I n d i v i d u a l H o u s e h o l d B e h a v i o u r M o d e l l i n g
529
The MCMC technique is also adopted by Widén, who developed a three state
occupancy model (Widén, 2009) in which users can either be absent, present and active
or present and inactive. Later, the model is extended to a nine state model that
combines occupancy and activities by adding six activities that may lead to electrical
energy consumption (Widén, 2010). In addition to the electricity consumption that is
derived from the household activities, the power demand for artificial lighting is estimated
in a lighting model.
In a Markov chain the current state typically only depends on the previous state and, in
case of a non-homogenous chain, on the time. The main disadvantage of the limited
memory of the Markov chain is that it may lead to improbable results. For example, most
people are expected to wake up somewhere between 6:00 AM and 10:00 AM, which
implies that the probability of waking up at each time step of this interval is high.
Nevertheless, at each of these time steps the possibility of not waking up still exists,
which may lead to a pattern where an individual never wakes up. To avoid malicious
patterns, control mechanisms are needed to ensure a certain average expected duration
and reoccurrence time for each state. Furthermore, from a computational point of view
the MCMC method should be optimized, since it leads to a high number of redundant
calculations whenever there is no transition between two steps (Robinson, 2011).
In order to bypass these issues, Wilke proposes the use of three time-dependent
quantities, namely the probability to be at home, the conditional probability to start the
activity whilst someone is at home and the probability distribution for the duration of the
activity (Wilke, 2011). Similar to the previous models, activities may only be started if the
individual is at home and awake. However, instead of checking at every subsequent time
step whether a transition to another activity may occur, a duration is assigned to the
activity or state at its start, which is more interesting from a computational point of view.
Model outline
In this work, an occupancy and activity model is presented. We start with the definition of
respondent- and household types based on a number of key variables. Next, a
description of the occupancy model and activity is provided in which the modelling
methodology is explained. Finally, we shortly address some seasonal effects.
In order to obtain accurate user and household behaviour patterns, we attempted to
define respondent types and household types in such a way that similar behaviour within
these types is to be expected. A wide range of individual- and household parameters are
available in the TUS and many of them could be influencing factors for household
behaviour that ultimately determine the energy consumption. In literature, building
related parameters such as the dwelling type, surface area of the dwelling and appliance
holdings are frequently proposed to have strong explanatory power. Household related
variables that are often put forward are household income, number of household
members, household composition and age. However, it is suggested by McLaughlin
(McLoughlin, 2012) that not these variables but their underlying implications are the
source of their explanatory power. For example, the age of the head of the home (HoH)
shows to have a negative effect on energy consumption. This can be due to the fact that
middle-aged HoH’s have more children living in the home, resulting in a higher energy
consumption. Retired HoH’s obviously spend more time at home, explaining higher
energy consumption.
Ultimately, as suggested by Yao and Steemers (Yao, 2005), it all boils down to the
number of people and the amount of time they spend in the house. As a result, we
decided to define respondent types based on their role in the household and their
