Volume 49

Wind speed ensemble forecasting framework based on deep neural network and dropout mechanism Zhendong Zhang, Huichao Dai, Qing Zhang

Abstract

With the depletion of non-renewable energy, renewable energy such as wind energy has received more and more attention. Wind speed prediction plays an important role in promoting the utilization of wind energy. This paper focuses on how to realize the wind speed ensemble prediction at multiple stations. First, convolutional neural networks are introduced to wind speed prediction because of its ability to mine input features. Then, the dropout mechanism is incorporated into the model so that multiple runs can obtain multiple predictions. Next, the kernel density estimation method is used to obtain the probability density function of wind speed prediction. At the same time, in order to complete the multi-station wind speed prediction at the same time, the four-dimensional input-output structure tensor is proposed for wind speed prediction. Finally, the model proposed in this study is verified on two datasets of Tibet, China. The experimental results show that: (1) The model proposed in this study can obtain high-accuracy deterministic prediction results and appropriate ensemble prediction intervals. (2) The appropriate dropout ratio is important to neither overfit nor reduce the prediction accuracy.

Keywords wind speed, ensemble forecasting, Convolutional neural network, dropout, Kernel density estimation