基于LSSVM实现CO_2转化率的软测量建模

Modeling of soft sensing transformation efficiency of carbon dioxide based on LSSVM

将最小二乘支持向量机LSSVM用于中压联尿尿素生产中CO2转化率的软测量建模,并与BPNN、RBFNN模型进行比较,研究表明,最小二乘支持向量机建模快速、准确,泛化性能好。将该模型用于生产试验,预测结果与事后分析值的均方误差MSE<0．006,最大绝对误差为0．280 1,相对误差绝对值在0．1％以内的约占90％,最大相对误差绝对值仅为0．42％,模型表现稳定,效果令人满意。

Soft sensor modeling method based on Least Square SVM（LSSVM） is proposed in this paper. The sensor is used to predict the transformation efficiency of carbon dioxide in medium pressure diurea process. Compared with BPNN and RBFNN the research indicates that this method is featured with higher speed, more accuracy and better generalization ability and performance. The experiment results shows that the mean square errors of estimation out!puts and analytical values are less than 0. 006 ,and about 90% absolute relative errors are less than 0. 1%. The maximal absolute error is 0. 262 1, and the maximal absolute relative error is only 0. 42%. The model is stahle and the result is satisfactory.