Impacts of charging behavior on BEV charging infrastructure needs and energy use

Battery electric vehicles (BEVs) are vital in the sustainable future of transport systems. Increased BEV adoption makes the realistic assessment of charging infrastructure demand critical. The current literature on charging infrastructure often uses outdated charging behavior assumptions such as universal access to home chargers and the ‘‘Liquid-fuel’’ mental model. We simulate charging infrastructure needs using a large-scale agent-based simulation of Sweden with detailed individual characteristics, including dwelling types and activity patterns. The two state-of-art archetypes of charging behaviors, ‘‘Plan-ahead’’ and ‘‘Event-triggered’’, mirror the current infrastructure built-up, suggesting 2.3–4.5 times more public chargers per BEV than the ‘‘Liquid-fuel’’ mental model. We also estimate roughly 30–150 BEVs served by a slow charger may be needed for non-home residential overnight charging.


Introduction
Electric vehicles (EVs) will play a vital role in the sustainable transformation of future transport systems.Global EV sales reached 6.75 million in 2021, doubling the sales number of 2020.The share of EVs in newly sold light-duty vehicles in 2021 reached 8.3%, most of which were battery electric vehicles (BEVs) (71%) (EV-volumes.com, 2022).European Union has passed an Executive Order that bans the sale of fossil-fuel cars by 2035 (electrive.com, 2022).Despite the need to increase the use of (electrified) public transport, biking and walking, driving private vehicles are still likely to be an important part of future travel.A future scenario of 100% BEV adoption will require careful consideration and infrastructure planning to accelerate the adoption of BEVs.
The infrastructure deployment should consider realistic charging behaviors (Chakraborty et al., 2019), which enables travelers to maximize their desired daily activities (Metais et al., 2022).When designing the location and sizing of charging infrastructure, oversimplified assumptions are typically made in the literature about charging behaviors and access to home chargers.With 100% BEV adoption, people living in all types of dwellings will demand charging.How much public charging will be needed remains a key policy question.
This study brings insights into charging infrastructure planning using state-of-the-art understanding of BEV charging & discharging dynamics, and the corresponding quantification of charging demand for BEV owners in different dwellings.The study's unique contribution is to deploy an advanced large-scale synthetic population in which we preserve realistic socioeconomic attributes and heterogeneous activity plans and implement a unique taxonomy of charging behaviors.Our results demonstrate that the spatiotemporal use patterns and the number of charging points at home, work, and other places for private BEV owners depend on the consideration of different charging strategies.
Y. Liao et al.

