Automatic Detection Method for Meter Installation and Electrical Energy Metering Faults Based on Deep Reinforcement Learning

Most existing fault detection methods for electric energy metering rely on static feature extraction and traditional algorithm modeling, which have poor ability to capture complex fault features in dynamic operating scenarios, and are prone to missed or false detections, resulting in detection accuracy and stability that cannot meet practical needs. In response to this situation, a deep reinforcement learning based automatic detection method for meter connected electric energy metering faults is proposed. Firstly, construct a fault characteristic state space covering the three core dimensions of voltage, power, and phase, and transform the physical measurement state into a standardized vector that the intelligent agent can perceive. Further utilizing deep reinforcement learning agents to select the optimal action in the preset action space, driving dynamic state transitions of the system and amplifying differences in fault features. Finally, by combining the state transition results with targeted reward functions, a dual judgment logic is designed to verify the authenticity of the fault and output the results. The experimental results show that the recall rate of this method is 0.978, the F1 score is 0.980, the AUC value is as high as 0.993, and the false detection rate is only 0.018. The core accuracy index is comprehensively better than the comparative method, and it can efficiently and accurately identify various types of meter connected electric energy metering faults.