Efficient Optimization of Inertial Energy Networks Using Physics-Informed Spatio-Temporal Model

Taha Boussaid, François Rousset, Marc CLAUSSE, Vasile-Marian Scuturici

Keywords: Spatio-temporal model, predictive control, sustainable energy systems, reinforcement learning

TL;DR: Enabling full dynamic optimization of inertial energy networks through a physics-informed spatio-temporal surrogate model while reducing the computational cost

Abstract: Mitigating climate change calls for a transition to more sustainable energy systems. One of the key levers is the efficient deployment of district energy networks which integrate diverse low-carbon sources. Inertial energy networks, characterized by their large-scale infrastructure, are an example of systems designed to provide heating and cooling to various consumers. However, the complexity of their dynamics such as multi-timescale responses, nonlinear behaviors and intermittence of renewable sources result in high computational cost for accurate simulation and optimization. To overcome this, we introduce a system-agnostic modeling framework that leverages the spatio-temporal structure of these systems to develop an efficient and scalable physics-informed state-space surrogate model. Coupled with model-based and learning-based optimizers, the framework enables full dynamic optimization and faster decision-making while drastically reducing computing time. Benchmarks against rule-based and physics-based strategies demonstrate that our approach achieves competitive energy cost savings while cutting the runtime from over a month to just few hours.

Format: Short paper or extended abstract, up to 4 pages.

Submission Number: 7