Offshore Wind Power Forecasting Method Based on Hybrid Model Optimization

The accurate and real-time offshore wind power forecasting can effectively mitigate the impact of offshore wind power integration on the power system, enhancing its safety, stability, economy, and controllability. However, offshore wind power forecasting currently faces three major challenges: difficulty in identifying anomalous data, high dimensionality of input vectors, and instability in multi-step forecasting. To address these issues, this paper optimizes existing wind power forecasting technologies from three aspects: "data", "feature" and "model", proposing an ultra-short-term power forecasting method for offshore wind power based on DBSCAN-Pearson-Informer. First, a wind power data preprocessing method based on the adaptive DBSCAN algorithm for different wind speed intervals is proposed, which effectively identifies various types of abnormal wind power data and corrects them. Second, a feature selection method for wind power forecasting based on Pearson correlation analysis is proposed, which selects factors highly correlated with wind power as input data for the forecasting model. Finally, an ultra-short-term offshore wind power forecasting model based on Informer is constructed, and the trained model is used to carry out offshore wind power forecasting work. Experimental results demonstrate that, compared to classical time-series forecasting models such as LSTM, TCN, and Transformer, the proposed method achieves average percentage improvements in MAE, RMSE, and MBE of 66.11%, 63.02%, and 90.73%, respectively, while also exhibiting higher forecasting stability and faster execution efficiency.