An evolutionary approach for resource constrained project scheduling with uncertain changes

In Resource Constrained Project Scheduling Problems (RCPSPs), it is usually assumed that the activity durations are known and integers. This assumption helps to conveniently develop a standard mathematical model, using discrete time steps. However, in reality, activity durations may not only be integer, and they may not be known with certainty at the time of project planning. The consideration of real-valued activity durations would increase the complexity in modelling of RCPSPs. In this paper, we consider that activity duration can be either integer or real-valued or both, and they are uncertain. To solve the optimization problem with uncertainty, scenario-based approaches are a popular choice. However, such a solution method is computationally very expensive. Therefore, in this research, we propose a simulation assisted evolutionary framework, that consists of two multi-operator based EAs and two heuristics to deal with the optimization process, and a simulation approach to deal with the uncertainty components. In the simulation, a range of problem instances is evaluated that are generated based on uncertain durations. The framework also proposes a new strategy to reduce the number of simulation runs. In the approach, the solution representation is different from the one required in the mathematical programming approach for RCPSP, and it does not require any discretization of the time periods. More than 1600 test problems, including some industrial problems, with up to 120 activities, have been solved using this proposed approach and the results have been compared with a set of state-of-the-art algorithms. The results obtained by the proposed approach were found to be of acceptable quality with a significant reduction of computational time. (C) 2020 Elsevier Ltd. All rights reserved.

Automated summary

This study considers the uncertainty of activity durations in Resource Constrained Project Scheduling Problems and proposes a simulation-assisted evolutionary framework to solve the optimization problem. A range of problem instances generated based on uncertain durations is evaluated using simulation, with a new strategy proposed to reduce the number of simulation runs.