基于聚类分析和可视化的增强遗传算法—I.算法的引出、原理与分析

Cluster Analysis and Visualization Enhanced Genetic Algorithm—I.Eduction,Principle and Analysis

提出了一种基于数据可视化的聚类分析法(ClusterConstrainedMapping,CCM)和人机结合的增强遗传算法,可保证进化过程在可行域中进行,不需要任何惩罚函数参数,可有效地进行带约束问题的优化.

Genetic Algorithms (GA) based on penalty function methods have been the most popular approach to constrained optimization problems because of their simplicity and ease of implementation. But how to find appropriate penalty parameters needed to guide the search towards the constrained optimum in the penalty function approaches is very difficult. A new cluster analysis based on visualization is proposed to address the constrained optimization problems. First, a Cluster Constrained Mapping (CCM) method based on feed-forward Artificial Neural Network (ANN) is proposed for dimension-reduction mapping from the original n-D space to 2-D, conserving the cluster information in the reduced dimensional space. Then the agglomerative algorithm that works in 2-D space is called upon for cluster analysis. Its parameters are provided through visualization and subsequent interaction with the user. Finally, the cluster information is derived from 2-D back into n-D to obtain the feasible region knowledge in the original dimensions, which is used in the IGA. The enhanced GA, incorporating a new cluster analysis method through data visualization (CCM) and user interaction guarantees the process of evolution in feasible regions without requiring any penalty parameters.