The offensive and counterattack around the maritime island can be regarded as a multi-stage sequential counterattack process, which is usually modeled as a zero-sum game with imperfect information, the game model elements are analysized. Aiming at the security game problem of maritime base protection, the imperfect information sequential game model is constructed. A configurable counterfactual regret (CogCFR) minimization algorithm based on base CFR variants and meta-controller is proposed to solve the approximate Nash equilibrium strategy quickly, which can dynamically control the CFR hyperparameters. For the experimental background of multiple maritime island protection, the CogCFR minimization algorithm is used to solve the resource allocation strategy of maritime island protection. This paper presents a robust opponent exploitation strategy updating method with unilateral trust region considering constraints for bounded rational opponent. The experimental results show that the CogCFR minimization with meta-learning is more efficient and has fewer parameters than the dynamic weighted CFR minimization. The algorithm has good application feasibility and domain generalization, and can provide reference for strategy solving of sequential interactive game.