CNN-BiLSTM combined with Bayesian optimization for short-term wind power prediction

Abstract Aiming at the difficulty of wind power prediction due to the volatility and uncertainty of wind power generation, this paper proposes a hybrid model based on Convolutional Neural Network (CNN) and Bidirectional Long Short-Term Memory Network (BiLSTM), and optimises the model hyper-parameters using Bayesian Optimization Algorithm (BO) in order to improve the prediction accuracy. Firstly, the input features that are highly correlated with wind power are screened using the Pearson Coefficient method (PCC). Then, CNN is used to extract features from the screened data. Next, the features extracted by CNN are further processed using BiLSTM to capture the long-term dependence and bi-directional information of the time series data. Finally, the hyperparameters of the model are adjusted by BO to obtain the best prediction performance. The experimental results show that the proposed BO-CNN-BiLSTM model reduces the RMSE by 35.3%, the MAE by 46.7%, and the R 2 improves by 1.5% compared with the BiLSTM model.