提出者：

编者注：本文发表在 E&P 时事通讯中。在这里订阅 。

挪威是西欧最大的石油生产国。该国正在加强其电网并扩大其绿色能源产业，同时规划通过继续开发其石油和天然气资源来实现能源结构的可持续发展。那里生产的石油是碳足迹最低的石油之一。 

除其他原因外，一些油田将其 CO ₂重新注入地下进行封存，附近油田基础设施集中，并且海上设施大规模电气化（尤其是在 2020 年）。此外，议会刚刚同意到2030年该行业的排放量必须减少50%。因此，在挪威气候政策的框架内，挪威大陆架（NCS）的能源行业必须通过寻找开采出来的新石油排放量更低。

NCS 已经被广泛探索了大约 60 年。碳氢化合物生产始于 1971 年，当时巨型 Ekofisk 开始试生产。随后出现了一长串的发现，许多大型油田投入运行，总体产量不断增加，直到 2001 年，根据挪威石油管理局的数据，石油产量达到了峰值。

该地区的勘探仍在继续，钻井仍在继续，但随着 2010 年 Johan Sverdrup 的豁免，经济发现及其相关规模不断下降。现有基础设施的闲置产能开始增加，并开始成为一种负担；随着相关油田产量的衰退，操作和维护它的成本效益会降低。一些基础设施的生命周期已经通过延长寿命工程流程得到延长。其余的则标记为退役（这是一种标准做法，但有重要的缺点：时间长、成本高并且阻碍了维持生产的目标）。

发现可以与现有基础设施联系起来的新油田可以延迟退役，并提供以极低的碳足迹进行极低成本生产的可能性。 

为了在保持生产的同时降低排放，需要在成熟、多产的地区寻找隐藏的水库。这种必要性要求： 

1. 新数据; 
2. 对现有数据进行新的处理/分析；
3. 来自历史数据的新见解；
4. 新的工具箱；或者
5. 以上全部。 

在 Utsira High 和 South Viking Graben 之间的地区，我们正在采用方案 5。

TGS 和 AGS 于 2018 年至 2019 年通过 Utsira 海底节点 (OBN) 勘测获取了新的地震数据。这是有史以来第一次为勘探目的进行的大规模（超过 1,500 平方公里）、密集采样的 OBN 调查。区域体积采用最先进的技术进行处理。

为了使处理更进一步，TGS 采用目标导向的方法（重点关注北部地区 200 平方公里），使用动态匹配全波形反演（DM-FWI）和超长波反演技术，完成了新的模型构建工作。偏移（17公里分裂传播）潜水波能量和第二次反射DM-FWI，通过断层扫描和各向异性更新增强，使用井、表面和解释者的地质输入的未连接。

现代人工智能 (AI) 和机器学习 (ML) 技术可以利用历史数据，就像地球科学分析 (ESA) 在其 EarthNET 平台中所做的那样。因此，ESA 和 TGS 决定将 ML 和历史数据的力量与现代 Utsira OBN 结合起来。

当谈论人工智能和机器学习时，人们会想到大数据的概念。NCS 60年的地震、井、解释和经验，结合政府驱动的发布、存储成本的降低和云计算，提供了具有大数据特征的地理数据的积累。换句话说，它具有数量、展示价值、多样性、可访问性和及时性。

大数据和人工智能技术的引入，为挖掘海量地学数据中隐藏的地质规律提供了更加多样化的数据处理方式和可能性。

为了了解该区域中存在的数据量，图 1 显示了北海北部地图，以橙色突出显示了 Utsira OBN 勘测区域，其上覆盖着其他现有 3D 地震勘测区域（每个勘测多边形都标有轮廓）黑色和蓝色；深蓝色表示 TGS 的调查）。Utsira OBN 内或附近有 106 口已释放井（98 口来自 NCS，8 口来自英国大陆架），70 处勘测与其多边形相交，勘测总面积为 21,513 平方公里。

AI/ML 用于根据大 Utsira High 地区所有已发布的井数据来预测缺失的测井曲线、岩石和流体特性曲线，并通过稳健的交叉验证和盲测试方案来减少和控制不确定性。然后将这些特性曲线与 Utsira OBN 的角度叠加数据进行对齐和集成。 

机器学习模型经过训练，可以量化地震数据和目标属性之间的关系：岩性、孔隙度、密度、Vp 和 Vs 速度。此外，还进行了人工智能驱动的构造地震解释来解释断层和注入砂。后者通常是在大乌齐拉高地地区可预测的速度和密度异常。它们是富含沙子的沉积物，渗透性易于流化，尺寸从几百米到一公里不等，形状不规则，从V形到水平窗台。注入砂岩的 3D 测绘可用于通过处理进一步提高数据质量并识别预期的岩石体积。 

