本論文提出一新的改良式基因演算法（KGA），此演算法透過由傳統基因演算法的結果搭配屬性的辨識去收集知識，並利用知識引導KGA的過程與交配時基因優良度的評估。此外，為了避免因為知識的應用而使演算過程容易落入區域最佳解，本研究利用突變的方式做區域搜尋，並在結果確定落入區域最佳解時，重新置換母體及替換舊有知識，藉由這些改變來使得本演算法可以同時兼顧集中性和多樣性。最後透過實驗的結果證實，本演算法確實是穩定的且可以找出不錯的排程，而知識的應用也可有效的提供引導的資訊。 This study presents a novel use of attribution for the extraction of knowledge from job shop scheduling problem. Our algorithm improves the traditional GA and using knowledge to keep the quality of solution. Based on the knowledge, the search space will be leaded to a better search space. In addition, this study uses mutation to do local search and refresh the knowledge and population when the solution fall into local minimum. Based on those methods, our algorithm will have the intensification and diversification. Those can make the algorithm have good convergence and leap for the search space to find the better solution. The experiment results show that algorithm steadily and can find the approximate optimal solution. And the knowledge is useful in provide the gene selection information.
