Constellation reconfiguration is a critical issue to recover from the satellite failure, maintain the regular operation, and enhance the overall performance. The constellation reconfiguration problem faces the difficulties of high dimensionality of design variables and extremely large decision space due to the great and continuously growing constellation size. To solve such real-world problems that can be hardly solved by traditional algorithms, the evolutionary operators should be promoted with available domain knowledge to guide the algorithm to explore the promising regions of the trade space. An adaptive innovationdriven multi-objective evolutionary algorithm (MOEA-AI) employing automated innovation (AI) and adaptive operator selection (AOS) is proposed to extract and apply domain knowledge. The available knowledge is extracted from the final or intermediate solution sets and integrated into an operator by the automated innovation mechanism. To prevent the overuse of knowledge-dependent operators, AOS provides top-level management between the knowledge-dependent operators and conventional evolutionary operators. It evaluates and selects operators according to their actual performance, which helps to identify useful operators from the candidate set. The efficacy of the MOEA-AI framework is demonstrated by the simulation of emergency missions. It was verified that the proposed algorithm can discover a non-dominant solution set with better quality, more homogeneous distribution, and better adaptation to practical situations.