The application of big data analysis in the optimizing and selecting artificial lift methods

The application of big data analysis in the optimizing and selecting artificial lift methods

Maintenance

Energy efficiency

Life cycle

Deep learning

Big data

Oil wells

Pumps

Losses

Reinforcement learning

Data handling

Petroleum reservoir evaluation

Function evaluation

Information analysis

Production efficiency

Neural network models

A reasonable method of artificial lift is the guarantee of efficient production of mechanical producing wells during the whole life cycle, however, there are many factors affecting the choice of artificial lift methods and most factors cannot be described with mathematical model. As a result, artificial lift method selecting mainly depends on expert experiences, and incorrect decision will lead to huge economic loss. In China, there are a large number of mechanical producing wells, and the large amount of data generated by artificial lift is of great value, which contains the adaptability of different types of oil wells to different artificial lift methods. This paper selects 11 parameters, such as well reservoir properties, maximum fluid production, well trajectory, pump depth, production gas oil ratio, pump inspection period, and production maintenance frequency as influence factors. The total data of 30,000 Wells of CNPC was initially selected as the big data analysis sample. An effect evaluation function is established taking pump efficiency, power consumption with lifting per ton liquid 100m, annual operating maintenance cost into account, then the oil well samples is selected further according the value of effect evaluation function, which will participate in machine learning. The paper present a neural network model based on deep reinforcement learning here, which can select the optimal artificial lift method by using influence factors and effect evaluation function. This model uses strategy gradient algorithm to train the samples. Experimental results show that the neural network model has fast convergence and high prediction accuracy. With the application of this model, artificial lift method selecting and effect analysis have been conducted for more than 4,000 oil wells of CNPC with different reservoir characteristics, different wellbore structures, and different fluid characteristics. The coincidence rate between calculation results of this model and actual production situation is 88.43%. Big data analysis provides a reliable, practical and intelligent method for optimizing and selecting artificial lift. 2018 Society of Petroleum Engineers. All rights reserved.

Shi, Junfeng

Han, Qiqing

Ren, Xianghui

Zhang, Xishun

Zhao, Ruidong

Chen, Shiwen

Li, Qiming

Zheng, Xiaoxiong

SPE Middle East Artificial Lift Conference and Exhibition 2018, MEAL 2018, November 28, 2018 - November 29, 2018

2018

conferencePaper

10.2118/192482-ms

http://dx.doi.org/10.2118/192482-ms