Evaluating the role of behavior and social class in electric vehicle adoption and charging demands

The code supporting the current study has not been deposited in a public repository because they have been developed with support from a commercial partner. Requests for access may be made to the lead author, but will be subject to funder approval for the first 12 months following publication.

Understanding electric vehicle (EV) adoption rates and charging patterns is critical in enabling grid operators to maintain quality of supply and offers the potential to procure network services and avoid or postpone capital investments. Agent-based models have separately been shown to be useful in modeling EV adoption, policy options, behavioral influences, and grid impacts. In this work, we bring together these threads with real world travel data to present a multi-scale, behaviour-based EV adoption and use model able to replicate historical changes in vehicle fleets and match the most recent real world EV charging profile data. We have shown how our model can be used to simulate the impact of policies and consumer behavior on the rate of EV adoption across socio-economic groups and the locational grid impacts of EV charging, and as such we believe it to be of value to policy makers, grid operators, and demand response aggregators.

Introduction

Governments around the world are seeking to reduce emissions from the transport sector in line with national and international commitments. The deployment of plug-in electric vehicles (EVs) has the potential to reduce carbon emissions and also to improve local air quality through reduced particulate and NOxamongst others (Department for Business Energy and Industrial Strategy, 2017; Hampshire et al., 2018). The EU has imposed manufacturer fleet-based average emissions limits reducing to 95 g CO2/km by 2021 (European Commission, 2017), a target requiring some degree of electrification. In China, the government’s ‘New Energy Vehicle Mandate’, similar to that employed in California, provides a strong incentive for manufacturers to develop electrified models; China is currently the largest EV market globally with some 47% of all EVs on the road (International Energy Agency, 2020). As of March 2020, 16 countries (12 European) had taken various actions to phase out Internal Combustion Engine (ICE) cars (Burch and Gilchrist, 2018). The UK Government, for example, proposed a target of 2040 for a ban on the sale of pure ICE cars in 2018 (Burch and Gilchrist, 2018), but in 2020 announced plans to bring this forward to 2030. Despite these proposals, there are few studies that seek to forecast individual country EV adoption and those that do tend to pre-date the introduction of EV models by many car manufacturers, which provide vehicles at lower cost points across a wider range of market segments. In the case of the UK, the only recent work in this area appears to be that of National Grid ESO (NGC), the electricity system operator, where various forecast cases are provided in the Future Energy Scenarios report (National Grid ESO, 2020).

EVs are substantially more efficient from battery to wheel than the equivalent tank-to-wheel efficiency of ICE vehicles and thus the electricity required for charging is likely to be some 30% of the petrol/diesel needs of the current fleet (Hass et al., 2014). While spare generation capacity exists in most grid systems, unmanaged charging of EVs could be problematic. Conversely, managed charging could assist grid operators to avoid wind and solar curtailment at times of high generation, a policy that would enhance the environmental credentials of EVs. Within local distribution systems EVs could also be problematic, with the potential to significantly increase local demands, overloading cables and transformers within low voltage systems (Papadopoulos et al., 2012; Richardson et al., 2013). Within the UK, the typical power rating of residential EV chargers is 7.2kW compared to a design After Diversity Maximum Demand (ADMD) of 1.5kW for typical households. A UK joint industry and academia project, ‘Customer-Led Network Revolution’ (Barteczko-Hibbert, 2015), including a small sample of EV drivers, showed that ADMD can vary between different social groups and explored the impacts that EVs and heat pumps might have. This work notes that the number of customers on the circuit has a significant impact on the extent to which demand is smoothed, but suggests an ADMD of around 2kW with 100 customers.

There have been many studies exploring the impacts of EV charging on distribution networks; from those using simplified assumptions about journey completion times, with some stochastic analysis overlaid (Papadopoulos et al., 2012) through averaging of travel survey data (Ihekwaba et al., 2017) to the use of agent based models (ABMs) generating their own probabilistic travel patterns based on survey data (Torres et al., 2015; Olivella-Rosell et al., 2015). ABMs are a class of simulation which attempt to model individual consumer behavior and aggregate this to reveal emergent behavior across a population. However, charging studies to date have not taken into account the socio-economic status of locales within the network, identified as relevant in the ‘Customer-Led Network Revolution’ project (Barteczko-Hibbert, 2015), and have not implemented EV adoption models to any great extent, relying instead on assuming various levels of penetration.

Shafiei et al. (Shafiei et al., 2012) developed an Alternative-Fuel Vehicle (AFV) growth ABM using consumer choice input data to parameterize their agent decision algorithm. While this model incorporates elements of peer communication and range anxiety, it assumes that purchasers are able to undertake total cost of ownership (TCO) calculations and that this forms a key element of the purchase decision. In practice, studies by Axsen et al. (Axsen et al., 2013) and Schuitema et al. (Schuitema et al., 2013) show that social influence, hedonic and symbolic vehicle attributes play a significant influencing role in car purchase choices. Eppstein et al. (Eppstein et al., 2011) add further social influence in their ABM, which focuses on plug-in hybrid vehicles (PHVs), through a parameter designed to adjust individual agent’s susceptibility to media and to their social network, which is selected based on homophily criteria with other agents. The model also assumes that some owners make rational assumptions of fuel costs while others do not, thus reducing the reliance on an accurate TCO in the decision making process. For simplicity, the model assumes uniform daily driving patterns and assumes daily recharging is always available. This degree of simplification is unlikely to be suitable for pure battery-electric vehicle (BEV) adoption given range anxiety and need for charging infrastructure. Krupa et al. (Krupa et al., 2014) confirm this in a consumer survey focusing on PHVs in which 77.8% of respondents noted that electric driving range was not limited unlike that of a BEV; despite this, study participants reported that the availability of public recharging infrastructure would have a positive influence on their purchase decision. The study also notes that the value of future fuel savings is probably insufficient to persuade most consumers to pay the additional up-front cost and that rational financial analysis is rarely applied to vehicle purchase decisions. Indeed, while the survey response indicated some 69.7% regarded seeing other similar vehicles on the road as having no influence, later questioning indicated that participants would only consider a purchase after PHVs had reached a certain level of penetration, suggesting social influence is important. Vehicle segment was also found to be important, with drivers of larger cars generally unwilling to trade down where PHVs were not available in the same segment as their existing vehicle.

