A Data-Mining Approach for Energy Behavioural Analysis to Ease Predictive Modelling for the Smart City

Analyzing urban districts to promote energy efficiency and smart cities control could be very complex as big data have to be analyzed filtered to discover unpredictable patterns. Using clustering method, specifically K-means algorithm, allows to create an energy profiling characterization of urban district models with multiple advantages: as first, large quantity of data can be managed and synthesized, easing the creation of algorithm patterns that could be replicable. Thus, it is possible to operate in a large scale and in a small scale in the same time, choosing the level of detail that is more appropriate for the specific analysis. In the large scale, the disadvantages are the dependence from data, i.e. if there are missing input values, it is hard to rebuild them because of the quantity of data. Missing values can confuse the analysis because scripts cannot identify the entire row missing it. Working with clustering analysis it is thus useful when large amount of data should be organized and interpreted and the technique can help the planner to make faster the analyses process. The research aims at demonstrate the efficiency of clustering methods when adopted for energy consumption issues at city level. In the paper, the clustering process concerning building energy profiles of a European city for the identification of building models is described. This means that an energy template on urban scale is used and clusters are applied on energy profiles based on architectural and energy similarity in order to find representative models. In particular, the study is focused on the relationships between building characteristics and actual building energy profiles.