A Novel Integrated Approach for Short-Term Wind Power Forecasting

For a sustainable environment, countries are planning to increase the share of Renewable Energy Resources (RERs) in their energy mix. Wind is the type of RERs that has the potential to overcome fossil reserves. However, due to the uncertain nature of wind, its integration with existing power systems brings instability. This instability can be overcome by implementing some accurate wind power forecasting techniques. Therefore, in this proposed study, we present a novel integrated approach for hour-ahead wind power forecasting. The proposed approach has two levels. At level 1, three boosting algorithms: Extreme Gradient Boosting (XgBoost), Categorical Boosting (CatBoost), and Random Forest (RF) are implemented. Level 2 consists of Linear Regression (LR) which adjusts the weights and biases and gives an hour-ahead forecast. The datasets of two different wind turbines are used in this study. For the robustness analysis, the proposed approach is compared with the XgBoost, CatBoost, RF, Long Short-Term Memory (LSTM), and Gated Recurrent Unit (GRU) networks. The findings demonstrate that the proposed integrated approach outperforms other models. For the Texas turbine dataset, the proposed approach records improvement of 11.51% and 39.1% in MAE than XgBoost, and CatBoost models, respectively. In the case of Turkey turbine dataset, the MAE of the proposed integrated network is 1.9% and 26.65% better than the MAE recorded by XgBoost, and CatBoost models. respectively.