A Very Deep One Dimensional Convolutional Neural Network (VDOCNN) for Appliance Power Signature Classification

Abstract

Estimating appliance specific power consumption using a single measuring device, known as Non-Intrusive Load Monitoring (NILM), is a challenging Blind Signal source Separation (BSS) problem. For the past two decades, numerous mathematical and pattern recognition techniques, including Fractional Hidden Markov Model (FHMM), Gaussian Mixture Model (GMM) and Mean Shift Based Clustering Techniques (MSBCT) have been proposed to decompose the total power consumption of a household into appliance specific power signals. The measurement sampling rate, operating characteristic of individual appliances and an unknown number of mixed signals create a big challenge in separating them. The main challenge is to design an algorithm that can learn appliance features accurately, before applying the algorithm to disaggregate the main power signals. To address this problem, A Very Deep One dimensional Convolutional Neural Network (VDOCNN) for appliance power signature classification is proposed in this research. As a first step, we have applied VDOCNN in learning appliance features from a given set of labeled training data. VDOCNN has achieved accuracy up to 98% in detecting appliance from its power signature using a UK Domestic Appliance-Level Electricity (UK-DALE) dataset. Using this algorithm, we are working towards disaggregation of power signatures for different appliances from a single power signal in future research.