A fluid identification method for buried hill reservoirs based on a BP neural network model using NMR

The geological structure of buried hill reservoirs is highly complex. This study aims to develop a new reservoir fluid identification method for buried hill reservoirs by integrating nuclear magnetic resonance (NMR) logging techniques with a backpropagation (BP) neural network model. NMR logging data were used as input features, including parameters such as T 2g (geometric mean of the transverse relaxation time), A (amplitude of the last peak in the T 2 spectrum), S 3 /S 1 , and S 2 /S 1 . A BP neural network model was constructed with a single hidden layer consisting of eight neurons. The ReLU activation function was employed to accelerate the learning process, and the Softmax function was selected as the output layer activation function to accommodate multi-class classification requirements. Results show that the BP neural network model achieves superior performance in terms of precision, recall, and F1-score across four fluid types: oil zones, oil-water bearing zones, oil-bearing water zones, and water zones, outperforming other similar models. In Well HZ26-6-1 located in the ancient buried hill area, the model’s predictions are largely consistent with the results from well testing. This study demonstrates that the BP neural network-based approach not only significantly enhances the accuracy of fluid identification in reservoirs but also provides a scientifically sound and effective solution for fluid discrimination under complex geological conditions. Notably, the model exhibits strong classification capability in distinguishing between oil-water bearing zones and oil-bearing water zones, which are typically difficult to differentiate.