在初步研究中，将预测的 Vp 和提取的注入物与现代 DM-FWI 和各向异性更新模型进行比较，以提供更详细的地下图像。ESA 和 TGS 认为，处理工具箱与 AI/ML 工具箱并不相互排斥或替代；综合分析可以提取附加值并减少不确定性。

例如，图 2 显示，基于 DM-FWI 的速度捕获了由于速度增加而导致的异常。红色人工智能模型注射物被红色覆盖，确认了它们的存在并细化了它们的大小。 

该地区的几口井已经遇到了未知的充满盐水的注入岩，而潜在的注入岩仍处于勘探不足状态，几乎没有发现和生产油田。

图 3 将 DM-FWI 与 AI/ML 预测的 Vp 进行了比较。深度切片与钻出的充满盐水的注入物（A.1 和 A.2）和较低速注入物（B.2 和 B2）在钻出的充满 HC 的注入物区域相交。DM-FWI 和预测速度模型都支持观察结果，表明与预期注射物相关的较高速度。由于用于推导这些模型的算法的性质，这些模型具有明显不同的分辨率；然而，他们提供的地质见解相互印证。DM-FWI 提供了更多的成像细节以及地质体和复杂地质的更好定位，而 AI/ML 预测则利用这些信息、井中的地面实况以及历史知识来提供更详细的解析模型。  

Utsira OBN 的研究正在进行中，该研究将 DM-FWI 和先进处理与结构、井处的岩石物理信息和高分辨率空间预测特性相结合。包含最终 DM-FWI 模型的新迁移卷即将推出。

此外，可行性PS成像和高分辨率迁移测试正在进行中。所有这些结果将与使用 EarthNET 预测的特性和提取的结构结合起来进行分析。这是未来出版物的主题。 

评论

北海已完全成熟。尽管有现代技术，新的经济发现却变得稀缺。该地区很大一部分油气田将在十年内达到其经济生命周期的终点。在如此成熟的地区发现巨大油田的期望并不高，但昂贵的现有投资的可持续性取决于边际发现，这为聚集和聚集新产量提供了机会。 

当目标是在现有基础设施附近识别高价值的桶时，新技术提供了使用历史数据来提取以前在不投入大量资源（例如时间、人员和计算）的情况下无法获得的见解的能力。基于井的物性预测研究、3D 物性预测模型和自动地震解释的输出提供了宝贵的新见解，可以推动对这一多产区域的进一步勘探，特别是当这些新信息与现代处理数据并排分析时以及利用超长偏移潜水波和反射 DM-FWI 导出的高端速度模型。 

Presented by:

Editor's note: This article appears in the E&P newsletter. Subscribe here.

Norway is Western Europe’s biggest oil-producing nation. The country is strengthening its power grid and expanding its green energy industry while planning for a sustainable approach to the energy mix by continuing to exploit its oil and gas resources. The oil produced there is among the lowest in carbon footprint. 

Among other reasons, some fields reinject their CO₂ into the subsurface for storage, nearby fields infrastructure is clustered, and there has been a massive electrification of offshore facilities (especially in 2020). Moreover, the parliament just agreed that the industry’s emissions must be cut by 50% by 2030. So, in the framework of the Norwegian climate policy, the energy industry in the Norwegian Continental Shelf (NCS) must continue to deliver production by finding new oil that is extracted with ever lower emissions.

The NCS has been extensively explored for about 60 years. Hydrocarbon production started in 1971, when the giant Ekofisk started trial production. A long list of discoveries followed, and many large fields were brought online increasing overall production until 2001, when according to the Norwegian Petroleum Directorate, peak oil production was reached.

Exploration in the area has continued and wells are drilled, but economic discoveries and their associated sizes keep declining—with the exemption of Johan Sverdrup in 2010. The spare capacity of the existing infrastructure starts to increase and begins to become a liability; operating it and maintaining it becomes less cost-effective as the production of their associated fields decay. Some infrastructures’ life cycles have been extended with lifetime extension engineering processes. The rest is marked for decommissioning (a standard practice with important downsides: lengthy, expensive and hinders the goal of maintaining production).

Discovering new fields that can be tied back to existing infrastructure delays decommissioning and offers the possibility of very low-cost production with an extremely low carbon footprint. 

To maintain production while lowering emissions, there is a need to find the hidden reservoirs in mature, prolific areas. That necessity calls for: 

1. New data; 
2. New processing/analysis of existing data;
3. New insights from historical data;
4. New toolbox; or
5. All the above. 

In the area between the Utsira High and the South Viking Graben, we are working with Option 5.

New seismic data, via the Utsira ocean-bottom node (OBN) survey, were acquired by TGS and AGS in 2018-2019. It was the first large-scale (more than 1,500 sq km), densely sampled OBN survey ever acquired for exploration purposes. The regional volume was processed with state-of-the-art technology.

