Dynamics forecasting for electrical characteristics of spacecraft solar arrays by means of computational intelligence

We consider the problem of forecasting the degradation process of spacecraft's solar arrays according to their available parameter changes measured together with the corresponding parameters of solar activity during the fulfilment of the real spacecraft's mission. The application of ANN-based predictors is proposed because of their generalization ability. In this paper, predictors automated design with self-adaptive evolutionary algorithms is suggested because of the ANN efficiency dependence on the choice of an effective structure and the successful tuning of weight coefficients. The adaptation of evolutionary algorithms is implemented on the base of the algorithms' selfconfiguration. In self-configuration technique, setting variants were used instead of the adjusting real parameters, namely types of selection, crossover, population control and level of mutation. Each of these has its own initial probability distribution which is changed as the algorithm executes. ANN-based predictors are automatically designed with the self-configured genetic algorithm and the self-configured genetic programming algorithm. In both cases the most informative features are selected for each neural network. Besides, the self-configuring genetic programming is used for the automated design of ensembles of ANN-based predictors. The performance of developed algorithms for automated design of ANN-based predictors is estimated on real-world data and the most perspective approach is determined. Self-configuring genetic programming algorithm for automated ANN design has the best performance among non-ensemble techniques. Self-configuring genetic programming for the automated design of ensembles presents the best result among all considered approaches.