Reduced-Order Modeling of Thermal Dynamics in District Energy Networks using Spectral Clustering

TL;DR

This paper introduces a graph theory-based reduced-order modeling method for thermal dynamics in large district energy networks, enabling efficient simulations with minimal accuracy loss.

Contribution

A novel spectral clustering approach for model order reduction that handles varying flow conditions and improves interpretability over existing methods.

Findings

Achieved 2.3% relative RMS error in temperature predictions.

Reduced computational complexity for city-scale energy grid simulations.

Applicable to dynamic flow scenarios with multiple producers.

Abstract

Simulation of thermal dynamics in city-scale district energy grids often becomes computationally prohibitive for long simulation runs. Current model order reduction methods offer limited interpretability with regards to the non-reduced system, and are not in general applicable for e.g., varying flow rates, multiple producers, or changing flow directions. This article presents a novel method based on graph theory that approximates the solution of an optimization problem that minimizes the local truncation error for heat transport in the grid. It is shown that the method can be used to reduce the thermal dynamic model of a city-scale energy grid, resulting in a coarser temporal and spatial resolution. The relative root mean square error was 2.3\% for the temperature in the evaluation scenario, comparing the reduced-order system with the non-reduced system at the instances of the coarser time-step.