基于时空特征融合的水力裂缝形态代理模型构建与应用OA

To address the engineering challenges of difficult hydraulic-fracture network morphology prediction and time-consuming numerical simulation,this study develops an FC-UNet surrogate model for two-dimensional fracture morphology prediction in horizontal wells.The model uses U-Net as the backbone and integrates fast Fourier transform(FFT),parallel convolution(PC),and a channel attention mechanism(CAM)to adaptively fuse pumping time-series signals with spatial geological features,thereby transforming fracture propagation prediction into a pixel-level segmentation task.Based on an unconventional fracture model(UFM)calibrated with field micro-seismic data,310 forward-modeling samples were generated.A joint loss function combining weighted cross-entropy and Dice loss was introduced to alleviate class imbalance caused by sparse fracture pixels.Independent test results show that FC-UNet can accurately characterize main-fracture penetration and fracture deflection features,achieving an intersection over union(IoU)of approximately 0.84 and an F1-score consistently above 0.91.The prediction time for a single case is about 5 s,indicating that the model can serve as a fast surrogate for UFM-based fracture morphology forward modeling under similar geological conditions.