An Ensemble Model for Short-Term Wind Power Forecasting using Deep Learning and Gradient Boosting Algorithms

Abstract

An accurate wind power forecasting is crucial for day-to-day grid operation. It plays a significant role in decision making for wind power producers playing in the electricity market. Deep learning and gradient boosting decision tree-based algorithms have been proven efficient in feature extraction from the time-series data and provide considerably good forecasting results. Since wind is highly intermittent and significantly varies both temporally and geographically, most algorithms perform well on certain wind datasets and poorly with a different dataset. Seasonal variability also affects the performance of the forecast model. Therefore, it is prudent to use an ensemble model rather than relying on a single model. In this paper, we propose an ensemble forecast model that takes the benefits of four machine learning algorithms, which include Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), Light Gradient Boosting Machine (LightGBM), and eXtreme Gradient Boosting (XGBoost). To demonstrate the effectiveness of the proposed model, we applied it to the wind power data taken from an actual wind farm located in Gujarat, India. The results illustrate that the proposed model outperforms the individual models and predicts the wind power for different months with significant accuracy.