时序示功图驱动的抽油机井结蜡预测及清蜡效果评价

Prediction of paraffin deposition and evaluation of paraffin removal effect for pumping wells driven by timing indicator diagram

檀朝东 陈培堯 杨亚少 于洋 宋健 冯钢 孙向飞

TAN Chaodong CHEN Peiyao YANG Yashao YU Yang SONG Jian FENG Gang SUN Xiangfei

中国石油大学(北京)人工智能学院 中国石油大学(北京)石油工程学院 中国石油长庆油田分公司油气工艺研究院 长庆油田分公司第三采油厂 北京雅丹石油技术开发有限公司

School of Artificial Intelligence, China University of Petroleum (Beijing), Beijing 102249, China School of Petroleum Engineering, China University of Petroleum (Beijing), Beijing 102249, China Research Institute of Oil and Gas Technology, CNPC Changqing Oilfield Company, Xi’an 710000, Shaanxi, China No.3 Oil Production Plant, CNPC Changqing Oilfield Company, Yinchuan 750000, Ningxia, China Beijing Yadan Petroleum Technology Development Co., Ltd., Beijing 100049, China

抽油机井结蜡是一个渐变的过程，序列示功图变化可以反映油井结蜡的程度。现场根据经验来预测结蜡程度和确定结蜡井热洗清蜡制度，决策能力低、效果差。应用人工智能技术认识结蜡程度与抽油机井示功图、电机运行参数、井口生产参数的关联关系，开展数据驱动的抽油机井结蜡预测预警方法和热洗效果评价的研究。应用残差卷积神经网络(ResNet)提取结蜡井示功图特征，使用聚类算法确定其结蜡等级，融合提取的示功图图形特征和12项生产参数建立样本集，利用长短时记忆神经网络(LSTM)构建序列到序列网络结构模型对样本集进行训练，建立结蜡等级预测模型，定量预测抽油机井的结蜡等级，并构建了油井清蜡效果评价指数Q。研究结果表明，建立的抽油机井结蜡预测模型和清蜡效果评价指数实现了油井结蜡等级的定量化预测、洗井周期的决策、清蜡效果的有效评价，对精准确定清蜡时机、评价清蜡效果具有较好的指导作用，有效避免了蜡卡躺井，同时延长了油井免洗周期。

The paraffin deposition on pumping well is a gradual process, and the changes in timing indicator diagram can reflect the degree of paraffin deposition in oil wells. Normally, it is usually to predict paraffin deposition degree and to determine thermal washing system for paraffin removal by using on-site experience, with low decision-making ability and poor paraffin removal effect. With the help of artificial intelligence technology, the correlation relationship between paraffin deposition degree and production parameters such as indicator diagram of pumping wells, motor operating parameters, and wellhead production parameters can be understood. Then, the researches on predicting, warning paraffin deposition and on evaluating thermal washing effect for pumping wells were performed driven by data. Extract the features of indicator diagram for paraffin deposition wells by using the residual convolutional neural network (ResNet), determine the paraffin deposition level by using the clustering algorithm, establish a sample set by combining the graphic features of the extracted indicator diagram with the 12 production parameters, train the sample set with a network structure model that constructed from sequence to sequence with long and short-term memory neural network (LSTM), establish a paraffin deposition level prediction model, quantitatively predict the paraffin deposition level for pumping wells, and construct an index evaluating paraffin removal effect for oil wells. The research results show that the established model predicting paraffin deposition and the constructed index evaluating paraffin removal effect for paraffin deposition wells have achieved quantitative prediction of paraffin precipitation level, decision-making of well cleaning cycle, and effective evaluation of paraffin removal effect. There is a perfect role for the established models to accurately guide paraffin removal timing and to guide paraffin removal evaluation, which may effectively avoid the paraffin stuck in the well and prolonging the no-clean cycle for oil wells.

10.13639/j.odpt.2022.01.019