中文摘要：

软测量技术是石化生产过程中在线监测油品难测性质的重要手段。本文提出了一种基于NARX神经网络的软测量仪表用于原油蒸馏装置中油品关键性质的在线预测。首先,利用流程模拟软件建立了原油蒸馏过程的动态模型。然后,基于动态模型的阶跃实验数据,建立了以装置操作变量为输入、油品关键性质为输出的NARX神经网络预测模型,并提出能有效减少模型预测误差的修正方法。仿真实验结果表明,所提出的误差修正方法可明显减少预测结果中的“大误差点”,降低根均方误差,因此,所建立的软测量仪表可用于油品关键性质的在线预测。

英文摘要：

Soft-sensing technique is an important method to online monitor key products properties in petrochemical process .This paper proposed a kind of soft sensor based on nonlinear autoregressive with external input （NARX） neural network which is used in online prediction of key products properties in the process of crude distillation. Firstly, a dynamic model of crude distillation unit was estab- lished. Then the NARX neural network prediction model, of which the inputs and the outputs were following operation conditions and key products properties respectively, was built using the data from step-test experiment of the dynamic model. Then a kind of correction method was presented to reduce the prediction error. The simulation results showed that the correction method can effectively reduce the number of bad predictions and the root mean square errors. And the soft sensor can precisely estimate the key products properties.