GHI forecast based on nonlinear autoregressive neural network

This paper proposes a short-term GHI forecast model based on nonlinear autoregressive dynamic neural network. At first, this paper proposes a kind of training sample in parallel structure to guarantee the time correlation within training sample. Secondly, by comparing the forecast accuracy of 511 combinations of 9 meteorological parameters, as model inputs, the best input combination is identified. Finally, this paper tests model's adaptiveness to four different typical weather conditions. By comparing with forecast results of the traditional forecast model based on focus time delay neural network, the forecast model based on nonlinear autoregressive neural network can effectively reduce the normalized root mean squared error. © 2019, Editorial Board of Acta Energiae Solaris Sinica. All right reserved.