When architecture meets RL+EA: A hybrid intelligent optimization approach for selecting combat system-of-systems architecture

Abstract

In recent decades, various domains such as military, aerospace and supply chain are becoming increasingly complex, forming system-of-systems (SoSs). The effectiveness of SoSs is mainly determined by their architecture, thus selecting proper architecture for SoSs is crucial to maximize their effectiveness. While most studies focus on specific issues, such as mission planning and SoSs meta-architecture selection, few papers deal with the two issues together. This paper introduces the task oriented SoSs architecture selection problem (TSASP) and proposes an combination of reinforcement learning (RL) and evolutionary algorithm (EA) (CRE) to address this problem. CRE comprises two layers: the inner layer utilizes a DQN-based algorithm to solve the mission planning problem, and the outer layer with improved genetic algorithm optimizes the SoSs architecture selection based on the output of inner one. A synthetic air and missile defense (AMD) example is conducted to test the effectiveness of proposed method, and results show that CRE can improve the mission planning effect by select a feasible architecture.