Selection of representative slices from historic load and generation time series using regular decomposition

Niina Helistö (Corresponding Author), Juha Kiviluoma, Hannu Reittu

Research output: Contribution to journalArticleScientific

19 Downloads (Pure)

Abstract

A generalized clustering method, referred to as regular decomposition, was applied to a long-term power system planning study covering countries in the Northern Europe. The aim of the method was to reduce the temporal resolution by selecting representative periods concurrently from load time series and variable generation capacity factor time series. Two versions of the method were employed: one with overlapping candidates and one without overlapping candidates. The method was compared to other selection methods as well as using full year time series in the planning problem. Furthermore, the comparison was repeated with various number of representative periods and in three CO2 price scenarios in order to provide more robust results. The method was also demonstrated to scale well when applied to inter-annual time series. When selecting four weeks or more, the regular decomposition method without overlapping candidates was shown to perform relatively well compared to a modified k-means method. However, regular decomposition was outperformed by a random sampling method in terms of both the computational time and total costs resulting from the power system model runs. The results highlight the need to include net load peaks in the selected periods and to carefully estimate their position in the time series.
Original languageEnglish
Number of pages22
JournalNot yet accepted into a journal
Publication statusSubmitted - 2019
MoE publication typeB1 Article in a scientific magazine

Fingerprint

Time series
Decomposition
Planning
Sampling
Costs

Keywords

  • generation expansion planning
  • representative periods
  • power system planning
  • regular decomposition
  • variable renewable energy

Cite this

@article{35cb1ff8177a4f6e826a4c2a4a336037,
title = "Selection of representative slices from historic load and generation time series using regular decomposition",
abstract = "A generalized clustering method, referred to as regular decomposition, was applied to a long-term power system planning study covering countries in the Northern Europe. The aim of the method was to reduce the temporal resolution by selecting representative periods concurrently from load time series and variable generation capacity factor time series. Two versions of the method were employed: one with overlapping candidates and one without overlapping candidates. The method was compared to other selection methods as well as using full year time series in the planning problem. Furthermore, the comparison was repeated with various number of representative periods and in three CO2 price scenarios in order to provide more robust results. The method was also demonstrated to scale well when applied to inter-annual time series. When selecting four weeks or more, the regular decomposition method without overlapping candidates was shown to perform relatively well compared to a modified k-means method. However, regular decomposition was outperformed by a random sampling method in terms of both the computational time and total costs resulting from the power system model runs. The results highlight the need to include net load peaks in the selected periods and to carefully estimate their position in the time series.",
keywords = "generation expansion planning, representative periods, power system planning, regular decomposition, variable renewable energy",
author = "Niina Helist{\"o} and Juha Kiviluoma and Hannu Reittu",
year = "2019",
language = "English",
journal = "Not yet accepted into a journal",

}

TY - JOUR

T1 - Selection of representative slices from historic load and generation time series using regular decomposition

AU - Helistö, Niina

AU - Kiviluoma, Juha

AU - Reittu, Hannu

PY - 2019

Y1 - 2019

N2 - A generalized clustering method, referred to as regular decomposition, was applied to a long-term power system planning study covering countries in the Northern Europe. The aim of the method was to reduce the temporal resolution by selecting representative periods concurrently from load time series and variable generation capacity factor time series. Two versions of the method were employed: one with overlapping candidates and one without overlapping candidates. The method was compared to other selection methods as well as using full year time series in the planning problem. Furthermore, the comparison was repeated with various number of representative periods and in three CO2 price scenarios in order to provide more robust results. The method was also demonstrated to scale well when applied to inter-annual time series. When selecting four weeks or more, the regular decomposition method without overlapping candidates was shown to perform relatively well compared to a modified k-means method. However, regular decomposition was outperformed by a random sampling method in terms of both the computational time and total costs resulting from the power system model runs. The results highlight the need to include net load peaks in the selected periods and to carefully estimate their position in the time series.

AB - A generalized clustering method, referred to as regular decomposition, was applied to a long-term power system planning study covering countries in the Northern Europe. The aim of the method was to reduce the temporal resolution by selecting representative periods concurrently from load time series and variable generation capacity factor time series. Two versions of the method were employed: one with overlapping candidates and one without overlapping candidates. The method was compared to other selection methods as well as using full year time series in the planning problem. Furthermore, the comparison was repeated with various number of representative periods and in three CO2 price scenarios in order to provide more robust results. The method was also demonstrated to scale well when applied to inter-annual time series. When selecting four weeks or more, the regular decomposition method without overlapping candidates was shown to perform relatively well compared to a modified k-means method. However, regular decomposition was outperformed by a random sampling method in terms of both the computational time and total costs resulting from the power system model runs. The results highlight the need to include net load peaks in the selected periods and to carefully estimate their position in the time series.

KW - generation expansion planning

KW - representative periods

KW - power system planning

KW - regular decomposition

KW - variable renewable energy

M3 - Article

JO - Not yet accepted into a journal

JF - Not yet accepted into a journal

ER -