In this work, we study the uncertainty modelling, reliability analysis and optimization problem for a black-box aeronautic system. Both aleatory and epistemic parameter uncertainties are considered. We propose a bootstrap-aggregation approach and a two-stage optimization scheme to obtain the regression system parameters for each observation data sequence, and, then, determine the uncertain models according to the obtained parameter set. The interval-presented uncertain system reliabilities are computed by a Monte-Carlo Simulation based inner-approximation method. For the parameter regression, we propose an elitism co-evolutionary algorithm as the learning algorithm for the regression process. We also proposes another genetic algorithm to find the robust optimal system design point, which maximizes the worst-case realization of the uncertain system reliability. In the end, we also proposed a modified uncertainty model, which takes small risks in the uncertain system reliability evaluation.