Detecting Abnormal Power Usage In Smart Grids With Fusion Matrix And Clustering

The advancement of new technologies such as the Internet of Things (IoT) and big data has made smart grids a key focus for the future of power systems. However, the increased communication between various participants in smart grids raises the risk of privacy breaches. Additionally, malicious attacks involving the injection of abnormal power data can result in financial losses and security issues. This study introduces a two-part solution to tackle these challenges: a matrix completion-based privacy protection scheme and an unsupervised learning-based anomaly detection scheme. The privacy protection scheme addresses data privacy concerns between different trust domains in the smart grid by using matrix completion to fill in missing data. It also improves privacy by adding noise with statistical properties similar to the original data. To improve anomaly detection, an enhanced K-means clustering algorithm is employed to eliminate outliers and refine clustering accuracy. Experimental results demonstrate that the proposed algorithm achieves an accuracy of 0.9, an F1 score of 0.7, and a Bayesian detection rate of 0.61 — all surpassing the performance of other algorithms. The AUC (Area Under the Curve) value of 0.79 confirms the effectiveness of the proposed method in detecting irregular electricity consumption patterns.