Short-Term Load Forecasting Of Resident User Cluster Based On ASTSGCN

With the extensive deployment and use of advanced measuring devices such as smart electricity meters in the user side,massive multi-source heterogeneous residential user data can be collected and stored,which provides a good data basis for user-level load prediction.Accurate residential user cluster load forecasting is an important basis for promoting demand side management of intelligent distribution network and assisting power grid companies to achieve peak cutting and valley filling.The shortterm load forecasting method of resident user cluster is studied in this paper.The main contents are as follows:(1)Most of the existing user-level load forecasting methods use the time correlation of historical load series to build data-driven models,but ignore the potential spatial correlation between neighboring users’ power consumption behaviors,In this paper,an ultra-short term load forecasting method for resident user cluster based on Kmeans clustering and adaptive spatiotemporal synchronous graph convolution neural network is proposed.Firstly,K-means clustering was used to divide residential user clusters into different groups according to the similarity of electricity consumption behavior.Then,based on the number of groups in the resident user cluster,the historical load data of each group and the correlation between the load series of each group,the spatial and temporal graph data of the resident user cluster load prediction is constructed.Finally,the adaptive spatiotemporal synchronous graph convolution neural network is used to predict the short-term load of resident user cluster.(2)Aiming at the problem that the adaptive spatio-temporal synchronous graph convolutional neural network cannot integrate the influence of meteorological and other factors on short-term load prediction in the process of load prediction,this paper improved the model of the adaptive spatio-temporal synchronous graph convolutional neural network and proposed a new conditional adaptive spatio-temporal synchronous graph convolutional neural network.Thus,a day-ahead short-term load forecasting method for resident user clusters is proposed based on K-means clustering and conditional adaptive spatio-temporal synchronization graph convolution neural network.Firstly,the above method is used to construct the spatio-temporal graph data for the cluster load prediction of resident users,and then the spatio-temporal graph data of meteorological factors and other factors are constructed on the basis of the graph structure.Finally,the conditional adaptive spatio-temporal synchronization graph convolution neural network is used to realize the day-ahead short-term load prediction of the cluster of resident users.In this paper,the accuracy and validity of the proposed forecasting method are tested and verified by real data sets.The experimental results show that the proposed user grouping,graph data construction and improvement methods are highly effective,and compared with each benchmark forecasting method,the proposed method can fully exploit and utilize the temporal and spatial correlation between different residential users’ power loads.The improvement of the forecasting model can excavate the complex correlation between meteorological factors and load forecasting,so as to further improve the precision of short-term load forecasting of residential user cluster.