Error compensation is an important guarantee method to ensure the accuracy of clock synchronization in underwater sensor networks. Existing research methods mainly use linear fitting and least square method to compensate for clock synchronization parameters. Underwater wireless sensor network nodes are mobile, which leads the network nodes to be always in a time-varying state. In the process of synchronous forwarding, the position where the node sends and receives data packets will change, resulting in a relative moving distance, leading the dynamic delay to an increase in. In this way, as the number of forwarding nodes increases, the error of the clock gradually increases, causing the synchronization accuracy of the underwater sensor wireless network to gradually decrease. The existing underwater wireless sensor network clock synchronization algorithm does not fully consider the dynamic time delay caused by the movement of the node with the ocean current. It only uses the time stamp mechanism to solve the clock synchronization parameters, and then uses the traditional linear fitting to refine the synchronization parameters. The accurate solution of dynamic time delay is a key factor of synchronization accuracy. The use of traditional optimization algorithms to refine the synchronization parameters can easily fall into a local optimum, which makes the synchronization accuracy not high. Therefore, the existing traditional research on clock synchronization algorithms cannot well solve the problem of clock synchronization accuracy caused by node mobility. However this type of method does not consider the clock synchronization accuracy of node movement affected by ocean currents. To solve this problem, this paper proposes a clock synchronization error compensation algorithm based on BP neural network model. First, the deep-sea Lagrangian ocean current model is used to describe the movement of underwater nodes and simulate the movement speed of underwater nodes, and then a clock synchronization parameter model is established, and finally a BP neural network clock synchronization error compensation model is build, which conforms to the underwater environment, and the excitation function is defined, and regular term factor and compensatory factor are introduced to avoid model over-fitting. The BP neural network model clock synchronization error compensation algorithm is established for error back propagation. Simulation experiments show that compared with the comparison algorithm TSHL, MM-sync, and MU-sync, the accuracy of clock synchronization, namely the error between clock synchronization time and standard time, increased by 37.42%, 17.29% and 21.86%, and the mean square error is significantly reduced.