A Decision-Making Framework of Hybrid System Based on Modified Hybrid Stochastic Timed Petri Net and Deep Learning

For designing the decision model of hybrid systems, there are sophisticated quantitative and qualitative attributes such as discrete events, continuous processes, stochasticity, and time delay. Meanwhile, the inherent heterogeneity and concurrency existing in hybrid systems trigger multiple decision challenges, including behavior uncertainty and states consistency. It is difficult to express the complicated nonlinear mapping relationship between decision results and hybrid situations. This proposes the decision-making scheme M-HSTPN-DL with a three-tier architecture based on modified hybrid stochastic timed Petri net (M-HSTPN) and deep learning (DL). Among them, M-HSTPN describes the various hybrid situations, and DL models are taken as the decision model to express the nonlinear relationship between decision results and hybrid situations. Then the training and calling mechanism of decision models is introduced. Taking the hybrid system of bearing fault diagnosis as an example, we compare the decision-making ability of multiple decision models and analyze the advantage of M-HSTPN-DL. It has proven to be that the M-HSTPN-DL architecture can adequately represent the hybrid situation and solves the complex decision- making problem of hybrid systems.