Smart charging management for shared autonomous electric vehicle fleets: A Puget Sound case study

Transportation Research Part D 78 (2020) 102184

Contents lists available at ScienceDirect

Transportation Research Part D journal homepage: www.elsevier.com/locate/trd

Smart charging management for shared autonomous electric vehicle fleets: A Puget Sound case study Tony Z. Zhang, T. Donna Chen

T

⁎

Department of Engineering Systems and Environment, University of Virginia, P.O. Box 400742, Charlottesville, VA 22904, United States

A R T IC LE I N F O

ABS TRA CT

Keywords: Shared mobility Autonomous vehicle Electric vehicle Smart charging Demand side management Vehicle grid interaction

Increasingly, experts are forecasting the future of transportation to be shared, autonomous and electric. As shared autonomous electric vehicle (SAEV) fleets roll out to the market, the electricity consumed by the fleet will have significant impacts on energy demand and, in turn, drive variation in energy cost and reliability, especially if the charging is unmanaged. This research proposes a smart charging (SC) framework to identify benefits of active SAEV charging management that strategically shifts electricity demand away from high-priced peak hours or towards renewable generation periods. Time of use (TOU), real time pricing (RTP), and solar generation electricity scenarios are tested using an agent-based simulation to study (1) the impact of battery capacity and charging infrastructure type on SAEV fleet performance and operational costs under SC management; (2) the cost reduction potential of SC considering energy price fluctuation, uncertainty, and seasonal variation; (3) the charging infrastructure requirements; and (4) the system efficiency of powering SAEVs with solar generation. A case study from the Puget Sound region demonstrates the proposed SC algorithm using trip patterns from the regional travel demand model and local energy prices. Results suggest that in the absence of electricity price signals, SAEV charging demand is likely to peak the evening, when regional electricity use patterns already indicate high demand. Under SC management, EVs with larger battery sizes are more responsive to low-electricity cost charging opportunities, and have greater potential to reduce total energy related costs (electricity plus charging infrastructure) for a SAEV fleet, especially under RTP structure.

1. Introduction Innovations are revolutionizing the transportation and electricity sectors. In the transportation sector, three technological trends are vehicle automation, vehicle electrification, and shared mobility, and many researchers are forecasting the future of transportation to be shared, autonomous and electric (Sperling, 2018; Chen et al., 2016; Loeb et al., 2018; Bauer et al., 2018). When these new mobility technologies become widely available, they will have significant impacts on urban transportation, energy consumption, and land use patterns. Fulton et al. (2017) suggest that such a combination could cut transportation energy use by 70%, CO2 emission by 80%, cost of transportation by 40%, and achieve savings approaching $5 trillion per year globally in 2050, compared to the current private vehicle ownership dominant transportation system. In the electricity sector, improvement are being made to the existing infrastructure so that it is capable of delivering reliable, affordable, and clean energy, while managing the increasing complexity and needs of electricity. Pratt et al. (2010) quantify the

⁎

Corresponding author. E-mail addresses: [email protected] (T.Z. Zhang), [email protected] (T.D. Chen).

https://doi.org/10.1016/j.trd.2019.11.013

1361-9209/ © 2019 Elsevier Ltd. All rights reserved.

Transportation Research Part D 78 (2020) 102184

T.Z. Zhang and T.D. Chen

benefit of the smart grid via nine mechanisms, including the integration of electric vehicles (EVs), and estimate these mechanisms can directly reduce energy use and emissions for the U.S. electricity sector by 12%, with the potential for an additional 6% reduction indirectly. The growth of EVs connects the transportation sector and the electricity sector in a significant way. As the two systems converge, more research is needed to prepare for the opportunities and challenges that may arise. Existing literature has already explored how EV smart charging (SC) can improve energy system efficiency, but these studies focus on privately-owned EVs and individual driver behavior. Fleet managed shared autonomous electric vehicles (SAEVs) that are continually in-service cannot employ the same SC strategies prescribed to privately-owned EVs which have a significantly lower daily utilization rate. In this study, we take a first exploration at the pairing of SAEVs with SC management and investigate its potential impacts on urban mobility and energy systems. Time of use (TOU), real time pricing (RTP), and renewable energy sources (RES) SC strategies are applied to a discrete time, agentbased model of a SAEV fleet serving 10% of travel demand in the Puget Sound case study region to quantify the range of impacts of vehicle and charging infrastructure choices on fleet energy-related costs. 2. Literature review Many transportation researchers are actively studying how shared autonomous vehicles (SAVs) and SAEVs can improve the urban transportation system and affect travel behavior. Meanwhile, energy researchers are discovering solutions for large scale EV deployment and their impact on energy infrastructure and grid operation. This section summarizes previous research relevant to this work: shared autonomous mobility and EV-grid interaction. The former body of research primarily focuses on the transportation impacts (congestion, emissions, competing modes, etc.) of shared autonomous mobility and does not consider the EV-grid interaction and active charging management, while the latter body of research generally examines EV-grid interaction and energy system optimization under private EV ownership scheme or small-scale commercial EV fleets, ignoring the potential impacts of shared mobility. List of acronyms. SAEV TOU RTP SC LMP SOC RES LR SR FC LV2 PV UMG DIS

Shared Autonomous Electric Vehicle Time-of-Use Real-time Pricing Smart Charging Locational Marginal Price State of Charge Renewable Energy Sources Long Range EV (200+ mile) Short Range EV (100+ mile) DC Fast Charging Level 2 Charging Photovoltaic Unmanaged Charging Distributed Charging

