Machine Learning Approaches to Non-Intrusive Load Monitoring
Samenvatting
Research on Smart Grids has recently focused on the energy monitoring issue, with the objective of maximizing the user consumption awareness in building contexts on the one hand, and providing utilities with a detailed description of customer habits on the other. In particular, Non-Intrusive Load Monitoring (NILM), the subject of this book, represents one of the hottest topics in Smart Grid applications. NILM refers to those techniques aimed at decomposing the consumption-aggregated data acquired at a single point of measurement into the diverse consumption profiles of appliances operating in the electrical system under study.
This book provides a status report on the most promising NILM methods, with an overview of the publically available dataset on which the algorithm and experiments are based. Of the proposed methods, those based on the Hidden Markov Model (HMM) and the Deep Neural Network (DNN) are the best performing and most interesting from the future improvement point of view. One method from each category has been selected and the performance improvements achieved are described. Comparisons are made between the two reference techniques, and pros and cons are considered. In addition, performance improvements can be achieved when the reactive power component is exploited in addition to the active power consumption trace.
Specificaties
Inhoudsopgave
<p>2 Non-Intrusive Load Monitoring</p>
<p> 2.1 Problem statement</p>
2.2 State of the Art<p></p>
<p> 2.3 Datasets</p>
<p> 2.4 Evaluation metrics</p>
<p> 2.5 Remarks</p>
<p>3 Background</p>
<p> 3.1 Hidden Markov Model (HMM)</p>
<p> 3.1.1 Baum-Welch algorithm</p>
<p> 3.1.2 Factorial HMM</p>
3.2 Deep Neural Network (DNN)<p></p>
<p> 3.2.1 Stochastic gradient descent (SGD)</p>
<p> 3.2.2 Autoencoder</p>
<p>4 HMM based approach</p>
<p> 4.1 Additive Factorial Approximate Maximum A-Posteriori (AFAMAP)</p>
4.1.1 Appliance modelling<p></p>
<p> 4.1.2 Rest-of-the-World model</p>
4.2 Algorithm improvements<p></p>
<p> 4.2.1 Experimental setup</p>
<p> 4.2.2 Results</p>
<p> 4.3 Exploitation of the reactive power</p>
<p> 4.3.1 AFAMAP formulation</p>
<p> 4.3.2 Experimental setup</p>
<p> 4.3.3 Results</p>
4.4 Footprint extraction procedure<p></p>
<p> 4.4.1 Experimental setup</p>
<p> 4.4.2 Results</p>
<p>5 DNN based approach</p>
<p> 5.1 Neural NILM</p>
<p> 5.2 Denoising AutoEncoder approach</p>
<p> 5.3 Algorithm improvements</p>
<p> 5.3.1 Experimental setup</p>
5.3.2 Results<p></p>
<p> 5.4 Exploitation of the reactive power</p>
<p> 5.4.1 Experimental setup</p>
<p> 5.4.2 Results</p>
<p>6 Conclusions</p>
6.1 Future Research Topics<p></p>
<p>7 References</p>