies. It portrays molecular phenotypes by generating individualized, easy to interpret images of the data landscape in combination with comprehensive analysis options. Our image-based, reductionist machine learning methods provide one interesting perspective how to deal with massive data in the discovery of complex diseases, gliomas, melanomas and colon cancer on molecular level. As an important new challenge, we address the combined portrayal of different omics data such as genome-wide genomic, transcriptomic and methylomic ones. The integrative-omics portrayal approach is based on the joint training of the data and it provides separate personalized data portraits for each patient and data type which can be analyzed by visual inspection as one option. The new method enables an integrative genome-wide view on the omics data types and the underlying regulatory modes. It is applied to high and low-grade gliomas and to melanomas where it disentangles transversal and longitudinal molecular heterogeneity in terms of distinct molecular subtypes and progression paths with prognostic impact.

Keywords: integrative bioinformatics, machine learning, molecular mechanisms of cancer, gliomas and melanomas

Procedia PDF Downloads 148

Authors: Ehsan Javanmardemamgheisi, Javad Mehr, Jason Ali-Lavroff, Damien Holloway, Michael Davis

Abstract:

During the last decades, a growing interest in faster and more efficient waterborne transportation has led to the development of high-speed vessels for both commercial and military applications. To satisfy this global demand, a wide variety of arrangements of high-speed crafts have been proposed by designers. Among them, high-speed catamarans have proven themselves to be a suitable Roll-on/Roll-off configuration for carrying passengers and cargo due to widely spaced demi hulls, a wide deck zone, and a high ratio of deadweight to displacement. To improve passenger comfort and crew workability and enhance the operability and performance of high-speed catamarans, mitigating the severity of motions and structural loads using Ride Control Systems (RCS) is essential.In this paper, a set of towing tank tests was conducted on a 2.5 m scaled model of a 112 m Incat Tasmania high-speed catamaran in irregular head seas to investigate the effect of different ride control algorithms including linear and nonlinear versions of the heave control, pitch control, and local control on motion responses and passenger comfort of the full-scale ship. The RCS included a centre bow-fitted T-Foil and two transom-mounted stern tabs. All the experiments were conducted at the Australian Maritime College (AMC) towing tank at a model speed of 2.89 m/s (37 knots full scale), a modal period of 1.5 sec (10 sec full scale) and two significant wave heights of 60 mm and 90 mm, representing full-scale wave heights of 2.7 m and 4 m, respectively. Spectral analyses were performed using Welch’s power spectral density method on the vertical motion time records of the catamaran model to calculate heave and pitch Response Amplitude Operators (RAOs). Then, noting that passenger discomfort arises from vertical accelerations and that the vertical accelerations vary at different longitudinal locations within the passenger cabin due to the variations in amplitude and relative phase of the pitch and heave motions, the vertical accelerations were calculated at three longitudinal locations (LCG, T-Foil, and stern tabs). Finally, frequency-weighted Root Mean Square (RMS) vertical accelerations were calculated to estimate Motion Sickness Dose Value (MSDV) of the ship based on ISO 2631-recommendations. It was demonstrated that in small seas, implementing a nonlinear pitch control algorithm reduces the peak pitch motions by 41%, the vertical accelerations at the forward location by 46%, and motion sickness at the forward position by around 20% which provides great potential for further improvement in passenger comfort, crew workability, and operability of high-speed catamarans.

Keywords: high-speed catamarans, ride control system, response amplitude operators, vertical accelerations, motion sickness, irregular waves, towing tank tests.

Procedia PDF Downloads 82

Authors: Ruth Hegarty, Noel Connaughton

Abstract:

Sustainability reporting has become a priority for many global multinational companies. This is associated with ever-increasing expectations from key stakeholders for companies to be transparent about their strategies, activities and management with regard to sustainability issues. The Global Reporting Initiative (GRI) encourages reporters to only provide information on the issues that are really critical in order to achieve the organisation’s goals for sustainability and manage its impact on environment and society. A key challenge for most reporting organisations is how to identify relevant issues for sustainability reporting and prioritise those material issues in accordance with company and stakeholder needs. A recent study indicates that most of the largest companies listed on the world’s stock exchanges are failing to provide data on key sustainability indicators such as employee turnover, energy, greenhouse gas emissions (GHGs), injury rate, pay equity, waste and water. This paper takes an indepth look at the approaches used by a select number of international sustainability leader corporates to identify key sustainability issues. The research methodology involves performing a detailed analysis of the sustainability report content of up to 50 companies listed on the 2014 Dow Jones Sustainability Indices (DJSI). The most recent sustainability report content found on the GRI Sustainability Disclosure Database is then compared with 91 GRI Specific Standard Disclosures and a small number of GRI Standard Disclosures. Preliminary research indicates significant gaps in the information disclosed in corporate sustainability reports versus the indicator content specified in the GRI Content Index. The following outlines some of the key findings to date: Most companies made a partial disclosure with regard to the Economic indicators of climate change risks and infrastructure investments, but did not focus on the associated negative impacts. The top Environmental indicators disclosed were energy consumption and reductions, GHG emissions, water withdrawals, waste and compliance. The lowest rates of indicator disclosure included biodiversity, water discharge, mitigation of environmental impacts of products and services, transport, environmental investments, screening of new suppliers and supply chain impacts. The top Social indicators disclosed were new employee hires, rates of injury, freedom of association in operations, child labour and forced labour. Lesser disclosure rates were reported for employee training, composition of governance bodies and employees, political contributions, corruption and fines for non-compliance. The reporting on most other Social indicators was found to be poor. In addition, most companies give only a brief explanation on how material issues are defined, identified and ranked. Data on the identification of key stakeholders and the degree and nature of engagement for determining issues and their weightings is also lacking. Generally, little to no data is provided on the algorithms used to score an issue. Research indicates that most companies lack a rigorous and thorough methodology to systematically determine the material issues of sustainability reporting in accordance with company and stakeholder needs.

Keywords: identification of key stakeholders, material issues, sustainability reporting, transparency

Procedia PDF Downloads 306

Authors: Gaurav Kumar Sinha

Abstract:

In an era defined by the proliferation of data and the utilization of cloud computing environments, the efficient transfer and processing of big data workloads across multi-cloud platforms have emerged as critical challenges. This research paper embarks on a comprehensive exploration of the complexities associated with managing and optimizing big data in a multi-cloud ecosystem.The foundation of this study is rooted in the recognition that modern enterprises increasingly rely on multiple cloud providers to meet diverse business needs, enhance redundancy, and reduce vendor lock-in. As a consequence, managing data across these heterogeneous cloud environments has become intricate, necessitating innovative approaches to ensure data integrity, security, and performance.The primary objective of this research is to investigate strategies and techniques for enhancing the efficiency of data transfer and processing in multi-cloud scenarios. It recognizes that big data workloads are characterized by their sheer volume, variety, velocity, and complexity, making traditional data management solutions insufficient for harnessing the full potential of multi-cloud architectures.The study commences by elucidating the challenges posed by multi-cloud environments in the context of big data. These challenges encompass data fragmentation, latency, security concerns, and cost optimization. To address these challenges, the research explores a range of methodologies and solutions. One of the key areas of focus is data transfer optimization. The paper delves into techniques for minimizing data movement latency, optimizing bandwidth utilization, and ensuring secure data transmission between different cloud providers. It evaluates the applicability of dedicated data transfer protocols, intelligent data routing algorithms, and edge computing approaches in reducing transfer times.Furthermore, the study examines strategies for efficient data processing across multi-cloud environments. It acknowledges that big data processing requires distributed and parallel computing capabilities that span across cloud boundaries. The research investigates containerization and orchestration technologies, serverless computing models, and interoperability standards that facilitate seamless data processing workflows.Security and data governance are paramount concerns in multi-cloud environments. The paper explores methods for ensuring data security, access control, and compliance with regulatory frameworks. It considers encryption techniques, identity and access management, and auditing mechanisms as essential components of a robust multi-cloud data security strategy.The research also evaluates cost optimization strategies, recognizing that the dynamic nature of multi-cloud pricing models can impact the overall cost of data transfer and processing. It examines approaches for workload placement, resource allocation, and predictive cost modeling to minimize operational expenses while maximizing performance.Moreover, this study provides insights into real-world case studies and best practices adopted by organizations that have successfully navigated the challenges of multi-cloud big data management. It presents a comparative analysis of various multi-cloud management platforms and tools available in the market.

