Realistic Synthetic Benchmark Instances for Multi-Objective Optimization of District Heating Systems

This dataset provides 100 large-scale synthetic benchmark instances for lexicographic multi-objective optimization of district heating systems, based on the Berlin network. The instances capture both the network topology and the underlying MIP structures of the unit commitment problem, enabling detailed analysis of operational planning decisions. The model, the method used to generate the instances, and a computational study are described in the accompanying paper "Benchmarking Realistic Synthetic Instances Against a Large-Scale District Heating Network: A Multi-Objective Optimization Study for Berlin": https://arxiv.org/abs/2606.02195 The open-source code to generate the instances is available on GitLab: https://git.zib.de/abuchhol/multi-energy-instance-generator Optimization is performed lexicographically across three objectives: Minimize operational cost Minimize emissions (subject to the minimum cost solution) Maximize delivered heat (subject to the previous objectives) Each instance is available both as: JSON files (graph-based network representation with topology, parameters, and time series) MPS files (complete mixed-integer programming formulations for direct solver use, provided for each objective) Instance design The dataset varies key parameters across instances: time horizon T (10 or 25 years, 4-hour resolution) number of demand nodes and production sites number of converter units inclusion of storage number of fuel markets Instances are grouped into configurations (10 per group), with stochastic variations in time series and technical parameters. Instance Timesteps Demand Prod. Sites Converters Storage Fuel Markets uc_000-009 54750 3 5 20 0 2 uc_010-019 54750 1 5 20 0 2 uc_020-029 54750 3 3 10 0 2 uc_030-039 54750 3 5 20 0 2 uc_040-049 21900 3 5 20 0 4 uc_050-059 54750 3 5 20 1 2 uc_060-069 54750 5 5 20 1 2 uc_070-079 54750 3 5 20 1 4 uc_080-089 54750 1 5 20 1 6 uc_090-099 21900 3 5 20 1 2 Complexity: The number of variables range from 5.3 million to 15.4 million and the number of constraints ranges from 6.6 million to 18.8 million, with 19.3 million to 64.4 million non-zeros.