Related work
Charging infrastructure is vital to support the diffusion of BEVs.However, deploying charging infrastructure is expensive and faces the chicken-and-egg dilemma between infrastructure operators (investment costs and profitability) and BEV consumers (charging demand) (Metais et al., 2022).Optimal locations and sizing of charging points are crucial to furthering BEV adoption and guaranteeing a good driving experience for BEV users without compromising their planned activities.The taxonomy of BEV charging options to fulfill BEV users' planned activities can be based on technology, e.g., non-contact charging, power rating, e.g., slow, intermediate, fast, and even ultra-fast charging (Suarez and Martinez, 2019), occasion, e.g., destination charging (Schmidt et al., 2020), etc.
Charging behavior is essentially a spatiotemporal concept, i.e., when and where BEV users access a specific charging point and for how long.Highlighting the behavioral aspect, we categorize the main charging options of BEV users into three groups: (1) overnight charging (slow e.g., 3-11 kW) at a home garage or non-home residential parking spot, (2) daytime charging (slow to intermediate e.g., 7-22 kW) at workplaces, and other places such as shopping malls etc., and (3) fast charging at public places (e.g., 50-150 kW), mostly along travel corridors to facilitate long-distance travel.Access to a home charger or a non-home residential parking spot leads to less dependence on public charging infrastructure.Recent studies suggest that 50%-80% of all charging events occur at home (Option 1) (Hardman et al., 2018).For daytime charging (Option 2), public charging infrastructure complements home charging only in densely populated areas (Funke et al., 2019), and workplace charging accounts for 15%-25% of BEV commuters' charging events (Hardman et al., 2018).Most of the evidence today is based on early adopters who are more likely to own detached houses with garages than the average population (Chakraborty et al., 2019).However, in places like the Netherlands with low availability of private parking space (Hardman et al., 2018), non-home residential parking spots are needed for charging at night.In a future of 100% BEV adoption, how much non-home charging is required for all dwellers remains an important policy question.
There are three broad categories of methods to determine the optimal locations and sizing of charging points: node-based, pathbased, and tour-based (Deb et al., 2018).Node-based and path-based have mainly been used for planning fuel stations, but they suffer from the lack of behavior dynamics and unrealistic charging speed of BEVs equivalent to refueling an internal combustion engine car (Metais et al., 2022).Tour-based (activity-based) methods more realistically reflect driving behaviors and charging needs (Metais et al., 2022) by using data at the level of individual car trajectories.These trajectories come from detailed real-world driving records (Gnann et al., 2018) or simulated data from agent-based models (ABMs) (Zhuge and Shao, 2018;Márquez-Fernández et al., 2021).
Most studies to date do not account for realistic charging behaviors (Patil et al., 2022) and varying access to a home charger.For example, some studies locate the candidate charging stations at today's fuel stations (Zafar et al., 2021).Other studies assume that BEV users will only charge when the state of charge (SOC) drops below a certain threshold (Wang et al., 2019;Kong et al., 2019).SOC ranges between 0 and 1, indicating how full the battery is with energy (1 = fully charged).These assumptions are based on our understanding of how we refuel ICEVs despite the distinct characteristics of BEVs.For instance, BEVs need a significantly longer time than ICEVs to charge.Therefore, charging behaviors would differ in how drivers plan their charging events to anticipate planned travels (Miralinaghi et al., 2020).The lack of complexity of charging behaviors in the literature is partly due to the lack of data to make appropriate assumptions, and studies generally extrapolate from early adopters who mostly have access to home chargers (Greaves et al., 2014).Future BEV users will develop different strategies for charging, depending on their activity plans, mental models, pricing models, and willingness to pay.It remains unclear how these charging strategies affect charging demand and infrastructure.Last but not least, studies that assume all private BEV owners have access to home chargers (Pan et al., 2020) find that their need for daytime charging at work and other places is minimal (Greaves et al., 2014).
Charging infrastructure planning literature is often abstracted into an optimization problem of charge point placement and sizing.Most studies aim to find the required number and location of charge points necessary to fulfill the travel demand of today's car users. 1 The optimization goal varies, including maximized distance traveled (Shahraki et al., 2015), minimized CO2 emissions (Liu et al., 2019), and other aspects such as infrastructure costs.The optimal charging infrastructure calculated from solving spatial optimization problems is often found to be minimal, far below what is being deployed in reality.The mismatch is caused by comfort, bounded rationality, and range anxiety toward, e.g., infrequent though highly valued long-distance and weekend trips, etc. (Funke et al., 2019).This mismatch partly contributes to the chicken-and-egg dilemma.A question is worth asking to shed light on solving the dilemma: what would be the charging demand if we provide BEV users with charge points when they want to, instead of need to, charge their cars?The broader diffusion of BEVs calls for the sensible integration of this BEV users' perspective in quantifying charging infrastructure demand.
ABMs can bridge the gaps in the literature by simulating realistic charging strategies, heterogeneous home charger access of a large population, and integrating charging decisions from the BEV user's perspective.ABMs are suitable for this study because they capture fundamental mechanisms of individual travel behaviors, the interactions between them and the environment, and the emergence of aggregate patterns.For instance, transport energy demand is simulated in future scenarios of a varying adoption rate of EVs (Novosel et al., 2015).For these types of problems, ABMs are superior because of their complete population coverage with socioeconomic attributes and behavior characteristics such as activities that can be modified for future scenarios (Metais et al., 2022).These characteristics empower simulating future scenarios and informing policymaking for charging infrastructure planning.

Outline of this study
This study aims to bridge the above-mentioned research gaps by applying agent-based modeling with a synthetic population of Västra Götaland, Sweden.We simulate BEV driving & charging given their planned activities for typical weekdays.We calculate the spatiotemporal patterns of charging demand considering different BEV charging strategies.We compare our results with today's charging infrastructure, highlighting the remaining infrastructure needs to support 100% BEV penetration to better inform future planning.
The remainder of this paper is organized as follows.In Methods, we describe the analytic framework including the datasets, simulation modules, and the methods for calculating charging demand.In Results, we present the outcomes of the simulation in three parts: charging demand of individuals; charging infrastructure demand; and a spatial comparison with today's charge points.We discuss the results in Discussion where we also summarize the major contributions and limitations of this paper.To complement the main body of this paper, Appendix A describes the mobility patterns of the applied car agents, while Appendix B presents the sensitivity test results.The codes are available at https://github.com/TheYuanLiao/synthetic-sweden.The main data input and output are publicly available (Liao et al., 2023).

