Case Studies

Current Status and Development Trends of Key Technologies for Intelligent Oilfield (Part 2)

2.1.2 Intelligent Analysis Technology for Oil and Gas Migration

Oil and gas migration is one of the fundamental theoretical topics in petroleum geology, and it is also a key issue in oil and gas exploration and development. However, the migration and accumulation of oil and gas have always been a relatively weak link in China's petroleum geological research. Traditional manual and qualitative research on oil and gas migration has problems such as cumbersome manual statistical data, large workload, low accuracy in qualitative characterization of transport systems, and lack of intuitive planar static descriptions, making it difficult to achieve quantitative evaluation.

In the application of intelligent methods to study the quantitative characterization of oil and gas migration, some scholars have combined traditional dynamic simulation with neural network simulation. Based on the dynamic simulation of three-dimensional structure formation body, the heterogeneous complex channel system is transformed into a finite number of simple homogeneous bodies based on the unit body model. Traditional dynamic simulation is used to solve the phase and driving forces, and neural network and other technical methods are used to solve the oil and gas migration direction, migration rate, and migration amount between unit bodies. A solution is provided for the quantitative simulation of three-dimensional oil and gas migration.

The transport system, as a bridge connecting source rocks and traps, plays a crucial role in the process of oil and gas migration and accumulation. Shengli Oilfield proposes a quantitative evaluation method for the opening and closing of faults at different positions in three-dimensional space using fault connectivity probability by analyzing the influence of multiple factors such as fluid pressure, mudstone coating, and normal stress on the fracture surface. By analyzing the relationship between the transport elements of skeletal sand bodies and oil and gas display, the main controlling factors (dip angle, physical properties) and their quantitative characterization parameters of sand body transport performance are determined, forming a quantitative evaluation formula and calculation model for transport elements. On this basis, based on the transport capacity of skeletal sand bodies and faults, taking into account the heterogeneity of the transport layer, as well as the influence of factors such as fluid potential and sandstone percentage, the streamline method can be used to obtain information such as the migration intensity of oil and gas in the transport layer, the simulated accumulation amount of traps, and the migration trajectory of oil and gas. This method has been applied in the exploration practice of Shengli Oilfield. In terms of screening the dominant migration paths between Sanhe Village, Kenxi Slope Belt, and Bonan Depression, it has been clarified that there are mainly two dominant migration paths for oil and gas generated from the source rocks of Sha4 to Sha3 in the Bonan Depression in this area (Figure 2): one is the dominant migration path from Well Yi 633 to Well Luo 358, and the other is the dominant migration path from Well Yi 96 to Well Luo 651, which is consistent with the macroscopic understanding of oil and gas migration and accumulation in this area from a planar perspective.


2.1.3 Automatic Seismic Layer Interpretation Technology

Seismic layer interpretation is a fundamental task in oil and gas exploration and development, and the identification and tracking of layers is an important part of seismic layer interpretation. Its accuracy directly affects the rationality of the final seismic interpretation. There are currently two main approaches to using artificial intelligence methods to interpret layers, with the main difference being in sample construction. One is based on the combination of well logging interpretation results and seismic data, while the other is based solely on seismic data or well logging data itself. In terms of machine learning methods, supervised learning is the main approach, and some unsupervised learning algorithms based on classification ideas have also been studied.

Shengli Oilfield has proposed a new method for automatic interpretation of layers based on reflection structure constraints, which is essentially a supervised learning method based on seismic data. The main idea is to divide the waveform units based on the seismic data of the target layer, obtain partial layer interpretation results through manual interpretation, and label them as sample labels in each waveform unit. At the same time, time-frequency analysis is performed on each waveform unit to elevate the one-dimensional time series signal of a single waveform unit to a two-dimensional space. Based on this, a layer recognition sample library is constructed, and the DFCNN deep learning algorithm is used to automatically identify the entire layer in the area. Quantitative reflection structure constraint information such as layer optimization based on dip angle and layer recognition anomaly point discrimination based on trend fitting function are introduced to optimize the DFCNN automatic recognition results. The application results show that the layer automatic interpretation technology based on reflection structure constraints ensures the continuity of the dip angle trend and the trend law of the spatial distribution of the formation (Figure 3). Compared with manual interpretation results, the overall agreement rate of the automatic interpretation results reaches over 86%. Compared with conventional layer automatic interpretation methods, the accuracy and reliability have been effectively improved.