Keywords: multi-cloud environments, big data workloads, data transfer optimization, data processing strategies

Procedia PDF Downloads 67

Authors: Zhao Gao, Eran Edirisinghe

Abstract:

The production of an accurate sketch of a suspect based on a verbal description obtained from a witness is an essential task for most criminal investigations. The criminal investigation system employs specifically trained professional artists to manually draw a facial image of the suspect according to the descriptions of an eyewitness for subsequent identification. Within the advancement of Deep Learning, Recurrent Neural Networks (RNN) have shown great promise in Natural Language Processing (NLP) tasks. Additionally, Generative Adversarial Networks (GAN) have also proven to be very effective in image generation. In this study, a trained GAN conditioned on textual features such as keywords automatically encoded from a verbal description of a human face using an RNN is used to generate photo-realistic facial images for criminal investigations. The intention of the proposed system is to map corresponding features into text generated from verbal descriptions. With this, it becomes possible to generate many reasonably accurate alternatives to which the witness can use to hopefully identify a suspect from. This reduces subjectivity in decision making both by the eyewitness and the artist while giving an opportunity for the witness to evaluate and reconsider decisions. Furthermore, the proposed approach benefits law enforcement agencies by reducing the time taken to physically draw each potential sketch, thus increasing response times and mitigating potentially malicious human intervention. With publically available 'CelebFaces Attributes Dataset' (CelebA) and additionally providing verbal description as training data, the proposed architecture is able to effectively produce facial structures from given text. Word Embeddings are learnt by applying the RNN architecture in order to perform semantic parsing, the output of which is fed into the GAN for synthesizing photo-realistic images. Rather than the grid search method, a metaheuristic search based on genetic algorithms is applied to evolve the network with the intent of achieving optimal hyperparameters in a fraction the time of a typical brute force approach. With the exception of the ‘CelebA’ training database, further novel test cases are supplied to the network for evaluation. Witness reports detailing criminals from Interpol or other law enforcement agencies are sampled on the network. Using the descriptions provided, samples are generated and compared with the ground truth images of a criminal in order to calculate the similarities. Two factors are used for performance evaluation: The Structural Similarity Index (SSIM) and the Peak Signal-to-Noise Ratio (PSNR). A high percentile output from this performance matrix should attribute to demonstrating the accuracy, in hope of proving that the proposed approach can be an effective tool for law enforcement agencies. The proposed approach to criminal facial image generation has potential to increase the ratio of criminal cases that can be ultimately resolved using eyewitness information gathering.

Keywords: RNN, GAN, NLP, facial composition, criminal investigation

Procedia PDF Downloads 159

Authors: Aleksander Ejsmont, Aleksandra Galarda, Joanna Goscianska

Abstract:

Smart porous carriers with defined structure and physicochemical properties are required for releasing the therapeutic drug with precise control of delivery time and location in the body. Due to their non-toxicity, ordered structure, chemical, and thermal stability, mesoporous carbons can be considered as modern carriers for active pharmaceutical ingredients (APIs) whose effectiveness needs frequent dosing algorithms. Such an API-carrier system, if programmed precisely, may stabilize the pharmaceutical and increase its dissolution leading to enhanced bioavailability. The substance conjugated with the material, through its prior adsorption, can later be successfully applied internally to the organism, as well as externally if the API release is feasible under these conditions. In the present study, ordered mesoporous carbons of different morphologies and structures, prepared by hard template method, were applied as carriers in the adsorption and controlled release of active pharmaceutical ingredients. In the first stage, the carbon materials were synthesized and functionalized with carboxylic groups by chemical oxidation using ammonium persulfate solution and then with amine groups. Materials obtained were thoroughly characterized with respect to morphology (scanning electron microscopy), structure (X-ray diffraction, transmission electron microscopy), characteristic functional groups (FT-IR spectroscopy), acid-base nature of surface groups (Boehm titration), parameters of the porous structure (low-temperature nitrogen adsorption) and thermal stability (TG analysis). This was followed by a series of tests of adsorption and release of paracetamol, benzocaine, and losartan potassium. Drug release experiments were performed in the simulated gastric fluid of pH 1.2 and phosphate buffer of pH 7.2 or 6.8 at 37.0 °C. The XRD patterns in the small-angle range and TEM images revealed that functionalization of mesoporous carbons with carboxylic or amine groups leads to the decreased ordering of their structure. Moreover, the modification caused a considerable reduction of the carbon-specific surface area and pore volume, but it simultaneously resulted in changing their acid-base properties. Mesoporous carbon materials exhibit different morphologies, which affect the host-guest interactions during the adsorption process of active pharmaceutical ingredients. All mesoporous carbons show high adsorption capacity towards drugs. The sorption capacity of materials is mainly affected by BET surface area and the structure/size matching between adsorbent and adsorbate. Selected APIs are linked to the surface of carbon materials mainly by hydrogen bonds, van der Waals forces, and electrostatic interactions. The release behavior of API is highly dependent on the physicochemical properties of mesoporous carbons. The release rate of APIs could be regulated by the introduction of functional groups and by changing the pH of the receptor medium. Acknowledgments—This research was supported by the National Science Centre, Poland (project SONATA-12 no: 2016/23/D/NZ7/01347).

Keywords: ordered mesoporous carbons, sorption capacity, drug delivery, carbon nanocarriers

Procedia PDF Downloads 176

Authors: Andreas Kerstan, Darren Robey, Wesam Alvan, David Troiani

Abstract:

Over the last 3.5 years, Quantum Cascade Lasers (QCL) technology has become increasingly important in infrared (IR) microscopy. The advantages over fourier transform infrared (FTIR) are that large areas of a few square centimeters can be measured in minutes and that the light intensive QCL makes it possible to obtain spectra with excellent S/N, even with just one scan. A firmly established solution of the laser direct infrared imaging (LDIR) 8700 is the analysis of microplastics. The presence of microplastics in the environment, drinking water, and food chains is gaining significant public interest. To study their presence, rapid and reliable characterization of microplastic particles is essential. Significant technical hurdles in microplastic analysis stem from the sheer number of particles to be analyzed in each sample. Total particle counts of several thousand are common in environmental samples, while well-treated bottled drinking water may contain relatively few. While visual microscopy has been used extensively, it is prone to operator error and bias and is limited to particles larger than 300 µm. As a result, vibrational spectroscopic techniques such as Raman and FTIR microscopy have become more popular, however, they are time-consuming. There is a demand for rapid and highly automated techniques to measure particle count size and provide high-quality polymer identification. Analysis directly on the filter that often forms the last stage in sample preparation is highly desirable as, by removing a sample preparation step it can both improve laboratory efficiency and decrease opportunities for error. Recent advances in infrared micro-spectroscopy combining a QCL with scanning optics have created a new paradigm, LDIR. It offers improved speed of analysis as well as high levels of automation. Its mode of operation, however, requires an IR reflective background, and this has, to date, limited the ability to perform direct “on-filter” analysis. This study explores the potential to combine the filter with an infrared reflective surface filter. By combining an IR reflective material or coating on a filter membrane with advanced image analysis and detection algorithms, it is demonstrated that such filters can indeed be used in this way. Vibrational spectroscopic techniques play a vital role in the investigation and understanding of microplastics in the environment and food chain. While vibrational spectroscopy is widely deployed, improvements and novel innovations in these techniques that can increase the speed of analysis and ease of use can provide pathways to higher testing rates and, hence, improved understanding of the impacts of microplastics in the environment. Due to its capability to measure large areas in minutes, its speed, degree of automation and excellent S/N, the LDIR could also implemented for various other samples like food adulteration, coatings, laminates, fabrics, textiles and tissues. This presentation will highlight a few of them and focus on the benefits of the LDIR vs classical techniques.

Keywords: QCL, automation, microplastics, tissues, infrared, speed

Procedia PDF Downloads 66

Authors: Arup Dutta, Badrul Alam, Sayed M. Wazed, Taslima Newaz, Srobonti Dutta

Abstract:

Background: In today's world, with modern science and contemporary technology, a lot of diseases may be quickly identified and ruled out, but children's fever of unknown origin (FUO) still presents diagnostic difficulties in clinical settings. Any fever that reaches 38 °C and lasts for more than seven days without a known cause is now classified as a fever of unknown origin (FUO). Despite tremendous progress in the medical sector, fever of unknown origin, or FOU, persists as a major health issue and a major contributor to morbidity and mortality, particularly in children, and its spectrum is sometimes unpredictable. The etiology is influenced by geographic location, age, socioeconomic level, frequency of antibiotic resistance, and genetic vulnerability. Since there are currently no known diagnostic algorithms, doctors are forced to evaluate each patient one at a time with extreme caution. A persistent fever poses difficulties for both the patient and the doctor. This prospective observational study was carried out in a Bangladeshi tertiary care hospital from June 2018 to May 2019 with the goal of identifying the epidemiological patterns of fever of unknown origin in pediatric patients. Methods: It was a hospital-based prospective observational study carried out on 106 children (between 2 months and 12 years) with prolonged fever of >38.0 °C lasting for more than 7 days without a clear source. Children with additional chronic diseases or known immunodeficiency problems were not allowed. Clinical practices that helped determine the definitive etiology were assessed. Initial testing included a complete blood count, a routine urine examination, PBF, a chest X-ray, CRP measurement, blood cultures, serology, and additional pertinent investigations. The analysis focused mostly on the etiological results. The standard program SPSS 21 was used to analyze all of the study data. Findings: A total of 106 patients identified as having FUO were assessed, with over half (57.5%) being female and the majority (40.6%) falling within the 1 to 3-year age range. The study categorized the etiological outcomes into five groups: infections, malignancies, connective tissue conditions, miscellaneous, and undiagnosed. In the group that was being studied, infections were found to be the main cause in 44.3% of cases. Undiagnosed cases came in at 31.1%, cancers at 10.4%, other causes at 8.5%, and connective tissue disorders at 4.7%. Hepato-splenomegaly was seen in people with enteric fever, malaria, acute lymphoid leukemia, lymphoma, and hepatic abscesses, either by itself or in combination with other conditions. About 53% of people who were not diagnosed also had hepato-splenomegaly at the same time. Conclusion: Infections are the primary cause of PUO (pyrexia of unknown origin) in children, with undiagnosed cases being the second most common cause. An incremental approach is beneficial in the process of diagnosing a condition. Non-invasive examinations are used to diagnose infections and connective tissue disorders, while invasive investigations are used to diagnose cancer and other ailments. According to this study, the prevalence of undiagnosed diseases is still remarkable, so extensive historical analysis and physical examinations are necessary in order to provide a precise diagnosis.