To take the processing one step further, new model building work has been done by TGS with a target-oriented approach (focusing in 200 sq km in the northern area), using dynamic-matching full waveform inversion (DM-FWI) with ultralong-offset (17-km split-spread) diving wave energy and a second pass of reflection DM-FWI, enhanced by tomography and anisotropy updates, using miss-ties to wells, surfaces and interpreter’s geological input.

Historical data can be leveraged by modern artificial intelligence (AI) and machine learning (ML) technology, like Earth Science Analytics (ESA) has done in its EarthNET platform. Thus, ESA and TGS decided to combine the power of ML and historical data with the modern Utsira OBN.

When talking about AI and ML, the concept of Big Data comes to mind. The 60-year history of seismic, wells, interpretations and experience in the NCS, combined with government-driven releases, reduction of storage costs and cloud computing provides an accumulation of geo-data that has the characteristics of Big Data. In other words, it has quantity, demonstrated value, diversity, accessibility and timeliness.

The introduction of Big Data and AI technology has provided more diverse data processing methods and the possibility for mining the geological rules hidden in the massive geoscientific data.

To give a sense of the amount of data that exist in this area, Figure 1 shows a map of the Northern North Sea highlighting the Utsira OBN survey area in orange, overlain by the area of other existing 3D seismic surveys (each survey polygon is outlined in black and colored in blue; dark blue indicates TGS’ surveys). There are 106 released wells (98 from NCS, 8 from the U.K. Continental Shelf) within or nearby Utsira OBN and 70 surveys intersect its polygon—a total area of 21,513 sq km of surveys.

AI/ML was used to predict missing well logs, rock and fluid property curves from all the released well data in the Greater Utsira High area with robust schemes for cross validation and blind testing to reduce and control uncertainties. These property curves were then aligned and integrated with the angle-stack data from Utsira OBN. 

ML models were trained to quantify the relationships between the seismic data and target properties: lithology, porosity, density, Vp and Vs velocities. Additionally, AI-driven structural seismic interpretation was performed to interpret faults and injected sands. The latter are commonly velocity and density anomalies that can be prospective in the Greater Utsira High area. They are sand-rich sediments with permeabilities prone to fluidization, and the sizes vary from a few hundred meters to a kilometer and have irregular shapes from v-brights to horizontal sills. The 3D mapping of the injected sandstones can be used for further improvements in data quality via processing and for identifying prospective rock volumes. 

In an initial study, predicted Vp and extracted injectites were compared against the modern DM-FWI and anisotropy updated model to provide a more detailed picture of the subsurface. ESA and TGS argue that the toolboxes, processing versus AI/ML, do not exclude or replace each other; a combined analysis can extract added value and reduce uncertainties.

Figure 2, for example, shows that the DM-FWI-based velocity captures the anomalies due to injectites as a velocity increase. The red AI-modeled injectites are overlain in red, confirming their presence and refining their size. 

Several wells in the area have encountered unprospective brine-filled injectites, whereas the prospective injectites still remain underexplored with few discoveries and producing fields.

Figure 3 compares the DM-FWI to the AI/ML predicted Vp. The depth slices intersect a drilled brine-filled injectite (A.1 and A.2) and lower velocity injectites (B.2 and B2) in an area of drilled HC-filled injectites. Both DM-FWI and predicted velocity models support observations indicating higher velocity associated with unprospective injectites. These models have distinctly different resolution due to the nature of the algorithms used to derive them; however, the geological insights that they provide confirm each other. The DM-FWI provides increased imaged detail and better placement of the geobodies and the complex geology, and the AI/ML prediction leverages this information, the ground truth from the wells, and the historic knowledge to deliver a much detailed, resolved model.  

The study of Utsira OBN, combining DM-FWI and advanced processing with structure, petrophysical information at the wells and high-resolution spatially predicted properties is ongoing. A newly migrated volume with the final DM-FWI model will be soon available.

In addition, feasibility P-S imaging and high-resolution migration tests are underway. All these results will be analyzed in combination with the predicted properties and extracted structure using EarthNET. This is the subject of future publications. 

Remarks

The North Sea is reaching its full maturity. Despite modern technology, new economic discoveries have become scarce. A significant portion of the oil and gas fields in this area will reach the end of their economic life cycle within a decade. There is not much expectation in discovering huge fields in such a mature area, but sustainability of the expensive existing investments depends on marginal discoveries, which offer opportunities for clustering and aggregating new volumes. 

When the goal is to identify highly valuable barrels nearby existing infrastructure, new technologies provide the ability to use historical data to extract insights that could not have previously been gained without investing large resources (e.g., time, personnel and compute). The output from the well-based property prediction study, the 3D-property prediction models and the automatic seismic interpretation provide invaluable new insights that can fuel further exploration of this prolific area, especially when this new information is analyzed side by side with modern processed data and high-end velocity models derived with ultralong-offset diving wave and reflection DM-FWI.