Geophysical Nonlinear Joint Inversion Method

The integration interpretation or inversion using manifold geophysical data is the total developing direction for geophysical data processing. The joint inversion has better resolution and can more effectively resolve the problem of non-uniqueness than the inversion using only one geophysical method data. Because the geophysical linear joint inversion have many problems such as complicated calculation, unstable iteration, insufficient using the information of every geophysical prospecting method data, and not being able to effectively integrate experts knowledge, this paper puts forward a new non-linear geophysical integration inversion method using FasART fuzzy inferential neural network.The relation between the data of geophysics and models usually is very complex nonlinear, which can be regarded as a map from data space to model space. If the model space is treated as geophysical property space, the geophysical inversion is regarded as an understanding or interpretation of geophysical property from field data and the inversion of multiple geophysical data is regarded as the fusion of multiple sensor data. Therefore, on the basis of the neural-fuzzy modeling, this paper introduces the concept of fuzzy set, aiming at the fuzzy features of geophysical data. By extracting all kinds of feature patterns of geophysical data and using easily calculated triangle membership function to fuzzify the feature patterns, the author forms a uniform linguistic fuzzy set and realizes the joint inversion of multiple geophysical data, using FasART fuzzy inferential neural network to fuse the linguistic fuzzy set. Only the pattern features of geophysical data are extracted owing to the use of neural pattern recognition to process the nonlinear data. FasART fuzzy inferential neural network has carried out characteristic level data fusion, and obtained the object's parameters using defuzzification method. The method can integrating interpret manifold geophysical data, and effective integrating expertsknowledge. It has higher resolution and can more effectively resolve the problem of non-uniqueness than the generalized linear joint inversion method. It proved to be fast, accurate, which that once the system has been trained it can research the best pattern in a moment without any more training. It also proved to be available for integration inversion in inhomogeneous medium.