Load Profiles Clustering and Knowledge Extraction to Assess Actual Usage of Telecommunication Sites

Deep awareness of a particular industry sector represents a fundamental starting point for its energy efficiency enhancement. In this perspective, a huge amount of industrial facilities' energy measurements are collected thanks to the widespread usage of monitoring systems and Internet-of-Things infrastructures. In this context, data mining techniques allows an effective exploitation of data for knowledge extraction to automatically analyse such enormous amount of data. This paper investigates a large data set including real telecommunication sites' aggregate electrical demand provided by the largest telecommunication service provider in Italy. The goal is the assessment of the actual usage category of telecommunication sites, aiming at supporting the facility management of the company and the energy knowledge discovery of each site category. A novel methodology is proposed that includes i) a proper normalisation method focused on energy Key Performance Indicators for telecommunication network energy management, ii) a time series decomposition tool to extract trends and periodical fluctuation of telecommunication sites' aggregated electric demand, and iii) the application of a k-Means clustering algorithm to assess sites' actual usage. The proposed methodology results in accurate outcomes, which witness the potential for practical application and discloses opportunities for further developments.