The Application Of Neural Network Predictor In Heating System

Along with the rapid progress of city construction, the central heating has become one of schemes adopted in the construction to reach city modernization in north China. The problem of how to control and manage the whole heating system effectively and thus enhance its economic and social benefits has become important subject that every heating enterprise urgently wants to resolve while the heating system is being enlarged gradually.The technology of the heating control is at lower level compared with developed country and there are many aspects including designing, installation, operation, management and production of facilities are need to be studied and this technology is also to be improved. According to the difference of the using characters of buildings, the heating system is divided into two categories, the continuous heating system and the intermittent heating system. To the intermittent heating system, it is essential to estimate accurately the preheat time. Overestimation of preheat will cause unnecessary energy wastage whilst underestimation will lead to a loss of comfort conditions at the start of occupancy.In this thesis, the law of change process of the inner part of building affected by heat resource and environment was detected. From the perspective of thermodynamic system, the mathematical model of the process of intermittent heating was established on the basis of similitude principle and a hybrid simulation experiment was conducted systemically. The hardware of this experiment was realized by using the analogue machine while the software was realized by adopting Visual Basic 6.0 language. The LonWorks technique of Echelon Corporation was used in interface communication and date collection. The hybrid simulation approach's main purpose was to gain input and output data for following construction of model for predictor.In this thesis, the predictor was implemented by using Neural Network that is widely applied in each field as a new method. By using Neural Network the time-predict model full of non-linearity was established and the improved BP (Back-Propagation) Algorithms was introduced in this work. The Back-Propagation network is the core part of Ahead- Propagation network in Artificial NeuralNetwork and is widely applied in many aspects such as function approach, mode distinguishing and data condensation. MATLAB is a kind of very powerful project language whose inner Neural Network tools make the realization of Back-Propagation Algorithms more easily and convenient.The time-predicting model constructed on the basis of Neural Network has been put into practice on system in this thesis and the ratio of error was less then 3%, which would be a good result and indicated that the aim of saving energy could be reached perfectly.