Multi-objective optimization in modeling and control for rotary inverted pendulum

Abstract

This article shows the application of multi-objective optimization techniques, both for the identification of parameters of a nonlinear model and for the adjustment of controllers. In particular, we propose a technique to identify the parameters of a first principles model of a rotational inverted pendulum (RIP) applying a methodology of multi-objective optimization and experimental data. Also the methodology extends to the tuning of PID and PI controllers for the mentioned system. For multiobjective optimization, an implementation based on evolutionary algorithms has been used, ev-MOGA (Herrero et al., 2007). For the analysis phase of the front solutions, we use the Pareto front visualization tool called Level Diagram (Blasco et al., 2017), which allows to successfully explore a set of Pareto optimal solutions and select one of them according to the preferences of the designer. The article does not try to study different control structures for the RIP system, but, given a control structure, analyze how to maximize the possibilities of the same to meet some conflicting objectives established by the designer. The advantage offered by this methodology is the easy understanding of the conflicts that appear among the design objectives, allowing to select a compromise solution according to the preferences of the designer, without losing sight of the set of optimal solutions found.