Beyond Static Near-Wellbore Data: Optimization of Hydraulic Fracturing Parameters Under Co-Evolution Constraints of Fracture Networks and Geological Spatial Fields

Abstract Data-driven fracturing optimization relies on conventional logging data that primarily reflect near-wellbore conditions. However, fracture propagation often extends hundreds of meters beyond the wellbore, making it highly challenging to characterize geological variations and stress shadow effects in far-field regions. Existing methodologies rely on simplified static assumptions. To address these challenges, we propose a physics-data fusion framework that integrates post-fracturing geological information to optimize fracturing parameters based on full-field dynamic responses rather than localized near-wellbore data. First, the unconventional fracture propagation model is employed for physical numerical simulation to construct a three-dimensional geological property field that characterizes the evolution of geological attributes after fracturing. Subsequently, a multi-channel three-dimensional convolutional neural network extracts geological spatial distribution features across nine channels, including permeability and porosity, which are fused with pre-fracturing logging parameters and fracturing treatment parameters to form integrated features, thereby achieving deep fusion of geological and engineering information. On this basis, a high-precision prediction model for cumulative oil production over 180 days is developed within the automated machine learning system, which automatically stacks multiple models and minimizes manual intervention, significantly enhancing the model's ability to represent complex nonlinear coupling relationships. Compared with traditional manual modeling methods, this model achieves a 20.2 percent reduction in root mean squared error on the test dataset. Based on this high-precision prediction model, the multi-objective evolutionary algorithm NSGA III combined with the TOPSIS decision method is employed to jointly optimize fracturing parameters, balancing production improvement and economic benefits. Results from six field trial wells demonstrate that wells optimized through this intelligent approach achieve a 25 percent increase in cumulative oil production over 180 days compared to conventionally designed wells, verifying the method's effectiveness and engineering applicability. The proposed method significantly enhances spatial characterization of geological changes compared to traditional data-driven fracturing optimization models, effectively overcoming the limitations of conventional logging data and enabling a more comprehensive and dynamic portrayal of reservoir responses within the fracturing influence zone, thus providing a new technical pathway for intelligent fracturing parameter design.