Temperature Prediction Model Identification Using Cyclic Coordinate Descent Based Linear Support Vector Regression

Temperature prediction plays an important role in ring die granulator control,which can influence the quantity and quality of production. Temperature prediction modeling is a complicated problem with its MIMO, nonlinear, and large time-delay characteristics. Support vector machine( SVM) has been successfully based on small data. But its accuracy is not high,in contrast,if the number of data and dimension of feature increase,the training time of model will increase dramatically. In this paper,a linear SVM was applied combing with cyclic coordinate descent( CCD) to solving big data regression. It was mathematically strictly proved and validated by simulation. Meanwhile,real data were conducted to prove the linear SVM model's effect. Compared with other methods for big data in simulation, this algorithm has apparent advantage not only in fast modeling but also in high fitness.

Temperature prediction plays an important role in ring die granulator control, which can influence the quantity and quality of production. Temperature prediction modeling is a complicated problem with its MIMO, nonlinear, and large time-delay characteristics. Support vector machine （SVM） has been successfnlly based on small data. But its accuracy is not high, in contrast, if the number of data and dimension of feature increase, the training time of model will increase dramatically. In this paper, a linear SVM was applied combing with cyclic coordinate descent （ CCD） to solving big data regression. It was mathematically strictly proved and validated by simulation. Meanwhile, real data were conducted to prove the linear SVM model＇s effect. Compared with other methods for big data in simnlation, this algorithm has apparent advantage not only in fast modeling but also in high fitness.