1,720,964 research outputs found
Deep Learning Techniques for Edge-Centric Non-Intrusive Load Monitoring
No full textIl cambiamento climatico è una delle sfide più significative affrontate in questo secolo. Per limitare l’aumento della temperatura media globale al di sotto di 1.5 °C, è fondamentale ridurre l’uso di energia elettrica ed è vitale promuovere l’efficienza energetica. I dispositivi elettronici dovrebbero essere utilizzati in modo efficiente per evitare sprechi di energia. Nell’ambito residenziale, gli utenti possono contribuire significativamente al risparmio energetico, soprattutto se consapevoli dei loro consumi. La costante disponibilità di dati di consumo energetico ha portato allo sviluppo di tecniche avanzate per monitorare i carichi all’interno degli edifici e fornire agli utenti residenziali una maggiore consapevolezza delle loro abitudini di consumo elettrico. Il monitoraggio non intrusivo del carico stima lo stato e il consumo energetico dei singoli elettrodomestici
nell’edificio, basandosi solo sulle letture aggregate del contatore. Oggigiorno, la maggior parte degli approcci proposti in letteratura si basa sul Deep Learning, poiché si è dimostrato superiore ad altri metodi inizialmente sviluppati. Tuttavia, devono essere ancora affrontati aspetti legati all’applicabilità nel mondo reale. In primo luogo, c’è il problema della disponibilità di dati annotati per addestrare approcci supervisionati. Chi fornisce le annotazioni spesso coincide con l’utente finale e il processo di annotazione può essere lungo, scomodo e soggetto ad errori. In secondo luogo, l’inferenza viene solitamente eseguita nel cloud su macchine ad alte risorse, quindi lontano da dove vengono acquisiti i
segnali. Questo richiede la trasmissione dei dati e può comportare ritardi del servizio e problemi di privacy. Per mitigare le suddette problematiche, questa tesi propone metodologie basate sui paradigmi dello Human-centred Computing e dell’ Edge Computing. Le strategie sviluppate mirano a ridurre lo sforzo richiesto all’utente per fornire le annotazioni mantenendo stabili o migliorando le prestazioni. Inoltre, i metodi mirano a diminuire la complessità strutturale e computazionale delle architetture utilizzate. Gli approcci sono stati sviluppati e valutati su dataset pubblici, dimostrando la loro superiorità rispetto allo stato dell’arte. Le prestazioni finali risultano superiori utilizzando meno dati, annotazioni più deboli e strutture di rete più semplici che risultano adatte per l’installazione su dispositivi a basse risorse. Ricerche future considerano di addestrare le reti neurali localmente, per promuovere l’adattabilità e l’affidabilità
del monitoraggio. Inoltre, possono essere sviluppate strategie di monitoraggio ibride e integrate con sistemi di gestione dell’energia.Climate change is one of the most significant challenges faced in this century. To limit the rise in global average temperatures below 1.5\,°C, it is crucial to decrease the electrical energy usage. Therefore, it is vital to promote energy efficiency through sustainable practices. Devices should be used efficiently to avoid energy waste. In the residential setting, individuals can significantly contribute to energy saving, especially if they are aware of their consumption. The constant availability of energy consumption profiles has led to the development of advanced techniques to monitor loads inside buildings and provide residential users with improved awareness of their energy consumption and usage habits. One such technique is Non-Intrusive Load Monitoring (NILM), which detects the states of appliances and estimates the power consumption of individual loads in the building based only on the building’s aggregate meter readings. Nowadays, most of the approaches proposed in the literature are based on deep learning, which has proven superior to other methods. Nonetheless, they still have to deal with aspects related to real-world applicability. Firstly, there is the issue of the availability of labeled datasets. Labels should be provided by annotators, who are often end-user, but this process is time-consuming and prone to errors. Second, computation is usually done in the cloud, which is far from where the data are acquired. This requires data transmission and can result in latency in the service output.
To mitigate the above issues, this thesis proposes several methodologies that follow the Human-Centred Computing and the Edge Computing paradigms. As a consequence, the developed strategies aim to lighten the effort requested to the user for providing labels while enhancing the performance. At the same time, the computation is lightened by reducing algorithms complexity while maintaining performance.
