Ultra-short-term wind power prediction method based on CEEMDAN-GAAD-HLSTM modeling

As a renewable energy source, accurate wind power prediction plays an important role in grid maintenance and power system planning. In order to improve the accuracy of ultra-short-term wind power prediction, a CEEMDAN-GAAD-HLSTM combined ultra-short-term wind power prediction model is proposed. Firstly, CEEMDAN is used to decompose the original wind power data, and a number of eigenmode functions are extracted from the original nonlinear data. Then dynamic adjustment and elite retention strategies are introduced into the genetic algorithm to prevent falling into local optimality and accelerate the convergence speed of the algorithm. The dynamic adjustment threshold strategy is introduced in HuberLoss to improve the model robustness. Finally, the improved genetic algorithm (GAAD), the improved HuberLoss and LSTM are built into a combined model GAAD-HLSTM, and the decomposed sequences are predicted separately with the combined model, and the combined prediction results are obtained as the final prediction results. The experimental results show that compared with the single LSTM model, the combined CEEMDAN-GAAD-HLSTM model has improved in MAE, RMSE and R2, and reduced 65.914 and 83.129 in MAE and RMSE, respectively, while the R2 is improved from 0.923 to 0.987, which effectively improves wind power prediction, and confirms that the method is effective in wind power prediction, confirming the effectiveness of the method in wind power prediction.