Spatial Modelling of Electric Vehicle Uptake and Electricity Grid Impacts in Australia

Spatial Modelling of Electric Vehicle Uptake and Electricity Grid Impacts in Australia Phillip Paevere Principal Research Scientist CSIRO Australia Phillip.Paevere@ csiro.au Andrew Higgins Principal Research Scientist CSIRO Australia Andrew.Higgins@ csiro.au Summary Wide spread adoption of electric vehicles (EVs) in the form of battery electric vehicles (BEV s) and plug-in hybrid electric vehicles (PHEV) is anticipated in Australia over the next 30 years. There are potentially significant benefits of a shift to elctricfied passenger transport in terms of emissions, air pollution, energy security, and the ability to better utilise the electricity network. This paper presents an innovative diffusion model for spatially projecting the uptake of electric vehicles. The model incorporates features of multi-criteria analysis and choice modelling, to estimate the adoptions of different competing passenger vehicle options across a landscape of heterogeneous consumers in Victoria. Seven criteria were included in the modelling: performance, upfront cost, annual cost, household income, demographic suitability, driving distance and risk. Annual change of criteria values and their elasticity to adoption were incorporated. The model was partly calibrated using data from a large scale survey conducted in Victoria. In this report we test the diffusion model on the vehicle stock across all 1.5 million households in Victoria, Australia, to forecast market share of BEV s, PHEV s, traditional hybrid electric vehicles (HEV s) and internal combustion engine vehicles (ICE s) through to 2030. Geographical differences in uptake of BEV s were observed, and these are primarily due to driving distance, demographic suitability, and household income, with urban areas having about three times the proportional uptake of BEV s compared to regional areas. This urban/regional variability was not as strong in the PHEV projections, primarily due to the fact that the range of PHEV s is not limited by battery size. By testing the model for rebate incentives of $5000 and $10000 for the upfront cost of a plug-in EV, along with technological improvements to increase the EV range, we demonstrate its capability to inform and optimise policy options. Keywords: electric vehicles, choice model, multi criteria analysis, electricity distribution network 1. Introduction Most of the world s major vehicle manufacturers are currently developing, or have plans to develop plug-in electric vehicles (EV s) for the mass market. Widespread adoption of these vehicles is anticipated in Australia over the next 30 years [1]. Two types of plug-in EV s are coming to market soon, or are on the horizon: battery-only electric vehicles which run entirely on a battery charged from the grid (BEV); and plug-in hybrid electric vehicles which can run on batteries charged from the grid alone (PHEV), liquid fuel (petrol), or a combination of both. Given typical EV battery sizes, and driving patterns in Australia, these two types of EV s used for passenger transport may require somewhere in the order of 2kWh to as much as 20 kwh of electrical energy from the grid to meet their driving distance requirements [2]. This potentially represents a significant extra electrical load on the grid which needs to be planned for in the future. Conversely, the electrical energy storage capacity available in electric vehicles, when not needed

for transportation, can potentially be harnessed and used to support the needs of the electricity grid at times of high demand, or constrained generation. In order to plan for the extra load from EV s, and to understand the impact and opportunities of extra loads and storages on an electricity distribution system with temporally and spatially variable capacity, it is critical to be able to project the magnitude, rate, and location of EV uptake by consumers at a fine spatial scale. This paper presents a methodology, based on diffusion modelling, for estimating the uptake of electric vehicles, at an appropriate spatial scale for determining the impacts on individual feeder lines in the electrical distribution system. Our diffusion methodology to analyse future uptake of EVs in Victoria, incorporate several enhancements to existing diffusion approaches [3,4]. Firstly we incorporate a highly granular differentiation of location by multiple demographic variables. This will allow future capacity planning of the electricity network in light of increased peak demand for charging. To accommodate and trade off a large number of financial and non-financial features of EV s, we also extend the multi-criteria analysis approach of the single product model in [5] to a multi-product model. Our method also captures the dynamic relationship between some features and market share, such as the case where consumer risk and some costs are reduced as the adoption (or market share) of the vehicle increases. Lifespan of a vehicle is considered in our model, after which it is replaced with a product amongst the competing options. Whilst EV s are the application in this paper, the methodology is adaptable to other applications for diffusion of new product options that are replacing a technology that is being phased out. In the context of improving energy efficiency, it is a problem faced by consumers when selecting a hot water system (instant gas, tank gas, solar to replace electric) or more efficient light bulbs and appliances such as airconditioners. 2. Description of diffusion model Our model combines features of choice modelling, multi-criteria analysis and diffusion models. This provides a capability to analyse adoption patterns of the competing vehicle options under a range of features that a buyer would consider for a purchase. In the generic model, the total stock of all vehicles (at each location and for the entire region under consideration) is known over time but the market share of the different competing vehicle options is to be estimated. A household will purchase a vehicle amongst the competing options available, when they are ready to replace their existing vehicle. Lifespan of a vehicle is considered in the model, after which it is replaced with a product amongst the competing options. The adopter can be a household, individual or vehicle. We will use the term vehicle for this paper, since a household can have more than one vehicle. Inputs to the model include parameters which represent the following: different categories of building by location by demographics. the set of competing vehicle types (e.g. electric, hybrid, diesel) the set of discrete time intervals in the planning horizon. Intervals are 3-monthly for this paper. expected life span of each vehicle option, in terms of number of time intervals. forecast of new households that will purchase a vehicle in each time interval. This represents additional vehicle stock that was not present during the previous time period. total market share of stock of each vehicle option in each demographic by location category at the beginning of the planning horizon. The main outputs from the diffusion model are: total market share of stock of each vehicle option at a given time interval market share of each vehicle option for each demographic by location category at each time interval.. Full details of the model and it s component equations are described in [5,6].

