Blizard & Stockar: District heating partitions show 2.8% cost penalty

A new arXiv paper by Audrey Blizard and Stephanie Stockar explores the robustness of performance-based partitioning in district heating networks. The authors leverage a physics-based distributed model predictive control framework and a learning-enhanced branch and bound method to systematically cull the search space for optimal partitions. Their study evaluates how sensitive a nominally optimal partition is to 12 different parameter perturbations, including supply temperature, operating season, building flexibility, pipe characteristics, and building type. The results show that a well-designed nominal partition incurs an average cost increase of only 2.8% relative to a fully centralized control system across 11 of the 12 cases. Moreover, in three of those cases the nominal partition was actually globally optimal under the perturbed conditions, demonstrating surprising robustness.

However, the analysis of the Optimality Loss Metric (OLM) reveals important caveats. In five of the twelve cases, the case-specific OLM-minimizing partitions actually underperformed the nominally optimal one, due to shifts in the relative magnitude of heat loss versus flexibility costs. This indicates that proper tuning of cost function weights and initial conditions is essential for reliable partition selection. The November operating case, in particular, showed that seasonal repartitioning becomes necessary when demand profiles deviate substantially from the nominal design. The work provides practical guidance for engineers designing distributed control systems for district heating, balancing computational tractability with control performance.