Generative AI for Field Development Planning
Reservoir engineers in the oil and gas industry are tasked with optimizing field development plans through reservoir simulators like Intersect™ high-resolution reservoir simulator and Eclipse™ industry-reference reservoir simulator. This task requires them to modifying numerous parameters, whether for history matching or optimizing various strategies such as infill drilling, waterflooding, and artificial lift. The syntax-based nature of these models necessitates deep expertise, often involving extensive use of user manuals. Additionally, engineers must generate many different field management scenario variants to maximize reservoir recovery. The comprehensive analysis required for these scenarios demands a significant amount of time and specialized knowledge, which is a bottleneck that hampers productivity and delays decision-making. This process is time-consuming and prone to human error, limiting the ability to explore all potential scenarios effectively, especially for large-scale models with numerous wells. Consequently, engineers often settle for suboptimal development plans which has direct impact on production efficiency and profitability.
Operators require a system that simplifies the modification and generation of new reservoir simulation model scenarios and provides insights to optimize field development plans. Such a system should automatically generate simulation syntax, adjust parameters based on insights, and provide comparative analytics to highlight improvements or gaps in the newly suggested development plans. They seek to reduce the iterative, manual burdens of simulation model adjustments, and increase the efficiency and efficacy of their planning processes.
The solution
Our approach leverages the advanced capabilities of Generative AI, particularly large language models (LLMs) coupled with an agent approach to carrying out domain-specific tasks. Agents are autonomous objective-driven systems built around a LLM to handle specific tasks. They use tools and planning modules to break down complex problems and access information they need. An internal memory allows them to keep track of past interactions and data used in solving problems. LLMs integrated into powerful workflows and leveraging external tools are adept at processing extensive datasets, and converting complex data into manageable, natural language outputs with insights and development options. This capability is pivotal for bridging the gap between technical reservoir simulation and development planning analysis and practical implementation by engineers.
The SLB solution, developed at our INNOVATION FACTORI. seamlessly integrates multi-agent systems, including AI agents which target specific aspects of field development planning.
Simulation syntax generation is designed to automate the generation of simulation syntax to streamline model adjustments and generation of new simulation models. Its core component, leveraging the Retrieval-Augmented Generation (RAG) framework, utilizes a vector database of syntax sourced from simulation guidelines and manuals. By employing semantic search, the LLM efficiently retrieves relevant context, aiding in comprehension and code generation. It's crucial to ensure these resources are comprehensive, covering a wide range of syntax rules and cases, to ensure accurate code generation. Another method explored involves fine-tuning the LLM. This process involves creating a diverse dataset of examples to refine the LLM's proficiency in adhering to simulation syntax. However, it's important to note the necessity of a substantial dataset for effective fine-tuning before implementing this approach and we typically employ a combination of these methods to achieve desired performance.
Reservoir model analysis and insights generation provides in-depth assessments of well and reservoir performance, offering potential optimization strategies. This module uses various agents, including advanced analytics and domain-specific, which can leverage external functions closely linked to simulation model data for analysis and actionable insights. These agents use various methodologies, including a chain of thought approach, to comprehensively address user queries. Integration with domain functions and tools is crucial as the agent's efficiency relies on this specialized knowledge. Without access to these functions, agents would struggle to navigate various reservoir analysis and simulation workflows. These domain functions, models and tools have been refined over years to cater to unique reservoir and petroleum engineering challenges. Guiding agents with relevant context and the right tools is imperative to avoid inaccuracies and hallucinations that may result from the lack of domain expertise.
Operational Guidelines Compliance ensures all scenarios align with strict reservoir management guidelines and standards. This module incorporates an advanced RAG agent, leveraging a fusion of simulation documentation and manuals, reservoir modeling and management guidelines, and the reservoir simulation model. Through its connectivity with a vector database created from comprehensive simulation manuals, the agent adeptly comprehends the intricacies of the user's reservoir simulation model and reservoir modeling and management guidelines. Subsequently, the agent meticulously evaluates the model against predefined guidelines and performs any necessary modifications based on the user query. A crucial facet of this approach entails conducting a thorough evaluation of the entire simulation model, as assessing only a portion of the model would yield an incomplete analysis.
Our workflow is further improved by the active involvement of reservoir engineers, who rigorously test and guide the agents, ensuring adherence to domain principles informed by their extensive industry experience. By combining facets of field development planning (FDP) and interfacing with existing digital capabilities, the multi-agent system furnishes reservoir engineers with a comprehensive application, offering indispensable support throughout the workflow life cycle and facilitating the resolution of intricate challenges.
These modules or agents are integrated seamlessly with existing SLB simulation tools, such as Intersect™ high-resolution reservoir simulator, thus enhancing their functionality and reducing the time engineers spend on manual configurations.
The result
One notable use case asked the system to analyze the performance of new infill wells in a forecast reservoir simulation scenario. It is a crucial exercise that ensures new planned infill wells are economically viable and drilled in optimal locations. The solution successfully identified the wells with the lowest cumulative oil production and categorized them based on their water cut and gas-oil ratio behavior, and also compared production profiles to the historical performance of existing wells drilled in the same region providing a potential reason on the underperformance of these wells. This level of detailed analysis was quite unexpected and was carried out in record time. Such comprehensive insights enable a deeper understanding of reservoir performance and assist in crafting scenarios that are not only realistic but also optimized to maximize recovery from the reservoirs.
The deployment of Generative AI in field development planning has led to improvements in the following areas:
→ Time savings: We have seen a significant reduction in the time required for simulation setup, modification, and scenario analyses.
→ Insights generation: Evaluation of complex FDP scenarios, by providing quick, tailored insights to specific queries related well performance, waterflood, gas injection, and gas lift performance.
→ Development plan optimization: Development plans with increased recovery, with substantial time reduction and decision quality compared to traditional methods.
For our customers, the implementation of Generative AI translates into tangible benefits. Engineers are now able to focus more on high level decision-making rather than tedious manual setup, leading to more optimized field development plans and more exhaustive exploration of possible options. This shift enhances operational efficiency, boosts profitability, and ensures that projects meet both technical and economic viability, within a shorter timeframe. Additionally, our solution's ability to quickly adapt to new data and changing conditions means that customers can maintain a competitive edge in a rapidly evolving industry.
Conclusion
The integration of Generative AI into reservoir simulation and field development planning signifies a major technological advancement for the energy industry. By automating complex processes and providing rapid, accurate insights we are creating a new standard for efficiency and precision in reservoir management. This initiative is just the beginning, as we continue to refine and integrate AI capabilities to address ongoing and future challenges of the industry, paving the way for a more sustainable and economically viable future in energy development.
