Resistivity restoration using Spiking Neural Networks for water-flooded zone interpretation in high water-cut sandstone reservoirs

Addressing the technical bottleneck of distorted dynamic resistivity responses in water-flooded zones during the development of high water-cut oilfield, where traditional static models and machine learning methods struggle to accurately characterize resistivity evolution patterns, this paper proposes a dynamic resistivity restoration and water-flooded zone interpretation method based on Spiking Neural Networks (SNNs). By simulating the impulse transmission mechanisms of biological neurons, the proposed method converts time-series logging data into spatiotemporal spike sequences, adaptively capturing the non-linear dynamic characteristics of resistivity evolution with changing water saturation. Comparative analysis with traditional static machine learning models demonstrates that this approach significantly enhances the accuracy of restoring the original resistivity of oil layers and effectively resolves the resistivity inversion challenges associated with "low-resistivity oil layers" and "high-resistivity water layers". Building upon the restored oil-layer resistivity model and integrating Archie's equation, the variations in oil saturation before and after water flooding were interpreted. Based on the established water-cut evaluation model, the water production rate at different saturation levels was predicted. Finally, a multi-indicator-involved classification model for water-flooded zones was developed by synthesizing resistivity changes, oil saturation variations, and water production rates. This model enables the accurate identification of strong, moderate, and weak water-flooded zones. Its reliability was validated using core analysis data from infill wells. Field application in an actual oilfield demonstrated that the water-flooded zone classification model achieved an accuracy rate of 92.7%. Perforation operations deployed based on the interpretation results yielded an incremental oil production of 9.7 t/d and a reduction in water cut of 16.2%. The research results successfully guided the identification of remaining oil enrichment areas and the optimization of injection-production strategies in the target block, providing a novel dynamic intelligent interpretation for precise potential tapping in mature oilfields with high water cuts.