Estimating the electric vehicle charging demand of multi-unit dwelling residents in the United States

Early battery electric vehicle (EV) adopters can access home chargers for reliable charging. As the EV market grows, residents of multi-unit dwellings (MUDs) may face barriers in owning EVs and charging them without garage or parking availability. To investigate the mechanisms that can bridge existing disparities in home charging and station deployment, we characterized the travel behavior of MUD residents and estimated their EV residential charging demand. This study classifies the travel patterns of MUD residents by fusing trip diary data from the National Household Travel Survey and housing features from the American Housing Survey. A hierarchical agglomerative clustering method was used to cluster apartment complex residents’ travel profiles, considering attributes such as dwell time, daily vehicle miles traveled (VMT), income, and their residences’ US census division. We propose a charging decision model to determine the charging station placement demand in MUDs and the charging energy volume expected to be consumed, assuming that MUD drivers universally operate EVs in urban communities. Numerical experiments were conducted to gain insight into the charging demand of MUD residents in the US. We found that charging availability is indispensable for households that set out to meet 80% state of charge by the end of the day. When maintaining a 20% comfortable state of charge the entire day, the higher the VMT are, the greater the share of charging demand and the greater the energy use in MUD chargers. The upper-income group requires a greater share of MUD charging and greater daily kWh charged because of more VMT.


Introduction
Vehicle electrification is a crucial step toward improving energy efficiency and reducing passenger travel's environmental impact, given that transportation accounts for the largest portion (29%) of total US greenhouse gas emissions [1]. A projected compound annual growth rate of 40.7% from 2020 to 2027 in the global electric vehicle (EV) market [2] implies the need to deploy home, workplace, and public charging stations to support EV operations. The existing literature on the installation of EV chargers overwhelmingly focuses on public and workplace charging station locations and management, often assuming universal home charging availability [3]. For instance, Xie et al [4] strategically planned for intercity fast-charging infrastructure, Dong et al [5] optimized the placement of public charging stations and analyzed the impact of the chargers' power and level on range anxiety, and Kontou et al [6] measured the level of public charging opportunities in a subset of US metropolitan areas. Such research provides a variety of models and assessment methods for public charging station siting, whereas the analyses of Li et al [7] and Wu [8] aids decisions on workplace charging station deployment. Even though home charging plays the most significant role in enabling the charging and use of EVs ( [9,10]), its deployment at residences has not been studied. Hardman et al [11] showed that 50%-80% of all charging events occur at home, while the second most frequent charging location is the workplace. However, less than 50% of household vehicles have access to dedicated parking in residences where charging installations can be undertaken [12]. Limited access to home charging, particularly for renters and those residing in apartment complexes, can slow the decarbonization of

Travel patterns of MUD residents 2.1. Data description and merging
To uncover the travel profiles and characteristics of MUD residents, we analyzed the latest data available from the 2017 NHTS [17], the 2019 AHS [18], and the 2019 ACS [19].
The NHTS is the primary database used for daily travel profile analysis. It is regularly used to infer EV daily mileage and charging needs in the US [20][21][22] in scenarios where EVs dominate the US vehicle market. However, this data source lacks information on vehicle owners' housing types (i.e. MUDs, single-family houses, etc), which can significantly impact assumptions on home charging availability and access [12]. Type of residence shares are available through AHS data; 71% of the US structures have one housing unit, while only 24% are MUDs, showing that single-family housing is the most prevalent housing type in the US. We gain insights into the spatial distribution of MUDs shares per census division, as shown in figure 1. The greatest share of MUD housing is in the Middle Atlantic division (34.1%) and the lowest in East South Central (15.4%). We performed a cross-tabulation analysis using AHS MUD share data, 14 income categories, and nine census divisions. We observe that a greater MUD percentage is found in lower-income categories, whereas the lowest shares are concentrated in the highest-income categories. The lowest income classes in the New England, Middle Atlantic, and Pacific divisions have the greatest share of MUDs. The highest income category ($120 000 or more) of divisions, such as West North Central and East South Central, has the lowest share of MUDs.
To match the NHTS household-level travel data with the MUDs share information, we fit a statistical model that predicts the share of MUDs as a function of variables included in both datasets. Variables included in both the AHS and NHTS span socio-demographic characteristics such as income and location. Evidence from the literature [23,24] demonstrates that income and location characteristics are significant covariates in predicting the housing unit type of a household. Additional characteristics such as population density or urban/rural development could be a good predictor of housing type; however, this granular cross-tabulation is not available by AHS, and thus, these characteristics do not enter the model. The ordinary least squares (OLS) method was used, and the linear regression model predicted the share of MUD residents with income and census division as independent variables. We tested interaction terms and effects and we found them not statistically significant. After conducting multiple robustness checks with the census division variables as ordinal and binary, as well as testing the entry of interaction terms, we propose the following model: where cd i ∈ {1, 0} is a binary variable for each census division; x is the continuous income variable; and β 0 , β 1 , and β 2i are the estimators. We used eight dummy variables for each census division, except for the Pacific division. The Pacific division is the reference category when all other binary variables are 1, denoting that, by  default, the variable would be representative of the Pacific. Table 1 presents the regression results. Given that R 2 = 0.827, we proceed to utilize this model for our analysis. As expected, average household income is negatively associated with the share of MUDs in a census division, while the coefficient β 2i shows the effect of each location compared to the reference case. The spatial aggregation level of the MUD share, which is the dependent variable in our regression, is illustrated in figure 1 as the US census divisions. Household income in the AHS dataset is classified into 14 groups. Since the US is divided into 9 census divisions, the total number of observations is 126 (14·9). Using the OLS model fitted with the AHS data, we apply function (1) to the NHTS observations and estimate the share of MUDs in the region of and for each household. Then, we synthesize the MUDs data as follows: (1) assign a binary indicator (1 designates a MUD resident) to each household in the NHTS trip database based on the location and income level, and (2) confirm the robustness of 100 simulation results by examining the time-of-day vehicle miles traveled (VMT) distribution.

