The geography of domestic energy consumption Anastasia Ushakova PhD student at CDRC UCL Ellen Talbot PhD student at CDRC Liverpool
Some important research questions How can we classify energy consumption over time, alone or as a part of lifestyle or household activities? Can we predict energy consumption under a range of what if? scenarios? How smart meter data both inform and respond to the decisionmaking in energy policy? Is it possible to understand aggregate daily mobility patterns through household energy usage?
Motivation and Research Context Energy sector continues to grow despite increasing energy efficiency Patterns of energy consumption correspond with household activity/movement patterns We do not fully understand the interactions between dwelling attributes and household characteristics The dynamics of energy consumption are integral to lifestyle analysis An understanding of consumption patterns can usefully inform energy savings initiatives Energy consumption can inform us about vulnerable and customers at risk of fuel poverty
Smart Meter Data What makes these consumer data special? Complete consumer records pertaining to fixed locations Fine temporal granularity of measures Direct correspondence with a single household Acceptable spatial granularity for linkage to administrative data Major resource for spatio-temporal data mining
Understanding energy consumption Spatial, Temporal and Social Determinants of Energy Use Financial Incentives Consumption Dwelling Characteristics Income and Wealth Household Type and Size Health Geography and Culture Society and Behaviour
Smart Meter Data Presentation Two ways to represent the time series sequence for energy data: look at intra-day consumption (i.e. how does energy usage depend on hour of day) tend to be non stationary or locally stationary Or inter-day consumption (how does usage vary across days) weakly stationary
Is there even a typical profile? Average temporal profile for owner-occupied homes in England (2010-11) Source: UK Housing Energy Fact File (2012)
Energy consumption variation While analyzing energy it is important to account for the decisions which drive energy consumption The decisions can be different at each t for each of the energy source The combination of electricity and gas consumption may aid our understanding how one affect another and how both respond to time..
Our data set
Some very simple investigations Example of national variation in the average annual gas consumption at 18:30 Wh
Effect of different spatial granularity Example of national variation in the average annual gas consumption at 18:30 OA sample Wh Postcode Sector sample
Attempting to group the patterns (annually) Cluster 1: 51% (average and stable) Cluster 2: 1 % (very high and variable users) Cluster 3: 38% ( high-less variable) Cluster 4: 10% (high and variable)
Cluster 1: 51% (average and stable) Cluster 2: 1 % (very high and variable users) Cluster 3: 38% ( high-less variable) Cluster 4: 10% (high and variable)
Consumption on a regular winter day..
We are picking up more differences in variation when we consider different time Cluster 1: 4% (very Low users) Cluster 2: 33% (average-high users) Cluster 3: 50% (low-stable users) Cluster 4: 6% (high users) Cluster 5 : 7% (average with tendency for higher usage)
Cluster 1: 4% (very low users) Cluster 2: 33% (average to high users) Cluster 3: 50% (low but stable users) Cluster 4: 6% (high users) Cluster 5 : 7% (average with tendency for higher usage)
How about Christmas?
Looks like people in 17 postcode sectors are consuming quite randomly.. On average consumption is not much higher for all clusters compared to weekday winter day but very consistent throughout the day
January weekday gas consumption intensity(total per day) Christmas day gas consumption intensity(total per day)
Could we look only at those who may stay at home?
Consumption outside peak hours: Cluster 1 almost no consumption to Cluster 4 high consumption levels which may mean that people are at home
Out of peak hours winter gas consumption differences by ONS classification super group in Bristol 1 - Rural Residents 3.2% 2 - Cosmopolitans 2.9% 3 - Ethnicity Central 0.4% 4 - Multicultural Metropolitans 9.0% 5 - Urbanites 28.8% 6 - Suburbanites 33.5% 7 - Constrained City Dwellers 4.5% 8 - Hard-Pressed Living 17.6%
Out of peak hours winter gas consumption differences by ONS classification group in Bristol 1a - Farming Communities 0.5% 1b - Rural Tenants 1.8% 1c - Ageing Rural Dwellers 1.0% 2a - Students Around Campus 1.4% 2b - Inner-City Students 0.0% 2c - Comfortable Cosmopolitans 0.5% 2d - Aspiring and Affluent 0.4% 3a - Ethnic Family Life 0.2% 3b - Endeavouring Ethnic Mix 0.0% 3c - Ethnic Dynamics 0.0% 3d - Aspirational Techies 0.0% 4a - Rented Family Living 6.7% 4b - Challenged Asian Terraces 0.8% 4c - Asian Traits 1.5% 5a - Urban Professionals and Families 20.5% 5b - Ageing Urban Living 8.6% 6a - Suburban Achievers 8.5% 6b - Semi-Detached Suburbia 25.2% 7a - Challenged Diversity 3.1% 7b - Constrained Flat Dwellers 0.0% 7c - White Communities 0.9% 7d - Ageing City Dwellers 0.5% 8a - Industrious Communities 5.6% 8b - Challenged Terraced Workers 2.2% 8c - Hard-Pressed Ageing Workers 5.1% 8d - Migration and Churn 4.9%
SPECTRUM classification Total per day consumption (up right) Out of peak hours consumption differences (bottom right) 1a Affluent Comfort 2.4% 1b Affluent Professionals 2.8% 1c Affluent Established Families 11.9% 1d Affluent Aspiring Families 4.8% 2a City Skilled 1.2% 2b City Workers 5.2% 2c City Students 0.5% 3a Modest Stability 15.6% 3b Modest Young Families 6.9% 3c Modest Opportunities 8.7% 3d Modest Blue Collar 8.1% 4a Deprived Workforce 6.8% 4b Deprived Estates 1.4% 4c Deprived Strugglers 0.6% 4d Deprived Elderly 4.1% 5a Multiethnic Black/Diverse 2.3% 5b Multiethnic South Asian 1.1% 6a Secluded Village Life 6.9% 6b Secluded Farming 0.0% 6c Secluded Ageing Prosperity 4.0% 6d Secluded Retirees 4.8%
House size measured by number of bedrooms and energy consumption on a winter day in Bristol Note: the fitted line is locally weighted regression fit
Age, house size and energy Some relationship exists But we also may have a bias towards customer base?
Some thoughts and conclusions Smart meter data is an example of a new form of data Aggregation may not necessarily be informative for spatio-temporal analyses Market segmentation is challenging compared to some other consumer data (i.e. loyalty card). Vast literature on clustering electricity exist but still no defined method.. Time chosen for the analysis plays an important role for characterizing consumers Census data may be helpful for the understanding of the customer base when there is no additional data Main limitations are associated with sample size relative to that of chosen geography (i.e. Postcode Sector or Census OA) and validation processes Other datasets available such as Energy Performance Certificates may be useful to understand the covariates of energy variation at first place Finally, is there less where there is more? Big data = big uncertainty
Thank you Anastasia Ushakova PhD student at CDRC UCL Anastasia.ushakova.14@ucl.ac.uk Ellen Talbot PhD student at CDRC Liverpool E.Talbot@liverpool.ac.uk We thank the advice we received on this work: Cheshire, J., Lansley, G., Longley, P., Mikhaylov, S., Murcio, R., and Singleton, A