Keywords: children, diagnostic challenges, fever of unknown origin, pediatric fever, undiagnosed diseases

Procedia PDF Downloads 27

Authors: Dhanuj M. Gandikota

Abstract:

Sensor-derived three-dimensional (3D) point clouds of trees are invaluable in remote sensing analysis for the accurate measurement of key structural metrics, bio-inventory values, spatial planning/visualization, and ecological modeling. Machine learning (ML) holds the potential in addressing the restrictive tradeoffs in cost, spatial coverage, resolution, and information gain that exist in current point cloud sensing methods. Terrestrial laser scanning (TLS) remains the highest fidelity source of both canopy and below-canopy structural features, but usage is limited in both coverage and cost, requiring manual deployment to map out large, forested areas. While aerial laser scanning (ALS) remains a reliable avenue of LIDAR active remote sensing, ALS is also cost-restrictive in deployment methods. Space-borne photogrammetry from high-resolution satellite constellations is an avenue of passive remote sensing with promising viability in research for the accurate construction of vegetation 3-D point clouds. It provides both the lowest comparative cost and the largest spatial coverage across remote sensing methods. However, both space-borne photogrammetry and ALS demonstrate technical limitations in the capture of valuable below-canopy point cloud data. Looking to minimize these tradeoffs, we explored a class of powerful ML algorithms called Deep Learning (DL) that show promise in recent research on 3-D point cloud reconstruction and interpolation. Our research details the efficacy of applying these DL techniques to reconstruct accurate below-canopy point clouds from space-borne and aerial remote sensing through learned patterns of tree species fractal symmetry properties and the supplementation of locally sourced bio-inventory metrics. From our dataset, consisting of tree point clouds obtained from TLS, we deconstructed the point clouds of each tree into those that would be obtained through ALS and satellite photogrammetry of varying resolutions. We fed this ALS/satellite point cloud dataset, along with the simulated local bio-inventory metrics, into the DL point cloud reconstruction architectures to generate the full 3-D tree point clouds (the truth values are denoted by the full TLS tree point clouds containing the below-canopy information). Point cloud reconstruction accuracy was validated both through the measurement of error from the original TLS point clouds as well as the error of extraction of key structural metrics, such as crown base height, diameter above root crown, and leaf/wood volume. The results of this research additionally demonstrate the supplemental performance gain of using minimum locally sourced bio-inventory metric information as an input in ML systems to reach specified accuracy thresholds of tree point cloud reconstruction. This research provides insight into methods for the rapid, cost-effective, and accurate construction of below-canopy tree 3-D point clouds, as well as the supported potential of ML and DL to learn complex, unmodeled patterns of fractal tree growth symmetry.

Keywords: deep learning, machine learning, satellite, photogrammetry, aerial laser scanning, terrestrial laser scanning, point cloud, fractal symmetry

Procedia PDF Downloads 102

Authors: Mayank Gupta, Siba Prasad Samal, Vasu Kakkirala

Abstract:

The internet world and its priorities have changed considerably in the last decade. Browsing on smart phones has increased manifold and is set to explode much more. Users spent considerable time browsing different websites, that gives a great deal of insight into user’s preferences. Instead of plain information classifying different aspects of browsing like Bookmarks, History, and Download Manager into useful categories would improve and enhance the user’s experience. Most of the classification solutions are server side that involves maintaining server and other heavy resources. It has security constraints and maybe misses on contextual data during classification. On device, classification solves many such problems, but the challenge is to achieve accuracy on classification with resource constraints. This on device classification can be much more useful in personalization, reducing dependency on cloud connectivity and better privacy/security. This approach provides more relevant results as compared to current standalone solutions because it uses content rendered by browser which is customized by the content provider based on user’s profile. This paper proposes a Naive Bayes based lightweight classification engine targeted for a resource constraint devices. Our solution integrates with Web Browser that in turn triggers classification algorithm. Whenever a user browses a webpage, this solution extracts DOM Tree data from the browser’s rendering engine. This DOM data is a dynamic, contextual and secure data that can’t be replicated. This proposal extracts different features of the webpage that runs on an algorithm to classify into multiple categories. Naive Bayes based engine is chosen in this solution for its inherent advantages in using limited resources compared to other classification algorithms like Support Vector Machine, Neural Networks, etc. Naive Bayes classification requires small memory footprint and less computation suitable for smartphone environment. This solution has a feature to partition the model into multiple chunks that in turn will facilitate less usage of memory instead of loading a complete model. Classification of the webpages done through integrated engine is faster, more relevant and energy efficient than other standalone on device solution. This classification engine has been tested on Samsung Z3 Tizen hardware. The Engine is integrated into Tizen Browser that uses Chromium Rendering Engine. For this solution, extensive dataset is sourced from dmoztools.net and cleaned. This cleaned dataset has 227.5K webpages which are divided into 8 generic categories ('education', 'games', 'health', 'entertainment', 'news', 'shopping', 'sports', 'travel'). Our browser integrated solution has resulted in 15% less memory usage (due to partition method) and 24% less power consumption in comparison with standalone solution. This solution considered 70% of the dataset for training the data model and the rest 30% dataset for testing. An average accuracy of ~96.3% is achieved across the above mentioned 8 categories. This engine can be further extended for suggesting Dynamic tags and using the classification for differential uses cases to enhance browsing experience.

Keywords: chromium, lightweight engine, mobile computing, Naive Bayes, Tizen, web browser, webpage classification

Procedia PDF Downloads 163

Authors: Michal Geca, Grzegorz Baranski, Ksenia Siadkowska

Abstract:

The work is a part of the project which aims at developing innovative power and control systems for the high power aircraft piston engine ASz62IR. Developed electronically controlled ignition system will reduce emissions of toxic compounds as a result of lowered fuel consumption, optimized combustion and engine capability of efficient combustion of ecological fuels. The tested unit is an air-cooled four-stroke gasoline engine of 9 cylinders in a radial setup, mechanically charged by a radial compressor powered by the engine crankshaft. The total engine cubic capac-ity is 29.87 dm3, and the compression ratio is 6.4:1. The maximum take-off power is 1000 HP at 2200 rpm. The maximum fuel consumption is 280 kg/h. Engine powers aircrafts: An-2, M-18 „Dromader”, DHC-3 „OTTER”, DC-3 „Dakota”, GAF-125 „HAWK” i Y5. The main problems of the engine includes the imbalanced work of cylinders. The non-uniformity value in each cylinder results in non-uniformity of their work. In radial engine cylinders arrangement causes that the mixture movement that takes place in accordance (lower cylinder) or the opposite (upper cylinders) to the direction of gravity. Preliminary tests confirmed the presence of uneven workflow of individual cylinders. The phenomenon is most intense at low speed. The non-uniformity is visible on the waveform of cylinder pressure. Therefore two studies were conducted to determine the impact of this phenomenon on the engine performance: simulation and real tests. Simplified simulation was conducted on the element of the intake system coated with fuel film. The study shows that there is an effect of gravity on the movement of the fuel film inside the radial engine intake channels. Both in the lower and the upper inlet channels the film flows downwards. It follows from the fact that gravity assists the movement of the film in the lower cylinder channels and prevents the movement in the upper cylinder channels. Real tests on aircraft engine ASz62IR was conducted in transients condition (rapid change of the excess air in each cylinder were performed. Calculations were conducted for mass of fuel reaching the cylinders theoretically and really and on this basis, the factors of fuel evaporation “x” were determined. Therefore a simplified model of the fuel supply to cylinder was adopted. Model includes time constant of the fuel film τ, the number of engine transport cycles of non-evaporating fuel along the intake pipe γ and time between next cycles Δt. The calculation results of identification of the model parameters are presented in the form of radar graphs. The figures shows the averages declines and increases of the injection time and the average values for both types of stroke. These studies shown, that the change of the position of the cylinder will cause changes in the formation of fuel-air mixture and thus changes in the combustion process. Based on the results of the work of simulation and experiments was possible to develop individual algorithms for ignition control. This work has been financed by the Polish National Centre for Research and Development, INNOLOT, under Grant Agreement No. INNOLOT/I/1/NCBR/2013.

