Efficient Project Scheduling with Autonomous Learning Opportunities

We consider novel project scheduling problems in which the experience gained from completing selected activities can be used to accelerate subsequent activities. Given a set of potential learning opportunities, our model aims to identify the opportunities that result in a maximum reduction of the project makespan when scheduled in sequence. Accounting for the impact of such learning opportunities causes significant complications, due to the cyclic nature of the learning relations and their interference with the precedence network. We propose additive and subtractive algorithms that iteratively reschedule the project using an enhanced topological sorting algorithm. Learning opportunities are integrated, activated, and potentially deactivated in each step by maintaining the acyclicity of the combined precedence and learning network. To illustrate the challenges that arise in this setting, we first consider the special case where activities can learn from at most one other activity. Subsequently, we extend our approach to the general case that admits multiple learning opportunities. We show that our approaches guarantee the construction of an optimal solution in polynomial time. In a computational study using 340 small and large resource-unconstrained PSPlib instances, we analyze the model behavior under various scenarios of learning intensity and learning opportunity. We demonstrate that significant project speedups can be obtained when proactively accounting for learning opportunities.