Anomaly detection based on data stream monitoring and prediction with improved Gaussian process regression algorithm

Condition monitoring has gradually become the necessary part of the diagnostics and prognostics for the complex systems. Especially, with the rapid development of data acquisition and communication technology, the appearing of large scale data set and data stream brings great challenges to model and process the condition monitoring data As a result, anomaly detection of the streaming monitoring data attracts more attention in the fields of prognostics and health management (PHM). Hence, in this study, Gaussian process regression (GPR) is applied for the abnormal detection in data stream; and on this basis a real-time abnormal detection method is proposed based on the improved anomaly detection and mitigation (IADAM) strategy and GPR which realizes incremental detecting for future data samples and requires no pre-classification labels of anomalies. Anomaly detection tested on an artificial data set and actual mobile traffic data set indicates the effectiveness and reasonability of IADAM-GPR model compared with naïve and Multilayer Perceptron (MLP) models.