Prediction of formation damage in fractured tight reservoirs based on machine learning

Abstract Reservoir damage incurred during oil and gas drilling operations will significantly reduce hydrocarbon productivity. Consequently, accurate prediction of reservoir damage severity before drilling operations is crucial for designing and deploying optimized reservoir protection strategies. Using the core analysis and damage evaluation experimental results of fractured tight reservoirs in eastern China oilfields, the correlation between each characteristic parameter and reservoir damage severity was analyzed by Pearson correlation analysis. This methodology identified dominant controlling factors of reservoir damage in the target block. Subsequently, the reservoir damage prediction model was established by using both Backpropagation Neural Network (BPNN) and Support Vector Machine (SVM) algorithms. The research results indicate that the top ten main factors affecting reservoir damage severity are: the concentration of drilling fluid temporary plugging agent (TPA), roughness coefficient of rock fracture surface, D90, D50, and the maximum size of TPA particles, clay mineral content, fracture aperture, gas permeability, D10 of TPA, and porosity. These ten dominant characteristic parameters can be used to construct prediction models. Comparative analysis shows that the error and mean square error of the prediction model based on BPNN are smaller than those of the SVM model, indicating that using the BPNN method for reservoir damage prediction under small sample conditions has higher accuracy and stability. Experimental verification shows that the prediction error is within 12.3%, which indicates that it can reliably predict the damage severity in fractured tight reservoirs, providing a basis for designing and implementing reservoir protection strategies in drilling and completion operations.