Keywords: radial engine, ignition system, non-uniformity, combustion process

Procedia PDF Downloads 366

Authors: Ahmed Elaksher, Islam Omar

Abstract:

Accurate geometric relation algorithms are imperative in Earth and planetary satellite and aerial image processing, particularly for high-resolution images that are used for topographic mapping. Most of these satellites carry push-broom sensors. These sensors are optical scanners equipped with linear arrays of CCDs. These sensors have been deployed on most EOSs. In addition, the LROC is equipped with two push NACs that provide 0.5 meter-scale panchromatic images over a 5 km swath of the Moon. The HiRISE carried by the MRO and the HRSC carried by MEX are examples of push-broom sensor that produces images of the surface of Mars. Sensor models developed in photogrammetry relate image space coordinates in two or more images with the 3D coordinates of ground features. Rigorous sensor models use the actual interior orientation parameters and exterior orientation parameters of the camera, unlike approximate models. In this research, we generate a generic push-broom sensor model to process imageries acquired through linear array cameras and investigate its performance, advantages, and disadvantages in generating topographic models for the Earth, Mars, and the Moon. We also compare and contrast the utilization, effectiveness, and applicability of available photogrammetric techniques and softcopies with the developed model. We start by defining an image reference coordinate system to unify image coordinates from all three arrays. The transformation from an image coordinate system to a reference coordinate system involves a translation and three rotations. For any image point within the linear array, its image reference coordinates, the coordinates of the exposure center of the array in the ground coordinate system at the imaging epoch (t), and the corresponding ground point coordinates are related through the collinearity condition that states that all these three points must be on the same line. The rotation angles for each CCD array at the epoch t are defined and included in the transformation model. The exterior orientation parameters of an image line, i.e., coordinates of exposure station and rotation angles, are computed by a polynomial interpolation function in time (t). The parameter (t) is the time at a certain epoch from a certain orbit position. Depending on the types of observations, coordinates, and parameters may be treated as knowns or unknowns differently in various situations. The unknown coefficients are determined in a bundle adjustment. The orientation process starts by extracting the sensor position and, orientation and raw images from the PDS. The parameters of each image line are then estimated and imported into the push-broom sensor model. We also define tie points between image pairs to aid the bundle adjustment model, determine the refined camera parameters, and generate highly accurate topographic maps. The model was tested on different satellite images such as IKONOS, QuickBird, and WorldView-2, HiRISE. It was found that the accuracy of our model is comparable to those of commercial and open-source software, the computational efficiency of the developed model is high, the model could be used in different environments with various sensors, and the implementation process is much more cost-and effort-consuming.

Keywords: photogrammetry, push-broom sensors, IKONOS, HiRISE, collinearity condition

Procedia PDF Downloads 63

Authors: Armin Nabaei, M. Omair Ahmad, M. N. S. Swamy

Abstract:

Many facial expression and emotion recognition methods in the traditional approaches of using LDA, PCA, and EBGM have been proposed. In recent years deep learning models have provided a unique platform addressing by automatically extracting the features for the detection of facial expression and emotions. However, deep networks require large training datasets to extract automatic features effectively. In this work, we propose an efficient emotion detection algorithm using face images when only small datasets are available for training. We design a deep network whose feature extraction capability is enhanced by utilizing several parallel modules between the input and output of the network, each focusing on the extraction of different types of coarse features with fined grained details to break the symmetry of produced information. In fact, we leverage long range dependencies, which is one of the main drawback of CNNs. We develop this work by introducing a Dynamic Soft-Margin SoftMax.The conventional SoftMax suffers from reaching to gold labels very soon, which take the model to over-fitting. Because it’s not able to determine adequately discriminant feature vectors for some variant class labels. We reduced the risk of over-fitting by using a dynamic shape of input tensor instead of static in SoftMax layer with specifying a desired Soft- Margin. In fact, it acts as a controller to how hard the model should work to push dissimilar embedding vectors apart. For the proposed Categorical Loss, by the objective of compacting the same class labels and separating different class labels in the normalized log domain.We select penalty for those predictions with high divergence from ground-truth labels.So, we shorten correct feature vectors and enlarge false prediction tensors, it means we assign more weights for those classes with conjunction to each other (namely, “hard labels to learn”). By doing this work, we constrain the model to generate more discriminate feature vectors for variant class labels. Finally, for the proposed optimizer, our focus is on solving weak convergence of Adam optimizer for a non-convex problem. Our noteworthy optimizer is working by an alternative updating gradient procedure with an exponential weighted moving average function for faster convergence and exploiting a weight decay method to help drastically reducing the learning rate near optima to reach the dominant local minimum. We demonstrate the superiority of our proposed work by surpassing the first rank of three widely used Facial Expression Recognition datasets with 93.30% on FER-2013, and 16% improvement compare to the first rank after 10 years, reaching to 90.73% on RAF-DB, and 100% k-fold average accuracy for CK+ dataset, and shown to provide a top performance to that provided by other networks, which require much larger training datasets.

Keywords: computer vision, facial expression recognition, machine learning, algorithms, depp learning, neural networks

Procedia PDF Downloads 74

Authors: Raptis Sotirios

Abstract:

Health and social care (HSc) services planning and scheduling are facing unprecedented challenges due to the pandemic pressure and also suffer from unplanned spending that is negatively impacted by the global financial crisis. Data-driven can help to improve policies, plan and design services provision schedules using algorithms assist healthcare managers’ to face unexpected demands using fewer resources. The paper discusses services packing using statistical significance tests and machine learning (ML) to evaluate demands similarity and coupling. This is achieved by predicting the range of the demand (class) using ML methods such as CART, random forests (RF), and logistic regression (LGR). The significance tests Chi-Squared test and Student test are used on data over a 39 years span for which HSc services data exist for services delivered in Scotland. The demands are probabilistically associated through statistical hypotheses that assume that the target service’s demands are statistically dependent on other demands as a NULL hypothesis. This linkage can be confirmed or not by the data. Complementarily, ML methods are used to linearly predict the above target demands from the statistically found associations and extend the linear dependence of the target’s demand to independent demands forming, thus groups of services. Statistical tests confirm ML couplings making the prediction also statistically meaningful and prove that a target service can be matched reliably to other services, and ML shows these indicated relationships can also be linear ones. Zero paddings were used for missing years records and illustrated better such relationships both for limited years and in the entire span offering long term data visualizations while limited years groups explained how well patients numbers can be related in short periods or can change over time as opposed to behaviors across more years. The prediction performance of the associations is measured using Receiver Operating Characteristic(ROC) AUC and ACC metrics as well as the statistical tests, Chi-Squared and Student. Co-plots and comparison tables for RF, CART, and LGR as well as p-values and Information Exchange(IE), are provided showing the specific behavior of the ML and of the statistical tests and the behavior using different learning ratios. The impact of k-NN and cross-correlation and C-Means first groupings is also studied over limited years and the entire span. It was found that CART was generally behind RF and LGR, but in some interesting cases, LGR reached an AUC=0 falling below CART, while the ACC was as high as 0.912, showing that ML methods can be confused padding or by data irregularities or outliers. On average, 3 linear predictors were sufficient, LGR was found competing RF well, and CART followed with the same performance at higher learning ratios. Services were packed only if when significance level(p-value) of their association coefficient was more than 0.05. Social factors relationships were observed between home care services and treatment of old people, birth weights, alcoholism, drug abuse, and emergency admissions. The work found that different HSc services can be well packed as plans of limited years, across various services sectors, learning configurations, as confirmed using statistical hypotheses.

