DESIGNING OPTIMAL TRANSPORTATION NETWORKS: A KNOWLEDGE-BASED COMPUTER-AIDED MULTICRITERIA APPROACH

The dissertation investigates the applicability of using knowledge-based expert systems (KBES) approach to solve the single-mode (automobile), fixed-demand, discrete, multicriteria, equilibrium transportation network design problem. Previous works on this problem has found that mathematical programming methods perform well on small networks with only one objective. Needed is a solution technique that can be used on large networks having multiple, conflicting criteria with different relative importance weights. The KBES approach developed in this dissertation represents a new way to solve network design problems.; The development of an expert system involves three major tasks: knowledge acquisition, knowledge representation and testing. For knowledge acquisition, a computer-aided network design/evaluation model (UFOS) was developed to explore the design space. This study is limited to the problem of designing an optimal transportation network by adding and deleting capacity increments to/from any link in the network. Three weighted criteria were adopted for use in evaluating each design alternative: cost, average V/C ratio and average travel time. The best nondominated design is determined by a multicriteria evaluation technique called concordance analysis. A design exercise was conducted by a group of students. They were asked to find a set of link capacity changes that would improve the overall performance of each design. The results were carefully examined and transformed into facts and rules for the knowledge-base of the network design expert system (EXPERT-UFOS). It has two phases of analysis. The macro-level analysis recommends a total budget using the tradeoff functions between the criteria. The micro-level analysis provides advice about how to add or delete capacity on each link to avoid paradoxes. Test results show that EXPERT-UFOS found better designs with fewer design cycles than the students. EXPERT-UFOS may have enough simplicity to deal with large networks. The major findings of this dissertation suggest: first, the KBES approach is appropriate to tackle the computational complexities of the network design problem; second, the simulation-based knowlege acquisition method is useful for the engineering-oriented applications.