No abstract

Shunbei Oilfield is a fractured carbonate reservoir with complex geological structures that are influenced by fault movements and prone to collapse and leak incidents. Precisely predicting leakage pressure is crucial for conducting fracturing operations in the later stages of production. However, current fracture-related leakage pressure prediction models mostly rely on statistical and mechanical methods, which require the consideration of factors such as fracture aperture and parameter selection, thereby leading to limitations in prediction efficiency and accuracy. To enhance the accuracy of reservoir leakage pressure prediction, this study leverages the advantages of artificial intelligence methods in dealing with complex nonlinear problems and proposes an optimized Long Short-Term Memory (LSTM) neural network prediction approach using the Particle Swarm Optimization (PSO) algorithm. Firstly, the Spearman correlation coefficient is used to evaluate the correlation between nine parameter features and leakage pressure. Subsequently, an LSTM network framework is constructed, and the PSO algorithm is applied to optimize its hyper-parameters, establishing an optimal model for leakage pressure prediction. Finally, the model’s performance is evaluated using the Coefficient of Determination (R2), Root Mean Squared Error (RMSE), and Mean Absolute Percentage Error (MAPE). The evaluation results demonstrate that the PSO-optimized LSTM model achieved an R2 of 0.828, RMSE of 0.049, and MAPE of 3.2, all of which outperformed the original model. The optimized LSTM model showed an average accuracy approximately 12.8% higher than that of the single LSTM model, indicating its higher prediction accuracy. The verification results from multiple development wells in this block further confirmed that the deep learning model established in this study surpassed traditional methods in prediction accuracy. Consequently, this approach is beneficial for drilling engineers and decision-makers to plan drilling operations more effectively and achieve accurate risk avoidance during the drilling process.