MICROCOMPUTER-BASED VEHICLE ROUTING AND SCHEDULING

THE INVESTIGTION HAS THE FOLLOWING OBJECTIVES: 1) ESTABLISH THAT MICROCOMPUTERS CAN QUICKLY PROVIDE HIGH QUALITY SOLUTIONS TO LARGE-SCALE VEHICLE ROUTING AND SCHEDULING PROBLEMS, 2) DEMONSTRATE THAT NEURAL NETWORKS AND GENETIC SEARCH CAN WORK SYNERGISTICALLY WITH HEURISTIC MATHEMATICAL ALGORITHMS TO PROVIDE SUPERIOR SOLUTIONS TO VEHICLE ROUTING PROBLEMS. THE METHODS BUILD UPON THE PROVEN STRENGTHS OF GENERALIZED ASSIGNMENT ALGORITHMS FOR VEHICLE ROUTING, AS WELL AS THE NEURAL NETWORK AND GENETIC SEARCH PARADIGMS THAT ARE NEW TO THE VEHICLE ROUTING AND SCHEDULING PROBLEM SOLVING DOMAIN. IN ESSENCE, THE NEURAL NETWORKS ACT AS A KNOWLEDGE SOURCE THAT ASSISTS IN MODEL AND ALGORITHM SELECTION AND IN SPECIFYING PARAMETERS FOR THE MATHEMATICAL METHODS. THE GENETIC SEARCH PROVIDES A DYNAMIC CAPABILITY, ALLOWING THE HEURISTIC SOLUTION PROCEDURE OF THE MATHEMATICAL MODEL TO BE ADAPTIVELY STEERED IN RESPONSE TO PARTIAL SOLUTIONS THAT ARE COMPUTED. PRELIMINARY TESTING SUGGESTS THAT SOLUTION QUALITY IS SIGNIFICANTLY BETTER THAN WHAT ANY OF THE UNDERLYING MATHEMATICAL METHODS WORKING INDIVIDUALLY CAN ACHIEVE. THE PRIMARY ISSUES TO BE RESOLVED INCLUDE THE TOPOLOGY OF THE NEURAL NETWORKS, THE PARAMETERS AND DEGREES OF FREEDOM OF THE GENETIC ALGORITHMS, AND SOME DETAILS OF THE UNDERLYING HEURISTIC BEING CONTROLLED.