The Application Of Neural Network Modeling Method And Data Mining In Coal Gasification Process

The gasification layer temperature and its state is the key parameter on enhancing the ability of gas making and reducing energy consumption in gasification furnace with coal as raw materials, and the measurement or prediction of these variables is the prerequisite to realize closed loop control. Due to the complex working conditions, difficult test conditions, directly testing thermometry tends to be more difficult. Therefore, this paper puts forward a modeling method which combines data mining with neural network method using ascending temperature, descending temperature and the sum of them, which indirectly reflect the temperature of the gasification layer. The main work is as follows:1. In industry, the data from gasification process has the characteristic of irregular, small changing, having noisy data, and having the coupling between variables. Base on the data, this paper presents a dynamic linear modeling method which combines with BP neural network and RBF neural network. On the one hand, use the description of dynamic system model in order to more accurately reflect the gasification layer temperature and its changing trends, on the other hand, effectively use the variables which can be controlled in the system. The algorithm do the correlation analysis on the data firstly, depending on the result establish dynamic linear model, and the error between the dynamic linear model and real data is described by neural network.2. Meanwhile, some commonly used modeling methods are studied in this paper, including ARMAX model, multiple linear regression model, BP neural network model, RBF neural network model, BP neural network model (PCA-BP) based on principal component analysis and multiple linear regression combines with BP and RBF neural networks models. In addition, through the learning and research on the hierarchical clustering theoretical knowledge, a data compensation algorithm based on hierarchical clustering is proposed in this paper, and have a simulation with the missing data from gas-making process.3. Using the real data from the gasification plant, this paper does lots of simulation researches on modeling methods with the ascending temperature, the descending temperature and the sum of them in gasification system, and contrasts the results between the proposed algorithm and BP neural network, RBF neural network, PCA-BP model, ARMAX model, multiple linear regression model to illustrate the validity and superiority of the algorithm mentioned herein. The results show that the proposed modeling method has the characteristic of high accuracy than other models, reducing training time of BP neural network, and because of the variables of the modeling methods proposed which are controlled in gasification system, so that the model can be used in practical applications. Therefore, the research of this paper is in favor of predicting and controlling data from coal gasification systems, and establishs the prerequisites for the simulation optimization of factory control system.