Designing evolutionary algorithms for dynamic environments

The robust capability of EAs to find solutions to difficult problems has permitted them to become the optimization and search techniques of choice by many industries. Despite the success of evolutionary techniques, the resultant solutions are often fragile and prone to failure when the problem changes. Many real problems in engineering, economics, and information technology require systems that can adapt to changes over time. This research provides an analysis of what an EA needs to do to solve dynamic problems, focusing on detecting changes in the problem environment and responding to those changes. In the course of this research we identify and quantify a key attribute needed to improve the detection and response performance of EAs in dynamic environments. We then create an enhanced evolutionary algorithm, designed explicitly to exploit this new understanding. This enhanced EA is shown to have superior performance on some types of problems. Our experiments evaluating this enhanced EA indicate some previously unknown relationships between performance and diversity that may lead to general methods for improving EAs in dynamic environments. Along the way, several other important design issues are addressed involving computational efficiency; performance measurement, and the testing of EAs in dynamic environments.