Research On Application Of Temperature Control System With Nonlinear Prediction And Control Algorithm

The application that need to measure the temperature and control is more and morein practical projects,measurement and control temperature of the process with large inertia,pure hysteresis, time-varying of nonlinear characteristics. This paper chooses a sealed spacesuch as a barn or a closed cabin ( such as mine refuge cabin) temperature measurement andcontrol as an example, the predictive control method can effectively regulate and control theclosed cabin temperature, has reached the set value that people want or the purpose ofpredictive control.In this paper,the sensor is used to measure indoor temperature with heating、cooling measures to meet the control requirements, to simulate the sealed space temperaturepredictive control.Because of the temperature measurement and control system of large inertia,pure hysteresis, time-varying characteristics, the traditional control methods can’t meet thecontrol accuracy,can’t achieve good control quality requirements. In order to better carry outhigh precision temperature measurement and control, also access to meet the technologicalrequirements of good precision quality control under the condition of the minimalcomprehensive consumption, this paper studies that using nonlinear prediction and controlmethod monitors and controls temperature of sealed cabin and achieves a good control effect.In this paper,nonlinear recursive augmented matrix least squares method, BP neuralnetwork algorithm and the extended Kalman filter(EKF) algorithm combined that is used topractice nonlinear predictive control for the temperature measurement and control system.Thepaper centers around the use of prediction model for model predictive control and spread out,first expounded the nonlinear predictive control theory, and then using the dynamic adaptiveBP neural network set up prediction model according to the principle; in the establishment ofprediction model before system model should also know, but here the object model is unknown, due to the Randomness of interference noise, using the nonlinear least squaresalgorithm set up nonlinear model of the sensor(NARMA); the use of dynamic adaptive BPneural network to establish the prediction model, system presents the dynamic characteristics,coupled with the various factors of noise and the temperature measurement and controlsystem with unknown noise interference, system modeling process is not reflected, so in orderto accurately nonlinear temperature predictive control, it is necessary to dynamiccompensation in the process of system prediction control, because the system is nonlinear, sowhen designing of dynamic compensator, using the extended Kalman filter algorithm to builddynamic compensator for nonlinear system effectively. Of course, information collection isnecessary before the entire system identification and modeling, so the article also discussesthe temperature corresponding to the voltage data acquisition system and data collectionprocess.Using the related theory established the dynamic model of sensor with the nonlinear leastsquares fitting in this paper;designing dynamic compensation filter that is to stretchtemperature sensor working band, on the temperature sensor signal for dynamic errorcompensation;using nonlinear neural network method predicts and controls temperature.Thesubject by using of nonlinear control theory for the large inertia temperature control objectconducts a series of nonlinear algorithm design and rolling optimization prediction targetvalue. Finally, through the above algorithm designed model and controller are combined toobtain the whole design of the system frame, and on this basis, using MATLAB simulationexperiment, to draw the relevant conclusion, and the simulation results are given. Simulationresults show that by using dynamic adaptive BP neural network predictive control systemdesign effect than the traditional control method of predictive control are better, comparingwith target value from table can be seen. The necessary to improve is linearized of nonlinearobject model and the BP neural network parameters of design.