Travel profiles overview
We proceeded by analyzing MUDs and other housing type residents' travel patterns by randomly selecting one scenario from our runs. In figure 2, the average VMT is based on those who are taking trips, indicating that households with zero VMTs are not counted. The aim of our research is to investigate the prospective charging needs and energy use of EV drivers and MUD residents; therefore, we focus on those trips taken by passenger vehicles with positive VMTs. In figure 2, we aim to demonstrate the mileage per trip of only these  vehicles that are driven. This figure along with the ones that comment on the location of the vehicle per time of day and the dwell time distributions (i.e. figures 3 and 4) provide a comprehensive overview of how passenger vehicles are being utilized by MUD and other housing types' residents. Few households take trips in the early morning and at midnight. We found no significant difference in the average daytime VMT between residents of MUDs and other housing types. The VMT of predicted MUD residents, represented by the blue line in figure 2, are lower than those of other residents. The variation of VMT is consistently higher in the early morning hours, after midnight, for households residing in housing types other than MUDs. The average VMT was low during the daytime because most trips occurred for commuting purposes. The distance of trips in the early morning and midnight is higher, which suggests that these trips are non-habitual and would require EVs with greater state-of-charge to be completed.
Understanding the differences between MUDs and other housing type residents in terms of the share of time spent driving, at home, or in other locations is of great importance because it could present opportunities for EV charging. The median share of time spent at home over 24 h of a day is not significantly different between housing types, while for all census divisions and housing types, this share falls within the (60%, 70%) interval. The share of daily time spent at home is higher for MUDs than for other residents in divisions such as New England, Middle Atlantic, and East South Central. The distributions of the dwell time spent parked at different trip destinations, including home, work, and public locations, are presented in   Figure 4 shows the share of vehicles driven or parked at home, work, or other public locations per day. More than 80% of vehicles are idle at home from 10:00 p.m. to 7:00 a.m., whereas approximately 20% are at their workplace by 9:00 a.m. The share of vehicles of those residing in MUDs parked at workplaces was lower than that for those residing in single-family housing. We only estimate the dwell time of the one primary household vehicle in the possible locations of home, workplace, and other public places and do not account for the trips that are conducted with other modes of travel. However, households might be conducting additional trips with transit mode alternatives or active transportation. Vehicles dwelling between 4 and 10 h parked at work premises. Driving occurs during the daytime, from 9:00 a.m. to 8:00 p.m., while the share of vehicle driving does not significantly differ between MUDs and other housing types' drivers.