Keywords: class, cohorts, data frames, grouping, prediction, prob-ability, services

Procedia PDF Downloads 231

Authors: Ursula Das, Manavjit Singh Dhindsa, Kshirasagar Naik, Marzia Zaman, Richard Purcell, Srinivas Sampalli, Abdul Mutakabbir, Chung-Horng Lung, Thambirajah Ravichandran

Abstract:

Wildland fires, also known as forest fires or wildfires, are exhibiting an alarming surge in frequency in recent times, further adding to its perennial global concern. Forest fires often lead to devastating consequences ranging from loss of healthy forest foliage and wildlife to substantial economic losses and the tragic loss of human lives. Despite the existence of substantial literature on the detection of active forest fires, numerous potential research avenues in forest fire management, such as preventative measures and ancillary effects of forest fires, remain largely underexplored. This paper undertakes a systematic review of these underexplored areas in forest fire research, meticulously categorizing them into distinct phases, namely pre-fire, during-fire, and post-fire stages. The pre-fire phase encompasses the assessment of fire risk, analysis of fuel properties, and other activities aimed at preventing or reducing the risk of forest fires. The during-fire phase includes activities aimed at reducing the impact of active forest fires, such as the detection and localization of active fires, optimization of wildfire suppression methods, and prediction of the behavior of active fires. The post-fire phase involves analyzing the impact of forest fires on various aspects, such as the extent of damage in forest areas, post-fire regeneration of forests, impact on wildlife, economic losses, and health impacts from byproducts produced during burning. A comprehensive understanding of the three stages is imperative for effective forest fire management and mitigation of the impact of forest fires on both ecological systems and human well-being. Artificial intelligence and machine learning (AI/ML) methods have garnered much attention in the cyber-physical systems domain in recent times leading to their adoption in decision-making in diverse applications including disaster management. This paper explores the current state of AI/ML applications for managing the activities in the aforementioned phases of forest fire. While conventional machine learning and deep learning methods have been extensively explored for the prevention, detection, and management of forest fires, a systematic classification of these methods into distinct AI research domains is conspicuously absent. This paper gives a comprehensive overview of the state of forest fire research across more recent and prominent AI/ML disciplines, including big data, classical machine learning, computer vision, explainable AI, generative AI, natural language processing, optimization algorithms, and time series forecasting. By providing a detailed overview of the potential areas of research and identifying the diverse ways AI/ML can be employed in forest fire research, this paper aims to serve as a roadmap for future investigations in this domain.

Keywords: artificial intelligence, computer vision, deep learning, during-fire activities, forest fire management, machine learning, pre-fire activities, post-fire activities

Procedia PDF Downloads 72

Authors: V. V. Fomin, A. P. Mikhailovich, E. M. Agapitov, V. E. Rogachev, E. A. Kostousova, E. S. Perekhodova

Abstract:

Ground-level landscape photos can be used as a source of objective data on woody vegetation and vegetation dynamics. We proposed a method for processing, analyzing, and presenting ground photographs, which has the following advantages: 1) researcher has to form holistic representation of the study area in form of a set of interlapping ground-level landscape photographs; 2) it is necessary to define or obtain characteristics of the landscape, objects, and phenomena present on the photographs; 3) it is necessary to create new or supplement existing textual descriptions and annotations for the ground-level landscape photographs; 4) single or multiple ground-level landscape photographs can be used to develop specialized geoinformation layers, schematic maps or thematic maps; 5) it is necessary to determine quantitative data that describes both images as a whole, and displayed objects and phenomena, using algorithms for automated image analysis. It is suggested to match each photo with a polygonal geoinformation layer, which is a sector consisting of areas corresponding with parts of the landscape visible in the photos. Calculation of visibility areas is performed in a geoinformation system within a sector using a digital model of a study area relief and visibility analysis functions. Superposition of the visibility sectors corresponding with various camera viewpoints allows matching landscape photos with each other to create a complete and wholesome representation of the space in question. It is suggested to user-defined data or phenomenons on the images with the following superposition over the visibility sector in the form of map symbols. The technology of geoinformation layers’ spatial superposition over the visibility sector creates opportunities for image geotagging using quantitative data obtained from raster or vector layers within the sector with the ability to generate annotations in natural language. The proposed method has proven itself well for relatively open and clearly visible areas with well-defined relief, for example, in mountainous areas in the treeline ecotone. When the polygonal layers of visibility sectors for a large number of different points of photography are topologically superimposed, a layer of visibility of sections of the entire study area is formed, which is displayed in the photographs. Also, as a result of this overlapping of sectors, areas that did not appear in the photo will be assessed as gaps. According to the results of this procedure, it becomes possible to obtain information about the photos that display a specific area and from which points of photography it is visible. This information may be obtained either as a query on the map or as a query for the attribute table of the layer. The method was tested using repeated photos taken from forty camera viewpoints located on Ray-Iz mountain massif (Polar Urals, Russia) from 1960 until 2023. It has been successfully used in combination with other ground-based and remote sensing methods of studying the climate-driven dynamics of woody vegetation in the Polar Urals. Acknowledgment: This research was collaboratively funded by the Russian Ministry for Science and Education project No. FEUG-2023-0002 (image representation) and Russian Science Foundation project No. 24-24-00235 (automated textual description).

Keywords: woody, vegetation, repeated, photographs

Procedia PDF Downloads 89

Authors: M. Eickermann, F. Ronellenfitsch, J. Junk

Abstract:

Oilseed rape (Brassica napus L.) is one of the most important crops throughout Europe, but pressure due to pest insects and pathogens can reduce yield amount substantially. Therefore, the usage of pesticide applications is outstanding in this crop. In addition, climate change effects can interact with phenology of the host plant and their pests and can apply additional pressure on the yield. Next to the pollen beetle, Meligethes aeneus L., the seed-damaging pest insects, cabbage seed weevil (Ceutorhynchus obstrictus Marsham) and the brassica pod midge (Dasineura brassicae Winn.) are of main economic impact to the yield. While females of C. obstrictus are infesting oilseed rape by depositing single eggs into young pods, the females of D. brassicae are using this local damage in the pod for their own oviposition, while depositing batches of 20-30 eggs. Without a former infestation by the cabbage seed weevil, a significant yield reduction by the brassica pod midge can be denied. Based on long-term, multisided field experiments, a comprehensive data-set on pest migration to crops of B. napus has been built up in the last ten years. Five observational test sides, situated in different climatic regions in Luxembourg were controlled between February until the end of May twice a week. Pest migration was recorded by using yellow water pan-traps. Caught insects were identified in the laboratory according to species specific identification keys. By a combination of pest observations and corresponding meteorological observations, the set-up of models to predict the migration periods of the seed-damaging pests was possible. This approach is the basis for a computer-based decision support tool, to assist the farmer in identifying the appropriate time point of pesticide application. In addition, the derived algorithms of that decision support tool can be combined with climate change projections in order to assess the future potential threat caused by the seed-damaging pest species. Regional climate change effects for Luxembourg have been intensively studied in recent years. Significant changes to wetter winters and drier summers, as well as a prolongation of the vegetation period mainly caused by higher spring temperature, have also been reported. We used the COSMO-CLM model to perform a time slice experiment for Luxembourg with a spatial resolution of 1.3 km. Three ten year time slices were calculated: The reference time span (1991-2000), the near (2041-2050) and the far future (2091-2100). Our results projected a significant shift of pest migration to an earlier onset of the year. In addition, a prolongation of the possible migration period could be observed. Because D. brassiace is depending on the former oviposition activity by C. obstrictus to infest its host plant successfully, the future dependencies of both pest species will be assessed. Based on this approach the future risk potential of both seed-damaging pests is calculated and the status as pest species is characterized.

Keywords: CORDEX projections, decision support tool, Brassica napus, pests

Procedia PDF Downloads 382

Authors: David F. Nettleton, Elodie Bugnicourt, Christian Wasiak, Alejandro Rosales

Abstract:

In this presentation, preliminary results are given for the modeling and calibration of two different industrial winding MIMO (Multiple Input Multiple Output) processes using machine learning techniques. In contrast to previous approaches which have typically used ‘black-box’ linear statistical methods together with a definition of the mechanical behavior of the process, we use non-linear machine learning algorithms together with a ‘white-box’ rule induction technique to create a supervised model of the fitting error between the expected and real force measures. The final objective is to build a precise model of the winding process in order to control de-tension of the material being wound in the first case, and the friction of the material passing through the die, in the second case. Case 1, Tension Control of a Winding Process. A plastic web is unwound from a first reel, goes over a traction reel and is rewound on a third reel. The objectives are: (i) to train a model to predict the web tension and (ii) calibration to find the input values which result in a given tension. Case 2, Friction Force Control of a Micro-Pullwinding Process. A core+resin passes through a first die, then two winding units wind an outer layer around the core, and a final pass through a second die. The objectives are: (i) to train a model to predict the friction on die2; (ii) calibration to find the input values which result in a given friction on die2. Different machine learning approaches are tested to build models, Kernel Ridge Regression, Support Vector Regression (with a Radial Basis Function Kernel) and MPART (Rule Induction with continuous value as output). As a previous step, the MPART rule induction algorithm was used to build an explicative model of the error (the difference between expected and real friction on die2). The modeling of the error behavior using explicative rules is used to help improve the overall process model. Once the models are built, the inputs are calibrated by generating Gaussian random numbers for each input (taking into account its mean and standard deviation) and comparing the output to a target (desired) output until a closest fit is found. The results of empirical testing show that a high precision is obtained for the trained models and for the calibration process. The learning step is the slowest part of the process (max. 5 minutes for this data), but this can be done offline just once. The calibration step is much faster and in under one minute obtained a precision error of less than 1x10-3 for both outputs. To summarize, in the present work two processes have been modeled and calibrated. A fast processing time and high precision has been achieved, which can be further improved by using heuristics to guide the Gaussian calibration. Error behavior has been modeled to help improve the overall process understanding. This has relevance for the quick optimal set up of many different industrial processes which use a pull-winding type process to manufacture fibre reinforced plastic parts. Acknowledgements to the Openmind project which is funded by Horizon 2020 European Union funding for Research & Innovation, Grant Agreement number 680820

