Deep Learning-based Estimation Of The Target Appliance's Electricity Consumption

In smart grid,measurement of the electricity consumption of major appliances in different time segments is of crucial significance to demand-side management and energy conservation.Non-intrusive load monitoring(NILM)can infer the target appliances' power use information by only collecting and analyzing the current and voltage value at the single power entrance point.Inspired by the success of deep learning(DNN)in other fields,some researchers have applied it to NILM with promising results.For improving the accuracy and practicability(reduce requirement of priori knowledge)of estimation of the target appliance's electricity consumption,this paper is focused on the input lost and inadequacy of training data and establishes a method only using synthetic data for training neural network.The innovation is as follow: 1)A neural network method with extended input is proposed.This method extends the input window length on both sides and hence supplements the lost information;2)A background load-based synthetic method is proposed,which generates abundant and reasonable training data;3)A background filtering is designed to accurately extract the background load required by the synthetic method in 2)without time synchronized submetering.On basis of improving the accuracy of estimation of electricity consumption,the proposed entire method only needs unlabeled real aggregate data and several operation curves of the target appliance,where the difficulty of training data acquisition is reduced.According to comparison experiments,the proposed method performs better than current DNN-based method and HMM-based method.However,background filtering only fits for the appliances that the working duration is much shorter than the idle duration such as kettle and washing machine.