Granular computing is usually considered as a representative method for solving complex problems, which can be solved quickly through freely switching among different granular models. In this paper, a genetic programming method based on the concept of granular computing is proposed to provide an efficient solution for optimizing the topology and parameters of a satellite system simultaneously. According to the coupling relationship of multiple physical fields, the multi-granularity description method of the satellite system scheme is defined and a multi-granularity digital satellite model is constructed. The genetic programming method is improved according to the principle of falsity preserving in granular computing. The concept and calculation method of granular risk factor are proposed to allow different individuals of the current population to switch among different granularities. The convergence difficulty caused by the complexity, hugeness and high integration of satellites is effectively alleviated. The application to design and optimize an earth observation satellite proves the effectiveness of the proposed method.