# Inference of Tampered Smart Meters with Validations from Feeder-Level Power Injections

## TL;DR

This paper proposes a graph theory-based reconfiguration scheme to detect and localize tampered smart meters in electrical distribution networks by analyzing feeder-level power injections and anomalies.

## Contribution

It introduces a novel graph-based switching scheme incorporating distributed generators for efficient tampering detection and localization in smart meter networks.

## Key findings

- Effective localization of tampered meters demonstrated
- Improved detection speed with graph reconfiguration
- Accurate anomaly identification using probabilistic analysis

## Abstract

Tampering of metering infrastructure of an electrical distribution system can significantly cause customers' billing discrepancy. The large-scale deployment of smart meters may potentially be tampered by malware by propagating their agents to other IP-based meters. Such a possibility is to pivot through the physical perimeters of a smart meter. While this framework may help utilities to accurately energy consumption information on the regular basis, it is challenging to identify malicious meters when there is a large number of users that are exploited to vulnerability and kWh information being altered. This paper presents a reconfiguration switching scheme based on graph theory incorporating the concept of distributed generators to accelerate the anomaly localization process within an electrical distribution network. First, a data form transformation from a visualized grid topology to a graph with vertices and edges is presented. A conversion from the graph representation to machine recognized matrix representation is then performed. The connection of the grid topology is illustrated as an adjacency or incidence matrix for the following analysis. A switching procedure to change elements in the topological matrix is used to detect and localize the tampered node or cluster. The procedure has to meet the electrical and the temporary closed-loop operational constraints. The customer-level anomaly detection is then performed in accordance with probability derived from smart meter anomalies.