隨著人工智慧的發展以及人類對於自動推理的需要，現今電腦科學發展出了龐大的演算法系統，各種形式的演算法在不同領域自動完成麻煩且困難的任務。本論文所改良的演算法為最佳化演算法中多粒子搜尋裡的粒子群演算法，粒子群演算法透過模仿鳥類覓食的移動方式來進行最佳化搜尋，透過自身最佳解和群體最佳解的帶領，使粒子有效且快速的收斂到區域最佳解。我們希望透過簡單且重要的改良使現有的粒子群演算法達到更佳的效能，同時不會增加程式實作上的複雜度，因此提出了跳躍式粒子群演算法，此演算法結合了位置公式改良以及慣性權重設計，提升了粒子收斂時解的精準度，適合用於多粒子搜尋的實作當中，其中位置公式改良有P-PSO和G-PSO兩種版本，權重設計也分成自適應型慣性權重和常態分布累積遞減型慣性權重，使用者可以依照自身需求來決定要使用的組合，本論文也同時探討初速度的有無對於疊代時的影響，讓使用者在不同的情況下採用不同的初速度，最後，透過實驗模擬我們驗證了這些方法的效果及性能，在16種測試函數中，跳躍式粒子群演算法於大部分函數裡有著最佳表現，能夠讓使用者在實作及應用上達到良好的效能及發揮。;With the development of artificial intelligence and the need of automatic reasoning, humans have created a great number of algorithms in computer science. Various forms of algorithms complete troublesome and difficult tasks automatically in different fields. The improved algorithm in the thesis is the particle swarm optimization in the multi-particle search in the optimization algorithm. Particle swarm optimization uses a mobile approach that mimics bird foraging to perform optimal searches. Through the guiding of the particle best solutions and the group best solution, the particles can effectively and quickly converge to a local optimum. We modify the existing particle swarm optimization algorithm to achieve a better performance through simple and important improvements without increasing the complexity of the programming implementation. An improved algorithm called Jumping Particle Swarm Optimization based on adaptive inertia weight is proposed. This particle swarm optimization method combines position formula improvements and inertia weight design to improve the accuracy of the solution. It is suitable for the implementation of multi-particle searches. There are two versions of position formula improvements: P-PSO and G-PSO. The weight design also has adaptive inertia weight and normal distribution cumulative decreasing inertia weight. Users can decide the combination to be used according to their own needs. The thesis also discusses the influence of the initial velocity on the iterations, allowing users to use different initial velocities in different situations. Finally, through the experimental simulation, we verify the effect and performance of the proposed methods. Among the 16 test functions, the jumping particle swarm algorithm has the best performance in most functions, which enables users to achieve an excellent performance in implementation and application.