Methods
The simulation framework is shown in Fig. 1.We use the synthetic population and their activity plans for an average weekday from the Synthetic Sweden Mobility Model (SySMo) (Tozluoğlu et al., 2022).SySMo is an agent-based decision support framework for modeling and analyzing transport scenarios.SySMo connects agents' socioeconomic characteristics with heterogeneous daily activity plans while preserving privacy.In this study, we focus on the 1.7 million residents in the Västra Götaland (VG) region where the second largest city of Sweden, Gothenburg, is located.
We use MATSim to simulate realistic daily activity plans of the agents.MATSim is an agent-based framework that provides a microscopic description of the travel demand (W.Axhausen et al., 2016).MATSim simulates agents' movement trajectories given their activity plans by optimizing agents' utility scores, which considers activity participation as positive while being late or stuck in traffic negative.We feed the agents' daily activity plans and the road network into MATSim for replanning until they converge on a set of optimal activity plans for all agents.Next, we extract the mobility trajectories of individual agents from the MATSim simulation.
The BEV simulation is implemented on the individual agents' travel trajectories considering overnight charger access, today's car fleet composition in Sweden, road network, and BEV charging & discharging dynamics.The parking times are the charging opportunities, and we do not consider rerouting to search for charging points.If an agent cannot visit all the planned places, despite charging the BEV the whole time it is parked, we define this situation as a failure.A failure can be caused by long travel distances and short parking time windows.
In the BEV simulation, we simulate three charging strategies: (1) Liquid-fuel strategy, (2) Plan-ahead strategy, and (3) Eventtriggered strategy.These charging strategies have varying SOC thresholds and conditions to start charging when the agents' BEVs are parked for more than 10 min.On the supply side, we provide three types of charging points according to the length of their parking and the SOC of their BEVs: (1) fast charger (50 kW), (2) intermediate charger (22 kW), and (3) slow charger (11 kW).We conduct extensive sensitivity analysis (see Appendix B) and summarize the results on charging demand overnight, at work, and in public spaces.

Simulating mobility trajectories
The two key inputs for MATSim to simulate the synthetic population's mobility trajectories are the car drivers with their daily activity plans, and the road network (detailed below).The configuration follows the benchmark scenario of MATSim 13.02 with minor modifications.The replanning strategy is a combination of BestScore (60%), TimeAllocationMutator (30%), and ReRoute (10%), where the percentage in the brackets indicates the share of agents who adopt these strategies. 3After 200 iterations, we see utility scores stabilize for all agents.We take the output of the 200th iteration as the trajectories for the next module, BEV simulation.Detailed input descriptions of this module are the following.

Synthetic population and their activity plans
In the 1.7 million residents in VG, we use 284,000 agents who are car users.They account for 35% of all the car users and 18% of the total population (Fig. 2a).They are proportionately sampled by the demographic statistics areas (DeSO zones) (Statistikmyndigheten SCB, 2022).Each agent has a set of socioeconomic attributes such as age, gender, income level, employment status, dwelling type, etc., together with a daily activity plan covering four activities: home (H), work (W), school (S), and other (O).Most of these car users live in detached houses (62.4%) and the rest in apartments (37.6%).More car users live in suburban areas than in Gothenburg city center (Fig. 2a).Their mobility patterns in the simulation day are summarized in Appendix A.

Road network with slope
A road network with slope is prepared for a more accurate estimation of BEV energy consumption than assuming an average energy efficiency for BEV discharging.We download the road network from GEOFABRIK (Geofabrik GmbH and OpenStreetMap Contributors, 2022) and extract the following road segments using osmosis (osmosis, 2022): all the road segments within the VG area (Fig. 2b) and all the main road segments for the rest of Sweden (Fig. 2c).The main road segments cover motorway, trunk road, and primary, secondary, and tertiary roads.
After getting the road network, we compute the road slope for the BEV simulation.For those road links longer than 500 m, we break them down into multiple connected links to ensure all processed road links are no longer than 500 m.We get the elevation of each road link's start and end points using the European Digital Elevation Model (DEM) data (Copernicus Programme, 2022).From the elevation information, we calculate the average slope of each road link.

BEV simulation
The BEV simulation concerns charging infrastructure technology, battery energy density, and BEV energy consumption.We combine the current status and near-future projections in designing a future scenario of 100% BEV adoption.From this setup, one can also test different technology development scenarios.
The BEV simulation assumes the agents drive BEVs to finish the activities on the simulation day following the trajectories from the MATSim simulation.The inputs are BEV fleet and discharging & charging dynamics, charging strategies, and charging access and initial SOC.Detailed input descriptions of this module are the following.

BEV fleet and discharging & charging dynamics
Using the data from a previous study on future EV charging infrastructure scenarios in Sweden (Márquez-Fernández et al., 2021), we consider a BEV fleet composed of three sizes of BEVs, B-(12%), C-(50%), and D-segment (38%) reflecting today's composition, and they have battery sizes of 40, 60, and 100 kWh, respectively.Another study on charging strategies for urban private vehicles in Berlin shows a similar selection of battery sizes (Jahn et al., 2020).For discharging, each BEV segment has a lookup table of energy efficiency (kWh/km) as a function of speed (m/s) and road slope (%).For charging, the power delivered by the charge point is limited by SOC, i.e., the higher the SOC, the slower the charging (Fig. 2d).The detailed characteristics of these vehicle segments, their simulated energy efficiency maps, and SOC-dependent charging profile were generously provided by Márquez-Fernández et al. (2021).
BEVs are assigned to the agents depending on their income levels.The higher the income, the larger the assigned BEV battery sizes.The proportions of BEV assigning are randomized and heuristically determined, as illustrated in Fig. 2e.

