Research On Network Based Model Predictive Control

Due to the increasing complexity of the networking of industrial control systemsand the modularity, economic efficiency, easy maintenance features of the modern com-puter network, the network based control systems has been widely used in industries inrecent years. Special issues on network control systems such as network induced delayand packet dropouts have received considerable attention due to its effects of control per-formance degradation and even instability, and are the major research problems in the pastyears.First, we consider the analysis and synthesis of a class of uncertain time-varyingfuzzy systems under the input output framework. Due to the better approximation abil-ity of two-term approximation method, we use two constant delay items to estimate thetime varying delay item, which results in the system transformation of feedback intercon-nected form. The scaled small gain theorem is further used to reduce the analysis andsynthesis problem of the original time varying system to that of the constant delay system.The widely used delay partitioning technique and fuzzy weighting dependent matrices areemployed to deal with the constant time delay. The difficulty in controller design prob-lem lies in the coupling of the Lyapunov matrices and controller gains matrices. A noveldecoupling method which is based on the Finsler’s lemma is proposed, and this methodreduces the free weighting matrix technique to its special case. The extensive simula-tion validates the less conservatism of the proposed method over the existing literatures.Second, we consider the network based model predictive control problem. The proposedalgorithm is applied to the tracking of the industrial economic objective. The predictivemodel is reformulated to explicitly include a class of dynamic random network dropoutmodel. Thecontrollergainsareobtainedthroughtheoptimizationoftheupperboundofaninfinite performance objective. The flotation process model is employed to demonstratethe effectiveness of the proposed method.