基于改进型PSO-BP神经网络的SVI软测量  

Soft Measurement of SVI Based on PSO-BP Neural Network

作　　者：郭晓燕[1] 郭民[1] 韩红桂[1] 

机构地区：[1]北京工业大学电子信息与控制工程学院,北京100124

出　　处：《控制工程》2014年第6期873-877,共5页Control Engineering of China

基　　金：国家自然科学基金项目(61203099;61034008;61225016);北京市自然科学基金项目(4122006);教育部博士点新教师基金项目(20121103120020);北京市科技新星计划(Z131104000413007);香江学者计划(XJ2013018)

摘　　要：针对曝气池污泥体积指数(SVI)难于在线测量的情况,提出了一种基于改进型的BP神经网络对SVI软测量的模型,该模型利用改进粒子群优化算法对BP神经网络进行权值调整。为了提高BP神经网络的学习性能,使粒子群的惯性权值按照对数规律变化,同时引入自适应变异算子对局部最优的粒子进行变异,然后利用粗糙集理论消除冗余信息,得到输入变量。该模型对实际污水厂的SVI值进行软测量预测,并与其它软测量模型进行比较,结果表明了该软测量模型具有较好的精度。Due to the aeration tank sludge volume index （SVI） is difficult to measure online. A BP neural network is used for SVI soft sensor in this paper. To improve the predictive accuracy of the BP neural network, BP neural network based on the improved particle swarm optimization ways has been put forward, in which the inertia weight varies according to a logarithmic scale, while the adaptive mutation operator is introduced for the variation of the local optimum particle to improve the particle swarm algorithm optimization abili- ty. The rough set theory is used to eliminate redundant information and BP neural network based soft sensor model is established. Then the model is used to simulate the actual wastewater treatment plant SVI, and it is compared with other soft sensor . The results prove the reliability and validity of the soft sensor model.

关 键 词：污泥膨胀 污泥体积指数 软测量 改进型PSO-BP神经网络 

分 类 号：TP27[自动化与计算机技术—检测技术与自动化装置]