Impact of the pre-simulation process of occupant behaviour modelling for residential energy demand simulations

Abstract

Occupant behaviour models play an important role in building energy demand modelling. Useful simulation algorithms have been developed in previous studies; however, the pre-simulation process to prepare modelling parameters for simulated occupants has received less attention. This study elaborated on the pre-simulation process and evaluated how it may alter model performance. We selected the activity-starting probability using American time use survey data as an example. The model performance was compared under three cases representing different numbers and types of variables together with three parameter preparation methods: multinomial log-linear regression, support vector machine, and artificial neural network. All the methods considering basic demographic and time-related variables performed well in reproducing the average probabilities. An increase in significant variables contributed to the reproduction of inter-occupant diversity. All the methods showed similar performances within the given dataset, although they were practically different. The results offer practical guidance for shaping the pre-simulation process.

KEYWORDS:

• Occupant behaviour modelling
• pre-simulation process
• time use
• inter-occupant diversity
• cross comparison

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 The error of Yamaguchi and Shimoda (2017) was 10%–20% for multiple activities. The error of Naspi et al. (2018) for the window-closing action was 15%. The RMSE for modelling the energy use of appliances in a low-energy house in Candanedo, Feldheim, and Deramaix (2017) was greater than 65%.

Additional information

Funding

This work was supported by JSPS KAKENHI (grant number 20H02312) and JST SPRING (grant number JPMJSP2138); Japan Science and Technology Agency.