Creating Electrical Load Profiles Through Time Series Clustering
(2019)Department of Automatic Control
- Abstract
- Challenged by new problems ranging from new renewable production methods to novel sources of loads, the Swedish electrical system is facing a time of change. To keep up with this process, having knowledge about what trends are present in electrical consumer behavior will be key. At the same time a new tool is emerging in the form of more available electrical load data. Motivated by this development, this thesis comprises the work of finding a data driven time series clustering method that can provide meaningful and intuitive profiles to describe the behaviors of consumers in the local electrical grid level.
Part one of this work consists of a broad survey where five clustering methods from literature as well as one novel method based on... (More) - Challenged by new problems ranging from new renewable production methods to novel sources of loads, the Swedish electrical system is facing a time of change. To keep up with this process, having knowledge about what trends are present in electrical consumer behavior will be key. At the same time a new tool is emerging in the form of more available electrical load data. Motivated by this development, this thesis comprises the work of finding a data driven time series clustering method that can provide meaningful and intuitive profiles to describe the behaviors of consumers in the local electrical grid level.
Part one of this work consists of a broad survey where five clustering methods from literature as well as one novel method based on ordinary least square modeling are applied to real electrical load data with the goal of finding the most useful one. The data consists of hourly measured consumption from varied consumer types
in the local grid in southern Sweden. Applying the methods to a small set of consumers, the linear model method is deemed to give the profiles that provide the most intuitive and useful information. With models based on time of day and outdoor temperature, the model generates profiles showing the consumers most typical load
patterns.
In part two of this work the linear model method is presented with improvements where most importantly consumers that are poorly modelled by the time of day and temperature are excluded from the clustering process. To test the performance of the improved method it is applied to two large data sets containing varied
consumer types and only small house consumers respectively. The method is shown to generate intuitive profiles that even in a homogeneous data set consisting of only housing consumers can distinguish different types of daily routines. As a method that is quick, simple and easily understandable, the method is suggested as a tool
both in decision making processes as well as an exploratory tool for more advanced consumer modeling. (Less)
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/8994319
- author
- Agner, Felix
- supervisor
- organization
- year
- 2019
- type
- H3 - Professional qualifications (4 Years - )
- subject
- report number
- TFRT-6092
- ISSN
- 0280-5316
- language
- English
- id
- 8994319
- date added to LUP
- 2019-09-06 13:23:45
- date last changed
- 2019-09-06 13:23:45
@misc{8994319,
abstract = {{Challenged by new problems ranging from new renewable production methods to novel sources of loads, the Swedish electrical system is facing a time of change. To keep up with this process, having knowledge about what trends are present in electrical consumer behavior will be key. At the same time a new tool is emerging in the form of more available electrical load data. Motivated by this development, this thesis comprises the work of finding a data driven time series clustering method that can provide meaningful and intuitive profiles to describe the behaviors of consumers in the local electrical grid level.
Part one of this work consists of a broad survey where five clustering methods from literature as well as one novel method based on ordinary least square modeling are applied to real electrical load data with the goal of finding the most useful one. The data consists of hourly measured consumption from varied consumer types
in the local grid in southern Sweden. Applying the methods to a small set of consumers, the linear model method is deemed to give the profiles that provide the most intuitive and useful information. With models based on time of day and outdoor temperature, the model generates profiles showing the consumers most typical load
patterns.
In part two of this work the linear model method is presented with improvements where most importantly consumers that are poorly modelled by the time of day and temperature are excluded from the clustering process. To test the performance of the improved method it is applied to two large data sets containing varied
consumer types and only small house consumers respectively. The method is shown to generate intuitive profiles that even in a homogeneous data set consisting of only housing consumers can distinguish different types of daily routines. As a method that is quick, simple and easily understandable, the method is suggested as a tool
both in decision making processes as well as an exploratory tool for more advanced consumer modeling.}},
author = {{Agner, Felix}},
issn = {{0280-5316}},
language = {{eng}},
note = {{Student Paper}},
title = {{Creating Electrical Load Profiles Through Time Series Clustering}},
year = {{2019}},
}