A Unified Co-Design Framework Integrating Predictive AI Control and CyberResilience for Autonomous Microgrid Operation

Abstract

The increasing integration of renewable energy and distributed resources makes accurate forecasting essential for reliable and efficient microgrid energy management; however, existing TFT-based forecasting studies primarily focus on prediction and lack direct integration with predictive control and cyber-resilient operation. Therefore, this study proposes a TFT-MPC-CR framework that combines Temporal Fusion Transformer (TFT) forecasting, Model Predictive Control (MPC), AI-based False Data Injection (FDI) detection, and adaptive resilient control. The framework is implemented in Python using PyTorch, with the Microgrid PV-EV Charging Dataset obtained from Kaggle, where PV generation, load demand, battery SOC, EV charging, and grid variables are used for learning. TFT predicts future PV generation and energy demand, and MPC optimizes battery, EV, and grid power allocation, while the cybersecurity layer detects manipulated measurements and activates resilient MPC. As a target experimental outcome, the proposed framework is designed to achieve at least 5–10% lower forecasting error than conventional TFT and improve operational resilience under FDI attacks. The framework is expected to provide accurate prediction, efficient energy management, rapid attack response, and secure autonomous microgrid operation.