3. Electric Vehicle Adoption Case Study Victoria, Australia Victoria is a State of Australia with 5.2 million people (2006) and a residential vehicle stock of 2.21 million. Data was gathered from multiple sources to set up the model parameters. The first step was to construct a typology to represent all possible categories. To do this, information from the Australian Bureau of Statistics (ABS) was available at a granular spatial scale, called census collection district (CCD) where each CCD represents about 250 households. Victoria contains 9300 CCD s which define the set of geographical locations of the analysis. Each location was partitioned via some key drivers that impact adoption of the vehicle combinations: housing type by ownership by household income by number of vehicles. For each combination, the number of households was extracted using the ABS TableBuilder on-line tool. To avoid having an extremely large number of categories and due to privacy restrictions of identifying individuals in the ABS data, the following categories were used: Housing type Detached house, Other (including townhouses and apartments) Ownership Own (including mortgage), Rent Household income - $0-$30000/yr, $30000-$75000/yr, >$75000/yr Vehicles -0,1,2+ Four different vehicle options were considered: BEV; PHEV; HEV; and ICE vehicle. Seven criteria were identified as essential drivers for vehicle choice and where there were sufficient data to populate and calibrate. These are: Performance; Annual costs (including electricity and fuel); Upfront cost; Household income; Driving Distance; Demographic suitability; and Risk (including hassle). Based on CSIRO surveys of consumers in Victoria (awaiting publication), it was shown that individuals in household incomes of >AU$75000/yr would pay a higher upfront price for a vehicle compared to the lower income categories. Based on the Oliver Wyman Study E-Mobility 2025 (www.oliverwyman.com), we assumed the annual change in EV costs to be -3%, whilst - 1.5% and 0% for Hybrid and ICE vehicle respectively. For the Demographic criteria, an EV demographic suitability score was generated for each location (CCD) in Victoria using several ABS variables for each location. The variables used were: Age (age group categories); Number of residents in the household, Employment (full time, part time, unemployed); Education (high school, diploma, degree etc); Occupation (career categories), and Transport mode (car, train, bus etc). For each of these variable categories, an EV suitability score was allocated, based on suitability categories inferred from the survey. A planning horizon of 20 years (2011 to 2030) was used for the case study. The initial vehicle stock was set to the values in the 2006 ABS data, since updated values by CCD are not available until late 2012. For this current report, the initial market share of HEV s was 3% (2010), and the remaining market share was ICE. For households that do not have a vehicle, we assumed they remained without a vehicle throughout the planning horizon. Our case study is based on a static population and demographic breakdown through to 2031. 4. Indicative Results This section is divided in two parts. Firstly the base case adoption trends and market shares are shown, including the implications across different parts of Victoria. The second part demonstrates the capability of the model to assess the adoption sensitivity to price changes and incentives. For all analysis, Fortran 95 was used to code the model under a PC Windows XP environment. 4.1 Uptake projections base case Figure 1 shows the overall market share of the four vehicle options, which is calibrated to the scenario in [1]. The larger upfront adoption of HEV s is due to the lower risk and costs in 2011. As the adoption of BEV s and PHEV s increased, the risk reduced substantially thus increasing the

