A reinforcement learning and H∞ hybrid control framework for 0–500 Hz full-band vibration suppression in active suspension systems

Abstract

With the rapid development of intelligent vehicles, active suspension systems have become crucial for enhancing ride comfort and handling stability. However, most existing control strategies focus on low-frequency vibrations below 20 Hz while neglecting high-frequency oscillations induced by discrete control signals and hardware delays. These high-frequency vibrations can degrade system performance and compromise operational safety. To address this issue, this study proposes a dual-layer control framework integrating mixed-sensitivity H∞ robust control with a reinforcement learning (RL)-based adaptive compensator. The robust controller ensures system stability and accelerates policy convergence, whereas the RL agent provides an adaptive weighting policy that complements the robust controller in real time. A safety supervision mechanism is also incorporated to monitor control actions and override potentially unsafe outputs by reverting to robust controller commands when necessary. Furthermore, a novel 0–500 Hz vibration evaluation framework is developed to comprehensively assess suspension performance, covering conventional suspension control metrics and vibration responses within the in-cabin structure-borne noise frequency band (50–500 Hz). Simulation results on a quarter-car model with a rigid ring tire and a strut-top-mount bushing indicate that the proposed RL-H∞ method achieves a 10.42 % reduction in body acceleration from 0 to 20 Hz. Compared with pure RL-based control, it achieves a 20.11 % improvement in vibration suppression within the 50–500 Hz band. Under complex and varying road excitations, the proposed RL-H∞ controller consistently demonstrates robust vibration suppression across the 0–500 Hz frequency range. By addressing vibration responses in the in-cabin acoustic control band, this study provides a solid theoretical and methodological basis for the integrated optimization of suspension control and in-cabin road noise mitigation in intelligent vehicles.