【摘要】 为实现边底水气藏CO₂注入气藏采收率最大、CO₂封存量最高的目的，以国内某典型火山岩气藏为例，综合采用拉丁超立方采样（LHS）、长短期记忆神经网络（LSTM）与多目标粒子群（MOPSO）优化算法等，建立了基于代理模型的气藏注CO₂提高采收率与地质封存多目标优化方法，实现了CO₂注入及生产井制度的优化。在该方法中，首先采用LHS方法及油气藏数值模拟器，在给定的注采制度范围内生成1 000组训练与测试样本；然后，采用LSTM进行训练，构建CO₂注入生产动态预测代理模型，提高计算效率；最后，利用MOPSO算法获取最优结果构成的帕累托前沿，根据实际需求，求取最优方案下注采制度。研究结果表明，构建的代理模型对气藏采收率、CO₂封存率的预测准确度大于0.98，显示其预测准确度较高；对比优化前注采方案，优化后方案气藏采收率、CO₂封存率分别提升了9.03、5.53个百分点，证实了方法的可靠性，对类似气藏开发后期注CO₂提采-封存方案的编制具有借鉴意义。更多还原

【Abstract】 To maximize gas recovery and CO₂ storage through CO₂ injection in a gas reservoir with edge and bottom water, a typical volcanic gas reservoir in China was taken as an example. By integrating latin hypercube sampling（LHS）, long short-term memory neural network（LSTM）, and the multi-objective particle swarm optimization（MOPSO） algorithm, a surrogate model-based multi-objective optimization method was established to enhance gas recovery and geological CO₂ storage in the gas reservoirs. This method enabled the optimization of CO₂ injection and production well strategies. First, LHS and reservoir numerical simulation were used to generate 1 000 training and testing samples within a specified range of injection and production strategies. Next, an LSTM was trained to construct a surrogate model for predicting CO₂ injection and production dynamics, significantly improving computational efficiency. Finally, the MOPSO algorithm was applied to obtain the Pareto frontier composed of optimal solutions, from which the best injection-production strategy was selected based on practical requirements. The results demonstrate that the constructed surrogate model achieves a high prediction accuracy, with R² coefficients exceeding 0. 98 for both gas recovery and CO₂ storage rates. Compared to the injection-production strategy before optimization, the optimized solution increases gas recovery and CO₂ storage rates by 9. 03 and 5. 53 percentage points, respectively, confirming the reliability of the method. This study also provides valuable insights for designing CO₂ injection-enhanced gas recovery and storage schemes in other similar gas reservoirs during late-stage development.更多还原