论文部分内容阅读
提出了一种基于误差高斯混合模型(EGMM)的高斯过程回归(GPR)软测量方法.首先,选择合适的变量组成误差数据集,利用贝叶斯信息准则优化得到合适的高斯成分的个数;然后用EGMM对误差数据进行拟合计算得到条件误差均值和方差的表达式;最后当新的数据到来时,用建立的GPR模型进行输出预测,并利用EGMM模型得到的条件误差均值对输出进行补偿,从而得到更加精确的建模结果.通过数值仿真及硫回收装置(SRU)的H2S浓度的软测量,进一步验证所提算法的可行性和有效性.
A Gaussian Processes Regression (GPR) soft-sensing method based on Error Gaussian Mixture Model (EGMM) is proposed.Firstly, an appropriate set of error data sets is selected and the number of suitable Gaussian components is optimized by Bayesian information criterion. Then, the error data is fitted by EGMM to obtain the expression of mean and variance of conditional errors. Finally, when new data arrives, the output is predicted by the established GPR model and the output is compensated by the mean value of the conditional error obtained by the EGMM model , So as to obtain more accurate modeling results.Furthermore, the feasibility and validity of the proposed algorithm are verified by numerical simulations and soft measurement of H2S concentration in sulfur recovery unit (SRU).