Keywords: data model, machine learning, industrial winding, calibration

Procedia PDF Downloads 241

Authors: B. R. Campomanes-Alvarez, P. Quiros, B. Fernandez

Abstract:

Automatic Speech Recognition (ASR) is a machine-based process of decoding and transcribing oral speech. A typical ASR system receives acoustic input from a speaker or an audio file, analyzes it using algorithms, and produces an output in the form of a text. Some speech recognition systems use Hidden Markov Models (HMMs) to deal with the temporal variability of speech and Gaussian Mixture Models (GMMs) to determine how well each state of each HMM fits a short window of frames of coefficients that represents the acoustic input. Another way to evaluate the fit is to use a feed-forward neural network that takes several frames of coefficients as input and produces posterior probabilities over HMM states as output. Deep neural networks (DNNs) that have many hidden layers and are trained using new methods have been shown to outperform GMMs on a variety of speech recognition systems. Acoustic models for state-of-the-art ASR systems are usually training on massive amounts of data. However, audio files with their corresponding transcriptions can be difficult to obtain, especially in the Spanish language. Hence, in the case of these low-resource scenarios, building an ASR model is considered as a complex task due to the lack of labeled data, resulting in an under-trained system. Semi-supervised learning approaches arise as necessary tasks given the high cost of transcribing audio data. The main goal of this proposal is to develop a procedure based on acoustic semi-supervised learning for Spanish ASR systems by using DNNs. This semi-supervised learning approach consists of: (a) Training a seed ASR model with a DNN using a set of audios and their respective transcriptions. A DNN with a one-hidden-layer network was initialized; increasing the number of hidden layers in training, to a five. A refinement, which consisted of the weight matrix plus bias term and a Stochastic Gradient Descent (SGD) training were also performed. The objective function was the cross-entropy criterion. (b) Decoding/testing a set of unlabeled data with the obtained seed model. (c) Selecting a suitable subset of the validated data to retrain the seed model, thereby improving its performance on the target test set. To choose the most precise transcriptions, three confidence scores or metrics, regarding the lattice concept (based on the graph cost, the acoustic cost and a combination of both), was performed as selection technique. The performance of the ASR system will be calculated by means of the Word Error Rate (WER). The test dataset was renewed in order to extract the new transcriptions added to the training dataset. Some experiments were carried out in order to select the best ASR results. A comparison between a GMM-based model without retraining and the DNN proposed system was also made under the same conditions. Results showed that the semi-supervised ASR-model based on DNNs outperformed the GMM-model, in terms of WER, in all tested cases. The best result obtained an improvement of 6% relative WER. Hence, these promising results suggest that the proposed technique could be suitable for building ASR models in low-resource environments.

Keywords: automatic speech recognition, deep neural networks, machine learning, semi-supervised learning

Procedia PDF Downloads 339

Authors: Rohit T. P., Sahil Athrij, Sasi Gopalan

Abstract:

Currently, the content entertainment industry is dominated by mobile devices. As the trends slowly shift towards Augmented/Virtual Reality applications the computational demands on these devices are increasing exponentially and we are already reaching the limits of hardware optimizations. This paper proposes a software solution to this problem. By leveraging the capabilities of cloud computing we can offload the work from mobile devices to dedicated rendering servers that are way more powerful. But this introduces the problem of latency. This paper introduces a protocol that can achieve high-performance low latency Augmented/Virtual Reality experience. There are two parts to the protocol, 1) In-flight compression The main cause of latency in the system is the time required to transmit the camera frame from client to server. The round trip time is directly proportional to the amount of data transmitted. This can therefore be reduced by compressing the frames before sending. Using some standard compression algorithms like JPEG can result in minor size reduction only. Since the images to be compressed are consecutive camera frames there won't be a lot of changes between two consecutive images. So inter-frame compression is preferred. Inter-frame compression can be implemented efficiently using WebGL but the implementation of WebGL limits the precision of floating point numbers to 16bit in most devices. This can introduce noise to the image due to rounding errors, which will add up eventually. This can be solved using an improved interframe compression algorithm. The algorithm detects changes between frames and reuses unchanged pixels from the previous frame. This eliminates the need for floating point subtraction thereby cutting down on noise. The change detection is also improved drastically by taking the weighted average difference of pixels instead of the absolute difference. The kernel weights for this comparison can be fine-tuned to match the type of image to be compressed. 2) Dynamic Load distribution Conventional cloud computing architectures work by offloading as much work as possible to the servers, but this approach can cause a hit on bandwidth and server costs. The most optimal solution is obtained when the device utilizes 100% of its resources and the rest is done by the server. The protocol balances the load between the server and the client by doing a fraction of the computing on the device depending on the power of the device and network conditions. The protocol will be responsible for dynamically partitioning the tasks. Special flags will be used to communicate the workload fraction between the client and the server and will be updated in a constant interval of time ( or frames ). The whole of the protocol is designed so that it can be client agnostic. Flags are available to the client for resetting the frame, indicating latency, switching mode, etc. The server can react to client-side changes on the fly and adapt accordingly by switching to different pipelines. The server is designed to effectively spread the load and thereby scale horizontally. This is achieved by isolating client connections into different processes.

Keywords: 2D kernelling, augmented reality, cloud computing, dynamic load distribution, immersive experience, mobile computing, motion tracking, protocols, real-time systems, web-based augmented reality application

Procedia PDF Downloads 72

Authors: Caroline Atef Shoukry Tadros

Abstract:

Massive Urbanization growth threatens the sustainability of cities and the quality of city life. This raised the need for an alternate model of sustainability, so we need to plan the future cities in a smarter way with smarter mobility. Smart Mobility planning applications are solutions that use digital technologies and infrastructure advances to improve the efficiency, sustainability, and inclusiveness of urban transportation systems. They can contribute to meeting the needs of Urbanization growth by addressing the challenges of traffic congestion, pollution, accessibility, and safety in cities. Some example of a Smart Mobility planning application are Mobility-as-a-service: This is a service that integrates different transport modes, such as public transport, shared mobility, and active mobility, into a single platform that allows users to plan, book, and pay for their trips. This can reduce the reliance on private cars, optimize the use of existing infrastructure, and provide more choices and convenience for travelers. MaaS Global is a company that offers mobility-as-a-service solutions in several cities around the world. Traffic flow optimization: This is a solution that uses data analytics, artificial intelligence, and sensors to monitor and manage traffic conditions in real-time. This can reduce congestion, emissions, and travel time, as well as improve road safety and user satisfaction. Waycare is a platform that leverages data from various sources, such as connected vehicles, mobile applications, and road cameras, to provide traffic management agencies with insights and recommendations to optimize traffic flow. Logistics optimization: This is a solution that uses smart algorithms, blockchain, and IoT to improve the efficiency and transparency of the delivery of goods and services in urban areas. This can reduce the costs, emissions, and delays associated with logistics, as well as enhance the customer experience and trust. ShipChain is a blockchain-based platform that connects shippers, carriers, and customers and provides end-to-end visibility and traceability of the shipments. Autonomous vehicles: This is a solution that uses advanced sensors, software, and communication systems to enable vehicles to operate without human intervention. This can improve the safety, accessibility, and productivity of transportation, as well as reduce the need for parking space and infrastructure maintenance. Waymo is a company that develops and operates autonomous vehicles for various purposes, such as ride-hailing, delivery, and trucking. These are some of the ways that Smart Mobility planning applications can contribute to meeting the needs of the Urbanization growth. However, there are also various opportunities and challenges related to the implementation and adoption of these solutions, such as the regulatory, ethical, social, and technical aspects. Therefore, it is important to consider the specific context and needs of each city and its stakeholders when designing and deploying Smart Mobility planning applications.

