Abstract: To address the issues of slow response and low interception efficiency in traditional task allocation methods in co-operative missile interception, this paper proposes an air-sea cooperative missile interception task allocation meth-od based on Heterogeneous-Agent Proximal Policy Optimization (HAPPO). First, models of radar detection and tracking, missile dynamics, and damage effectiveness are established, and the missile interception allocation prob-lem is formally formulated within a heterogeneous multi-agent reinforcement learning framework. On this basis, a step adjustment mechanism based on event-triggered is developed to achieve task reconfiguration upon triggering events. To enhance the training stability of the interception task allocation policy, a staged training algorithm based on HAPPO is introduced, which progressively incorporates an enhanced reward from an interception hit-rate pre-diction network, thereby enhancing the cooperative decision-making capability of heterogeneous agents. Simula-tion results demonstrate that the proposed algorithm effectively overcomes the challenge of sparse rewards, ena-bles distributed cooperative missile interception, and significantly improves the interception success rate against adversarial anti-ship missiles.

Key words: Task Allocation, Heterogeneous-Agent Reinforcement Learning, Missile Interception, HAPPO Algorithm, Hit-rate Prediction