Electric power demand forecasting using multivariate time series analysis

OData support
Supervisor:
Barta Gergő
Department of Telecommunications and Media Informatics

Forecasting electric power consumption is a really effective tool in the hands of service providers, which can help them to optimize their own strategies, gain some advantage on their competitors, while they can protect the enviroment, and stop wasting energy.

Data mining provides a solution, which can help analyzers to retrieve real, and precious informations out of a fuzzy set of big data, with statistical, and matematical methods. For that, you only need a pair of experienced eyes, and a few good ideas to start the process.

During my work on this paper, I used a data set collected by a british electric service provider, and I tried to create the best model for the task I got, which was to predict long term electric power consumption. I made myself familiar with some more complex version of regression, k-nearest neighbour technique, support vector machines, and neural nets. While I was closing on my best solution for the problem, I started to understand these algorithms. I used the software RapidMiner, which was not completely new for me, I worked with it on my last paper, but this time I had the chance to go deeper in its capabilities.

I had success with outperforming my primitive model, the baseline, which was my main goal, but besides thats, I also got a good feedback on comparing my numbers, with the results of the competition, that was the origin of my dataset.

Downloads

Please sign in to download the files of this thesis.