Research On Resident User Clustering And Short-Term Load Forecasting In Smart Grid

With the continuous improvement of the demand for electric energy in various fields of society,the rapid development of smart grid technology and the comprehensive implementation of national renewable resource policy,the power system is facing the transformation to be more intelligent,more flexible and more interactive.The number of users participating in information interaction in the process of power system transformation is gradually increasing.However,due to the different types of energy users,the electricity load presents the characteristics of volatility and scheduling,which has a certain impact on the stable operation of power grid.The ability to accurately predict short-term electricity loads in a region plays an important role in power grid planning and operation.This paper presents a cluster-based LSTM short-term load forecasting method.The research work includes the following aspects:1.Aiming at the problem of determining the optimal cluster number k and selecting the initial cluster centers for K-Means algorithm,an improved K-Means algorithm is presented.The algorithm uses the PCA algorithm to reduce the feature space of the sample set to the visual three-dimensional space,and determines the optimal cluster number of the K-Means clustering algorithm.By defining the density-distance weight,the sample points corresponding to the larger k weights are sequentially selected in the data set according to the weight value as the initial cluster centers.Finally,based on the multi-dimensional electricity feature set of each household in the smart grid,the improved K-Means algorithm is used to determine the number of resident categories and complete the clustering analysis of resident users.The simulation results show that the improved algorithm can accurately select the number of resident categories,and has higher clustering accuracy and stronger anti-noise performance.2.Based on the results of resident user clustering analysis,an LSTM short-term electricity load forecasting method is presented.Firstly,the residence community electricity load data is used to determine each hyperparameter of the LSTM network.Then,different LSTM forecasting models are constructed according to the categories of residents.Finally,all resident user prediction results are summarized as the total electricity load forecast value of the community.The simulation results show that the LSTM neural network has the highest accuracy for electricity data forecasting compared to other neural network models.At the same time,the accuracy of the LSTM load forecasting model combined with K-Means clustering analysis is higher than that of the non-resident user clustering model.3.A JavaWeb visualization system is built up for the clustering results of resident users and short-term electricity load forecasting results.The system uses SSH architecture to design each module function,and draws visual images through JFreeChart library,which realizes the function of visualizing the electricity data analysis results,which is convenient for power managers to view and operate.