Improving Electricity Short-term Load Forecast with Smart-meter Data
No Thumbnail Available
URL
Journal Title
Journal ISSN
Volume Title
Perustieteiden korkeakoulu |
Master's thesis
Ask about the availability of the thesis by sending email to the Aalto University Learning Centre oppimiskeskus@aalto.fi
Authors
Date
2016-06-13
Department
Major/Subject
Machine Learning and Data Mining
Mcode
SCI3015
Degree programme
Master’s Programme in Machine Learning and Data Mining (Macadamia)
Language
en
Pages
79 + 0
Series
Abstract
The main goal of the thesis is to improve the short-term electricity load forecasting using smart-meter data. Due to the onset of deregulated energy market in Finland and in most European countries in the 1990’s, the energy market has become more competitive. The usage of smart-meters in measuring and transmitting at most hourly energy data has enabled us to improve the short-term load forecasting, there by playing a great role in the energy market. There are two broad short-term forecasting categories that were analysed in this thesis: the first one is clustering of metering points as a preprocessing technique before feeding to forecasting models, where as, the second one is using the aggregated values from the metering points as an input to the forecasting models. Temperature, together with historical consumption behaviour, has been among the factors considered in affecting the short-term electric energy consumption. Besides, the factors used in clustering the individual metering points have been studied well so that the models created would benefit from them. The effect of the diverse nature of the clusters on the need for more diverse forecasting models to address the individual clusters has been studied. The effect of cluster numbers in the overall accuracy of the forecast has also been analysed. There are six methods implemented: Support Vector Machines (SVM), Auto Regressive Moving Average (ARMA), non-linear multinomial regression, Neural Network (NN), exponential regression and non-seasonal regression. During model selection, for each cluster segment, the models developed will compete each other and the best model is selected using 10-fold cross validation.Description
Supervisor
Rousu, JuhoThesis advisor
Hollmen, JaakkoHirsimäki, Teemu
Keywords
time-series, forecasting, energy data, regression