Abstract:

With the rapid development of the natural gas industry， the demand for efficient and reliable gas well control technologies is increasing day by day. Among the key technologies for improving the gas production efficiency of low-pressure gas wells， the research on the intelligent evaluation system of plunger gas lift is of great significance. In order to improve the accuracy and efficiency of the evaluation system， an intelligent evaluation algorithm of the plunger gas lift based on deep learning was proposed. Employing Convolutional Neural Network （CNN） as the core of the algorithm， the comprehensive evaluation of performance of the plunger lift control algorithm was realized by analyzing key features such as oil pressure， casing pressure， switching well status， and production regulations. Extensive test results on actual gas well data demonstrated that the algorithm can effectively improve the accuracy and stability of evaluation. Specifically，compared with Bayesian neural network （BNN） and Attention-LSTM algorithm， the algorithm improves the prediction accuracy of normal rate by 14% and 5%， respectively， and improves the prediction accuracy of stable running rate by 6% compared with backpropagation neural network （BPNN）.

Key words: plunger gas lift, deep learning, Convolutional Neural Network (CNN), intelligent evaluation, gas well control

摘要：

随着天然气工业的快速发展，对高效可靠的气井控制技术的需求日益增长。在提高低压气井产气效率的关键技术中，柱塞气举的智能评价系统研究具有重要意义。为提高评价系统的准确性和效率，提出一种基于深度学习的柱塞气举智能评价算法。所提算法采用卷积神经网络（CNN）作为核心，通过分析油压、套压、开关井状态和生产制度等关键特征，实现对柱塞气举控制算法性能的综合评估。在实际采集的气井数据上进行广泛测试结果表明，所提算法能有效提高评价的准确率和稳定性。具体来说，所提算法在正常率预测精度方面，相较于贝叶斯神经网络（BNN）和带注意力机制的长短期记忆网络（Attention-LSTM）算法分别提高了14%和5%，在稳定运行率预测精度方面，相较于反向传播神经网络（BPNN）提高了6%。

关键词: 柱塞气举, 深度学习, 卷积神经网络, 智能评价, 气井控制