Community Structure Detection Based Subsystem Decomposition And Distributed Model Predictive Control Of Large Scale System

With the development of social economy,a large-scale system has emerged with characteristics of complex structure,multi-objective,mode uncertainties and randomness.In recent years,the requirements of industrial production process have been continuously improved,and higher requirements for control indicators and performance have been put forward at the same time.Due to the geographical location of complex system,it is difficult to communicate with each local system,which will increase the communication cost and reduce the system reliability.Model predictive control has received great attention with advantages in dealing with constraints,multivariable and coupled systems.Model predictive control can solve many problems that conventional controllers cannot solve.At present,the application of model predictive control mainly based on centralized mode.That is,it designs a control center to measure the state or output of the system by the sensor.The controller optimizes the global input based on the measurement information,and then sends it to the actuator.However,the research object of model predictive control is becoming more and more complex.On the one hand,because of the requirements for receding optimization of system input,there is a problem of centralized optimization when predictive control processes complex systems with fast sampling rate.On the other hand,the traditional centralized model predictive control signal depends on a single controller,if the connection between the centralized controller and the actuator of the system fails,the whole control system will fail.To this end,the robustness and flexibility of its control structure are poor.In recent years,the control of complex network structure system has gradually changed from centralized to networked distributed coordinated implementation.The common complex large-scale system is composed of many subsystems which are coupled with each other.It is necessary to decompose these subsystems reasonably first.Then,we can design distributed state estimation and control for each subsystem.The research of distributed model predictive control is of great theoretical significance and practical industrial application value.This topic focuses on the division of community discovery subsystem and the problem of distributed predictive control of large system.The main contents include the research on the decomposiition method of large system,the design of distributed state estimation,predictive control and coordination scheme.The research contents are as follows:1)The method of complex system decomposition based on weighted graph theory community detection algorithm is studied.First,the weighted digraph of complex large-scale system is constructed.The state variables and the measured output variables are regarded as network nodes,which are connected by the weighted edge reflecting the connection strength of the state and the measured output variables.Then,based on the weighted digraph,the community structure detection algorithm is used to decompose the global system into smaller groups to make the connection strength ratio within each group.The coupling between different groups is much stronger.Finally,the subsystem decomposition algorithm for distributed state estimation is studied.The subsystem decomposition method is applied to the decomposition of process system,and several distributed state estimation algorithms are designed to show the effectiveness.2)The design method of distributed predictive control for uncertain systems with random packet loss is studied.Under the structure of information physical system,model uncertainty and information packet loss are two unavoidable problems.For the information physical system with model uncertainty and packet loss,this chapter considers the model uncertainty of the random packet loss with Markov process.The whole system is decomposed into multiple subsystems for distributed predictive control design.The coordination of each subsystem controller is achieved through iterative process algorithm.Finally,a random distributed predictive control design method is presented.3)The distributed predictive control design method of stochastic input saturation constraint system is studied.Because the input saturation constraint is an inevitable problem,the influence of state delay should be considered for the uncertain system with stochastic input saturation constraint.The whole system is divided into several subsystems,and a distributed MPC structure based on stochastic input saturation constraint system is proposed to relax the input saturation constraint.The stochastic input saturation constraint is described by Bernoulli distribution fashion.The optimization problem of random distributed MPC is designed,and the distributed processing of random input saturation constraint is realized by designing an iterative coordination algorithm.4)The distributed predictive control method for information coordination is studied and simulated on the cutterhead system of hard rock driving equipment.The dynamic model of the cutter head system is established,and the parameters of the model are estimated based on the construction data.The cutterhead system model which can be used for control is obtained.In view of the speed control problem of the cutter head system,firstly,the cutter head system is divided into several parts to get the subsystem model.Secondly,the distributed control structure is designed,and the independent distributed predictive control is designed for the torque setting of the motors in each part.The existing model uncertainty is dealt with by using the feedback correction.Finally,a distributed predictive control system is designed.A sequential iterative coordination algorithm is designed.