Research On Non-intrusive Load Monitoring And Identification

The development of smart grid,smart home and advanced measurement technology provides a good platform for the application of load monitoring.The research on load monitoring and decomposition has broad prospects.Load monitoring data contains valuable user energy information.After mining and analyzing and refining the power information into internal,it has high application value in two-way interaction,demand-side management,power network construction,and optimal management of user load in smart grid.It can be used to formulate business plans for power companies.Provides auxiliary reference for users to master their own power consumption and actively participate in grid interaction.The main process of non-invasive load monitoring is divided into three steps:event detection,feature extraction and load identification.This paper has made some innovations in the monitoring process and achieved good res ults.First,the technical points of several key processes for non-invasive load decomposition are studied in detail.Several commonly used event detection methods,feature extraction methods and load decomposition algorithms are analyzed,and the core part of the technology is discussed in detail,which paves the way for the subsequent method research of each process.Secondly,an innovative transient event detection algorithm based on standard deviation multiple is presented.The algorithm detects the prese nce of a change point by a multiple of the active power deviation from the mean relative to the standard deviation,and accurately locates the change point by deviating from the last time the multiple crosses zero.Compared with the sliding window based mut ation detection algorithm,this method does not need to reset the detection threshold according to the load every time a load is detected,and it can detect multiple mutations for different characteristic loads.Compared with the detection algorithm based o n sliding window,the detection speed and accuracy are greatly improved.Then,the harmonic characteristics,V-I track characteristics and power characteristics of the events are extracted separately for the identification of the neural network.The harmonic inclusion rate HRI and distortion rate THD,grid graphical characteristics of V-I track images,average active power and reactive power in a stable period are extracted from the high frequency sample data of the device.This feature extraction process enables the extracted features to better adapt to the training of convolution and BP networks.Finally,a neural network load identification method based on triple features is presented.The convolution neural network and BP network are used to fuse the extracted image features and numerical features to form a composite feature,which is used as a new feature,and then the BP network is trained to achieve non-invasive load identification.The algorithm has good recognition performance for different home appliances,and its accuracy has obvious advantages compared with different feature combinations and different classification algorithms.