演化計算環境中個體間合作行為之研究

Abstract

傳統遺傳演算法中，染色體間並沒有「合作」的關係。每代都會挑選 出一些表現較好的染色體參與演化以產生下一代，最後只從中挑選一條出 來當作最佳解。換句話說，染色體間只有「競爭」的關係存在。而在模糊 推理系統中，常有多條規則被激發，在做解模糊化時，會依各規則的激發 度做整合性的處理，所以簡單來說，要產生正確的輸出，需要各規則間的 「合作」。本論文將資源有限的觀念引進傳統的遺傳演算法，以競爭資源 來刺激個體的進化。在本模型中，整個族群（環境）是由一些小群體組合 而成的。每個小群體都是一個模糊規則庫，而每個規則庫則是由一些規則 （染色體）組成。本模型的特色為：群體間的關係是競爭，群體內個體間 的關係則是合作，由此達到快速自動建立規則庫的目標。 In tranditional Genetic Algorithms, there is no cooperative relation between chromosomes. In each generation, we select those chromosomes with better performance for participating in evolution to produce the next generation. At last, we select one of those chromosomes as the best solution. On the other hand, only the relation of competition exists between chromosomes. In Fuzzy Inference Systems, when many rules are fired, we process all of these rules according to their firing strength during defuzzification. Simply speaking, we need cooperation between rules to generate correct outputs. In this thesis, we introduce the concept of restricted resource into the tranditional Genetic Algorithms. It stimulates the evolution of the individuals by competing resources. In our model, the whole group (the environment) is consisted of several small groups (the rulebases) which are the collections of some rules (the chromosomes). The primary characteristics of the model, the competition between groups and the cooperation within the group, help us to achieve the goal of fast and automatic construction of a rulebase.