2.1. Shared autonomous mobility Shared autonomous mobility has significant implications for transportation energy consumption and emissions. Many researchers have found that SAVs/SAEVs can be effective at replacing privately-owned vehicles. Fagnant and Kockelman (2014, 2015) presented an agent-based model for SAVs which simulated environmental benefits of such a fleet as compared to conventional vehicle ownership and use in a dense urban core area. Simulation results indicated that each SAV can replace 11 privately owned vehicles, but generates up to 10% more travel distances due to unoccupied VMT, without ridesharing. When the simulation was extended to a case study of low market penetration (1.3% of trips) in Austin, Texas, each SAV was found to be able to replace 9 conventional vehicles and on average generated 8% more VMT due to unoccupied travel. Brownell and Kornhauser (2014) evaluated the necessary autonomous vehicle fleet size for personal rapid transit and smart paratransit. The model predicts a fleet size of 1.6 to 2.8 million sixpassenger vehicles to serve state-wide demand in New Jersey. Martinez and Viegas (2017) presented a study to evaluate the impacts of automated shared taxis with ride-sharing in Lisbon, Portugal. The authors developed an agent-based simulation model to simulate 1.2 million trips and scenarios reflecting a situation where private car, taxi and bus trips are replaced by automated taxis. The study indicates a decrease in cost by 55%, highly increased transportation accessibility in the city, and carbon emission reductions of almost 40%. When the SAVs are electric, both travel and charging activities need to be considered in the fleet operation. Chen et al. (2016) proposed an agent-based model that simulates a fleet of SAEVs, considering electric vehicle (EV) and charging infrastructure decisions. They found that each SAEV can replace 3.7–6.8 privately owned vehicles, due to the additional charging and vehicle range constraints. In this study, charging is governed by rule-of-thumb applied to remaining battery range and number of rejected trips due to insufficient range. Bauer et al. (2018) developed an agent-based model to predict the battery range and charging infrastructure requirements of a fleet of SAEVs operating on Manhattan Island using taxi-trip data. They found that such operation will reduce green house gas emissions by 73% and energy consumption by 58% compared to an automated gasoline vehicles. Loeb et al. (2018) proposed a model simulating performance characteristics of SAEV fleets that focuses on charging station and charging time 2

Transportation Research Part D 78 (2020) 102184

T.Z. Zhang and T.D. Chen

requirements, using realistic vehicle speeds, allowing flexible charging strategies, and requiring all demand for trips under 47 miles to be met. The study concluded that reducing charge times lowers fleet response times, however fleet size increase offers significant improvement in response times in Austin, Texas. Both Bauer and Loeb assumed continuous charging when the SAEV is idling to ensure the fleet always has adequate range remaining, the difference being that Loeb adds a 30 min time delay before idling vehicles are assigned to charging. While these studies have explicitly modeled SAEV operations and charging activities, they do not address the potential challenges in the EV-grid interaction process, let alone the potential for SAEV fleet charging management.

2.2. EV-grid interaction In a survey developed by National Renewable Energy Laboratory and Lawrence Berkeley National Laboratory (Karali et al., 2017), industry respondents were asked to prioritize various topics related to EV-grid interaction. Topics with the highest priority include long term EV effects on distribution grids, business models for EV charging infrastructure, demand side management in smart grids, grid storage and load-shifting, effects of time-of-use (TOU) rate on charging costs, and interactions between renewable energy generation and EV charging activities. Langton and Crisostomo (2014) estimated that achieving California’s target of 1.5 million zero-emission vehicles by 2025 would increase summer peak load by 16%, if vehicle charging were unconstrained (compared to 2013 electricity usage); while Graabak et al. (2016) estimated that 100% PEV penetration of the passenger fleet in the Nordic region by 2050 could increase electricity demand by 7.5%. Due to the increasing EV penetration in the transportation market, many studies have researched the possibility of adopting EV for grid services, including ancillary services, RES generation storage, and energy supply via vehicle-to-grid technology, assuming EV owners have monetary incentives to use their EV for ancillary services. However, such opportunities exist largely because privately-owned EVs are utilized for only 5% of the day for transportation, on average (Shoup, 2005), and have relatively low energy demand. With the growth of shared mobility, the average vehicle utilization and need for electricity will increase significantly. Future smart grids are expected to incorporate smart pricing for effective demand side management and benefit both EV owner and grid operator. Many previous studies examine EV SC from the energy system operator’s perspective. Jian et al. (2017) proposed an efficient centralized valley-filling charging strategy that takes advantage of periods when electricity demand is below supply for EVs using capacity margin index (the surplus power of the electricity network) the charging priority index (the charging priority of the n-th EV at the k-th time slot), and found that it can alleviate the negative impacts arising from the EV charging loads on power grids. Zhang et al. (2014) proposed a decentralized valley-filling charging strategy using day-ahead pricing scheme, and achieved a 28% reduction in generation costs. Schneider et al. (2011) model different charging scenarios, including unmanaged, central network operator, and TOU scenarios. The study showed that controlled charging can reduce generation costs by 45%. Similarly, Kara et al. (2015) found a 24.8% energy cost reduction for EV owner while decreasing the contribution to the system peak load by 37%, using TOU pricing. However, Moon and Kim (2017) point out that when the EV charging is guided by the three-tier TOU rates, it yields an energy saving of 44.1% for EV owners from the uncontrolled scenario, but the system costs are not minimized since the TOU signal cannot reflect the system load state accurately. As a result, the shifted demand reaches another peak load around 24:00 and increases system generation costs. In Canada, Behboodi et al. (2016) proposed a charging strategy that enables EV owners to participate in realtime pricing (RTP) electricity markets to reduce their charging costs, where consumers can take advantage of inexpensive renewable resources, and utilities can avoid overloading the distribution system. These works inspired us to consider both TOU pricing and RTP schemes in our study. Another goal of SC management is to increase grid capacity for distributed energy generation such as photovoltaic (PV) and wind generation. Due to the scale and complexity of the distribution grid, many researchers choose a microgrid context to study the EVgrid interaction for renewable generation sources. Kavousi-Fard and Khodaei (2016), Aluisio et al. (2017), Bhatti et al. (2016), Mortaz and Valenzuela (2017), Anastasiadis et al. (2017a,b) found improvement in the microgrid’s viability from both operation costs and reliability perspectives, when coordinated control of distributed energy resources and EVs is achieved in a smart microgrid operation. Meanwhile, EV charging costs are reduced and RES self-consumption is improved. vanan der Kam and van Sark (2015) proposed a simulation of microgrid operations including solar panels, EVs, and load demand to study different EV charging control algorithms and the impact of SC and vehicle-to-grid (V2G) technologies on PV self-consumption and peak reduction. The PV selfconsumption with SC controls exceed 75%, representing a significant improvement compared to under 55% with no SC control. Ghofrani et al. (2014) proposes a collaborative strategy between the PV participants and EV owners to reduce the generation forecast uncertainties, and found the total PV power utilization increases from 72% to 97% under SC. Some researchers have considered the long term scenarios of EV-grid interaction, assuming high EV penetration and RES generation that is not confined in a microgrid. Fattori et al. (2014) study the combination of PV energy and EVs under uncontrolled charging regime and under the application of SC and V2G strategies, with a maximum of 50% EV market penetration. They found that intelligent control of EV charging could better accommodate the PV energy generation. Forrest et al. (2016) developed a simulation framework to study the potential impact of EV charging, considering high renewable penetration (PV and wind) and EV penetration (28.88% to 80% ). They found that EV SC can effectively reduce the system storage capacity and the required non-renewable energy capacity. These existing literature have examined the effectiveness of SC strategies in allowing privately-owned EV to harness renewable energy generation and increase grid capacity for RES. Inspired by the potential demonstrated in these studies to use SC to balance demand from EVs with RES, we also add a PV generation scenario in our analysis, to study the SC management with shared-used EVs.

