Cost Function Reduction Using Stability-Informed Bayesian Optimization for the Model Predictive Control of a Semi-Active Suspension System

Traditional Model Predictive Control (MPC) performance is very sensitive to the state-weighting matrix and the control-weighting , resulting in time-consuming and suboptimal results. This work introduces a smart control approach for semi-active suspension systems that combines MPC with Stability-informed Bayesian Optimization (SiBO). The Bayesian framework uses a neural network surrogate model to approximate the cost function, significantly reducing the number of iterations required for optimization. Using a quarter-car model with a Magnetorheological (MR) damper, the method tackles the challenge of non-linear damping by linearizing it and directly integrating it into the MPC cost function. This reduces the computational load and makes the control more suitable for real-time use. The proposed strategy optimizes the damping coefficient to balance ride comfort and stability. Compared to traditional methods, it delivers clear improvements. The system showed a 15% reduction in Root Mean Square (RMS) body acceleration and a 12% improvement in suspension travel over standard Proportional–Integral–Derivative (PID)-tuned damping. It also achieved faster convergence, reaching optimal performance in just 10 iterations. In contrast, Particle Swarm Optimization (PSO) and Genetic Algorithms (GAs) required more than 50 iterations to reach such results.