基于ART-SVR的过程建模及在干点软测量中的应用

针对石油化工生产过程通常呈高度非线性，且生产过程数据呈非连续、具有一定类别特性等特征，提出基于自适应谐振神经网络（adaptive resonance theory，ART）和支持向量回归（support vector regression, SVR）相结合的建模方法（ART-SVR）。首先，基于建模样本，通过ART将样本模式空间分割成若干模式特性相近的子空间;然后，对各子空间分别采用SVR建立各自模型，实现基于样本模式空间分割的“分段”建模。仿真试验和在石脑油干点软测量建模的实际应用表明：ART-SVR模型的拟合精度和预测精度均优于全局SVR模型。

Process modeling based on ART-SVR and its application in dry point soft measurement

The petrochemical process is highly nonlinear and the observation data of the petrochemical process are non-continuous and have classified characteristics.A novel process modeling method, which combined adaptive resonance theory (ART) with support vector regression (SVR), was proposed.Firstly, ART was used to separate the input pattern space into several sub-spaces based on a modeling sample.Then, SVR was used to build up each sub-model for each sub-space.The results of simulation experiment and an application in dry point soft measurement of naphtha showed that ART-SVR could reduce the nonlinear degree of the sub-models and its fitting accuracy and prediction accuracy were both better than those of a single SVR model.