ReservoirFlow

ReservoirFlow redefines the way we do reservoir modeling and engineering by integrating modern scientific computing with the power of artificial intelligence (AI) to provide next-generation tools for different applications. By combining scientific machine learning with agentic AI models, we refer to our approach as scientific-agentic AI. Moreover, we focus on scientific machine learning based on physics-informed neural networks (PINNs) and its advanced evolutions to ensure that our models are consistent with the underlying physical laws. This allows us to develop AI models that can generalize well even with limited data, making them suitable for real-world applications where data may be scarce or expensive to obtain.

Our main mission is to develop AI agents that can understand the underlying physics to provide quick solutions even by chatting or talking to them (e.g. Hi ReservoirFlow! based on latest production profiles, give me a forecast for the next 12 months!). This is a step forward in making reservoir simulation and engineering more accessible and user-friendly, especially for non-experts and managers who are either not familiar with the technical details or do not have the time to go through complex simulation models.

ReservoirFlow is also built to be common platform for both AI specialists and domain engineers (e.g. reservoir engineers). Our tools are designed to be flexible, modular, and extensible to support a wide range of problems related to subsurface energy storage systems including: (1) predicting rock properties based on seismic data and sparse well logs, (2) predicting reservoir performance based on available field data, and (3) optimizing production and drilling strategies to empower efficient real-time reservoir management. We are designing our tools to be accessible in different environments including local machines, on-premise servers, and Cloud platforms.

To empower applications in local machines, ReservoirFlow is also provided as a modern open-source Python library developed by Zakariya Abugrin at Hiesab. ReservoirFlow allows to study and model the process of fluid flow in porous media related to subsurface energy storage systems, reservoir simulation and engineering. ReservoirFlow is the first reservoir simulator based on physics-informed neural network models and one of its kind in a sense that it allows comparing and combining analytical solutions, numerical solutions, and neurical solutions (i.e. solutions based on artificial neural networks). We are also developing a cloud platform that will allow users to benefit from our services and upload or link data from various sources.

ReservoirFlow is designed based on the modern Python stack for data science, scientific computing, machine learning, and deep learning with the objective to support high-performance computing including multithreading, parallelism, GPU, and TPU. Throughout our computing problems based on large simulation models, intensive benchmarking well be carefully designed and carried out to evaluate the performance of computing software (i.e. frameworks) and hardware (e.g. GPUs). The outcome of these tests will be used to further improve the performance of ReservoirFlow and to provide materials with recommendations about available computing tools, techniques and frameworks. ReservoirFlow is planned to support different backends including NumPy, SciPy, JAX, PyTorch, TensorFlow, and more.

ReservoirFlow brings reservoir simulation and engineering to the Python ecosystem to empower automation in intelligent fields where engineers and specialists can deploy their models in containers that will be ready to make real-time optimization for any well in the field. In contrast to commercial black-box software where reservoir simulation studies are relatively isolated, important actions can be immediately predicted and made available for the field hardware to execute. A special attention well be given to provide solutions for environmentally friendly projects with a clear objective to reduce emissions. We are committed to extend our tools to cover the topic of Carbon Capture and Storage (CCS) especially $C{O}_2$ Underground Storage. In addition, we are looking forward to covering a wider range of topics from Reservoir Engineering including: Pressure Transient Analysis (PTA) and Rate Transient Analysis (RTA), Enhanced Oil Recovery (EOR), Improved Oil Recovery (IOR), Pressure-Volume-Temperature (PVT), Equation-of-State (EOS), etc.

An open-source reservoir simulation and engineering library within the Python ecosystem is also very important to students, universities, researchers, engineers, and practitioners. Unlike the common monopolistic approach in the Oil and Gas industry where software is usually offered as a closed black-box at a high cost, we plan to make our tools accessible and freely available to everyone except for commercial-use where an explicit authorization will be required. We aim to offer our sponsors the commercial-use license with other benefits including trainings, custom features, studies, and more. On the other hand, our license allows universities, students, academics, and researchers to use our tools directly for teaching or publication just with a proper referencing. Therefore, the growth of this tool can only be taken as a positive growth for a new community that we try to create. However, this requires a huge support to meet the upcoming challenges that we are looking for, see Support Us.