When solving the multiple agile satellites cooperative scheduling problem, the metaheuristics faces many problems due to their low intelligence, such as premature or late convergence, poor stability, etc. To solve these problems more efficiently, a data-driven adaptive parallel search algorithm is proposed. Firstly, some task allocation operators are designed to based on domain knowledge, with the purpose of transforming the multiple agile satellites cooperative scheduling problem into multiple single-satellite task scheduling subproblems. Then, multiple threads are started to parallelly and independently solve each single-satellite task scheduling problem. During algorithm iterations, each thread selects different neighborhood operators based on probability, and dynamically updates the probability of neighborhood operators and elites. Next, the frequent pattern mining method is applied to extract knowledge from the elites to construct new solutions. Finally, all single-satellite task scheduling results are fed back to the task allocation layer to start a new allocation. The simulation results show that the proposed algorithm can obtain high-quality solutions within a limited time, and has good applicability and optimization effects in different scenarios.