Gravity Anomaly Inversion Based On Neural Networks

Gravity exploration is an applied subject which is developed on the basis ofgravity surveying, it is one of the geophysical exploration methods that is usedfreguently and the results of it is good. Usually the geological structure of theexploration area is quite complex, gravity anomalies collected in the field will beaffected by various geological factors and then produce superposition or interference,so if we simply observe and analyze subsurface anomalies by virtue of abnormalcontour map,the results will be inadequate. The gravity abnormal inversion becomesan important means,we can obtain the subsurface information with it. At present,mostinversion method is shape inversio,but underground anomalies are often unevendensity distribution space.The neural network gravity inversion is not only inverse thegeological body’s shape, size, location. And also inverse the density distribution ofgeological bodies. This is helpful for us to identify, understand, and research.The interpretation of the geophysical anomalies can be divided into physicalinterpretation and geological interpretation. According to the distribution of theabnormal conditions of the work area to determined the geophysical body’s shape,size, depth and location of the projection on the ground,if the conditions are allowedwe can furtherly determine the factor of the occurrence and the residual density bytheoretical calculations.T his is mathematical physics explanation. Methods forsolving inverse problems of gravity anomalies have been many. There is not a matureview on the classification, and we can roughly divide it into four categories: linearinversion, nonlinear inversion,direct method and statistical analysis. Thus thegeophysical inversion and mathematics are closely linked,and a good mathematicalapproach can help us to better solve the inverse problem.The inversion can be divied into linear inversion and nonlinear inversion. Themost optimized selection method is the most used in the linear inversion.There aretwo questions existing in it.The fist one is how to assess the degree of compliancebetween the theoretical curve and the measured curve and how to reach thestandards of accuracy.The second one is how to modify the parameters of geological models. The more inversion parameters,the better its result.Nonlinear inversionmethod include genetic algorithms,the simulation annealing algorithm and neuralnetwork algorithm,and so on.While the neural network is used in geophysical moreand more widely,especially in the oil and gas geophysical it has made breakthroughprogress, but in the aspects of solid minerals it is still in its infancy. Neural networkhas a strong nonlinear mapping function,and it can simulate non-linear approximationfrom input to output.Because of this function we are able to introduce it to theinversion of gravity anomaly,we can used the gravity anomaly observations as neuralnetworks input, and the underground geological parameters is its output. Through asufficient number of samples,we can be trained forward we need to simulate theinversion function. Neural network has many advantages,for example,it has a highspeed when it handle with inversion problem.It can reduce the possibility of theinversion and inverse the three-dimensional object’s physical parameters.In this paper,I design many models to verify the effect and resolution of theinversion, and it is applied to data processing. This paper firstly introduces somebasic theoretical knowledge about gravity forward, then introduces the theory of theneural network and the neural network theory on gravity anomaly inversion. In orderto verify the effect of the neural networks,I designed several monomers modelsincluding horizontal and vertical rectangular parallelepiped model.In the two models,I designed two different sizes, different depths, different locations and differentresidual density. Based on the inversion data,I drawed a three-dimensional renderingsof geological density distribution profile, and then I compared the observation valueswith the theoretical model data and achieved good results.It verify the neural networkinversion’s feasibility. In the actual data processing, I used it in the national charityproject "Kangdian axis iron-copper exploration techniques demonstration pool ".Aftercompared with the work of exploration, neural networks inversion method was provedthat it has a good effect in actual data processing. It has practical value.Finally,Isummarized the advantages and disadvantages of the neural network and somequestions should be payed attention to.