A Novel Approach For Anomaly Detection In Time Series Based On LSTM

Time series is o a group of data point arranged according to the sequence of time occurrence under the same measurement index.It reflects the dynamics of a subject on the time axis.At present,it has been widely studied in many fiels,including fincance,ecological environment,industrial production,network security,etc.In the time series,there are some data points that do not conform to the law of change,and their characteristics are significantly different from other sample data.We always name them “outliers”.Abnormal samples are also called outliers and ambiguity points.Although these abnormal data points are small proportion and easy to be ignored,however,they are often more valuable rather than normal signal.For example,network attack detection,credit card fraud detection,spacecraft fault detection,electrical anomaly detection,are very typical time series anomaly detection.Therefore,anomaly detection of time series is of self-evident significance in real life.In the previous studies,researchers have proposed various computational methods to realize the classification and detection of time series anomalies.However,the current methods still have a lot of limitations,and their performances are not good in practical applications.In recent years,the huge development of deep learning has provided a new idea for time series anomaly detection.In this paper,a time series anomaly detection method(LSTMAD)based on long and short term memory network(LSTM)is proposed to solve the problems existing in traditional methods,such as high time consumption,low accuracy and poor adaptability to unknown distribution of data.To verify the effectiveness of the LSTMAD model,we tested it on multiple datasets.Simulation results show that LSTMAD can not only detect abnormal data in time series,but also greatly reduce the complexity of calculation.