Research And Application Of Molecular Ratio Prediction Algorithm In Aluminum Electrolysis

Molecular ratio is a very important technological parameter in aluminum electrolysis production,and it is an important basis for controlling electrolyte composition and deciding additive weight.In the production process of aluminum reduction cell,the electrolyte has been in the environment of high temperature and strong corrosion,so the molecular ratio can not be detected online,which seriously affects the accuracy of aluminum reduction production decision-making.Therefore,the molecular ratio prediction algorithm model of aluminum electrolysis based on multiple time Windows is studied in this paper to solve the problem that the molecular ratio cannot be estimated online.Firstly,paper analyzes the relevant process flow in the process of aluminum electrolysis and the main relevant factors affecting the molecular ratio.According to the process characteristics of molecular ratio,the relationship between molecular ratio and fluoride,average voltage,alumina concentration and other parameters is discussed;Secondly,the data characteristics of molecular ratio and its related parameters in the process of aluminum electrolysis are studied.On the basis of data preprocessing,combined with the characteristics of time series,the stationarity test and feature selection are carried out;Finally,in view of the current situation that the molecular ratio of each electrolytic cell is only sampled once or twice a week for assay in the actual electrolytic production,the prediction algorithm model of molecular ratio is studied and developed through multiple time windows.For least squares support vector machine(SVM),hyper-parameters determination of impact to the problem of the prediction accuracy and generalization ability of the data using artificial colony algorithm combining with molecular ratio characteristics of least squares support vector machine was optimized parameters combination,at the same time introduced a tabu search list more easily solved the artificial colony algorithm in the shortcoming of local optimum;Based on the advantage that autoregressive recurrent neural network is good at predicting time series data,and combined with the characteristics of molecular ratio process,the moving weighted average is introduced to improve the cell unit in autoregressive recurrent neural network;Finally,multiple linear regression algorithm was used to synthesize the prediction results of three algorithms based on the data of different time Windows: the improved artificial bee colony least-squares support vector machine,the improved autoregressive recurrent neural network and gaussian process regression,and the experiment proved that the prediction results were better.The experimental results show that it has achieved good prediction results.On this basis,the molecular ratio prediction subsystem of aluminum reduction cell applied to electrolytic aluminum production is developed.