四季循環演算法於全域最佳化問題之應用

Season-Cycling Algorithm for Global Optimizations

本文源於自然四季循環，農業社會耕值與日日新的生活模式之啟示，發展四季循環演算法，採用粒子群演算法（PSO）處理粒子群的進化資訊及鑑定搜尋空間的機制。於每次演化後，進行粒子群偏好統計分析，試圖找出於下一循環時，粒子群的最佳搜尋空間。透過四個30維／100維的函數最佳化問題的測試，由搜尋結果顯示，本文四季循環演算法已有相當高的搜尋效果與穩定性；同時驗證改善PSO搜尋性能的另一種新思維，並提供族群式進化演算法於鑑定搜尋空間資料之系統架構。 An innovative season-cycling algorithm far global optimizations has been developed based on each season's lifestyle in agricultural society, seeding in spring, cultivating in summer, harvesting in autumn, and storing in winter. The population-based algorithm, particle swarm optimization (PSO), is used in the proposed algorithm to generate swarm particles and simultaneously provides particle's evolutional information. The particles data in each evolution have been analyzed statistically to End their evolution tendency and then identify the most promising region with the possible global optimal solution. Four benchmark problems with 30/100 dimensions were tested. The results demonstrated the novel season-cycling algorithm proposed in this study as an optimization algorithm with high efficiency and reliability. The proposed algorithm not only introduced a new thinking on improving the PSO performance but offered the structure of identifying the promising region while population-based algorithms are applied.