IEEJ Transactions on Electronics, Information and Systems
Online ISSN : 1348-8155
Print ISSN : 0385-4221
ISSN-L : 0385-4221
<Neural Network, Fuzzy and Chaos Systems>
Individual Evaluation Scheduling for Experiment-Based Evolutionary Multi-Objective Optimization
Hirotaka KajiHajime Kita
Author information
JOURNAL FREE ACCESS

2008 Volume 128 Issue 6 Pages 986-996

Details
Abstract

Since the pioneer work of Evolution Strategies, experiment-based optimization is one of the promising applications of evolutionary computation. Recent progress in automatic control and instrumentation provides us with a smart environment called Hardware In the Loop Simulation (HILS) for such application. However, since optimization through experiment has severe condition of limited evaluation time and fluctuation of observation, we have to develop methodologies that overcome these problems. This paper discusses application of Multi-Objective Evolutionary Algorithms (MOEAs) to Experiment-Based Multi-objective Optimization of parameters of dynamical systems. In such applications, we have to apply various parameter candidates spreading near the Pareto frontier to the system, and it causes fluctuation of the observed criteria due to the transient response by parameter switching. For reduction of loss time caused by such transient response in evaluation of criteria, we propose techniques called Evaluation Order Scheduling and Evaluation Time Scheduling. Numerical experiments using a formal test problem and experiment in a HILS environment for real internal-combustion engines have demonstrated the effectiveness of the proposed methods.

Content from these authors
© 2008 by the Institute of Electrical Engineers of Japan
Previous article Next article
feedback
Top