A 2016 study by Adepetu, Keshav and Arya (Adepetu et al., 2016) focused on San Fransisco; their consumer agents include a ‘greenness’ variable to represent a person’s tendency to incorporate lower carbon footprint into their purchase decision. Their model also incorporates a function representing the ability of individual agents to accurately estimate TCO of a vehicle and thus the relative benefits of lower running costs for EVs vs. lower initial capital cost for conventionally fueled vehicles. The study employs temporal and spatial information based on National Renewable Energy Laboratory survey data of 366 San Francisco residents covering work day and non-work day driving patterns, with simulated routes within the city. This enables the impact of charging station location to be considered and, on the assumption that there is immediate un-controlled charging when available, predicts the impact on electricity demand at various locations. However, the adoption element of the model includes only private vehicle ‘cash’ purchases whereas some 59% of new car purchases are made by fleet buyers in the UK (DVLA, 2019); only two models of EV are included and no long-distance trips are simulated, thus, users do not experience range anxiety. Ahkamiraad and Wang (Ahkamiraad and Wang, 2018) have presented an agent-based EV adoption model that aims to simulate adoption at zip-code level in New York. Each zip-code is an agent with differing adoption rates based on its population characteristics and communication also occurs between neighboring areas. An adoption and annual energy consumption spatial model is developed, but this does not resolve to half-hourly demand data, which is essential in planning network capacities.

The ‘Consumat’ model, which was first put forward by Jager, Jannssen and Vick (Jager et al., 1999) in 1999, brought together various theories relevant to understanding consumer behavior in a methodology applicable to ABM. A Consumat is an agent within an ABM that participates in four processes: deliberation, social comparison, imitation, and repetition of previous actions. The Consumat model was further updated by Jager and Jannssen in 2012 (Jager and Janssen, 2012), this introduced variability in the capability of agents to, for example, assess the utility of a purchase decision.

Kangur et al. (Kangur, 2014; Kangur et al., 2017) have developed an agent-based simulation for diffusion of EVs known as STECCAR (Simulating the Transition to Electric Cars using the Consumat Agent Rationale) based on a Consumat model. Consumats here are car drivers who each week evaluate their four needs–financial, functional, social, and environmental–against the performance of the vehicle. The primary focus is to satisfy financial and functional needs, but where possible to optimize social and environmental needs. Each agent has a different set of personal attributes which determine whether the agent is satisfied and certain. Here ‘certain’ means that they are comfortable in their mental state, for example, in regard to how they fit into their social group. The evaluation of needs and mental state results in four different agent conditions as illustrated in .

The STECCAR model evaluates purchases of the three main car groups (Internal Combustion Engine (ICE), PHV, and BEV) and allows agents to purchase vehicles from the used car market as some owners dispose of their vehicles. The authors have also validated the model against a number of market metrics including the rate of turnover of vehicles (used car market), duration of vehicle ownership, average vehicle age and market penetration. This model does not however take into account brand or segment loyalty and does not attempt to project AFV uptake onto electricity network demands.

Numerous studies have shown the effects of brand loyalty on purchasing decisions and May (May, 1971) showed the degree of brand loyalty in relation to Californian car purchasing is also related to social status while Danish et al. (Danish et al., 2018) showed how a number of aspects of brand are correlated with purchasing decisions. Although there are no recent UK academic studies specifically looking at brand loyalty and car purchase choices, a 2017 industry survey by AutoTrader (AutoTrader/AM-Online, 2017) suggests around 25% of volume-brand purchasers repeat buy but that fell to 16% for premium brands. The same survey showed that buyers exhibit even greater body-type (segment) loyalty, with an average of 41% repeat purchasing. The survey also noted an increase in SUV and 4×4 repeat purchasing but a reduction in super-mini repeat purchasing; this may thus, in part, reflect a general desire to purchase higher status (typically more expensive) vehicles. These levels of brand and segment loyalty are significant given that EVs, in the early years of deployment, have not been available across all segments and brands.

In this study, we bring together these different strands to develop an ABM, the Behaviour-based Electric Vehicle Grid Integration (BEVI) model that is able to simulate the uptake of AFVs, including HEVs (non-plugin hybrid-electric), PHVs, and BEVs, taking into account social interactions and limited TCO calculation ability. This work extends beyond existing ABM EV modeling by combining a behavioral model, incorporating charging knowledge acquisition, with a representative suite of vehicles and detailed travel diaries to provide a simulation able to both project EV adoption at a societal and social group level together with evolving impacts on the electricity grid on highly granular time and geographic scales. Applying this model to the UK market as an exemplar, we investigate the impacts of brand and segment loyalty and how availability of price parity with ICE vehicles and larger ranges impact on diffusion rate. We show that such a model can be effectively used to forecast the impact of AFV uptake on electricity demands both nationally and critically at a local distribution level taking into account socio-economic factors.

The methodology behind the BEVI model, and the datasets employed are described in the STAR Methods section. In the Results and Discussion section, we first analyse the performance of the model and subsequently explore the impacts of tax policy, brand loyalty, range, and pricing. We also validate the charging demand against recent real-world data and show how the model can be used to forecast demands in different locations. We finish with general conclusions, areas for further work and a review of the study limitations.