Applying Genetic Algorithm to Multi-objective Combinatorial Optimization Problem- A Study on Traveling Salesman Probelm

• Main Authors: Wang, Jih-Chang, 王日昌
• Other Authors: Tzeng Gwo-Hshiung
• Format: Others
• Language: zh-TW
• Published: 1997
• Online Access: http://ndltd.ncl.edu.tw/handle/16034022778155562745

博士 === 國立交通大學 === 交通運輸研究所 === 85 === The most methods of Multiple Criteria Decision Making (MCDM) will cause acomputational burden tremendously in applying to the combinatorial optimization problem. Besides, the most methods of MCDM ignore the new-coming information and cannot deal the weight of decision maker dynamically. Using the parallel- searching, this research develops a Genetic Algorithm (GA) to overcome theseshortages. There are three primary parts in this research: developing a dynamic weight assessing method, proving the effectiveness of GA in the traveling salesman problem (TSP), developing a method to overcome the shortages above. Theimportant results are listed in the below.(1) To develop a weight assessingmethod by Habitual Domain.(2) To analyze the genetic Crossover operator in TSP,and to classify and implement 23 relative operators, and to propose 3 improving operators.(3) To develop the "Doubly-Nearest-Available neighborhood" crossover operator, and to compare its performance with others by 24 well- knownnetworks.(4) To propose the concept of template database for improving thequality of crossover.(5) To improve a Space- Filling Curve of Chaos Theoryas a quick procedure of tour initialization.(6) To propose a Genetic Algorithm to solve the TSP by combining Doubly-Nearest-Available neighborhoodand template database. And, to measure its performance by 35 networks that are from 48 to 657 nodes and are obtained from TSPLIB. The result is excellent: this method finds 27 exact solutions in 35 networks with only 0.03%error rate in average.(7) To combine the weight assessing method and theparallel-searching of GA, and propose a multiple-objectives GA. The mainproperties are (a) modifying weight dynamically with the results in searching process; (b) and reducing to computational requirement to the levelof single-objective; (c) and computing the weight directly.(8) To apply theabove method to the post-office in Taipei, and to demonstrate the detail by a numeric example.(9) To analyze and write the computer language ofmethods of combinatorial optimization problem and TSP, and to review thehundred''s reference of them. And to build a homepage in the Internet forsupplying the source code and relative inquiry service.