Long-term Wind Power Forecasting Based on Bayesian Optimization-Based CNN-GRU Model

The safe and stable operation of the power grid, the optimization of power system dispatching, and the improvement of the economic benefits of wind farms all rely on long-term wind power prediction. Long-term wind power prediction requires handling complex time series and spatial dependencies. Insufficient data availability, accumulation of prediction errors, complexity of the model, limitations of computing resources, volatility of wind energy, and the accuracy of long-term numerical weather forecasts can all affect the accuracy of predictions. Therefore, long-term wind power prediction is more difficult than short-term prediction. This paper uses the Bayesian optimization-based CNN-GRU combined model (BO-CNN-GRU) to predict the long-term wind power in a low-wind-speed wind farm. CNN is good at extracting local spatial features from the input data. GRU focuses on modeling long-term dependencies in time series. The two complement each other. The instability and nonlinearity of wind power can be handled effectively. Bayesian optimization can handle datasets with sharp fluctuations and avoid overfitting. Compared with the CNN-GRU model, the BO-CNN-GRU model has higher prediction accuracy: the mean square error (MSE) is reduced by an average of 19.27%, the root mean square error (RMSE) is reduced by an average of 11.10%, and the mean absolute error (MAE) is reduced by an average of 17.24%. The generalization ability of the model has also been enhanced: the MSE, RMSE and MAE values have decreased by 20.45%, 10.45%, and 17.85%, respectively. Meanwhile, the model has good robustness and strong seasonal universality. The new model has a good application effect in our case.