Multi Feature and Multi Model Fusion for Improved Prediction of Offshore Wind Power Output

Accurately forecasting offshore wind power output is essential for enhancing the reliability and stability of renewable energy integration into power grids. This paper presents a novel approach that leverages multi-feature and multi-model fusion to improve the prediction of offshore wind power output. Using actual power output data from wind farms, we developed and tested models using basic atmospheric features (BAF) and an extended set of multi-dimensional features (MUF) that include lag features, difference features, and rolling window statistics to capture complex dynamic relationships affecting wind power generation. These features were then applied to Support Vector Regression (SVR), Random Forest (RF), Light Gradient Boosting Machine (LGB), and a proposed Multi-Model Fusion (MMF) method. Comparative analysis of RMSE and MAE across all wind farms consistently demonstrates that the MUF models outperform the BAF models, highlighting the substantial enhancement in predictive accuracy achieved by incorporating multi-dimensional features. Furthermore, the MMF model consistently achieved the lowest error metrics among all models, validating its effectiveness in leveraging the strengths of individual models for improved accuracy. The results highlight the potential of multi-feature and multi-model fusion in achieving more reliable and accurate offshore wind power predictions, ultimately contributing to the optimization of renewable energy integration into existing power systems.