Edge-center hybrid optimization multi-agent reinforcement learning method for fault diagnosis

Abstract

The application relates to the computer technical field and discloses an edge-center hybrid optimization multi-agent reinforcement learning fault diagnosis method, which comprises the following steps: S1, collecting equipment data and preprocessing; S2, hierarchical fault diagnosis: an intelligent diagnosis agent adopts a hierarchical reinforcement learning structure, which is composed of a high-level policy network and a low-level policy network; S3, edge-center hybrid optimization: a center strategy optimization agent designs and trains the high-level policy network and the low-level policy network in stages, adopts a teacher-student policy structure to compress and distill the trained policy network, executes policy fusion and global updating based on received policy execution information; and S4, policy migration. The application continuously learns and updates the policy in the scene where fault samples are extremely scarce, utilizes a multi-agent collaborative mechanism to realize diagnosis knowledge sharing, policy synchronization and cross-device migration, supports efficient deployment and operation at an edge device end, and adapts to changes in a field state in real time.

Images