Anomaly Time Series Detection And Application Based On Segmentation Feature Representation

At present,time series is widely used in system traffic changes and is the most direct reflection of monitoring system normal operation and performance.It is of great research significance to achieve anomaly time series detection in the field of intelligent operation and maintenance by monitoring the trend of data changes.Due to the real-time nature of time series data,the large amount of data is difficult to iterate,and the data has the characteristics of nonlinear and even high-dimensional.Direct anomaly detection results in significant impact on the accuracy and efficiency of the algorithm.Therefore,in response to the above issues,this article conducted relevant research on the detection of anomaly time series,and the main work is as follows:(1)A time series feature extraction method based on boosted multi-scale fuzzy entropy is proposed to address the complexity of the current time series feature extraction process and the inability to efficiently extract the trend features of time series data.Firstly,using demeaning to reconstruct the original time series effectively simplifies the process of feature extraction in the time series.Secondly,by calculating the similarity and relationship dimensions of the reconstructed data,the quality of the change trend features is effectively improved.Finally,based on the calculation of multi-scale fuzzy entropy(MFE)under this relational dimension,and through the coarsening process,a boosted multi-scale fuzzy entropy(BMFE)model is constructed for feature extraction to ensure the integrity of time series feature information and the stability of entropy values.The experimental results show that this method has low fitting error and improves the efficiency of time series feature extraction.(2)A supervised anomaly detection method that consumes a lot of time for data annotation and is difficult to efficiently solve the annotation problem in anomaly series detection is proposed,which uses segmented feature representation for anomaly series detection.Firstly,based on the time series feature extraction method,the standardized calculation of segmented aggregation is used to obtain the feature representation of the data,which improves the reliability of anomaly detection in unlabeled time series.Secondly,the represented features are divided into anomaly series related features and irrelevant features,and anomaly series irrelevant features are pruned to reduce the adverse impact of these features on the detection results.Then,a time series similarity measurement method for time weight analysis is proposed,and a similarity matrix of the time series is constructed to effectively quantify the differences between different series.Finally,the similarity matrix is set,and the average similarity deviation value is calculated as the anomaly score for each sub-series.Threshold comparison is performed to effectively reduce the algorithm’s running time and achieve the detection of anomaly time series.The experimental results show that this method saves computational overhead and improves the efficiency and accuracy of the algorithm.(3)Based on the above research results,a network anomaly traffic detection prototype system based on time series analysis was designed and implemented using Java+Eclipse as a development tool.A detailed description was given from the aspects of system requirements analysis,functional modules,architecture design,and system implementation.By ensuring the effectiveness of network anomaly traffic detection through system operation,the goal of network anomaly traffic detection is achieved,and theoretical research is applied to practice,achieving the transformation of research results.