US 12,237,981 B2
Traffic anomaly detection method, and model training method and apparatus
Yanfang Zhang, Nanjing (CN); Gang Li, Chengdu (CN); Li Xue, Nanjing (CN); and Wei Lin, Shenzhen (CN)
Assigned to HUAWEI TECHNOLOGIES CO., LTD., Shenzhen (CN)
Filed by Huawei Technologies Co., Ltd., Shenzhen (CN)
Filed on Feb. 11, 2022, as Appl. No. 17/669,638.
Application 17/669,638 is a continuation of application No. PCT/CN2020/107627, filed on Aug. 7, 2020.
Claims priority of application No. 201910752193.9 (CN), filed on Aug. 15, 2019.
Prior Publication US 2022/0166681 A1, May 26, 2022
Int. Cl. H04L 41/14 (2022.01); H04L 43/08 (2022.01)
CPC H04L 41/145 (2013.01) [H04L 43/08 (2013.01)] 20 Claims
OG exemplary drawing
 
1. A method comprising:
obtaining a target time series comprising N elements, wherein the N elements correspond to N moments, and wherein each of the N elements is traffic data received at a corresponding moment;
obtaining a target parameter of the target time series based on the target time series, wherein the target parameter comprises at least one of a periodic factor or a jitter density, wherein the periodic factor represents a wave-shaped change that is presented in the target time series and that is about a trend that represents an overall change of the target time series, and wherein the jitter density represents a deviation between an actual value of the target time series and a target value of the target time series within a target time;
obtaining, based on a first mapping relationship and a first type of the target time series, a first-type decision model, wherein the first mapping relationship comprises correspondences between a plurality of types and a plurality of first-type decision models, wherein the first type of the target time series is based on the plurality of types and on the target parameter, wherein each of the types corresponds to one parameter set and corresponds to a type of decision model, and wherein the target parameter belongs to a parameter set that corresponds to the first type; and
detecting an anomaly of the target time series based on the first-type decision model that corresponds to the first type.