A Hybrid QPSO-LSTM-CEEMDAN-Attention Model for Wind Power Forecasting

In response to the issue of wind power generation being easily affected by random factors such as weather conditions, leading to insufficient prediction accuracy, this paper proposes a composite model for wind power prediction based on QPSO-LSTM-CEEMDAN-Attention. First, employ Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) on wind power time series data, and then utilize the Attention mechanism to optimize the LSTM algorithm to accurately capture the key information affecting model predictions and enhance the weights of key features. Next, use the Quantum Particle Swarm Optimization (QPSO) algorithm to globally optimize the hyperparameters of the LSTM, and then build the LSTM model based on the obtained optimal hyperparameters to complete the prediction. Finally, the predicted results are fused to obtain the final prediction result, constructing the QPSO-LSTM-CEEMDAN-Attention wind power prediction model. Using the historical monitoring data of actual wind farms as the dataset, train and test the model. Experimental results show that the composite model effectively integrates the advantages of different algorithms. Compared with the LSTM model and the LSTM-Attention model, this model reduces RMSE to 8.5042 and MAE to 5.3372, significantly lowering prediction errors. The R2value increases to 0.97953, indicating a substantial improvement in goodness of fit. It demonstrates superior accuracy and stability in wind power forecasting, providing a new methodological reference for precise wind power prediction.