Integrating Machine Learning in District Heating Networks

The unit commitment (UC) problem consists in scheduling power generators in order to minimize their operating cost. Power generators may include thermal generation units, as well as renewable energy generation units, the latter introducing uncertainty regarding their production level when solving UC problems.
The UC problem must be solved several times a day in short time frames. Therefore, machine learning (ML) models can be used at different stages to speed up computation time, e.g. by finding first approximate solutions, or to alleviate errors due to uncertainty, e.g. by
forecasting the network load.
Parallels can be drawn from UC with district heating networks (DHN) problems. In a DHN, an operator seeks to supply customers with heated water. To heat the water, thermal units are employed, such as gas boilers or heat pumps (HP). In combined UC and DHN, combined
heat and power (CHP) units are considered.
To reduce greenhouse gas (GHG) emission of the energy sector, UC/DHN models can consider emission constraints, and multi-objective formulations both minimizing exploitation cost and emissions are developed.
In this talk, we present the contribution of ML models to solve UC/DHN and UC/DHN+GHG problems.