Aiming at the problem that parameter selection depends on the accumulation of expert knowledge and the anomaly detection type is single in current UAV flight data anomaly detection methods,an anomaly detection method based on parameter selection and data reconstruction is proposed.Firstly,Spearman is utilized to mine the correlation between parameters to remove redundant parameters and thus improve the model anomaly detection performance.Secondly,an AE-LSTM model is designed,and the selected parameters are reconstructed in unsupervised manner to realize multi-type anomaly data detection.Finally,a filter is introduced to smooth the residuals,considering the effect of random noise.Using real UAV flight data for validation,the results show that the true positive rates of the three anomaly types detected by the proposed method are 98.04%,97.80%and 98.00%,respectively,which is superior to single-class support vector machine,long and short-term memory network and self-encoder,and verifies that the proposed method has good robustness and generalization ability for the detection of different types of anomalies.
维普
万方数据