Charging strategies
We abstract three charging strategies (Sprei and Kempton, 2022): • Liquid-fuel strategy: Wait until the gauge shows low, then refill.
• Plan-ahead strategy: Plan ahead for when charging is needed.For example, when parking, think ahead to the next trip and check if BEV has enough charge for it.If not, plug in.• Event-triggered charging: Plug in to charge whenever parking at a specific location, e.g., workplace, meal stops, sometimes without considering whether stop duration only allows a partial refill (opportunistic partial charging).
These three strategies are translated into rules for simulating the agents' charging decisions, as shown in Fig. 3a.Once an agent decides to charge, if parking time allows, the battery will be fully charged by an intermediate charger or 80% charged by a fast charger.As an example of BEV simulation outputs, Fig. 3b shows that if this agent does not use daytime charging, they will run out of battery before going to the last activity location (light blue curves).If adopting the Plan-ahead strategy, this agent will charge the battery during the first parking event (green curve) so that they are able to complete the activities of the simulation day.

Charging access and initial SOCs
Most public and workplace charging stations today can deliver 11 kW-22 kW (Mathieu et al., 2020).Newly installed home chargers for Tesla have a power of around 11 kW, as do most recent home chargers in Sweden and Germany, some even 22 kW.The increasing BEV adoption also comes with the expectation that the power output of workplaces and public chargers will increase in 2030 (Mathieu et al., 2020).Therefore, in this study, the charging options are (1) slow charger (11 kW) for overnight charging at a detached house or a non-home residential parking spot nearby an apartment, (2) intermediate charger (22 kW) for charging at We initialize the BEV fleet SOCs according to whether the agents live in a detached house or not.For simplification, an agent living in a detached house always starts the simulation day with a fully-charged battery.This may lead to an overestimation of the overnight charging demand because people may only need to charge BEVs once every three days, according to a simple estimation (Wang et al., 2019).The SOCs of the rest of the agents are randomly drawn from a skewed normal distribution with skewness of −4 (Azzalini and Capitanio, 1999), ranging between 0.2 and 0.9.An arbitrary initial SOC distribution biases the simulation results (Hipolito et al., 2022).Therefore, we run five consecutive simulation days with the same planned activities, considering overnight charging and different daytime charging strategies.If an agent lives in an apartment and fails to finish driving or ends up with SOC below 0.2, the agent is assigned a non-home residential charger to charge the BEV overnight so that the next day will start with a fully charged battery.
Different charging strategies will lead to varying distributions of initial SOCs.For example, the Event-triggered strategy will keep SOC at a high level at the end of the simulation day and therefore a high SOC at the start of the following simulation day.After a continuous simulation spanning multiple activity days, the initial SOCs patterns stabilize and we use the fifth day's results for further analysis.In summary, we simulate charging access shown in Table 1.The three charging strategies have varying initial SOCs (Fig. 3c).

Quantifying charging demand
Based on the output of the BEV simulation module, we aggregate charging demand from two perspectives: individual charging patterns and spatiotemporal patterns of charging needs.We then compare the distribution of simulated charging points with today's charging infrastructure.

Individual charging patterns
We first summarize the charging demand of individual BEV users by two characteristics: whether they do daytime charging and finish all the planned activities without draining out the battery.Next, for those who use charging points at workplaces or other places, we summarize their charging behaviors in terms of charging duration, the share of charging duration of parking time, and the total energy from charging points to their BEVs.

Spatiotemporal patterns of charging needs
Charging points are provided when the agents want to charge their BEVs.Therefore, the required number of charging points in each DeSO zone is calculated as the maximum number of plugged-in BEVs in this area at each minute during the simulation day.
We quantify required charging points at workplaces, other places, and near apartments (non-home residential chargers) using five measures: (1) the total number of charging points, (2) the number of cars served by a charge point (the agent number divided by the number of wanted charging points), (3) the number of charging points in each DeSO zone, (4) the number of charging points per km 2 , and (5) hourly power demand.

