Short-term Electricity Load Forecasting Based on TPE-optimized CNN-BIGRU-Attention

Abstract

To fully exploit the time series characteristics of the electricity load data and further improve the prediction accuracy for the intelligent dispatch of the power grid, this paper proposes a combined model of short-term electricity load forecasting based on the Bayesian optimization algorithm (TPE) optimized convolutional neural network (CNN) — bi-directional gated recurrent network (BiGRU) — Attention mechanism (Attention). First, the data are pre-processed by normalization and removal of missing values. Then, a CNN-BiGRU-Attention prediction model is built and the TPE algorithm is used to determine the hyper-parameters. Finally, the trained model is used to complete the load prediction. This paper takes a region of China’s electricity load data as an example for seasonal forecasting, and experiments show that the model in this paper has stronger adaptability and higher forecasting accuracy compared with CNN, GRU, CNN-GRU, and CNN-BiGRU networks.