6 research outputs found

    Evaluating neural load disaggregation for ambient assisted living applications

    No full text
    Non-intrusive Load Monitoring (NILM), also known as Load Disaggregation, is the problem of identifying the power consumption related to individual appliances through only a single metering point. It gained significant attention during the past decade, notably with deep neural networks. These models offer a promising alternative to traditional activity monitoring of older adults living alone, relying on more intrusive sensing technologies. Yet, this adoption faces several challenges. First, load disaggregation techniques are perceived as privacy-violating technologies since they can reveal sensitive information, notably with data sampled at the high frequency required by deep models. Second, the comparability problem in NILM scholarship is still a research gap. Little is known about state-of-the-art performance and models suitable for each appliance due to heterogeneous evaluation setups. Third, despite the fact that a handful set of frameworks for activity monitoring based on NILM have already been proposed in the literature, none of them evaluated the potential errors propagated by the disaggregation approaches.The current thesis addresses these issues. First, an open-source energy measurement system is suggested leveraging open-source technologies to enable local processing of the data with a NILM module. The system is evaluated on two real energy datasets and one synthetic energy dataset. Second, the author addresses the problem of comparability in NILM through the adoption of machine learning best practices implemented and made available to the research community through an open-source python package. Furthermore, the author develops an open-source NILM benchmark repository containing both baselines and recent models to facilitate comparability in future work. A simulation study of federated learning for local processing and privacy-preserving NILM is furtehr presented, highlighting promising results and potential future research directions. Finally, a novel activity monitoring framework is presented and evaluated on a synthetic dataset quantifying the error propagated by a NILM algorithm.Hafsa BousbiatDissertation Universität Klagenfurt 202

    Evaluating neural load disaggregation for ambient assisted living applications

    No full text
    Non-intrusive Load Monitoring (NILM), also known as Load Disaggregation, is the problem of identifying the power consumption related to individual appliances through only a single metering point. It gained significant attention during the past decade, notably with deep neural networks. These models offer a promising alternative to traditional activity monitoring of older adults living alone, relying on more intrusive sensing technologies. Yet, this adoption faces several challenges. First, load disaggregation techniques are perceived as privacy-violating technologies since they can reveal sensitive information, notably with data sampled at the high frequency required by deep models. Second, the comparability problem in NILM scholarship is still a research gap. Little is known about state-of-the-art performance and models suitable for each appliance due to heterogeneous evaluation setups. Third, despite the fact that a handful set of frameworks for activity monitoring based on NILM have already been proposed in the literature, none of them evaluated the potential errors propagated by the disaggregation approaches.The current thesis addresses these issues. First, an open-source energy measurement system is suggested leveraging open-source technologies to enable local processing of the data with a NILM module. The system is evaluated on two real energy datasets and one synthetic energy dataset. Second, the author addresses the problem of comparability in NILM through the adoption of machine learning best practices implemented and made available to the research community through an open-source python package. Furthermore, the author develops an open-source NILM benchmark repository containing both baselines and recent models to facilitate comparability in future work. A simulation study of federated learning for local processing and privacy-preserving NILM is furtehr presented, highlighting promising results and potential future research directions. Finally, a novel activity monitoring framework is presented and evaluated on a synthetic dataset quantifying the error propagated by a NILM algorithm.Hafsa BousbiatDissertation Universität Klagenfurt 202

    Detecteurs pyroelectriques sur polyvinylidene bifluore et copolymere, integres sur silicium: approche technologique de capteurs unitaires et matrices bidimensionnelles infrarouges

    No full text
    SIGLEAvailable from INIST (FR), Document Supply Service, under shelf-number : T 79422 / INIST-CNRS - Institut de l'Information Scientifique et TechniqueFRFranc

    Abstracts of 1st International Conference on Computational & Applied Physics

    Get PDF
    This book contains the abstracts of the papers presented at the International Conference on Computational & Applied Physics (ICCAP’2021) Organized by the Surfaces, Interfaces and Thin Films Laboratory (LASICOM), Department of Physics, Faculty of Science, University Saad Dahleb Blida 1, Algeria, held on 26–28 September 2021. The Conference had a variety of Plenary Lectures, Oral sessions, and E-Poster Presentations. Conference Title: 1st International Conference on Computational & Applied PhysicsConference Acronym: ICCAP’2021Conference Date: 26–28 September 2021Conference Location: Online (Virtual Conference)Conference Organizer: Surfaces, Interfaces, and Thin Films Laboratory (LASICOM), Department of Physics, Faculty of Science, University Saad Dahleb Blida 1, Algeria
    corecore