New Method For Prediction And Determination Of Key Physicochemical Parameters Of Complicated Aluminum Electrolyte

The electrolyte is the carrier medium for aluminum electrolysis production to dissolve alumina and produce aluminum through electrochemical reduction,in which various physical and chemical reactions occur,therefore,the electrolyte directly affects the power consumption,production quality and cell life.Compared with the traditional aluminum electrolyte system,the chemical composition of complicated aluminum electrolyte system is beyond the range of standard electrolyte components,and the changes of physical and chemical properties have led to lower efficiency,higher energy consumption,more precipitation and control difficulties in electrolysis production.Focused on the strategic objectives of improving quality and efficiency,energy saving and consumption reduction,transformation and upgrading of electrolytic aluminum industry,exploring the prediction and measurement of key physicochemical parameters such as liquidus temperature and cryolite ratio of complicated electrolyte is of great significance for improving the production process of aluminum electrolysis,realizing precise production control and promoting the intelligent upgrading of aluminum smelting.In this paper,the physical and chemical properties of complicated aluminum electrolyte system were analyzed by means of total element analysis to get the chemical composition,phase composition,physical and chemical properties such as element embedding and thermal stability,therefore,the regional distribution characteristics of complicated aluminum electrolyte system were revealed and the mapping relationship between raw materials,auxiliary materials and the formation of complicated electrolyte systems were established.By using machine learning algorithm,prediction model of liquidus temperature of complicated aluminum electrolyte system based on multi matrix types and wide composition range of training samples was established.Using laser induced breakdown spectroscopy(LIBS)technology,a chemometric method based on feature extraction and machine learing fusion has realized the quantitative analysis and determination of CR in complicated aluminum electrolyte.In situ on-line detection of CR,Ca and Mg content in molten complicated aluminum electrolyte was carried out,and for the first time,the LIBS in-situ on-line detection and analysis of the main components of a complex aluminum electrolyte system was realized.The main research results are as follows:(1)The multi-dimensional,large-capacity,multi-source data of electrolytes and raw materials combined with the analysis method of raw material regional supply coordination realized the regional mapping and association of the components between the complicated aluminum electrolyte system and raw materials.The typical physical and chemical properties of complicated aluminum electrolyte were analyzed based on the collecting samples from multiple regions in China,and the regional physicochemical characteristics of complicated aluminum electrolyte system were revealed from the macro level.The origin of complicated aluminum electrolyte system was analyzed from the alumina,carbon anode,anode covering material and carbon slag.The distribution law of impurity elements in alumina,carbon anode and anode covering material was explored.The basic mapping relationship between impurity elements in auxiliary materials and formation of complicated aluminum electrolyte.(2)A modeling method based on the coupling of total elements of complicated aluminum electrolyte components with large sample size and machine learning analysis has been used to accurately predict the primary crystal temperature of complex aluminum electrolyte system.The application scope of the model was widened and the prediction accuracy was improved.Moreover,the nonlinear relationship between liquidus temperature and chemical composition of complicated aluminum electrolyte system was revealed.The RMSE of BP-ANN model with LOOCV is 6.77,MRE is 0.54%,and the average relative error of prediction results for 39 samples is 0.39%;For the SVM(Rbf)model with LOOCV,RMSE=6.90,MRE=0.49%.The average relative error of prediction results for 39 external samples is 0.43%.The accuracy of the forecast is high,and it has the value of industrial application.(3)The LIBS experimental device was designed and built,and combined with single factor experiment to optimize the key experimental parameters of LIBS detection.The plasma temperature and electron density were calculated by characteristic analysis spectral line,the validity of plasma spectrum was confirmed,the LIBS experimental conditions were optimized,and a reasonable combination of experimental parameters was obtained.Combined with the criterion of Mc-Whirter,the laser plasma temperature is 5353K and the electron density is 1.55×1018 cm-3,which indicates that the complicated aluminum electrolyte plasma satisfies the local thermodynamic equilibrium state and the LIBS plasma spectrum is effective.The optimization conditions of LIBS parameters were determined by single factor experiment as follows:argon atmosphere,laser delay time of 4 ?s,laser energy of 133 mJ,grinding time of 30 s,electrolyte sample pressure of 8 MPa,laser pulse accumulation of 50 times,which laid a foundation for the quantitative determination of the main component for complicated aluminum electrolyte employing LIBS technology.(4)A fusion method based on spectral variable feature extraction and machine learning combined with LIBS technology was proposed for the first time to achieve quantitative determination and analysis of complicated aluminum electrolyte CR.The hyperpolyhedron method was used to screen the spectral characteristic variables,the machine learning algorithm was used to train the modeling based on the screened characteristic variables as the new data set.For the SVM(Liner)model with LOOCV,RMSE=0.062,MRE=1.79%,and the SVM(RBF)model with LOOCV,RMSE=0.027,MRE=0.93%;The average relative error of CR prediction results by SVM(Liner)and SVM(Rbf)models was 0.33%and 0.43%,and the HyperPolyhedron-SVM method showed excellent predictive ability for both training samples and validation samples of complicated aluminum electrolyte.(5)A LIBS in-situ online detection device combined with a stoichiometric analytical method was established to achieve the quantitative analysis of the main components of the complicated aluminum electrolyte in the strongly disturbed and heterogeneous molten state under high temperature environment for the first time.It is found that the SVM correction model based on full spectrum has good prediction ability,and the average relative error of CR prediction for 20 external electrolyte samples is 2.62%.The calibration curve of Ca and Mg in complicated electrolyte system was established by traditional calibration method.The calibration curve of Ca was y=6208.43x-8654.59,calibration model R=0.94,RSD=1.89%;The calibration curve of Mg is y=7120.13x+1312.60,calibration model R=0.95,RSD=3.28%;The average relative standard deviations of Ca and Mg are 5.40%and 13.0%,respectively by analyzing 10 external independent test electrolyte samples.The detection limits of Ca and Mg are 8.54 mg·g-1 and 15.50 mg·g-1,respectively.