Spatial Load Forecasting Method Based On Multi-level Clustering And Improved Elman Neural Network

Spatial load forecasting(SLF) needs to divide the area to be predicted into power supply block at different levels (also called cellular). It is not only to predict the future total load of this area,but also predict the number of users,the time of the load and load distribution in each power supply block. As the premise and foundation of distribution network planning,whether the prediction results are accurate will directly affect the development and construction of power system in the future.This paper summarizes the commonly used methods of spatial load forecasting at home and abroad. Then the classification and spatial distribution characteristics of the load are summarized. At the same time, the effect of spatial resolution on the accuracy of load forecasting is analyzed. According to the load characteristics and change rules,this paper proposes a method of spatial load forecasting based on multilevel clustering analysis(MCA) and improved Elman neural network in order to meet the demand of modern distribution network planning.Firstly, distribution network block is divided to generate cellular. The partition of distribution network block is the key point of the spatial load forecasting,which is directly related to the layout planning of power grid in the future. In this paper, a new method of distribution network block partitioning is proposed based on improved K-mean clustering algorithm. According to the specific location of the transformer substation, the area to be predicted is divided into the irregular distribution blocks.And each block corresponds to a first-level cellular. The construction land corresponding to the first-level cellular is subdivided according to its function,generating the second-level cellulars. Finally,the multilevel cluster analysis of SLF is completed.The core of spatial load forecasting method is to forecast the load value of each cellular based on the partition of distribution network block. In this paper, a spatial load forecasting method based on improved Log-Elman neural network is proposed.As a feedback neural network algorithm,Elman neural network has the advantages of feed-forward neural network algorithm, such as BP. And it has broad prospects in the field of SLF. In this paper, the Elman neural network is improved, and the Logistic algorithm is applied to the Elman neural network. It can adapt to the change rules of the load in different blocks,and improve the flexibility and accuracy of spatial load forecasting.The proposed spatial load forecasting based on multilevel clustering algorithmm and Log-Elman neural network algorith is applied to load forecasting in a city of Shandong Province, and the prediction results show that the proposed method has the advantages of accuracy, flexibility and practicability.