以混合演化式演算法處理最佳化多目標合成氣體問題

A Hybrid Evolutionary Algorithm for Multi - Objective Optimization of Synthesis Gas Production

在化學工業中，常因為不同需求條件所以聯合使用數種反應模型來生產合成氣體。本研究主要運用整合Nelder-Mead單體法與粒子群體最佳化演算法（particle swarm optimization, PSO），簡稱NM-PSO來解決合成氣體的多目標問題，希望藉由合併兩種演算法之優點，並加上常見的資料暫存概念來保留每次演算過程中所找到的Pareto解。採用相同測量方法，比較NM-PSO所求出的Pareto解與PSO的解兩者之間解的差異性及收斂性，並將實驗結果與其他文獻資料比較之後，本研究證明NM-PSO結果顯著的優於PSO的方法，因此證明了本研究演算法能有效的處理多目標最佳化問題並且能快速收斂到最佳解。

A combined process of gas reforming is often used in chemical industry for different demands. In this work, hybridized Nelder-Mead simplex method and particle swarm optimization algorithm is employed to solve multi-objective problems. Some effects were expected by hybrid strengths and weaknesses of the two algorithms. The use of archive controller keeps each Pareto solution found during computing. By using the same measurement method, it was shown that hybrid evolutionary algorithms outperform general evolutionary algorithms. In addition, the experimental results compared favorably with those found in the literature in terms of the degree of convergence and the dispersion of particles. This study demonstrates that the hybrid method is superior to PSO, and that the hybrid algorithm can effectively handle multi-objective optimization problems.