Research On Non-intrusive Load Disaggregation Of Residential Electricity Consumption Based On Deep Learning

Residential electricity non-intrusive load disaggregation is a technology that disaggregate the user’s total load information into the energy consumption information of each electrical appliance.The obtained energy consumption information can be used to analyze the user’s electricity consumption behavior and guide the user to manag e energy consumption rationally,thereby saving electricity expenses and reducing unnecessary power consumption.At the same time,based on the energy consumption information of electrical appliances,it is possible to further carry out refined power distribution work,optimize demand side management,effectively enhance the interactive response capability of the power grid and users,and promote the construction of smart grids.At present,the research of non-intrusive load disaggregation has achieved some results,but most of the methods still have some problems,such as low efficiency of model training,difficult to popularize the model and insufficient disaggregation accuracy of complex power consumption period.In view of the above problems,this paper uses deep learning method to deeply study the non-intrusive load disaggregation of residential electricity:1.Aiming at the problems of low model training efficiency and insufficient load disaggregation accuracy,a non-intrusive load disaggregation method based on temporal convolutional neural network is proposed.Firstly,the basic process of non-intrusive load disaggregation is given by combination with the deep learning method;secondly,the load disaggregation model is constructed by using the temporal convolutional neural network by analyzing the changes in the operating power of electrical appliances;and then using the optimized sliding window method to generate training data and normalize it;finally use the generated training data to train the load disaggregation model,and save the trained model for load disaggregation.The analysis of the calculation example shows that the proposed method improves the training efficiency of the model and further improves the load disaggregation accuracy of the complex power consumption period.2.Aiming at the promotion problem caused by insufficient model generalization ability,a load disaggregation method based on domain adversarial adaptive learning is proposed.Firstly,analyze the essence of the promotion problem,Transform it into a data scarcity problem in the target field,and combine the transfer learning method to define the concept of source domain family,target domain family,and cross-family load disaggregation;then,use the sliding window method to generate the training data of the source domain family and the target domain family,and normalize them;secondly,construct an adversarial domain adaptive load disaggregation model composed of feature extractors,predictors,and domain discriminators;thirdly,use the generated training data to train the disaggregation model with the method of adversarial domain adaptation;finally,save the trained model,and combine the feature extractor and predictor to disaggregate the household load in the target domain.The analysis of the calculation example shows that the proposed method improves the load disaggregation accuracy of the households in the target domain.