Enhancing predictive models for short-term forecasting electricity consumption in smart buildings

Manuel Herrera, Daniel Fosas De Pando, Bruno M Beltran, David Coley

Research output: Contribution to conferencePaper

14 Downloads (Pure)

Abstract

Lighting, heating, and air conditioning systems are instances of how electricity use at buildings is of key importance for occupants comfort and well-being. Since the electricity can be produced but cannot be stored, for utility companies it is important to reliably forecast energy supply almost in near real-time. Nowadays, smart grid technologies development also require a high resolution forecasting to eliminate blackouts and to optimally adapt energy supply to customers’ needs. These are the reasons why the finest Machine Learning and Data Science based methods have been developed and applied to approach as much accurate as possible predictive models for short-term electricity consumption. This paper proposes to enhance those predictive models by using weather and calendar information to configure a more complete working database. In addition, a cluster-based forecasting methodology will augment any predictive model with learning from other buildings. Thus, predicting future values for one smart meter is approached by utilising not only its own historical electricity consumption values, but working with a multivariate time series on weather and calendar data and information from other buildings at the same cluster. This proposal has been tested with measures from smart meters collected every 30 minutes during one year for 5 selected buildings in Bristol (UK). The enhancing methodology can predict electricity consumption data with higher accuracy than using data from just one building.
Original languageEnglish
Pages26-30
Number of pages5
Publication statusPublished - 29 Jun 2018
Event1st International Conference on Data for Low Energy Buildings - University of Murcia, Murcia, Spain
Duration: 28 Jun 201829 Jun 2018
http://dataleb18.inf.um.es/

Conference

Conference1st International Conference on Data for Low Energy Buildings
Abbreviated titleDATALEB18
CountrySpain
CityMurcia
Period28/06/1829/06/18
Internet address

Fingerprint

Intelligent buildings
Electricity
Smart meters
Air conditioning
Learning systems
Time series
Lighting
Heating
Industry

Keywords

  • electricity consumption
  • smart meters
  • predictive models
  • machine learning

Cite this

Herrera, M., Fosas De Pando, D., Beltran, B. M., & Coley, D. (2018). Enhancing predictive models for short-term forecasting electricity consumption in smart buildings. 26-30. Paper presented at 1st International Conference on Data for Low Energy Buildings, Murcia, Spain.

Enhancing predictive models for short-term forecasting electricity consumption in smart buildings. / Herrera, Manuel; Fosas De Pando, Daniel; Beltran, Bruno M; Coley, David.

2018. 26-30 Paper presented at 1st International Conference on Data for Low Energy Buildings, Murcia, Spain.

