Solving The Heat Transfer Problem Of Gasifier Wall Surface Based On Data-driven Method

Coal gasification technology is a very important part of the topic of efficient and clean coal utilization,and it is also an important means for my country to cope with the deformed energy structure of "rich coal,less oil,and poor gas" and reduce environmental pollution caused by coal combustion.It is very difficult to monitor slag layer thickness parameters in cold-wall gasifiers,and too thick slag layer can easily cause slag blockage,and too thin slag layer will corrode the wall surface of the gasifier,so it is very important to monitor the slag layer thickness parameters.The existing engineering monitoring data is limited,and the calculation of numerical simulation is relatively cumbersome.This paper proposes to establish a simple and fast slag layer prediction model through a data-driven method combined with the data of numerical simulation.The data-driven method is to complete the prediction of new working conditions through the learning of historical data.Its advantage is that it is simple and efficient to put aside the cumbersome equation derivation process.Due to its advantages in solving high-dimensional and nonlinear problems,neural networks have become the most popular data-driven modeling tool.The main research contents and conclusions of this article are as follows:1.A modeling method using neural network to solve the temperature field is proposed.The main idea is to learn the corresponding relationship between the temperature field nodes.The trained neural network can be used for iterative solution of the temperature field.The temperature field of different boundary conditions under the heat conduction model is established by the finite difference method,and the temperature field data is processed according to the relationship between the nodes.The temperature value to be solved is selected as the output value of the neural network,and the nodes and thermophysical parameters related to the node are used as the input values,and the neural network is used for learning.The boundary conditions are changed and the neural network is used for iterative calculations.The calculation results show that this method has one-dimensional flat-wall heat conduction problems,twodimensional flat-wall heat conduction problems,and cylindrical heat conduction problems.The mean square error is used to describe the calculation results,and the error values are all less than 1,indicating that the neural network has strong calculation ability in different boundary conditions and different coordinate systems.2.Propose a modeling method of using neural network to establish a "temperaturephysical parameter" prediction model,and analyze the hollow cylinder.The temperature parameter can be predicted by solving the temperature field model,and the predicted value can be corrected to obtain the final calculation result.In the onedimensional steady-state problem,if one side temperature boundary and the temperature value of a certain internal node are known,the boundary temperature value can be predicted,and the calculation error is in the order of 10-4.In the one-dimensional transient problem,one side is the heat flow boundary and the other side is the adiabatic boundary.The temperature value on the adiabatic boundary can be predicted,and the maximum mean square error does not exceed 2.5,indicating that the calculation results are more accurate.In the process of predicting other thermophysical parameters,neural networks are used to learn the corresponding relationship between temperature,thermal conductivity and boundary heat flow,and the prediction of temperature on thermophysical parameters is completed under specific working conditions.3.In the slag layer prediction model,the solution model of the slag layer temperature value is first established.According to the relevant parameters of the slag layer given in the literature,the heat conduction field on the slag layer and the wall can be solved.The calculation model still uses the nerve in the previous article.Network model.When the temperature value near the wall of the furnace is determined,the thickness of the slag layer is different,and the temperature value on the side of the water wall is also different.Using the neural network to learn the corresponding relationship,when the gas near the wall of the furnace is at this temperature value,the thickness of the slag layer can be directly estimated through the temperature value on the water side,and a single-temperature prediction slag layer thickness model can be obtained.In this working condition,if the critical temperature is known,the thickness of the solid slag layer can also be predicted by the outer wall temperature.In order to better apply the model to engineering practice,the temperature near the wall of the furnace and the temperature on the water side are used to determine the temperature field.Different temperature combinations will correspond to different slag layer thicknesses.By learning the corresponding relationship through the neural network,a dual-temperature model can be obtained for calculating the slag thickness by the temperature of the near wall of the furnace and the temperature of the outer wall.If the critical temperature value is introduced,the thickness of the solid and liquid slag layer can be predicted by the three-temperature model.The neural network model chosen in this article is a more general and stable BP neural network.The neural network model is established by Matlab software.