market share by 2030. Figure 2 partitions the market share by household income categories, showing the high income household having the greatest exodus from ICE vehicles by 2030, with low income earners primarily staying with ICE s. Figure 3 shows the modelled market share of BEV and PHEV for 2030 for the greater Melbourne region. It highlights the granularity of the analysis as well as the rural to urban differences in adoption. The biggest drivers for the rural to urban differences in uptake were driving distance (due to current range of BEV s), occupation and education. For precinct to precinct differences, the largest drivers were household income and employment status, which were often substantially different for neighbouring suburbs. In Figure 4 the spatial distribution of vehicle uptake across Victoria for 2030 is shown for all vehicle options. The rural versus urban differences of BEV uptake are very profound, with PHEV and HEV s having the greatest impact in per-urban areas. Whilst the average uptake of PEV is 12.56% (Figure 1, Table 1), rural CCD s were about 7%, whilst urban areas were up to 22%. In both Figures 3 and 4, a CCD of about 250 households will have a much greater land area in the rural areas compared to the urban areas. Whilst ICE vehicles are projected to have a significantly decreasing market share to 2030 (Figure 1), their geographical market share in Figure 4 is still dominant. 4.2 Sensitivity The aim of the sensitivity analysis is to demonstrate how the model can be used to inform future policies/incentives that may be introduced to stimulate additional uptake of EV s, and to test the impact of more rapid technological advances. Federal and state government incentives, in the form of financial rebates to up front costs, have been a major strategy over the past five years to accelerate the uptake of solar PV s and solar hot water systems [7]. In the case of Solar PV s, [1] showed that a rebate of $8000 introduced in 2009, led to more than a 120% increase in uptake by the end of 2012. Whilst, the Australian government have not yet introduced financial incentives for BEV s, we will evaluate the following hypothetical scenarios 1. Rebates of up to $5000 from 2010 to 2030. 2. Rebates of up to $10000 from 2010 to 2020. All rebates are introduced at the beginning of the planning horizon and for each scenario, we calculate the total cost to the government (Victorian or federal). In Table 1, a rebate, of $5000 from 2010 to 2030 had a much large impact on adoption compared to a rebate of $10,000 from 2020 to 2023, whilst only costing the government about 20% more. The likely reason is because the second 10 years (2020-2030) is when the lower upfront costs and performance improvements meant the rebate became far more attractive to prospective adopters. The next scenario tests the impact of improved upfront performance of BEV s relative to the other vehicle performance. The base performance score was 4.0, and Figure 5 compares the sensitivity of the initial score of between 0 and 10. In practice, higher initial performance scores can represent more rapid advances in battery technology for Li-ion and LiFePO4, and public EV recharging infrastructure. This would lead to significant increases in the effective range of EV s and thus the performance relative to other vehciles. When the effective range of BEV s reaches that of ICE, we assume the performance will also reach that of ICE. In Figure 5, an initial performance score of >6 leads to the performance matching that of ICE s by 2020, thus the plateau of market share these values. Overall, the effect of increased vehicle performance on market share of BEV s was significantly greater than the upfront costs in Tables 1 and 2.

Market Share % 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024 2025 2026 2027 2028 2029 2030 Market Share % 100 90 80 70 60 50 40 30 ICE HEV PHEV BEV 20 10 0 Fig. 1 Overall market share trend from 2011-2030 100 90 80 70 60 50 40 30 BEV <$30000 BEV $30000 to $75000 BEV >$75000 PHEV <$30000 PHEV $30000 to $75000 PHEV >$75000 HEV <$30000 HEV $30000 to $75000 HEV >$75000 ICE <$30000 ICE $30000 to $75000 ICE >$75000 20 10 0 Year Fig. 2 Market share of vehicle options partitioned by household income

0% BEV 25% 0% PHEV 30% Fig. 3 Market share of BEV s and PHEV s at 2030 at the suburb-scale for metropolitan Melbourne

0% BEV 25% 0% PHEV 30% 0% HEV 45% 0% ICE 100% Fig. 4 Market share of vehicles across Victoria at 2030, top =EV, middle=hybrid, bottom= ICE Table 1. Market share (%) for rebates of up to $5000 from 2010 to 2030 Rebate BEV PHEV HEV ICE Total Cost $0 12.56 17.10 17.99 52.34 0 $1,000 13.14 16.92 17.81 52.13 $290,591,569 $2,000 13.72 16.74 17.63 51.91 $607,151,125 $3,000 14.33 16.56 17.44 51.67 $950,888,379 $4,000 14.96 16.38 17.26 51.39 $1,324,202,244 $5,000 15.66 16.21 17.09 51.03 $1,732,850,879 Table 2. Market share (%) for rebates of up to $10,000 from 2010 to 2020 Rebate BEV PHEV HEV ICE Total Cost $0 12.56 17.10 17.99 52.34 0 $1,000 12.58 17.10 17.99 52.33 $278,307,884 $2,000 12.60 17.09 17.98 52.33 $557,611,870 $3,000 12.63 17.08 17.97 52.32 $838,273,125 $4,000 12.66 17.07 17.96 52.31 $1,120,795,602 $5,000 12.70 17.05 17.94 52.30 $1,405,881,759