The methods have been developed and evaluated on publicly available datasets, demonstrating their superiority compared to benchmark strategies. Moreover, the final performance is increased, even with less data and simpler structures.
Future directions considers to train networks locally to promote adaptability and reliability. Additionally, hybrid monitoring strategies can be investigated and integrated with energy management systems or demand-response programs based on the user requirements
Knowledge Distillation for Scalable Nonintrusive Load Monitoring
Get PDFSmart meters allow the grid to interface with individual buildings and extract detailed consumption information using nonintrusive load monitoring (NILM) algorithms applied to the acquired data. Deep neural networks, which represent the state of the art for NILM, are affected by scalability issues since they require high computational and memory resources, and by reduced performance when training and target domains mismatched. This article proposes a knowledge distillation approach for NILM, in particular for multilabel appliance classification, to reduce model complexity and improve generalization on unseen data domains. The approach uses weak supervision to reduce labeling effort, which is useful in practical scenarios. Experiments, conducted on U.K.-DALE and REFIT datasets, demonstrated that a low-complexity network can be obtained for deployment on edge devices while maintaining high performance on unseen data domains. The proposed approach outperformed benchmark methods in unseen target domains achieving a F1-score 0.14 higher than a benchmark model 78 times more complex
Appliance-Incremental Learning for Non-Intrusive Load Monitoring
No full textNon-Intrusive Load Monitoring (NILM) is an algorithmic-based approach for monitoring appliances inside a building using a single metering point, enabling better traceability of energy use within the grid. Most state-of-the-art approaches, that are based on Deep Neural Networks (DNN), are “static” and present serious limitations in handling changes in appliance usage patterns or the addition of new appliances. This limits the adaptability of NILM in practical scenarios. To address this, we propose an Appliance-Incremental Learning (AIL) method that can continuously adapt a DNN-based NILM algorithm to monitor new appliances without introducing new networks. AIL utilizes a multi-label classification framework, sharing the same network structure for all appliances. We mitigate catastrophic forgetting by distilling knowledge of previous tasks and dynamically selecting network parameters better related to new tasks. This allows us to efficiently train the network on new tasks while maintaining stable performance on previous appliances. We compare our method with the static NILM approach and the Learning without Forgetting (LwF) method in a real-world scenario, where a pre-trained model is adapted to new environments post-deployment. The results demonstrate the effectiveness of our method in handling new appliances, achieving improved adaptability and performance with respect to comparative methods. The proposed AIL method offers a reliable solution for NILM, enabling the monitoring of new appliances and accommodating variations in usage patterns
Weakly Supervised Transfer Learning for Multi-label Appliance Classification
No full textNon-intrusive Load Monitoring refers to the techniques for providing detailed information on appliances’ states or their energy consumption by measuring only aggregate electrical parameters. Supervised deep neural networks have reached the state-of-the-art in this task, and to improve the performance when training and test data domains differ, transfer learning techniques have been successfully applied. However, these techniques rely on data labeled sample-by-sample (strong labels) to be effective, which can be particularly costly in transfer learning since it requires collecting and annotating data in the target domain. To mitigate this issue, this work proposes a cross-domain transfer learning approach based on weak supervision and Convolutional Recurrent Neural Networks for multi-label appliance classification. The proposed method is based on the concept of inexact supervision by modeling NILM as a Multiple Instance Learning problem, exploiting different and less costly annotations called weak labels. The learning strategy is able to exploit weak labels both for pre-training and fine-tuning the models.UK-DALE and REFIT are used in the experiments as source and target domain datasets to train, fine-tune, and evaluate the networks. The results demonstrate the effectiveness of the proposed method compared to the related pre-trained models. In particular, when the model is pre-trained on strongly and weakly labeled data of UK-DALE and then fine-tuned only on REFIT weak labels, the performance improves by 20.3%
Improving knowledge distillation for non-intrusive load monitoring through explainability guided learning
