| With the development of information technology such as the Internet of Things and the Industrial Internet,time series data has exploded,and research on time series databases has become a hot topic in recent years.At present,most mainstream time series databases are all from abroad.The representative product of the domestic time series database is Io TDB,which has rich functions and excellent performance.Most mainstream time-series databases have supporting visual management tools to help users manage and operate time-series databases more conveniently,improving user experience and work efficiency.A company’s actual project "Io TDB Workbench" is a visual management tool to be designed and implemented for Io TDB.Monitoring and alarming is a very important function of the tool.Its main job is to analyze the operational data of Io TDB and its servers,detect anomalies in multivariate temporal data,and provide timely alerts to help users discover problems in a timely manner,improving system reliability and availability.The time series data anomaly detection algorithm based on deep learning has excellent performance,but requires a large amount of historical training data,and is not suitable for monitoring and alarming online systems that are frequently deployed and updated.In recent years,Jump Starter,a jump-starting multivariate time series anomaly detection algorithm proposed by Ma et al.,is based on compressed sensing technology,requires no training,and only needs a small amount of historical data to achieve excellent detection performance.After research,it still has the following problems: 1.The algorithm used to calculate the sampling confidence in the outlier-resistant sampling step does not perform feature selection,which is easily affected by noise data,and has a large amount of computation under high-dimensional data sets,which affects the overall detection accuracy and efficiency;2.The convex optimization tool library used in the compression sampling reconstruction step has a large amount of calculation,which will affect the overall detection efficiency.Based on the above analysis,this thesis takes a company’s actual project "Io TDB Workbench" as the background,conducts research around the multivariate time series data anomaly detection algorithm,and designs and implements an Io TDB visual management tool on this basis.The main work is as follows:1)Aiming at the problems of Jump Starter algorithm,this thesis researches and improves,and propose a multivariate time series anomaly detection algorithm based on compressed sensing(CS-TSAD).Perform shape-based clustering on the original multivariate time series data to obtain multiple multivariate time series data groups;perform outlier-resistant sampling on the data in each time window of each group to obtain multiple sampling matrices,and add the feature selection process is simplified,which can reduce the influence of noise data and improve detection efficiency;use a more efficient compressed sensing reconstruction method to reconstruct data for each sampling matrix,and obtain multiple reconstructed multivariate time series data groups;Multiple reconstructed multivariate time series data are grouped and spliced into a reconstructed multivariate time series data,and the difference between the reconstructed multivariate time series data and the original multivariate time series data is compared to calculate anomaly scores;thresholds are dynamically adjusted through extreme value theory(EVT)to compare abnormalities Score and the size of the abnormal threshold to determine whether it is abnormal.Comparative experiments on three public datasets show that the proposed algorithm has better detection performance.2)Taking a company project "Io TDB Workbench" as the background,design and realize an Io TDB visual management tool according to the software engineering specification,and apply the proposed multivariate time series anomaly detection algorithm based on compressed sensing to the monitoring and alarm function of the tool.The main tasks include: 1.Requirements analysis,including functional requirements and non-functional requirements,to meet the needs of efficient management and application of Io TDB;2.Overall design,including technical architecture design and functional architecture design,dividing the system into six functions Modules: connection management module,metadata management module,data management module,query editor module,authority management module and monitoring and alarm module;3.Detailed design and implementation,detailed design and implementation of each module,in which the monitoring and alarm module uses the proposed multivariate time series anomaly detection algorithm based on compressed sensing;4.Testing,conducting functional and performance tests on the system,and the results show that the requirements are met.At present,the Io TDB visual management tool has been deployed and operated on a company’s online server,with normal functions and fast response. |
