This paper combines a derivative-free hybrid optimization algorithm, generalize pattern search (GPS), with Treed Gaussian Processes (TGP) to create a new hybrid optimization algorithm. The goal is to use the method for top design of satellite system, in which the objective or constraint functions usually are computationally expensive black-box functions. TGP model partitions the design space into disjoint regions, and employs independent Gaussian Processes (GP) in each partition to represent the time consumption of true problem responses. Utilizing the TGP, we generate the new “promising” points, which are the combination of model-predicted values and estimated model errors. Then, these points are used to guide GPS search in the design space efficiently. The hybrid optimization method is applied to top design of multi-satellites cooperated observation. The results demonstrate that the proposed method can not only increase the chance of obtaining optimal solution but also cut down the cost of function evaluations.