Dynamic self-recurrent wavelet neural network for solar irradiation forecasting

This chapter presents a dynamic neural-wavelet network with online parameter adjusting by a dissipation approach for the solar irradiation and solar daily hours forecast. The main idea of the results shown in this study is based on the design of a dynamic wavelet neural network considered a discrete dynamic system to ensure stability and accuracy in forecasting solar irradiance and solar daily hours. Then, by implementing a Lyapunov approach and considering the system’s energy dissipation properties, an adaptive adjusting law is obtained for the wavelet parameters and the entire network. Therefore a robust forecasting strategy is done considering that, in opposition to other neural network approaches, the network provides optimal performance when the training data are unstable and incomplete or when the training dataset is reduced; some attributes make this approach a suitable contribution to the field. This study focuses on these solar variables, forecasting in some Honduras regions, so before the theoretical and experimental derivations, a previous analysis is provided in this study.