本論文中我們提出了一種改良的粒子群演算法，名為維度經驗重心分享粒子群演算法 SDGPSO(Standard Dimension Searching with Center of Gravity PSO algorithm)，其特殊的單維度搜索機制，讓其必須擁有較一般粒子群演法不同的經驗分享機制，且在前期能夠有優秀且快速的收斂能力，化簡的計算程序也能減少計算消耗時間，並且在多數函數都能夠有優良的最終收斂值，但在後期跳脫區域最優解的能力還是不如標準型粒子群演算法SPSO，所以在本論文又對SDGPSO作一個新的改良融合命名為SDGPSO-MSP (Standard Dime- nsion Searching with Center of Gravity PSO algorithm Mixing SPSO)，其中本文提出的銜接機制讓SDGPSO-MSP擷取SDGPSO前期快速收斂的能力和SPSO 後期優秀的跳脫能力，讓其優缺能夠達到良好的互補。最後我們使用測試函數對SDGPSO 和SDGPSO-MSP作性能測試並且與幾個已提出的擁有優良校能粒子群演算法作比較。經由模擬結果顯示，本文所提出的單維搜索與經驗重心分享機制在銜接SPSO後整體測試都均具有優越的表現，並且SDGPSO-MSP表現出兩邊所擷取的優點甚至更加優良。 In this thesis we have presented an improved algorithm for Particle Swarm Optimization (PSO) named Standard Dimension Searching with Center of Gravity PSO algorithm (SDGPSO). The SDGPSO algorithm needs to have a special experience-sharing mechanism to coordinate with single-dimensional searching mechanism. These mechanisms cause the faster convergence ability in the pre-convergence and have less computing time than SPSO. However, The ability to escape local optimal solution is worse than SPSO in the post-convergence. Because of this shortcoming, we further proposed a hybrid version named SDGPSO-MSP which takes advantages of fast and excellent convergence ability from SDGPSO and obtaining the better convergent solution from SPSO. The performances of SDGPSO and SDGPSO-MSP are fairly demonstrated by applying sixteen benchmark problems and compared it with several popular PSO algorithm. The simulations of results show that our proposed methods are effective and gain better performance than other compared PSO algorithms.
