With the large-scale constellation, the importance of a single satellite is reduced. When formulating a replacement strategy for faulty satellites, multiple replacement methods and the retention of faulty satellites should be considered. In response to this demand, a method for formulating constellation orbit satellite replacement strategy based on Markov decision process is proposed. The orbital state is divided according to the number of satellites working in orbit, in-orbit backup and ground backup. On this basis, combined with the different replacement strategies of faulty satellites and the corresponding ordering, storage and penalty costs, a return function model representing the expected maintenance cost is constructed. In order to minimize the expected maintenance cost of each decision cycle, a sequential replacement strategy is developed and applied to a constellation orbit. The results show that when the ground backup is used to replace the fault satellite, it tends to supplement the consumed backup satellite. The longer the mean time between failure, the lower the expected total maintenance cost, but the effect is gradually weakened. The increase of penalty cost will lead to the solidification of replacement strategy, and the upward trend of total maintenance cost is expected to be greatly slowed down. This method can make a sequential decision on the replacement mode of the faulty satellite according to the state of the constellation orbit, which makes the analysis more comprehensive and has certain value for the formulation of the constellation spare strategy.