一种判别实测资料中异常点的方法及应用

在原始实测资料中经常出现与绝大部分点据体现出来的规律不相符合的离散异常点,需要特别标出。人工神经网络原理分析认为:如果有少量点据的训练值与实测值存在较大差别,而其他绝大多数点的训练值与预测值十分接近,则差别较大的点是异常点。为了证实该方法的有效性,用半理论半经验公式计算异常点的结果,与实测值也存在很大差异,而正常点则符合甚好,表明文章提出的方法较为合理。对方法的适用性和优缺点进行了探讨。

A Method to Distinguish Unconventionality Points from Real-tested Data and Its Application

Unconventionality points which are not fit for the rule manifested by most points are appeared frequently in testing data.These points shall be lined out specially.Based on Artificial Neural Net(ANN) theory,a new method is put forward in this paper.If several points' training results exist sharp contrast with the real-tested ones while the training results of most points tally well with the forecast values,then the points of sharp contrast are unconventionality points.The mathematical foundation of this method is demonstrated,and the applicability and merits and disadvantages are also discussed.