Offshore Wind Power Prediction Integrating Wind Region Clustering and CNN-Transformer

Accurate power prediction is essential for the optimal scheduling of large-scale offshore wind power integration. To learn the power variation patterns from highly fluctuating and stochastic offshore wind power data, this paper designs a short-term power prediction framework that integrates clustering embedding, convolutional neural network (CNN), Transformer encoder, and adaptive attention. Firstly, based on correlation analysis, key meteorological elements affecting wind power are screened, and deep features including wind speed-wind direction interaction features and wind direction change rate are constructed to establish a high-quality feature dataset. Secondly, a strategy combining wind direction pre-classification and K-means clustering is adopted to optimize the distribution of training data and enhance the identification ability for multiple wind condition patterns. Furthermore, a CNN-Transformer prediction model is constructed, where CNN is used to extract multi-scale temporal features from local meteorological sequences, and then the Transformer encoder is utilized to capture global correlations. An adaptive attention pool is introduced at the output layer to dynamically learn the importance of features at each stage, achieving higher accuracy in feature fusion. Finally, this study selects an offshore wind farm for case analysis. The test results confirm that this method can accurately analyze the deep temporal correlations in data, significantly improving prediction accuracy and robustness, providing an efficient and feasible solution for offshore wind power prediction.