Deep Learning-Accelerated Surrogate Optimization for High-Dimensional Well Control in Stress-Sensitive Reservoirs

Abstract

Production optimization in stress-sensitive unconventional reservoirs is governed by a nonlinear trade-off between pressure-driven flow and stress-induced degradation of fracture conductivity and matrix permeability. While higher drawdown improves short-term production, it accelerates permeability loss and reduces long-term recovery. Identifying optimal, time-varying control strategies requires repeated evaluations of fully coupled flow-geomechanics simulators, making conventional optimization computationally expensive. We propose a deep learning-based surrogate optimization framework for high-dimensional well control. Unlike prior approaches that rely on predefined control parameterizations or generic sampling, our method treats well control as a continuous, high-dimensional problem and introduces a problem-informed sampling strategy that aligns training data with trajectories encountered during optimization. A neural network proxy is trained to approximate the mapping between bottomhole pressure trajectories and cumulative production using data from a coupled flow-geomechanics model. The proxy is embedded within a constrained optimization workflow, enabling rapid evaluation of control strategies. Across multiple initializations, the surrogate achieves agreement with full-physics solutions within 2-5 percent, while reducing computational cost by up to three orders of magnitude. Discrepancies are mainly associated with trajectories near the boundary of the training distribution and local optimization effects. This framework shows that combining surrogate modeling with problem-informed sampling enables scalable and reliable optimization for high-dimensional, simulator-based problems, with broader applicability to PDE-constrained systems.

Figures (9)

• Figure 1: The unconventional oil reservoir model with multistage hydraulic fractures: (a) 3D view of the model, (b) top view of the model, and (c) zoomed view of one hydraulic fracturing stage.
• Figure 2: Evolution of sampling strategies used for proxy training: initial linear-decline-with-noise and moving-uniform sampling schemes (top row), intermediate variable-decline-with-noise schemes (middle row), and the final problem-informed sampling strategy consisting of linear-decline, constant-or-decline, and combined trajectory classes (bottom row).
• Figure 3: Schematic representation of the proxy model mapping time-varying bottomhole pressure (BHP) trajectories to cumulative oil production. The neural network surrogate approximates the nonlinear relationship between control inputs and production response.
• Figure 4: Schematic representation of the optimization framework, showing iterative updates of control variables using a gradient-based optimizer (fmincon), where the objective function is evaluated using either the full-physics simulator or the neural network proxy model.
• Figure 5: Cumulative oil production as a function of control step for different stress sensitivity levels and flow-only conditions. Higher stress sensitivity leads to increased permeability degradation and reduced cumulative production, with differences becoming more pronounced at later control steps.