Short-Term Wind Speed Forecasting With Deep Learning

Wind speed forecasting is crucial for planning and investing in renewable wind energy. In addition, Wind speed forecasting is essential for planning renewable wind energy, optimizing wind power production, and enhancing transmission line capacities. However, intermittent and stochastic fluctuations of wind speed pose a significant problem for high quality wind speed forecasting. In this study, a deep learning-based approach is proposed for wind power plant planning and feasibility studies that can provide wind speed prediction more easily. In this approach, firstly, wind speed time data were converted into color images using continuous wavelet transform. The obtained images were applied to the pre-trained AlexNet CNN model and wind speed prediction was performed. In the study, hourly speed data from the Elazig meteorology regional directorate between 2018-2019 were used. In the experimental studies, three different horizon forecasts were made: 1-hour, 2-hour and 3-hour. Metrics like correlation coefficient (R), mean absolute error (MAE), and root means square error (RMSE) were utilized to assess the proposed forecasting models performance. In the experimental studies, the whole dataset images were randomly divided into three parts as training, validation and test at the rates of 70%, 10% and 20% respectively for transfer learning. In the 1-hour horizon forecast, the best wind speed prediction was achieved with experimental results of 0.0335, 0.0275 and 0.9517 for RMSE, MAE and R metrics, respectively. In this respect, the proposed AlexNet model shows that it is an effective model in wind speed forecast since the 1-hour horizon forecast is more reliable and accurate.