Dynamic reconfiguration is critical to improving safety of the flight, minimizing application loss, and balancing partition load. Reconfiguration optimization is a promising way to achieve superior flight safety because of the finite system resource and random failure. To optimize the reconfiguration strategy, we propose a deep reinforcement learning algorithm. The problem is formulated as a Markov decision process with five-tuple, and solved by our proposed deep deterministic policy gradient algorithm. The simulation result shows that our proposed algorithm outperforms the Actor-Critic algorithm and optimized reconfiguration strategy not only restores successfully the system application when failure occurs, but also balances the partition load, further improving the system security.