摘要
为解决网络智能运维中智能模型构建门槛高、效率底的问题,提出了一种异常检测算法选择框架,通过对提取的时序数据特征进行波形分类,根据数据分类结果进行异常检测算法的最优匹配,并依据指令对本地选定的模型进行训练、更新,依靠选择的检测算法以及更新后的模型进行实时异常检测。本系统可以针对现实环境中各种KPI数据自动适配异常检测算法,减少专家经验的介入,提高生成效率,降低维护成本。
In order to solve the problem of high threshold and low efficiency in intelligent model construction in network intelligent operation and maintenance,this paper proposes an anomaly detection algorithm selection framework,which first extracts features from time series data,classifies waveforms based on data features,and then classifies based on data As a result,the optimal matching of the anomaly detection algorithm is performed,and the locally selected model can be trained and updated according to the instructions,and finally,the selected detection algorithm and the updated model are used for real-time anomaly detection.The system can automatically adapt the anomaly detection algorithm to various KPI data in real environment,reducing the involvement of expert experience,improving generation efficiency and reducing maintenance cost.
出处
《信息技术与标准化》
2021年第5期40-45,50,共7页
Information Technology & Standardization
基金
国家重点研发计划项目,项目编号:2019YFB1405000
2019国家自然科学基金项目,项目编号:61873309。
关键词
时序数据
特征提取
异常检测
模型训练
专家经验
time series data
feature extraction
anomaly detection
model training
expert experience
