Wind Prediction Performance of Complex Neural Network with ReLU Activation Function

The complex neural network expands on the complex plane more than the real neural network taking the real axis. So its figure transformation ability and learning speed are very good. However, the neural network needs to determine nonlinear transformation to be used in the hidden neurons, to construct an optimal network structure according to the data to be handled, and to analyze prediction ability and initial value dependency. In this paper, we propose new four kinds of Rectified Linear Unit (ReLU) function on complex plane. Wind speed and direction are simultaneously predicted using the complex neural network with the proposed ReLU function. As a result of the numerical experiments, the optimal number of degrees of freedom was determined, and the complex neural network predicts more accurately than the real neural network does.