Research And Implementation On Cascade Predictive Control Of Multi-Zone VAV Air Conditioning Systems

There are several control loops in Variable air volume (VAV) systems, and these loops usePID control scheme. PID parameter setting has a great impact on the control systemperformance.There are several equipments in this complex system, which should adjust PIDparameter according engineer’s experience in the different applications. There is noself-learning ability on PID control algorithm,which makes a difficulty in field application.This paper presents a neural network cascade predictive control, which combine thesuperiority of both predictive control and cascade control. In this method, two neural networkcontrollers are used as inner and outer controller in a cascade structure. Outer loop controlleris used to calculate the set air flow of input air-conditioned area. the inner loop controller isused to control air-valve. Inner and outer control are based on feedback control and the Euler-Lagrange optimal control algorithm, which adopt multi-step prediction optimizationperformance index to overcome the uncertainties and complexity of change. The predictionmodel based on the BP network, the outer loop predictive model is a temperature model,theinner loop predictive model is a air flow model.After researching the control strategy，experiments were carried out in the multi-zoneVAV system Laboratory of Beijing University of Civil Engineering and Architecture. Thefactors influencing on the sensible cooling load and coupling between zo nes are analyzed andconsequently the structure of the neural network model is determined. Then use the embeddeddevelopment board and I/O modules to acquire data of two zone,including temperature, windspeed,static pressure pipes, etc. In order to fully demonstrate the dynamic characteristics ofthe VAV system, neural network training samples cover all the VAV dynamic range. Aftercomparing the generalization ability of the BP network and the RBF network, and this paperselect the BP neural network to establish the temperature model and the air flow model.Basedon the established model, programming the algorithm operated in Windows CE6.0is used tocontrol the VAV air-conditioning system in Laboratory.Experimental results show that the neural network cascade predictive control couldrealize decoupling control, which ensure temperature and air flow control accuracy ofzone.Self-learning ability of neural networks can solve the problem of debugging PIDparameters, and the control system is portable.