摘要：将基于分解的多目标进化算法(Multi-objective Evolutionary Algorithm Based on Decomposition,MOEA/D)应用于工程优化问题时,由于各目标函数在数量级及量纲上的不同,需要对目标函数进行归一化处理.首先,采用一种自适应ε约束差分进化算法(εConstrained Differential Evolution,εDE)寻找各个目标在Pareto前沿上的最大值和最小值,利用这些值对各目标进行归一化处理;然后,用MOEA/D进行求解,并在算法中加入了自适应ε约束处理技术;最后,采用一个标准测试问题和一个焊接梁设计优化问题对该算法进行测试,并与其他两种归一化方法进行了比较.根据提出的方法,MOEA/D能对Pareto前沿的一端进行集中优化,因而能处理一些Pareto前沿两端难以优化的问题.

摘要：This paper attempts to set a unified scene for various linear time-invariant (LTI) control system design schemes, by transforming the existing concept of “computer-aided control system design” (CACSD) to novel “computer-automated control system design” (CAutoCSD). The first step towards this goal is to accommodate, under practical constraints, various design objectives that are desirable in both time and frequency domains. Such performance-prioritised unification is aimed at relieving practising engineers from having to select a particular control scheme and from sacrificing certain performance goals resulting from pre-commitment to such schemes. With recent progress in evolutionary computing based extra-numeric, multi-criterion search and optimisation techniques, such unification of LTI control schemes becomes feasible, analytical and practical, and the resultant designs can be creative. The techniques developed are applied to, and illustrated by, three design problems. The unified approach automatically provides an integrator for zero-steady state error in velocity control of a DC motor, and meets multiple objectives in the design of an LTI controller for a non-minimum phase plant and offers a high-performance LTI controller network for a non-linear chemical process.