Short-term LOAD Forecasting Method of TPA-LSTNet Model Based on Time Series Clustering

Abstract

To provide a stronger guarantee for the power system's stable operation, improving the accuracy of short-term load peak prediction is necessary. This paper proposes a short-term load prediction model TPA-LSTNet that combines TPA (Temporal Pattern Attention) and LSTNet and combines the K-Shape time series clustering method. Firstly, collect external information on the corresponding date of the data, such as daily temperature, humidity, wind direction, whether it is a holiday, Etc. Secondly, using the characteristics of high precision and high efficiency of the K-Shape algorithm, cluster analysis is carried out on the electricity load data in the station area. Then combine the data with external information and input it into the TPA-LSTNet model to extract time series features and train the model. Finally, the prediction of short-term power load is realized using the trained model. The predicted results on an existing urban distribution network verify the prediction accuracy of the method.