人工神经网络中隐含层节点与训练次数的优化.pdfVIP

第 35 卷 第 2 期 哈 尔 滨 工 业 大 学 学 报 VOI.35 NO.2 ， 2 0 0 3 年 2 月 JOURNAL OF HARBIN INSTITUTE OF TECHNOLOGY Feb. 2003 !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 人工神经网络中隐含层节点与训练次数的优化 高大文，王 鹏，蔡臻超 （哈尔滨工业大学 市政环境工程学院，黑龙江 哈尔滨 150090） 摘 要：目前构建定量构效关系人工神经网络模型中隐含层节点数和网络训练次数大多是依靠试验方法来 确定，针对该方法运算工作量较大、模型质量和预测精度没有保证等问题，通过编写程序获得有关网络的预 测精度和百分误差与网络隐含层节点数和训练次数之间关系的大量数据，采用 MatIab 语言分别绘制预测精 度和百分误差与网络隐含层节点数和训练次数之间的三维关系图，从图中可以很容易判断出达到最佳预测 精度和最小百分误差的隐含层节点数和训练次数 . 该方法和技术从根本上提高了选择人工神经网络隐含层 节点数和训练次数方法的效率 . 关键词：隐含层节点；训练次数；优化；人工神经网络 中图分类号：Tp183 文献标识码：A 文章编号：0367 - 623（4 2003）02 - 207 - 03 Optimization of hidden nodes and training times in artificial neural network GAO Da-wen，WANG peng，CAI Zhen-chaO （SchOOI Of MunicipaI and EnvirOnmentaI Engineering，Harbin Institute Of TechnOIOgy，Harbin 150090，China） Abstract：In Order tO find Out the number Of hidden nOdes and the training times Of netwOrk in OSAR，a Iarge num- ber Of data has been gathered fOr the reIatiOnships between the predictabiIity and percentage errOr Of netwOrk and the number Of hidden nOdes and the training times by running prOgrams，and a three - dimensiOn pIOt has been drawn by appIying MatIab. The number Of hidden nOdes and the training times can be easiIy fOund with the Ieast percent- age errOr and the best predict precisiOn and the efficiency in seIecting the number Of hidden nOdes and the training times can therefOre easiIy imprOved. Key words：nOde in hidden Iayer；training time；OptimizatiOn；artificiaI neuraI netwOrk 近几年来，人工神经网络在环境工程中的应 ［ ］