以關聯法則對於合作式粒子族群演算法訂立族群分類規則

Abstract

本論文主要目的在於訂立合作式粒子族群演算法的族群分類規則，以頻率項成長(FP-growth)找尋函數各維度間的關聯性，關聯性的強弱將決定族群分類規則是否成立。合理的族群分類規則可使演算法在效能上有良好的表現。因此對合作式粒子族群演算法來說，族群分類應為整個演化過程中相當重要的一部分。此外，使用最廣泛的粒子族群演算法及分割因子合作式粒子族群演算法，在使用上皆有其限制，兩者皆有適合解決的問題種類，但也有表現較差的部分，並無法以一種方法涵蓋所有最佳化的問題。本論文因族群分類規則具有彈性，且可根據維度關聯性訂立族群分類規則，可視問題不同做調整，在使用上可適用於各種問題類型。

The purpose of this thesis is to define the group-based evolution classification rules for cooperative particle swarm optimization (CPSO). FP-growth is adopted in this thesis to find the connectivity between each dimension, in which the strength of the connectivity determines whether the classification rules apply or not. Generally, reasonable classification rules cause the algorithm work effectively. Hence, for the algorithm, the group classification plays a very important role in the evolutionary process. Besides, there are both restrictions between the most widely-used particle swarm optimization (PSO) and the CPSO. These two algorithms both work well under certain optimization tasks, but weak under others. None of these algorithms can cope with all the optimization tasks. In this thesis, the group classification rules are more flexible. The classification rules defined in this thesis are determined according to the connectivity between each dimension, and they are adjustable under different tasks, which greatly enhance the applicability of proposed mining based CPSO.