摘要
平均粒径是气固流化床反应器运行过程中需要监控的重要参数之一,首次提出了利用声波信号对床内颗粒平均粒度进行检测的方法,该方法安全环保不侵入流场,能克服传统方法不能实时在线测量的缺陷.对于接收仪获得的声发射信号,先用sym8小波变换进行六尺度分解,求出各细节信号小波系数的绝对值加和,标准化之后进行主成分分析,主成分分析可以消除原自变量间的复共线性,减少变量的个数.以所得主成分作为自变量,颗粒的平均粒度作为因变量,并由一4-8-1结构的三层前传神经网络为预测模型,所建神经网络结构简洁,根据声信号对平均粒度的预报准确性高于98%.
Average particle size is one of the key parameters, which need to be monitored for fluidized bed reactors. A novel method to detect the average particle size in the fluidized bed reactors by acoustic emission (AE) signals was firstly put forward, which is secure and non-invasive. The method overcome the disadvantage of traditional methods that the average particle size can't be measured at real time and on line. Original AE signals received from a detector were firstly decomposed by the discrete wavelet transform (DWT) with symmetric level eight wavelet (sym8). The absolute wavelet coefficients of the decomposed signals were then summed. After the normalization for summation, the principal component analysis (PCA) was applied in order to eliminate co-linearity between the input variables and reduce the number of variables. A 4-8-1 regression neural network was used to predict the average particle size with the principal components as independent variable and the average particle size as dependent variable. Structure of this neural network is simple and the accuracy for predicting the average particle size is higher than 98%.
出处
《中国矿业大学学报》
EI
CAS
CSCD
北大核心
2008年第2期220-224,共5页
Journal of China University of Mining & Technology
基金
国家自然科学基金重大项目(20490200)
关键词
多尺度
离散小波变换
流化床
神经网络
multiscale
discrete wavelet transform(DWT)
fluidized bed neural network