Simulation and energy management of an experimental solar system through adaptive neural networks

Abstract

In this study, the authors consider a solar system which consists of a solar trainer that contains a photovoltaic panel, a DC centrifugal pump, flat plate collectors, storage tank, a flowmeter for measuring the water mass flow rate, pipes, pyranometer for measuring the solar intensity, thermocouples for measuring various system temperatures and wind speed meter. The various efficiencies of the solar system have been predicted by an artificial neural network (ANN) which was trained with historical data. The ANN fails to predict the efficiencies accurately over the long-time horizon because of system parts degradation, environmental variations, date changes within the year from the modelling date and presence of modelling errors. Therefore the ANN is adapted using the error between the ANN-predicted efficiency and the efficiency measurement from the appropriately selected sensors and efficiency laws to update the network's parameters recursively. The adaptation scheme can be performed online or occasionally and is based on the Kaczmarz's algorithm. The adaptive ANN capability is demonstrated through computer simulation.