In western culture it is common to view humans and by extension human behavior as highly individual. This view often ensues a perspective on behavior that emphasizes its high variability between individuals. When it comes to including information on human behavior in energy models, how to deal with such variability is often viewed as an important challenge.

1. 1.

   An example of the employed diaries is included in Statistisches Bundesamt (2016).

2. 2.

   When talking about the consequences of modeling energy using behavior as part of a bottom-up approach, the term “energy system” is employed in an abstract meaning which encompasses all possible aggregation levels from households to grid sections, to areas and so forth.

3. 3.

   In the 2012 article Torriti analyzes for 15 European countries variations in occupancy levels and aims to deduce DSM strategies for shifting user behavior. Although to my knowledge this is the first study to introduce a concept of occupancy variation, unfortunately, the study does not offer (theoretical) ideas on the meaning of variability and how its related to energy using flexibility. Furthermore, the used indicators for behavioral variability and deduced DSM strategies are not well justified. The proposed indicator for flexibility is peak variance and given for two time periods within a day which are identified as peak events. Peak variance is calculated for peak events restricted to 40 min for each period per country between 7 a.m. and 8 a.m. and 19:30 p.m. and 20:30 p.m. (exact times per country not reported) according to the following formula:

   \(\mu_{T,T + 1} = \frac{{\upomega _{T} }}{{\upomega _{T + 1} }}\) where\(\upomega _{T}\) is the level of occupancy in time period\(T\) (Torriti, 2012, p. 201).

   As this indicator gives the changes in occupancy status from one time period to the next, it does reflect variation in behavior sequences in a peak event period in relation to the following time period but it seems not suitable as indicator of behavioral flexibility. This is because variation in occupancy status gives the amount of changes occurring in occupancy but cannot relate whether or not those changes are timely fixed changes in a behavior. High variance in peak periods means that it is more likely that changes in occupancy status and thereby in electricity consumption occurs but an aggregate description of variation cannot indicate flexibility of behavior in that time period because nothing is known about the possible restrictedness or structure of individual behavior sequences.

4. 4.

   Aerts, Minnen, Glorieux, Wouters and Descamps (2014) reference a third influential occupancy model developed by Widén, Nilsson and Wäckelgård (2009), which is excluded in the above list of bottom-up simulations using TUD for modelling household appliance using behavior because it models lighting demand.

5. 5.

   Time dependence is operationalized by (Torriti, 2017, p. 39) as follows:

   $$T_{DEP} = \frac{{Max\left[ {x_{i} - m\left( X \right)} \right]}}{m\left( X \right)}$$

   where\(x_{i}\) is the number of minutes associated with the practice\(x\) at the time of the day\(i\) and\(m\left( X \right)\) is the mean number of minutes of practice\(x\).

6. 6.

   Operant is the theoretical term for behavior which is defined by its consequences (as was described in the section theoretical analysis of behavioral variability).

7. 7.

   A verbal description of consequences is also not the correct way to identify operants in a behavior theoretical analysis, but in keeping with the advantages of the TUS, it is a compromise which could improve upon the information one can obtain from TUSs.

8. 8.

   A display of frequency distributions for the 22 activities for weekday and weekend data is given in Appendix A.

9. 9.

   The distance matrix is calculated using the stringdist R package version 0.9.5.2 (van Der Loo, 2014).

10. 10.

    The TUD analysis by Palm et al. (2018) also used a clustering method, but from their description it is not clear which method they employed: “The clustering was done in R version 3.2.3 (R Core Team, 2014) using Ward’s distance and the TraMineR (Gabadinho, Ritschard, Müller, & Studer, 2011) and WeightedCluster (Studer, 2013) packages.” p. 102. It is probable that they used the ward method for clustering, leaving the distance measure unspecified.

11. 11.

    The graphical display has the advantage of relating the frequency of the differentiating activities to time of day, which is an important feature for interpretation and also for recognizing differences, which might be unseen in comparison of mean values in activities between clusters. Nonetheless mean values, standard deviations and results of a robust ANOVA for trimmed means (Wilcox, 2012) for all activities for weekday and weekend data are reported in Appendix C. The results are in accordance with the description of the plotted activity patterns pointing towards the same significant differences in activities between clusters. That is, occupational and educational activities and sleeping for weekday clusters and occupational and social activities, hobbies, watching TV and sleeping for weekend clusters.

12. 12.

    For weekend data preparing meals is homogeneous between all clusters except for weekend cluster 5, which in correspondence to its high peak in physiological recreation in the morning also has a steeper rise in the morning for preparing meals and cleaning and a very high frequency in the midday peak: 23% maximum compared to 8% maximum (weekend cluster 3); view Appendix EFigure E.2.

13. 13.

    The overall frequencies of the doing laundry activity in terms of mean values and standard deviations are not meaningfully different between clusters: weekday cluster 1 (M = 0.66,SD = 1.69); weekday cluster 2 (M = 0.31,SD = 1.42); weekday cluster 3 (M = 1.22,SD = 2.74).

14. 14.

    ICT is used by the authors as an abbreviation for ‘information and communications technologies’ (Nicholls & Strengers, 2015).

15. 15.

    Even though the general term building model is used which can include many different components, here it is only regarded in reference to shifting energy using behavior in time, i.e., without considering other components such as thermal components or transmission losses which would be relevant for a complete description of a building model.

16. 16.

    Supported by the Lower Saxony Ministry of Science and Culture through the ‘Niedersächsiches Vorab’ grant program (grant ZN3043). Final Project report in Blaufuß et al. (2019).

17. 17.

    In addition to the assumptions from TUD, driving schedules are implemented into the model (Reinhold, 2019). In this simulation for 60% of the people in each cluster, further influencing the presence times at home.

Authors and Affiliations

1. Braunschweig, Germany

   Farina Wille

Corresponding author

Correspondence toFarina Wille.

Reprints and permissions

© 2021 The Author(s), under exclusive license to Springer Fachmedien Wiesbaden GmbH, part of Springer Nature

Policies and ethics