A carregar...
Publicação
Clustering-Based Identification of Energy Retrofitting Strategies in Italian Buildings: Integrating Weather Data, Spatial Data and Autoencoder-Enhanced Models
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 4.57 MB | Adobe PDF |
Autores
Orientador(es)
Resumo(s)
In Europe, there is still a critical problem being faced regarding building energy inefficiency.
Over the last years people have been more alert about the need for retrofitting as a solution
to reduce energy consumption (EC) and support sustainability goals. The focus of this work is
to understand how the integration of Geographic Information Systems (GIS) and weather data
can help the identification of retrofitting strategies for buildings. Nonetheless, to support this
goal, it is studied the effectiveness of the usage of autoencoders to detect anomalies within
building clusters. The focus of this study was the region of Lombardy, Italy and three datasets
were used. Energy Performance Certificates (EPC) for the buildings with yearly data, a weather
dataset containing daily details on weather conditions for the region and geospatial data.
Different cluster techniques were integrated such as K-means and Self Organizing Maps (SOM)
comparing both with and without weather data results. Then autoencoders were used to
address the city-level problem presented by the weather features. The results showed that
there was a temporal mismatch between the weather data and the EPC data which introduced
biases into the results. The autoencoders showed a great capacity to mitigate this problem
leading then to more interpretable and balanced clusters. The GIS visualizations enabled clear
interpretations of the results, primarily of possible retrofit measures and the return on
investment (ROI) associated. The conclusions aligned the initial hypothesis of the importance
of the use of spatial data to support energy policy and retrofit decision-making. This work
highlights the role of cluster and anomaly detection in the optimal identification of retrofitting
strategies for buildings and suggest that the temporal alignment between datasets it is
mandatory to unlock the maximum predictive capability of the models.
Descrição
Dissertation presented as the partial requirement for obtaining a Master's degree in Data Science and Advanced Analytics, specialization in Data Science
Palavras-chave
Energy Retrofit Energy Performance Artificial Intelligence Machine Learning Autoencoders GIS Spatial Factors Certification Buildings SDG 7 - Affordable and clean energy SDG 9 - Industry, innovation and infrastructure SDG 11 - Sustainable cities and communities SDG 13 - Climate action