Mitigating the Weaknesses of Machine Learning in Short–Term Forecasting of Aggregated Power System Active Loads

Pekka Koponen, Harri Niska, Antti Mutanen

Research output: Chapter in Book/Report/Conference proceedingConference article in proceedingsScientificpeer-review

Abstract

Machine learning methods predict accurately in situations that are adequately included in the learning data and do not require detailed domain knowledge based model development. They have their weaknesses compared with other
forecasting methods, however. For example, they may fail in many new situations not experienced before. Hybrid models are increasingly popular as they are capable of combining the strengths of several modelling methods and mitigating the weaknesses. We study short–term forecasting of aggregated
electricity demand that includes dynamically controlled thermal storage. Purely measurement data driven models tend to fail in forecasting power in rarely occurring situations, such as dynamic load control actions and extreme weather. The thermal dynamics of the loads, large outdoor temperature variations,
and changes in the energy technologies contribute to this challenge. Combining various information sources and the strengths of different modelling approaches is needed. We study the following approach using field trial data covering over 7500 houses and 27 months. We forecast control responses and load
saturation using models that have physically based model structures. Then we forecast the residual using data driven models, such as machine learning models designed and tuned to learn also system dynamics. The load forecast is the sum of these component forecasts. We further improve the forecast by using
ensemble forecasting and physically based range forecasts. We find that the hybrid methods are more accurate than their component methods alone and combining several hybridization approaches can improve the performance and reliability.
Original languageEnglish
Title of host publicationIEEE International Conference on Industrial Informatics (IEEE-INDIN 2019)
PublisherInstitute of Electrical and Electronic Engineers IEEE
Number of pages8
Publication statusAccepted/In press - 22 Jul 2019
MoE publication typeA4 Article in a conference publication
Event 17th IEEE International Conference on Industrial Informatics, IEEE-INDIN 2019: IEEE-INDIN 2019 - Aalto University, Espoo, Finland
Duration: 23 Jul 201925 Jul 2019
https://www.indin2019.org/

Conference

Conference 17th IEEE International Conference on Industrial Informatics, IEEE-INDIN 2019
CountryFinland
CityEspoo
Period23/07/1925/07/19
Internet address

Fingerprint

Learning systems
Dynamic loads
Model structures
Dynamical systems
Temperature
Hot Temperature

Keywords

  • forecasting
  • hybrid intelligent systems
  • machine learning
  • multilayer perceptrons
  • power demand
  • support vector

Cite this

Koponen, P., Niska, H., & Mutanen, A. (Accepted/In press). Mitigating the Weaknesses of Machine Learning in Short–Term Forecasting of Aggregated Power System Active Loads. In IEEE International Conference on Industrial Informatics (IEEE-INDIN 2019) Institute of Electrical and Electronic Engineers IEEE.
Koponen, Pekka ; Niska, Harri ; Mutanen, Antti. / Mitigating the Weaknesses of Machine Learning in Short–Term Forecasting of Aggregated Power System Active Loads. IEEE International Conference on Industrial Informatics (IEEE-INDIN 2019). Institute of Electrical and Electronic Engineers IEEE, 2019.
@inproceedings{b95c21fee0ac4909a8b344f479fab595,
title = "Mitigating the Weaknesses of Machine Learning in Short–Term Forecasting of Aggregated Power System Active Loads",
abstract = "Machine learning methods predict accurately in situations that are adequately included in the learning data and do not require detailed domain knowledge based model development. They have their weaknesses compared with otherforecasting methods, however. For example, they may fail in many new situations not experienced before. Hybrid models are increasingly popular as they are capable of combining the strengths of several modelling methods and mitigating the weaknesses. We study short–term forecasting of aggregatedelectricity demand that includes dynamically controlled thermal storage. Purely measurement data driven models tend to fail in forecasting power in rarely occurring situations, such as dynamic load control actions and extreme weather. The thermal dynamics of the loads, large outdoor temperature variations,and changes in the energy technologies contribute to this challenge. Combining various information sources and the strengths of different modelling approaches is needed. We study the following approach using field trial data covering over 7500 houses and 27 months. We forecast control responses and loadsaturation using models that have physically based model structures. Then we forecast the residual using data driven models, such as machine learning models designed and tuned to learn also system dynamics. The load forecast is the sum of these component forecasts. We further improve the forecast by usingensemble forecasting and physically based range forecasts. We find that the hybrid methods are more accurate than their component methods alone and combining several hybridization approaches can improve the performance and reliability.",
keywords = "forecasting, hybrid intelligent systems, machine learning, multilayer perceptrons, power demand, support vector",
author = "Pekka Koponen and Harri Niska and Antti Mutanen",
note = "The IEEE INDIN 2019 procedings will be soon published by IEEE and will be available via IEEEE Xplore",
year = "2019",
month = "7",
day = "22",
language = "English",
booktitle = "IEEE International Conference on Industrial Informatics (IEEE-INDIN 2019)",
publisher = "Institute of Electrical and Electronic Engineers IEEE",
address = "United States",

