The goal of the Project Management is to organise project schedules to complete projects before their completion dates, specified in their contract. When a project is beyond its completion date, organisations may lose the rewards from project completion as well as their organisational prestige. Project Management involves many uncertain factors such as unknown new project arrival dates and unreliable task duration predictions, which may affect project schedules that lead to delivery overruns. Successful Project Management could be done by considering these uncertainties. In this PhD study, we aim to create a more comprehensive model which considers a system where projects (of multiple types) arrive at random to the resource-constrained environment for which rewards for project delivery are impacted by fees for late project completion and tasks may complete sooner or later than expected task duration.
In this thesis, we considered two extensions of the resource-constrained multi-project scheduling problem (RCMPSP) in dynamic environments. RCMPSP requires scheduling tasks of multiple projects simultaneously using a pool of limited renewable resources, and its goal usually is the shortest make-span or the highest profit. The first extension of RCMPSP is the dynamic resource-constrained multi-project scheduling problem. Dynamic in this problem refers that new projects arrive randomly during the ongoing project execution, which disturbs the existing project scheduling plan. The second extension of RCMPSP is the dynamic and stochastic resource-constrained multi-project scheduling problem. Dynamic and stochastic represent that both random new projects arrivals and stochastic task durations. In these problems, we assumed that projects generate rewards at their completion; completions later than a due date cause tardiness costs, and we seek to maximise average profits per unit time or the expected discounted long-run profit. We model these problems as infinite-horizon discrete-time Markov decision processes.