Multiple point geostatistical simulation with adaptive filter derived from neural network for sedimentary facies classification

Sedimentary facies distribution is a vital reference for oil and gas exploration in an offshore area. One of remaining questions in a marine exploration area is to acquire an accurate sedimentary facies map with limited loggings and a set of indirect information. Previous studies are largely based on geological parameters and structures for estimating the sedimentary distributions, while some other researches employ machine learning algorithm to participate the classification tasks based on accessible datasets. However, the application of either method alone still cannot overcome the information limitation or the low accuracy of final results. This study is a kind of attempt to introduce machine learning algorithm into traditional multiple point geostatistical simulation (MPS) with seismic and logging information for increasing the applicability of multiple point geostatistical simulation and solving the previous mentioned problems of sedimentary classification. The filter-generation auto-encoder (FAE) using in this work is constructed by an encoder, a decoder, and a classifier, which are designed for generating adaptive filters of MPS, maintaining the stability of data extraction, and introducing the constraints of sedimentary facies, separately. For applying these adaptive filters into MPS, the distance function need to be modified to meet the requirement of indirect information without prototypes. Then the multi-distances function is able to measure the relationship among a known pattern, a nearby layer, related simulated points, and seismic attributes. After that, the sedimentary maps can be obtained by this method with great continuity in horizontal and vertical direction at the mean time, and its effectivity has been verified by the loggings in this area. Furthermore, the 3-D sedimentary facies map have valuable meanings for both petroleum exploration and geological engineering in an area with limited information.