A FRAMEWORK FOR MULTIRESOLUTION OPTIMIZATION IN A PARALLEL DISTRIBUTED ENVIRONMENT - SIMULATION OF HIERARCHICAL GAS

A novel framework for multi-resolution optimization methodology is dev eloped for parallel/distributed simulation environments. The architect ure is constructed with multiple clusters hierarchically arranged with each level solving different degrees of abstracted problems. Creation and deletion of clusters are executed dynamically during the search o peration. A higher level cluster evaluates a wider search space with l ow resolution, whereas a lower level cluster investigates a smaller se arch space which is more promising for containing the global optimum. Each cluster consists of a controller (expert system) and agents (GA), where the agents evaluate the parameters of the problem of variable s tructure which allocates more computing resources to promising search subspaces. This article describes the prototyping of the hierarchical distributed genetic algorithms (HDGA) in an object-oriented simulation environment and provides preliminary experimental results. The result s are promising, and many theoretical questions concerning robustness of the approach are raised for future research. (C) 1996 Academic Pres s, Inc.