3

Transportation Research Part D 78 (2020) 102184

T.Z. Zhang and T.D. Chen

Fig. 1. SAEV simulation framework.

3. Methodology 3.1. SAEV framework overview Chen et al. (2016) proposed an agent-based model to simulate SAEV fleet operations, which include charging station generation (phase 1), vehicle generation (phase 2), and vehicle operation (phase 3). However, the charging assumption in this model is rather simplistic and ignores vehicle-grid interaction. In order to better understand SAEV operation in a complex system where energy system and transportation demands are both prioritized, there is need for a modified simulation framework that allows advanced SAEV charging management and testing against various energy scenarios and the corresponding charging strategy. We adopt this agent-based model to study the effectiveness of SC strategies for a SAEV fleet by adding the active charging management framework to the operations phase (phase 3) (Fig. 1). To complete a full simulation, the simulator will sequentially execute phases 1, 2, and 3 of the framework. Phases 1 and 2 are described in Chen et al. (2016). At the beginning of phase 3 (day 1), the fleet generally has a positive net flow of electricity (fleet average state-of-charge [SOC] at the end of the day is higher than the SOC at the beginning). However, SOC reaches equilibrium after approximately 3 days. Therefore, the simulation will run continuously for 5 days and the results from the last day are reported. Three energy schemes are developed: TOU static pricing scheme (with respect to current retail energy market structure), RTP scheme (with respect to the future smart grid), and RES scheme (with respect to SAEV SC potential under PV generation). The TOU scenario in this study focuses on overall fleet performance and the impacts of EV battery size and charging infrastructure type, while RTP scenario targets the usefulness of a dynamic SC algorithm under fluctuating energy prices due to seasonal variations of energy supply and demand, load uncertainties, and extreme events in a smart grid system. Lastly, the RES scenario explores the impacts of battery size and charging rate on self-consumption potential under PV generation. In this study, an aggregator-based approach is used to manage SAEV charging. This approach assumes that centralized charging coordination can be achieved by the fleet operator. In other words, the fleet operator can dispatch any SAEV to charge as long as the vehicle is not occupied by a traveler. A charging management module that provides SC recommendations to the fleet operator is added in phase 3. Because the SAEV simulator updates at 5-min intervals, the charging recommendations (the ideal number of vehicles to charge in the next time step) are updated accordingly every 5 min. Then, the fleet operator sorts all available vehicles by their individual battery SOC, and assigns the SAEVs with the lowest SOCs to charge. Regardless of SC recommendations, vehicles that have less than 10 mile range or have been rejected for two consecutive trip requests due to insufficient range are still assigned to charging. Such implementation is intended to prioritize transportation service over charging management because energy costs have previously been estimated to only account for 10% of the overall SAEV operation costs (Chen et al., 2016). 3.2. Smart charging assignment algorithm The SC assignment framework is aimed at reducing electricity costs of the SAEV fleet by shifting SAEV charging demand to hours when electricity prices are low, while continuing to meet the transportation demand and maintaining mobility service quality. Under this logic, the first step is to determine the ideal number of charging vehicles during each time step (Eq. (1)).

Nt = St × XInf

(1)

where: Nt is the ideal number of charging vehicles at time step t; St is a smart charging signal to indicate the probability of a vehicle charging at time step t, St ∈ [0, 1]; and XInf is the charging infrastructure constraint (maximum number of concurrent charging vehicles). In the second step, the number of new vehicles that should be sent to charge (in addition to vehicles already charging in the 4

Transportation Research Part D 78 (2020) 102184

T.Z. Zhang and T.D. Chen

previous time step) is determined (Eq. (2)). (2)

Nn = Nt − Ne

where Nn is the number of new SAEVs sent to charge at time step t; Ne is the number of charging SAEVs that are continuing charging activity. The total number of concurrently charging SAEVs is constrained by the charging infrastructure capacity to regulate charging intensity (Eq. (3)).

Xinf = C ×

x × Dd nc × R charging

(3)

where C is a constant to ensure correct unit conversion; x is the charging multiplier to set the number of charging ports available to the SAEV fleet. Essentially, the charging multiplier is the inverse of the charging port utilization rate (x = 2 translates to 50% utilization rate); Dd is the total daily SAEV fleet energy demand in kWh; nc is the total number of simulation time steps per day, for the case study nc = 288; R charging is the SAEV recharge rate (mile/time step/veh), refer to SAEV recharge rates in Section 4.3. Finally, the fleet operator assigns a set of individual EVs that are low in battery to charging stations (Eq. (4)). The set of available SAEVs are sorted by SOC and the SOC of the Nn th vehicle becomes the recharge threshold (if Nn is greater than the total number of available vehicles then all available vehicles will be assigned to charge). Additionally, a minimum charging constraint of 80% SOC (no vehicle with greater than 80% SOC will be sent to charge) is imposed to avoid consecutive charging activities in a short period of time and to minimize battery life degradation.

SOCev ⩽ SOCNn, s. t . SOCev < 80%

(4)

where SOCev represents the SOC of each individual SAEV; SOCNn represents the charging threshold (the SOC of the Nn th available SAEV sorted in ascending order). With a two-tier TOU pricing structure, we set x to equal the ratio of the length of the day to the length of the off-peak hours, so that the optimal charging infrastructure can theoretically allow the SAEV fleet to shift all charging demand to off-peak periods. Since the off-periods in the TOU pricing structure are static, we use a binary function to assign value to St , the smart charging signal, based on whether the time step falls into the off-peak period. Unlike the TOU pricing structure where a charging multiplier can be derived directly, the RTP pricing structure requires an explicit charging multiplier as another independent control variable, due to price fluctuation and price prediction inaccuracy. Thus, we simulate a full range of charging multipliers (from 2 to 5), and use a sigmoid function to dynamically determine the value of St , as a means to adjust for potential price prediction errors and uncertainties (Eq. (5)). The sigmoid function essentially maps the real numbers space (electricity price rank Rankt ) into probability space (the probability of SAEV charging St ) when considering the charging multiplier and overall SAEV fleet SOC. This step is not required in the TOU scenario since the mapping is discrete and deterministic.

