UAV Flight Data Anomaly Detection Based on Parameter Selection and Multiple Regression

Unmanned aerial vehicles (UAVs) are important in military and civilian applications. Flight data anomaly detection is an essential part of ensuring the safety and reliability of UAV s and has received extensive attention and research. The UAV system is complex and has many parameters. However, many unrelated parameters do not positively impact the model performance. In addition, since there are many anomaly types of UAV flight data, the adaptability of different anomaly types for anomaly detection methods remains a significant challenge. This paper proposes a data-driven method based on parameter selection and multiple regression for UAV flight data anomaly detection, which integrates correlation analysis and long short-term memory (CA-LSTM). First, a correlation-based analysis is performed to select parameters with correlation to reduce the model complexity. Second, an LSTM-based multiple regression model is designed to map flight parameters. Finally, the effectiveness of the proposed method is verified on real UAV flight data injected with two common anomaly types.