A novel filtering method for underwater cooperative localization with mixed acoustic measurement anomalies

This paper addresses the positioning error divergence problem in autonomous underwater vehicle (AUV) cooperative navigation caused by mixed abrupt and gradual anomalies in acoustic measurements. An improved dual-threshold fault-tolerant cooperative localization method is proposed to detect and mitigate such mixed anomalies. The kinematic model of the cooperative localization system with alternating pilotage between leader and follower AUVs is first established. The Mahalanobis distance of the innovation sequence from the adaptive cubature Kalman filter is used as the anomaly test statistic, with the upper threshold determined by the confidence level. A long short-term memory (LSTM) neural network is then trained using navigation parameters and filter states as inputs to predict acoustic measurement information, and the lower threshold is derived from the root mean square error of the network training results together with statistical principles. The acoustic ranging data are classified into three categories by the improved dual-threshold method, and fault-specific correction strategies are applied accordingly. Experimental results from lake trials demonstrate that, compared with the conventional chi-square detection algorithm, the proposed algorithm reduces the root mean square error by 24.2%, effectively resolving the mixed anomaly problem in underwater acoustic measurements for cooperative localization.