St = (1 −

1 ) 1 + e c1× Rankt + c2× SOCavg, t + c3× x + c4

(5)

where Rankt is the predicted percentile rank of electricity price at time step t within a given day, Rankt ∈ [0, 1], where 0 represents the predicted highest price period and 1 represents predicted the lowest price period within a day; SOCavg, t is the average SAEV fleet SOC at time step t , SOCavg, t ∈ [0, 100%];c1, c2, c3 and c4 are parameters to adjust the sensitivity of the charging index with respect to Rankt , SOCavg, t , and XInf . For the case study, these parameters are set to 20, −2, −2, and −5, respectively. Lastly, EV-RES coupling is considered here. Unlike the electricity price-based SC strategies where the goal is to minimize charging costs, the goal under the RES scenario is to manage fleet charging behavior to adapt to the renewable generation pattern and increase energy self-consumption rate. To demonstrate SC behavior within a high PV penetration environment, we specify the PV-based renewable system capacity so that the system is capable of producing the amount of energy needed for daily SAEV fleet operation, on average. Thus, the energy generated from the PV system at each time step becomes a soft constraint for fleet charging capacity (Eq. (6)), while the energy generated at peak time step is translated into a hard constraint for SAEV charging (Eq. (7)).

St =

Gt , s. t . Gmax

Xinf = C ×

288

∑

Gt = Dd

(6)

t=0

Gmax R charging

(7)

where Gmax is the renewable generation in kWh within the peak generation time step. Finally, the problem can be fitted under the proposed SC framework (Eqs. (1), (2) and (4)). The system efficiency is evaluated using the self-consumption metric, which is defined as the amount of PV generation consumed by SAEV fleet without the need for intermediate energy storage (Eq. (8)). 288

Self _Consumption =

∑t = 0 min (Dt , Gt ) 288

∑t = 0 Gt

(8)

where Gt is the renewable generation in kWh within time step t; Dt is the SAEV fleet energy demand in kWh within time step t. 5

Transportation Research Part D 78 (2020) 102184

T.Z. Zhang and T.D. Chen

Fig. 2. Coordinate transformation to Manhattan distance.

4. Case study In order to systematically examine and compare SC management under the three electricity schemes, we select the Seattle metropolitan region (King, Kitsap, Pierce, and Snohomish Counties) as the case study area. Regional transportation demand, electricity price, and PV generation datasets used in the simulation model are described in this section. 4.1. Travel demand data In the model proposed by Chen et al. (2016) simulating SAEV travel in Austin, Texas, trips are generated using a Poisson process with average rates based on population density and the National Household Travel Survey trip length and time distributions. In order to better reflect local travel demand in the case study area, we update the trip generation module with Puget Sound Regional Council (PSRC) Regional Travel Demand Model (Puget Sound Regional Council, 2017) data. We use a projection method to convert trip locations from a zonal system (traffic analysis zones [TAZs]) to a 2-d Cartesian coordinate grid system (0.25 mile by 0.25 mile cell identified by [X,Y] integer coordinates) so that it is compatible with the SAEV simulator that operates in Manhattan distance, as shown in Fig. 2. Trip generation rates are obtained by aggregating trips between each cell pair (origin-destination) by departure time at every hour. If the TAZ is represented by multiple cells, the trip generation rate is prorated accordingly; or if multiple TAZs fall into one cell, the trip generation rate of the cell is aggregated from all TAZs. After coordinate transformation, the 3700 TAZs in the case study area are represented by 193,600 cells in a 110 mile by 110 mile area. Approximately 12 million passenger trips are modeled on a typical weekday within the study area. 10% of passenger trips are randomly selected to be served using SAEVs (similar to Fagnant and Kockelman, 2015; Chen et al., 2016), reflecting what Shaheen et al. (2006) estimates as market potential for carsharing in the U.S. Average hourly link travel time by time-of-day from the travel demand model is used to model SAEV speed and represents the typical hourly network congestion conditions. Since the induced unoccupied VMT from SAEV operations is not reflected in the regional travel demand model, we assume the additional VMT (8.7% to 11.1% of SAEV VMT, translating to 1% increase of total region-wide VMT) is negligible in terms of impact on congestion. 4.2. Electricity data For the case study, TOU pricing rates come from Seattle City Light as the company serves roughly 40% of the TAZs in the study area. The 2017 general service TOU rates include two tiers: off-peak (0.0497 $/kWh, from 10 pm to 6 am) and on-peak (0.0746 $/kWh, from 6 am to 10 pm). In addition to the energy charge, a demand charge (based on peak demand usage within a given month) is applied at a rate of $0.27 per kW per month off-peak and $3.05 per kW per month on peak (Seattle City Light, 2018). RTP is a time-varying rate that generally applies to electricity usage on an hourly basis (U.S Department of Energy, 2018). Since RTP is not currently offered in the case study area, we translate the wholesale locational marginal price (LMP) into hourly energy prices to simulate a RTP environment. In the case study area, the hourly LMP is published by ColumbiaGrid (2018). Since it is unrealistic to model all possible price scenarios, we uniformly sampled 10 days of LMP data (the Wednesday of every fifth week) in 2017 and simulated SAEV operations based on the price profiles from these 10 days. To account for sampling errors, we sorted all LMP into four distinctive price categories: no peak (hourly price variance is less than 15 ("$"/ MWh)2 ); peak (the daily maximum price deviates more from the daily average compared to the daily minimum price); spike (daily maximum price exceeds 100 $/MWh, compared to an average price of 40 $/MWh); and off-peak (the daily minimum deviates more from the daily average compared to the 6

Transportation Research Part D 78 (2020) 102184

T.Z. Zhang and T.D. Chen

Fig. 3. Four categories of RTP days, each line represents one sample day (spike in log scale).

