The Research On Multi-Skilled Project Scheduling And Robust Optimization Methods

The multi-skilled resources are common in modern projects,such as multifunctional digital controlled lathes,multi-skilled employees,etc.These multi-skilled resources can perform different types of tasks in the project.Compared with the traditional project scheduling problem,not only does the project scheduling problem involving multi-skilled resources need to assign resources to activities,but also it needs to determine the skills that the resources perform in these activities,which makes it a complicated problem.In addition,modern projects are also faced with many uncertain factors,such as changes in the natural environment and social environment,and the instability of resources within the project.These uncertainties can make project execution deviate from the pre-established schedule,incurring additional costs.Therefore,this dissertation studies the multi-skilled project scheduling problem in a deterministic environment and the methods of robust optimization of multi-skilled project schedule in an uncertain environment.Firstly,two scheduling algorithms for the multi-skilled project in a deterministic environment are designed.A heuristic algorithm based on dynamic resource priority is proposed to solve the problem quickly.During the scheduling process,the algorithm dynamically updates the resource assignment rules according to the resource attributes and the demand for skills of the activities to be scheduled.Compared with the traditional heuristic algorithm,the proposed heuristic algorithm can allocate resources more reasonably and reduce the project duration.Furthermore,a general paradigm of applying a reinforcement learning algorithm to project scheduling problems is proposed,based on which a reinforcement learning-based multi-skilled project scheduling algorithm is developed for better solutions.The methods of converting the Markov chain and designing the state,the action,and the reward are described,as well as the updating strategy of the value function.The experimental results on a single project instance and a project instance library show that both perform better than traditional ones.Secondly,two time-based robust project schedule optimization algorithms are proposed considering the uncertain factors in the project.The defects of the existing robust scheduling algorithms are analyzed,based on which an improved decentralized time buffer setting algorithm and a reinforcement learning-based time buffer inserting algorithm are developed.The former algorithm avoids inaccurate assessment of uncertain risks and overprotection of some activities,which are the defects of the traditional robust scheduling algorithm,through the Monte-Carlo simulation and buffer back-off mechanism.The latter uses the simulated deviation costs as the reward to guide the optimization,which can avoid the defect of the short-sighted optimization in the traditional buffer setting methods.The experimental results on a single project instance and a project instance library show that both of the two proposed algorithms can improve the robustness of project schedule in a better way than traditional algorithms.Thirdly,a resource flow-based robust optimization algorithm and integrated time buffer and resource flow-based robust optimization algorithms are developed.The influence mechanism of resource flow on the robustness of project schedule is investigated,according to which an arc goodness-based resource flow robust optimization algorithm is developed.In addition,the interaction between time buffer and resource flow is deeply analyzed,based on which two integrated time buffer and resource flow-based robust optimization algorithms are designed.The simulation results show that the integrated algorithms have obvious advantages compared with traditional robust optimization methods.Finally,a practical case of a chip R&D project involving multi-skilled employees is introduced to test the practicability of the algorithms proposed in this dissertation.The traditional robust optimization methods are difficult to apply in this case because of its complicated uncertainties.Nevertheless,the algorithms proposed in this dissertation are capable of solving the case.The application of the proposed algorithms indicates that the algorithms in this dissertation have the characteristics of strong practicability and accessibility,which are testified by the simulation results.