Comparison with today's infrastructure
We scale up our results from the simulation of 35% VG car users to the demand of all the VG car users.Here, we focus on the spatial disparity of daytime charging points, i.e., the value difference in the number of charging points per zone between the simulated and today's infrastructure (2022).A negative value difference indicates a region needs to build more charging points to support BEV users in VG.At the same time, a positive value suggests that today's infrastructure is sufficient for 100% BEV adoption for VG's private car owners.
Sweden currently has 14,339 public charging points (Power Circle AB, 2022).However, there are no official data on private charger deployment (Xylia and Joshi, 2022) and the statistics of charging points deviate between different data sources.Therefore, the comparison indicates the magnitude of the charging point deficit toward a 100% BEV adoption future and its spatial patterns.However, the absolute numbers of charging point deficits in analysis zones are less reliable due to the discrepancy between available sources.We got the data on today's infrastructure in Sweden from a free internet service that helps EV drivers find charging stations (CHARGEX AB, 2022).

Individual charging patterns
The share of agents who access daytime charging points is small, especially for those with access to a home charger (Table 2).Compared with the Liquid-fuel strategy, the Plan-ahead strategy significantly reduces the number of agents who fail to finish all their activities.The Event-triggered strategy further reduces the failure rate, but only slightly, compared to the Plan-ahead strategy; however, it results in considerably more charging demand for daytime charging.
Figs. 4a-b summarize the total charging duration for daytime charging.Charging times are 30-125 min for intermediate charging and 3-30 min for fast charging (Fig. 4a).For the Event-triggered strategy, the charging time is reduced due to more agents with higher SOC choosing to charge their cars.Living in an apartment without a home charger induces ∼15 min longer charging time.Fig. 4b suggests that charging takes around 30% of the total parking time.Fast charging only happens when the parking time is below 30 min, but the share is rather small, especially for those with home chargers.
Figs. 4c illustrates the total energy consumption for daytime and overnight charging.Overnight charging consumes a larger amount of energy than daytime charging, especially for those with home chargers, because they always top up to 100% charged.Agents without home chargers get more energy during the day than those with home chargers, especially for Strategies 1 and 2. But the relationship is reversed for overnight charging because only a small number of apartment dwellers have access to a non-home residential charger.The Event-triggered strategy reduces the gap between the two types of dwellers and relies more on daytime charging than overnight charging.

Spatiotemporal patterns of charging needs
Table 3 summarizes the charging points required by the 284,000 car users according to the charger type, occasion, and charging strategy.The number of daytime charging points increases in the order of Strategies 1, 2, and 3, and the reverse for overnight charging.The number of charging points at workplaces is 54.6% of the number at other places for Strategy 1, 56.8% and 65.2% for Strategies 2 and 3, respectively.Regardless of the strategy, there is a smaller charging demand in terms of energy at workplaces than at other places.This is because Other activity happens more frequently than Work (Table A.1).When BEV users adopt a more conservative (and frequent) charging like Strategies 2 and 3, the demand gap between the two occasions decreases.Compared with intermediate charging points, the demand for fast charging points is minimal, especially at workplaces.Fig. 5 shows the spatiotemporal distributions of daytime charging points.Workplace charging is concentrated in central Gothenburg, while charging points for Other activities are more spread out in southern Sweden.
From a density perspective (Fig. 6), all the charging strategies have many zones where only 1 or 2 charging points are needed.But for most zones, different charging strategies result in vastly different charging point densities seen from median and 50th-percentile and 95th-percentile values.
The temporal patterns of power demand shown in Fig. 7 suggest that the Plan-ahead strategy has a more concentrated power demand during the daytime than the Liquid-fuel strategy.The Event-triggered strategy creates ∼60% more daytime power demand in MW than the Plan-ahead strategy.The three charging strategies all peak at around 8 AM but the demand for the Event-triggered strategy remains high throughout the day.

Comparison with today's infrastructure
Fig. 8 shows the difference between the simulated results and today's infrastructure.The number of zones with sufficient charging points for VG BEV users decreases in the order of Strategies 1, 2, and 3 (smaller number of blue lines).From left to right, the gap between this study's simulated required charging points and today's infrastructure becomes greater.If all agents adopt the Liquid-fuel strategy, today's charging points at central Gothenburg are sufficient for 100% BEV adoption in VG.Some surrounding cities along the coast already have enough to support all VG car users (100% BEV), despite today, only 118,400 passenger cars in Sweden are BEVs (2.2%) (Xylia and Joshi, 2022).When agents are more dependent on charging at public places and work (Strategies 2-3), more DeSO zones need to install additional charging points, not only in densely populated areas such as Gothenburg but also in its surrounding urban areas.

Discussion
This study evaluates the charging infrastructure needs from the user's perspective in a future scenario of 100% BEV adoption.We take a synthetic population of Sweden for an agent-based simulation exploring the charging demand overnight and during daytime, corresponding to different charging strategies and dwelling types.The results describe individual charging patterns at a high spatial   and temporal granularity enabling detailed and local-level explorations.We show how many intermediate and fast charging points are required when and where at work, other public places, or near home, and how the infrastructure demand varies between the three charging strategies, Liquid-fuel, Plan-ahead, and Event-triggered, and two types of dwellings, detached houses, and apartments.
The chicken-and-egg dilemma in charging infrastructure planning leads to a paradox between the perceived lack of infrastructure for a shift to BEVs and the analyses that suggest that adequate charging infrastructure is required to support BEV driving.By integrating realistic charging strategies and access to a home charger (w/ and w/o), our study contributes unique insights into this paradox of EV charging infrastructure by quantifying the charging infrastructure that BEV users want instead of need.