}

Koponen, P, Niska, H & Mutanen, A 2019, Mitigating the Weaknesses of Machine Learning in Short–Term Forecasting of Aggregated Power System Active Loads. in IEEE International Conference on Industrial Informatics (IEEE-INDIN 2019). Institute of Electrical and Electronic Engineers IEEE, 17th IEEE International Conference on Industrial Informatics, IEEE-INDIN 2019, Espoo, Finland, 23/07/19.

Mitigating the Weaknesses of Machine Learning in Short–Term Forecasting of Aggregated Power System Active Loads. / Koponen, Pekka; Niska, Harri; Mutanen, Antti.

IEEE International Conference on Industrial Informatics (IEEE-INDIN 2019). Institute of Electrical and Electronic Engineers IEEE, 2019.

Research output: Chapter in Book/Report/Conference proceedingConference article in proceedingsScientificpeer-review

TY - GEN

T1 - Mitigating the Weaknesses of Machine Learning in Short–Term Forecasting of Aggregated Power System Active Loads

AU - Koponen, Pekka

AU - Niska, Harri

AU - Mutanen, Antti

N1 - The IEEE INDIN 2019 procedings will be soon published by IEEE and will be available via IEEEE Xplore

PY - 2019/7/22

Y1 - 2019/7/22

N2 - Machine learning methods predict accurately in situations that are adequately included in the learning data and do not require detailed domain knowledge based model development. They have their weaknesses compared with otherforecasting methods, however. For example, they may fail in many new situations not experienced before. Hybrid models are increasingly popular as they are capable of combining the strengths of several modelling methods and mitigating the weaknesses. We study short–term forecasting of aggregatedelectricity demand that includes dynamically controlled thermal storage. Purely measurement data driven models tend to fail in forecasting power in rarely occurring situations, such as dynamic load control actions and extreme weather. The thermal dynamics of the loads, large outdoor temperature variations,and changes in the energy technologies contribute to this challenge. Combining various information sources and the strengths of different modelling approaches is needed. We study the following approach using field trial data covering over 7500 houses and 27 months. We forecast control responses and loadsaturation using models that have physically based model structures. Then we forecast the residual using data driven models, such as machine learning models designed and tuned to learn also system dynamics. The load forecast is the sum of these component forecasts. We further improve the forecast by usingensemble forecasting and physically based range forecasts. We find that the hybrid methods are more accurate than their component methods alone and combining several hybridization approaches can improve the performance and reliability.

AB - Machine learning methods predict accurately in situations that are adequately included in the learning data and do not require detailed domain knowledge based model development. They have their weaknesses compared with otherforecasting methods, however. For example, they may fail in many new situations not experienced before. Hybrid models are increasingly popular as they are capable of combining the strengths of several modelling methods and mitigating the weaknesses. We study short–term forecasting of aggregatedelectricity demand that includes dynamically controlled thermal storage. Purely measurement data driven models tend to fail in forecasting power in rarely occurring situations, such as dynamic load control actions and extreme weather. The thermal dynamics of the loads, large outdoor temperature variations,and changes in the energy technologies contribute to this challenge. Combining various information sources and the strengths of different modelling approaches is needed. We study the following approach using field trial data covering over 7500 houses and 27 months. We forecast control responses and loadsaturation using models that have physically based model structures. Then we forecast the residual using data driven models, such as machine learning models designed and tuned to learn also system dynamics. The load forecast is the sum of these component forecasts. We further improve the forecast by usingensemble forecasting and physically based range forecasts. We find that the hybrid methods are more accurate than their component methods alone and combining several hybridization approaches can improve the performance and reliability.

KW - forecasting

KW - hybrid intelligent systems

KW - machine learning

KW - multilayer perceptrons

KW - power demand

KW - support vector

M3 - Conference article in proceedings

BT - IEEE International Conference on Industrial Informatics (IEEE-INDIN 2019)

PB - Institute of Electrical and Electronic Engineers IEEE

ER -

Koponen P, Niska H, Mutanen A. Mitigating the Weaknesses of Machine Learning in Short–Term Forecasting of Aggregated Power System Active Loads. In IEEE International Conference on Industrial Informatics (IEEE-INDIN 2019). Institute of Electrical and Electronic Engineers IEEE. 2019