Forecasting underheating in dwellings to detect excess winter mortality risks using time series models

Ahmed I. Ahmed, Robert Scot McLeod*, Matej Gustin

*Korrespondierende/r Autor/-in für diese Arbeit

Publikation: Beitrag in einer FachzeitschriftArtikelBegutachtung

Abstract

Advanced forecasting of impending low temperatures in dwellings could play a transformative role in predicting energy poverty in real-time, thereby helping to prevent excess winter morbidity and mortality. A novel recursive time series model combining AutoRegressive with eXogenous inputs was developed to provide multi-step ahead predictions of the wintertime internal temperatures of homes. A stepwise regression approach was adopted to automate the optimal model selection process based on the minimisation of the Akaike Information Criterion. The model was validated using three case study homes located in Loughborough, UK. Prediction intervals, at the 95% probability level, were used to define a credible interval for the forecasted temperatures at different time horizons during periods of cold weather. The AutoRegressive with eXogenous inputs model proved capable of producing reliable forecasts for 1, 3 and 6 h ahead, achieving Mean Absolute Errors below 1.38 °C for these horizons. The results showed that this model consistently outperformed the more complex AutoRegressive Moving Average with eXogenous inputs model. The study provides the first evidence of the potential for using time series forecasting as part of a high-resolution indoor Winter Early Warning Response System which could be used to identify homes at imminent risk of cold-related health impacts.
Originalspracheenglisch
Aufsatznummer116517
FachzeitschriftApplied Energy
Jahrgang286
DOIs
PublikationsstatusVeröffentlicht - 15 März 2021

ASJC Scopus subject areas

  • Maschinenbau
  • Energie (insg.)
  • Management, Monitoring, Politik und Recht
  • Bauwesen

Fingerprint

Untersuchen Sie die Forschungsthemen von „Forecasting underheating in dwellings to detect excess winter mortality risks using time series models“. Zusammen bilden sie einen einzigartigen Fingerprint.

Dieses zitieren