Quasi-BP Neural Network Inversion Of Gravity Gradient Tensor

Compared to the vertical component of gravity field measurement in traditional gravity measurement, gravity gradient tensor is the second derivative of gravity potential, which more sensitive about the density abnormal.It's useful to delineate mineral deposits or Geological structure. Tensor gravity gradient measurements have been carried out in foreign countries more than ten years, and achieved some results, but the interpretation of measurement technology and related data in our study is still small.The anomaly of gravity gradient tensor, which is related to the coordinate system, are complex.If a single tensor component is used for inversion and interpretation individual, it is easy to lose useful information and result in misinterpretation.It's used more more information of the source in the inversion with full gravity gradient tensor which include All the five independent components. Therefore, compared with single-component inversion and inversion of gravity anomalies, the resolution of its results is higher,and it's better able to identify the source characteristics.Among the many nonlinear inverse algorithms, BP neural network algorithm is one of the most successful algorithms,and used widely. Based on previous research, this paper combine the BP learning algorithm and and RPOP algorithm.And it'used for single-component inversion and full tensor inversion.This article describes the process of Quasi-BP neural network algorithm and its advantages compared with traditional methods, then analyzes the characteristics of full tensor inversion. Firstly, through the inversion of potential field data in different contrast, this paper sums up the advantage of full tensor inversion,and then through more detailed comparative analysis and discussion,we show the influence of data quality, source characteristics and the initial model in full tensor inversion.It's hoped to make a comprehensive analysis of the inversion of gravity gradient tensor with Quasi-BP neural network algorithm in the paper,and lay a solid foundation for practical application.