2.2 Intelligent Development Technology

2.2.1 Intelligent Updating Technology for Reservoir Models

With the continuous deepening of oil reservoir development, geological modeling and numerical simulation of oil reservoirs have become the main technical means for comprehensive development research. Currently, most oil field development in China has achieved the promotion and application of integrated modeling, numerical simulation and research. However, the geological conditions of old onshore oil fields in China are complex, and most of them have entered the late stage of development. Various process measures are adjusted frequently. How to quickly, dynamically, and accurately update the geological model of the reservoir, guide the deepening understanding of remaining oil, and optimize development plans has become a key factor affecting the quality of oil field development. The traditional methods for updating reservoir geological models and historical fitting are mainly carried out manually, relying heavily on human experience, taking a long time, and the timeliness of model tracking and field application is poor, which restricts the pace of high-quality development. In recent years, in-depth research has been conducted both domestically and internationally on the intelligent updating of reservoir models or the automatic (or auxiliary) historical fitting of reservoir numerical simulations. In particular, optimization algorithms have gradually transitioned from traditional gradient algorithms and evolutionary algorithms to neural network methods, ensemble Kalman filtering methods, and even various hybrid methods. Intelligent algorithms are increasingly being applied in the research of automatic historical fitting, making the intelligent updating of reservoir models gradually move from theoretical exploration to practical application.

In the research of automatic updating methods for reservoir models, it is necessary to first determine the relevant static and dynamic parameters that affect the fitting accuracy of reservoir model calculation indicators, and then develop optimization algorithms for different optimization objectives. In response to the problem of automatic historical fitting in numerical simulation of oil reservoirs, Shengli Oilfield starts from the measured formation pressure, oil, gas and water production, and fluid PVT basic data of the oilfield. Taking the flow unit of the reservoir as the object, global sensitive parameters such as reserves and water parameters are obtained through material balance calculation, and the direction and range of historical fitting adjustment are clarified. On the basis of the three-dimensional three-phase fully implicit black oil model, calculate the sensitivity coefficients of wellbore pressure, production oil gas ratio, water content, or other objective functions to grid block permeability, porosity, skin factor, and relative permeability. Based on the adjoint system theory, introduce the Lagrangian operator, and establish an adjoint model with the adjoint variable independent of the simulated calculation variable to avoid directly solving the gradient equation. A mathematical model for reservoir simulation history fitting was established based on Bayesian statistical theory, and the historical fitting problem was solved using gradient free optimization method and data assimilation method, respectively, to solve the automatic historical fitting problem of ultra-high water cut oil reservoirs. The application of this technology in the Ng63+4 sand layer group in Block 7 of Gudong Oilfield (Figure 4) resulted in consistent numerical simulation results with the shape and trend of the actual water content change curve of the block, achieving high fitting accuracy and increasing fitting efficiency by more than three times.


2.2.2 Integrated Research Process and Synergy Technology for Reservoir Development

With the deepening of reservoir development research, professional division of labor is becoming more and more detailed. However, from the perspective of research processes, offline serial processes are mainly used, and data and results are not standardized and cannot be shared in a timely manner. From the perspective of research methods, geological research, numerical simulation, dynamic analysis and other different links use a variety of professional software, which is inconvenient for researchers to call and lacks comprehensive visual analysis methods. As a result, the development and comprehensive research cycle for specific mining blocks is becoming longer, affecting work efficiency. In response to this issue, Liu Xianmei et al. proposed to study and build a networked collaborative work system for oilfield development in 2006. However, limited by the research methods at that time, only the computer online operation of process management links such as online scheme issuance, supervision and management, and data acceptance was achieved, without further research on comprehensive development. For many years, there have been few literature reports on the progress of China's oil and gas fields in this area. However, large foreign oil service companies have continued to carry out integrated and collaborative software platform research, among which Schlumberger's PetroRE is the most mature, achieving exploration and development process oriented and collaborative research based on the same software platform. However, this model is not suitable for the current situation of comprehensive research in China's oil and gas industry. One key reason is that China's exploration and development professional division and personnel division are more detailed, and the software mastered by researchers in various processes and links is not consistent. It is difficult to meet the requirements with only one set of PetroRE software.

