Learning algorithms for non intrusive load monitoring

Abstract

Non-Intrusive Load Monitoring (NILM), also known as Energy Disaggregation, is the process of segregating a building’s total electricity consumption into its appliances by appliance consumption from the smart meter data. In developed countries, Smart meters are currently being rolled out on a large scale, mainly due to the two most important benefits of energy disaggregation: 1)helps consumers understand their energy usage by providing itemized bills; 2)helps grids with capacity planning. These benefits ultimately lead to energy saving andcost-cutting. Existing algorithms for NILM consist of a training phase in which sub-metered appliance data is used to build models of the appliances. In the test phase, these models are used to disaggregate the total electrical energy consumption. To collect the appliance-level data, we need to put sensors on appliances present in the building. This makes the training phase intrusive. Due to this, such methods do not provide a scalable solution for NILM. This thesis has three main objectives: 1)To propose more accurate algorithms for NILM than state of the art; 2)To propose the methods which make the training phase completely non-intrusive, i.e., to dodge the requirement of the submetered data ; 3)To propose an algorithm that can work with compressed energy signals in order to save the bandwidth and avoid network congestion. First, we propose a dictionary-learning based algorithm called Deep Sparse Coding for NILM. The usual technique is to learn a dictionary for every device and use the learned dictionaries as a basis for blind source separation during disaggregation. Prior studies in this area are shallow learning techniques, i.e., they learn a single layer of dictionary for every device. In this work, we learn multiple layers of dictionaries for each device. These multi-level dictionaries are used as a basis for source separation during disaggregation. We show that this algorithm outperforms the benchmark techniques like Factorial Hidden Markov Model and Discriminative Disaggregating Sparse Coding. Second, we follow the multi-label classification based paradigm for NILM and determine the state(On/Off) of the appliances present in the building. For this, we propose several algorithms that adapted Transform Learning, Sparse Representation Classifier, Restricted Boltzmann Machine and Long Short Term Memory to perform multi-label classification and subsequently disaggregating the appliance-level load. Third, we propose a compressed sampling(CS) approach. The high-frequency power signal from a smart meter is encoded (by a random matrix) to very few samples making the signal suitable for WAN transmission without choking network bandwidth. CS guarantees the recovery of the high-frequency signal from the few transmitted samples under certain conditions. This work shows how to recover the signal and simultaneously disaggregate it. The motive of the work presented in this thesis is to propose NILM algorithms independent of the sub-metered data and make advancements in state-of-the-art in the field of energy disaggregation.