Research On Dynamic Stochastic Resource-Constrained Multi-Project Scheduling Problem

Order is the basis for the survival and development of enterprises.The enterprises receive new orders,usually in the form of multi-project management,to develop the implementation plan.In this case,project scheduling plays an important role in project management planning.With the increase of customization requirements,it gradually brings new challenges to resource-constrained multi-project scheduling problems.For example,the products may require further quality inspection and even need to be reworked with constant new projects added in and unknown actual duration of them.These random factors affect the normal implementation of the plan,and the project is with ease to face the impact of procrastination in the implementation.Therefore,the project manager needs to update the scheduling plan dynamically and frequently,which brings a considerably high requirement to project managers.From the perspective of project managers,this thesis analyzes the multi-project scheduling problem with dynamic resource constraints,and further illustrates the scheduling algorithm of multi-project scheduling problem with random resource constraints with the optimization goal of minimizing weighted delay.The work carried out in this thesis is as follows:(1)An example of multi-project scheduling problem with dynamic resource constraints is proposed.The existing resource-constrained multi-project scheduling problem is studied based on the deterministic project duration.In order to carry out the research on dynamic stochastic resource-constrained multi-project scheduling,this thesis needs to build appropriate and standardized dynamic multi-project calculation examples.Firstly,on the basis of the comparison of several standard example generation methods,this thesis constructs an appropriate example of dynamic resource-constrained multi-project scheduling problem based on multi-project resource index and network index.Secondly,in this thesis,based on the references in the competent responsible for multiple purposes and from two perspectives of the single purpose project director,the samples are simulated in the optimization goal of minimizing the weighted tardiness to reveal a fact that various objective indicators can affect the performance of precedence rules.(2)Based on the dynamic multi-item example proposed in Part I,the influence of different random factor levels on the performance of priority rules is studied.To consider in actual situations where the projects are in different level of random rework and random time limit for a project is mathematically distributed,in this thesis,the examples of different random environment are simulated obeying the precedence rules.Besides,this thesis also analyzes the effect of random level on the properties of precedence rules and summary and analysis on the impact of random factors on precedence rules according to the conclusion part of the project more indicators,(3)To explore a better performance algorithm for multi-project scheduling problems with dynamic resource constraints.This thesis presents an approximately dynamic programming algorithm based on short-sighted strategy and Markov decision process modeling.Also,with the same short-sighted strategy of intelligent algorithm variable domain search and priority rules as the control,dynamic multi-project in two scheduling environments of fixed type and random type is carried out to compare and verify the performance and operating efficiency of the algorithm in this thesis.(4)Supervised learning is used to solve the infinite time-domain multi-project scheduling problem.The example constructed in this thesis is to simulate the infinite time domain stochastic resource constrained multi-project scheduling problem,to analyze with a small number of samples,and finally to apply the conclusion to the actual infinite time domain stochastic environment and verify it.In order to shorten the simulation time of approximate dynamic programming,the supervised learning classification model is proposed to train the solutions of approximate dynamic programming,and the obtained model is applied to a dynamic random multi-item example to verify its validity.In an attempt to improve the performance of model,the method of classification model fusion is proposed in this thesis.Finally,an infinite time-domain multi-item example is constructed,and the supervised learning model with the most efficient priority rule and the best performance is used to conduct simulation experiments to verify the performance of the supervised learning classification model under different arrival rates and resource loads.Based on the above research content,this thesis finally summarizes the scientific method that can be applied to the actual infinite time domain dynamic stochastic resource-constrained multi-project scheduling problem,which has certain guiding value for improving the enterprise multi-project management level.