Study On Intelligent Control Method Of Networked Control Systems

With the continuous expansion of system scale and the rapid development of information technology,such limitations of geographical area and resources are broken in networked control systems.The network control systems become much more important due to their modularization,integration,digitization,and intellectualization properties.In the networked control systems,the factors of channel disturbance,packet loss and inter-node information transmission delay of network communication mechanism will lead the network control systems lose their steadiness,integrity,causality and certainty.The analysis and design method of traditional point-to-point direct control system is no longer suitable for the network control systems.Therefore,it is necessary to restudy the analysis and control method in view of the characteristics of network control systems.Intelligent control is an interdisciplinary subject integrating artificial intelligence,fuzzy set theory,operations research and cybernetics.The intelligent control technology is used to study the control method,it can solve the complexity problem which caused by the model uncertainty,and then provide an effective way for establishing the analysis and control strategy of network control systems.In this doctoral dissertation,the main problem of analysis and control in networked control systems and intelligent control technology are summarized comprehensively,the characteristics and mechanism of BP network and genetic algorithm are analyzed deeply.In view of such problem as time-delay influence performance of networked control systems,the control methods for the networked PID control based on fast BP network,the networked predictive control based on gradient weighted genetic algorithm-BP network and the variable sampling period networked control based on fuzzy algorithm are proposed.The main work is as follows:(1)For the problems of time-delay compensation and PID controller parameter tuning in network control systems,a class of networked PID control method based on fast BP network is proposed.Considering the time-consuming problem of obtaining the number of hidden layer nodes for BP network time-delay prediction model with large dataset,a dimension projection algorithm is proposed.Cross-validation and repeat experiments are not required for the dimension projection algorithm.The number of hidden layer nodes in BP network can be directly calculated.Considering the time-consuming problem of learning method in the BP network time-delay prediction mode,a hybrid learning method based on parameter adaptive genetic algorithm is proposed.The fast BP network time-delay prediction model is composed by the dimension projection algorithm and the hybrid learning method based on parameter adaptive genetic algorithm for improving the speed and accuracy of the system.Considering the problems of integral saturation,PID controller parameter tuning and model mismatch,the integral term of PID controller is improved,and a PID parameter regulator based on on-line BP network is designed.A class of networked PID control method is given based on the fast BP network time-delay prediction model and the on-line BP network parameter regulator.(2)For the problems of time-delay compensation and unknown controlled object model,a class of networked predictive control method based on gradient weighted genetic algorithm-BP network is proposed.Considering the time-consuming problem of obtaining the number of hidden layer nodes for BP network time-delay prediction model with small dataset,a weight analysis algorithm is proposed.The weight analysis algorithm relies on the weights statistics information of BP network rather than the test result.Therefore,it is not sensitive to dataset partition which can save time by sample replacement.Considering the problem of local optimum in the BP network time-delay prediction model,a hybrid learning method based on gradient weighted genetic algorithm is proposed.The gradient weighted genetic algorithm-BP network time-delay prediction model is composed by the weight analysis algorithm and the hybrid learning method based on gradient weighted genetic algorithm for improving the control accuracy of the system.Considering the linear and nonlinear control problems in the network control systems,a single-step predictive controller based on on-line BP network is designed.A class of networked predictive control method is given based on the gradient weighted genetic algorithm-BP network time-delay prediction model and the single-step predictive controller.(3)In view of such problem as on-line adjusting time-delay in the linear and nonlinear networked control systems,a variable sampling period networked control method based on fuzzy algorithm is proposed.A variable sampling period method based on fuzzy algorithm is used for on-line adjusting time-delay,a fuzzy controller for the linear and nonlinear controlled object is designed,and the variable sampling period networked control method based on fuzzy algorithm is also given.The genetic algorithm is used to optimize and improve BP network.Some control methods such as networked PID control(based on fast BP network),networked predictive control(based on gradient weighted genetic algorithm-BP network)and variable sampling period networked control(based on fuzzy algorithm)are proposed.Those control methods are good to solve such problems as model uncertainty,time-delay prediction and compensation in the network control systems,and the simulation platform is used to verify the experiment.The results in this doctoral dissertation are great significant to enrich and perfect the theory of network control and accelerate the application of network control systems.