Based on an in-depth analysis of the personnel, processes, and software status of China's oil and gas development research, Shengli Oilfield has established a comprehensive research standard process system for oil reservoirs, which includes eight primary businesses: stratigraphic comparison and division, structural research, reservoir research, reservoir characteristics research, reservoir model establishment, development effect evaluation, remaining oil analysis, and scheme optimization deployment. In response to the needs of data integration and result transformation for each process, functional modules such as data integration and processing, and result management have been designed. Based on cloud sharing mode, different types of professional software applied in each process have been virtualized and installed, achieving online operation of the entire development comprehensive research process and business collaboration among different positions. The process oriented and collaborative technology of comprehensive research on reservoir development has supported the formulation and optimization of plans for 15 new and old areas in Shengli Oilfield, improving the quality and efficiency of comprehensive research on development. For example, in the construction plan for the production capacity of the Fengshen Xie101 new area, researchers follow a customized research process, call platform functions and professional software to conduct comprehensive research, and the results data formed by various business functions are shared and dynamically updated in real-time, which improves work efficiency by more than twice compared to traditional offline work modes.

2.2.3 Intelligent Diagnosis Technology for Oil Well Working Conditions

Monitoring and diagnosing oil well operating conditions is of great significance for tapping into the potential of oil wells, predicting development risks, and optimizing development measures. In recent years, with the rapid development of oilfield industrial control systems, real-time collection of dynamic monitoring data for oil well production, such as indicator diagrams and electrical diagrams, has been achieved. Compared with other oil and gas collection data, indicator diagrams have become the first direction for the practical application of big data and artificial intelligence technology in the oil and gas industry, as they meet the data characteristics of the big data general industry, such as strong uniformity, strong certainty, low interference, and strong real-time performance. For example, Wang Xiang and others prepared a set of oil well condition diagnosis samples covering 5 major categories and 37 types of working conditions based on historical dynamic data and indicator diagrams. They designed a dedicated convolutional neural network (OWDNet) for model training and completed over 5 million working condition diagnoses on site, with an accuracy rate of 90%.

However, based on the current research results of intelligent diagnosis of working conditions, the diagnosis of working conditions mainly focuses on identifying potential problems in the wellbore or formation of oil wells based on indicator diagrams, such as pipe leakage, piston leakage, insufficient liquid supply, sand production, etc. There is relatively little research on diagnosis and prediction related to oil and gas development indicators, such as dynamic liquid level measurement, which is still mainly manual. Although indicator diagrams can be used to calculate liquid production, existing methods have problems such as large measurement errors and require manual intervention, and have not achieved true intelligent oil recovery. Therefore, based on the electric power diagram, which can reflect all the information from the surface to the wellbore of the pumping well, and with the characteristics of high accuracy in electrical parameter testing, accurate data, and convenient data collection, Shengli Oilfield has developed an intelligent calculation method for liquid production and dynamic liquid level based on the electric power diagram. The main idea is to achieve precise description of different node operating conditions of the pumping well through waveform separation and feature information extraction of the electric power diagram, and compare it with the standard diagram under normal working conditions, as a judgment basis for intelligent diagnosis of oil well operating conditions; Then, the effective stroke is calculated using wavelet signal separation through the electrical power diagram, and the real-time dynamic liquid level is calculated by the relative change in the work done by the lifting liquid in the upward work, achieving real-time monitoring of changes in the formation's liquid supply capacity. This algorithm was applied on site in Block 1 of the Gudao Oilfield, using real-time automatic production data collected from 11 oil wells in the block to calculate and analyze the liquid production and dynamic liquid level (Figure 5). The accuracy of the calculated liquid production by the electric power diagram reached 92.12%, and the error of the calculated dynamic liquid level by the electric power diagram was 6.69%, greatly improving the accuracy compared to the indicator diagram method.