daily maximum price) and they account for 31%, 31%, 16%, 22% of all price data, respectively. Finally, we compare the sampled LMP data (sample rate at 30%, 20%, 20%, 30% on each category) against the categorized LMP dataset to ensure the sample is representative (Fig. 3). Note that the absolute costs from the TOU pricing scheme and the RTP scheme are not directly comparable because the TOU Price is a retail rate while the LMP is a wholesale rate. There is currently no utility-scale PV generation within the case study area. Thus, NREL solar integration data sets (National Renewable Energy Laboratory, 2018) is used to model RES generation patterns within the study area. The dataset provides simulated PV generation data for a one year period. Approximately 100 sites are randomly selected within the case study area to form a network of renewable generation sources that supply energy to power the SAEV fleet. For the case study, a generalized PV generation curve is obtained by averaging generation from all selected sites within one year, at 5-min intervals. The simulated PV system supplies 980 MW to the SAEV fleet in the case study. Each site’s generation capacity ranges between 0.2 MW and 22 MW. 4.3. EV technology assumptions Scenarios examined include combinations of two types of EVs and two types of charging infrastructure. The short range (SR) EV has a battery capacity of 40 kWh (similar to 2017 Nissan Leaf), while the long range (LR) EV has a battery capacity of 90 kWh (similar to 2018 Tesla Model 3). The charging rates vary by battery size and charger type. For SR EVs, level 2 (LV2) EV chargers can charge the battery at 7 kW/hr, while DC fast charger (FC) can charge the battery at 70 kW/hr. For LR EVs, LV2 chargers can charge at 20 kW/hr, while FC can charge the battery at 120 kW/hr (Smith and Castellano, 2015). Due to the nonlinearity of the charging rate, a 20% range reduction is applied when FC is used, similar to Chen et al. (2016) and Loeb et al. (2018). Department of Energy Office of Energy Efficiency and Renewable Energy (2018) estimates the current average EV efficiency is 126 MPGe (27 kWh/100 mi), an assumption applied here. To prepare for the additional energy consumption from vehicle automation hardware and software, the total estimated energy consumption is increased by 20%, assuming Medium Connected and Automated Vehicle subsystem power of 240 W estimated by Gawron et al. (2018). As a result, the final energy efficiency assumption is 0.33 kWh/mi for both LR and SR EVs. Combining the battery capacity and the energy efficiency assumptions, the final SAEV ranges at full charge are the following: 107 mi. (SR-FC), 133 mi. (SR-LV2), 218 mi. (LR-FC), and 273 mi. (LR-LV2). 5. Results 5.1. Baseline charging scenarios Two baseline charging scenarios are defined in order to evaluate SC results. The first one is the unmanaged (charge-as-needed) scenario, where SAEVs recharge if the battery level is below 20% SOC threshold. This strategy minimizes charging frequency and unoccupied VMT for charging. In the absence of electricity price signals, SAEV charging activities tend to concentrate during evening hours when the PM travel peak occurs. Fig. 4 shows the unmanaged charging profile for a LR-FC SAEV fleet, where charging activity 7

Transportation Research Part D 78 (2020) 102184

T.Z. Zhang and T.D. Chen

Fig. 4. Baseline charging behaviors (Left: Unmanaged; Right: Distributed).

is low overnight (midnight to 6am) but peaks at 7 to 8 pm. In such a scenario, the utility providers are responsible for adjusting energy supply to level the energy market imbalance and match the additional energy demand from EVs. Ultimately, this will increase grid operation costs and threaten grid reliability. The second baseline scenario is the distributed (minimum charging infrastructure) scenario. In this case, the SAEV charging is managed (in phase 3) so that the number of chargers needed to support the SAEV fleet is minimized. As a result, chargers are continuously used in order to meet SAEV energy demand. 5.2. Fleet charging behavior under TOU smart charging Fig. 5 shows the fleet charging behavior under TOU price SC scheme for the four different range and charging infrastructure combinations. Model results indicate SR EVs reach full charge (above 90% SOC) in all scenarios before 3 am and are unable to further take advantage of the off-peak electricity rate. When transportation demand increases in peak AM travel hours, SR EVs exhaust energy (under 50% SOC) shortly thereafter and requires intermediate battery recharge at midday. In other words, SR EVs do not have the battery capacity to avoid charging on peak under a typical two-tier TOU price scheme, especially when EV utilization is not uniformly distributed across the fleet (some vehicles incur significantly more daily VMT than others). As a result, SR EVs exhibit suboptimal charging behavior under TOU electricity pricing. LR EVs, however, show the ability to adopt a more desirable charging profile that complies with the off-peak charging schedule. In the LR-FC scenario, charging activities resume around 6 pm, suggesting that some EVs (about 3%) are running out of battery and need to recharge on-peak; while in the LR-LV2 scenario, on peak charging activities are negligible. Comparing these results with the charging profile in the unmanaged scenario, we find that SC can shift 77%

