Reservoir Fluid Identification Method Based on Multi-Feature Fusion

The resistivity of tight sandstone reservoirs in Sulige gas field is greatly affected by the mineralization degree of formation water and siliceous cementation, resulting in the phenomenon of “high resistivity water layer” and difficulty in identifying reservoir fluids. We extract frequency domain information from well logging curves using time-frequency analysis methods, construct a set of time-frequency images, and extract local binary pattern (LBP) and histograms of oriented gradients (HOG) features from time-frequency image data to construct a fused feature matrix of LBP and HOG. Based on support vector machine (SVM), the fluid information of the fused feature matrix is recognized, and a reservoir fluid recognition method based on fused multiple features is established. This method has a good recognition effect on the properties of gas and water fluids, with an average recognition accuracy of 83.2% on the test set. Field applications have shown that compared with the optimized logging interpretation method and the time-frequency analysis method alone, the reservoir fluid identification method based on fused multiple features improves the accuracy of fluid identification in tight sandstone reservoirs, with an average identification accuracy of 90.4%, verifying the feasibility of using this method to identify fluids. The reservoir fluid identification method based on fusion of multiple features has improved the accuracy of gas water identification in tight sandstone reservoirs, and can effectively identify the fluid properties of tight sandstone reservoirs.