Short-term forecast method of wind power output based on multi-scale CNN-LSTM in extreme weather

• Hybrid CNN-LSTM Model: Combines multi-scale CNN and LSTM to improve wind power prediction under extreme weather. • Extreme Weather Modeling: Uses truncated normal distribution to quantify weather impact intensity. • Kalman Filter Correction: Enhances wind speed data accuracy for better input reliability. • XGBoost Error Compensation: Estimates prediction errors to refine final power output forecasts. • High Prediction Accuracy: Achieves close alignment with measured values, minimizing errors near zero. It is often difficult to accurately capture wind power variation rules under extreme weather conditions, resulting in large forecasting errors. Therefore, a short-term forecast method for wind power output based on multi-scale CNN-LSTM under extreme weather conditions was designed. The truncated normal distribution was used to model the impact intensity of extreme weather, the Kalman filter was applied to wind speed correction, and the prediction equation coefficient of the current moment was modified by real-time feedback of the prediction error of the previous moment. The XGBoost model was introduced to estimate the prediction error of the wind power output, and the prediction models of the multi-scale CNN and LSTM were combined. Power residuals were used as learning targets to predict the power to be corrected using convolution kernels of different sizes to extract features from different span temporal neighborhoods, and the final predicted power was obtained. The experimental results showed that the predicted power of the design method was consistent with the measured power. The predicted value of the predicted wind speed at all sample points is very close to the real value, and the relative frequency distribution of the error is more concentrated and close to zero, which can improve the accuracy of the prediction.