Regression-based Online Anomaly Detection for Smart Grid Data

TL;DR

This paper presents a real-time anomaly detection system for smart grid data using supervised learning and statistical methods, leveraging Spark Streaming for scalability and prompt detection of unusual energy consumption behaviors.

Contribution

It introduces a Lambda-based system combining supervised learning and statistical techniques for scalable, real-time anomaly detection in smart meter data.

Findings

System is effective in detecting anomalies in real-time.

The approach is scalable to large data streams.

Empirical results demonstrate high detection accuracy.

Abstract

With the widely used smart meters in the energy sector, anomaly detection becomes a crucial mean to study the unusual consumption behaviors of customers, and to discover unexpected events of using energy promptly. Detecting consumption anomalies is, essentially, a real-time big data analytics problem, which does data mining on a large amount of parallel data streams from smart meters. In this paper, we propose a supervised learning and statistical-based anomaly detection method, and implement a Lambda system using the in-memory distributed computing framework, Spark and its extension Spark Streaming. The system supports not only iterative detection model refreshment from scalable data sets, but also real-time detection on scalable live data streams. This paper empirically evaluates the system and the detection algorithm, and the results show the effectiveness and the scalability of the proposed lambda detection system.