Research On Hybrid Modeling Method For600MW Unit And The Energy Consumption Sensitivity

The energy-saving and emission-reduction is on a new starting point with the development of national economy. As the source of the power system, the power plant is always the most important part of the energy production. At the same time, the energy-saving and cost-reducing of the power plant is a heat subject of scientific research at home and abroad. The energy consumption model for generator is the theoretical and date principle of the energy-saving and cost-reducing in a power plant, it is also the important foundation of the design and maintaining for the thermal equipments. At the same time, it can be the instructor for in-service operating and adjustment. With the development of the technology and equipments in the power plant, it expects more and more of the definition and accuracy of the thermo-economic analysis for unit.Currently the modeling method is divided into two kinds of forward modeling and reversed modeling. Forward modeling that mechanism modeling is derivation of the functional relationship between the main variables and auxiliary variables by theorem as energy conservation and mass conservation, after the comprehensive understanding of the operation mechanism of the entire system. Reverse modeling can be seen as a "black box" modeling method without understanding the internal operation mechanism and the device structure of the system in details. It creates the mathematical relationship between the needed main variables and auxiliary variables by the large amounts of existing operating data in order to predict the value of the main variable. In this paper some wide and classic modeling methods are fully introduced from both forward and reversed two angles. After that, based on the available information, the advantages and disadvantages of all the methods are compared. The feasibility and scope of the models are analyzed. Neural network algorithm is a kind of reversed modeling method based on the error back-propagation theory modeling. It is suitable for analog input and output approximate relationship. It can approximate arbitrary continuous function with arbitrary precision. That is the nonlinear approximation of unknown function by training samples. In practice, it can be done by using neural network toolbox in Matlab software package quickly. In this paper a coal consumption model is established with higher accuracy and wider application by the application of forward modeling and reversed modeling two theories, while the forward modeling is for turbine system which is that the internal structure has an important influence on the model, and the reversed modeling is for boiler system as "black box" by using neural network algorithm. Then the energy consumption sensitivity of the important parameters was analyzed according to the established model. It is important for operator to adjust the unit.