Detecting Patterns in Energy Use and Greenhouse Gas Emissions of Cities Using Machine Learning
Cities are expected to play a major role in managing climate change in the coming decades. The actual environmental performance of urban centres is difficult to predict due to the complex interplay of technologies and infrastructure with social, economic, and political factors. Machine learning (ML)...
Guardado en:
Autores principales: | , , , , , |
---|---|
Formato: | article |
Lenguaje: | EN |
Publicado: |
AIDIC Servizi S.r.l.
2021
|
Materias: | |
Acceso en línea: | https://doaj.org/article/ffabfbf8f0bc43bb85f16eb1d471f469 |
Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
Sumario: | Cities are expected to play a major role in managing climate change in the coming decades. The actual environmental performance of urban centres is difficult to predict due to the complex interplay of technologies and infrastructure with social, economic, and political factors. Machine learning (ML) techniques can be used to detect patterns in high-level city data to determine factors that influence favourable climate performance. In this work, rough set-based ML (RSML) is used to identify such patterns in the Sustainable Cities Index (SCI), which ranks 100 of the world’s major urban centres based on three broad criteria that cover social, environmental, and economic dimensions. These main criteria are further broken down into 18 detailed criteria that are used to calculate the aggregate SCI scores of the listed cities. Two of the environmental criteria measure energy intensity and greenhouse gas (GHG) emissions. RSML is used to generate interpretable rule-based (if/then) models that predict energy utilisation and GHG emissions performance of cities based on the other criteria in the database. Attribute reduction techniques are used to identify a set of 7 non-redundant criteria for energy use and 9 non-redundant criteria for GHG emissions; 6 criteria are common to these two sets. Then, RSML is used to generate rule-based models. A 10-rule model is determined for energy intensity, while an 11-rule model is found for GHG emissions. Both models were reduced further by eliminating rules with weak generalisation capability. A key insight from the rule-based models is that social, environmental, and economic attributes are associated with energy intensity and GHG emissions due to indirect effects. |
---|