Candidate
Sagnik De
Title
Energy Efficient Energy Analytics
Supervisor
Wojciech Golab
Abstract
In the presence of global warming the governments around the world has looked to move away from fossil fuels and towards renewable sources of energy like wind, solar, hydro and nuclear power. However these other sources of power are not without problems related to maintenance cost, power distribution and power transmission cost. Some of these renewable sources of energy also have an environmental cost. The upkeep of the infrastructure of the power generation is determined by peak load of the power generation. Thus a lot of times the electrical infrastructure is left unused during non-peak hours. Thus, it is in the interest of the energy supplier to reduce peak load as much as possible to decrease cost and improve the efficiency of the infrastructure. In order to narrow down and focus on peak power consumption, energy suppliers have started to collect fine grained power data from the customer using smart meters. Smart meter allows for hourly grained data collection related to the power consumption of the customer. Thus it is a time series data. However this results in thousands of data points, which hides the broader trends in power consumption and makes it difficult to make decision on the part energy supplier with regards to a specific customer or to large number of customers. Since data without analysis is useless data, various algorithms has been proposed to which lowers the dimensionality of data, discovers trends (eg. regression), studies the relationship between different types (e.g. temperature and power data) of collected data, summarizes data (e.g. histogram). This allows for easy consumption by the end user. The smart meter data is very compute intensive with large number of large number of houses and each house has the data is collected over a few years. To speed up the smart meter data analysis, computer clusters has been used. Ironically these clusters use up a lot of power. Studies have shown that about 10 % of power is consumed by the customers is used up by the computing infrastructure. In this thesis a GPU will be used to perform analysis of smart meter data and it will compared to a baseline CPU implementation. It will also show that not only GPUs are faster than the CPU, but they are also more power efficient compared to the CPU.