A systematic approach to using artificial intelligence as surrogate and generative models in energy systems : applications in carbon capture and electricity markets

Abstract

Artificial intelligence (AI) is revolutionizing numerous fields, yet its application in energy systems has been slow. This dissertation introduces and validates a systematic approach to leveraging AI, focusing on the development of surrogate models to replace computationally intensive engineering simulations and the subsequent use of these surrogates in generative design and predictive analytics. The proposed framework yields fast, lightweight, and robust models that are not prone to convergence issues, enabling gradient-based optimization for more reliable design space exploration. This data-driven approach also reduces the need for extensive model development and calibration, while allowing for seamless adaptation to changing system conditions.

This approach is first demonstrated in post-combustion carbon capture (ACC), a critical but energy-intensive technology. A surrogate model was developed to predict the energy consumption of an ACC plant, and this model was integrated into a novel generative design framework that rapidly identifies optimal process configurations to minimize specific reboiler duty. The investigation produced a visual design guide for practitioners and open-sourced the models and data to facilitate further research.

The second application addresses European electricity markets. An open-access deep learning toolkit, Deepforkit, was created for large-scale, day-ahead price forecasting. A comprehensive analysis across 19 bidding zones challenged the conventional assumption that high price volatility equates to market unpredictability, demonstrating that advanced models can maintain high forecast accuracy even during major events like the recent global energy crisis.

Ultimately, this dissertation provides a validated, systematic framework for applying AI in energy systems. Through impactful and open-sourced contributions to carbon capture process design and electricity market analysis, it demonstrates how surrogate and generative models can lead to more efficient, adaptive, and economically viable energy solutions.