LAUSR

Abstract The influence of paleoenvironments on the lithofacies of lacustrine mudstone is important for both paleoenvironment reconstruction and hydrocarbon exploration. Compared with the machine learning approaches, the methods to find coupling relationships are too qualitative to be convincing. This study used the artificial neural network (ANN) approach to better understand the primary factors that control lacustrine shale lithofacies development and how paleoenvironments influence depositional processes. The neural network was trained with the back propagation training algorithm based on identified shale lithofacies (desired output) and geochemical indicators (input). The lithofacies deposited under different paleoenvironments were identified from an 80-m long core based on observations of the sediment texture and structure and biota and analyses of the mineralogy and trace elements. To confirm the primary factors of lithofacies development in J1d of the Upper Yangtze Platform, five groups of elements were selected as geochemical indicators (Sr/Ba, Zr/Rb, Sr/V, V/V + Ni, and Ba/Ca) to monitor the changes in paleosalinity, paleohydrodynamic condition, paleoclimate, paleoredox, and paleoproductivity, respectively. The element groups were selected and parameterized as inputs to constrain the lithofacies (output) boundaries using artificial neural networks (ANNs). The results of the ANN models indicate that the lacustrine lithofacies development of J1d was controlled by paleosalinity, paleohydrodynamic condition, paleoclimate, and paleoredox but not by paleoproductivity. The results of this study indicate that the primary factors of shale lithofacies development can be effectively assessed via geochemical, sedimentological, and ANN analyses.