Anomaly Identification of Flight Test Time Series Data Based on Improved LSTM

Real-time monitoring is an effective measure to ensure safety and reduce risk in flight test. Aiming at the problems in real-time monitoring, such as over-reliance on manual interpretation data, difficulty in detecting slow-changing anomalies in time, and inability to accurately predict the state of key parts, an anomaly recognition method based on improved LSTM is proposed. First, a convolutional variational autoencoder is pre-trained to learn potential representations of normal flight timing data, forming stable local features in each window. Secondly, the low-dimensional feature vector generated by the encoder is used as the input of the LSTM model to learn the correlation of longer time series and extract the feature. Finally, the anomaly index of the window is calculated based on the sample reconstruction, so as to realize the data anomaly identification. The experiment results show that the proposed method can realize the interpretable reconstruction process and effectively identify the common types of telemetry data anomalies, which can provide assistant decision support for flight safety in the real-time monitoring process.