In a recent conversation with The Guardian’s Ifeanyi Ibeh, petroleum engineer and researcher Emmanuel Obasi shared insights on how Physics-Informed Neural Networks (PINNs) are transforming the shale oil and gas industry, particularly highlighting their applications in reservoir characterisation, hydraulic fracturing, CO₂ storage, and production forecasting. Obasi emphasises the transformative potential of PINNs and advocates strongly for increased interdisciplinary collaboration among petroleum engineering, geoscience, and artificial intelligence communities. Such collaborative efforts, he argues, are essential to maximising the benefits of PINNs across the petroleum engineering sector, particularly within Nigeria’s evolving energy landscape, ultimately contributing to sustainable development and economic prosperity.

Can you briefly explain what Physics-Informed Neural Networks (PINNs) are and their significance in shale development?

Certainly. PINNs integrate fundamental physical laws directly into artificial intelligence and deep learning models. Their significance lies in bridging the gap between purely physics-based and purely data-driven approaches. PINNs enhance accuracy, efficiency, and physical realism, making them highly valuable in complex subsurface environments like shale reservoirs. These networks essentially combine the strengths of traditional simulations with the adaptive capabilities of machine learning, resulting in superior performance in solving complex engineering problems.

What specific advantages do PINNs offer compared to traditional modelling methods?

Traditional physics-based simulations, while accurate, tend to be computationally expensive and data-intensive, often requiring substantial computational resources and prolonged simulation times. On the other hand, purely data-driven models, such as conventional machine learning algorithms, lack physical realism and can struggle significantly when generalising predictions beyond observed data. PINNs overcome these limitations by embedding physical laws directly into neural networks. This approach drastically improves predictive reliability and computational efficiency, enabling accurate modelling even with limited data, which is particularly beneficial in sparsely explored or data-poor regions.

Can you provide a detailed example of how PINNs have improved reservoir characterisation?

Absolutely. In reservoir characterisation, PINNs have been highly effective in estimating key properties such as permeability, porosity, and fluid distribution. Seminal studies, notably by Tartakovsky et al. (2020), demonstrated that PINNs could accurately infer these complex subsurface characteristics with significantly fewer data points compared to conventional inversion methods. For instance, PINNs have successfully been used to estimate heterogeneous permeability fields and relative permeability curves by embedding the governing flow equations directly into their training processes. This significantly accelerates the characterisation workflow while ensuring physical fidelity and reliability of the results.

How have PINNs influenced hydraulic fracturing practices?

PINNs have revolutionised hydraulic fracturing by providing real-time predictive capabilities that outperform traditional numerical simulators. These networks allow for the rapid optimisation of fracture designs, accurate prediction of proppant distribution, and robust uncertainty quantification. Such capabilities directly lead to improved operational outcomes, substantially reduced costs, and optimised resource utilisation. For example, PINNs can accurately predict fracture propagation paths and proppant transport dynamics in real-time, significantly enhancing the efficiency and effectiveness of fracturing operations.

Given Nigeria’s growing interest in sustainable energy practices, how can PINNs support CO₂ storage initiatives?

PINNs are particularly suited for modelling CO₂ injection and storage in depleted shale reservoirs, an essential component of Carbon Capture, Utilisation, and Storage (CCUS) projects. They can accurately predict complex behaviours such as CO₂ migration, trapping, and potential leakage scenarios rapidly, enabling effective feasibility assessments and reliable long-term storage projections. These predictive capabilities position PINNs as strategic tools in Nigeria’s energy transition efforts, helping mitigate environmental risks while enhancing the viability of CCUS initiatives.

You’ve mentioned production forecasting as a critical area impacted by PINNs. Could you elaborate further?

Production forecasting in shale reservoirs faces unique challenges due to their rapid production declines and complex flow regimes, including interactions between fractures and reservoir boundaries. PINNs significantly enhance the reliability of these forecasts by integrating advanced physical knowledge into predictive models. Compared to traditional empirical methods, like decline curve analysis, PINNs provide markedly more accurate and generalizable forecasts. By accounting for the physics of reservoir depletion and well interference, these networks deliver crucial insights for effective resource management, investment decisions, and financial planning in shale operations.

Finally, what future developments do you foresee for PINNs in the shale industry, particularly in Nigeria?

The future of PINNs is exceptionally promising, with ongoing research efforts focused on enhancing computational efficiency, refining uncertainty quantification techniques, and integrating more sophisticated neural network architectures, such as advanced neural operators and deep operator networks. For Nigeria, adapting these cutting-edge methods could significantly accelerate sustainable shale resource development, supporting national energy independence, economic diversification, and overall growth. Additionally, PINNs could facilitate more comprehensive assessments and planning strategies, optimising reservoir management practices tailored specifically for Nigerian shale resources.