Wind Power Output Forecasting Based on AC-BiLSTM

Abstract

Wind power output prediction is crucial to the balanced dispatch and market operation of the power system, and accurate prediction of wind power output can guarantee the security and stability of power grid operation. In order to realize highly accurate wind power output prediction to support the low-carbon transformation of the power system, a wind power output prediction method integrating convolutional neural network (CNN), bi-directional long and short-term memory network (BiLSTM) and Attention mechanism is proposed. To address the nonlinear and temporal characteristics of wind power output data, the CNN layer extracts local nonlinear features, the BiLSTM layer captures bidirectional temporal dependencies, and the Attention mechanism reduces redundant interference and improves the model interpretability by dynamically weighting the key temporal information. Finally, the final output of the output power prediction result is output through the fully connected layer. The example analysis is carried out with the actual wind power data of China region in 2012, and the model generalization ability is optimized by using normalization processing and Dropout technique, and the parameters are trained by using the mean square error (MSE) as the loss function in combination with Adam’s algorithm, which verifies the model’s validity and robustness in complex fluctuation scenarios, and provides technical support for the planning and stable operation of the power system. The model can provide technical support for power system planning and stable operation.