Authors:
Hasina Rahman
1
;
Priyadarsi Nanda
1
;
Manoranjan Mohanty
1
and
Nazim Sheikh
2
Affiliations:
1
University of Technology Sydney, Sydney, NSW, Australia
;
2
Torrens University, Sydney, NSW, Australia
Keyword(s):
Smart Meter Security, Anomalies, Intrusion Detection System, False Data Injection, Deep Learning Model, Energy Data, LSTM-DAE Model.
Abstract:
Smart meters, intelligent devices used for managing energy consumption of consumers, are one of the integral components of the smart grid infrastructure. The smart metering infrastructure can facilitate a two-way communications through the Internet to leverage home energy management and remote meter reading by the service providers. As a consequence, the smart meters are extremely susceptible to various potential security threats, such as data tampering, distributed denial of services (DDoS) attack and spoofing attacks. In this paper, we put forward a scheme to detect anomalies in energy consumption data using real-world datasets. Thereby, addressing data tampering attacks. We have adapted an unsupervised machine learning method to distinguish the anomalous behaviour from the normal behaviour in energy consumption patterns of consumers. In addition, we have proposed a robust threshold mechanism for detecting abnormalities against noise, which has not been used in smart grids before.
Our proposed model shows an accuracy of 94.53% in detecting anomalous patterns in energy consumption data. This accuracy surpasses the existing benchmark in anomaly detection in energy consumption data using machine learning models (Huang and Xu, 2021).
(More)