基于多相关性的传感数据离群点检测与处理

Outlier Detection and Processing of Sensor Data Based on Multiple Correlation

针对传感数据中离群点造成结果精确度低的问题,设计了一种基于局部异常检测方法的改进离群数据检测算法。利用节点数据的时空相关性建立了时空变化因子模型,对局部异常检测算法中距离计算进行改进,使弥散数据更加密集,提高检测精确性。此外,还利用了不同数据间的属性相关性构建了一个线性回归模型,用于处理误差数据。为了验证算法的可行性,选用多个数据集进行对比测试。实验结果表明,该算法能够在实现高检测精度的前提下,将虚警率控制在较低水平。

To solve the problem of low accuracy caused by outliers in sensor data,an improved outlier detection algorithm based on local anomaly detection method is designed.A spatio-temporal change factor model was established by using the spatio-temporal correlation of node data,which was used to improve the distance calaculation in the local anomaly detection algorithm to make the dispersed data denser and improve the detection accuracy.Furthermore,linear regression model was established by using attribute correlation between different data to process the error data.To verify the feasibility of the algorithm,multiple data sets were selected for comparison testing.Experimental results show that the algorithm can control the false alarm rate to a low level on the premise of achieving high detection accuracy.