Electricity Consumption Data Quality Improvement Based on Low-Rank Matrix Decomposition

Abstract

Measurement data from smart meters can be affected by issues such as missing values and noise due to communication problems, power fluctuations, or interference. These anomalies reduce the reliability and utility of electricity consumption data, thereby hindering the efficient operation of power systems. To address the issues of missing data and noisy anomalies in electricity consumption data, this paper proposes a low-rank matrix recovery model based on the truncated nuclear norm. By leveraging the low-rank property inherent in electricity consumption data, the truncated nuclear norm is used as an approximation of the matrix rank. Simultaneously, the L1- norm is employed to constrain the sparse noise, enabling the completion of the low-rank components and the separation of sparse components. Finally, a two-step iterative strategy based on the alternating direction method of multipliers (ADMM) is applied to solve the model. Comparative experiments conducted on real datasets against commonly used methods for missing data imputation demonstrate the effectiveness of the proposed approach.