Researchon Non-invasive Household Electric Load Decomposition Basedon Optimized Neural Network

With the development of the national economy and the advancement of science and technology,the society has higher and higher requirements for the quality of electricity.This also puts forward higher requirements for the development of smart grid,advanced measurement technology,load monitoring and other technologies.The non-intrusive load decomposition technology can mine the user’s internal information to obtain the load information of various electrical equipment,which is convenient for users to formulate power consumption plans,and makes the smart grid closer to daily life.Aiming at the problems of load factor characteristic selection,load identification accuracy and scalability in non-invasive load decomposition,this paper constructs the load characteristic set of electrical equipment before load identification,and proposes a load decomposition method based on shallow neural network model and a load decomposition method based on deep learning framework,as well as solutions to scalability problems.First of all,In this paper,by screening the load types of electrical equipment,based on the steady-state characteristic data of six typical electrical equipment,the semi-supervised nearest neighbor propagation clustering(SAP)algorithm is used to establish a data set of electrical equipment load characteristics.This is done to reduce computational complexity and provide data support for subsequent load identification and energy consumption fitting.In addition,the factor characteristics of the PI and PQ input clustering algorithms are compared,and the clustering algorithm has better performance under the combination of PI factor characteristics.Then,for the accuracy of the identification model,this paper uses the Drosophila optimized generalized regression neural network(FOA-GRNN)model.This paper uses the FOA to find the optimal smoothing parameters in the GRNN model,so that the performance of the GRNN model is optimal.By Verifying,the FOA-GRNN model proposed in this paper has a higher load identification accuracy.Combining the decomposition results of the FOA-GRNN model with the SAP clustering load characteristic data set to complete the user energy consumption fitting,it is found that the SAP-FOA-GRNN method can effectively solve the problem of non-intrusive load decomposition,and not only can correctly identify the user’s electrical equipment,Also has the ability to estimate.In addition,this article discusses the selection of factors in the input layer of the neural network model.Finally,this paper uses the deep learning framework based on the SAP clustering feature data set to establish a seq2 seq deep convolutional neural network(DCNN)model.Validation using the same data set as before,Compared with other deep learning models,although the DCNN model takes a little longer time,the load identification accuracy rate is higher;compared with the shallow neural network model(FOA-GRNN),the DCNN model has better performance,but The actual application of the deep neural network model requires higher computer hardware.In addition,a solution to the problem of scalability in non-intrusive load decomposition is proposed,After the data set scalability test,it is found that the two non-intrusive load decomposition methods proposed in this paper can solve the scalability problem,and the method based on SAP clustering and seq2 seq DCNN has stronger data mining capabilities.