基于模糊神经网络的球团密度在线测量  

On-line Measurement of Pellet Density Based on Fuzzy Neural Network

作　　者：谢又成[1] 章兢[1] 任萍[2] 樊绍胜[1] 

机构地区：[1]湖南大学电气与信息工程学院,湖南长沙410082 [2]长沙理工大学能源与动力工程学院,湖南长沙410076

出　　处：《湖南大学学报（自然科学版）》2005年第6期52-56,共5页Journal of Hunan University:Natural Sciences

基　　金：教育部科学技术研究重点资助项目([2001]224)

摘　　要：提出了一种在线测量球团密度的新方法.该方法以模糊神经网络技术为基础,综合考虑制团过程各因素对球团密度的影响,建立起球团密度的软测量模型.在软测量模型中,采用模糊神经网络模型描述球团密度变化的非线性过程,并提出了一种改进的模型辨识算法,利用减法聚类法确定合适的聚类组数目,并用实数编码的遗传算法优化全局参数,从而获得了结构简单、具有较高精度的模糊神经网络软测量模型.根据此方法,设计了测量装置,并进行了现场试验,试验结果表明软测量模型输出与实验室测量值基本一致,平均误差较低且最大误差未超过0.05 g/cm3.A new on-line measurement approach for pellet density was presented. In the approach, fuzzy neural network modeling was adopted to construct the measurement model for pellet density. In the course of the modeling, fuzzy neural network model was employed to approximate the non-linearity of pellet density. An improved model-identifying algorithm was presented. Subtractive-clustering algorithm was used to determine the optimum number of clusters, and a real coded genetic algorithm was adopted to optimize model parameters. All these techniques made the fuzzy neural network model simple and accurate. Based on this, an instrument was developed and the test on an actual briquetting machine was conducted. The results showed that the proposed approach provided a result similar to laboratory measurement. The error was low and tolerable. It greatly improved the efficiency of pelletizing and laid a foundation for the optimal control of density.

关 键 词：球团密度 软测量 模糊神经网络 遗传算法 

分 类 号：TN216[电子电信—物理电子学]