A novel method for solar panel temperature determination based on a wavelet neural network and Hammerstein-Wiener model

Abstract

Accurate prediction of solar panel temperature can help keep on-orbit satellites in good condition. Traditional physical models have the ability to describe and predict temperature; however, the effect is not entirely satisfactory. To produce a better forecast, the panel current of the solar panel, which is strongly correlated with temperature signals, is chosen as the input, and a novel system identification model between the two signals is established. The model we propose is based on a Hammerstein-Wiener model and integrates wavelet neural networks that adopt self-constructed wavelet bases. In addition, a complete process of training and parameter optimization is designed to be less time consuming than previous methods. The result from a test set of real telemetry data demonstrates the efficiency and accuracy of our method. Moreover, the proposed prediction model, which is based on historical data, can be used in on board self-learning and our subsequent autonomous health management.