Fig. 5. Fleet charging behavior and SOC pattern (TOU). 8

Transportation Research Part D 78 (2020) 102184

T.Z. Zhang and T.D. Chen

(LR-FC) and 92% (LR-LV2) of on-peak charging demand to off-peak periods. In fact, if the SAEV battery capacity can reach 290-mile range, on-peak charging activities can be completely avoided in this two-tier structure. When comparing the different types of charging infrastructure (LV2 and FC), simulation results show that LV2 chargers require up to 78% of SAEVs to charge simultaneously in a single time step, compared 13% for DC FC. These results suggest that FC infrastructure requires significantly fewer charging ports compared to LV2 chargers under SC management. For example, for the SR-LV2 combination, more than 40% of the SAEVs are charging when the transportation demand is still high during the PM peak. This offer some explanation on why the SR-LV2 combination requires a fleet size that is 40% larger compared to other combinations (Table 2). 5.3. Mobility impacts and cost analysis under TOU smart charging In terms of SC impacts on SAEV mobility service, simulation results show that FC infrastructure with SC slightly increases passenger wait time, while LV2 charging infrastructure with SC can reduce average passenger wait times by up to 21.7%, compared to unmanaged charging (Table 2). This is due to the charging strategy assumed in the unmanaged charging base case, where EVs are sent to charge only when battery is running low, and remain in charging status until the battery is full. In practice, fleet operators are likely to anticipate travel demand and send EVs to charge when they are not in use (regardless of state-of-charge) and send charging EVs into operations as needed (even if batteries are not full). Deploying LV2 chargers magnifies this problem in the base case by holding up charging vehicles for a long period of time, thus deteriorating service level and increasing passenger wait time; while the impact on FC is less obvious since vehicles are recharged to full quickly. These results suggest that SC can shift charging demand and allow more SAEVs to stay in service during hours with high transportation demand. However, SC management increases unoccupied VMT, due to increased number of trips to charging stations. As a result, the average percentage of unoccupied miles for charging increases from 1.4% to 1.9% (of total fleet VMT) across all SC scenarios compared to the unmanaged scenarios (Table 2).With the exception of the SR-LV2 combination, all other managed charging scenarios are able to reduce the number of unserved trips compared to unmanaged charging (Table 2). The benefit of SC is that electricity costs per vehicle mile are reduced by 10% to 34.2% (Table 2) by maximizing off-peak charging activity. Within the four vehicle range-charging infrastructure combinations, SR EVs achieve a maximum 11.5% reduction in average electricity cost with SC, which reflects smaller-battery vehicles’ inflexibility in charging schedule (mandatory on-peak charging is unavoidable). In contrast, LR EVs achieve a minimum 32.8% reduction in average electricity cost. While simulation results offer insight into how SC with combinations of battery capacity and charging infrastructure impact fleet operations and electricity costs, a complete financial analysis is necessary to truly grasp the trade-off between vehicle/charger choice and total operational costs per mile. Table 1 summarizes capital and recurring costs associated with SAEV operations in three scenarios: high-, medium-, and low-cost. In recent years, the costs of EV, battery, and charging infrastructure has decreased as market EVs increased exponentially. Medium- and high-cost scenarios for EVs are based on the current retail values; and the low-cost based on a predicted battery cost decrease to 100 $/kWh (Howell et al., 2016). Smith and Castellano (2015) summarize the costs of EV supply equipment in the U.S., upon which we developed the high-, medium- and low-cost scenarios based on charging infrastructure in non-residential settings. However, the cost of Level 5 (full) vehicle automation (and associated maintenance [insurance, inspection, repair, etc.]) is highly uncertain for a SAEV fleet. For this analysis, various documents from vehicle manufacturers were reviewed to come up with the estimated cost of the on-board autonomous system and the annual maintenance costs. In terms of service life, each SAEV is expected to drive 231,000 miles before replacement, with the battery being replaced once during the vehicle’s service life, and the charging infrastructure is expected to have a 10 year service life, same as the assumptions in Chen et al. (2016). In the TOU pricing SC scenario, model results estimate vehicle-related costs (vehicle, battery, autonomous system, and maintenance) is still the dominant cost component (87% to 93%) compared to the energy costs (7% to 13%) on a per occupied-VMT basis (Table 2). While LR EVs are more effective at reducing electricity costs under SC compared to SR EVs, such cost savings cannot overcome the additional capital cost of larger battery EVs in the medium-cost scenario. For scenarios with SR EVs, distributed charging yields a lower per occupied VMT cost compared to unmanaged charging and TOU SC; for LR EVs, TOU SC outperforms both baseline scenarios. As previously mentioned, LR EVs favor SC with larger battery capacities to take advantage of the low off-peak energy prices. With LR EVs, SC can reduce the average costs per occupied mile by up to 3.1% (high-cost), 3.9% (medium-cost), and 6.2% (low-cost), compared to unmanaged charging. Meanwhile, SR EVs require a large percent of the fleet to charge simultaneously under SC, requiring additional charging ports whose costs outweigh the cost savings from the off-peak charging activity, making SC strategy suboptimal compared to distributed charging. With SR EVs, distributed charging can reduce the average costs per occupied mile by up to 2.2%, (high-cost), 2.5% (medium-cost), and 2.9% (low-cost), compared to unmanaged charging. Model results also Table 1 Vehicle & charger cost assumptions. Item

High

Medium

Low

Note

Short Range EV Long Range EV Autonomous System Level 2 Charger DC Fast Charger Vehicle Maintenance

30000 45000 25000 5000 25000 3000

30000 45000 10000 3000 16750 1500

25000 30000 5000 1000 8500 800

$/vehicle, similar to Nissan Leaf $/vehicle, similar to Tesla M3 $/vehicle, sensors and controller $/charger, non-residential (Smith and Castellano, 2015) $/charger, non-residential (Smith and Castellano, 2015) $/vehicle/year, include registration, inspection etc.

9

Transportation Research Part D 78 (2020) 102184

T.Z. Zhang and T.D. Chen

Table 2 Simulation results and cost breakdown, TOU.

Total trips Unserved trips Fleet size Number of chargers Daily VMT per veh (mi.) Avg % unoccupied VMT Avg % VMT for charging Average wait time (min) Electricity cost per veh mile, $

SR-FC umg

SR-FC dis

SR-FC offpeak

LR-FC umg

LR-FC dis

LR-FC offpeak

SR-LV2 umg

SR-LV2 dis

SR-LV2 offpeak

LR-LV2 umg

LR-LV2 dis

LR-LV2 offpeak

163 50,295 4,446 168.5 9.8% 2.1% 2.05 0.029

148 50,295 2,224 171.9 11.1% 3.0% 2.12 0.025

142 50,295 6,677 171.3 11.0% 2.6% 2.20 0.026

156 49,644 2,852 167.8 8.7% 1.1% 1.87 0.029

142 49,644 1,134 168.2 8.9% 1.2% 1.88 0.024

1288099 130 53 49,644 69,217 3,402 33,211 170.2 121.3 9.9% 9.1% 1.4% 1.7% 2.00 1.88 0.019 0.027

48 69,217 19,891 123.8 10.1% 3.3% 1.55 0.024

88 69,217 56,587 122.2 9.6% 2.1% 1.56 0.023

303 49,639 11,980 168.3 9.3% 0.9% 2.33 0.027

198 49,639 6,141 168.4 9.1% 0.9% 2.24 0.024

251 49,639 18,574 169.9 9.7% 1.4% 1.93 0.018

0.130 0.044 0.033 0.002 0.235 0.263

0.130 0.045 0.034 0.005 0.240 0.266

0.195 0.033 0.024 0.001 0.284 0.313

0.195 0.033 0.024 0.001 0.278 0.306

0.195 0.032 0.024 0.002 0.272 0.301

0.108 0.022 0.018 0.001 0.175 0.196

0.108 0.022 0.018 0.002 0.177 0.195

0.130 0.016 0.013 0.000 0.190 0.210

0.130 0.016 0.013 0.000 0.185 0.204

0.130 0.016 0.013 0.001 0.178 0.197

0.130 0.110 0.066 0.003 0.334 0.375

0.130 0.112 0.067 0.008 0.344 0.380

0.195 0.081 0.049 0.002 0.358 0.394

0.195 0.081 0.049 0.001 0.352 0.387

0.195 0.081 0.048 0.003 0.346 0.383

Costs Per Mile based on Average Cost Assumption ($) Vehicle Autonomous system Maintenance Charger Total cost per veh mile Total cost per occupied mile

