基于PCA和PSONN的动态软测量建模与应用

Dynamic soft sensor modeling based on PCA and PSO neural network and application

针对催化裂化主分馏塔过程的数据特点和当前静态软测量技术的不足之处,提出了一种新的基于主元分析和微粒群神经网络的动态软测量方法,该方法首先利用PCA技术对过程变量进行主元分析,降低了输入变量的维数且消除了输入变量之间的线性相关性,然后根据汽油干点及其对应时刻的过程变量建立静态神经网络,利用该静态神经网络估计历史时刻的汽油干点。最后利用估计的干点值建立动态神经网络,该建模方法反应了系统的动态特性。还将微粒群算法(PSO)用于神经网络的学习过程,提高了模型的预测精度。对比结果表明了该方法的有效性。

Based on the characteristics of the ges of static soft sensor, a new dynamic soft sensor data of FCCU main fractionator process and the disadvanta- based on principal component analysis （PCA） and particle swarm optimization neural network （PSONN） is presented. The PCA method is incorporated into the network for principal component analysis of process variable, which solves the linear correlation of the input. Based on the laboratory data of gasoline end point and corresponding process variables, a static neural network soft sen- sor model is established, which is used to compute the values of the history gasoline endpoint. A dynamic soft sensor model is finally built based on the predicted endpoint. This method not only reflects the dynamic char- acteristics of main fractionator of the fluid catalytic cracking unit, but also applies the particle swarm optimiza- tion （PSO） in dynamic neural network training, which improves the model~ prediction accuracy. The results of comparison show that the proposed method is feasible and effective in soft-sensor of gasoline endpoint.