Mining of Data from Evolutionary Algorithms for Improving Design Optimization

摘要

This paper focuses on integration of computational methods for design optimization based on data mining and knowledge discovery. We propose to use radial basis function neural networks to analyze the large database generated from evolutionary algorithms and to extract the cause-effect relationship, between the objective functions and the input design variables. The aim is to improve the optimization process by either reducing the computation cost or improving the optimal. Also, it is hoped to provide designers with the salient design pattern about the problem under consideration, from the physics-based simulations. The proposed technique is applied to both academic problems and real-world problems, including optimization of an airfoil and the turbopump of a cryogenic rocket engine. Our results demonstrate that these techniques can further improve the design already achieved by the evolutionary algorithms with a slightly additional cost.