Abstract：Aiming at the features of styrene, soft sensor based on the real-time data was applied to predict some key process variables. Two methods, BP neural network and partial least squares, were used to model the unmeasured process variables. At the same time, real-time process data were used to train the model in order to make sure the constructed model endure a good fault-tolerance. The experiments for total conversion, first reactor conversion, second rector conversion of ethylbenzene and selectivity of styrene were given based on BP neural networks and PLS, which showed the proposed methods could predict the dynamic performance of some key variables. It is very useful to monitor the important variables for the advanced control and optimization of styrene production.
