Due to the limited coverage of base stations in the deep-sea areas of the South China Sea,water depth estimation are affected by environmental factors such as tides and waves,making it impossible to directly use the BeiDou Navigation Satellite System(BDS) positioning technology for real-time water depth estimation.To overcome these challenges,this paper proposes a method that leverages Precise Point Positioning(PPP) function and multi-beam bathymetry data samples to achieve real-time precise water depth data acquisition without relying on base stations.The methodology involves training a neural network using the Earth Gravity Model(EGM) and multi-beam bathymetry data samples to preserve the seabed's geoid height fitting surface.A Particle Swarm Optimization Back Propagation(PSO-BP) network model is then developed to determine the geoid height at the mud line.Water depth is obtained by real-time calculation of the mud line and positioning data.To evaluate the adaptability of the proposed method,tests were conducted at a selected sea area of the South China Sea.By comparing the fitted water depth with the multi-beam measured water depth results,it shows that the application of the PSO-BP neural network fitting model and PPP technology significantly reduces the error in real-time water depth estimation compared to the EGM 2008 fitting method.Specifically,at a test site with a water depth of 400 meters,the accuracy of average water depth estimation without tidal measurements can achieve ±1 meter.This performance meets the realtime water depth estimation requirements for vessels operating in certain sea areas without depth measurement equipment.This study offers substantial support for navigation safety and exploration of resources in the South China Sea and other deep-sea areas.