A Novel Change Severity Detection Mechanism for the Dynamic 0/1 Knapsack Problem

Abstract. Evolutionary algorithms (EA) are commonly used for solving optimization problems that can be stationary or dynamic. From an EA viewpoint, dynamic problems differ from stationary problems, mainly because in a changing environment, the EA needs to be able to track the optima through generations. Performance of an EA in dynamic environments is improved if the EA is modified to address the special requirements of such environments. In this paper, a novel environment-quality measuring technique and a reliable change detection mechanism is proposed for the 0/1 knapsack problem. Possible improvements in an EA performance using the proposed methods are discussed. An example application is demonstrated through the hypermutation approach. It is shown that by using the environment quality information, the hypermutation technique becomes more adaptive to the nature of the change in the environment and its performance is increased. Promising preliminary results promote further study to extend the experiments to other EA approaches and problem domains.