Cooperative evaluation of multiple wind farms based on information in neighboring wind farms and improved TCN

To enhance the cooperative evaluation performance of multiple wind farms, this paper proposes a novel short-term wind power forecasting framework that integrates information from neighboring wind farms and an improved temporal convolutional network (TCN). The primary research gap addressed in this work is the lack of effective utilization of spatial–temporal correlations between wind farms and the failure of traditional models to optimize feature selection from neighboring farms, which limits forecasting accuracy. First, this study analyzes the correlation between the comprehensive wind speed series and wind power series. And calculates comprehensive similarity scores between neighboring and target wind farms to construct a high-dimensional feature dataset; Next, floating search feature selection algorithm is introduced to optimize the features. Finally, on the basis of TCN, multi-scale convolutional neural network (CNN) is used to extract local features, and the dual multi-head self-attention of features and temporal are introduced to mine the internal correlations between different input features and different time-steps in the feature matrix, respectively. These innovations address the gap in accurately capturing dependencies across both spatial and temporal dimensions. Case analysis is carried out based on the actual data in a certain area. The results demonstrate that, compared to traditional benchmark models, the proposed model reduces the MAE and MSE by at least 17.79 % and 3.91 %, respectively, significantly improving prediction accuracy.