Design And Implementation Of State Awareness Method Based On Multivariate Time Series

State Awareness is an important research direction.Detecting the anomalies in the systems accurately and timely will provide important help to a lot of fields,such as system monitoring,disaster precaution and intrusion detection.Since data is transmitted in time series in most fields,it is especially important to detect anomalies in time series data.With the development of big data and machine learning in recent years,the dimensional and magnitude of data has increased significantly.The traditional State Awareness algorithm for univariate time series cannot process current complex data samples,and simply applying it to multidimensions data cannot maintain the accuracy and real-time of the algorithm at the same time.As a result,it is necessary to design an State Awareness algorithm specifically for multivariate time series data.Based on this situation,we propose LGMAD,a real-time State Awareness algorithm,based on Long Short-Term Memory(LSTM)and Gaussian Mixture Model.Firstly,based on the improved LSTM,this paper proposes LSTM-BP,an State Awareness algorithm for univariate time series data to detect each dimension anomaly respectively.By comparing the difference between the predicted values and the true values of time series,the anomalies are preliminarily screened and a list of anomaly score is formed,which is the basis for multivariate time series State Awareness.Secondly,this paper calculates current system healthy factors at those possible anomaly points which are screened at the first step.A multidimensional joint detection in Gaussian model are given to some anomalies that be decided to further detection according to the current system health status.Based on this,multi-dimensional joint anomalies can be detected,and multivariate time series State Awareness results can be got.Lastly,this paper verifies the performance of LGMAD algorithm based on experiments.By experiments on two different data sets,it is been proved that the effect of State Awareness by LGMAD algorithm is better than that of the singledimensional State Awareness algorithm commonly used in the field,as well as the leading multi-dimensional State Awareness algorithm.