Abstract:

Due to the influence of shale reservoir characteristics and special seepage mechanisms, hydraulically fractured horizontal wells are commonly used for shale oil development. Production prediction is an important prerequisite for the optimization of fracturing schemes and the evaluation of economic indicators. However, commonly used production prediction methods still have limitations. Therefore, an intelligent post-fracturing production prediction method for fractured horizontal wells based on physical constraints and Fourier Neural Operator (FNO) is proposed to address the above issues. First, a mathematical model for fluid flow in fractured horizontal wells was established to derive typical solution of transient production rate. Validation using the Unsteady Production Analysis (RTA) module within commercial software confirmed high consistency between the model-generated and RTA transient production data. Subsequently, production data across diverse parameter ranges, generated from this validated seepage model, served as the FNO training dataset. Second, an FNO network was constructed and trained. Results demonstrate exceptional generalization ability, with coefficients of determination (R²) exceeding 0.99 for both training and validation sets. Finally, taking Well H1 in the Gulong shale oilfield as an example, the prediction of its oil production for the next two years was conducted.The specific steps are as follows: (1) The actual production data, geological data, and fracturing construction data of the well were collected, and the input parameters were obtained combined with well test interpretation; (2) The daily oil production of Well H1 was predicted using RTA and FNO methods respectively; (3) The two prediction methods were compared, and it was found that the prediction results of the FNO network are highly consistent with those of RTA (the coefficient of determination R² of discrete points is 0.95), which proves the accuracy and reliability of the FNO network prediction results. Moreover, the FNO network is more efficient and much faster in speed than RTA (FNO: 1-2 seconds, RTA: 600 seconds). In addition, we evaluated different intelligent production prediction models and compared the training errors of the FNO model with those of the Bi-LSTM, LSTM, and CNN models on the same dataset. The results show that the FNO model has the smallest error on the validation set and the strongest generalization ability. Therefore, the post-fracturing production prediction method for fractured horizontal wells based on the FNO network is expected to provide theoretical support for the optimization of hydraulically fractured shale oil wells and economic evaluation contributing to the efficient development of oil and gas reservoirs.

Key words: artificial intelligence, fractured horizontal wells, production prediction, FNO, well test interpretation

摘要：

页岩油井受页岩储层特征及特殊渗流机制影响，常使用压裂水平井开发，产量预测是其压裂方案优化与经济指标评价的重要前提，然而，目前常用的产量预测方法还不够完善。因此，针对上述问题提出一种基于物理约束和傅里叶神经算子(FNO)的压裂水平井压后智能产量预测方法：首先，建立压裂水平井渗流数学模型，求得瞬时产量典型解后，利用商业软件的不稳定产量分析方法(RTA)模块验证，结果显示两者瞬时产量典型解高度一致，再基于该渗流模型生成不同参数范围下的产量数据作为FNO网络数据集；其次，构建FNO网络并进行训练，训练结果显示，在训练集与验证集中，预测值与真实值的决定系数R²均超过0.99，表明网络具备优异的泛化能力；最后，以古龙页岩油H1井为实例，对其未来两年产量进行预测，具体步骤有：(1)收集该井实际生产资料、地质资料、压裂施工资料，结合试井解释得到输入参数；(2)分别采用RTA与FNO方法对H1井日产油量进行预测；(3)对比两种预测方法，发现FNO网络预测结果与RTA预测结果具有高度一致性(离散点决定系数R²为0.95)，证明了FNO网络预测结果的准确性与可靠性，并且FNO网络速度远快于RTA(FNO：1-2s，RTA：600s)。此外，对不同产量智能预测模型展开评估，对比了FNO与Bi-LSTM、LSTM、CNN模型在同一数据集上的训练误差，结果显示，FNO模型在验证集上误差最小，泛化能力最强。因此，基于FNO网络的压裂水平井压后产量预测方法有望为页岩油井压裂优化和经济评价提供理论支撑，助力油气藏高效开发。

关键词: 人工智能, 压裂水平井, 产量预测, 傅里叶神经算子, 试井解释

CLC Number: 

• TE319

• P618.13