Machine Learning-Accelerated Multi-Objective Design of Fractured Geothermal Systems

Guodong Chen, Jiu Jimmy Jiao, Qiqi Liu, Zhongzheng Wang, Yaochu JinNexus 2024 · DOI · OpenReview

Abstract

Multi-objective optimization has burgeoned as a potent methodology for informed decision-making in enhanced geothermal systems, aiming to concurrently maximize economic yield, ensure enduring geothermal energy provision, and curtail carbon emissions. However, addressing a multitude of design parameters inherent in computationally intensive physics-driven simulations constitutes a formidable impediment for geothermal design optimization, as well as across a broad range of scientific and engineering domains. Here we report an Active Learning enhanced Evolutionary Multi-objective Optimization (ALEMO) algorithm, integrated with hydrothermal simulations in fractured media, to enable efficient optimization of fractured geothermal systems using few model evaluations. We introduce a probabilistic neural network as a classifier to predict the Pareto dominance relationship between candidate samples and reference samples, thereby facilitating the identification of promising but uncertain offspring solutions. We then use an active learning strategy to conduct hypervolume-based attention subspace search with a surrogate model by iteratively infilling informative samples within local promising parameter subspaces. Results demonstrate that the ALEMO approach achieves a remarkable reduction in required simulations, with a speed-up of 1-2 orders of magnitude (10-100 times faster) than traditional evolutionary methods, thereby enabling accelerated decision-making.

Introduction

Enhanced Geothermal Systems (EGS) are a cornerstone of the transition to renewable energy, but designing them is an engineering challenge of immense complexity. Engineers must balance conflicting goals—such as maximizing short-term revenue versus ensuring long-term reservoir stability—while navigating a massive parameter space that includes well placement and injection rates.

Traditionally, finding the optimal set of designs (the Pareto front) requires thousands of high-fidelity hydrothermal simulations. Because each simulation can take hours, the process is often prohibitively slow. This "simulation bottleneck" has historically limited the ability of engineers to explore the full range of design possibilities in fractured geothermal systems.

The ALEMO Framework

To overcome these limitations, researchers developed ALEMO (Active Learning enhanced Evolutionary Multi-objective Optimization). This framework integrates machine learning surrogates with physics-based modeling to accelerate the discovery of optimal engineering trade-offs.

The ALEMO framework integrating machine learning model updates, evolutionary optimization, and hydrothermal simulations for geothermal design.

As shown in the framework overview, ALEMO consists of three sophisticated components designed to bypass the traditional bottleneck:

  1. Probabilistic Neural Network (PNN) Classifier: Instead of attempting to predict exact physical values, ALEMO uses a PNN to predict the relationship between designs. It classifies candidate solutions into categories (e.g., "promising" vs. "poor") based on whether they are likely to dominate current best designs.
  2. Surrogate Modeling: The framework uses Radial Basis Functions (RBF) to create a computationally inexpensive mathematical approximation of the complex reservoir physics.
  3. Hydrothermal Simulation: High-fidelity finite element methods are used selectively to validate the most promising designs identified by the ML models.

Optimization via Active Learning

The core innovation of ALEMO is its active learning strategy. Rather than uniform sampling, the algorithm identifies "attention subspaces"—local regions in the parameter space that are most likely to contain optimal designs.

The active learning loop in ALEMO, showing the interaction between the evolutionary generator, ML discriminator, and hydrothermal simulator to refine the Pareto front.

By focusing on these subspaces and using hypervolume-based attention, ALEMO strategically chooses new samples to simulate only where the model is most uncertain or where the potential for improvement is highest. This allows the framework to handle high-dimensional problems with over 100 variables—a task that typically defeats standard optimization methods due to the "curse of dimensionality."

Key Results and Impact

The ALEMO framework was rigorously tested across mathematical benchmarks and large-scale, field-scale EGS models featuring over 1,200 fractures. The results were transformative:

  • Massive Speedup: ALEMO achieved a 10x to 100x speedup compared to industry-standard algorithms like NSGA-II and MOEA/D.
  • Efficiency: While traditional methods struggled to converge after 5,000 simulations, ALEMO found superior, diverse Pareto-optimal designs in just 300 to 500 evaluations.
  • Economic Performance: In field-scale tests, ALEMO-designed control schemes maintained higher thermal production rates and net present value (NPV) throughout a 6,000-day project lifespan.

Future Directions

While ALEMO represents a major leap forward, future work will focus on incorporating geo-mechanical effects and managing seismic risks. The ultimate goal is to integrate ALEMO with real-time sensor data to create "digital twins" of geothermal plants for autonomous, real-time management, further securing geothermal energy's role in a sustainable future.

Summary updated

Abstract

Multi-objective optimization has burgeoned as a potent methodology for informed decision-making in enhanced geothermal systems, aiming to concurrently maximize economic yield, ensure enduring geothermal energy provision, and curtail carbon emissions. However, addressing a multitude of design parameters inherent in computationally intensive physics-driven simulations constitutes a formidable impediment for geothermal design optimization, as well as across a broad range of scientific and engineering domains. Here we report an Active Learning enhanced Evolutionary Multi-objective Optimization algorithm, integrated with hydrothermal simulations in fractured media, to enable efficient optimization of fractured geothermal systems using few model evaluations. We introduce probabilistic neural network as classifier to learns to predict the Pareto dominance relationship between candidate samples and reference samples, thereby facilitating the identification of promising but uncertain offspring solutions. We then use active learning strategy to conduct hypervolume based attention subspace search with surrogate model by iteratively infilling informative samples within local promising parameter subspace. We demonstrate its effectiveness by conducting extensive experimental tests of the integrated framework, including multi-objective benchmark functions, a fractured geothermal model and a large-scale enhanced geothermal system. Results demonstrate that the ALEMO approach achieves a remarkable reduction in required simulations, with a speed-up of 1-2 orders of magnitude (10-100 times faster) than traditional evolutionary methods, thereby enabling accelerated decision-making. Our method is poised to advance the state-of-the-art of renewable geothermal energy system and enable widespread application to accelerate the discovery of optimal designs for complex systems.
BibTeX
@inproceedings{chen2024machine,
  author = {Guodong Chen and Jiu Jimmy Jiao and Qiqi Liu and Zhongzheng Wang and Yaochu Jin},
  title = {{Machine Learning-Accelerated Multi-Objective Design of Fractured Geothermal Systems}},
  booktitle = {Nexus},
  year = {2024},
  doi = {10.1016/j.ynexs.2024.100044},
}
Figure