| This thesis uses a non-traditional estimating tool, Neural Networks, to provide an effective cost-data management for highway projects and accordingly develops a realistic cost estimating model. In the present study, the characteristic factors that affect the cost of highway construction have been identified and actual cases of highway and bridge projects constructed in Newfoundland during the past five years have been used as the source of cost data. The structure of a Neural Network template has been formed on a spreadsheet and three different techniques, Backpropagation training, Simplex Optimization and Genetic Algorithms, have been utilized to determine the optimum Neural Networks model. The resulting optimum model has been coded on Microsoft Excel in a user-friendly program to predict the outcomes for new cases. In addition, the proposed model provides a methodology to account for uncertainty in the user's assessment of project factors by measuring the sensitivity of the model to changes in cost-related parameters. It also enables the user to re-optimize the model on new historical encounters and accordingly adapt the model to new environments. (Abstract shortened by UMI.). |
