Neural network models for a solar power plant

This paper deals with the identification of a nonlinear solar power plant using neural networks. The nonlinear identification problem is tackled by decomposing the complex system in two main components: an active part and a passive part. For the active part of the solar power plant a model based on the parallel connection of ten neural networks; is built, while for the passive part a white box model and a neural network black box model are developed. All models are identified and validated using measurement data collected at Plataforma Solar de Almeria. Practical aspects regarding inputs selection and neural networks training are discussed, and physical modelling principles are explained. A comparison between the two overall models is also provided.