Temporal Convolutional Kolmogorov–Arnold Networks-Based Multivariate Regression for Unmanned Aerial Vehicle Flight Data Anomaly Detection and Recovery

With the rapid development and wide application of unmanned aerial vehicles (UAVs), concerns about their safety and reliability are increasing. Anomaly detection of flight data and reasonable recovery of detected anomalies are important for the safety and reliability of UAVs. However, the complex high-dimensional nonlinearity, dynamic time-varying behavior, and random noise in flight data severely challenge the accuracy of current anomaly detection and data recovery approaches. This article proposes a temporal convolutional Kolmogorov–Arnold networks (KANs)-based anomaly detection and recovery (TCKANs-ADR) framework for UAV flight data. First, a multivariate regression model based on temporal convolutional network (TCN) and KANs (TCN-KANs) is designed, leveraging TCN’s ability to capture local features and contextual dependencies, along with KANs’ strengths in nonlinear modeling and high-dimensional function approximation to adequately extract the spatiotemporal features of flight data and effectively realize the mapping of the monitored parameter. Second, a residual smoothing method is used to overcome the disturbance of random noise followed by modeling the extremes of the smoothed residuals using the extreme value theory to obtain dynamic thresholds for improving the anomaly detection accuracy. Finally, the effectiveness of the proposed approach is verified through extensive experiments on simulated and real UAV flight data. The experimental results demonstrate the significant advantages of the proposed method in UAV flight data anomaly detection and data recovery.