Definition of hosting capacity maps in MV distribution networks

(MV) distribution networks by combining probabilistic power-flow assessment, GIS-based territorial characterisation, and machine-learning (ML) techniques. A Monte Carlo procedure is first adopted to compute, for each primary substation (PS) in an MV network, a probabilistic indicator of hosting capacity that accounts for uncertainty in operating conditions under predefined technical limits and risk criteria. The same networks are then characterised through a compact set of electrical descriptors and territorial descriptors extracted from GIS layers (e.g., land use/cover and built environment). These features are used to train an ML model that finds a relationship between territorial features and hosting-capacity risk, enabling large-scale screening and the rapid generation of preliminary hosting-capacity maps. The approach is validated on 110 MV real Italian distribution networks. Preliminary results show that hosting capacity is correlated with territorial features and can be predicted over time with engineeringgrade accuracy using a limited set of GIS-derived descriptors, supporting scalable DSO-level screening and prioritisation of detailed studies.