Introduction 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. From the design of jet engines to the scheduling of airline crews, EAs, in their various forms, are routinely solving a multitude of complex, multi-dimensional, multi-modal optimization problems. But what happens if the information that has been provided to the EA changes? Despite the obviously successful application of evolutionary techniques to complex problems, the resultant solutions are often fragile, and prone to failure when subjected to even minor changes in the problem. Many practical engineering, economic, and information technology problems require systems that adapt to changes over time. Examples of problems where environmental changes could cause the fitness landscape to be dynamic include: target recognition, where the sensor performance varies based on environmental conditions; scheduling problems, where available resources vary over time; financial trading models, where market conditions can change abruptly; investment portfolio evaluation, where the assessment of investment risk varies over time; and data mining, where the contents of the database are continuously updated. Our motivation for designing EAs for solving dynamic problems is simple. With the continuing increases in available processing power, it is becoming computationally possible to assign an EA to continuously solve actual dynamic problems without the need for human intervention. This requires that the EA continuously provide a “satisfactory” level of performance when subjected to a dynamic fitness landscape, and we strive to ensure such performance. | |
|