Improving system performance for stochastic activity network: A simulation approach

An activity network with returning loop activities has a wide variety of applications, but can cause a heavy computational burden for large networks. Moreover, if an activity processing time and/or the probability of taking a particular activity changes when the number of activity visits is added, the computation is very complicated and difficult. We propose a simulation approach to deal with stochastic activity networks consisting of multiple terminal nodes, no limit on looping activities, non‐constant activity selection probabilities, and non‐deterministic activity times following arbitrary distributions. Probability and time control functions are introduced to reflect the acceleration, or learning effect, of repeated activities. Performance measures such as system success/failure probabilities, time to completion/success/failure times, time between success/failure, and the pth percentile times of a project are obtained. A series of sensitivity analysis was performed to understand the trend and behavior of system performance. A cost function is developed to find an optimal strategy by manipulating control factors. To illustrate the efficacy of this simulation approach a new drug discovery and development project was analyzed. The Promodel simulation language was used for performance evaluations, and the SimRunner optimization tool for obtaining the optimal solution.