Research On Non-intrusive Residential Household Electricity Load Monitoring Technology

Non-intrusive load monitoring(NILM)is a technology that monitors and identifies power-using devices in the electrical system in real time through a non-intrusive approach.This technology acquires voltage and current signals by installing sensors at the residential home’s electrical inlet.And through the analysis and processing of the signals,it identifies the characteristic signals of different residential home electric devices,so as to realize the real-time monitoring and analysis of residential home electric devices.Based on deep learning technology,this paper conducts research on load decomposition algorithm for different application scenarios of residential household electricity load,mainly including two application scenarios of load decomposition under the influence of noise interference and insufficient accumulation of household historical electricity consumption data.In the scenario of low accuracy of load decomposition under noise interference,a NILM method based on attention mechanism is proposed.The coding-decoding structure is introduced,and the residential electricity consumption load information,which plays a key influence on the decoder,is highlighted by introducing the SE attention mechanism to calculate the most important information in the sequence at the current moment before the decoder.5% Gaussian white noise is added to the test data as experimental data to simulate the effect of noise interference on load decomposition,and experiments are carried out to verify the effectiveness of this method.To address the problem of poor effect of NILM method on load decomposition of unknown households,the effect of NILM method based on attention mechanism on load decomposition of unknown households is verified in the test of non-participating training households.The experiments show that this method also has good load decomposition effect for unknown households.To address the problem that the load data of residential households may be partially missing,data filling experiments are conducted by simulating a 1%-10% missing data rate,and the mean filling and linear interpolation methods are used to fill in the data for different cases.The experiment proves that the data filled by this method basically does not affect the accuracy of load decomposition and can still better reflect the real time-series characteristics of household energy loads.In the scenario where the accumulated historical electricity consumption data of some households are insufficient,a model migration-based NILM method is proposed.The introduction of an embedding layer with time vectors(Time2vec)as an encoder converts the time series into a continuous vector space,which improves the computational efficiency of the model for power time series information.Then,by introducing the channel attention mechanism,the favorable residential electricity load characteristics can be effectively obtained.After that,the model migration method is used to fine-tune the pre-trained model based on the complete data so as to overcome the problem of insufficient accumulation of historical electricity consumption data for some households.The REDD,UK-DALE and REFIT datasets are used to conduct experiments for different application scenarios.The experimental results show that the proposed method can significantly shorten the training time and improve the accuracy of load decomposition under the application scenarios of simulated noise interference load decomposition and insufficient accumulation of some household historical electricity consumption.