Innovative Convolutional Neural Networks Applied in Dynamometer Cards Generation
- Yi Peng (PetroChina RIPED) | Ruidong Zhao (PetroChina RIPED) | Xishun Zhang (PetroChina RIPED) | Junfeng Shi (PetroChina RIPED) | Shiwen Chen (PetroChina RIPED) | Qingming Gan (Oil & Gas Technology Research Institute Of Changqing Oilfield Company, PetroChina) | Gangyao Li (Xinjiang Oilfield Company, PetroChina) | Xiaoxiong Zhen (Oil Production Technology Institute Of Dagang Oilfield Company, PetroChina) | Tao Han (Oil Production Technology Institute Of Dagang Oilfield Company, PetroChina)
- Document ID
- Society of Petroleum Engineers
- SPE Western Regional Meeting, 23-26 April, San Jose, California, USA
- Publication Date
- Document Type
- Conference Paper
- 2019. Society of Petroleum Engineers
- Sucker Rod Pumping, Artificial Lift, Dynamometer Card, Artificial Intelligence, Deep Learning
- 2 in the last 30 days
- 97 since 2007
- Show more detail
- View rights & permissions
|SPE Member Price:||USD 5.00|
|SPE Non-Member Price:||USD 28.00|
There are nearly 200,000 oil wells producing in PetroChina and 94% are sucker-rod pumping wells. Dynamometer card is essential for well diagnosis and optimization, which cost huge investment on sensor installations and maintenance. This paper present an innovative artificial intelligence method to get the card data directly from electrical parameters.
With the development of artificial intelligent technology, many challenges in oil and gas industry have the potential to be addressed by developing artificial intelligent and Big Data analytic tools. This paper presents a method to get the dynamometer card using the electrical data collected during production. In our method, we start with building our training datasets consist of dynamometer card and corresponding electrical data with various well performance conditions. We using the state of art artificial intelligent deep learning algorithm to training the data and make a general model to generate the dynamometer card from electrical data.
This method is based on image processing and deep learning, which eliminates the disadvantage of conventional physical-mechanism based model. After tested training datasets in several oilfields in PetroChina, the model demonstrates very high accuracy with almost 90% similarity to corresponding dynamometer card. The calculated dynamometer card from our model are good consistent with the field data too. In addition, this method has been testing on hundreds of wells newly produced well utilizing sucker-rod pumping system, which would reduce millions of investment on the sensors and equipment on average 5000 newly producing wells in PetroChina each year.
As a result, with the fast development of artificial intelligence technology and expanding training datasets, the deep learning method will be fully utilized in more sucker-rod pump wells to generate the dynamometer card, which is also part of the intelligent well strategy. In the low oil price trend, artificial intelligence is a good choice to lower the investment and maintenance cost for mature fields and newly produced fields.
|File Size||1 MB||Number of Pages||9|