Shaping the Future of Oilfields with Data and AI

Amid the relentless tide of technological innovation, AI has emerged as a pivotal driver of the digital revolution and industrial transformation, serving as an engine for cultivating and developing new productive forces. Today, AI development has entered the era of large models and the competition inside the global energy industry is rapidly evolving. Energy companies are increasingly partnering with leading tech firms to secure a strategic edge in innovation, aiming to transform vast amounts of complex industry data and business insights into a competitive advantage for future growth.

Moving forward, the goal of smart oilfield construction is shifting from mere digitization and networking toward full intelligence and autonomy. At the heart of this transformation lies the development of large model-powered capabilities—spanning the entire chain of oil and gas (including coalbed methane) exploration, development, storage, transportation, production, and safety—for intelligent perception, knowledge-driven operations, and autonomous decision making, ultimately achieving full intelligence.

As a globally leading energy enterprise, China National Petroleum Corporation (CNPC) remains steadfast in its commitment to serving national strategies and embracing its role in supporting the country's overarching priorities. With AI as the cornerstone for its strategic transformation, CNPC vigorously implements three major initiatives—better IT infrastructure, digital transformation, and intelligent development—for the creation of a "Digital and Intelligent CNPC." Throughout this journey, CNPC uses its scenarios across the entire value chain to help create an innovative Kunlun Model tailored to the industry. This contributes to an intelligent new infrastructure that serves as a benchmark for implementing the national "AI+" strategy.

In May 2024, CNPC, together with Huawei and other leading tech companies, officially launched the Kunlun Model. By August of the same year, the model had received national certification, becoming the first in China's energy and chemical industry to gain such recognition. Employing a Mixture of Experts (MoE) architecture, the model integrates diverse and heterogeneous data from the energy and chemical sectors. It also boasts billions of parameters and superior generalization capabilities. In fact, the current version features 330 billion parameters, spanning 26 business lines, 119 business fields, and 100 core scenarios. This signifies the creation of an intelligent ecosystem with "thousands of applications, hundreds of scenarios, and dozens of domains", forming a robust technical loop. This positions the Kunlun model as not only a key driver for building a "Digital and Intelligent CNPC" but also a leader in the "AI+" initiative. The Kunlun model is now one of the largest and most comprehensive energy industry models globally, setting a new benchmark for industry intelligence.

As the largest onshore oil and gas production base in China, Changqing Oilfield has maintained an annual oil and gas production of over 60 million tons for years, playing a crucial role in ensuring national energy security. Spanning five provinces and regions—Shaanxi, Gansu, Ningxia, Inner Mongolia, and Shanxi, Changqing Oilfield covers a 360,000 square kilometer area within the Ordos Basin, serving as the primary force in exploring and developing this super basin. The Oilfield faces significant challenges, including managing a complex production environment with 126,000 oil and gas wells, over 3,000 stations, and 133,000 kilometers of pipelines across vast areas. Additionally, low-permeability, low-pressure, and low-yield oil and gas reservoirs, coupled with high-risk operations such as seismic exploration, digital rock analysis, drilling and fracturing, hot work, confined space operations, inspections in sensitive areas of the Yellow River basin, and unattended operations, make traditional manual oversight inadequate for modern development needs.

To address these challenges, the Intelligent Identification for Hidden Risks in Special Operations project (the "Project") was launched. Guided by "integrated training and inference, cloud-edge-end collaboration", the Project focuses on high-risk oil and gas operation scenarios, proactively seeking ways to integrate intelligent systems with the energy industry. Drawing valuable insights from Changqing Oilfield, the project supports the Kunlun Model's development, tackles safety management challenges, and improves production models in the industry with intelligence. In response, researchers from Changqing Oilfield, Huawei, and Kunlun Digital have united their efforts, leveraging digital and intelligent technologies to inject new vitality into the development of the Oilfield.

In line with the "Digital and Intelligent CNPC", Changqing Oilfield has set its sights on Digital Transformation 2.0. Leveraging Huawei's "Kunpeng + Ascend + Kylin OS" technology stack, we have laid a robust foundation, achieving significant technological advancements from the foundational hardware to the upper layer applications. An innovative three-tier collaborative AI architecture has been developed: CNPC Group cloud-based centralized training + Changqing Oilfield regional unified management of inference + real-time inference at edge wells and stations. This structure encompasses an intelligent algorithm matrix spanning five main categories and 46 sub-scenarios, greatly enhancing intelligent control for safe production at oilfields.

Changqing Oilfield applies the Kunlun Model to multiple core scenarios, such as four-dimensional reservoir modeling, horizontal drilling optimization, intelligent decision-making on fracturing, intelligent well and station control, production operation optimization, safety and environmental protection monitoring, and intelligent leakage detection.

By analyzing multimodal data from major initiatives such as the efficient development of 30 billion cubic meters of tight gas in Sulige, production of 3.5 million tons of shale oil in Longdong, establishment of an intelligent gas production plant in Jingbian (Yulin) with a capacity of billions of cubic meters, implementation of a smart gas processing plant for upper paleozoic natural gas, creation of a demonstration area for intelligent management in Zhenbei Oilfield, development of intelligent pipelines, and execution of safety monitoring, we have improved the detection accuracy in certain scenarios from 78% with traditional methods to 92% using the model. The predictive maintenance feature for oil and gas processing facilities has reduced unplanned downtime by 50%, driving the intelligent transformation of the energy industry. Currently, Changqing Oilfield is sparing no efforts to advance the Project. By integrating AI, big data, cloud computing, and other cutting edge information technologies with the traditional energy sector, we are charting a unique course for intelligent development tailored to Changqing Oilfield.

