| Summary: | Nowadays, huge quantities of metering data of each consumer are being taken from the electric distribution network through smart meters and stored in databases. These metering data are consumption readings useful for many data analysis applications as, for example, fraud detection. However, a large number of possible analysis on this data are still unexplored. In this dissertation, we explore smart meters as a way of improving Energy Efficiency. To do that, we need to understand more about the way clients consume and find behavioural patterns on their consumption. The identification of all these profiles of consumption is essential since it will allow EDP Distribuição (EDPD) to know more about its types of clients, providing focused feedback and consumption advice. Clustering algorithms are useful to understand the distribution of patterns in large data sets. By creating several groups/clusters, we will be able to understand the profile of a specific client by the characteristics of its cluster. It makes it possible to categorise a customer from its group behaviour rather than expecting that each customer as it own specific profile. This allows to save a lot of time analysing the types of consumers. In this dissertation, we intend to analyse clients from several perspectives in order to capture different types of behaviours. For example, we may want to analyse the clients based on their absolute consumption values, in order to compare their scales or compare them from the consumption regularity point of view. So, we use the clustering algorithms with the appropriate features as a data mining approach to find structure in our data. The results that we present highlight the strengths and weaknesses of each clustering algorithm and validate their applicability to the EDP Distribuição (EDPD) use case. So, this dissertation will bring added knowledge about clustering techniques and analysis over smart metering data.
|
|---|