Study On Control Strategy Of Central Heating System

Central heating system is one of the city’s critical infrastructures. Datashows that building energy consumption accounts for30%of total energyconsumption in our country, and the heating energy consumption per unit area isa multiple of other developed countries. Along with the development of centralheating system, vast coal and electric energy have been consumed every year.Therefore, the research on energy-saving control in central heating system is ofgreat significance.System model is the premise of running state analysis and control of thecentral heating system. The central heating system of a community in Taiyuan isused as study subject in this paper. Step response of the central heating systemcan be obtained by step increasing the setting temperature of direct-firedmachine. The parameters such as temperature of supply water and return waterare recorded. Hammerstein model is used as the model set of the central heatingsystem to identify. The inputs and outputs of Hammerstein model are settingtemperature of direct-fired machine and temperature of return water. Theparameters of Hammerstein model are obtained by the identification of recorded data with recursive least squares method. The order of Hammerstein model isacquired by F-test method based on residual of the model. The validation resultsshows that the Hammerstein model of central heating system obtained byidentification traces the output of the system very well and can reflect thedynamic characteristic of heat-supply system.Heating load forecasting is premise and foundation of on-demand heating.On the analysis of the heating load sequences, concluded that heating load haveapparent periodicity, though it can be disturbed by random factors. The WaveletNetwork is introduced to predict heating load. For the problems of trainingalgorithm of wavelet network, extended ant colony algorithm is used to train thenetwork. The simulation results show that, compared with the BP neuralnetwork and Wavelet network, wavelet neural networks optimized by expandAnt Colony algorithm have higher forecasting accuracy.Because of the characteristics of model calibration and rolling optimizationet.al of Generalized predictive control, the strong nonlinear， large-lag，time-varying of heating process can be overcome. Generalized PredictiveController is designed based on the Hammerstein model, and used to Simulinkthe control process of central heating system by contrast with conventional PIDcontrol. Results shows that the generalized predictive controller is able to trackthe set point the temperature of return water and have a certain degree ofanti-interference ability.