Applying Learnable Evolution Model to Heat Exchanger Design

Abstract

A new approach to evolutionary computation, called Learnable Evolution Model (LEM), has been applied to the problem of optimizing tube structures of heat exchangers. In contrast to conventional Darwinian-type evolutionary computation algorithms that use various forms of mutation and/or recombination operators, LEM employs machine learning to guide the process of generating new individuals. A system, ISHED1, based on LEM, automatically searches for the highest capacity heat exchangers under given technical and environmental constraints. The results of experiments have been highly promising, often producing solutions exceeding the best human designs.