Methodology for a global sensitivity analysis with machine learning on an energy system planning model in the context of thermal networks

Thermal networks have gained attention in recent research as a means for the European Union to reach its climate targets. The potential has been well established for a multitude of regions. However, there is more to learn regarding the quantity and behaviour of the decision boundary between the use of thermal networks and competing alternatives. This study aims to describe that decision boundary with a parameter analysis of an energy system planning model. First, parameters of the energy system planning model (e.g. the linear heat density) describe a situation. Second, an energy system planning model finds the optimal (economic) energy supply mix for a given situation. Third, a sampling method with machine learning manipulates the values of the parameters to generate new situations, which improves the information on the as of yet unknown decision boundary. And fourth, the final location of the decision boundary is then predicted with machine learning from the optimised set of situations. Illustration of the concept is provided by means of a simplified case study. For that case study it is found that the decision for district heating is bound by a minimum value of several parameters instead of a single parameter. [Display omitted] •An alternative to classic scenarios to support decisions regarding district heating.•A novel global sensitivity analysis on an energy system planning model.•Machine learning to form a decision boundary between centralised and local heating.•Multiple parameters; the decision boundary is not solely characterized by the linear heat density.