Case Studies

Current Status and Development Trends of Key Technologies for Intelligent Oilfield (Part 3)

2.3 Intelligent Production Operation

"IOT" is the foundation of intelligent oilfield construction. Major oilfields in China, such as Daqing and Shengli, have achieved real-time collection and processing of production site data based on years of IoT construction, and have the conditions for real-time perception of oil reservoir production and operation data. A series of intelligent application research has been carried out based on real-time production data. 

2.3.1 Intelligent Early Warning Technology for Oilfield Production Sites

With the continuous improvement of security requirements and the continuous development of information technology, major oil field production sites in China have achieved extensive coverage of video surveillance, reducing the time and cost of on-site well inspections. However, with the continuous generation of multiple massive video data, if we continue to use manual mode to analyze and judge videos, higher requirements will be placed on the number of monitoring personnel, personal experience, and analytical ability, which will greatly affect the overall efficiency of video applications. Therefore, intelligent recognition technology based on video big data is developing rapidly. Through advanced technologies such as video data interface integration, video data decoding and format conversion, virtual matrix architecture technology, cloud storage technology, GIS technology, and automatic recognition, intelligent expansion applications of video systems can be achieved.

Due to the lack of recognition models for specific oilfield scenarios and action features, existing video intelligent analysis functions are greatly affected by changes in working conditions, resulting in a significant reduction in the timeliness and accuracy of video applications. To this end, Shengli Oilfield has focused on specific scenarios and action characteristics and conducted research on intelligent video applications. One is to study and establish a comprehensive anti-terrorism warning model based on motion target detection and recognition. Based on the specific business characteristics of comprehensive governance and anti-terrorism, including perimeter prevention, regional intrusion, vehicle deployment, personnel deployment, etc., determine the structured characteristics of the analysis targets in each scenario. The background modeling method using a mixed Gaussian model improves the running speed and recognition accuracy of the background modeling algorithm, and improves the detection efficiency for moving targets. By using deep learning algorithms, deep feature extraction and cross comparison are performed on moving targets in the scene, and the accuracy of recognition is improved through multiple iterations. The second is to study and establish a safety event warning model based on standardized detection of direct work processes. By intelligently analyzing videos, effective early warning of safety hazards in direct work processes can be provided. According to the safety regulations of Sinopec for seven direct operational processes, including blind flange plugging, high-altitude operations, hot work, earth work, confined space operations, temporary electricity operations, and lifting operations, a structured model is established for various operational scenarios, which is used as a standard comparison template for visual intelligent analysis. By using deep learning algorithms, structured processing is carried out on videos from direct homework sites, extracting various key attributes in the scene, and comparing and analyzing them with 7 standard structured models to achieve intelligent identification of violations. Applied in business scenarios such as oilfield production and operation, safety and environmental protection, comprehensive governance and stability maintenance, emergency command, etc., the accuracy rate of standardized detection and alarm in the direct operation process is 75%, the accuracy rate of regional intrusion is 90%, and the accuracy rate of behavior recognition and analysis is 65%. It has achieved active detection, rapid analysis, early warning and alarm of safety events, video assisted analysis of equipment operation abnormalities, and comprehensive anti-terrorism intelligent warning for key locations and critical parts (Figure 6).

The traditional oil well production management mode is to take measures after problems occur. How to use massive real-time production data to achieve advanced warning of abnormal production problems is the key to improving the intelligent management of oil field production. Based on the collected real-time data, using trend analysis of historical oil well data, a production fault multi parameter warning model is established. The offset variable calculation engine is used to quickly access massive real-time data. Through clustering analysis and model recognition, abnormal wells are intelligently screened for fault warning. We have constructed and applied 30 warning models, including wellbore conditions, surface equipment, and surface pipeline networks, as well as over stress on pumping rods, wax buildup in oil wells, and belt breakage in pumping units. We have accumulated over 15000 oil wells and achieved abnormal data change exceeding limit alarms and multi parameter combination trend tracking warnings. In the process of optimizing oil well timing, in-depth analysis of influencing factors is carried out, and treatment work is carried out according to factors. For the main factors such as oil well waxing, pipeline freezing and blockage, and equipment failure, the trend of parameter changes is studied. Multi parameter combination warning tools are applied to construct three warning models: load fluctuation, oil pipe leakage, and mechanical equipment transmission failure. This effectively supports early detection and treatment of problems, reduces the "cure before disease" of oil well lying, reduces the oil well lying rate from 1.9% to 1.7%, and increases the oil well timing rate from 96.8% to 97.3%, making a significant contribution to stable production and increased production.

