论文部分内容阅读
在给定Pareto解附近，用神经网络建立了Pareto曲面的近似模型，以探索新的Pareto解。在给定Pareto解附近随机产生一组Pareto解，利用可视化工具将它们展示给设计者，并用定性和定量相结合的方法评定它们的分值。利用神经网络，建立Pareto解到评分值的映射，以表达设计者在给定Pareto解附近的局部偏好。然后用遗传算法进行优化，找到最佳符合设计者偏好的Pareto解。以这个Pareto解为期望点，求解折衷规划，从而得到最终的优化设计方案。
In the vicinity of a given Pareto solution, an approximate model of Pareto surface is established by neural network to explore the new Pareto solution. A set of Pareto solutions are randomly generated near a given Pareto solution, visualized to designers using visual tools, and their scores are assessed qualitatively and quantitatively. Using neural networks, a mapping of Pareto solutions to scoring values is constructed to express the designer’s local preferences near a given Pareto solution. Then use genetic algorithm to optimize and find the Pareto solution that best fits the designer’s preference. Taking this Pareto solution as the expectation point, the compromise planning is solved to get the final optimization design.