The advancement of control, computer, and communication networks technologies, lead to the development of fieldbus that labels the emergency of networked control systems (NCSs). Meanwhile, the technologies on general computer networks especially Ethernet have progressed rapidly. With widespread usages (80% of LAN market), increasing speed, decreasing price, and well-established infrastructure, Ethernet becomes major competitors to the industrial fieldbus. In response to the corporation information technology application and global optimization, Internet has been also used to NCSs. In the systems based on Ethernet and Internet, TCP/IP Protocol is widely adopted and these systems can be intituled as TCP/IP system. The randomicity and indetermination of network-induced time delay in TCP/IP system makes the analysis and design of an NCSs complex. So it has significance to research on the design of NCSs over TCP/IP network.PID control has a simple structure and can perform more reliably than many advanced controllers, so it's the most popular controller in industry. So it is very important to improve the performance of the PID controllers in NCSs.In this paper, campus net linked to Internet, one instance out of TCP/IP networks, is adopted to estibilish the NCSs. The network-induced delays are measured and their statistical characteristics are analyzed. The design methods and compensation schemes of PID controller over campus net are presented.First of all, a measurement environment based on application layer for network-induced delay over campus net is established and sufficient RTT (Round Trip Time) delay data are obtained. The analysis about these measurements reveal that the statistical characteristic of the RTT delays based on application layer over campus net is self-similar, so it is very difficult to design controller over campus net. Pareto distribution is the most suitable distribution al candidate by Chi-square test method. Because of the complex of RTT delay over campus net, it's not precise to model the RTT delay by only one probability distribution function in a short period of time. In a RTT delay sliding widow (the length is 150), Pareto distribution and generalized exponential distribution can be used to model the RTT delay in different actual network traffic condition by chi-squre test. But the disadvantage of the chi-square test is that it requires a sufficient sample size in order for the chi-square approximation to be valid. Thus, the idea and method that use a support vector machine classifier to pick the suitable distribution from the Pareto distribution and the generalized exponential distribution online are put forward.In campus net, TCP/IP protocol is adopted, and data packet loss is impossible. But the packets can arrive at a network node in a wrong order, so the driven modes in controller and executer node need to be choiced carefully. Considering the characteristic of control system, the arithmetic that the late packet is discarded is provided.The results about RTT delays over campus net are realible through measuremen, analysis and modeling. Theses results can be applied to design NCSs based on TCP/IP network.Then, aiming at lower order time dely linear plant in industrial process, the region of admissible gain of parameterization PI controller is presented, and the relationship between this region and time delay is derived. In order to use this controller as networked controller over campus net and maintain the best possible performance, the optimal gain need to be scheduled in real time with respect to network delay. This paper uses Pareto distribution and generalized exponential distribution to describe round-trip time (RTT) delays over campus net, and a support vector classifier is used to select the appropriate distribution model in real time. The optimal gain under different distribution model parameters is off-line evaluated by genetic algorithm (GA). Simulation results show that the methodology proposed provides better performance.Next, the network-induced delay over campus net is modeled. The delay is divided into two parts, one is constant delay and another is uncertain delay varying from the constant delay. The delay is approximated by the first-order Pade approximation, and the uncertain delay part is treated as the simultaneous multiplicative perturbation. The robust PID controllers are designed by H_∞framework andμ-analysis. Through simulation, the robust PID controller with invariable parameters can't fulfil performances in all network traffic conditions. The parameters of delay model need to be computed by the mean and median of RTT delay. The parameters of PID controller are computed offline in all network traffic conditions, and the optimal parameters are determined online by the actual delay.Finally, considering to traditional neural network approach have suffered difficulties with generalization, producing models than can overfit the data, and the support vector machines has been showen to be superior to neural network, least squares support vector machines (LS-SVM) is proposed to model the complex industrial process. A predictive PID controller based LS-SVM is presented which is mathematically equivalent to generalized predictive control (GPC). For using this GPC-based PID controller to campus net, a predictive algorithm and a control selector are introduced to the controller side and the plant side, respectively. Simulation results are presented to highlight the principles and effectiveness of this control scheme. |