Market share (%) 90 80 70 60 50 40 30 20 BEV PHEV HEV ICE 10 0 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 6 Weight w j - Performance Fig 5. Sensitivity to Performance criteria weight 5. Application of Model to Electrical Grid Impacts Analysis The diffusion model described herein has been used to spatially model the future impacts of EV usage on the electrical grid, based on EV uptake rates, spatially distributed travel patterns, and different charging (and discharging) scenarios. An example of this application is given in Figure 6, which shows the impacts of EV charging on the grid in 2030 in metropolitan Melbourne, Victoria. The left hand diagram in Fig 6 represents the %increase in household peak electrical load, when EV s charge on demand (as soon as they arrive home). The right hand diagram represents the the %increase, or %decrease in household peak load, under a scenario where the EV will discharge energy back into the grid when it arrives home, so long as there is enough capacity in the battery to do so, and still meet the household travel requirements. DEMAND CHARGING Melbourne, Australia, 2030 Vehicle-to-House Charging/Discharging Melbourne, Australia, 2030 Average EV Uptake = 10% Average EV Uptake = 10% -10% 0% % Change in Peak Load +20% -10% 0% % Change in Peak Load +20% Fig. 6 Spatially projected impact of different EV-grid integration scenarios on electrical grid peak load in metropolitan Melbourne, Australia.

6. Conclusion and Further Research In this paper we implemented an innovative diffusion model, incorporating features of multi-criteria analysis and choice modelling, to estimate the market share of BEV s versus other vehicle options, across the landscape of heterogeneous consumers. A characteristic of the model was its ability for a highly granular geographical by demographic analysis, allowing adoption rates to be assessed at a sub-precinct level. This provides a useful capability for energy providers to better understand capacity constraints across their electricity grid as adoption of BEV s increase at different locations. A case study of 9300 locations across all of Victoria from 2011 to 2030 as used to show high spatial resolution uptake of BEV s and PHEV s compared to HEV s and ICE vehicles, along with the precinct to precinct and rural to urban differences. Key drivers of these differences were driving distance, employment status and household income. By testing the model on scenarios of government financial rebates and BEV range improvements, we demonstrated the powerful capability of the model to inform or optimise various government policy schemes targeted towards increasing adoption of BEV s at minimal cost. In the future, the model described herin will be used to simulate the impacts of various BEV uptake scenarios on the electricity distribution system, and to model the effects of various BEV policy options identified in Supporting Electric Vehicle Adoption in Australia project [8]. Policies to be considered include: developing standards for BEV s and charging equipment in priority areas; time of use pricing; price on carbon emissions; and purchase price and operational subsidies such as green registration discounts. 7. References [1] Graham P., Reedman L., And Poldy F., Modelling of the future of transport fuels in Australia, A report IR1046 to the Future Fuels Forum. www.csiro.au, 2008. [2] Usher J.,, Horgan C., Dunstan C.,, Paevere P., CSIRO Electric Driveway Project: Technical, Economic, Environmental & Institutional Assessments Phase 1 Report, Prepared for Australian Commonwealth Scientific and Industrial Research Organisation (CSIRO), by the Institute for Sustainable Futures, UTS: Sydney, 2011. [3] Lin Z., and Greene D., Who will more likely buy PHEV: A detailed market segmentation analysis, The 25 th World Battery, Hybrid and Fuel Cell Electric Vehicle Symposium & Exhibition, Shenzhen, China, 2010. [4] Axsen J., Mountain D.C., and. Jaccard M., Combining stated and revealed choice research to simulate the neighbour effect: The case of hybrid-electric vehicles, Resource and Energy Economics, Vol. 31, 2009, pp. 221 238. [5] Higgins A., Foliente G., and McNamara C., Modelling intervention options to reduce GHG emissions in housing stock a diffusion approach, Technological Forecasting & Social Change, Vol. 78, 2011a, pp. 621 634. [6] Higgins A., Paevere P., Gardner J., and Quezada G., Diffusion of competing vehicle options across a landscape of heterogeneous consumers, Technological Forecasting & Social Change, 2011b, Under review. [7] Australian Government. Renewable Energy Bonus Scheme Solar Hot Water Rebate. Department of Climate Change and Energy Efficiency. http://www.climatechange.gov.au/government/programs-and-rebates, 2010. [8] Dunstan C., Usher J., Ross K., Christie L., Paevere, Supporting Electric Vehicle Adoption in Australia: Barriers and Policy Solutions (An Electric Driveway Project Report), Prepared for Australian Commonwealth Scientific and Industrial Research Organisation (CSIRO), by the Institute for Sustainable Futures, UTS: Sydney, 2011.