Research On Time Series Classification Method For Electrical Smelting Magnesium Working Conditions Based On BRB-XGB

Electrical smelt magnesite has the characteristics of fire resistance,corrosion resistance and high strength,and is widely used in metallurgy,building materials,aerospace and other fields.The power demand of electrical smelt magnesium is different under different operating modes,and the accurate prediction of operating modes is conducive to improving product quality and power efficiency.The electrical smelt magnesium is affected by periodic,intermittent and random factors,whose operating mode changes dynamically,and has the characteristics of incomplete sample information,imbalanced categories,and difficult to predict the operating mode shift.To address this problem,this paper designs a two-layer prediction model based on BRB(Belief Rule-Base Inference)pre-training and XGBoost(e Xtreme Gradient Boosting),and processes imbalanced samples through R-IWSMOTE(Rate-IWSMOTE),so as to improve the accuracy of time series classification prediction.The specific research work includes:Firstly,a time-series classification model based on BRB pre-training is proposed to solve the problem of incomplete sample information.In the pre-training of BRB,a generative belief rule base from feature space to category discrimination is constructed,and the activation weight of the belief rule is calculated based on the weighted confidence distribution method with reliability,and the inference results of the activation rule are synthesized.The training objective is to maximize the classification accuracy of samples,and the rule weights,premise attribute weights and confidence parameters in BRB model are learned and optimized to ensure the inference’s objectivity and classification accuracy.BRB pre-training can extract coarse-grained knowledge,and its classification results can be used as a new feature to expand the feature space of samples,and the further time series classification can be performed on this basis.Four representative classifiers,i.e.,LGB(Light GBM),RF(Random Forest),SVM(Support Vector Machine)and XGBoost(e Xtreme Gradient Boosting),are selected to design four time-series classification methods based on BRB pre-training.In order to verify the effectiveness of the method,in addition to the electrical smelt magnesium dataset,the Trace and Distal Phalanx Outline Age Group public datasets are used to carry out comparative experiments on the above algorithms.Through experimental comparison,the designed BRB-XGB model has the best prediction accuracy and better effect than other single models.Secondly,a resampling algorithm R-IWSMOTE(Rate-IWSMOTE)is proposed to solve the problem of category imbalance.This algorithm uses a noise filtering mechanism based on accurate estimation of confusion information to prevent noise propagation during the generation of SMOTE synthetic instances.A soft-weighted strategy is adopted to allocate appropriate opportunities for the minority class instances to become seed instances in the SMOTE process,which further highlights the boundary.Considering the relationship between the minority class and the majority class in the training set,the rate of the minority class sample number to the majority class sample number is introduced to modify the sampling rate,which can further improve the balance of sample categories after resampling.The results show that the imbalance can be reduced and the prediction accuracy can be improved after processed by the R-IWSMOTE algorithm.Thirdly,the dynamic change of the operating modes of electrical smelt magnesium is analyzed from the two perspectives of mechanism and data,and the operating mode prediction is decomposed into two problems of operating mode shift prediction and operating mode category prediction.Six premise attributes such as the operating mode duration,the preceding operating mode,the current setting error and the current change rate are extracted as the input of BRB pre-training model,based on taking whether the electrical smelt magnesium operating mode is shifted or not as the result attribute,then the electrical smelt magnesium operating mode shift prediction based on BRB pre-training model is established.The BRB pre-training model can extract the cycle shifting rules between operating modes,and the classification results are used as a new feature to expand the feature space of electrical smelt magnesium samples to complete the sample information.On this basis,the operating mode prediction model is established based on LGB,RF,SVM and XGBoost.The electrical smelt magnesium industrial dataset is used for experimental study.To address the problem of imbalance in the ratio of operating mode shift,as well as the imbalance between operating mode categories,the R-IWSMOTE algorithm is adopted to resample the dataset to improve the data balance.A two-layer classification model based on BRB pre-training and operating mode category prediction is established to realize the time series prediction of electrical smelt magnesium’s operating mode.The experimental results show that the R-IWSMOTE-BRB-XGBoost model can meet the performance requirements of the prediction for the electrical smelt magnesium’s operating mode.