摘要
为了探究适合全射流喷头多因素下射程的预测模型,通过改变喷头工作压力、安装高度、喷嘴直径、喷头仰角共4个参数,对射程进行测量.基于BP神经网络和广义径向基(RBF)神经网络的基本原理和算法,建立了全射流喷头射程预测的BP和RBF神经网络模型,并分析BP和RBF神经网络的预测性能.结果表明射程与工作压力、喷嘴直径呈非线性关系;当喷头在1.2 m安装高度、27°仰角、4～10 mm喷嘴直径时,压力增大到0.4 MPa,射程趋于极限,并且安装高度与射程呈正相关关系.BP与RBF神经网络均能较好地表达全射流喷头射程与主控因素之间的非线性关系.在训练时间方面,RBF网络比BP网络慢8.05 s;预测过程中,BP网络在每次运行程序时的预测结果不一定相同,而RBF网络则不会出现此问题,且RBF网络预测值与实测值之间的平均绝对误差比BP网络的小3.55%.从网络预测总体效果观察,RBF神经网络预测喷头射程具有更好的推广能力.
To explore a suitable multi-parameter model for predicting ranges of completely fluidic sprinkler,the ranges of a fluidic sprinkler were measured by varying four parameters such as installation height,diameter,nozzle tilt angle and working pressure.Based on the basic principles and algorithms of BP neural network and generalized radial basis (RBF) neural network,BP and RBF neural network models for range prediction of the fluidic sprinkler were established to analyze the prediction performance.The results show that the range is related nonlinearly to the working pressure and nozzle diameter.For the nozzle installed at 1.2 m height,27° tilt angle and with a diameter ranged in 4-10 mm,the range tends to reach a limit with the working pressure increased to 0.4 MPa,and the installation height is correlated positively with the range.Both BP and RBF neural networks can better express the nonlinear relationship between the range and the main control parameters of the fluidic sprinkler.In terms of training time,the RBF network is 8.05 s longer than the BP network.In the prediction process,the BP network program can produce the same result unnecessarily in each run,however,this problem doesn′t exist in the RBF network.Further,the average prediction error of the RBF network against the measurements is 3.55% smaller than that of the BP network.Based on the overall effectiveness of both networks in prediction,the nozzle range predicted by RBF neural network model has an even better applicability.
作者
胡广
朱兴业
袁寿其
张林国
李扬帆
HU Guang;ZHU Xingye;YUAN Shouqi;ZHANG Linguo;LI Yangfan(National Research Center of Pumps,Jiangsu University,Zhenjiang,Jiangsu 212013,China)
出处
《排灌机械工程学报》
EI
CSCD
北大核心
2019年第3期263-269,共7页
Journal of Drainage and Irrigation Machinery Engineering
基金
"十三五"国家重点研发计划项目(2016YFC0400202)
江苏高校优势学科建设工程项目(PAPD)
关键词
全射流喷头
射程
预测
BP神经网络
RBF神经网络
complete fluidic sprinkler
range
prediction
BP neural network
RBF neural network
