基于PCA-BP神经网络的精馏塔产品组成软测量模型

A soft-sensing model built for acetic acid distillation columns based on the PCA-BP neural network

依据工艺机理和操作经验,初选了醋酸精馏塔产品组成的神经网络预测模型的输入变量,运用主元分析方法对输入变量进行主元分解,降低输入变量维数且消除了输入变量之间的线性相关性,再通过基于LM优化算法的BP神经网络进行建模。仿真结果表明,该模型具有较快的训练速率和较高的预测精度,可以满足精馏过程对出口物料组成的在线软测量要求。

The nonlinear and timevarying characteristics make distillation columns very difficult to build a softsensing model.To solve this problem,the paper puts forward a softsensing method based on the BP neural network used to estimate those unmeasurable signals that are important for controlling a distillation process in order to improve the system control performance.The PCA(Principal component analysis)method is incorporated into the network,which not only solves the linear correlation of the input,but also simplifies the network structure and improves the training speed.The model performance has been tested.The accuracy of predictive results can satisfy the demand of the online softsensing method.