A point-interval wind speed prediction model based on entropy clustering and hybrid optimization weighted strategy

Wind speed prediction is crucial for effective energy management, power dispatching, and optimizing wind energy conversion systems. However, its inherent randomness and instability pose significant challenges. This paper introduces a wind speed prediction method that enhances accuracy through entropy clustering and a hybrid optimization weighted strategy. Firstly, the training set is decomposed and reconstituted into multiple feature subsequences by the improved complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN). Secondly, the internal relationship between the training set and these subsequences is constructed through the gated recurrent unit (GRU). To prevent information leakage, this relationship is mapped to the testing set. Based on the characteristics of each subsequence, the optimal prediction model is selected. Finally, chaos game optimization (CGO) is used to weighted integrate the prediction results of each model to obtain the final point and interval prediction results. The proposed method is evaluated using data from six Chinese wind farms located in diverse geographical areas. Compared with other models, the mean squared error (MSE) of the proposed method on the six datasets is 0.882 m/s, 0.507 m/s, 0.174 m/s, 0.197 m/s, 0.362 m/s and 0.322 m/s, respectively. This fully proves its effectiveness and broad application prospects.