A reliability evaluation method for complex systems based on the editable GSPN and adaptive Monte Carlo simulation

This paper presents an innovative EGSPN (Editable Generalized Stochastic Petri Net) model, optimizing the size of GSPN (Generalized Stochastic Petri Net) while preserving its core benefits. Specifically, the EGSPN introduces a transition termed "editable transition," which allows for the establishment of any dynamic logic, thereby circumventing the intricacies associated with the EGSPN model structure. A Monte Carlo (MC) simulation technique is put forth for EGSPN analysis, enhanced by the integration of Sobol sequences with a Bayesian optimizer. This MC simulation approach, unbounded by any specific distribution constraints, exhibits accelerated convergence rates. To address the issue of preset sampling size, an adaptive sampling strategy rooted in confidence intervals has been proposed. By real-time computation of margin of errors under varying confidence levels, simulations can be halted prior to reaching the preset sampling size. This method strikes a balance between reduced simulation time and maintained accuracy. The applicability and efficacy of the proposed method are further elucidated through a numerical example of a flight control system.