Machine learning-based rapid fracture morphology prediction technology for horizontal wells in shale gas reservoir

During the development of shale gas reservoirs, the multistage fracturing technology of horizontal wells is a key technique for reservoir stimulation. Using downhole monitoring results to guide fracturing improvements helps enhance the shale gas fracturing technology. Currently, fracture fitting is mainly achieved through post-assessment by commercial software, but quickly quantifying and visualizing fracture propagation characteristics during fracturing is crucial for formulating and adjusting effective operational parameters, enabling efficient and economical fracturing. In this study, a new workflow based on JAYA optimization of least squares support vector machine (JAYA-LSSVM) algorithm is proposed for quickly predicting the downhole distribution characteristics of fractures in multistage fractured horizontal wells. Researchers selected operational parameters, perforation parameters, and pressure parameters that influence fracture propagation, conducted extensive numerical simulation experiments, and constructed a complete training dataset, with field data used as a test set to develop a real-time predictive model of fracturing curves. Based on the characteristic parameters inverted from the predictive curves, rapid visualization of fracture network morphology in shale reservoirs during fracturing was achieved. The inversion results were validated by micro-seismic data, confirming the reliability of the rapid fracture morphology prediction technology proposed in this study. This research provides guidance for parameter optimization design and fracturing technology adjustments for shale gas reservoir fracturing.