Clustering of MUD residents' travel patterns
After reporting descriptive trends of time-of-day variation in the automobile travel patterns of MUDs residents, we focus on clustering them to uncover distinctive driver behaviors. To do so, we evaluated alternate clustering techniques. K-means is a popular clustering algorithm in the literature, but the Gaussian mixture model is a more appropriate method when clusters have different sizes and correlations within them, which is the case for travel data (e.g. [25,26]). However, both clustering methods require the pre-determination of the number of clusters before implementing them. The elbow method [27] and silhouette coefficient [28,29] were used to determine and validate the number of clusters in a dataset. To avoid predetermining the optimal number of clusters, we used the hierarchical agglomerative clustering (HAC) method [30][31][32], which constructs a binary tree over the data from the bottom and merges groups until only one single cluster is left. To create clusters of time-of-day travel patterns, except for the variables of income, census division, and trip travel time, we generated new covariates, including dwell time at home, dwell time at work, and dwell time at a public location, as well as VMT and vehicle location per day. We denote the vehicle location as a categorical variable of 'home,' 'workplace,' 'public,' and 'driving' . The HAC classification algorithm is implemented to determine travel behavior classes after normalizing all variables using the min-max formula. Three clusters could best describe the travel patterns of MUD residents; thus, a threshold distance of four was set and applied in the classification algorithms.
The descriptive statistics for the three clusters are presented in table 2. Table 2(a) shows that clusters 1 and 2 had similar average dwell times at home. However, these clusters differ in terms of the average dwell time at work and in public locations. The greatest share of daily travel patterns of MUD residents belonged to cluster 3, for which the average dwell time at home was shorter, while the dwell times at the workplace or public location were greater than for the rest of the clusters. Cluster 2 MUD drivers spent their day primarily at home, while drivers of cluster 1, characterized as less outgoing than cluster 3 drivers, spent their day primarily between home and work locations.
To gain further insights into the MUDs drivers' daily travel clusters, we explored their dwell time distributions at locations such as their MUDs, workplaces, public locations, and their trip time distributions  in figures 5(a)-(d), respectively. Cluster 3 had the widest dwell-time distribution at home, with the lowest mean and the greatest standard deviation. The dwell time distribution of cluster 2 at the workplace and public locations was highly concentrated, with the lowest mean and standard deviation. The VMT presented in figure 5(d) was averaged for all MUD residents, including those who did not conduct any trips during the hour. Figure 5(d) shows that almost no vehicle miles were driven for all three clusters in the early morning and late evening hours. Cluster 2 represents relatively low VMT during the daytime compared to the other two clusters, while cluster 3 had relatively higher VMT during the daytime compared to the rest. Therefore, MUD residents and drivers of cluster 3 spend the least time at home, but the most time at the workplace and the public locations with the highest VMT. Clusters 1 and 2 were similar in terms of dwell time distribution at home and daily VMT. However, cluster 1 drivers spent more time at the workplace and in public locations, as indicated by their corresponding distribution. Given that the share of residents in MUDs across census divisions is greater in low-income groups, we examined the impact of a single factor, income, on travel pattern differences across MUD residents. MUD residents are grouped into three income groups (lower-income, middle-income, and upper-income groups). The categories were assigned to account for spatial differences per census division because the income range for the middle class differs spatially according to the U.S. Census Bureau's 2018 ACS [16]. Descriptive statistics of the MUD residents' travel patterns for the three income groups are provided in table 2(b). The middle-income class has the largest share of MUD residents, while the upper-income class has the least share. Table 2(b) shows that the upper-income class has the least dwell time at home and the largest dwell time at the workplace and other public locations because wealthier populations have access to a greater share of opportunities. The lower-income class has the largest dwell time at home, but the least at the workplace and public locations, associated with higher unemployment rates among the lowest-income class according to the U.S. Bureau of Labor Statistics [33].
Three groups of dwell time distributions at major locations (MUDs, workplaces, and public places) and trip time distributions for each income group are shown in figures 6(a)-(d). The middle-income group had the widest distribution of dwell time at home with the greatest standard deviation. Figure 6(d) shows that the low-income group travels lower vehicle miles during the daytime than the other two groups, while the upper-income group 3 has the highest VMT during the daytime. This aligns with expectations, given prior empirical evidence that income is a strong and positive predictor of VMT [34].
The travel pattern classes and income groups further assist in quantifying the unique EV MUD charging needs and understanding how the provision of MUD charging should be specifically tailored to these socioeconomic classes.

