Abstract: The three-dimensional (3D) gravity inversion based on the properties model is very popular in recent years. The inversion is composed of numerous forward calculations. With the increasing magnitude of target mesh, the time of the regular CPU (central processing unit) serial algorithm will exponentially increase. In this study, the authors use the GPU (graphic processing unit) parallel processing technology and put forward a new computing method of inversion to distribute the calculations of every observed point’s gravity anomaly among the GPU parallel threads. The new method takes full advantage of GPU’s operational capability and greatly accelerates the inversion. The authors applied this new method in the 3D inversion of gravity anomaly of Nansha’s target area by the genetic algorithm method. The results show that using the GPU parallel algorithm can reduce the time of inversion from more than 17 days to 7 hours in 600 generations. The inversion and observation of target’s gravity anomaly show the method is efficient. Then, the authors obtained the 3D density distribution of the target and made a density profile of the NH973-1 survey line. Through the analysis, the authors found the density range of the area with random reflection in the seismic profile is 2.24-2.53 g·cm^-3 and believed this area might be a Mesozoic deposit layer. At last, by locating some parameters the authors found some of the target area has a high level of oil and gas prospects due to their shallow depth (less than 1.5 km) and large thickness (more than 2.5 km) of Mesozoic deposit layer.

Key words: gravity inversion, GPU parallel processing technology, 3D density distributions, Mesozoic deposit layer