Abstract: Identification of abnormal ship behavior is an important part of the theoretical research of maritime safety science, and the identification of abnormal behavior is the main content of maritime supervision, which is of great significance to the safety of ships and maritime traffic. In this paper, a ship abnormal behavior identification method based on BO（Bayesian Optimization） improved GRU（BO-GRU） and AKDE（Adaptive Kernel Density Estimation） is proposed. The BO-GRU is used to predict the longitude and latitude, course and speed of the ship, and the error data set is obtained by comparing the predicted value based on the neural network with the actual value. The adaptive kernel density estimation is used to estimate the error data set non parametrically, so as to obtain the fluctuation interval of the ship track characteristic data under different confidence levels. The experiment is based on the data of the AIS（Automatic Identification System） of Tianjin Port. Compared with the basic GRU, LSTM（Long Short-Term Memory） and Bi-LSTM（Bidirectional Long Short-Term Memory）, the experiment verifies that BO-GRU has higher prediction accuracy; Compared with other methods, the adaptive kernel density can be better fitted and can detect the abnormal behavior of ships in time.