Genetic Algorithms, With Inheritance, VersusGradient Optimizers, And GA/Gradient Hybrids
Genetic Algorithms, With Inheritance, VersusGradient Optimizers, And GA/Gradient Hybrids
复制标题
具有继承性的遗传算法、与梯度优化器以及 GA/梯度混合算法
DOI：
10.2495/op970251
复制
发表时间：
1970
期刊：
影响因子：
--
通讯作者：
J. Finckenor
中科院分区：
文献类型：
--
作者：
J. Finckenor
本研究比较梯度求解器，遗传算法（GA），和两种GA梯度混合。梯度优化方法可以快速收敛到最优解。然而，该解决方案通常是局部最优的，特别是如果函数是噪声的。局部最优可能远离全局最优并且取决于用户输入起始点。梯度求解器也无法执行整数优化。遗传算法是有效的，在寻找全球的解决方案，但需要许多功能评估。GA是一种整数优化器，当对连续函数进行操作时，可能会丢失梯度优化器可用的分辨率。第一个混合使用最终的GA解决方案作为梯度求解器的起点。第二个混合使用每个GA个体作为起点。具有代表性的问题是蒙皮-长桁结构圆筒。这是一个相当嘈杂的设计空间，有几个不连续性。结果将最终解决方案的权重与所需的函数调用进行比较。最有效的求解器是一个混合的GA选择起点和一个相当小的人口50-100个人。
This study compares gradient solvers, Genetic Algorithms (GA's), and two kinds of GAgradient hybrids. Gradient optimization methods can rapidly converge to an optimum solution. However, the solution is often a local optimum, particularly if the function is noisy. The local optimum may be far from the global optimum and depends on the user input starting point. Gradient solvers are also unable to perform integer optimization. GA's are effective at finding global solutions, but require many function evaluations. The GA is an integer optimizer and can lose resolution available to gradient optimizers when operating on a continuous function. The first hybrid uses the final GA solution as a starting point for the gradient solver. The second hybrid uses each GA individual as a starting point. The representative problem is a skin-stringer construction cylinder. It is a fairly noisy design space with several discontinuities. Results compare the weight of the final solution against the function calls required. The most efficient solver is a hybrid with the GA selecting starting points and a fairly small population of 50-100 individuals.