Abstract: ‍ ‍Location estimation based on underwater wireless sensor networks is an effective method to solve the localization problem of autonomous underwater vehicle， however， since the sensor nodes in underwater wireless sensor networks are generally deployed in open water， the injection of malicious noise delays in the arrival time-based network channel by foreign malicious nodes can adversely affect the localization accuracy. To address this problem， in this paper， we proposed a Modified Joint Probabilistic Data Association （MJPDA） algorithm. First， we consider a heterogeneous network architecture， including surface buoy nodes， underwater beacon nodes and AUV target nodes to be located， and obtain distance measurements between the beacon nodes and AUV in the network architecture； then， we design a matrix similarity-based malicious noise identification mechanism to identify and discard distance measurements contaminated by malicious noise， and perform a distance measurement analysis on those detected as not contaminated by malicious noise. Then， in order to enrich the sample measurement points to improve the localization accuracy of AUV， we generate a series of Gaussian-distributed virtual measurements in the vicinity of the predicted points and the detected measurement points without malicious noise contamination； finally， we correlate and update these measurement points to complete the precise localization of AUV target nodes. Finally， these measurement points are correlated and updated to complete the accurate positioning of the AUV target nodes. The simulation results show that the proposed localization method can maintain better localization accuracy in the presence of malicious noise attacks compared with other works.