Feasibility of charging strategies for BEV driving
In the literature, charging strategies often refer to how charging is optimally managed to maximize the profitability or power stability (Sachan et al., 2020).In this study, we examine from BEV users' perspective by abstracting the patterns of how they charge their cars in daily life, while acknowledging the simplicity of the rules translated in Section 2.2.2.In reality, people are more likely to adopt a mix of these three strategies depending on their general preferences, timings, and occasions.However, our results provide a clear distinction between the simulated charging demand given different assumptions of charging behaviors.
The Liquid-fuel strategy, the predominant strategy assumed in the literature, is more sensible for those with fixed daily activities or secured overnight charging access.However, the assumption that BEV users will adopt this strategy might lead to more problems and less satisfaction with EVs, i.e., difficulties to adopt EVs (Sprei and Kempton, 2022).We see a much smaller share of agents with home chargers who charged their BEVs during the daytime when adopting the Liquid-fuel strategy (0.35% vs. 3.59%, Table 2).In contrast with the Liquid-fuel strategy, the Plan-ahead strategy seems more realistic and rational.It drastically reduces the failure rate for those with a home charger at their detached houses (from 1.17% to 0.37%).A more aggressive charging strategy like the Event-triggered one does not further reduce their failure rate.Despite being seemingly extreme, the Event-triggered strategy resembles the behaviors of BEV users who do not have fixed overnight charging access.They rely on public charging as finding an overnight charging point near their home maybe uncertain.
Regardless of charging strategy, 99% of BEV users can manage their weekdays by relying on charging during known parking events, provided access to charging infrastructure.However, the three charging strategies result in different numbers of charging points and how they are used.For today's Sweden (2022), we have 23 cars per charge point and 210 cars per fast charger (Power Circle AB, 2022).These numbers are close to our simulated results for the Plan-ahead and Event-triggered strategies (Table 3).Assuming that BEV users adopt the Liquid-fuel strategy leads to an underestimation of charging demand, the Event-triggered strategy results in a large number of charging points (Table 3) and shorter charging time (Fig. 4a-b), leading to potentially higher infrastructure costs.
A major simplification of simulating charging behaviors in the current work is the lack of charging costs including electricity price and parking fee.This simplification is made to separate the demand and supply sides providing a perspective on what BEV users want, free of other constraints.The literature indicates that consumers can be price sensitive and modify their charging behaviors accordingly (Yang et al., 2021).Chakraborty et al. (2019) show that plug-in electric vehicle drivers use workplace charging when the electricity cost is higher at home, and more so when charging at work is free.In our study, due to more activities in public places than work activities, there are more charging points wanted at Other occasions than at workplaces (Table 3).In reality, the difference may be smaller due to a better charging price at workplaces.
The current results regarding the three simplified charging strategies can work as a starting point to further building other scenarios, e.g., a more realistic representation of charging behaviors by integrating different supply-side pricing options and seeing how the charging cost affects charging demand.With 100% BEV adoption, the grid will need to significantly expand in capacity (Powell et al., 2022), and pricing can provide incentives to make BEV users adapt to more grid-friendly charging behavior.
Y. Liao et al.

Role of charging access at work, public places, and near home
Overnight charging contributes more energy than daytime charging at work and other public places for an average weekday (Fig. 4).The relationship between these two sources of energy depends on dwellings.For all the detached house dwellers (62.4%) of VG car users, overnight charging provides the energy of 1500-2600 MWh, about 1.5-54 times the energy supplied by daytime chargers .For all the apartment dwellers (37.6%), this number is around 0.2-0.9(150-460 vs. 511-996 MWh).Early adopters have predominantly access to home chargers (Hardman et al., 2018) but the demand of those who do not have home chargers in a future scenario remains unclear.Our study shows a distinct difference in charging infrastructure needs and charging demand between the two types of dwellers.Policymakers should be aware of such a difference to efficiently plan charging infrastructure for wider BEV adoption, especially to encourage those living in other types of housing than detached houses.The role of overnight charging also depends on the charging strategies for daytime charging.The more aggressively batteries are topped up during daytime, the less dependence on overnight charging there is (Table 3).
This study also reveals the spatiotemporal distribution of the charging needs at work and other places desired by BEV users with different charging strategies.When adopting the Plan-ahead strategy, agents think about planned activities before the first parking event, leading to early charging (Fig. 7).On the other hand, the Liquid-fuel strategy prolongs the time between starting the day and the first charging because people wait until batteries run low.
Compared with today's infrastructure in the study area (Fig. 8), significantly more charging points are needed for 100% BEV adoption according to the Plan-ahead and the Event-triggered strategies.A previous study (Funke et al., 2019) suggests that ''public charging infrastructure as an alternative to home charging is only needed in some densely populated areas''.However, our results show that, first, not all future BEV users will have stable access to home charging, which leads to a distinct demand split between overnight charging and daytime charging (mostly in public spaces).Second, to meet local BEV users' charging demand, additional 11-220 charging points per zone can be needed in many cities and surrounding areas, in addition to a few densely populated areas.Our synthetic population will be open source and available publicly.Others can, for example, use our work to look at the storage capacity available for vehicle-to-grid systems (Hipolito et al., 2022) or other applications.

