Neural network-based prediction of drilling fluid leakage

During the drilling process, reservoir fractures may lead to drilling-fluid loss, thereby slowing drilling progress and reducing well productivity. Therefore, it is necessary to choose the appropriate materials and formulations for plugging, and the leakage amount and rate are the most important indicators for selecting plugging agents. In this study, the amount of rigid mineral particles and plant fibers commonly used in drilling, as well as the width of formation fractures, were used as input variables, while leakage volume served as the output variable. By combining the multiple-population genetic algorithms (MPGA) and the backpropagation neural network (BPNN), an MPGA–BPNN prediction model was established to predict the leakage amount under different plugging formulations. The results showed that the correlation coefficient of the established prediction model reached 0.9741, indicating strong predictive accuracy for leakage volume and plugging performance under varying formulation conditions, providing useful reference and guidance for the optimization of plugging agents.