Research On Fault Diagnosis And Remaining Life Prediction Method Of Aluminum Electrolytic Cell Based On Big Data Mining Technology

The structure of aluminum electrolysis equipment is complex,and the production environment is harsh.It is affected by electric,magnetic,thermal,flow,stress,concentration and multi-field coupling.Frequent failures result in reduced current efficiency,sharp increase in energy consumption,and polluting gas emissions.Increased,causing major safety accidents such as equipment damage and casualties.The economic loss of aluminum electrolysis cell due to overhaul and shutdown is even greater.Predicting the remaining life of the aluminum electrolysis cell,and then formulating a scientific and reasonable aluminum electrolysis cell overhaul plan is a technical problem that the aluminum electrolysis industry focuses on.Based on the real historical production data of aluminum reduction cells,this paper uses big data mining technology and deep learning theory to study aluminum reduction cell anode effect fault diagnosis and cell remaining life prediction methods,and has achieved the following research results:(1)This article innovatively applies the deep learning theory to the anode effect fault diagnosis and prediction task in the aluminum electrolysis production process.Aiming at the cell parameters with different data characteristics,a new LSTM-SDAE-RF anode effect fault fusion prediction model was established.(2)In terms of model optimization,the Batch Normalization algorithm is introduced,and gradient checking is added to the model running process,which improves the stability and convergence speed of the model,improves the traditional Adam algorithm,introduces learning rate attenuation,speeds up the model learning speed,reduces noise,and improves Improved model performance.Simulation test results show that the model classification accuracy rate reaches 94.37%,the F1-score reaches 0.9254,and the advance forecast time reaches 16 minutes.Compared with the single LSTM network model,the fusion model classification accuracy rate and F1 score increase by 4.3% and 8.17% respectively;After field tests,the average classification accuracy of the fusion model reached 90.83%,and the average F1 score reached 0.8488.(3)Based on the life cycle data of aluminum electrolysis cell,data visualization analysis and research were carried out from the following three aspects: the change of main process parameters of aluminum electrolysis production with the running time of aluminum electrolysis cell;the clustering of parameters of electrolysis cell conditions of different cell ages;aluminum electrolysis The relationship between the overall distribution of the main process parameters of the cell and the age of the cell,the following conclusions are drawn: 1)The voltage,current,aluminum output and other parameters change significantly with the increase of the operating time of the aluminum reduction cell;2)In the same life span There are similarities among the electrolytic cells in the range of cell condition parameters;3)The main process parameters of aluminum electrolytic cells of different cell ages have significant differences in data distribution.(4)Innovatively proposed the maximum correlation minimum redundancy method based on mutual information to select the characteristics of the remaining life data of the aluminum reduction cell,select the effect coefficient,the amount of alumina,and the fluoride salt,etc.10 main characteristic parameters,and establish a support vector The regressed tank remaining life prediction model uses genetic algorithm to optimize its hyperparameters,and sets multiple evaluation indicators to comprehensively evaluate the performance of the model.The test results show that the average absolute error of the remaining life prediction of the support vector regression model optimized by the genetic algorithm is 175.92 days(the current industry average absolute error exceeds 400 days),and the coefficient of determination reaches 0.8722.Compared with similar regression models,the remaining life of the tank is The average absolute error of prediction is reduced by 21.4%,and the coefficient of determination is increased by14.7%.