A. Multi-Objective optimization
A. 多目标优化
Fig. 17 From: An integrated multi-objective optimization method with application to train crashworthiness design
图17:一种集成的多目标优化方法,应用于训练耐撞性设计
Assuming that there are M objective functions, the problem can be mathematically formulated as follows:
假设有 M 个目标函数,则该问题可以数学地表述如下:
where x is a vector of N variables, which can be both integer and continuous, defined in the search space Σ. The problem is generally subjected to equality and inequality constraints, which can be expressed, respectively, as:
其中 x 是 N 个变量的向量,可以是整数,也可以是连续变量,在搜索空间 Σ 中定义。该问题通常受到相等和不等式约束的影响,它们可以分别表示为:
For multi-objective optimization, the problem is thus to simultaneously minimize the components of a vector of functions (f=(f1, …, fM)T), which is subject to several constraints. The problem is typically not uniquely solvable, but a set of equally efficient alternative solutions is possible. These solutions, among all the found ones, constitute the Pareto-optimal front, when they are plotted in the objective space. When comparing individuals i and j, with objective vector functions fi and fj we affirm that i dominates j in the Pareto sense if and only if:
因此,对于多目标优化,问题在于同时最小化函数 (f=(f1, …, fM)T) 向量的分量,该向量受多个约束。该问题通常不是唯一可以解决的,但一组同样有效的替代解决方案是可能的。在所有找到的解中,当它们在目标空间中绘制时,它们构成了帕累托最优前沿。当使用客观向量函数 fi 比较个体 i 和 j 时, fj 我们确认 i 在帕累托意义上支配 j,当且仅当: