Gas-bearing Prediction in Tight Sandstone Reservoirs Based on Multi-Network Integration

Abstract

The tight sandstone reservoir gas is integral to unconventional natural gas exploration and production in China. However, traditional oil and gas assessment methods sometimes suffer from low accuracy. Therefore, this paper proposes a method for predicting gas-bearing in tight sandstone reservoirs. This method selects seismic attributes through Pearson coefficients, combines multiple attribute information, and inputs it into a deep neural network. This study constructed MultipleNet by combining a convolutional neural network, a bidirectional gated neural unit network, and a self-attention mechanism. This network takes advantage of the complementary advantages of the above network modules and can more effectively mine information on various seismic attributes and improve gas-bearing prediction accuracy. This method is applied to actual data from a tight sandstone gas exploration area in the Sichuan Basin. Experimental results show that the results of well sides predictions using this method are consistent with well data, providing a new approach and perspective for predicting gas-bearing in tight sandstone reservoirs.