Target assignment for multiple stages of weapons systems using a deep Q-learning network and a modified artificial bee colony method

Abstract

Multistage Weapon Target Assignment (MWTA) in defense systems is a critical process that focuses on minimizing deviations and inaccuracies during target assignment across stages like initialization, tracking, and counterattack. The research problem lies in efficiently assigning targets to weapon systems while accounting for distractions and tracking accuracy constraints. Existing approaches often face challenges related to computational complexity, real-time engagement limitations, and scalability issues, leading to inaccuracies in various stages within the system's perceived radius. This paper proposes a novel State-based Modified Artificial Bee Colony Method (SMABCM) that integrates Q-learning to generate adaptive target assignment decisions. SMABCM aims to maximize assignment accuracy by prioritizing low deviation rewards, enabling precise one-to-one target tracking and reducing the population of the artificial bee colony. The Q-learning network dynamically defines generation and reduction states based on deviations to minimize distracting targets and enhance tracking precision. Experimental results demonstrate SMABCM's superior performance over existing approaches in target precision comparison, hit probability, increase in distracting target detection, reduced deviation factor, and computation time. As the speed increases, the proposed SMABCM improves target accuracy by 9.54 %, hit probability by 9.17 %, distracting target detection by 12.35 %, while reducing deviation by 7.39 % and computation time by 10.11 %. This approach addresses complexities in multistage weapon systems through dynamic strategies, contributing to proactive defense against evolving security threats.