Abstract
Data-driven models of buildings could potentially reduce implementation barriers for demand forecasting and predictive control in the built environment. However, such models appear to be sensitive to the quality of the available input data. Here, we investigate the influence of sampling time, noise level and amount of available measurement data as well as the quality of the weather forecast on a heating demand forecast with online corrected Artificial Neural Networks. Based on a case study, we demonstrate that sampling time has a stronger influence on the prediction performance than noise level and the amount of available data. Furthermore, we show that using measured ambient temperatures for training appears to provide no benefit compared to using weather forecasts.
Highlights
Predictive control of building heating and cooling systems has the potential to significantly reduce the energy consumption and as a consequence the CO2 emissions of buildings [1]
We investigate the influence of sampling time, noise level and amount of available measurement data as well as the quality of the weather forecast on a heating demand forecast with online corrected Artificial Neural Networks
Based on a case study, we demonstrate that sampling time has a stronger influence on the prediction performance than noise level and the amount of available data
Summary
Predictive control of building heating and cooling systems has the potential to significantly reduce the energy consumption and as a consequence the CO2 emissions of buildings [1]. With increasing availability of high-resolution energy monitoring data in the built environment and the success of data-driven modelling in a variety of domains, the possibility to use these methods for demand forecasting [6] and predictive control [7] has arisen. In [8] we introduced a method based on Artificial Neural Networks (ANN) and forecast correction methods based on online learning and error auto-correlation to make a day-ahead forecast of the heating demand of a multi-use building in Switzerland in a 15-minute time interval. The method showed superior performance compared to other Machine Learning methods and to a variety of fitted simple resistor-capacitor models. The correction methods significantly improved the forecast quality and reliability and the approach outperformed a fitted 5R3C resistor-capacitor building model
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
