Using Unsupervised Learning to Determine Geospatial Clusters in Municipalities to Improve Energy Measurements

摘要

This paper presents a tool used to solve the Geospatial capacitated clustering problem applied to an energy company scenario. The billing process of an energy distributor in Brazil is connected to the spatially-aware logistics of collecting energy consumption data. Usually, consumer units are grouped into geospatial clusters that will be covered by meter readers. The process of creating those groups, in general, is carried out manually by analysts, which is an exhaustive process and prone to mistakes. In order to automatize this issue, this work presents a system that automatically generates reading groups for the collection of electrical energy consumption. The approach used to solve the capacitated clustering problem was based on a recursive K-Means. The results obtained with the proposed tool are promising.