Research On Adaptive Predictive Control Of Networked Control Systems

Recently there has been much interest in networked control systems (NCS), that is, control systems with a feed-back loop closed through a communications network or a field bus. The presence of the communication network, however, complicates the application of standard results and algorithms of control theory. The problem lies with the communication channel restricting access to the data held by the sensors; decisions must be made on old or on partial data and the intervals between updates are not regular as the network is used for many purposes besides the routing of feedback data. So there will inevitably be time delays in the communication net. As long as the sampling periods are long compared with these delays there is no need to consider the influence of the delays. As the demand on the control system increases it will be more and more important to take the delays into account in the analysis and the design of the control system.While many aspects in the NCS have been extensively studied in recent years, the adaptive predictive control of NCS has not been treated to any larger extent. This is thus an area where much can be gained by combining ideas from the fields of adaptive predictive control, real-time systems, and communication networks.Two principles are always presence on the research of NCS in this thesis. The first one is real-time abilities of control algorithm, which is very important to the NCS. So the most effective algorithm will be employed for NCS control. In most cases, online recursive algorithms are used for parameter estimation. The second principle is to control NCS with less information, e.g. only the I/O data of NCS, which will make less traffic among NCS nodes, and reduce time delay of NCS.In order to estimate random communication time delay in NCS, a six months' latency test, based on IETF's (The Internet Engineering Task Force) RFC2544, has been performed with application layer based test software. More than 1.7 million data have been obtained for correlation analysis. Two algorithms, a Least Mean Square (LMS) adaptive algorithm and a Least Square Support Vector Machines (LS-SVM) based online predictive algorithm, are proposed in this thesis respectively for the time delay estimatation and prediction in NCS. Based on the real time-dalay data, detailed performance evaluation and comparison have been made for the two proposed algorithms. The results show that the LMS adaptive algorithm takes less calculation time in time-delay estimation, while the LS-SVM algorithm can achieve a more accurate estimation.Four adaptive predictive control algorithms are proposed in this dissertation:(1)Adaptive predictive control algorithm based on time delay estimation is presented for NCS control. By modelling the time delay of NCS as a nonlinear time series, the algorithm updates the delay estimate using an online LS-SVM or a LMS adaptive algorithm. And then, an adaptive predictive control algorithm is presented to compensate and predicte for the time delay of networked control systems. Simulation results show that the proposed method has good performance in time delay predictation of networked control systems, and the plant output can trace desired output effectively.(2)By combining node buffer queuing management and adaptive predictive control, a novel approach on predictive control of NCS is proposed. Instead of solving the Diophantine Equation, the parameters of predictive controller are identified directly by an online recursive algorithm. Further more, an improved algorithm, with less computational time, is presented for NCS control. Convergence analyses show that the proposed algorithms can ensure NCS globally convergent and stable.(3)A multiple model approach is introduced for NCS adaptive predictive control. The multiple model schemes, with two fixed models and two adaptive models, are proved efficient to deal with special problems in NCS. Convergence analyses show that the proposed multiple model switch algorithms can ensure NCS globally convergent and stable.(4)Based on multi-step regressive generalized predictive algorithm and node buffer queuing algorithm, a multirate sample adaptive predictive control algorithm is proposed for NCS control in the thesis. By using this algorithm, the NCS will produce less network traffic with the same performance.