Enhancing wind power forecasting accuracy with hybrid temporal convolutional networks and adaptive stacked ensemble learning

The increasing demand for renewable energy sources highlights the importance of accurate wind power forecasting. This paper proposes a novel hybrid adaptive stacked ensemble-TCN approach for improving wind power forecasting accuracy by combining adaptive stacked ensemble learning and temporal neural networks. This approach preprocesses data by removing outliers and identifying important features for accurate forecasting, utilizing data from MERRA2 instruments. The variables of the hybrid adaptive stacked ensemble-TCN approach have been optimized to get optimal results in evaluating it on multiple wind farms in Europe (EuroWind 1, EuroWind 2, and EuroWind 3). The energy effectiveness metrics show that this hybrid approach significantly reduces errors on all three datasets. In particular, the hybrid approach RMSE obtained for EuroWind 1 is 0.0017 (as opposed to 0.0027 for random forest), and its MAE evaluation is 0.0011 (as opposed to 0.0013 for random forest). The RMSE number for EuroWind 2 is 0.0015, the comparison is 0.0019 for random forest, while the MAE value is 0.0006, as opposed to 0.0008 for random forest. Regarding EuroWind 3, the RMSE result is 0.0176 (as opposed to 0.1755 for random forest) while the MAE value is 0.0158 (as opposed to 0.0863 for random forest). These findings show that the hybrid adaptive stacked ensemble-TCN approach is more effective than the current techniques, with RMSE results from 0.0014 to 0.0176 and MAE results between 0.0010 and 0.0158. This approach greatly enhances grid stability, forecasts wind power, and encourages the production of renewable energy sources.