基于人工神经网络的铝熔体中氢的预测

孟庆格; 边秀房; 陈建奇

浏览量 : 0 下载量: 51 CSCD: 3

导出
分享
收藏
专辑

基于人工神经网络的铝熔体中氢的预测
Prediction of the Hydrogen Content in Molten Aluminum Based on Artificial Neural Network
2002年第5期页码：1-3
纸质出版日期： 2002 ，
DOI：

扫描看全文

[1]孟庆格,,,,,,,,,,边秀房,陈建奇.基于人工神经网络的铝熔体中氢的预测[J].特种铸造及有色合金,2002(05):1-3+3.

Meng Qingge Bian Xiufang Chen Jianqi. Prediction of the Hydrogen Content in Molten Aluminum Based on Artificial Neural Network[J]. Special Casting & Nonferrous Alloys, 2002,(5):1-3.
[1]孟庆格,,,,,,,,,,边秀房,陈建奇.基于人工神经网络的铝熔体中氢的预测[J].特种铸造及有色合金,2002(05):1-3+3. DOI：

Meng Qingge Bian Xiufang Chen Jianqi. Prediction of the Hydrogen Content in Molten Aluminum Based on Artificial Neural Network[J]. Special Casting & Nonferrous Alloys, 2002,(5):1-3. DOI：

摘要

用HyscanⅡ型测氢仪测定了铝熔体在不同温度和保温时间下的氢含量

通过对BP人工神经网络的分析和改进

采用了结构为 2 4 2 1的BP神经网络模型

用所获得的试验数据对其进行训练和测试

当BP神经网络经过 3× 10 5次学习后

最大训练误差（MaxTrainingError）和训练均方差（RMSTrainingError）分别为 0 .5 5 %和 0 .18%

同时相应的最大测试误差（MaxTestError）和测试均方差（RMSTestTraining）分别达到了 0 .72 %和 0 .3 3 %

对铝熔体中氢的预测达到了很高的精度

从而建立了熔炼条件（温度、保温时间）和氢含量的映射模型

Abstract

Hydrogen content in molten aluminum at different temperatures and holding times was examined by means of Hyscan Ⅱ hydrogen analyser. The improved BP neural network modeling with 2 4 2 1 structure was adopted for training and testing by experimental data obtained. After learning for 3×10 5 times

max training error and RMS training error of BP neural network are 0.55%and 0.18%

corresponding to 0.72% max test error and 0.33% RMS test training respectively. The prediction precision is so high for the hydrogen content in molten aluminum that the response model between smelting condition （temperature

holding time） and hydrogen content may be established.