Daytime charging demand and grid perspective
The overall daytime charging demand for 100% BEV adoption is high regardless of charging strategies.The peak demand in Gothenburg is 250 MW (Svenska Kräfnät, 2022), ca.750 MW in Västra Götaland, considering its population is three times that of Gothenburg.Our results indicate that the peak demand of daytime charging corresponds to roughly 6%-33% of today's grid capacity.This is related to the simulation assumptions.First, we assume a charging point is placed when an agent wants to charge the car.This is different from many previous studies, where the number of charging points is optimized according to various targets, e.g., minimized infrastructure cost.Based on this distinct simulation design of this work, the results highlight what BEV users want instead of need.Second, the three charging strategies are abstracted and simplified from real-world behaviors.In reality, electricity prices will affect BEV users' charging decisions.Given these assumptions, this study provides a baseline from a charging behavior point of view.
In general, charging infrastructure will affect the electricity grid.For example, a recent study suggests that the peak net electricity demand would increase by 50% in full electrification in the US (Powell et al., 2022).The study also indicates that daytime charging can be essential in reducing the load on the electric grid.Moreover, smart pricing incentives for demand shift and vehicle-to-grid strategies will significantly reduce the grid impact (Barthel et al., 2021;Tuchnitz et al., 2021).The daytime power demand according to different charging strategies given by the current study can contribute insights to designing good daytime charging experience from a user's perspective.

Limitations and future work
The first limitation is the lack of charging cost from the user's and the infrastructure's perspectives (as discussed in Section 4.1).Future directions include having more realistic charging strategies integrating charging cost models and minimizing charging infrastructure costs.The second limitation is the lack of feasibility constraints regarding land use and grid.We assume a charging point can be installed where the agents park and decide to charge their BEVs.However, in reality, the feasibility of a charging point placement is constrained by many factors such as zoning constraints, business models, etc.A more detailed depiction will better guide planning practice.The third limitation concerns using today's population and travel patterns for a future scenario.Our future work will incorporate changes in the population, including sociodemographic and behavioral changes, e.g., the share of car users, etc.We will in particularly consider interactions between driving and a shift to more sustainable modes of transport such as cycling and public transit, thus further reducing the need for parking and charging stations.By varying assumptions of travel demand such as trip distances, we could also explore the impacts of behavioral changes on charging travel demand in the future.
Moreover, we currently simulate an average weekday (Mon-Fri) and not an average day of the week (any day); thus, longdistance driving is under-represented, and so is the need for fast-charging, particularly along travel corridors.This study reveals that the need for fast charging to facilitate daily driving is small.One future direction would be to extrapolate one average weekday to multiple-day (Mon-Sun) travel distances and test the robustness of the model.

Appendix B. Sensitivity test results
The sensitivity test aims to reveal how the simulation results change corresponding to different SOC thresholds of the three charging strategies and fast charging powers.We ran seven additional simulations using a SOC threshold of 0.3 for Strategies 1 and 2 and the fast charging power of 150 kW.There are ten scenarios (Table B.1), where the main body of the manuscript presents scenarios No. 1-3, and the sensitivity test covers additional scenarios No. 4-10.We present the sensitivity test results in this section, complementary to the manuscript's main body.

B.1. Individual charging patterns
As an addition to Table 2, Table B.2 shows the sensitivity results of individual charging behaviors.Higher power for fast charging does not affect the share of agents using daytime chargers and the agents' failure rate.A higher SOC threshold of commencing daytime charging induces greater charging demand and a slightly lower failure rate for those without a home charger.
Table B.3 summarizes the median values of individuals' total daytime charging duration and charging time ratio and their total energy consumption for daytime and overnight charging (in addition to Fig. 4).Higher fast charging power does not affect intermediate charging due to its small share but reduces the number of fast charging points and the required fast charging time.A greater SOC threshold leads to increased daytime charging and decreased overnight charging.