0.130 0.033 0.024 0.002 0.223 0.247

0.130 0.032 0.024 0.001 0.214 0.241

0.130 0.032 0.024 0.004 0.220 0.247

0.195 0.033 0.024 0.002 0.288 0.315

0.195 0.033 0.024 0.001 0.279 0.306

0.195 0.032 0.024 0.002 0.273 0.303

0.130 0.045 0.034 0.003 0.243 0.267

Per Mile Costs based on Low Cost Assumption ($) Vehicle Autonomous system Maintenance Charger Total cost per veh mile Total cost per occupied mile

0.108 0.016 0.013 0.001 0.172 0.191

0.108 0.016 0.013 0.001 0.165 0.185

0.108 0.016 0.013 0.002 0.169 0.190

0.130 0.016 0.013 0.001 0.194 0.213

0.130 0.016 0.013 0.000 0.186 0.204

0.130 0.016 0.013 0.001 0.180 0.200

0.108 0.023 0.018 0.001 0.180 0.198

Costs Per Mile based on High Cost Assumption ($) Vehicle Autonomous system Maintenance Charger Total cost per veh mile Total cost per occupied mile

0.130 0.081 0.049 0.004 0.297 0.329

0.130 0.080 0.048 0.002 0.287 0.322

0.130 0.080 0.048 0.005 0.293 0.329

0.195 0.082 0.049 0.002 0.362 0.397

0.195 0.081 0.049 0.001 0.352 0.387

0.195 0.080 0.048 0.003 0.346 0.384

0.130 0.113 0.068 0.005 0.346 0.381

imply that FC outperforms LV2 chargers in terms of reducing total costs per mile under SC, by allowing faster turnaround between charging EVs, which reduces charging infrastructure capacity requirements. Overall, SR-FC combination with distributed charging strategy is able to offer the lowest operating cost ($0.241/occupied mile) compared to other medium-cost scenarios. However, the cost difference between the four vehicle-charger combinations are negligible in the low-cost scenarios. This result implies that if battery costs decrease as experts project, reducing electricity costs for operations will become a more significant factor in SAEV operator’s vehicle choice. It is worth noting that the cost estimates presented here only reflect operator-side costs, and do not account for changes in user costs and benefits (e.g. changes in passenger wait times and percent unserved trips) (Table 2).

5.4. Fleet charging behavior under RTP smart charging Based on the results from the TOU scenario, RTP SC scenarios were only examined with FC infrastructure assumption, due to LV2 chargers’ lack of responsiveness (a full charge requires 4 + hours) for dynamic SC. Fig. 6 shows two examples of SAEV fleet charging behavior under different charging capacity constraints, with a 24-h RTP profile that falls under the “peak” category. Broadly speaking, larger SAEV battery capacity is correlated with decreased on-peak charging intensity (number of concurrently charging SAEVs) due to the discordance between low-cost charging opportunity (generally overnight hours) and peak transportation demand (business hours). When the optimal charging opportunities are concentrated at night, the battery will need to store more energy to be able to serve travel demand through the PM peak. The results suggest that when the battery capacity is limited, the SAEV fleet struggles to store sufficient energy and is likely to resort to unmanaged charging behavior in order to maintain ability to serve transportation demand. This eventually leads to involuntary on-peak charging to meet the transportation demand, despite the high electricity prices on-peak, as seen in the RTP-SR scenario in Fig. 6, with heavy on-peak charging activity after 6 pm. The SC strategy based on RTP structure provides insight on the performance of SAEVs under various price category days (see Fig. 7). With LR-FC, if the price profile is categorized as ”peak” or ”off-peak”, the energy cost reduces as the charger capacity constraint is relaxed. However, if the grid experiences an energy price spike, the benefit of SC is not proportional to the number of 10

Transportation Research Part D 78 (2020) 102184

T.Z. Zhang and T.D. Chen

Fig. 6. Fleet charging behavior and SOC pattern (RTP).

charging ports. Such results are due to the different characteristics between valley-filling (concentrating charging overnight) and peak-shaving (avoiding energy price peaks) strategies. In order to achieve valley-filling, the fleet-average SOC oscillates with maximum amplitude. However, under RTP, the valley-filling strategy creates risk of involuntary on-peak charging if the battery capacity is limited and the vehicle usage rate is high. In the RTP scenario, the penalty of such activity is much higher than the potential rewards obtained from off-peak charging, since the peak electricity price can be higher than the average price by one order of magnitude. For days with small electricity price fluctuation (“no peak” days), model results suggest SC management yields at most 13% reduction in electricity costs, regardless of the charging infrastructure capacity. However, for the three other categories of RTP profiles, the weighted average annual energy cost savings with SC ranges from 36% to 43% for the LR-FC combination. While SC management was able to significantly reduce the energy cost with LR EVs under RTP, it is not effective for SR EVs. None of the scenarios with SR EVs are able to reduce the energy costs beyond 15%, under all types of RTP pricing categories. The weighted average annual energy savings with dynamic SC ranges from 1% to 12% for SR EVs, lower than the 14% reduction achieved by the distributed charging strategy. 5.5. Mobility impacts and cost analysis under RTP smart charging Like the TOU SC scenario, RTP SC strategy also increases unoccupied VMT due to additional travel for charging (up to 42%), which results in an overall unoccupied VMT increase of up to 15%, increasing overall unoccupied VMT from 8.7% to 9.9% (for LR vehicles) and 9.8% to 11.3% (for SR vehicles) of total fleet VMT, compared to the unmanaged baseline scenario. On average, RTP SC strategy also slightly increases average passenger wait time (up to 0.1 min/trip), compared to the unmanaged scenario. Because the RTP scenario is constructed based on wholesale (and not retail) rates, a comprehensive operator-side economic analysis (like the one performed for the TOU SC scenario) is not appropriate. Instead, an analysis is presented here focusing on the cost trade-off between electricity and charging infrastructure, to understand how infrastructure decisions affect total energy-related costs. In the LR EVs scenarios, model results suggest that the fleet is able to reduce its electricity costs significantly with SC management, with the total energy-related costs reduced by 43.4% on average when charger capacity is constrained at two times the minimum capacity (capacity required for distributed charging), compared to the unmanaged scenario, even while the cost per mile electricity is not the lowest. Dynamic SC is not effective for SR EVs under RTP for controlling overall energy-related costs, with the 11

