Abstract:<p>Calibration devices in an automatic verification system are used to verify the accuracy of electricity smart meters. As time goes by, calibration devices may experience the metering performance degradation, making results of error tests biased. Therefore, it is of great significance to ensure calibration devices have qualified metering performance. Conventionally, their performance could only be manually inspected at regular intervals, while keep unknown during the operation. To address this issue, an on-line anomaly identifying method based on calibration devices' parallel working mode is proposed to check their metering performance in real time. Instead of off-line inspections, a device's metering performance is associated with the corresponding results of error tests, and comparisons among distributions based on results corresponding to different calibration devices would provide evidences to identify the ones having metering performance unqualified without interrupting the verification tasks. In addition, a case study is conducted to prove the validity.</p>

What problem does this paper attempt to address?

The paper primarily addresses the issue of calibration equipment metrological performance degradation in the Automatic Verification System (AVS). Specifically, the paper proposes an online anomaly detection method to monitor the metrological performance of calibration equipment in real-time, ensuring it remains in a qualified state. During the large-scale automated testing of smart electric meters, calibration equipment is used to verify the accuracy of the meters. However, over time, these calibration devices may experience a decline in metrological performance, leading to deviations in test errors. Traditionally, the performance of calibration equipment can only be monitored through periodic manual inspections, during which the performance status of the equipment is unknown. To solve this problem, the paper proposes an online anomaly detection method based on the parallel working mode of calibration equipment. This method allows for real-time inspection of the metrological performance of calibration equipment without interrupting normal testing tasks. The core ideas of the method include: 1. **Correlation Establishment**: By analyzing the relationship between the calibration equipment and the meters under test, the metrological performance of the calibration equipment is correlated with the test errors of the meters. 2. **Distribution Comparison**: The cumulative distribution differences in test errors among different calibration equipment are used to reflect the metrological performance differences between them. 3. **Anomaly Detection**: The maximum distance measure is used to describe the differences between two cumulative distributions, and box plots are used for outlier detection to determine which calibration equipment has unqualified metrological performance. The effectiveness of this method is demonstrated through case studies. Additionally, the paper discusses the impact of dataset selection and volume on the reliability of conclusions, as well as how to adaptively identify anomalous calibration equipment through iterative strategies.

An on-line anomaly identifying method for calibration devices in an automatic verification system for electricity smart meters