The automatic verification device for energy meters conducts quality checks through periodic verifications and interim verifications, leading to issues such as low operational efficiency, long verification cycles, and the inability to promptly identify and rectify anomalies. Therefore, implementing online anomaly detection for the automatic verification device for electricity meters holds significant importance. In this study, a method for online anomaly detection based on a small number of unlabeled samples is proposed. It involves the online collection of positional error data from the automatic verification device for electricity meters, construction of error features, extraction of crucial data components using principal component analysis. Subsequently, it employs the isolation forest algorithm to mark anomalous meter positions. Following this, a support vector machine (SVM) model is trained based on the marked results, and the hyperparameters of the SVM are optimized using k-fold cross-validation and Bayesian optimization. With the use of only a small number of support vectors, the method achieves anomaly detection for meter positions with an accuracy rate of 99.44%, demonstrating effective anomaly detection.