B.2. Spatiotemporal patterns of charging
As an addition to Table 3, we show the sensitivity results of required charging points in Table B.4.Higher power for fast charging sometimes leads to more fast charging points (SOC=0.2for Strategies 1-2) but a reduced number of fast charging points in the other scenarios.We see more fast charging points required for Strategies 1-2 when the agents decide to charge on a low battery (SOC=0.2) because, in these scenarios, they are more likely to have short parking events with the battery SOC below 0.8, triggering the need for fast chargers.On the other hand, when the agents have a more conservative charging strategy (SOC=0.3for Strategies 1-2 or Event-triggered strategy), we see a declined demand for fast charging.However, a higher fast charging power always goes with a slightly smaller number of intermediate charging points.A higher SOC threshold of commencing daytime charging means more daytime charging points and fewer overnight charging points for apartment dwellers.B.4).A greater SOC threshold induces higher power demand in both Strategies 1 and 2, especially the Liquid-fuel strategy.

B.3. Comparison with today's infrastructure
Table B.5 suggests higher fast charging power only slightly reduces the demand for charging points.More conservative charging (higher SOC threshold) requires more additional charging points for Strategy 1 Liquid-fuel but is not significantly different in Strategy 2 Plan-ahead.

Fig. 2 .
Fig. 2. Simulation inputs: agents, road network, and BEVs.(a) Car users' home distribution.(b) The road network in VG.(c) The entire road network.(d) Charging time as a function of SOC by battery size and power.For slow charging overnight (11 kW), we assume SOC will reach 1 before the start of the next day.(e) Assumed distribution (%) of BEV types by battery size and by income group.Yearly income is measured in thousand (K) Swedish krona (1K Swedish kronor is about 92 Euro).

Fig. 3 .
Fig. 3. Simulating three charging strategies and their SOCs.(a) Charging strategies.(b) An example of SOC time history of a selected agent.The light blue curves indicate no daytime charging.The green curve shows adopting the Plan-ahead strategy, which leads to charging during the first parking event.(c) Initial SOCs for the three charging strategies.(For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

Y
.Liao et al.

Fig. 4 .
Fig. 4. Charging demand of individual agents.(a) Charging duration in minutes per agent per day.(b) Charging duration as the share of parking time per agent-day.Error bars indicate the range between the 25th-percentile and the 75th-percentile.(c) Energy flow from chargers to BEVs.

Fig. 5 .
Fig. 5. Spatial distributions of charging points for daytime charging by charging strategy.
Y.Liao et al.

Fig. 6 .
Fig. 6.Daytime charging point density by charging strategy and occasion (work or other).

Fig. 7 .
Fig. 7. Hourly power demand of all daytime charging points by charging strategy.

Fig. 8 .
Fig. 8. Spatial disparity in charging points between simulated results and today's infrastructure: difference in the number of charging points by DeSO zone.Bottom line charts indicate the magnitude of charging points disparity where each line represents a DeSO zone.

Fig. A. 1 .
Fig. A.1.Activity patterns.(a) Temporal profiles.(b) Distribution of parking duration for the activities reached by car.

Fig
Fig. B.1 shows the temporal profile of daytime charging power demand of the ten scenarios.Higher fast charging power has minimal impact because most wanted charging points are intermediate(Table B.4).A greater SOC threshold induces higher power demand in both Strategies 1 and 2, especially the Liquid-fuel strategy.

Table 1
Charging access.If an agent fails to finish driving or ends up with SOC below 0.2.b Depending on charging strategy, parking duration, and SOC.If an agent decides to charge the car given its charging strategy, fast chargers (50 kW) for daytime charging are provided when the parking time is below 30 min and SOC below 0.8, otherwise intermediate chargers (22 kW) are provided.workplacesand other places such as shopping malls, etc., where agents are engaged in other non-school or work activities, and (3) fast chargers (50 kW) for daytime charging depending on parking time, charging strategy, and SOC.Specifically, if an agent decides to charge the car given its charging strategy, fast chargers (50 kW) for daytime charging are provided when the parking time is below 30 min and SOC below 0.8, otherwise intermediate chargers (22 kW) are provided.

Table 2
Share of agents who use daytime chargers and failure rate.HC = home charger at a detached house.
a An agent is not able to finish all the activities with the assigned BEV and initial SOC.

Table 3
Number of charging points for daytime charging and overnight charging (non-home residential).Inter.= intermediate.

Table A .1
Activity statistics.Except for the share of activities, the rest of indicators are median values of all car agents.

Table B .1
Scenarios of charging strategies and fast charging powers.Daytime charger usage and failure rate.HC = home charger at a detached house.Charging demand of individual agents.HC = home charger at a detached house.
a An agent is not able to finish all the activities by the assigned BEV and initial SOC.

Table B .4
Number of charging points by scenario.Inter.= intermediate.

Table B .5
Statistics of charging point disparity between simulated results and today's infrastructure by scenario.