A Robust HIDRL-ISAC Framework for Energy-Efficient Optimization in Reconfigurable Intelligent Surface-Assisted Wireless Systems

A Reconfigurable Intelligent Surface (RIS) is a promising technology for enhancing the spectral and energy efficiency (EE) of next-generation wireless networks. Adapting to dynamic user mobility while jointly optimizing beamforming, phase shifting, and power control remains a challenge. This paper proposes a novelHierarchical Intelligent Deep Reinforcement Learning with Integrated Sensing and Communication (HIDRL-ISAC)framework for real-time, energy-efficient optimization in RIS-assisted networks. Our method features cooperative dual agents: a master agent that optimizes the base station (BS) beamforming and RIS phases based on the sum rate and a slave agent that adjusts the transmit power in response to ISAC-based user location feedback. The framework leverages an Extended Kalman Filter (EKF) for robust position tracking using ISAC sensing data. HIDRL-ISAC effectively addresses the non-convexity and temporal variability of the optimization problem by adapting to channel fluctuations and mobility patterns in real time. The simulation results demonstrated significant improvements in EE and convergence, establishing HIDRL-ISAC as a resilient and scalable solution for mobility-aware RIS-assisted wireless systems.