A Model for Updating Project S-curve by Using Neural Networks and Matching Progress

The S-curve is commonly used for project planning and control, but the traditional schedule-based method for estimating S-curves is not always accurate, so many alternative empirical models have been suggested as an aid. Using a polynomial function for generalizing S-curves, an earlier research developed a neural network model that maps the function parameters from project attributes for obtaining a preliminary S-curve estimate. To produce a subsequent S-curve estimate during construction, this research first adopts the concept of case-based reasoning and proposes a progress-matching method based entirely on matching actual progress so far against S-curves of historical cases and using similar ones for estimation. Then, to improve it, a model combining the preliminary estimate from the neural network model and the subsequent estimate from the progress-matching method is proposed. The two estimates are assigned gradually decreasing and increasing weights, respectively, in successive S-curve updates as the percent project time increases. Tests found that using such an integration model can produce accurate S-curves in the beginning or middle of a project, indicating that it can be used to help the schedule-based method in project control during construction.