The Soft-sensor Application For Temperature Of Aluminum Reduction Cells Based On Twin Support Vector Regression

With the rapid development of China’s socialist cause,the aluminum electrolysis industry plays an important role in the national economy.In the process of aluminum electrolysis,the temperature of aluminum reduction cells plays a role of control center and is one of the most important parametric variables affecting the current efficiency.However,the process of aluminum electrolysis is characterized by high temperature,strong corrosion,non-linearity,and time lag,causing the temperature of aluminum reduction cells difficult to measure.Therefore,measuring the temperature of aluminum reduction cells effectively is an important control target for the stable production of aluminum electrolysis.This paper adopts soft-sensor technology to implement measurement on the temperature of aluminum reduction cells.At present,support vector regression is one of the most popular methods for soft-sensor in aluminum reduction cells.The only shortcoming is that the training time is too long.Based on the support vector regression algorithm,this paper introduces the Twin Support Vector Regression(TSVR).Twin Support Vector Regression has obvious advantages over Support Vector Regression.TSVR is a lower time-complexity and stronger generalization of the soft-sensor model.Finally,based on the TSVR,the paper builds and regulates the soft-sensor model for the temperature of aluminum reduction cells.The study is as follows:1.A soft-sensor model for the temperature of aluminum reduction cells based on Improved Twin Support Vector Regression(ITSVR).TSVR only needs to solve two sets of smaller quadratic programming problems in the optimization solution,and its training time is only one quarter of the support vector regression.However,TSVR does not adopt the principle of minimizing structural risk.In order to decrease over-fitting risk,the L2 norm regularization term is introduced in the TSVR algorithm to limit the upper and lower boundaries of the TSVR and implemented the principle of minimizing structural risk.The paper uses ITSVR algorithm to build soft-sensor model for the temperature of aluminum reduction cells.Principal component analysis is used to reduce the dimension of auxiliary variable in the input variables,simplify the model.Finally,MATLAB experiments are used to prove that ITSVR algorithm has low time-complexity and high generalization.Soft-sensor model for the temperature of aluminum reduction cells based on ITSVR predict temperature effectively.2.The application of incremental ITSVR soft-sensor model for the temperature of aluminum reduction cells.Soft-sensor model may predict temperature invalid due to continuous changes in the production status of aluminum electrolysis.Therefore,it is necessary to regulate soft-sensor model for the temperature of aluminum reduction cells.During the modeling process,the incremental learning algorithm is used to dynamically update the soft-sensor model.At the same time,the traditional incremental learning process easily overlooks non-supported vector samples that may be converted to support vectors.In the process,the boundary region support vector is defined,and the boundary region support vector is used to better retain useful non-supported vectors.Finally,the validity of the incremental ITSVR in the correction process of the soft-sensor model for the temperature of aluminum reduction cells and boundary region support vector was proved by MATLAB simulation experiments.