Global optimum search method for nonlinear system with multiple local optima using genetic algorithm

Abstract

The success of the process control for nonlinear multimodal adjustment systems mainly depends on the effective estimation of the initial parameter error that is the unknown internal static state of the system. This thesis proposes a new kind of static state estimator based on Genetic Algorithms (GAs) in order to accomplish an estimation-based search technique for nonlinear multimodal adjustment processes. To adapt GA to the estimation problem, the general procedure of the conventional GAs is modified in two aspects; first, the reproduction strategy uses two different criteria (survival cost and parent cost) for the elimination of weaker members and parent selection to carry out the estimation recursively, and secondly, a new genetic operator based on a priori knowledge of the system is developed to provide fast convergence characteristics to the estimation. As a proper controller for the adjustment system, a simple integral type controller is adopted in this thesis. Some variations of the overall control scheme are discussed to improve performance. The appropriateness and performance of the proposed estimator is investigated via several numerical examples of the multimodal adjustment system, and the results obtained show that this estimator performs very well even with the presence of a large amount of measurement noise and with the existence of the system modeling error. In addition, the proposed control scheme is adopted for the automatic control of VCR tuning process in this thesis. The experimental results show that the scheme can carry out the task very well.