基于神经网络的高比例异常数据识别仿真

Simulation of High Ratio Anomal Data Identification Based on Neural Networks

针对因高比例异常数据的特征种类较多且存在状态转移特性,导致识别误差大的问题,提出一种基于神经网络的高比例异常数据识别算法。根据相邻数据集间具有时间和空间相关性特点,建立数据基础状态映射序列,计算数据样本在不同空间、不同时间点上的初始状态和结束状态,以及上一时刻和当下一时刻的数据空间和时间所属状态,根据前后变化对比得到状态转移概率,计算提取数据在不同时间点的具体特征。采用神经网络建立高比例异常数据识别模型,对待识别样本标准化处理,定义标准样本并提取特征因子,求解样本集中所有数据与标准样本的隶属度,存在正向相关关系为正常样本,反向则为异常数据样本,通过阈值对比实现高比例异常数据的有效识别。实验结果表明,所提算法识别精准度较高,误差率较低,具有一定的实用价值。

In this paper,an algorithm of identifying high proportion abnormal data based on neural networks was proposed.According to the characteristics of temporal and spatial correlation between adjacent data sets,we constructed a base state mapping sequence of data,and calculated the initial and end states of data samples at different spaces and time points,as well as the space and time status of the data at the previous and next moments.According to the comparison,we obtained the state transition probability.Meanwhile,we calculated and extracted the specific characteristics of the data at different time points.Moreover,we used the neural network build a model of identifying high proportion of abnormal data and standardized the samples to be identified.Furthermore,we defined the standard samples and extracted feature factors.In addition,we calculated the membership degree of all data in the sample set and standard samples.If there is a positive correlation,it is a normal sample,and if there is a reverse correlation,it is an abnormal sample.Finally,the effective recognition for the high proportion of abnormal data was achieved through threshold comparison.Experimental results show that the proposed algorithm has high recognition accuracy and low error rate,as well as certain practical value.