Using extreme value theory to take better account of peak demand when generating typical periods by clustering for district heating networks design optimisation

Reducing greenhouse gas emissions is a critical goal for many communities and countries worldwide, and fourth-generation district heating networks can be a key part of achieving this goal. However, optimising district heating networks has become more and more computationally demanding due to the increasing complexity of the systems through the last decades. One common solution identified by the community is to use time-clustering to reduce the amount of input data in the problem. However, these algorithms smooth out anomaly periods which result to be critical for an optimal choice of the district heating technology installation. This article introduces a new methodology based on an automatic process that creates representative periods and includes a detection of anomaly periods. This new approach is compared with a manual anomaly detection method, and then applied to a district heating sizing optimisation problem. Results show that the extreme value theory applied to a clustering method allows to speed up significantly the solving while limiting the duration of improper operation of the found configuration during a full year simulation.