摘要
针对人工神经网络在预测中出现的异常值现象,采用了回归分析模型得到的预测区间来控制异常值现象的方法.并且应用在黄河三门峡河段的水质预测中,氨氮通量预测的网络模型控制前平均精度仅有50.05%,这是因为2006年6月份预测值偏离真实值太大,预测相对误差达到214.88%,超出了回归预测区间,从而影响了整体精度.控制后该月的相对精度为90.08%,平均精度达到80.79%,整体预测精度明显提高.实践表明,该方法对于消除网络模型预测中出现的异常值现象是较为有效的.
As to the abnormal phenomenon in the forecast of artificial neural networks, the method, in which the forecast range from the regression analysis model is used to control the abnormal phenomenon, has been adopted. In the forecast of the water quality of Yellow River in Sanmenxia, the average accuracy of the quantity of ammonia and nitrogen before the control of ANN is only 50.05% , which is because the forecast number is very different of the accurate number in June 2006, the relative error of the forecast number reach up to 214.88 percent, beyond the forecast range of regression, in order to have effect on the whole accuracy. The accuracy of this month is 90.08% , the average accuracy reaches up to 80.79% ; the whole forecast accuracy is proved obviously. The practice shows that the method is effective to eliminate the abnormal phenomenon in the artificial neural networks.
出处
《郑州大学学报（工学版）》
CAS
2008年第1期106-109,共4页
Journal of Zhengzhou University（Engineering Science）
关键词
回归分析
人工神经网络
水质预测
regression analysis
artificial neural networks
water quality forecast
