Abstract:

In surface gathering and transportation systems for ultra-high water-cut oilfields, gathering without hot water mixing have emerged as an effective measure for energy saving and carbon reduction. However, the gelation characteristics of crude oil with high water content lead to complex and unpredictable wellhead pressure evolution, further constraining the low-carbon synergistic operation of the entire extraction-to-surface gathering process. Based on static and dynamic operational data from actual production wells in a certain oilfield, this paper considers the impact of time-domain nonlinear shifts on wellhead pressure evolution patterns. The Dynamic Time Warping algorithm (DTW) is employed to identify temporal characteristics of wellhead pressure in crude oil gathering pipelines without water mixed. Feature expansion reconstruction and mutual information-based dimensionality reduction are then applied to construct a nonlinear reconstructed dataset. Subsequently, by adopting a Mixture of Experts (MoE) model as the intelligent prediction framework, with Random Forest Regression (RFR), Neural Network Regression (NNR), and Gradient Boosting Regression (GBR) selected as sub-expert models, a DTW-MoE wellhead pressure prediction model was established, aiming to reveal the influencing factors of wellhead pressure in gathering pipelines and rank their degrees of impact. Results indicate that the four optimal path matches in the DTW-MoE wellhead pressure prediction model are ξ₁-GBR-72, ξ₂-GBR-6, ξ₃-RFR-3 and ξ₄-NNR-6. Particularly in the high pressure section where the wellhead pressure exceeds 0.5 MPa, the prediction errors are all controlled within 10%. The DTW algorithm identified four clusters with distinct temporal characteristics. Cluster ξ₁ is primarily governed by the synergistic effects of the production radius and liquids production rate on wellhead pressure, with corresponding composite SHAP values of 0.0386 and 0.0225 respectively. For Clusters ξ₂, ξ₃ and ξ₄, the key influencing factors are the production radius, liquids production rate, and pipe diameter, though their relative impacts vary significantly. The findings of these studies provide valuable insights for energy conservation and emission reduction oversight in oilfield gathering and transportation systems, particularly in the context of digital intelligence empowerment and green transformation.

Key words: wellhead pressure, gelation characteristics, gathering without hot water mixing, route matching, intelligent prediction

摘要：

停掺集输工艺已然成为特高含水油田地面集输系统节能减碳的有效举措之一，但高含水原油胶凝特性导致井口回压演变规律复杂且难以预测，进一步制约“采油工程—地面集输”全流程低碳协同运行。本文基于某油田实际生产运行井的静动态运行数据，考虑时域非线性偏移对井口回压演变规律的影响，采用动态规整聚类算法(DTW)识别原油停掺集输管道的井口回压时序特征，并通过特征拓展重构与互信息降维，构建非线性重构数据集，进而以混合专家模型(MoE)为智能化预测架构，选取随机森林回归(RFR)、神经网络回归(NNR)和梯度提升回归(GBR)为子专家模型，建立DTW-MoE井口回压预测模型，揭示原油停掺集输管道井口回压影响因素及影响程度排序。结果表明，DTW-MoE井口回压预测模型中4组最优路由匹配关联分别为ξ₁-GBR-72、ξ₂-GBR-6、ξ₃-RFR-3、ξ₄-NNR-6，在井口回压超过0.5 MPa的高压区段中预测误差均小于10%；DTW动态规整聚类划分了具有不同时序特征的4类聚类类别，聚类类别ξ₁的井口回压受集油半径与产液量协同主导，其综合SHAP值分别为0.0386、0.0225；聚类类别ξ₂、ξ₃和ξ₄的井口回压主导因素为集油半径、产液量、管径，但不同因素的影响程度存在较大差异。研究成果可为数智赋能与绿色转型下的油田集输系统调控过程节能减碳提供有益参考。

关键词: 井口回压, 胶凝特性, 停掺集输, 路由匹配, 智能预测

CLC Number: 

• TE88

• TE866