Analytical Hybrid Model-PINN for Geothermal Heat Exchangers

The design and optimization of geothermal heat exchangers (BBEs) represent a major technical challenge in the renewable energy sector. Traditionally, numerical methods such as finite elements or finite differences have been the standard tool for simulating the thermal behaviour of the subsurface, but their high computational cost limits their application in large-scale projects or in real-time monitoring environments. In this context, models based on neural networks emerge as a promising alternative, although they usually face convergence problems when they have to handle singularities —such as point sources of heat— and heterogeneities of the terrain. An innovative approach that is gaining traction is the combination of analytic solutions with physics-informed neural networks (PINN), resulting in hybrid models that take advantage of the best of both worlds: the precision of physical equations and the flexibility of machine learning.

The hybrid analytical-PINN model proposed in recent studies directly addresses the difficulty of representing singular sources of heat—such as geothermal heat exchangers—in a heterogeneous medium. The key is to decompose the total thermal response into an ideal homogeneous part (which can be calculated analytically using source line models) and a correction that captures the influence of soil variability. In this way, the neural network does not have to learn the singularity from scratch, but is trained to predict only the correction, which drastically reduces the complexity of the problem. This scheme also allows the thermal conductivity of the terrain to be parameterized, which is usually the most uncertain and determining property in the design of geothermal drilling fields.

From a computational standpoint, the hybrid model offers significant advantages. By eliminating uniqueness through the use of analytical solutions, the need to refine the mesh in the vicinity of the well is avoided, speeding up simulations without sacrificing accuracy. The neural network—trained with a loss function that combines terms based on differential heat equations (physics-informed) and observational data (data-anchored)—learns a universal corrector for an individual well with a unit extraction rate. Then, thanks to the principle of overlap, this corrector can be reused to model multiple wells and different extraction rates, making the model an extremely efficient tool for parametric studies and sensitivity analysis.

The practical application of this approach transcends the academic field. Geothermal engineering companies can benefit from a drastic reduction in simulation times, going from hours or days to minutes, without sacrificing the accuracy required for certifications and feasibility studies. In addition, the ability to learn thermal conductivity from field data allows models to be updated in real-time as new measurements are obtained, facilitating decision-making during drilling or operation of borehole fields.

Behind the implementation of these hybrid models is the need for a robust and scalable software ecosystem. At Q2BSTUDIO we understand that innovation in clean energy goes hand in hand with information technology. For this reason, we offer tailor-made applications to integrate geothermal simulation models into business management platforms, allowing engineers and managers to access real-time results without the need to be programming experts. In addition, our AI solutions for enterprises allow for the incorporation of intelligent agents that continuously monitor sensor data in the field, automatically adjusting the parameters of the hybrid model to maintain accuracy in the face of unforeseen changes in subsurface conditions.

The technological infrastructure that supports these developments is equally critical. AWS and Azure cloud services provide the elastic compute capacity needed to train and run neural networks, while business intelligence solutions, such as Power BI, make it easy for non-technical stakeholders to visualize complex outcomes. At Q2BSTUDIO we integrate these services in a coherent way, designing systems ranging from IoT data acquisition in the field to the executive dashboard, including machine learning models deployed in managed containers with orchestrators such as Kubernetes.

Of course, any system that handles critical data from energy infrastructures must ensure a high level of protection. That's why we incorporate cybersecurity measures from the design phase, protecting both simulation data and communications between sensors and servers. Our expertise in pentesting and vulnerability assessment ensures that the solutions implemented meet the most demanding standards, especially when it comes to strategic assets such as geothermal heat exchangers.

The convergence between computational physics and machine learning is opening up new frontiers in energy engineering. The hybrid analytical-PINN model for BHE is a clear example of how the intelligent combination of classical and modern methods can solve problems that until now seemed intractable from a computational point of view. As the demand for efficient and renewable HVAC systems grows, having fast, accurate and adaptable simulation tools becomes a decisive competitive advantage.

In this scenario, collaboration between geothermal specialists, machine learning experts and software development companies such as Q2BSTUDIO is essential to transfer cutting-edge research to the real market. Our team is working on building modular and extensible platforms that allow energy companies to adopt these hybrid models without having to invest in complex infrastructures or highly specialized profiles. From the creation of web interfaces for the definition of scenarios to the integration with SCADA systems, including the optimization of parameters using AI agents, we offer comprehensive support.

The future of geothermal simulation lies in models that learn from the data and update themselves in real time, while maintaining physical rigor as a basis. The hybrid analytical-PINN approach represents a firm step in that direction, and at Q2BSTUDIO we are committed to helping companies implement it efficiently and securely. If your organization is exploring the potential of geothermal energy or wants to modernize your simulation and monitoring processes, we invite you to learn about our custom software development solutions with built-in artificial intelligence capabilities.