Intelligent Parameter Inversions for Activated Fault Based on Well Testing Constraints and Transfer Learning

Summary Large faults widely exist in reservoirs, and it is of great significance for oil and gas production to accurately obtain the parameters of activated faults. In this paper, an intelligent inversion method of activated fault parameters based on well testing theory and transfer learning is proposed. Transfer learning is used to develop a 1D convolutional neural network (CNN) inversion model for obtaining five key fault parameters, including matrix permeability, fault conductivity, fault length, fault distance, and fault angle. In addition, the wellbore parameters were inverted by the feature point method, including the well storage coefficient and the skin factor. The proposed intelligent inversion method is applied to well test data from three real wells, and point-by-point mean squared errors (MSEs) of fitting curves are 0.015, 0.071, and 0.021, which are lower than those from manual inversion. Additionally, the time consumption for intelligent inversion is significantly shorter, with values of 0.014 seconds, 0.023 seconds, and 0.027 seconds. The results confirm that this paper provides a more accurate and efficient parameter estimation method.