Research output: Contribution to conferencePaper

Herrera, M, Fosas De Pando, D, Beltran, BM & Coley, D 2018, 'Enhancing predictive models for short-term forecasting electricity consumption in smart buildings' Paper presented at 1st International Conference on Data for Low Energy Buildings, Murcia, Spain, 28/06/18 - 29/06/18, pp. 26-30.
Herrera M, Fosas De Pando D, Beltran BM, Coley D. Enhancing predictive models for short-term forecasting electricity consumption in smart buildings. 2018. Paper presented at 1st International Conference on Data for Low Energy Buildings, Murcia, Spain.
Herrera, Manuel ; Fosas De Pando, Daniel ; Beltran, Bruno M ; Coley, David. / Enhancing predictive models for short-term forecasting electricity consumption in smart buildings. Paper presented at 1st International Conference on Data for Low Energy Buildings, Murcia, Spain.5 p.
@conference{d1e839ba3f0940789ec757a009d60309,
title = "Enhancing predictive models for short-term forecasting electricity consumption in smart buildings",
abstract = "Lighting, heating, and air conditioning systems are instances of how electricity use at buildings is of key importance for occupants comfort and well-being. Since the electricity can be produced but cannot be stored, for utility companies it is important to reliably forecast energy supply almost in near real-time. Nowadays, smart grid technologies development also require a high resolution forecasting to eliminate blackouts and to optimally adapt energy supply to customers’ needs. These are the reasons why the finest Machine Learning and Data Science based methods have been developed and applied to approach as much accurate as possible predictive models for short-term electricity consumption. This paper proposes to enhance those predictive models by using weather and calendar information to configure a more complete working database. In addition, a cluster-based forecasting methodology will augment any predictive model with learning from other buildings. Thus, predicting future values for one smart meter is approached by utilising not only its own historical electricity consumption values, but working with a multivariate time series on weather and calendar data and information from other buildings at the same cluster. This proposal has been tested with measures from smart meters collected every 30 minutes during one year for 5 selected buildings in Bristol (UK). The enhancing methodology can predict electricity consumption data with higher accuracy than using data from just one building.",
keywords = "electricity consumption, smart meters, predictive models, machine learning",
author = "Manuel Herrera and {Fosas De Pando}, Daniel and Beltran, {Bruno M} and David Coley",
year = "2018",
month = "6",
day = "29",
language = "English",
pages = "26--30",
note = "1st International Conference on Data for Low Energy Buildings, DATALEB18 ; Conference date: 28-06-2018 Through 29-06-2018",
url = "http://dataleb18.inf.um.es/",

}

TY - CONF

T1 - Enhancing predictive models for short-term forecasting electricity consumption in smart buildings

AU - Herrera, Manuel

AU - Fosas De Pando, Daniel

AU - Beltran, Bruno M

AU - Coley, David

PY - 2018/6/29

Y1 - 2018/6/29

N2 - Lighting, heating, and air conditioning systems are instances of how electricity use at buildings is of key importance for occupants comfort and well-being. Since the electricity can be produced but cannot be stored, for utility companies it is important to reliably forecast energy supply almost in near real-time. Nowadays, smart grid technologies development also require a high resolution forecasting to eliminate blackouts and to optimally adapt energy supply to customers’ needs. These are the reasons why the finest Machine Learning and Data Science based methods have been developed and applied to approach as much accurate as possible predictive models for short-term electricity consumption. This paper proposes to enhance those predictive models by using weather and calendar information to configure a more complete working database. In addition, a cluster-based forecasting methodology will augment any predictive model with learning from other buildings. Thus, predicting future values for one smart meter is approached by utilising not only its own historical electricity consumption values, but working with a multivariate time series on weather and calendar data and information from other buildings at the same cluster. This proposal has been tested with measures from smart meters collected every 30 minutes during one year for 5 selected buildings in Bristol (UK). The enhancing methodology can predict electricity consumption data with higher accuracy than using data from just one building.

AB - Lighting, heating, and air conditioning systems are instances of how electricity use at buildings is of key importance for occupants comfort and well-being. Since the electricity can be produced but cannot be stored, for utility companies it is important to reliably forecast energy supply almost in near real-time. Nowadays, smart grid technologies development also require a high resolution forecasting to eliminate blackouts and to optimally adapt energy supply to customers’ needs. These are the reasons why the finest Machine Learning and Data Science based methods have been developed and applied to approach as much accurate as possible predictive models for short-term electricity consumption. This paper proposes to enhance those predictive models by using weather and calendar information to configure a more complete working database. In addition, a cluster-based forecasting methodology will augment any predictive model with learning from other buildings. Thus, predicting future values for one smart meter is approached by utilising not only its own historical electricity consumption values, but working with a multivariate time series on weather and calendar data and information from other buildings at the same cluster. This proposal has been tested with measures from smart meters collected every 30 minutes during one year for 5 selected buildings in Bristol (UK). The enhancing methodology can predict electricity consumption data with higher accuracy than using data from just one building.

KW - electricity consumption

KW - smart meters

KW - predictive models

KW - machine learning

M3 - Paper

SP - 26

EP - 30

ER -