Context

District heating systems (DHS) are an ideal way of supplying heat to homes, as they allow heat to be exchanged between producers and consumers, ensuring optimum and efficient use of energy. However, DHS present a major challenge as they are very difficult to model. DHS are thermohydraulic systems composed of many elements such as pipes, valves, reservoirs, heat exchangers and so on. This makes them complex systems as they depend on several non-linear equations based on flow rates and continuity equations. The aim of my project was to investigate some data-driven modelling alternatives that should be much easier to implement and more adaptable.

Approach

I tried two modelling approaches: a grey box approach, which is a mixture of data-driven and physics-based models, and a black box approach, where I only used a data-driven model to predict the heat load.

The results showed that simple linear models, such as linear regression, aren't complex enough to model the complex structure of a DHS. On the other hand, the models using extra tree regressors showed promising results. We also found that a grey-box approach performed better for short-term forecasting, while the black-box approach performed better for long-term forecasting. It was also found that the performance of the model changed depending on the mode of operation of the building.

Finally, the use of the model was demonstrated by applying it to a control system.