The Quality Prediction Of Tabular Alumina Based On Support Vector And Ensemble Learning

Tabular alumina is a kind of refractory material,its production target has converted from increasing output to improving quality.Previous research has shown that the machine learning is effective in the similar chemical field,which improves the production efficiency and saves experimental costs.The machine learning methods can help enterprises predict the product quality and find out key process parameters of tabular alumina,which has important application value and theoretical guidance to the enterprises.In this thesis,we propose a method to predict the quality of tabular alumina based on support vector and integrated learning.First of all,we use Lagrange interpolation and other methods to integrate and clean data set according to the characteristics of production.Then,Adaptive-Lasso is used to reduce the number of features for the model of support vector regression.Besides,we build three predictive models by Support Vector Regression,Random Forest Regression and XGBoost.After the comparision and analysis of results,we choose Random Forest Regression as the best method for the prediction of corundum quality.Finally,the different feature selection results are given by these algorithms.We calculate the weighted average of results as the final rank of feature importance,which can achieve the purpose of providing key process parameters.The experimental results show that the regression models in this thesis have small mean square error.These models are able to meet the actual requirements basically.This thesis proposes the methods of predicting continuous quality index of tabular alumina,provides the top 10 important process parameters.The result is valuable and meaningful to improve tabular alumina quality and production efficiency for relevant decision-makers.