EVOLVING CONSTRUCTION SCHEDULES UNDER RESOURCE CONSTRAINTS

Abstract

This research studies the problem of project scheduling under resource constraints and employs a evolutionary algorithm called genetic algorithm (GA) to solve it. Previously existing models were studied to identify likely difficulties encountered in such problems. It was realized that the mathematically-based models were theoretically sound but bigger and complicated problems proved arduous to solve due largely to the exponentially increasing number of schedules to consider. Models based on the use of heuristic rules were easy to apply and could solve bigger problems efficiently. However, the rules themselves lacked generality- no heuristic rule could guarantee its performance a priori. Moreover, the use of heuristic rules lacked a sound theoretical basis and the "optimality" of the solutions could not be judged. GAs are a class of computational algorithms that are based on nature's rule of evolution and adaptation. In working with a GA-based model, a population of complete solutions is maintained at any point of time; this is in contrast to the other traditional methods which build the complete solution piecewise. The GA identifies good portions of solutions present on different "individuals" in the population and combines them on to a new individual through its crossover operator. Such a computational framework allows the GA to be adept in ·solving weakly-defined problems. The thesis discusses three GA-based models for the resource scheduling problem, namely the GA-Scheduler, the Contracts model and the Time-cost model. The core model is the GA-Scheduler which can handle problems with a wide range of objectives, including resource allocation, resource leveling and fixed completion date problems. GA- Scheduler typically produces a family of equivalent solutions, wherever the best solution is not uniquely optimal. The Contracts model goes further in solving two interacting sub-problems simultaneously: the resource availability and a corresponding "optimal" schedule that minimizes the sum of idle resources are determined concurrently. The Time-cost model recognizes the additional leverage that a flexible activity rate provides in determining "optimal" resource schedules and incorporates the time-cost trade-off decision into the scheduling problem. Finally, the Back Analysis module shows how the GA-Scheduler can interact with other existing scheduling methods. Schedules obtained from the other methods can be introduced into the initial population to see if the GA-Scheduler can improve on such schedules. The GA-Scheduler itself can also be used as a bootstrap mechanism to provide good starting solutions.