EV charging model of MUD residents
The main scope of this work is to uncover the needs of MUD residents for EV charging infrastructure under the assumptions of 100% EV adoption and use. We devise a model that enables estimating the demand for deployment of EV home charging and electricity consumption in MUDs, compared to other charging locations (i.e. workplace and public) while accounting for MUDs passenger vehicle users' travel patterns. Due to limited and proprietary data on EV adoption and use by residents of MUDs, this assumption becomes necessary. It is also of interest to decision-makers given that states like New York and California have set 2035 as the last year of selling fossil-fueled vehicles, signifying that this transition is underway. (New York A.4302/S.2758 [35], California Executive Order N-79-20 [36]).
Our optimization framework determines the optimal charging use for the hypothesized MUD charging hub. Such charging stations need to be deployed to account for drivers' decisions and charging availability among the locations of trip stops during the day (MUDs, workplaces, and public destinations). Therefore, the mathematical program presented below was developed to assist with determining the deployment of charging stations at MUDs and defining the optimal charging energy use of the EV drivers and MUD residents.

Model formulation
The framework has two groups of decision variables to determine (a) the necessity of charging stations at MUDs (i.e. a binary variable of MUD charging availability) and (b) the charging time at each location where  min ∑ ij e ij · P j · t ij subject to: Constraints (2) present the state transition functions that update the range miles of MUDs household vehicle i at the location j of vehicle stops. Constraints (3) ensure that the range miles of the MUD household's vehicle i at location j are not less than the comfortable driving range miles (i.e. assuming 20% of the full driving range of miles). Constraints (4) illustrate that the charged range miles of a MUD household's vehicle i at location j do not exceed the difference between full range miles and current range miles. Constraints (5) show that the charging time of the MUDs household vehicle i at location j does not exceed the current dwell time. Constraints (6) demonstrate that if there is charging availability at the MUD location for household i, the charging time can be non-negative; if there is no charging availability, the charging time can only be set to zero. Constraints (7) ensure that the charging station availability variables are binary and the charging time variables are nonnegative.

Parameters assumptions
For each household, the driver may have the opportunity to charge their EV at home, workplace, and public locations. Our proposed model aims to determine charging station deployment at MUDs, while charging availability at workplaces and public locations are assumed to be known parameters. We used the number of charging stations per state from the alternative fuel station locator [37]. Given the time constraints of charging at public locations, Level 1 and Level 2 chargers are regarded as offering workplace charging, whereas direct current fast charging (DCFC) stations are preferable for charging at public locations. We observe that the current placement of charging station infrastructure in workplaces and public locations is sparse. If we assume that the current gasoline station network can cover 100% of the gas vehicle refueling  [47] needs, an even denser network would be needed to cover the recharging needs of a fully electrified passenger vehicle fleet in the US [38]. We measured charging availability as the ratio of charging stations to gasoline stations, divided by a factor τ j , which captures the difference in efficiency between gasoline refueling and EV recharging. Charging is more time-consuming than refueling at a gasoline station. However, 50%-80% of all EV charging events occur at home [34], while all conventional gasoline vehicles refuel at gasoline stations. The average time to fill a gasoline vehicle's tank is approximately 5 min [39], whereas the average time to charge at a DCFC is approximately 20-60 min (i.e. 40 min on average) [40]. Therefore, the charging availability at public locations is derived from the division of the DCFC charging stations and gasoline stations multiplied by a coefficient of 5/40. Charging availability at the workplace is calculated by the division of assumed workplace charging stations and the number of firms [41] across the US, assuming an adequate number of charging plugs at the workplaces where chargers are available. Our assumptions regarding public and workplace charging availability are presented in table 4.
It would be preferable to assume that residential electricity costs vary by the time of use (TOU). However, due to our census division-based analysis (i.e. using the census division as the geographical unit of analysis), we use residential electricity prices that reflect the average daily costs per division in US dollars per kWh from the US Energy Information Administration [42], as shown in table 4. There is no available dataset that provides an average TOU residential volumetric pricing scheme in different census divisions and, thus, we are working with the best alternative. Residential TOU electricity rates are not ubiquitous in the United States. A survey by the Brattle Group [43] found that only 14% of the utilities in the US offer TOU rates and that, when a TOU rate is available, only 3% of residential customers are enrolled on average. The MUD charging stations are assumed to host Level 2 chargers. The electricity price for charging at the workplace is assumed to be 0.33$ kWh −1 [44]. The other parameters adopted in our model are listed in table 4.