Get PDFKnowledge distillation (KD) is a machine learning technique widely used in recent years for the task of domain adaptation and complexity reduction. It relies on a Student-Teacher mechanism to transfer the knowledge of a large and complex Teacher network into a smaller Student model. Given the inherent complexity of large Deep Neural Network (DNN) models, and the need for deployment on edge devices with limited resources, complexity reduction techniques have become a hot topic in the Non-intrusive Load Monitoring (NILM) community. Recent literature in NILM has devoted increased effort to domain adaptation and architecture reduction via KD. However, the mechanism behind the transfer of knowledge from the Teacher to the Student is not clearly understood. In this work, we aim to address the aforementioned issue by placing the KD NILM approach in a framework of explainable AI (XAI). We identify the main inconsistency in the transfer of explainable knowledge, and exploit this information to propose a method for improvement of KD through explainability guided learning. We evaluate our approach on a variety of appliances and domain adaptation scenarios and demonstrate that solving inconsistencies in the transfer of explainable knowledge can lead to improvement in predictive performance
Interpretability and reliability-driven knowledge distillation for non-intrusive load monitoring on the edge
Get PDFThe deployment of deep neural networks (DNNs) on resource-constrained edge devices necessitates efficient, low-complexity algorithms. Knowledge distillation (KD) addresses this through a student-teacher paradigm, transferring knowledge from complex teacher models to simpler student models. Current KD methods often optimize student performance without adequately addressing the reliability and interpretability of transferred knowledge, thus presenting challenges in maintaining both robustness and decision transparency. This paper introduces an Interpretability and Reliability-driven Knowledge Distillation (IR-KD) framework that enhances teacher model interpretability through perception-aligned gradients while leveraging hidden information from weak labels to optimize knowledge transfer. Our approach ensures compressed models remain computationally efficient while improving interpretability, which is essential for trustworthy edge AI deployment. We demonstrate improved predictive performance and model interpretability in non-intrusive load monitoring (NILM) applications as a case study. Quantitative explainability metrics confirm that perception-aligned gradients provide more faithful explanations, validating our approach's effectiveness in developing reliable and transparent edge AI systems
A Deep Cascade Framework for Non-Intrusive Power Disaggregation in Solar-Powered Households
No full textInverter-Based Resources are commonly installed behind the customer meters. Thus, non-intrusive power monitoring systems must handle power signals of different natures measured at the main meter and estimate power generation to ensure the observability of the power grid. This paper proposes a non-intrusive disaggregation approach that includes photovoltaic power production with load monitoring. The approach is based on an innovative cascade learning framework that exploits the solar power estimate to simplify the load monitoring task, thereby improving the overall disaggregation performance. Compared to five state-of-the-art models, our method achieves the lowest disaggregation error on two different real-world public datasets, with improvements of 20.86% and 8.67% over the runner-up benchmark. The code to reproduce the method is available on GitHub1
Going Beyond Counting First Authors in Author Co-citation Analysis
Get PDFThe present study examines one of the fundamental aspects of author co-citation analysis (ACA) - the way co-citation
counts are defined. Co-citation counting provides the data on which all subsequent statistical analyses and mappings
are based, and we compare ACA results based on two different types of co-citation counting - the traditional type that
only counts the first one among a cited work's authors on the one hand and a non-traditional type that takes into
account the first 5 authors of a cited work on the other hand. Results indicate that the picture produced through this non-traditional author co-citation counting contains more coherent author groups and is therefore considerably clearer. However, this picture represents fewer specialties in the research field being studied than that produced through the traditional first-author co-citation counting when the same number of top-ranked authors is selected and analyzed. Reasons for these effects are discussed
Variations on the Author
Get PDF“Variations on the Author” discusses two of Eduardo Coutinho’s recent films (Um Dia na Vida, from 2010, and Últimas Conversas, posthumously released in 2015) and their contribution to the general question of documentary authorship. The director’s filmography is characterized by a consistent yet self-effacing form of authorial self-inscription: Coutinho often features as an interviewer that rather than express opinions propels discourses; an interviewer that is good at listening. This mode of self-inscription characterizes him as an author who is not expressive but who is nonetheless markedly present on the screen. In Um Dia na Vida, however, Coutinho is completely absent form the image, while Últimas Conversas, on the contrary, includes a confessional prologue that moves the director from the margins to the center of his films. This article examines the ways in which these works stand out in the filmography of a director who offers new insights into the notion of cinematic authorship
- …
