Application of Neural Network Based on Real-Time Recursive Learning and Kalman Filter in Flight Data Identification

The process of obtaining flight data from flight test is complex and costly, which makes it difficult to identify aerodynamic parameters. Therefore, Cessna172 flight simulator was used for flight data extraction, which ensures the convenience, efficiency and economy of the test. To obtain aerodynamic model, based on the idea of machine learning, a recurrent neural network was used to process multi-dimensional nonlinear flight test data, and a real-time recursive learning algorithm was proved to be suitable for dynamic training. Due to the large amount of state parameter data generated by aircraft, which will cause the real-time recursive learning algorithm to train slowly. So, Kalman filter algorithm was introduced for system identification. Considering validity analysis, the comparative verification method was used to verify system identification model. Results show that the aircraft aerodynamic and aerodynamic moment models have good applicability and can be popularized and applied.