Anomaly detection frameworks for identifying energy theft and meter irregularities in smart grids / Yip Sook Chin

Yip, Sook Chin (2019) Anomaly detection frameworks for identifying energy theft and meter irregularities in smart grids / Yip Sook Chin. PhD thesis, Universiti Malaya.

	PDF (The Candidate's Agreement) Restricted to Repository staff only Download (195Kb)
Preview	PDF (Thesis PhD) Download (3769Kb) | Preview

Abstract

Non-technical losses including electricity theft and anomalies in meter readings are estimated to cost the utility providers losses of approximately $96 billion per annum globally. Although the implementation of smart grids offers technical and social advantages, the smart meters deployed in advanced metering infrastructure are susceptible to more sophisticated types of malicious attack as compared to conventional mechanical meters. To curb non-technical losses, utility providers are increasingly leveraging on real-time smart metering to identify theft and meter irregularities. In the first part of this study, a linear regression-based anomaly detection framework is put forward to study consumers’ energy utilization behavior and evaluate their anomaly coefficients to detect the localities of the compromised and defective smart meters. The main idea is to model the amount of stolen energy at a smart meter as an anomaly coefficient. Specifically, a zero-anomaly coefficient indicates a faithful meter while a non-zero one represents an anomalous/defective meter. However, some of the predicted elements of anomaly coefficient vector might show inaccurate values when energy theft/meter irregularities take place only during a certain period in a day. Thus, categorical variable and detection coefficient are introduced in the framework to identify the periods and localities of consumers’ malfeasance/faulty meters. By investigating the anomaly coefficients and detection coefficients, non-technical losses can be deduced whether they occur either all the time or only during a certain period in a day. However, the linear regression-based framework assume that power line losses are known. Therefore, in the second part of this study, the assumption of known power line losses is relaxed, and a new anomaly detection framework is designed to take into consideration the impact caused by technical losses and measurement noise. Similarly, the goal is to identify anomalous consumption patterns within the billing reports transmitted to utility provider by evaluating consumers’ anomaly coefficients. To improve detection accuracy and reduce false positives, metrics known as loss factor and error term are introduced. Linear programming is utilized to solve for anomaly coefficients and loss factors by minimizing the error terms. The linear programming-based anomaly detection framework can still detect irregularities in meter readings regardless of whether non-technical losses occur all the time or at varying rates during intermittent intervals in a day. In addition, it can estimate the percentage of technical losses based on measurements at the data collector and the knowledge of the distribution network. To evaluate the performance of the proposed frameworks, a diverse set of non-technical loss attack functions is investigated and generated such that the experiments are closely related to the possible real-world energy fraud/meter irregularities scenarios. Subsequently, an advanced metering infrastructure test rig is constructed in the laboratory to validate the reliability and performance of both anomaly detection frameworks. Results from simulations and test rig show that both anomaly detection frameworks can reveal the amount of under-reporting/over-reporting by smart meters based on a small volume of consumers’ energy consumption data samples regardless of the type of consumer, thereby reducing loss incurred.

Actions (For repository staff only : Login required)