Instability fault monitoring method with self-learning strong generalization capability and instability fault monitoring apparatus

Abstract

The present invention discloses an instability fault monitoring method with self-learning strong generalization capability and an instability fault monitoring apparatus. First, electrical volume acquisition nodes are identified and data composition features are collected to form samples, the samples including a small number of labeled data samples and a large number of unlabeled data samples. Then, the samples are input into a reinforcement learning network of actor-critic architecture, and self-learning can be realized by means of the network to form different diagnostic models, where actor selects a suitable network layer from action space (alternative models) to construct a deep belief neural network and formulate a corresponding strategy, critic is configured to evaluate the strategy currently formulated by actor, and the reinforcement learning network outputs the corresponding actor-critic architecture and diagnostic model. Finally, the trained diagnostic model is used to diagnose an instability state of a power electronic system online in real time. The method is capable of simultaneously solving the problems about instability monitoring and single-module faults for different power electronic system structures and fault diagnosis tasks, and has strong generalization capability and high adaptive use.