An Improved DDPG and Its Application in Spacecraft Fault Knowledge Graph

Abstract

We construct a spacecraft performance-fault relationship graph of the control system, which can help space robots locate and repair spacecraft faults quickly. In order to improve the performance-fault relationship graph, we improve the Deep Deterministic Policy Gradient (DDPG) algorithm, and propose a relationship prediction method that combines representation learning reasoning with deep reinforcement learning reasoning. We take the spacecraft performance-fault relationship graph as the agent learning environment and adopt reinforcement learning to realize the optimal interaction between the agent and the environment. Meanwhile, our model uses a deep neural network to construct a complex value function and strategy function, which makes the agent have excellent perceptual decision-making ability and accurate value judgment ability. We evaluate our model on a performance-fault relationship graph of the control system. The experimental results show that our model has high prediction speed and accuracy, which can completely infer the optimal relationship path between entities to complete the spacecraft performance-fault relationship graph.