2-D resistivity inversion using artificial neural network of subsurface pipeline’s data

Inverting geophysical data due to their nonlinear nature is a very complex process, especially when a very high resolution in low depth is considered. In the past two decades, nonlinear inversion algorithms such as neural networks and genetic algorithms with significant growth for the interpretation of geophysical data have been used. In this study, geophysical inverting subsurface pipeline's data with high resolution is done .Thus, back-propagation network helped us to interpreting two-dimensional resistivity tomography data. Network parameters, including input and output data types, number of layers, number of neurons in each layer, network optimal learning rate, momentum coefficient and their impact on the value of the network error. After 20 iterations the error is reduced to 0.001. Robust pipeline (1000 ohm m) in a homogeneous half- space (100 Ohm m) by dipole-dipole array and one meter electrode spacing was modeled. 36 data sets that contained 207 components, were considered in this study, the 18 data sets in the training phase, 9 data sets in the evaluation phase and other data sets were assigned to a validation phase and the field data sets after the interpretation were compared with conventional inversion method. in this study, Comparing pseudo-sections interpreted by the artificial neural network method and conventional methods showed that although both methods lead to the detection pipeline but the artificial neural network method has a capability to separate two tube in pipelines that perched in 1.2 meter distance from each other and have 32 centimeter diameter and even can offer an approximately estimation of the diameter of the pipelines in pseudo-sections.