Soft measurement for calorific value of cracking fuel gas based on Online SVM algorithm

Abstract

A newly adaptive online support vector regression machine(Online SVM)was proposed to improve the generalization ability of soft sensing model of calorific values of fuel gas in the cracker system that was constructed based on historical data.The approach combined the incremental support vector machine(ISVM)with approximate linear dependence(ALD)condition.New independent samples with ALD condition to update the SVM model were determined by calculating the approximate linear dependence(ALD)value between new samples and modeling samples.The influencing factors of calorific value of fuel gas of cracking furnace were analyzed,and an on-line soft sensing model of calorific values for fuel gas of the cracker system was established using Online SVM algorithm.This model consisted of off-line training module and on-line updating module.The off-line training module was mainly used to produce initially soft sensing model of calorific value based on historical data,and the on-line updating module was used to keep high predictive accuracy for on-line model of calorific value through making off-line training module to learn newly independent samples.A series of comparison simulation experiments were carried out between the proposed method and the conventional SVM and LS-SVM methods using synthetic data,benchmark data and calorific value data of cracking fuel gas.The simulation results show that the proposed method can adapt to new conditions with capability of learning new samples adaptively,and can be used for modeling of soft measurement for calorific values of fuel gas in cracker system with slow time-varying character.