Research Of Aluminum Electrolysis Temperature Control Based On Neural Network

Current efficiency is the most important technical and economic index to evaluate aluminum reduction cells in the aluminum electrolysis process.Temperature is the key factor affecting the current efficiency.In a working electrolytic cell,temperature can be maintained at the optimum range by adjusting some parameters.Then the current efficiency reaches the highest level and the electrolytic cell is in optimum condition.Therefore,it is very meaningful to control aluminum electrolysis temperature.The aluminum electrolysis process is a large time-delay process,the effect of temperature control can not be displayed in real time,the traditional control technology is not ideal.Based on the radial basis function neural network,the aluminum electrolysis prediction system is constructed according to the historical data of the normal operation of the electrolyzer.When the temperature of the aluminum cell is given,the value of AlF3 addtion and aluminum tapping volume can be calculated,so as to achieve the system requirements.This paper firstly analyzes temperature dependent parameters in the aluminum electrolysis process,results show that the parameters such as voltage,AIF3 addition,aluminum tapping volume,the molecular ratio and the aluminum level have an important effect on the temperature.The controllable parameters in the production are AIF3 addition and aluminum tapping volume.As the aluminum electrolysis is a large delay system,the author selected yesterday's voltage,AIF3 addition,aluminum tapping volume,the molecular ratio,temperature and the aluminum levels,and today's voltage,the molecular ratio,the aluminum levels and today's desired temperature to predict today's AIF3 addition and aluminum tapping volume.Therefore,a three-layer neural network model with ten-input and two-outputs is constructed.Secondly,using the data from the aluminum cell to train the network with four different algorithms(Gradient descent algorithm,k-means clustering algorithm,orthogonal least squares algorithm and subtractive clustering algorithm),the results shows that the network with OLS algorithm training can predict the AIF3 addition and the aluminum tapping volume precisely.In No.3507 electrolytic cell,the maximum error of predictive AIF3 addition is 1,the average error is 0.45 and the percentage of average error is 3.59%,the maximum error of predictive aluminum tapping volume is 211kg,the average error is 58.1kg and the percentage of average error is 2.53%.Finally,the author designed a prediction system by mixed programming among SQL Server,MATLAB and GUI toolbox.When the desired temperature value is modified,the decision value of AIF3 addition and aluminum tapping volume can be obtained.