In the early stages of the project, Changqing Oilfield utilized advanced proprietary technology in collaboration with top Chinese tech firms like Huawei and Kunlun Digital to establish a new computing infrastructure. Researchers ventured into field operations, gathering 2 million images and 150,000 hours of video footage, and developed a comprehensive safety knowledge base comprising 12,000 rules. This extensive dataset provided a solid foundation for algorithm training.

To build a robust technical system, we utilized Huawei's "Kunpeng + Ascend + Kylin OS" framework to achieve end-to-end independent innovation, from foundational hardware to upper-layer applications. By embracing openness and innovation, we successfully break through external tech barriers. The system now covers all wells, pipelines, stations and production management areas of Changqing Oilfield, delivering three key capabilities:

• A platform for the automatic annotation of millions of samples

• A distributed incremental learning framework for algorithms

• A lightweight model deployment engine for edge devices

During Phase I, the Project has rolled out 20 high-precision models across various scenarios within the oilfield to create an intelligent safety protection system for key oil and gas production areas. This effort has set new industry standards and delivered impressive results:

① Industry-leading recognition accuracy:

The models have achieved an impressive 94.1% accuracy in detecting hidden risks, marking an 18.7% improvement over traditional methods. They also demonstrate a recall rate of 90.75%, a 32.5% increase, and maintain a false alarm rate below 5% in specific scenarios. These advancements effectively decrease false positives and negatives, leading to a reduced misjudgment rate in high-risk operations and fewer unnecessary alerts.

② Closed-loop risk response management:

The inference latency on the device side is under 800 ms, enabling the realtime reporting of potential risks and facilitating a systematic closed-loop control process of "risk perception > decision making > action."

③ Efficient and autonomous development and O&M:

Automated and self-managed sample annotation triples development efficiency, reduces O&M costs by 40%, and ensures 100% independent control of core code.

④ Better operational efficiency:

The intervention rate for high-risk operations has dropped by 70%, easing safety management pressures and improving overall operational efficiency by over 25%.

⑤ Improved safety and costs:

The safety management capability for high-risk operations such as hot work is enhanced, and the overall O&M cost of the oilfield is reduced by 18%.

The Project establishes a new value paradigm in the energy sector:

• Safety: Achieves real-time detection of critical hidden risks by 100% , reducing the probability of major accidents by 90%.

• Efficiency: Increases the efficiency of well and station inspections fivefold, saving over 500,000 hours of safety work annually.

• Open innovation: Establishes a comprehensive set of technology standards for AI, lowering the threshold for intelligent transformation in the industry.

Central to this project is the "Changqing Paradigm"—founded on comprehensive independent innovation, a three-tier collaborative innovation architecture, an agile organization that integrates business and technology, and an open, mutually beneficial ecosystem. This paradigm offers a standardized, replicable model of success for the intelligent upgrade of the entire oil and petrochemical industry, as well as process industries. The autonomous innovation path and methodology validated by Changqing Oilfield have been rapidly adopted in multiple oilfields like Daqing and Tarim, effectively reducing the barriers and trial-and-error costs for intelligent transformation in the industry. This exemplifies the positive cycle of "application through innovation, innovation through application" as required by the "AI+" initiative. It signifies that China is leading globally in establishing an approach for intelligent innovation in the energy sector—"technology research and development, scenario validation, and then promotion for large-scale application"— providing a Chinese solution for global process industries. This approach holds significant strategic implications for bolstering national energy security.

During the "14th Five-Year Plan" period(2021-2025), Changqing Oilfield advanced the "Changqing Digital Intelligence 2.0" initiative, focusing on four key development areas: intelligent collaborative exploration and development, intelligent production management, intelligent field operation control, and data-driven business decision-making.

Breakthroughs were achieved in technologies such as intelligent analysis of 3D seismic data and intelligent "three-in-one" image recognition, enabling cloud storage and the shared application of basin-wide seismic data. Using the AI-powered dynamic reservoir analysis system, Changqing Oilfield tracked 240 key oil reservoir development metrics with prompt alerts and diagnosed the causes of production fluctuations in over 80,000 oil-water wells. We have established China's largest industrial internet platform, achieving centralized management of industrial production data.

By employing technologies such as autonomous optimization of oil, gas, and water well systems, intelligent well shutdown, intelligent drainage gas recovery, unmanned station operation, and intelligent pipeline inspection, we created a new operational model characterized by "centralized management, integrated supervision, and coordinated scheduling." As a result, Changqing Oilfield became one of the first in China to achieve Level 4 certification under the national standard for digital transformation maturity. Our transformation practices have been included in the white paper titled China's Energy Transition released by the China's State Council Information Office.

In short, the deep application of AI technologies in key oil and gas production processes has allowed Changqing Oilfield to achieve instant risk detection, halve high-risk misjudgments, vastly improve efficiency, and continuously optimize costs. More importantly, it has enhanced the safety of employees and strengthened the foundation of national energy security.

Changqing Oilfield, together with Huawei, Kunlun Digital, and other leading companies, has successfully created the world's first benchmark case of "AI foundation model + full independent innovation" that has been verified by large-scale production practices. We are dedicated to building a large, strong, robust, and sustainable oilfield. We look forward to working closely with global partners to establish international standards for intelligent oilfields. Riding the wave of AI, we strive to build future-proof intelligent oilfields and advance the safety, efficiency, and sustainability of the global energy sector

Disclaimer: The views and opinions expressed in this article are those of the author and do not necessarily reflect the official policy, position, products, and technologies of Huawei Technologies Co., Ltd. If you need to learn more about the products and technologies of Huawei Technologies Co., Ltd., please visit our product pages or contact us.