Ensemble LOF-based detection of false data injection in smart grid demand response system

Abstract

Demand response (DR) systems are prone to false data injection attacks (FDIA), which present substantial economic and operational hazards. Notwithstanding their significance, DR systems are remarkably vulnerable to such vulnerabilities, highlighting the pressing need for improved anomaly detection. In this research, we provide a novel technique called Ensemble Local Outlier Factor (ELOF), specifically designed to identify FDIAs in DR systems. ELOF integrates many local outlier factors (LOF) models with tuned hyperparameters and diverse distance metrics to enhance the identification of harmful activity. By examining an extensive dataset provided by Pecan Street, which encompasses data from 168 houses in Austin, Texas, our methodology outperforms current benchmarks in terms of performance. We also examine the constraints of traditional anomaly detectors, including their susceptibility to noise and biases and their restricted ability to distinguish intricate data patterns. Our investigation uncovers an innovative FDIA vector that enables authorized users to initiate assaults without jeopardizing the integrity of the hardware or communication connections. In addition, we demonstrate how these vectors can be managed for economic benefit through various optimization tactics, highlighting an increasing and noteworthy danger to consumer-focused energy systems. Finally, we show how our model outperforms state-of-the-art detection methods in a smart grid environment, demonstrating our superior technique.