Short‐term load forecasting based on CNN‐BiLSTM with Bayesian optimization and attention mechanism
Hui-Feng Shi, Kai Miao, Xiaochen Ren
Abstract
Abstract Accurate short‐term load forecasting can ensure the reliable power supply of the power system and the regular operation of the grid economy. This article proposes a hybrid model based on convolutional neural networks and bidirectional long short‐term memory (CNN‐BiLSTM) with Bayesian optimization (BO) and attention mechanism (AM) for short‐term load forecasting. CNN is used to capture the significant features of input data. BiLSTM is adept in time series forecasting and AM can reduce the computational complexity of model. BO can help to tune the hyperparameters automatically. The input features of models are recent load, time slot, date type, and meteorological factors. In order to eliminate the seasonality, the data set is divided into four subsets according to four seasons. The performance of the proposed model is compared with other models by MAE, RMSE, MAPE, and score. The forecasting results represent that the proposed model is most suitable for short‐term load forecasting among the contrast models and the meteorological factors have an impact on forecasting accuracy.