Efficient hybrid genetic algorithm for resource leveling via activity splitting

Resource leveling problem is an attractive field of research in project management. Traditionally, a basic assumption of this problem is that network activities could not be split. However, in real-world projects, some activities can be interrupted and resumed in different time intervals but activity splitting involves some cost. The main contribution of this paper lies in developing a practical algorithm for resource leveling in large-scale projects. A novel hybrid genetic algorithm is proposed to tackle multiple resource-leveling problems allowing activity splitting. The proposed genetic algorithm is equipped with a novel local search heuristic and a repair mechanism. To evaluate the performance of the algorithm, we have generated and solved a new set of network instances containing up to 5,000 activities with multiple resources. For small instances, we have extended and solved an existing mixed integer programming model to provide a basis for comparison. Computational results demonstrate that, for large networks, the proposed algorithm improves the leveling criterion at least by 76% over the early schedule solutions. A case study on a tunnel construction project has also been examined.