Load Prediction And Optimal Setpoint Based MPC Control Of Compression Refrigeration System

Based on the moving boundary method，nonlinear dynamic mathematical model ofrefrigeration system was established. The effect of void fraction of the evaporator andcondenser in the refrigeration system on the time-varying dynamic model parametersis considerated, and the void fraction mechanism and calculation is analysed anddisplayed. With the energy and mass conservation law constraints, each componentmodel is jointed to build a14-order dynamic mathematical model of the overallrefrigeration system.According to the cyclical nature of the load data, based on method of cycle timeseries differencing, smoothing of the periodic sequence process is established. Withthe AIC criterion, using maximum likelihood estimation algorithm, order andparameter of ARIMA time series model is estimated. Traditional load forecastingalgorithm does not take full advantage of the prediction error information to improvethe prediction accuracy. Using the error information to estimate future errors, the sumof initial prediction and the future error estimation is the proposed method in thispaper. The accuracy of the predictive value of24hours load is increased by6%.A set point optimal control program was designed, which the optimization objectivewere the total energy consumptions of the compressor, chilled water pump andcooling water pump, the working points were evaporating pressure, condensingpressure and the minimum stable superheat, and using the pattern search algorithmwith the external penalty function as constraints to solve the objective function.Usingadaptive methods of energy consumption model, according to the changing inworking conditions of the system, the static working points of the optimal controlloop values were optimized. Experiment shows that the system can adjust wellcorresponding to the conditions change under optimal control scheme and theenergy-saving effect is remarkable. Under part-load conditions, the energyconsumption of cooling system is reduced by11.8%.With the existence of pure hysteresis for superheat and evaporation temperature ofthe cooling system, the traditional model prediction algorithm can not effectivelysolve the serious problems of controlled variable overshoot. As forecasting model inmodel predictive control algorithm, SMITH predictor and generalized least squares algorithm for online identification are used to improve the model predictive controlalgorithm structure. In the feedback loop, with the small gain theorem, low-pass filterto improve the robustness of the control system in the prediction model mismatchcase is established. Compared with the traditional model predictive control algorithm,experimental results show that improved SMITH predictive model predictive controlalgorithm has a better dynamic precision and reduce the superheat and evaporationtemperature shock and overshoot. Meetting the minimum energy consumptiondemand, the system operates in more stable condition.