Than The Prediction Of Electrolytic Aluminum Process Temperature Of The Neural Network-based And Molecular Studies

Temperature and molecular ratio are two very important parameters in aluminum reduction production process, the two parameters directly reflect balance state of the cell's, for make the cell work in the best state which can heighten the current efficiency, deduce environment pollution and prolong the working life of this cell, so it is very important to get exact values of temperature and molecular ratio.But it is not easy to measure the temperature and molecular ratio with human power, because high temperature and strong causticity in aluminum reduction cell. And due to the aluminum reduction cell is a high non-linear, multi-coupling and big delay system, so we cannot create a precise mathematic model for temperature and molecular ratio in this situation. Therefore this paper mainly present a predictive method to predict the values of temperature and molecular ratio based on RBF neural network.Firstly, this paper analyze some mainly parameters and the control strategy of temperature and molecular ratio in aluminum reduction cell, so we can get the cell voltage, the addition amount of AlF3 and aluminum level are three mainly parameters to affect the temperature and molecular. In view of the delay in the cell, so the author selected today's cell voltage, the addition amount of AlF3 and aluminum level and yesterday's cell voltage, the addition amount of AlF3 and aluminum levels, and yesterday's temperature and molecular ratio to predict today's temperature and molecular ratio, so the network which we created is a eight-inputs, two out-outputs predictive neural network model, in here selected Gaussian function(RBF) as the hidden layer node function.Secondly, using the data from the aluminum reduction cell to train the network with two different algorithms in MATLAB software, the results shows that the RBF neural network can predict the values of temperature and molecular ratio precisely, the average error is small enough used in aluminum reduction cell. But clustering method can get more accurate predictive results than Gradient descent method, The predictive errors of Gradient descent method are 3.4163 and 0.053 respectively, but clustering method's errors are 1.9145and 0.0158 respectively in 20 days.So we can use clustering method to train this RBF network.Finally, the author designed the prediction system of temperature and molecular ratio for aluminum reduction cell based on RBF neural network, using Visual C++ software as development tools, SQL Server 2000 database as back-end Database and MTALAB software as computing tool. This system can achieve training function on-line and predictive function on-line. Meanwhile some auxiliary modules were designed for better operate this system, such as login module, user management module, cell number management module, and history data inquire module, and show data curve module. After testing, the system can normal operate, each function can be realized.