Authors:
Bing Qian
1
;
Chong Ma
1
and
Tong Zhang
2
Affiliations:
1
Beijing Research Institute, China Telecom Corporation Limited, Beijing, China
;
2
Intel Corporation, Santa Clara, California, U.S.A.
Keyword(s):
Multidimensional Time Series Data, Anomaly Detection, Unsupervised Learning.
Abstract:
With the continuous increase of network terminal equipment, the operation scenarios of 4G-LTE wireless networks are becoming more and more complex. The traditional manual method of analysis and screening of network cell equipment can no longer meet the needs of production. Therefore, an efficient wireless network cell abnormality diagnosis algorithm is needed to screen abnormalities of equipment to improve operation and maintenance efficiency. In view of the fact that the existing single-dimensional anomaly diagnosis algorithm cannot achieve fully automated detection and the existing multidimensional anomaly diagnosis algorithm has low detection efficiency on multidimensional time series data, there are a large number of errors and omissions. This paper proposes a multidimensional time series data based on 4G-LTE wireless network cell anomaly diagnosis optimization algorithm uses small-sample supervised algorithms to assist the training of massive-sample unsupervised algorithms, ther
eby improving the detection performance of unsupervised learning algorithms. This paper verifies the effectiveness of the optimization algorithm through experiments, and has a great improvement in the four commonly used unsupervised algorithms, which can well improve the anomaly detection capabilities of the existing algorithms.
(More)