针对SMB 色谱分离过程中组分纯度难于实时在线测量的问题,建立了一种基于BP 神经网络的软测量模型。首先对实际生产中的原始数据，经过过失误差剔除及滤波处理后得到一套训练数据和校验数据组成训练样本，然后采用BP 神经网络进行训练，得到组分纯度的非参数模型。为加快网络收敛速度，采用改进BP算法对其进行训练。在MATLAB 工作平台上进行了大量的仿真研究，对该模型进行验证，仿真结果表明，该方法的有效性。关键词：模拟移动床;软测量;人工神经元网络;BP 算法;改进BP 算法Abstract: In order to realize the online measurement in SMB separation process, a soft sensor model based on BP neural network which measures two components purity is built up. First by filter process of practical primal data, the training samples are gained. Then the nonparametric model on component purity is developed by use of BP neural network. Improved BP algorithm is applied to fast convergence. Much computer simulation on Matlab validates this method.Keywords: Simulated moving bed; Soft-sensor; Artificial neural network (ANN); BP algorithm; Improved BP algorithm