Short-Term Power Load Prediction Based on Cluster Analysis and Temporal Convolutional Networks of Attention Mechanism

Abstract

Short-term power load prediction has become one of the important contents of smart grid management. Accurate power load prediction can provide a safer, more reliable, and more efficient direction for power system operation. This article proposes a short-term power load forecasting method. Mainly based on the improved fuzzy c-means clustering (FCM) algorithm and a temporal convolutional network (TCN) model combined with an attention mechanism (AM). First, to cluster the load data with the same power consumption behavior into one class, a kernel FCM algorithm based on particle swarm optimization is used. Meanwhile, external factors with high correlation are selected as inputs. The Pearson correlation coefficient can be used to measure the degree of correlation between load data and external factors. Second, by analyzing the degree of correlation between external influencing factors and load data, the channel AM and time AM are introduced into the TCN model. Finally, the effectiveness of the proposed method was verified through a real electricity load dataset. The experimental results indicate that this method can accurately predict future changes in power load. Compared with other models, it also has high accuracy and generalization ability.