Keywords: smart mobility planning, smart mobility applications, smart mobility techniques, smart mobility tools, smart transportation, smart cities, urbanization growth, future smart cities, intelligent cities, ICT information and communications technologies, IoT internet of things, sensors, lidar, digital twin, ai artificial intelligence, AR augmented reality, VR virtual reality, robotics, cps cyber physical systems, citizens design science

Procedia PDF Downloads 73

Authors: Alfredo Fortunato, Carmelofrancesco Origlia, Sara Laurita, Rossella Nicoletti

Abstract:

In the post-pandemic recovery phase of tourism, the role of a Destination Management Organization (DMO) as a coordinated management system of all the elements that make up a destination (attractions, access, marketing, human resources, brand, pricing, etc.) is also becoming relevant for local territories. The objective of a DMO is to maximize the visitor's perception of value and quality while ensuring the competitiveness and sustainability of the destination, as well as the long-term preservation of its natural and cultural assets, and to catalyze benefits for the local economy and residents. In carrying out the multiple functions to which it is called, the DMO can leverage a collective intelligence that comes from the ability to pool information, explicit and tacit knowledge, and relationships of the various stakeholders: policymakers, public managers and officials, entrepreneurs in the tourism supply chain, researchers, data journalists, schools, associations and committees, citizens, etc. The DMO potentially has at its disposal large volumes of data and many of them at low cost, that need to be properly processed to produce value. Based on these assumptions, the paper presents a conceptual framework for building an information system to support the DMO in the intelligent management of a tourist destination tested in an area of southern Italy. The approach adopted is data-informed and consists of four phases: (1) formulation of the knowledge problem (analysis of policy documents and industry reports; focus groups and co-design with stakeholders; definition of information needs and key questions); (2) research and metadatation of relevant sources (reconnaissance of official sources, administrative archives and internal DMO sources); (3) gap analysis and identification of unconventional information sources (evaluation of traditional sources with respect to the level of consistency with information needs, the freshness of information and granularity of data; enrichment of the information base by identifying and studying web sources such as Wikipedia, Google Trends, Booking.com, Tripadvisor, websites of accommodation facilities and online newspapers); (4) definition of the set of indicators and construction of the information base (specific definition of indicators and procedures for data acquisition, transformation, and analysis). The framework derived consists of 6 thematic areas (accommodation supply, cultural heritage, flows, value, sustainability, and enabling factors), each of which is divided into three domains that gather a specific information need to be represented by a scheme of questions to be answered through the analysis of available indicators. The framework is characterized by a high degree of flexibility in the European context, given that it can be customized for each destination by adapting the part related to internal sources. Application to the case study led to the creation of a decision support system that allows: •integration of data from heterogeneous sources, including through the execution of automated web crawling procedures for data ingestion of social and web information; •reading and interpretation of data and metadata through guided navigation paths in the key of digital story-telling; •implementation of complex analysis capabilities through the use of data mining algorithms such as for the prediction of tourist flows.

Keywords: collective intelligence, data framework, destination management, smart tourism

Procedia PDF Downloads 121

Authors: Pinzhe Zhao

Abstract:

This study focuses on predicting university students' cheating tendencies using psychological traits and machine learning techniques. Academic dishonesty is a significant issue that compromises the integrity and fairness of educational institutions. While much research has been dedicated to detecting cheating behaviors after they have occurred, there is limited work on predicting such tendencies before they manifest. The aim of this research is to develop a model that can identify students who are at higher risk of engaging in academic misconduct, allowing for earlier interventions to prevent such behavior. Psychological factors are known to influence students' likelihood of cheating. Research shows that traits such as test anxiety, moral reasoning, self-efficacy, and achievement motivation are strongly linked to academic dishonesty. High levels of anxiety may lead students to cheat as a way to cope with pressure. Those with lower self-efficacy are less confident in their academic abilities, which can push them toward dishonest behaviors to secure better outcomes. Students with weaker moral judgment may also justify cheating more easily, believing it to be less wrong under certain conditions. Achievement motivation also plays a role, as students driven primarily by external rewards, such as grades, are more likely to cheat compared to those motivated by intrinsic learning goals. In this study, data on students’ psychological traits is collected through validated assessments, including scales for anxiety, moral reasoning, self-efficacy, and motivation. Additional data on academic performance, attendance, and engagement in class are also gathered to create a more comprehensive profile. Using machine learning algorithms such as Random Forest, Support Vector Machines (SVM), and Long Short-Term Memory (LSTM) networks, the research builds models that can predict students’ cheating tendencies. These models are trained and evaluated using metrics like accuracy, precision, recall, and F1 scores to ensure they provide reliable predictions. The findings demonstrate that combining psychological traits with machine learning provides a powerful method for identifying students at risk of cheating. This approach allows for early detection and intervention, enabling educational institutions to take proactive steps in promoting academic integrity. The predictive model can be used to inform targeted interventions, such as counseling for students with high test anxiety or workshops aimed at strengthening moral reasoning. By addressing the underlying factors that contribute to cheating behavior, educational institutions can reduce the occurrence of academic dishonesty and foster a culture of integrity. In conclusion, this research contributes to the growing body of literature on predictive analytics in education. It offers a approach by integrating psychological assessments with machine learning to predict cheating tendencies. This method has the potential to significantly improve how academic institutions address academic dishonesty, shifting the focus from punishment after the fact to prevention before it occurs. By identifying high-risk students and providing them with the necessary support, educators can help maintain the fairness and integrity of the academic environment.

Keywords: academic dishonesty, cheating prediction, intervention strategies, machine learning, psychological traits, academic integrity

Procedia PDF Downloads 20

Authors: Stanislav Babaev, Vladimir Cuk, Sjef Cobben, Jan Desmet

Abstract:

The procedures for performing harmonic studies for medium-voltage distribution feeders have become relatively mature topics since the early 1980s. The efforts of various electric power engineers and researchers were mainly focused on handling large harmonic non-linear loads connected scarcely at several buses of medium-voltage feeders. In order to assess the impact of these loads on the voltage quality of the distribution system, specific modeling and simulation strategies were proposed. These methodologies could deliver a reasonable estimation accuracy given the requirements of least computational efforts and reduced complexity. To uphold these requirements, certain analysis assumptions have been made, which became de facto standards for establishing guidelines for harmonic analysis. Among others, typical assumptions include balanced conditions of the study and the negligible impact of impedance frequency characteristics of various power system components. In latter, skin and proximity effects are usually omitted, and resistance and reactance values are modeled based on the theoretical equations. Further, the simplifications of the modelling routine have led to the commonly accepted practice of neglecting phase angle diversity effects. This is mainly associated with developed load models, which only in a handful of cases are representing the complete harmonic behavior of a certain device as well as accounting on the harmonic interaction between grid harmonic voltages and harmonic currents. While these modelling practices were proven to be reasonably effective for medium-voltage levels, similar approaches have been adopted for low-voltage distribution systems. Given modern conditions and massive increase in usage of residential electronic devices, recent and ongoing boom of electric vehicles, and large-scale installing of distributed solar power, the harmonics in current low-voltage grids are characterized by high degree of variability and demonstrate sufficient diversity leading to a certain level of cancellation effects. It is obvious, that new modelling algorithms overcoming previously made assumptions have to be accepted. In this work, a simulation approach aimed to deal with some of the typical assumptions is proposed. A practical low-voltage feeder is modeled in PowerFactory. In order to demonstrate the importance of diversity effect and harmonic interaction, previously developed measurement-based models of photovoltaic inverter and battery charger are used as loads. The Python-based script aiming to supply varying voltage background distortion profile and the associated current harmonic response of loads is used as the core of unbalanced simulation. Furthermore, the impact of uncertainty of feeder frequency-impedance characteristics on total harmonic distortion levels is shown along with scenarios involving linear resistive loads, which further alter the impedance of the system. The comparative analysis demonstrates sufficient differences with cases when all the assumptions are in place, and results indicate that new modelling and simulation procedures need to be adopted for low-voltage distribution systems with high penetration of non-linear loads and renewable generation.

Keywords: electric power system, harmonic distortion, power quality, public low-voltage network, harmonic modelling

Procedia PDF Downloads 158

Authors: Ksenia Volkova, Artur Petrosyan, Ignatii Dubyshkin, Alexei Ossadtchi

Abstract:

Brain-computer interfaces are a growing research field producing many implementations that find use in different fields and are used for research and practical purposes. Despite the popularity of the implementations using non-invasive neuroimaging methods, radical improvement of the state channel bandwidth and, thus, decoding accuracy is only possible by using invasive techniques. Electrocorticography (ECoG) is a minimally invasive neuroimaging method that provides highly informative brain activity signals, effective analysis of which requires the use of machine learning methods that are able to learn representations of complex patterns. Deep learning is a family of machine learning algorithms that allow learning representations of data with multiple levels of abstraction. This study explores the potential of deep learning approaches for ECoG processing, decoding movement intentions and the perception of proprioceptive information. To obtain synchronous recording of kinematic movement characteristics and corresponding electrical brain activity, a series of experiments were carried out, during which subjects performed finger movements at their own pace. Finger movements were recorded with a three-axis accelerometer, while ECoG was synchronously registered from the electrode strips that were implanted over the contralateral sensorimotor cortex. Then, multichannel ECoG signals were used to track finger movement trajectory characterized by accelerometer signal. This process was carried out both causally and non-causally, using different position of the ECoG data segment with respect to the accelerometer data stream. The recorded data was split into training and testing sets, containing continuous non-overlapping fragments of the multichannel ECoG. A deep convolutional neural network was implemented and trained, using 1-second segments of ECoG data from the training dataset as input. To assess the decoding accuracy, correlation coefficient r between the output of the model and the accelerometer readings was computed. After optimization of hyperparameters and training, the deep learning model allowed reasonably accurate causal decoding of finger movement with correlation coefficient r = 0.8. In contrast, the classical Wiener-filter like approach was able to achieve only 0.56 in the causal decoding mode. In the noncausal case, the traditional approach reached the accuracy of r = 0.69, which may be due to the presence of additional proprioceptive information. This result demonstrates that the deep neural network was able to effectively find a representation of the complex top-down information related to the actual movement rather than proprioception. The sensitivity analysis shows physiologically plausible pictures of the extent to which individual features (channel, wavelet subband) are utilized during the decoding procedure. In conclusion, the results of this study have demonstrated that a combination of a minimally invasive neuroimaging technique such as ECoG and advanced machine learning approaches allows decoding motion with high accuracy. Such setup provides means for control of devices with a large number of degrees of freedom as well as exploratory studies of the complex neural processes underlying movement execution.

