MANAGEMENT OF RESOURCE CONSTRAINED MULTI-PROJECTS WITH REATTEMPT AT FAILURE

在本研究中，我们考虑了资源受限的项目调度问题，其中每个活动都有几种不同的执行方式(模式)。每种模式都携带特定数量的资源，并具有不同的持续时间。在执行活动时，我们必须考虑资源的约束选择合适的模式，并且活动的执行成功概率会随着所选模式的不同而变化。我们针对上述6个问题进行了研究：(1)给出了最大化项目总成功概率的有效算法。(2)考虑了基于每个活动执行后的性能测试结果需要重试任务的情况，给出了最小化项目总预期完成时间的启发式随机方法。通过大量的数值实验，该算法比现有的调度规则有更好的评价。(3)针对上述问题，我们提出了一种遗传算法，在几乎相同的计算时间内使上述启发式算法的解更优。(4)我们还利用了现代启发式算法中最优秀的禁忌搜索算法。但数值实验表明，遗传算法优于禁忌搜索。(5)对于总资源受限的多模式、多项目问题，我们证明了类遗传算法也是有效的。(6)我们考虑了一个具有随机活动持续时间的随机PERT网络问题，并给出了一个很好的算法来估计期望完成时间的上下界。

In this study, we consider resource constrained project scheduling problems, in which each activity has a several different ways (modes) in their execution. Each mode is carried with specified quantities of resources and with a different duration. On executing an activity we have to select a suitable mode considering constraints on resources, And further success probability in performance of the activity is changing according to a mode selected.We studied the following 6 subjects on above problems :(1) We first give an efficient algorithm to maximize total success probability of the project.(2) We consider the case that a reattempt task is required based on result of performance test done just after execution of each activity, and we give a heuristic stochastic approach minimizing total expected completion time of the project. Through many numerical experiments the proposed algorithm gives much more nice evaluation than existing dispatching rules.(3) For the above stated problem, we propose a Genetic Algorithm which brings superior solutions to the above heuristic in almost equivalent computing time.(4) We also exploit Tabu Search Algorithm, which are said to be excellent among modern heuristics. However for the problem it is shown that Genetic Algorithm is superior to Tabu Search through numerical experiments.(5) We show that the analogous Genetic Algorithm is also effective for a multi-mode, multi-project problem with total resource constraints.(6) We consider a stochastic PERT network problem with random activity duration time and give a nice algorithm to evaluate good lower and upper bounds for the expected completion time.