A Machine Learning Based Methodology for Load Profiles Clustering and Non-Residential Buildings Benchmarking

Buildings benchmarking based on their electric profiles is a fundamental step to identify, evaluate and then possibly implement energy efficiency oriented actions. Indeed, benchmarking enables comparison among peer buildings or industrial sites and the identification of reference cases, either efficient and inefficient ones. In this regard, temporal data clustering is an effective and widely applicable benchmarking tool. In this work, we propose a novel Machine Learning based methodology, taking advantage of two fundamental tools, namely a decomposition algorithm and a clustering one. Several clustering algorithms have been tested to identify k-Means as the most suitable one. The proposed methodology includes the evaluation of energyKeyPerformance Indicators for effective analysis and comparison of buildings. The proposed framework has been tested on a real-world case study including around 2000 non-residential buildings. The classification of buildings based on K-Means achieved an accuracy of 99.7% with respect to their usage category. Furthermore, reference Key Performance Indicator values for each cluster are obtained and discussed to understand buildings' energy behaviour and possible reasons for inefficiencies.

Automated summary

Benchmarking buildings based on their electric profiles is an important step in identifying, evaluating, and implementing energy efficiency actions. Temporal data clustering is an effective tool for this purpose, and we propose a novel machine learning methodology that combines decomposition and clustering algorithms. The proposed framework achieved high accuracy in classifying buildings based on their usage category and provides reference key performance indicator values for each cluster to understand energy behavior and possible inefficiencies.