Safety-Comfort-Oriented Adaptive Safe Q-Learning for Networked Active Quarter-Vehicle Suspension Systems

The rapid development of the Internet of Things (IoT) facilitates data exchange and communication across numerous devices and sensors. As a cornerstone of intelligent vehicles, active quarter-vehicle suspension systems within IoT ecosystems critically enhance ride comfort and safety. This paper proposes a safe Q-learning (QL) algorithm based on a novel adaptive control barrier function (ACBF) to design a safe optimal controller. Periodically optimize policies in the cloud, download them, deploy them on edge devices, and implement millisecond-level correspondence in the edge processing layer. In the edge computing layer, the controller guarantees that constrained states of nonlinear systems remain strictly within a safe region while simultaneously achieving optimal performance. First, the ACBF with adaptive factor under the current state is introduced. Then, the Q-function is designed, integrating a positive-definite ACBF with the current state. This integration ensures the concurrent enforcement of safety and optimality constraints. This algorithm utilizes data collected within the safe region—rather than a mathematical model—to seek safe optimal policies. Third, rigorous theoretical guarantees are established for the monotonicity, safety, and optimality of the proposed algorithm. Furthermore, a critic neutral network and an action neutral network with provable approximation capabilities are established to facilitate the execution of the ACBF-based safe Q-learning (ACBF-SQL) algorithm. Finally, hardware-in-the-loop (HIL) simulation experiments validate the effectiveness and superiority of ACBF-SQL.