Abstract: Underwater positioning is the basic and critical work of automation equipment in the process of safety monitoring of underwater buildings. The filtering algorithm is an important part of underwater acoustic location algorithm, which helps to suppress noise and restore real data. Attenuated memory filtering can solve the problem of filtering divergence caused by inaccurate model, model change and rounding error accumulation to a certain extent. The memory of memory attenuation has a great impact on the performance of the algorithm. The current selection method has some problems, such as low flexibility, strong subjectivity and so on. In this paper, the memory attenuation mode of extended Kalman filter algorithm in underwater acoustic positioning is improved. The main idea is that the memory attenuation is carried out according to the observed value, after judging whether memory attenuation is needed. The underwater acoustic positioning simulation test results under three different working conditions verify the rationality and accuracy of the memory attenuation method in the improved algorithm. Compared with the existing memory attenuation algorithm simulation, the improved algorithm can improve the positioning effect and the continuous positioning accuracy.