Seismic Inversion Through the Application of Machine Learning Algorithms

Seismic inversion is one of the crucial step in life of all oil fields. Obtained rock property from recorded seismic response is closely associated with distribution velocity and density in the rock section which are products of acoustic impedance. Conventional inversion techniques have disadvantages such as knowledge requirement of wavelet in deterministic case, or construction of geological model in stochastic case as input data. Well logging is the most representative source of rock properties, especially sonic and density which provide with essential information about reservoir. This kind of information might be used to obtain porosity distribution through the well and further permeability, knowledge of which is crucial for appropriate field development, but these measurements conducted quite rarely in wells. Machine learning algorithms will be applied in order to obtain density and sonic logs from seismic trace. Applied algorithms mainly based on statistical data analysis and allow generalize approach for achieving problem to solve. Also algorithm permits to avoid choosing of wavelet which is a significant disadvantage of conventional seismic inversion techniques. From the machine learning point of view it is required to solve regression problem which implies data series prediction from the given one.