Keywords: brain-computer interface, deep learning, ECoG, movement decoding, sensorimotor cortex

Procedia PDF Downloads 177

Authors: Monica Racha, Siva Chandrasekaran, Alex Stojcevski

Abstract:

The engineering programs focus on promoting students' personal and professional development by ensuring that students acquire technical and professional competencies during four-year studies. The traditional engineering laboratory provides an opportunity for students to "practice by doing," and laboratory facilities aid them in obtaining insight and understanding of their discipline. Due to rapid technological advancements and the current COVID-19 outbreak, the traditional labs were transforming into virtual learning environments. Aim: To better understand the limitations of the physical laboratory, this research study aims to use a Machine Learning (ML) algorithm that interfaces with the Augmented Reality HoloLens and predicts the image behavior to classify and detect the electronic components. The automated electronic components error classification and detection automatically detect and classify the position of all components on a breadboard by using the ML algorithm. This research will assist first-year undergraduate engineering students in conducting laboratory practices without any supervision. With the help of HoloLens, and ML algorithm, students will reduce component placement error on a breadboard and increase the efficiency of simple laboratory practices virtually. Method: The images of breadboards, resistors, capacitors, transistors, and other electrical components will be collected using HoloLens 2 and stored in a database. The collected image dataset will then be used for training a machine learning model. The raw images will be cleaned, processed, and labeled to facilitate further analysis of components error classification and detection. For instance, when students conduct laboratory experiments, the HoloLens captures images of students placing different components on a breadboard. The images are forwarded to the server for detection in the background. A hybrid Convolutional Neural Networks (CNNs) and Support Vector Machines (SVMs) algorithm will be used to train the dataset for object recognition and classification. The convolution layer extracts image features, which are then classified using Support Vector Machine (SVM). By adequately labeling the training data and classifying, the model will predict, categorize, and assess students in placing components correctly. As a result, the data acquired through HoloLens includes images of students assembling electronic components. It constantly checks to see if students appropriately position components in the breadboard and connect the components to function. When students misplace any components, the HoloLens predicts the error before the user places the components in the incorrect proportion and fosters students to correct their mistakes. This hybrid Convolutional Neural Networks (CNNs) and Support Vector Machines (SVMs) algorithm automating electronic component error classification and detection approach eliminates component connection problems and minimizes the risk of component damage. Conclusion: These augmented reality smart glasses powered by machine learning provide a wide range of benefits to supervisors, professionals, and students. It helps customize the learning experience, which is particularly beneficial in large classes with limited time. It determines the accuracy with which machine learning algorithms can forecast whether students are making the correct decisions and completing their laboratory tasks.

Keywords: augmented reality, machine learning, object recognition, virtual laboratories

Procedia PDF Downloads 134

Authors: F. Tang, S. Boskovich, A. Raheja, Z. Aliyazicioglu, S. Bhandari, N. Tsuchiya

Abstract:

Calpoly Autonomous Transportation Experience (CATE) is a driverless vehicle that we are developing to provide safe, accessible, and efficient transportation of passengers throughout the Cal Poly Pomona campus for events such as orientation tours. Unlike the other self-driving vehicles that are usually developed to operate with other vehicles and reside only on the road networks, CATE will operate exclusively on walk-paths of the campus (potentially narrow passages) with pedestrians traveling from multiple locations. Safety becomes paramount as CATE operates within the same environment as pedestrians. As driverless vehicles assume greater roles in today’s transportation, this project will contribute to autonomous driving with pedestrian traffic in a highly dynamic environment. The CATE project requires significant interdisciplinary work. Researchers from mechanical engineering, electrical engineering and computer science are working together to attack the problem from different perspectives (hardware, software and system). In this abstract, we describe the software aspects of the project, with a focus on the requirements and the major components. CATE shall provide a GUI interface for the average user to interact with the car and access its available functionalities, such as selecting a destination from any origin on campus. We have developed an interface that provides an aerial view of the campus map, the current car location, routes, and the goal location. Users can interact with CATE through audio or manual inputs. CATE shall plan routes from the origin to the selected destination for the vehicle to travel. We will use an existing aerial map for the campus and convert it to a spatial graph configuration where the vertices represent the landmarks and edges represent paths that the car should follow with some designated behaviors (such as stay on the right side of the lane or follow an edge). Graph search algorithms such as A* will be implemented as the default path planning algorithm. D* Lite will be explored to efficiently recompute the path when there are any changes to the map. CATE shall avoid any static obstacles and walking pedestrians within some safe distance. Unlike traveling along traditional roadways, CATE’s route directly coexists with pedestrians. To ensure the safety of the pedestrians, we will use sensor fusion techniques that combine data from both lidar and stereo vision for obstacle avoidance while also allowing CATE to operate along its intended route. We will also build prediction models for pedestrian traffic patterns. CATE shall improve its location and work under a GPS-denied situation. CATE relies on its GPS to give its current location, which has a precision of a few meters. We have implemented an Unscented Kalman Filter (UKF) that allows the fusion of data from multiple sensors (such as GPS, IMU, odometry) in order to increase the confidence of localization. We also noticed that GPS signals can easily get degraded or blocked on campus due to high-rise buildings or trees. UKF can also help here to generate a better state estimate. In summary, CATE will provide on-campus transportation experience that coexists with dynamic pedestrian traffic. In future work, we will extend it to multi-vehicle scenarios.

Keywords: driverless vehicle, path planning, sensor fusion, state estimate

Procedia PDF Downloads 144

Authors: Pablo Oteiza, Ricardo Alvarez, Mehrdad Pirnia, Fuat Can

Abstract:

To effectively combat climate change, many countries around the world have committed to a decarbonisation of their electricity, along with promoting a large-scale integration of renewable energy sources (RES). While this trend represents a unique opportunity to effectively combat climate change, achieving a sound and cost-efficient energy transition towards low-carbon power systems poses significant challenges for the multi-year Transmission Network Expansion Planning (TNEP) problem. The objective of the multi-year TNEP is to determine the necessary network infrastructure to supply the projected demand in a cost-efficient way, considering the evolution of the new generation mix, including the integration of RES. The rapid integration of large-scale RES increases the variability and uncertainty in the power system operation, which in turn increases short-term flexibility requirements. To meet these requirements, flexible generating technologies such as energy storage systems must be considered within the TNEP as well, along with proper models for capturing the operational challenges of future power systems. As a consequence, TNEP formulations are becoming more complex and difficult to solve, especially for its application in realistic-sized power system models. To meet these challenges, there is an increasing need for developing efficient algorithms capable of solving the TNEP problem with reasonable computational time and resources. In this regard, a promising research area is the use of artificial intelligence (AI) techniques for solving large-scale mixed-integer optimization problems, such as the TNEP. In particular, the use of AI along with mathematical optimization strategies based on decomposition has shown great potential. In this context, this paper presents an efficient algorithm for solving the multi-year TNEP problem. The algorithm combines AI techniques with Column Generation, a traditional decomposition-based mathematical optimization method. One of the challenges of using Column Generation for solving the TNEP problem is that the subproblems are of mixed-integer nature, and therefore solving them requires significant amounts of time and resources. Hence, in this proposal we solve a linearly relaxed version of the subproblems, and trained a binary classifier that determines the value of the binary variables, based on the results obtained from the linearized version. A key feature of the proposal is that we integrate the binary classifier into the optimization algorithm in such a way that the optimality of the solution can be guaranteed. The results of a study case based on the HRP 38-bus test system shows that the binary classifier has an accuracy above 97% for estimating the value of the binary variables. Since the linearly relaxed version of the subproblems can be solved with significantly less time than the integer programming counterpart, the integration of the binary classifier into the Column Generation algorithm allowed us to reduce the computational time required for solving the problem by 50%. The final version of this paper will contain a detailed description of the proposed algorithm, the AI-based binary classifier technique and its integration into the CG algorithm. To demonstrate the capabilities of the proposal, we evaluate the algorithm in case studies with different scenarios, as well as in other power system models.

Keywords: integer optimization, machine learning, mathematical decomposition, transmission planning

Procedia PDF Downloads 85