論文主題：A Self-Adaptive Particle Swarm Optimization with Fitness Adjustment Parameters
基於目標函數績效之自適應粒子群最佳化演算法
指導教授：鄭辰仰副教授
摘要：
Particle swarm optimization (PSO) has been widely applied in solving optimization problems because of its simple execution with fast convergence and high solution quality as known. Previous research has observed the effect of PSO parameters on a particle’s movement and the solution performance. Most PSO variants constructed the search strategy of evolution by controlling the movement pace to be larger in early evolution for exploration and become smaller in the late evolution for exploitation. However, the movement pace is depended on three key parameters of PSO: the inertia weight (w), the acceleration of self-cognition (c1), and the acceleration of social cognition (c2). Therefore, this study proposes a novel PSO algorithm based on the fitness performance (PSOFAP) of each particle for rapid obtained suitable parameter values to converge to an approximate optimal solution. The experiment is verified through six benchmark problems and the results are compared with those of other PSO variants. Furthermore, a well-known nonparametric statistical analysis method, namely the Wilcoxon signed rank test, is applied to demonstrate the performance of the proposed PSOFAP algorithm. The results of the experiment and statistical analysis show that PSOFAP is effective in enhancing the convergence speed, increasing the solution quality, and accurately adapting the parameter value without performing parametric sensitivity analysis.
粒子群最佳化（PSO）演算法，因為具有高應用性、快速收斂與高品質的解決方案特點，已經廣泛應用於解決各式各的最佳化問題。以往研究為改善PSO的求解速度與品質，觀察PSO求解過程，其中三項參數對粒子運動和求解效能具有影響力。進而提出此論點:理想的粒子移動步距，在勘探的早期演化階段應該較大，而在後期的開採過程中，粒子的移動步距應相對較小。爾後，多數改良式PSO便通過控制粒子的移動步距，以構建粒子進化的搜索策略。因此，本研究提出了一種基於粒子適應度性能的新型改良式粒子群最佳化演算法 (PSOFAP)，使三項參數依著適應值的績效而自行調整至較佳的值，同時仍可快速收斂至近似最佳解。本研究進行十二個標竿問題的驗證，並將實驗結果和其他改良式PSO演算法比較與分析。接續，更進一步採用廣泛採用的非參數統計分析方法–Wilcoxon符號秩序檢定，證實本研究提出的PSOFAP演算法的性能。統計分析結果顯示PSOFAP在參數依目標函數適應值績效，仍可準確調整參數值，有效提高收斂速度，同時提高求解品質。
