Development of Hybrid Genetic Algorithms for Solving Large Parameter Optimization Problems

發展混合型基因演算法求解大型參數最佳化問題

劉振隆(Jenn-Long Liu)

《立德學報》 1卷2期 (2004/06) Pp. 79-91

摘要

本文提出三種基因演算法分別與一種以梯度為基礎之最佳化數值法來組合建構成三種混合型基因演算法，並且評價這些方法對於求解大型參數最佳問題的計算效能。混合型基因演算法的演算架構為先執行基因演算法特定演化代數後，提供良好的初始條件給區域性最佳數值法來使用，如此可以達到快速收斂以及得到較高的數值準確度。本文使用之基因演算法分別為簡單式基因演算法、微基因演算法以及一種使用田口直交表建構之智慧型基因演算法，而區域性最佳化數值法為在既定的目標函數下朝向可行與有用的方向去搜尋最小值。應用這三種混合型基因演算法於幾種多極值函數測試例，可檢視這些方法對於求解全域最佳值的能力與效率，近一步地，本文提出之混合型基因演算法亦應用於大型參數最佳化問題並評定這三種數值法之收斂率與數值準確性。

關鍵字

混合型基因演算法 ； 梯度為基礎之最佳化數值法 ； 田口直交表

並列摘要

This study presents three hybrid genetic algorithms (GAs), which are constructed from a combination of three types of GA and a gradient based optimization algorithm, and evaluates their performance from the numerical computations of large parameter optimization problems (LPOPs). Hybrid GAs, perform genetic algorithms with specified generations to provide good initial conditions for a local optimization. They can be used to achieve fast convergence and high numerical accuracy. The original GAs used here are simple GA, micro-GA and an intelligent GA using Taguchi's orthogonal arrays (OAs), The local optimization algorithm searches both feasible and usable directions to find a minimum for a specified objective function. The local optimization algorithm offers a wide variety of state-of-the-art optimization algorithms for nonlinear constrained function minimization. In this work, the hybrid GAs are applied to several multi-modal test functions to examine the ability and efficiency in finding the global optima. Furthermore, the proposed hybrid GAs are applied to LPOPs to evaluate the algorithmic convergence rate and numerical accuracy.

並列關鍵字

Hybrid Gas ； gradient-based optimization algorithm ； Taguchi OAs