基于PSO算法优化的BPNN天然气脱CO_2膜分离过程软测量模型  

Optimized BPNN Soft Sensor Model for Membrane Separation Process of CO_2 Separating From Natural Gas Based on PSO

作　　者：王磊[1] 李桂香[1,2] 王元麒[3] 

机构地区：[1]海南大学信息科学技术学院海南省特种玻璃重点实验室,海南海口570228 [2]广东省医疗器械研究所 广东省医用电子仪器及高分子材料制品重点实验室,广东广州510500 [3]意大利米兰理工大学,意大利米兰20133

出　　处：《海南大学学报（自然科学版）》2015年第1期28-33,共6页Natural Science Journal of Hainan University

基　　金：国家科技支撑计划课题(2012BAA10B03);国家自然科学基金(61463011);广州市科技计划项目(201334500076)

摘　　要：天然气脱CO2膜分离过程是一个非线性复杂系统.针对其过程中一些关键性能参数如脱碳气流量和尾气CO2浓度难以在线精确测量,建立了基于PSO算法优化的BP神经网络天然气脱CO2膜分离过程软测量模型.其中,为克服BP神经网络收敛速度慢、易陷入局部最小值等缺点,引入惯性权重算法和收缩因子对传统PSO算法进行改进后,用PSO算法优化BP神经网络的权值和阈值;然后,基于Matlab软件和采集的现场数据对天然气脱CO2膜分离过程进行建模仿真.结果表明,模型收敛速度快,模型测量的天然气脱CO2膜分离过程中的关键参数脱碳气流量和尾气CO2浓度值与实测值符合较好,误差小.The membrane separation process of CO2 separating from natural gas is a complex and non-linear system.Aimed at the problemes that some key performance parameters,such as decarburization gas flow and CO2 concentration of decarburization gas,is difficult to be measured,a PSO optimized BP neural network soft measurement model for gas membrane separation process was constructed.Firstly,BP neural network soft sensor model for membrane separation process of CO2 separating from natural gas based on PSO was constructed,and in which,in order to solve the problems of BP networks such as slow training speed and local minimum point,the inertia weight algorithm and shrinkage factor were used to optimize the weights and threshold parameter; then,based on MATLAB and the collected field data,a simulation experiment was performed.The results indicated that the model convergence very fast,and the prediction results of the key performance parameters of CO2 separating from natural gas membrane separation process were in reasonable agreement with the measurement values,and the maximal relative errors are very small.

关 键 词：气体膜分离 粒子群 BP神经网络 软测量 

分 类 号：TQ028.83[化学工程] TP183[自动化与计算机技术—控制理论与控制工程]