Research On Automaton Model Updating Using L~* Algorithm

With the rapid development of industrial technology,the complexity of the system has in-creased significantly.How to further control and optimize the complex system has become a key problem that people pay attention to.The formal modeling of discrete-event system can build the complex system into an intuitive visual model for the subsequent research and development,which has a powerful advantage in solving the problem of complex system.The modeling and simulation technology of discrete-event system is widely used to solve complex system optimization and management decision-making problems in manufacturing,supply chain,finance,medical and other fields.It has great practical value and technical advantages,so it has attracted more and more attention in academia and industry.Finite state automaton is an intuitive and easy to use discrete-event system model,which is widely used in the process of system modeling and analysis.However,the system is not fixed.In practical applications,the system is usually updated and optimized to meet the needs of users or improve the reliability and robustness of the system.In order to facilitate the development of future research work,the finite state automaton model con-structed according to the original system needs to be updated accordingly.At present,the tradi-tional discrete-event system modeling method and simulation technology can only be used to build the system model,and cannot be directly used to update the model.The reconstruction of the automaton model of the system will do a lot of repetitive work.Therefore,how to update the existing model into a model that can describe the logical behavior of the new system has become one of the problems to be solved.The updating operation of the automaton model of the system can be divided into three types:adding system state,deleting system state and modifying transition function.Based on the learning framework of L~*algorithm,this paper proposes two update method that can be used to determine the finite state automaton model.The main research contents are as follows:(1)This text proposes an automata model updating method based on counterexample guid-ance.This method includes three different model updating algorithms,and the corresponding al-gorithms are used to update the model for three different system changes.Through conformance test,the algorithm makes equivalence query on the original system model and the new system to be modeled to find out the difference between them,returns the information in the form of the counterexample string,and then updates the system model according to the counterexample.Be-cause different algorithms handle counterexamples differently,this method can only update one system change operation at a time.Therefore,this method is more suitable for the situation where only one kind of system change occurs and the system change rate is small.Experiments show that this method can update the model correctly.Compared with L~*algorithm,this method can effec-tively reduce the number of learning rounds and membership queries used in the modeling process.(2)This text proposes an automaton model update method based on information caching.The method stores part of the known information of the original system by referencing a cache.Spec-ulates and modifies the information in the observation table based on this known information,in order to make corrections to the erroneous data in the original observation table and complete the update of the original model.The final experiment verifies that the method is also effective in reducing the number of membership queries used for modeling,and the method is able to handle multiple change operations of the system model in a uniform manner,which is more suitable for the situation where multiple type change operations have occurred in the system.Different from the existing updating methods,the two model update methods proposed in this paper have no additional requirements on the system,only need to collect the input and output information of the system and complete the update of the original observation table data.There-fore,the methods in this paper are more universally adaptable.In addition,the two proposed meth-ods do not need to use a large number of observation tables of learned models.The update of the specified system model can be achieved by modifying only one observation table of the original system model,and it can reduce the number of learning rounds and the number of member queries used for modeling during the model update.