中文摘要：

生产装置的＂假液位＂现象严重影响了工业生产过程的正常运行。为此,运用软测量思想建立数学模型,监测直接测量结果,防止假液位的发生,保障生产安全。通过比较机理建模法与基于数据驱动建模法的优劣,提出采用PSO-SVM法建立数学模型,并以SBR泄料槽液位的监测为例进行分析。以集管进料压力P1、泄料槽入口温度TA丁二烯分离系统压力pO这3个量作为辅助变量来预测泄料槽液位L。结果表明,该模型预测值与实际值符合良好,具有较强的预测性能,能够较好地对SBR泄料槽液位进行监测,有效避免SBR泄料槽假液位的产生。

英文摘要：

This paper is aimed at introducing a renovated method for liquid level monitoring of the processing instruments based on the PSO- SVM principle in order to solve the problem on how to apply the idea of soft measurement to the monitoring of the industrial production.The said renovated method is intended to be used directly for measuring the production process to ensure the production safety by avoiding the false level interference in it.However,there still exist some assumptions made by using the method,saying that it cannot fully reflect the reality of the actual industrial production system.Therefore,as a means of supplementation,we have proposed a so-called PSO-SVM method for the liquid level monitoring of the processing instruments based on the data-driven modeling under the support of the vector machine sparse for the small samples with the nonlinear high dimensional data.In so doing,it would be possible to overcome the known ＂dimension disaster＂ by avoiding being trapped into the local minimum dots.However,on the whole,the method we have developed proves to perform successfully in generalization ability and stand out in the data-driven modeling methods.In order to optimize the functions of the model,we have managed to use the particle-swarming optimal algorithm to optimize the parameters based on the SVM model.Last of all,we have also established a mathematical model based on the PSO- SVM for analyzing the liquid level monitoring of the SBR discharge chute by taking into account the three auxiliary variables,that is,the feed pressure of manifold p_1,the inlet temperature of the discharge chute T_A and the pressure of the butadiene separation system p_0 so as to get the level of the discharge chute L.The reliability and validity of the application results of the model show that the predicted values prove to be well in accord with the actual ones,wheres the model turns to be successful in operation as is predicted.Thus,the model we have established can be taken as a reference to the effective monitoring of