Convolutional Neural Network for Short-term Wind Power Forecasting

Wind power generation is becoming increasingly relevant to the power supply system as it is clean and renewable. This paper proposes a novel methodology for short-term wind power forecasting, based on a convolutional neural network (CNN). In this work, we evaluate the CNN ability of predicting the wind power generation by comparing it to two benchmarking methods - ARIMA and gradient boosting machine (GBM). We prove that CNN is well suited for this purpose, outperforming the other tested techniques, specially when the prediction horizon is greater than 1-hour. Besides, this paper shows that additional features like meteorological forecasts provide fruitful information, powering the CNN performance.