Results
Our optimization problem was solved using Gurobi 9.1.0 [48] build v9.1.0rc0 in Python 3.8.3. for all MUD residents and their travel patterns, as described in the previous section.
We assume 100 miles as the EV driving range for our analysis. Based on the Department of Energy [49], the median driving range of EVs was 234 miles for a 2021 model. Using a 100 mile driving range is quite conservative compared to the current models' range availability. If we were to use 234 miles as the full driving range, at the end of the day, almost all the EVs do not need to recharge. This is the case even when we simulate three days of driving. NHTS daily travel data are insufficient for tracking such charging needs and weekly trip data and their variation would be more suitable for a 234 mile range EV home-charging needs assessment. On the other hand, current research finds that the state of charge, right before charging sessions, is approximately 50% [50]. This demonstrates that even if we use 200 miles of electric range, the remaining 100 miles are unutilized due to range anxiety issues. We conduct our MUD EV charging assessment with the actively utilized 100 miles of range.
The comfortable driving range of the drivers was set to 20% of the driving range, per literature evidence [51]. The model was applied to 24833 household vehicles across the nine census divisions. The problem is solved under two scenarios in general: (1) END20: EVs maintain a 20% state-of-charge as a comfortable range all day; (2) END80: EV drivers ensure that recharging meets or exceeds the boundary of 80% of full range at the end of the day but maintain a comfortable range during the intermediate trips. MUD residents were grouped into three clusters based on their travel profiles and into three income groups based on the income distribution of the census divisions. The heterogeneity of the MUDs charging station deployment was evaluated across these classes and groups. With the assumption of deployment of Level 2 chargers at the MUDs in table 4, we conducted numerical experiments with a remaining EV driving range of 30 miles as well as 50 miles at the beginning of the experiment's time horizon, to capture the day-starting battery state-of-charge effect on MUD charging and energy use.

Expected deployment share of MUDs charging stations 4.1.1. Base case: one-day run
We determine the deployment of MUDs charging stations and compute the optimal share categorized into three travel pattern clusters and three different income groups across the country when running the optimization model over a day; the results are depicted in figure 7.
In figure 7, we remark that I30E20 denotes the scenario of starting with an initial state of charge of 30% at the beginning of the day and sustaining a comfortable range of 20% all day; scenarios I30E80, I50E20, and I50E80 follow the same logic but with a different state of charge at the beginning of the day (30% vs. 50%) and end-of-day state of charge that the drivers aim to meet (20% vs. 80%). For some residents, accommodating their trips with EVs is infeasible because of the strict state-of-charge boundaries that we set or their travel patterns. We report modeling results as the share of MUD residents who require MUD charging station placement; however, there is also a share of MUD residents for whom the operation of an EV, as described here, is infeasible.
Recall that both travel cluster 3 and the upper-income group are characterized by the highest VMT and the least dwell time at home compared to the rest of the travel clusters and income groups. Clusters 1 and 2 had similar amounts of VMT and dwell time at home, while the lower-income group had less VMT and dwell time at home than the middle-income group. Both travel cluster 2 and the upper-income group represented the least share of MUD households.
In figures 7(a) and 8(b), overall, the END80 scenarios of all three travel clusters and income groups result in over 90% of MUDs charging deployment rates, which is significantly greater than that of the END20 scenarios where the MUD charger share has a wide range from 20% to 80%. We focus on the END20 scenario results in figures 7(a) and (b). In figure 7(a), travel clusters 1 and 2 have a similar share of MUDs deployment because of similar VMT and required EV energy volume. Because the number of households represented in cluster 2 is the least, travel cluster 2 has the largest variation in the MUD charging deployment share. In figure 8(b), the higher the income group level, the greater the share of MUDs charging deployment. Given that the upper-income group has the least number of households, we expect it to show the greatest variation.

Sensitivity analysis: reporting the middle day of a three-days run
We present the findings associated with running the optimization model in a three-days horizon experiment. The results are depicted in figures 7(c) and (d). For the END80 scenarios in both figures 7(c) and (d), all three clusters and income groups have almost 100% of MUDs deployment rate. As shown in figure 7(c), travel cluster 3 had the largest share of MUDs deployment for the END20 scenarios. In terms of the END20 scenarios, the shares of charger deployment in figure 7(c) for all three clusters are higher than those in figure 7(a). When we compare the END20 scenarios in figure 7(d) with those in figure 7(b), the three income groups exhibit the same relative relationships (i.e. the higher the income, the greater the share of MUDs charging placement). The share of MUDs charger deployment in figure 7(d) for all three income groups is higher than that in figure 7(b). The results suggest that three consecutive days of numerical experiments result in a greater need for MUD charging stations because of the longer distances and more realistic representation of the EV state of charge. EV charging becomes more flexible as the optimization horizon is expanded and our three-day run results are expected to better represent reality, given that EVs do not need to recharge every day [52,53].

