Causative Cyberattacks on Online Learning-Based Automated Demand Response Systems

Power utilities are adopting Automated Demand Response (ADR) to replace the costly fuel-fired generators and preempt congestion during peak electricity demand. Similarly, third-party (DR) aggregators leveraging controllable small-scale electrical loads provide on-demand grid support services utilities. Some have started employing Artificial Intelligence (AI) learn energy usage patterns of consumers use this knowledge design optimal DR incentives. Such AI frameworks open communication channels between utility/aggregator customers, which vulnerable causative data integrity cyberattacks. This paper explores vulnerabilities AI-based learning designs a data-driven attack strategy informed by collected from New York University (NYU) campus buildings. The case study demonstrates feasibility effects maliciously tampering with (i) real-time incentives, (ii) event sent (iii) responses customers