Investigation of Vulnerabilities of Machine Learning Models to Adversarial Attacks in Smart Grid Forecasting Systems

Abstract

This article examines how the use of machine learning in Smart Grid forecasting turns predictive models into targets of cyberattacks. Small, carefully chosen distortions of telemetry, training data, or federated updates can shift forecasts of load, renewable generation, reserves, and grid stability without clear signs of failure. The review systematizes adversarial attack scenarios by model access, impact goal, lifecycle stage, and implementation level in digital infrastructure. It also highlights what makes energy time series different from images and ordinary tabular data: temporal structure, cross-channel consistency, and physical plausibility must be preserved. The paper analyzes resilience measures — anomaly detection, adversarial training, secure aggregation, physics-informed hybrids, and data integrity monitoring — with attention to computational cost and real-time use. Further research should focus on realistic black-box attacks, cascading effects, and engineering criteria for trusting predictive services.

More by Mekhriddin Nurmamatov

• Lean Quantum Intelligence: A Compact Variational-Circuit Hybrid for Turbofan RUL Prediction with SHAP AttributionComputers · 2026
• Capacity-Constrained Multi-robot Task Allocation via Quantum Approximate Optimisation: A Reproducible Benchmark on a Solomon VRPTW C101-Derived InstanceCommunications in computer and information science · 2026
• Forecasting Electrical Load and Renewable Energy Generation Using Machine Learning MethodsAdvances in computational intelligence and robotics book series · 2026
• Explainable Variational Quantum Policy Gradient: SHAP-Based Interpretability of Quantum Reinforcement Learning Agents on CartPoleCommunications in computer and information science · 2026