Total energy use per vehicle at all charging stations 4.2.1. Base case: one-day run
We also determined the total energy use per vehicle at MUD charging stations and other locations. We plotted the total energy use per vehicle at all locations (i.e. home, workplace, and public) within a day's run as a base case, as shown in figure 8. The higher the VMT, the more energy is required daily to power the EVs  of MUD residents. In figures 8(a) and (b), the END20 scenarios have lower total energy use than END80 because END80 requires at least 80 miles at the end of the day. For both the END20 and END80 scenarios in figure 8, owing to the highest VMT, both travel cluster 3 and the upper-level income group had the greatest energy use at all locations. As shown in figure 8(b), the higher the income group, the greater the total energy used at all locations. Figures 8(c) and (d) plots the total energy use per vehicle at all locations on the middle day in a three-days run. Less total energy use per vehicle is expected for the middle day in the three-days run, given the low VMT compared to the driving range of EVs, which might not require frequent daily charging. By comparing figures 8(b) and (d), the END80 scenarios of the lower-, middle-, and upper-income groups imply that the total volume of energy use per vehicle for one-day runs is higher than that of three-days runs. MUD residents and EV drivers can effectively spread their recharging sessions over three consecutive days, demonstrating that daily charging deployment and management models may overestimate the demand of chargers.

Energy use per vehicle at MUDs compared to all charging stations 4.3.1. Base case: one-day run
The share of energy use per vehicle in MUDs is compared to that recharged in all locations (i.e. home, workplace, and public) within a daily run as a base case in figures 9(a) and (b). As shown in figure 9(a), the share of energy use per vehicle at home is comparable for all three clusters. In figure 9(b), the upper-income group had the highest share of energy use per vehicle at home. Overall, the share of energy use at the MUD charging station was approximately 40%. According to figures 9(c) and (d), we first observe that the upper-income group consumes over 80% of electricity volume per vehicle charging at home compared to other locations. For the lower-income group, the END20 scenarios of three-day runs in figure 9(d) consume less energy per vehicle at MUDs over the total energy use than that of one-day runs, as shown in figure 9(b). In the END80 scenarios of three-days runs (figure 9(d)), more energy was recharged per vehicle at MUDs over the one-day runs, as shown in figure 9(b). In the three days optimization horizon, there are more opportunities to recharge at home, with longer dwell time and lower electricity prices.

Discussion
This study uncovers the daily travel patterns of MUD residents based on the latest NHTS and AHS data. It shows the characteristics of the different travel clusters generated by a HAC method and income groups. Three clusters were identified and their travel profiles (i.e. dwell time, VMT, etc) were analyzed in this study. Given that a great share of MUD residents are characterized as low-and middle-income, income-grouped travel patterns are also summarized. An optimization model is developed to determine the demand for deploying charging stations in MUDs and to simulate the charging decisions of MUD residents' vehicles under a scenario of mass electrification. We find that for the cluster with the highest average daily VMT, charging availability in MUDs is needed because charging at home costs the least and the dwell time is high. The upper-income class requires a greater share of MUD charging and charges higher kWh daily owing to greater VMT. The higher the income level, the greater the share of MUDs deployment and energy use. Charging availability in MUDs is indispensable for EV drivers who wish to recharge their vehicles to meet 80 miles of range by the end of the day. When an EV driver needs to ensure a comfortable range of mileage all day, the energy MUD charger use could be less than 5 kWh, especially when the starting state of charge is greater than 30 miles. However, if the driver is inclined to recharge their vehicle and meet an 80 miles range boundary at the end of the day, the energy use at the residential MUD chargers is much higher and depends on the starting state of charge.
Next research endeavors should aim to use contemporary EV models' driving range (e.g. longer driving range), accompanied by synthesized weekly household trip data to capture EV MUD charging needs over a longer period. Future research should also move beyond the national-level characterization of MUD charging needs and leverage finer-resolution data for community-level MUD charging assessments. With the future deployment of charging stations in MUDs, multiple residents may share a limited number of stations. Therefore, smart charging-scheduling models and algorithms for managing residents' charging behaviors should be developed. In addition, techno-economic assessment modeling [54] can evaluate the financial viability of such community charging hubs shared by apartment complex residents to enhance home charging accessibility and economic viability.

Data availability statement
The data that support the findings of this study are openly available at the following URL/DOI: https://doi. org/10.13012/B2IDB-4230392_V1.