Previsione di Serie Storiche ad Alta Frequenza, Applicazione ai Consumi Energetici del Politecnico di Torino

Energy management is nowadays dramatically important for the operation of modern industrial and service economy. Firms must found out new technologies and new solutions to avoid the problems deriving from the growth of energetic consumptions and from the relevant environmental emergency due to greenhouse. Energy efficiency and green energy are easy and advantageous ways to improve the competitiveness of our businesses, reduce energy costs for consumers and combat climate change. We can save energy if we know where, when and how much of it is used, this is possible thanks to monitoring technologies. The "Applus Energie" project is connected with these trends. Among various objectives, it aims at monitoring the electricity consumption of the university building of a branch of Turin Polytechnic University in Verrès (Aosta Valley Region). The purpose of such monitoring is to determine the consumption behaviour of the building throughout the day in order to optimize consumptions and costs. In this thesis work, I analyze and discuss daily time series from this project with a statistical approach. The aim is twofold. On one side I extract qualitative information on the consumption trends and on daily and weekly seasonalities also considering the electricity costs. On the other side I construct a model which will be used to forecast future values on the basis of past information. To this end I compare two approaches. The first consists in a single sequential model in which consumptions are considered at the high frequency of one observation per hour. The second consists in considering parallel 24 time series, one for each hour, and independently analyze them. In both cases, the analysis is carried out with ARIMA-GARCH models given regressors. The methods are evaluated in terms of one week ahead prediction performance on a test set.