2.3.2 Predictive Maintenance Methods for Key Oilfield Equipment

Plunger type water injection pumps and centrifugal water injection pumps are key equipment in oil fields, which pose significant safety hazards due to their high-pressure operation. In the daily management process, if some hidden faults or malfunctions cannot be detected in a timely manner and continue to develop, it may lead to the failure and shutdown of the pump equipment, and even cause major repairs and scrapping of the equipment. At present, the monitoring methods only include video monitoring and single parameter operation monitoring, and it is often difficult to accurately and comprehensively monitor the pump operation status. Some faults still rely on on-site inspections to be discovered, and manual on-site inspections pose significant safety risks. Therefore, researchers propose the idea of conducting data mining methods based on production IoT data to improve efficiency and reduce costs. For example, Qin Tianfei et al. focused on the evaluation of the remaining life of temporary maintenance centrifugal pumps and obtained key indicators reflecting the degradation process of centrifugal pump health status through feature extraction and principal component analysis. They proposed a centrifugal pump overhaul threshold model based on optimized probability neural network (PNN) and a centrifugal pump remaining life evaluation model based on optimized limit learning machine (ELM), and optimized the training method through data dimensionality reduction processing and genetic particle swarm optimization algorithm. The application of practical cases shows that the proposed method has high prediction accuracy in the temporary repair prediction and remaining life assessment of centrifugal pump wells.

Since 2013, with the automatic collection and accumulation of real-time production data, Shengli Oilfield has conducted research on predictive maintenance of water injection pumps based on intelligent methods. Firstly, abnormal data in real-time data of high-pressure water injection pumps are automatically identified and processed. State feature information such as standard deviation, root mean, variance, kurtosis, and feature spectrum are extracted using feature extraction algorithms such as time-domain and frequency-domain. The faults of oilfield water injection pump equipment in the past five years are analyzed, and a sample dataset of four types of water injection pump faults, including power end faults, valve body assembly faults, plunger wear or corrosion, and belt faults, is established; Secondly, a diagnostic model combining expert knowledge with deep learning models based on deep belief networks (DBN) was constructed. For problems with clear mechanisms such as power end faults, valve body assembly faults, belt faults, etc., the knowledge model was applied for diagnosis. For problems with complex mechanisms such as balance plate wear, plunger wear or corrosion, a neural network model was applied for diagnosis. Since its application in oil fields, the diagnostic accuracy of this method has continuously improved, with a fault alarm rate of 95% (total number of alarms/number of alarms that should be triggered), a fault missed alarm rate of 5% (number of missed alarms/number of alarms that should be triggered), an alarm effectiveness rate of 90% (actual number of faults that occur/total number of alarms), and a daily alarm frequency of 0.3 for a single pump (number of alarms per day for a single pump). It has become an essential means for technical personnel in oil production management areas to manage water injection pumps.

3. Development Trend of Intelligent Oilfield Technology

Although China has made significant progress in the research and pilot construction of intelligent oilfield technology, there is still a certain gap compared to the advanced level abroad. There are still many technical problems that need to be addressed in supporting the digital transformation and intelligent development of the petroleum industry, mainly reflected in the following four aspects: Firstly, in terms of dynamic automatic monitoring and intelligent control technology for oilfield production sites, the IoT identification standards and analysis system for the production front-end have not been established, the key technology for highly reliable IoT in oil and water wells and wellbore has not yet been broken through, and the capabilities of edge intelligence and edge cloud collaboration technology are insufficient; Secondly, in terms of intelligent technology for the new generation of big data in oilfield industry, China has not yet formed a unified standard for data and achievement sharing applications, and the efficient management and intelligent service technology capabilities for massive exploration and development data are insufficient. A series of big data models that are suitable for diagnosing, analyzing, predicting, warning, and optimizing decision-making in various business processes of oilfield development and production have not been fully established; Thirdly, in terms of intelligent optimization technology for reservoir development, high-level modeling and modeling software is highly dependent on foreign countries. The automatic updating method of models based on big data and artificial intelligence is not suitable for reservoirs, and the development plan lacks technical means for automatic optimization and push; Fourthly, in terms of oilfield digital twin and intelligent operation command technology, there is a lack of automatic construction of oilfield digital twin models and integrated intelligent diagnosis technology for reservoirs, wellbore, and ground. The integration technology of production and operation is weak, and the intelligent identification technology for production process risks is single. A production operation command technology support system based on abnormal management of production and operation has not been established.

