Research On Task Scheduling And Resource Provision For Saas Applications

The characteristics of time series flow data are not only related to time,but also the characteristics of massive,high-dimensional and real-time updating make the data mining more difficult.Time series flow data are often closely related to daily life.Therefore,anomaly detection of time series data has become a hot issue in data analysis and mining.For this reason,the research on anomaly detection of time series flow data is carried out.The main work is as follows:Firstly,we propose a distributed time series anomaly detection model based on edge computing.Using the big data processing idea of edge computing,the corresponding data is processed as close as possible to the computing resources close to the data source,which can not only reduce the pressure of network transmission bandwidth,but also improve the overall efficiency of data processing.Based on the distributed anomaly detection model,we propose an Anomaly Detection algorithm for Streaming Time Series(ADSTS).This algorithm will detect the corresponding abnormal temporal points in streaming time series according to the outlier distance measurement and the correlation of time series.The experimental results show that compared with the benchmark algorithms,ADSTS has shorter detection time and higher abnormal detection rate.Secondly,we propose a time series anomaly detection method using feature-based symbolic representation(TSAD-FD).Initially,we obtain FD-SAX representation of the raw time series data,and then we propose an optimized heuristic ordering based on FD-SAX to accelerate anomaly detection process.Extensive experiments have been conducted to demonstrate that the pruning rate and the overall efficiency of TSAD-FD is higher than the benchmark algorithms.Finally,we propose a K-nearest neighbor anomaly detection algorithm based on Piecewise Aggregate Approximation(PAA)representation and High-Dimensional indexing for Time Series(HDITS).The PAA can help HDITS avoid the problem of sharply degraded retrieval performance caused by "dimensionality curse",and HDITS can quickly pruning effectively for dissimilar sequences,thereby accelerating the search efficiency of anomaly detection.The experimental results show that the algorithm can detect the abnormal sequence at a specific time or a certain time range while ensuring the detected abnormal pattern sequence has certain "timeliness".