An Anomaly Detection Algorithm of QAR Based on Spatial-Temporal Correlation

QAR (Quick Access Recorder) data contains numerous quality flaws such as anomalies and missing data. It will cause significant problems for subsequent data mining, model training, and analysis if it is not detected. To address these issues, this paper investigates QAR-specific anomaly detection (AD) algorithms before presenting a three-stage QAR data AD algorithm based on QAR spatial-temporal correlation, which includes single parameter AD, parameter correlation analysis, and multi parameter AD. The SST (Singular Spectrum Transformation) is used in this process to analyze the correlation between parameters based on the change point rather than the change trend. Simultaneously, a double K-means clustering algorithm that can automatically select the hyper-parameters K is proposed, followed by a relatively complete empirical experiment. The methods investigated in this paper are implemented in Python code, and their feasibility and effectiveness are demonstrated through simulation analysis. The accuracy rate is increased by 54% when compared to existing literature methods.