Data Driven Framework for Load Profile Generation in Medium Voltage Networks via Transfer Learning

This paper presents a framework to create daily active power load profiles adapted to the social demographic characteristics of the areas serviced by medium to low voltage (MV/LV) distribution transformers, to reduce the number of measurement devices to be installed in the distribution grid. The core concept is the use of transfer learning with a domain adaptation approach, which uses actual MV load consumption of the transformers where the data is available to transfer load patterns from one transformer to another. The framework has three main steps: clustering historical load profiles by consumption types; training a supervised classification model which relates consumption type and social demographic attributes; implementing the transfer learning method to generate a daily profile. The framework shows positive transfer learning between transformers, creating load profiles that correspond with activities in the servicing areas. The implementation is demonstrated with real data from two municipalities in the Netherlands.