Application of Neural Networks in Reservoir Classification of the Luliang Oil Field

Abstract

To address the reservoir classification requirements of the Hutubihe Formation in Luliang Oilfield, this paper constructs an intelligent classification model based on deep neural networks. By automatically learning the complex nonlinear relationships among seven parameters—porosity, permeability, gamma ray, resistivity, oil saturation, density, and thickness—the deep neural network replaces traditional manual weight assignment methods and achieves multi-parameter collaborative characterization of reservoir properties. Based on the data-driven concept, an adaptive deep neural network architecture is adopted to realize intelligent reservoir classification. The accuracy of this method is verified through core analysis and neural network methods. Finally, the classification results are presented in radar charts, significantly improving the objectivity and accuracy of reservoir classification. Research shows that the deep learning model can automatically extract deep features from reservoir parameters. Core experiments indicate that the matching degree between reservoir categories and pore structure and flow capacity exceeds 90%, while the classification error of key parameters (porosity and permeability) is less than 5%. This method breaks through the limitations of traditional classification methods, achieves intelligent and precise classification of multi-layer reservoirs in Luliang Oilfield, and provides a more reliable decision-making basis for optimizing development adjustment plans. It verifies the superior performance of deep learning algorithms in complex heterogeneous reservoir classification and demonstrates significant practical value for refined residual oil potential tapping and intelligent development decision-making in mature oilfields.