Skip to Main Content
A difficult application case of evolutionary algorithms is that in which individual fitness evaluations take several processor-minutes to a few processor-hours. The design of evolutionary algorithms with such expensive fitness evaluation differs substantially from the norm where fitness evaluation is rapid. In this paper we apply evolutionary algorithms to a thermal systems engineering design problem - the design of a biomas cook stove currently in use in Central America. Fitness evaluation involves the use of computational fluid dynamics (CFD) modeling of the flow of hot air and heat transport within the stove to equalize the surface temperature. The goal is to optimize the placement and size of baffles that deflect hot gasses underneath the cook top of the stove. Three techniques are used to permit evolutionary algorithm to function on this challenging problem using a population of relatively small size. First, computations are performed on a Linux cluster machine yielding a large, fixed performance increase. Second, the resolution of the mesh for CFD computations used a minimal; mesh that yields acceptable fidelity of CFD computations. Third, a diversity preserving technique called a graph based evolutionary algorithm (GBEA) is used to retain population diversity during evolution. A usable stove design, subsequently deployed in the field, was located by the evolutionary algorithm. In this paper we demonstrate that GBEAs preserve diversity on this baffle design problem and give evidence that highly connected graphs is a good choice for future work on analogous CFD problems. Diversity preservation is a function of both tournament size and the connectivity (geography) of the graph used.
Evolutionary Computation, 2004. CEC2004. Congress on (Volume:1 )
Date of Conference: 19-23 June 2004