Preference-driven parameter tuning of model predictive controllers

• Main Author: Xu, Zekai
• Other Authors: Ling Keck Voon
• Format: Thesis-Master by Coursework
• Language: English
• Published: Nanyang Technological University 2024
• Subjects: Engineering
• Online Access: https://hdl.handle.net/10356/173682
• Institution: Nanyang Technological University

Model Predictive Control (MPC) has become a cornerstone in numerous applications, ranging from industrial processes to autonomous systems. However, a key challenge lies in the development of a generalized and proceduralized method for implementing MPC strategies across diverse scenarios. Traditional trial-and-error tuning of MPC parameters is often time-consuming and inefficient. Addressing this, this dissertation introduces a novel approach to MPC parameter tuning, utilizing the recently developed global optimization algorithm, GLISp (preference-based GLobal minimum using Inverse distance weighting and Surrogate radial basis functions). Distinctive in its methodology, GLISp iteratively constructs an acquisition function based on preference information obtained from sampled candidates within the parameter search space. This approach is designed to efficiently tackle global optimization challenges within constrained steps and resources. Simulations were conducted to evaluate the efficacy of the GLISp algorithm in various MPC applications, benchmarking its performance against traditional non-preference-based algorithms. These comparative analyses shed light on the relative effectiveness of the algorithms in handling different nonlinear control scenarios. Moreover, this research offers generalized guidelines for implementing the tuning algorithm, paving the way for more streamlined and effective MPC tuning processes. The findings and methodologies presented in this dissertation hold valuable implications for advancing MPC tuning strategies, contributing insights to the field of control systems engineering.