The description of electricity demand mostly focusses on aggregated data. However, for some problems, more detailed data is required. Section 4.3.1 describes techniques to select customers in pilot projects. Section 4.3.2 focusses on how electricity demand is modelled.
4.3.1
Selection
The preferred way to sample customers for a pilot project is using random sampling, because it allows for a comparison between control and treatment groups [75]. A sufficient number of customers is needed to draw conclusions. A trade-off between the information value and the cost of sampling has to be made.
A Swedish study selected 500 households as part of a time of use pilot [76]. The study didn’t indicate how the households were selected. A set of 40 households was selected in a demand response pilot project in Norway [77]. The participants were recruited by means of a local news paper. The voluntary basis for participation suggests that people interested in demand response, over-responded.
Random selection is very costly (Section 3.4). The approach taken in [77] is convenience sampling. The technique is easy and less costly than simple random sampling, but comes with an over-representation of people willing to participate in such a project and who are interested in demand response. This problem is also addressed in the ‘Linear’ project (Section 7.1).
4.3.2
Modelling
Two strategies exist to model residential low voltage electricity demand: bottom up and top down. Bottom up approaches model electricity demand of each individual appliance. The sum over all appliances results in the electricity demand at the connection point. Top down approaches on the other hand start with the electricity demand at connection point and try to model it without knowledge of appliances.
The advantage of bottom up approaches compared to top down is that the electricity demand of the various appliances is known. Demand response and active demand strategies can be tested more easily. The downsides of bottom up approaches are the intensity of modelling and the risk of missing appliances to model [78].
Bottom up
Capasso et al. [79] base their household load model on two types of functions: ‘behavioural’ and ‘engineering’. The ‘behavioural’ functions include a probability of being at home, home activities, appliance ownership, appliance use and human resources. Appliance’s mode of operation, the maximum power at connection
point and technological penetration are ‘engineering’ functions. The sum of the individual loads of the appliances make up the electrical power demand. The time resolution is 15 minutes.
A dataset of 1200 homes where 175 households were equipped with appliance monitoring infrastructure, is the basis of the load modelling done by Stokes [80]. The electricity demand during the year of each appliance is modelled by sinus- functions. The appliance use is scaled by a weight determined by the number of inhabitants. Daily patterns ensure that appliances are started at the appropriate time. Appliance load cycles with a resolution of 1 minute are used to build electricity demand. The washing machine model for example uses 3 load cycles selected at random each time a washing machine is started. Both active and reactive power are modelled.
Energy demand (thermal and electrical) in households have two determinants according to Yao et al. [81]: behavioural and physical. The approach is hence similar to Capasso et al. [79]. The composition of the household and the occupancy pattern together with appliance ownership and usage constitute the behavioural determinant. The physical determinants are energy consumption of the various appliances and the energy consumption of domestic hot water and heating.
Widén et al. [82] use activity schemes to model electricity and hot water demand. An activity describes what a person is doing, vacuum cleaning for instance. The model for daily household activities is based on a diary of 431 persons in 169 households, filled in for one weekday and one weekend day between August and December by each person. An average power demand (on an hourly basis) is associated with each activity. The results are validated against two data sets: five households measured in a 10 minute interval during autumn and the aggregation of a dataset of 217 submetered households.
Markov chains are used to model the occupancy in households [83]. The occupancy models are combined with daily activity patterns to generate switch- on events [84]. The events trigger appliance load cycles. Most appliances have an assumed constant power, except for those with time-varying demand such as washing machines. No explicit explanation of the specific load cycles of time- varying appliances is given. The sum of the power demand of all appliances is the power demand at the connection point of the dwelling. The model is tested against 22 households, shows a slight underprediction of the variation between dwellings, an annual mean daily demand comparable to typical household profiles and an aggregated load duration curve similar to that of the measured profiles.
Top down
McLoughlin et al. have a top down approach to model residential electrical power demand [85]. The electricity load profiles of five households are modelled by a homogeneous Markov chain. The model is able to recreate the distribution of the electrical power. However, the autocorrelation, common in load profiles, could not be reproduced.