A cooperative jamming mode adjustment method based on Multi-Agent reinforcement learning

Abstract

With the advancement of multifunctional netted radar systems (NRS), traditional jamming decision-making strategies struggle to adapt to the nonlinear challenges of dynamic electromagnetic countermeasures environments, particularly against multifunctional NRS. To address this, we propose a Multi-agent Joint Collaborative Jamming Adjustment Method (MJCJMA). Firstly, a non-cooperative adversarial scenario model is constructed, employing an improved snow melting algorithm (GPSAO-LSSVM) for radar threat pre-evaluation. And a threat quantification model is developed using enhanced entropy weighting (IEWM) and improved TOPSIS (ITOPSIS). Then, a multi-agent reinforcement learning algorithm is designed, integrating prioritized experience replay, entropy regularization, and reward centering to improve efficiency and stability. Furthermore, an alternating training strategy is proposed, which significantly accelerates the convergence process. Extensive simulation results validate the superiority of MJCJMA, which significantly reducing radar detection probability (96.25% vs. baseline, 47.51% vs. non-alternating training) and threat levels, thereby enabling intelligent jamming decisions in adversarial scenarios.