On-line monitoring of single-phase smart meter operation error based on wavelet transform and machine learning

Smart electricity meters are widely deployed in modern power systems, where their operational accuracy directly influences billing equity and grid reliability. However, harmonic interference, particularly from higher-order harmonics, can significantly compromise the metrological accuracy of these devices. Conventional error-monitoring approaches often rely on periodic on-site calibrations, which are inefficient and lack real-time capability. To address the computational complexity associated with high-dimensional harmonic features, this paper proposes a hybrid architecture designed for embedded real-time monitoring. This architecture integrates Continuous Wavelet Transform (CWT) with the isolation forest algorithm (CWT-iForest). The core of the proposed method lies in its system-level co-design. CWT is leveraged for high-fidelity time-frequency feature extraction, while an optimized iForest enables efficient dimensionality reduction and anomaly detection in the feature space. This design maps high-dimensional features into a lower-dimensional space while preserving critical physical characteristics, thereby significantly enhancing computational efficiency and real-time capability. Simulation results indicate that the integrated scheme achieves markedly higher error identification accuracy than conventional methods. Crucially, its computational load and memory footprint demonstrate lightweight characteristics suitable for direct terminal deployment. The proposed solution thus provides a technically feasible and performant approach for online smart meter error monitoring, contributing substantially to enhanced power metrology. Furthermore, the methodology demonstrates extensible application potential across various power metering engineering scenarios, thereby amplifying its market value and implementation prospects.