Research On Some Theories And Methods Of Multiple-model Control For Thermal Process

The thermal process control system is a rather complicated nonlinear control system.Many advanced control algorithm have been found to have preliminary,successful applications in the system.Some control difficulties have been overcome.However,the problems encountered with the change of controlled system lead by frequent load change ,are still unsolved.In order to give some hints on the solution of these problems,we have studied in this dissertation the multiple model control scheme and its related theories. Our study addresses the following topics:1. In view of the large time-delay and fast time-varying characteristics of steam temperature system of Circulating Fluidized Bed Boiler (CFBB), steam temperature properties effected by spray-water and feed-water are studied in different steam load. Multi-model forward-feedback generalized predictive control is proposed. The new method has ability of overcoming some shortcomings while single model used, and eliminate the coupling property from feed-water varying.2. A PID-PID cascade control strategy is presented based on multi-model control strategy, which keeps both the merit of cascade control system and that of multi-model control. Fixed Proportional-Integral- Derivative (PID) control is used in inner loop, in order to eliminate disturbances, and the outer loop is designed by multi-model PID control. The method of scheduling among the local models is according to power network load that is the primary effect factor. Smooth switch is realized using fuzzy weighting function. This algorithm is simple and can be realized easily.3. The characteristics of the super-heated steam system are nonlinear time-varying,large inertia and time-delay. Based on this, a multi-model IMC-PID cascade control strategy is presented. The fixed models are set up at some typical operating points , the corresponding controller is desgined meanwhile. Smooth switch is realized by using fuzzy weighting function. The desgined system can keep the merit of cascade control system as well as multiple model control. Fixed PID controller is used in inner loop in order to elimitate disturbances , the time-delay of intera plant can be overcome by the internal model controller of outer loop.4. A self -tuning PID control strategy based on a two-level Diagonal Recurrent Neural Networks (DRNN) is proposed for SSTS (Superheated Steam Temperature System) and is extended to multi-variable system.The two-level NNs are called Static NN (SNN) and Dynamic NN (DNN) respectively. SNN is used for controller PID's arguments primary tuning . The PID arguments, designed based on local muiltple models which are set up according to different operating conditions, are used to train SNN. On the basis of SNN's coarse tuning to PID, DNN is used for fine tuning according to error and error rates to overcome the small range load changing, system parameters'slow variance and some disturbance.Combined with gery prediction, good dynamic regulation performance used in SSTS can be obtained by using the presented new method, and stronger robustness is obtained. The sucssesful application in CCS (Coordinated Control System) prove that this method can realize quasi—dynamic decoupling and complete static decoupling , simultaneously featured by quick response and strong robust capability.