Research And Application Of Time Series Data Anomaly Detection Method

Mining important information in time series data with more characteristic items is a common task in the field of industrial informatization,including network service,finance,electric power,water conservancy,enterprise automatic operation system,biological information and other fields.In order to improve the classification accuracy of anomaly detection,the proposed algorithm is trained and tested on Kaggle’s pump＿sensor,and the research of anomaly detection based on time series data is realized.This paper analyzes the advantages of data preprocessing and anomaly detection algorithms in the existing time series data,introduces the time series detection model,data processing theory and anomaly detection evaluation index and other related technologies,which lays a theoretical foundation for the time series detection model constructed in this paper.For the feature overload of Kaggle’s pump＿sensor time series data,combined with the analysis results of LSTM model,the Ant-lion optimization algorithm is used to optimize the LASSO feature selection algorithm.The obtained ALO-LASSO feature selection results are applied to the LSTM model to realize the optimization of the input feature parameters of the LSTM model,so as to construct the anomaly detection classification model based on the ALO-LASSO-LSTM time series data.The proposed model was simulated on the pump＿sensor time series data published by Kaggle.The results show that the average accuracy of the constructed anomaly detection model(ALO-LASSO-LSTM)reaches 95.3%,which improves the accuracy of anomaly detection of LSTM model.In view of the time characteristics of pump＿sensor data,combined with the characteristics of short-short time memory network suitable for processing sequence data,the LSTM network is optimized with the addition of Attention mechanism,and the LSTM-Attention detection model is constructed,which improves the weight of important features.It strengthens the classification function of important features in LSTM model.Finally,the strong classification model Adaboost integration algorithm can integrate multiple Attention-LSTM classification models,and then get a strong classification model,and further improve the accuracy of the classification model.The model is simulated on the pump＿sensor data set,and the Attention-LSTM-based Adaboost anomaly detection model has improved classification results with an average accuracy of 96.6% under the same characteristics.