Parameter Optimization of Chemical Plugging for Gas Channeling in CO2 Flooding Based on Multi-Proxy Collaborative Pre-screening

Driven by the goal of carbon neutrality, the integration of Carbon Capture, Utilization, and Storage (CCUS) with Enhanced Oil Recovery (EOR) technologies has demonstrated immense economic potential. However, in heterogeneous reservoirs, injected CO₂ is highly susceptible to forming gas channeling pathways along high-permeability zones, leading to a drastic decline in oil sweep efficiency and the ineffective cycling of greenhouse gases. The injection of a PPG-polymer composite for deep conformance control currently serves as a crucial countermeasure. To address optimization challenges such as the strong nonlinearity of chemical conformance control mechanisms, multi-objective conflicts—specifically the trade-off between incremental oil production and chemical costs—and the prohibitive expense of full-tensor numerical simulations, this paper proposes an Adaptive Heterogeneous Ensemble Surrogate-assisted Differential Evolution (AHES-DE) framework. This framework innovatively constructs a heterogeneous surrogate model pool comprising Kriging, Radial Basis Function (RBF), and Generalized Regression Neural Network (GRNN). It dynamically adjusts the weights of each model using the entropy weight method and prediction variance to adaptively fuse multi-source predictive information. Furthermore, a surrogate-based pre-screening mechanism is introduced to guide the differential evolution algorithm in invoking the high-fidelity numerical simulator (CMG STARS) exclusively for the most promising solutions, thereby significantly mitigating the computational burden. The results demonstrate that AHES-DE exhibits substantial advantages over traditional single-surrogate optimization approaches. Under a constrained computational budget, its convergence speed is remarkably accelerated, reducing time consumption by over 60%. Additionally, it displays extremely narrow standard deviation confidence bands across multiple independent runs, manifesting exceptional optimization robustness. Concurrently, the algorithm successfully generates a high-quality Pareto front. By optimally selecting Pareto solutions, it overcomes the traditional engineering reliance on high-concentration agents, yielding a notable increase in cumulative oil production and a substantial enhancement in the chemical utilization factor. Regarding the final Net Present Value (NPV), Standard DE and SOA-RBF converge to only 8.71 × 10⁸ CNY and 8.85 × 10⁸ CNY, respectively. In contrast, AHES-DE robustly captures a global maximum of 8.98 × 10⁸ CNY, representing improvements of 3.10% (approximately 27 million CNY) and 1.47% (approximately 13 million CNY) over the former two algorithms. This study refines the collaborative optimization methodology for conformance control following CO₂ gas channeling, providing a robust decision-making tool for the economic evaluation of complex CCUS projects.