Forecasting System of Wind Speed and Direction by Neural Network

This paper discusses the neural network that is used to forecast wind speed and direction by combining methods of three-dimensional convolutional neural networks (3D-CNN), convolutional long short-term memory (CLSTM), and two-dimensional CNN (2D-CNN). Its method is called 3CN-CLSTM-2CN. Combining the three methods is to diminish the forecasted error and training time compared to the 3CN-CLSTM method. The input of the forecast system is the wind vector in the form of image data that indicates the image of time-sequential which is given by the Automated Meteorological Data Acquisition System (AMeDAS), Japan. Evaluate the forecast system used by the root mean square error (RMSE) between observed data and forecast data. For verification, the usefulness of the proposed forecasted system is compared with the persistent system. The suggested forecasting system's objective is to increase prediction accuracy for each season, which is typical of Japan. The proposed forecasted system by 3CN-CLSTM-2CN for each season in one-year data can increase around 16.11-26.47% to wind speed and around 20.69-23.34% to wind direction over the persistent system, which has a greater accuracy than the 3CN-CLSTM system. So, it was suitable that the 3CN-CLSTM-2CN system be the most robust forecasting system.