Adaptive ε-greedy exploration for stable reconfiguration in next-gen aviation IMA systems

Abstract

Abstract The next-generation aviation Integrated Modular Avionics (IMA) system adopts an architecture based on container technology, offering higher resource utilization and task configuration flexibility while increasing system reconfiguration complexity. Efficient reconfiguration strategies enhance adaptability and fault tolerance, ensuring stable operation and reduced maintenance costs. However, existing manual and heuristic-based methods struggle to meet current fault tolerance requirements. We propose an embedded container reconfiguration method using a Double Dueling DQN with Adaptive $$\varepsilon$$ -Greedy Exploration (D3QNAE), which incorporates adaptive exploration to efficiently generate stable strategies in complex environments. Experimental results demonstrate that D3QNAE reduces the first feasible solution time by 34 $$\%$$ compared to the best baseline (D3QN) in large-scale deployments (500-task scenarios), while achieving a 15.6 $$\%$$ higher maximum reward value and a 100 $$\%$$ migration impact rate under continuous faults. This method provides enhanced fault tolerance for container-based IMA systems, significantly improving stability.