Short-term load forecasting based on temporal convolutional network and prophet algorithm

Abstract

With the deepening of electrification of load side, the diversity of comprehensive load weakens its variation law and increases the difficulty of prediction. It is urgent to improve the short-term load forecasting accuracy to deal with the power balance problem of new power system. This paper aims to study the application of artificial intelligence methods in load forecasting and the problem of improving the forecasting effect. First, on one hand, considering that the load data has obvious temporal characteristics, the TCN (temporal convolutional network) model is used to obtain a good overall forecasting effect. On the other hand, based on the obvious periodicity of load data, the Prophet algorithm is adopted to obtain the prediction results with significant local characteristics. Then, the particle swarm optimization algorithm is used to obtain the optimal weights of these two prediction results, and the combined model is constructed. Finally, the validity and advanced nature of the forecasting method are verified through the load data of a region in western China in 2017 and hourly power consumption data of Turkey in 2018. Experimental results prove that the method we proposed can effectively extract the characteristics of data and improves the forecasting accuracy.