Research on Anomaly Detection and Fault Prediction of Flight Data based on Support Vector Machine

The stable operation of the rocket power system is very important for the successful launch and mission completion of the rocket. In this paper, an anomaly monitoring method of rocket power system based on flight status data is proposed to solve the monitoring problems caused by insufficient sensors and low reliability of data. Firstly, the key flight state parameters are selected from the closed loop of the rocket control system, and the fault data set is constructed using these parameters. Then, long short-term memory (LSTM) autoencoder algorithm is used to extract features from the fault data set. On this basis, a support vector machine (SVM) classification and identification model are constructed to identify and classify faults of dynamic systems. In order to optimize the parameters of the SVM model and improve its classification performance, the artificial bee colony (ABC) algorithm was introduced in this study to continuously search for the optimal solution by simulating the foraging behavior of bees. In summary, this study proposed a rocket power system anomaly monitoring method combining flight status data, which provides a new idea and technical support for the rocket power system anomaly detection and fault prediction, and is of great significance to ensure the success of the rocket launch mission.