The mixing evolutionary algorithm-independent selection and allocation of trials

Abstract

When using an evolutionary algorithm to solve a problem involving building blocks, we have to grow the building blocks and then mix these building blocks to obtain the (optimal) solution. Finding a good balance between the growing and the mixing process is a prerequisite to get a reliable evolutionary algorithm. Different building blocks can have different probabilities of being mixed. Such differences can easily lead to a loss of the building blocks that are difficult to mix and as a result to premature convergence. By allocating a relatively large amount of trials to individuals that contain building blocks with a low mixing probability, we can prevent such effects. We developed the mixing evolutionary algorithm (mixEA) in which the allocation of trials is a more explicit procedure than in the standard evolutionary algorithms. Experiments indicate that the mixEA is a reliable optimizer on a set of building block problems that are difficult to handle with more traditional genetic algorithms. In the case that the global optimum is not found, the mixEA creates a small population containing a high concentration of building blocks.