Prediction Of Alumina Concentration In Aluminum Electrolytic Cell

Electrolytic alumina is widely used to prepare aluminum in modern aluminum industry the concentration of alumina,the alumina concentration directly affects the balance of materials and energy in aluminum electrolytic cell.Too low concentration of alumina will lead to frequent anode effect of electrolytic cell,sharp increase of energy consumption and serious fluctuation of electrolytic cell status;Too high concentration of alumina will lead to precipitation on the bottom of the aluminum electrolytic cell and crust on the side of the aluminum electrolytic cell,and affect the service life of the aluminum electrolytic cell.A study indicates that when the alumina concentration is between 1.5% and 3.5%,the aluminum electrolytic cell is stable and the fluctuation of the aluminum electrolytic cell state is less;When alumina concentration is 2.0%~3.0%,it has higher current efficiency and lowest energy consumption.Therefore,the accurate control of alumina concentration in aluminum electrolysis cell is the key to maintain stable the aluminum electrolytic cell state and reduce energy consumption.The aluminum electrolytic cell is high heat,strong corrosion,strong current,strong magnetic field environment,the concentration of alumina cannot be directly measured by sensor.The traditional method of measuring alumina concentration is to manually sample the electrolyte solution,and then use the chemical method to test its composition to obtain the accurate alumina concentration.This method of obtaining electrolyte concentration has a 2~4hour lag and consumes a lot of manpower and material resources,so the amount of alumina concentration data of a single electrolytic cell is very few.Thanks to the development of computer technology,modeling is often used in modern industry to predict the industrial parameters that are difficult to obtain.Most of the existing alumina concentration prediction models are characterized by easily detectable parameters such as slot resistance,and are predicted by neural networks or their improved models.The accuracy of these models is based on the data with a large amount of alumina concentration.When the data of alumina concentration is few,the prediction effect of the existing model on alumina concentration is poor and cannot meet the requirement of alumina concentration control.My research by comparison and analysis of the existing alumina concentration prediction model,In view of the lack of alumina concentration data in a single the aluminum electrolytic cell,a prediction model of alumina concentration was established with Bayesian theory as the core.The specific research contents are as follows:1.The existing data were synthesized and the literature related to alumina concentration in the aluminum electrolytic cell was reviewed,nine factors affecting the concentration of alumina in aluminum electrolytic cell,such as aluminum electrolytic voltage,aluminum electrolytic resistance and the material addition time interval,were selected in this study,which laid a foundation for the establishment of subsequent models.2.In addition to the alumina concentration,all the data obtained in the factory are image data.Firstly,the data on the image are extracted,and the extracted data are interpolated.According to the sampling time of alumina concentration.Secondly,the data of alumina concentration were corresponded with other parameters at that time.K-means algorithm was used to perform cluster analysis on aluminum electrolytic current,aluminum electrolytic voltage,aluminum electrolytic resistance and alumina concentration to obtain the distribution characteristics of data.Alumina concentration as the main object,analyze the clustering results of three kinds of the aluminum electrolytic cell.3.According to the results of the state analysis of aluminum electrolytic cell,a set of data with the most stable state and the best alumina concentration was selected for the regression experiment.The theory of linear regression,Bayesian regression and ridge regression was compared and analyzed,and the Bayesian linear regression was improved in view of the errors and noise in the prediction data of alumina concentration,the Bayesian ridge regression model was selected for experiment.The Bayesian ridge regression model and the linear regression model were established,and the accuracy of the two models was compared in the large data set and the small data set respectively.Taking the error less than 2.5% as the evaluation standard for the accuracy of single predicted value,the accuracy of Bayesian ridge regression model in 298 large data sets was 87.3%,which was basically the same as that of linear regression 87%.However,the prediction accuracy of Bayesian ridge regression in the small data set of 24 groups reached 83.5%,higher than the 75% accuracy in the small data set of linear regression.Bayesian regression has obvious advantages in small data sets.4.By analyzing the prediction model of alumina concentration and combining the influence of the material addition time interval and anode distance on alumina concentration in aluminum electrolytic cell,the adjustment scheme of the material addition time interval and anode distance for different alumina concentrations was proposed.