Research Of Hybrid System Modeling And Decision Methods Based On HSTPN

Because of the precise semantics,strong ability in expression brought by the graphical description and advantages in dealing with concurrent and heterogeneous problem,Petri nets are considered as one of the most commonly used modeling methods in hybrid dynamic system.Currently,there is a kind of hybrid system with discrete events,continuous processes,time-delay,stochastic phenomena and conflict problems simultaneously,such as cascade reservoir flood control system,missile interception game system,humanoid soccer robot system and fault diagnosis system.Higher requirements are placed on the modeling capabilities of model semantic rules,physical process modeling granularity,multi-domain modeling software development and complex decision-making.This dissertation focuses on the modeling and decision-making problems in such hybrid systems,the research contents and achievements are summarized as follows:1.For a class of hybrid system involving discrete events,continuous processes,stochastic phenomena,time-delay and conflict problems,HSTPN with five tuple is proposed based on the conditional/event network,five types of sub-place sets are defined and its static and dynamic characteristics are analyzed.Taking the cascade reservoirs flood scheduling hybrid system as a case,the reservoir gate scheduling logic and water level lifting process are analyzed.The results show that the HSTPN-based scheduling scheme can discharge floods safely,and then the effectiveness of HSTPN in hybrid system modeling are verified.2.In order to further prove the powerful modeling ability of HSTPN in hybrid system,multiple comparisons are executed between HSTPN and the traditional hybrid system modeling methods.Through analyzing the modelling ability of HPNs and HLPNs to the continuous,time-delay and stochastic characteristics,the advantage in simplify model structure and assurance to instantaneousness of transitions are verified.The semantic and rule complexity of HSTPN relative to HPNs is analyzed from the perspective of structure,dynamic behavioral of transitions and conflicting problems.3.Aiming at the problem of complicated operation process,staggered links,strong coupling degree of hybrid system and large modeling workload,inability to run independently,mismatch between software and models of current modeling tools,a kind of module modeling method and the corresponding modeling tool HSTPNSim based on HSTPN is proposed.Using the "top-down" modeling approach,the modeling process is decomposed to effectively reducing the modeling cost of hybrid systems,and granularity could be adjusted flexibly according to the importance of each link to the results.Then the achievement are applied in the missile interception hybrid game system.Taking the interception success rate and cost-effectiveness ratio as evaluation indicators,the double-value matrix and Nash equilibrium are given,and finally the effective of module modeling methods and rationality of HSTPNSim are verified.4.For the irrationality of decision-making problem result from ignoring the influence of physical process to the hybrid system with DEDS decision-making methods,the hybrid decision-making mechanism based on the controlled place in the HSTPN with the "event layer-physical layer" two-tier architecture is put forward and the hybrid confliction and hybrid deadlock problem are analyzed.Based on HSTPN model,the AGV automatic sorting hybrid system was established to analyze the sorting time,obstacle avoidance path selection,sorting efficiency,delay of each node and obstacle avoidance frequency distribution in the path planning process of single sorting process.The advantages of HSTPN model in hybrid system decision-making and process analysis are analyzed.5.Aiming at the problem of state space explosion and the inadequate decisionmaking ability of controlled place,the M-HSTPN with "decision layer-event layerphysical layer" architecture is proposed.By comparing with the controlled place,the advantages of the decision-making place in dealing with decision problems are analyzed.The action cycle of the humanoid soccer robot(HSR)and the multi-HSR confrontation hybrid systems are modeled,and the decision-making place is used to realize the complex distance calculation,behavior decision-making,task planning,behavior control.Compared with the HSTPN in completed time and the success rate of the ball,the superiority of the decision-making place when dealing with decision-making problems are verified.6.Aiming at the problem of decision-making model design caused by complex nonlinear mapping relationship between decision results and hybrid situation,a hybrid system decision-making scheme based on deep learning is proposed,decision results could be obtained with the aid of strong nonlinear fitting ability and classification ability of deep learning,thus improve the decision-making ability of hybrid systems.Taking the bearing fault data of the West Reserve University as an example,the hybrid system of bearing fault diagnosis is established.By comparing the decision-making ability of multiple hidden layer neural networks,stacking auto-encoders,convolutional neural networks and extreme learning machines,the effectiveness of hybrid system decision-making scheme of deep learning and M-HSTPN is verified.