Wind forecasting with artificial intelligence methods

In recent years, renewable energy sources have been preferred for meeting electricity demands in developed and developing countries. Wind energy is widely used among these energy sources because it is more efficient. However, accurately and reliably predicting wind speed is very important in the management and operation of wind energy power systems. Due to the intermittent and non-stationary nature of wind speed, modeling and predicting it becomes challenging. In this study, a deep learning-based two-stage model is proposed for effective wind forecasting for the planning and feasibility studies of wind farms. In this approach, firstly, wind speed time data was converted into color images using continuous wavelet transform. Then, effective wind forecasting was performed using these images with the help of the proposed convolutional neural network. The study utilized hourly wind speed data from 2018 to 2019 obtained from the Elazığ Meteorology Regional Directorate. In the experimental studies, three different forward horizon predictions were made: 1-hour, 2-hour, and 3-hour forecasts. Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and correlation coefficient (R) metrics were used for performance evaluation of the proposed model. According to the experimental results, in 1-hour ahead predictions, successes of 0.0615, 0.8757, and 0.0448 were achieved in RMSE, R, and MAE performance evaluation criteria, respectively. In 2-hour ahead predictions, it was observed that the performances in RMSE, R, and MAE metrics slightly decreased compared to the 1-hour ahead predictions, with values of 0.0804, 0.7397, and 0.0590, respectively. This trend continued in the 3-hour ahead predictions as well. All these experimental results show that the proposed model exhibits better prediction performance for shorter forecast periods. Additionally, for more detailed performance analysis, it was seen that the proposed model was more successful in the 1-hour ahead prediction when compared with the results of three different deep learning models like AlexNet, ResNet50, and GoogLeNet. Keywords: Wind Energy Forecasting, Continuous Wavelet Transform, Convolutional Neural Network, Deep Learning.