Therefore, this article believes that the next technological development of intelligent oil fields mainly includes the following four aspects. Firstly, dynamic automatic monitoring and intelligent control technology for oilfield production sites. Following the development of new technologies such as the Internet of Things, artificial intelligence and 5G, focusing on the development needs of oilfield intelligent production capacity, focusing on the needs of unmanned, few people and automated management of oilfield production sites, we studied the coding standards, identification carrier standards and key technologies of identification analysis of oilfield industrial Internet, supported the whole process, whole node Internet of Things of oilfield production, and realized real-time perception of oilfield production dynamics and full life cycle management of equipment, facilities, materials, etc; Research on dynamic intelligent monitoring technology for oil reservoirs and on-site electronic inspection and unmanned control technology, replacing traditional manual operations, researching a series of models for chain control and closed-loop optimization control of the entire process of wellbore lifting, surface injection and production, and tackling intelligent control technology for key links such as oil and water wells, stations, and pipelines, providing technical support for unmanned and unmanned management of oilfield sites; Research and form a series of oilfield edge computing technologies, tackle key technologies of edge cloud collaboration, support the efficient operation of oilfield remote real-time management and control system, and provide technical support for significantly reducing front-line labor, improving labor productivity, and ensuring efficient and long-life production of equipment and facilities.

Secondly, the new generation of intelligent big data technology for oilfield industry. Starting from big data and artificial intelligence technology, with the deep integration of intelligent technology and oil and gas exploration and development as the main strategic direction, based on the construction of data resource centers, and based on scientific matching of data, business, and algorithms, small tasks, multiple data, strong correlation, hybrid technology, and big data analysis are carried out to achieve artificial intelligence learning, memory, and recognition, empowering production practice with massive data. Based on the needs of petroleum big data management, advanced data management technologies such as data lakes and data warehouses are utilized to conduct research on the integration of lake and warehouse data fusion methods, establish a unified data model, data services, and data management standards for China's petroleum industry, and achieve data asset management. Based on the requirements of cross disciplinary data sharing and system linkage, research intelligent service technology for exploration and development application data, support cross disciplinary and cross type big data applications, study sharing standards between various professional software, conduct research on intelligent technology for oilfield application system linkage and process integration, and support intelligent optimization of business processes. Carry out research on big data mining and analysis models for oilfield industry, study sample calibration techniques and standards in core professional fields of exploration and development, construct big data mining models for petroleum professional fields, and support intelligent transformation in oilfield development research, analysis, optimization, and other fields. Research and development of artificial intelligence based seismic data automation processing, achieving intelligent noise suppression, fully automated first break picking, time-frequency domain well seismic fusion intelligent learning to improve resolution, intelligent evaluation of processing effectiveness, and conducting technical research on intelligent recognition of geological targets based on three-dimensional seismic bodies, intelligent description and evaluation of reservoirs and fluids, auxiliary exploration decision-making, and optimal deployment of exploration wells.

Thirdly, intelligent optimization technology for reservoir development. Research the knowledge system of geological modeling, combine big data and artificial intelligence technology, construct new methods and algorithms for reservoir modeling, form intelligent geological modeling technology for oil reservoirs, improve modeling efficiency and accuracy; On this basis, research on intelligent numerical simulation technology for oil reservoirs, break through the bottleneck of automatic historical fitting technology for oil reservoirs, develop a new generation of surface numerical simulators for oil reservoir wellbore, improve the automation and intelligence level of oil reservoir numerical simulation, and provide support for automatic simulation optimization of development plans. Research on dynamic intelligent analysis, scheme intelligent optimization, and effect intelligent evaluation technologies for reservoir development, to achieve automatic analysis of contradictions and potentials in reservoir development, as well as intelligent push of schemes, and to form a full lifecycle management model for reservoirs, including development deployment, analysis, adjustment, optimization, and evaluation. This provides support for optimizing reservoir development technology policies, dynamically implementing comprehensive adjustments, and reducing natural decline.

Fourthly, the digital twin of oil fields and intelligent operation and command technology. Based on the achievements of information technology construction in oil and gas production, research is being conducted on the construction technology of integrated "digital twins" for reservoirs, wellbore, and surface, in order to enhance the visualization and monitoring capabilities of oil fields. Research on risk warning technology for oilfield development and production process, to achieve early warning and prediction of development and production process risks such as indicator changes, production fluctuations, safety and environmental protection, and reservoir management. Research on risk warning technology for oilfield development and production process, to achieve early warning and prediction of development and production process risks such as indicator changes, production fluctuations, safety and environmental protection, and reservoir management. Taking oilfield development and production business as the main line, focusing on reservoir operation and management, covering all oilfield production and operation businesses, developing an intelligent operation and command system with functions such as production monitoring, dynamic management, collaborative management, evaluation and assessment, QHSSE, emergency response, etc., leveraging real-time data to support visual monitoring, anomaly diagnosis, early warning prediction, auxiliary decision-making, intelligent operation, etc., achieving vertical hierarchical connectivity, horizontal business collaboration, and gradually transitioning from production command to production operation, improving labor productivity, reducing development costs, enhancing the ability to respond to and prevent risks, and ensuring national energy security.