DHDrift: District heating concept drift benchmark

Zenodo (CERN European Organization for Nuclear Research)

Abstract

A synthetic district heating load dataset with documented ground truth drift occurrence timings for benchmarking concept drift detection and adaptive load forecasting algorithms. The dataset comprises 901 hourly heat load timeseries spanning four years (2020–2023), generated through dynamic building energy simulation of a 32-building residential district in Aachen, Germany. Nine concept drift scenarios representing realistic operational changes—network expansion, building refurbishment, and night setback control introduction—are provided at three severity levels, each with 100 randomized realizations to enable statistically robust algorithm comparison. Supplementary data including 256 individual building load profiles and building attributes support custom district configurations and scenario development.

Authors 3

1. K Derzsi Aachen

   RWTH Aachen University

   Affiliation as printed

   RWTH Aachen University

2. Jonas Klingebiel Aachen

   RWTH Aachen University

   Affiliation as printed

   RWTH Aachen University

3. Dirk Müller Aachen

   RWTH Aachen University

   Affiliation as printed

   RWTH Aachen University