Smart Meter Condition Monitoring in a Metropolitan City with Data Analytics and Machine Learning

Abstract

Smart Meters deployed in Advanced Metering Infrastructure (AMI) generate vast operational data that can be leveraged for condition monitoring in densely populated urban environments. This research presents a comprehensive Machine Learning (ML) approach to detect Smart Meter irregularities that addresses limitations of traditional rule-based expert systems through advanced data analytics. Using systematic feature engineering on multidimensional time-series data from over 10,000 Smart Meters across Hong Kong's metropolitan network, a binary classification framework utilizing Gradient Boosting is developed and achieves 93.8% precision on holdout testing, and demonstrated 100% detection rate for critical faults at operational threshold 0.7, thus considerably outperforming the existing expert rules system's 67.7% precision. This research addresses the inherent challenge of highly imbalanced datasets where faulty meters represent significantly less than 1% of the population, and provides a scalable solution for proactive Smart Meter asset management in metropolitan environments with over 2 million deployed units.

Date of Award	2026
Original language	English
Awarding Institution	The Hong Kong University of Science and Technology
Supervisor	Bertram Emil SHI (Supervisor), Kam Tim WOO (Supervisor) & Ryan Ka Lun Lam (Supervisor)