Noninvasive Power Load Disaggregation Based On Graph Signal And Hybrid Neural Network

Demand side energy management technology is the key technology of smart grid.It helps to achieve more effective utilization of power energy by reducing the energy demand of users during peak load.In the background of smart grid and smart meter,aiming at the problems of high investment cost and difficult popularization and application of traditional sensor invasive power monitoring method,non-invasive power monitoring method has become a hot spot of many scholars.In this paper,the non-invasive residential power load monitoring project is studied,and the simple and efficient active power is selected as the non-invasive load decomposition feature,which has stronger versatility,wider application range and lower cost.This paper studies how to use graph signal processing method for non-invasive power load decomposition.Through the detailed analysis of the research process of initial event detection and clustering analysis,cluster merging and pairing,as well as equipment working state and equipment power sequence matching,dynamic time warping is used in sequence matching Warping(DTW)algorithm can effectively reduce the matching failure problem caused by sequence drift when using Euclidean distance matching.The method used in this paper is demonstrated with relevant examples.The results show that it can improve the accuracy of non-invasive power load decomposition.Finally,the proposed method is compared with the hidden Markov load decomposition method in the related literature on the same data,which further verifies the effectiveness of the proposed method.Although the load decomposition method based on graph signal processing has achieved good results on low sampling rate data,the effect of this algorithm based on event detection will not be very ideal when the power change is close when the equipment is in use.Therefore,this paper studies how to use deep neural network for non-invasive power load decomposition.Combined with the idea of stack type hybrid neural network,the convolutional neural network with spatial feature extraction ability and the cyclic neural network with time sequence feature extraction ability are stacked to form a hybrid neural network model,which enhances the network's ability to extract features in load power series The extraction ability effectively solves the problem that the single form neural network model has limited ability to extract the features contained in the load power series,resulting in the low accuracy of the final load decomposition.Finally,an example is given to verify that the hybrid network model is better than the single neural network model for the non-invasive load decomposition problem.