ETH Zurich : Computer Science : Pervasive Computing : Distributed Systems : Education : Student Projects : Abstract

Distributed Consumption Analysis (M)

Status: Abgeschlossen

Worldwide electricity consumption is rising and, for years to come, the generation will mainly consists of non-renewable and CO2 producing sources. The residential sector consumes almost 1/3 of the total produced electricity, therefore, we can have an impact on overall electricity consumption if we manage to reduce it. The growing deployment of smart electricity meters, a tool that provides real time consumption data to consumers and electric utilities, allows the enactment of several measures (e.g. dynamic billing, real-time consumption feedback, load disaggregation) to help people consume less electricity. Electricity providers are motivated to provide clients with the computational power needed to perform these analyses because they offer an extra service that could rise customer fidelity. In addition, electric utilities gain information useful for targeted marketing and optimized billing plans. In order to analyze the large amount of electricity consumption data (several terabytes per day in case of 1,000,000 clients), electric utilities are in need of a method to execute consumption data analysis by distributing the computation over several machines. In this thesis we want to assess if consumption data analysis algorithm are suitable to be executed on an Apache Hadoop cluster. Therefore, the solution that we propose is a Distributed Consumption Analysis (DCA) system, for which developers and data analysts within an electricity providing company can write consumption data analysis routines and uses Apache Hadoop as a MapReduce distribution framework to parallelize the computation. Moreover, we provide utilities’ market strategists with a Web based user interface that allows them to effortlessly execute complex analysis algorithms. Finally, we present an evaluation of the proposed system on real world consumption data, with respect to executing the same computation using a single machine.