整合增長式自組織映射圖與遺傳演算法之發展與應用

The Development and Application of Integration of Growing Self-Organizing Map and Genetic Algorithm

王志峯(Chih-Fehn Wang)

指導教授 ： 郭人介

國立臺北科技大學/管理學院/工業工程與管理研究所/碩士(2008年)

摘要

近幾年來，集群分析（Clustering Analysis）已廣泛被應用於各個層面，如：商業和教育、社會科學、遺傳學、生物科學等，群集其主要目的就是利用群體中具有相同統計特性聚成同一群，使得同群之間的同質性較高，而不同群之間有顯著的差異。因此，本研究利用自組織映射圖（Self-Organizing Map, SOM）發展出一個兩階段群集分析方法。第一階段為增長式自組織映射圖（Growing Self-Organizing Map, GSOM），其將透過輸入資料本身之結構產生適合的網路拓樸，並透過傳統SOM之訓練來產生網路權重向量，而第二階段則再利用連續型遺傳演算法（Continuous Genetic Algorithm, CGA）找出全域最佳解。
為驗證本研究所提之集群分析方法，提出GA、GASOM（Genetic Algorithm Self-Organizing Map）和SOM+GA，首先採用4個已知群集分佈的基準資料集：Iris、Wine、Vowel及Glass來分析何種方法之網路效益為最佳。研究之結果顯示GASOM為最佳，不但可以加速收斂效果而且也可以獲得更佳的解。
此外，本研究亦透過GSOM與GASOM的結合，將其應用在電池評等分級上，結合實際的電池量測結果做一綜合評判，以作為電池自動化分級評等系統之基礎，提供電池相關業者在設計及製造上品質之提升及成本的降低。

關鍵字

自組織映射圖 ； 群集分析 ； 遺傳演算法 ； 增長式自組織映射圖

並列摘要

In recent years, clustering analysis has been widely applied in may areas, like engineering, management, and bioscience. The purpose is to segment the individuals with the same characteristics in the population into the same group. Thus, those belong to the same group are homogenous and have the same characteristics. On the hand, those belong to different groups are heterogonous and have different characteristics. Therefore, this study attempts to use Growing Self-Organizing Map (GSOM) with integration of Genetic Algorithm (GA) to accelerate searching global solution and reduce the phenomenon of falling into regional solution. The proposed Genetic Algorithm-Based GSOM (GASOM) is consisted of two stages. The first stage will determine the network topology using tradition GSOM while the weights are determined by using genetic algorithm with SOM operator in the second stage.
The proposed method is compared with other two clustering methods using four benchmark data sets, Iris, Wine, Vowel, and Glass. The simulation results indicate that GASOM not only has faster converging speed but also can find the better solution.
In practical application, the proposed method also has been successfully employed to grade Li-ion cells and characterize the quality inspection rules. The results may offer a great assistance to the battery manufacturers and it can provide for improving the quality and decreasing the costs of battery design and manufacturing.

並列關鍵字

Self-Organizing Map ； Clustering Analysis ； Genetic Algorithm ； Growing Self-Organizing Map

參考文獻

[6] Alahakoon, D., Halgamuge, S. K., & Srinivasan, B., “Dynamic self-organizing maps with controlled growth for knowledge discovery,” IEEE Transactions on Neural Networks, 11(3), pp.601-614, 2000.

[7] Bauer, H.U. and Villmann, T., “Growing a hypercubical output space in a self-organizing feature map,” IEEE Transactions on Neural Networks, 8(2), pp.218-226, 1997.

[8] Blackmore, J. and Miikkulainen, R., “Incremental grid growing: Encoding high-dimensional structure into a two-dimensional feature map” Technique Report: AI92-192. Austin, TX: The University of Texas at Austin, 1992.

[9] Chelouah, R. and Siarry, P., “A continuous genetic algorithm designed for the global optimization of multimodal functions,” Journal of Heuristics, Vol.6, pp.191-213, 2000.

[11] Fritzke, B., “Growing cell structures: a self-organizing network for unsupervised and supervised learning,” Neural Networks, 7(9), pp.1141-1160, 1994.