Transportation Research Part D 78 (2020) 102184

T.Z. Zhang and T.D. Chen

Fig. 7. SAEV average energy price (RTP).

limited battery capacity being the primary bottleneck. As a result, adding more charging infrastructure capacity only increases the overall energy-related expenses. In the SR EV scenarios, the energy related costs are minimized if SAEV charging is distributed, resulting in a reduction in energy-related costs by 18.2% compared to the unmanaged scenario. In theory, providing more charging infrastructure capacity provides more opportunity for vehicles to charge during low electricity cost periods. However, overall energy-related costs also increase as the number of chargers increases due to the additional capital and maintenance cost of adding charging capacity. As charging intensity increases, accurate price predictions become key in reducing energy-related costs. Due to the discordance between charging opportunity (overnight) and transportation demand (business hours), once charger capacity exceeds a certain threshold, increasing charger capacity is only useful if the energy price prediction is accurate and the battery capacity is sufficient to store energy when the electricity prices are relatively low. For example, providing additional chargers to LR EVs will not reduce the average energy price when the price profile is categorized as ”spike”, due to the relatively low price prediction accuracy during extreme price spikes. For the case study, the RTP are predicted using a decision tree algorithm that trains on one year of historical price data. In practice, the prediction accuracy can be improved given more data sources and real-time grid status.

5.6. Fleet charging behavior under PV smart charging Unlike charging using the main grid, where SC often implies overnight valley-filling, PV generation is only available during the day and peaks at solar noon. As a result, unmanaged SAEV charging demand already overlaps with PV generation by 40% to 60%. From the simulated SC behavior with PV generation, we find that LR-FC is the only scenario where the SAEV fleet energy demand can match the PV generation at peak due to the combination of high charging rate and larger batteries for energy storage. SR EVs exhibit the largest gap between solar generation and charging demand, as the smaller battery capacity constraints the PV self-consumption. As a result, some portion of fleet electricity demand is shifted towards hours where PV generation is not available, thus requiring additional generation sources to supplement PV generation (Fig. 8). Overall, both battery capacity and charging rate constrain the effectiveness of SC management in a high-PV generation scenario.

12

Transportation Research Part D 78 (2020) 102184

T.Z. Zhang and T.D. Chen

Fig. 8. Fleet charging behavior based on PV generation.

5.7. Mobility impacts and system efficiency under PV smart charging In all PV SC scenarios, SAEVs can achieve a self-consumption above 80%, suggesting that SC is effective at allowing the SAEV charging behavior to adapt to the PV generation pattern. Among the scenarios simulated, LR EVs are able to reach higher selfconsumption (from 93% to 99%); while SR EVs are only able to reach a self-consumption rate up to 81%, due to their limited battery capacity. These self-consumption rates are comparable to results from SC studies of privately owned EVs (vanan der Kam and van Sark, 2015; Ghofrani et al., 2014; Fattori et al., 2014) and significantly higher than self-consumption rates of high PV generation in building energy consumption at 50% on average (Luthander et al., 2015). The results also show that under PV generation, Level 2 chargers increase wait time for passengers by up to 54% compared to the unmanaged charging base case (Fig. 9). Meanwhile, wait times are not affected when DC fast chargers are deployed. This is due to the low charging rate of Level 2 chargers, which forces the fleet operator to send a large portion of the fleet to charge in order to absorb PV generation at peak. In the SR-LV2 scenario, the fleet operator needs to dispatch up to 60% of the SAEV fleet to charge during the peak PV generation time step, leaving the remainder 40% of the fleet for transportation service. Lastly, all SC scenarios require 10% to 30% increase in unoccupied miles for charging and 24% to 55% additional charging port capacity, compared to the unmanaged charging scenario (Fig. 9). 6. Conclusions and limitations Experts predict SAEVs will be integrated into the transportation system in the near future. Meanwhile, future smart grids are expected to incorporate smart pricing for effective demand side management. In this study, SC strategies are proposed and evaluated for a SAEV fleet for TOU pricing, RTP, and PV generation schemes, based on a case study of the Seattle metropolitan region. Findings suggest that grid operators should design effective price signals to balance real-time SAEV charging energy supply and demand, and

Fig. 9. Mobility impacts of PV SC. 13

Transportation Research Part D 78 (2020) 102184

T.Z. Zhang and T.D. Chen

to deter additional peak charging loads. Meanwhile, the SAEV fleet operator should consider the roles of vehicle range, charging infrastructure type and capacity, in response to the energy pricing structure to maintain mobility operations and reduce total costs. Disruptive mobility technologies and trends will change the way urban transportation systems interact with the electric grid. Based on simulated SAEV travel and charging behavior in this study, the SAEV unmanaged charging peak occurs between 6 pm and 8 pm, which corresponds to the end of PM transportation peak. In the same time period (6 pm to 8 pm), the real-time wholesale electricity price suggests that the grid is expected to experience high demand under the regional energy use pattern. As such, the addition of the unmanaged SAEV charging demand will worsen the grid strain and require additional energy capacity to accommodate such increase in demand. Under TOU electricity pricing structure, SC strategies with LR vehicles can reduce energy costs for the SAEV fleet operator while maintaining or improving the level of mobility service, despite increasing unoccupied miles of travel to charge. Furthermore, model results suggest greater EV battery capacity is essential in allowing flexible charging schedule under TOU SC implementation. Fleet operator electricity cost savings can be even greater under RTP pricing (especially with LR vehicles). Model results show the fleet operator should be focusing on peak-shaving rather than valley-filling to maximize electricity cost savings, if the price is dynamic. Results from both price scenarios suggest that battery capacity plays an essential role in the SAEV-grid interaction (perhaps more so than charging rate). Larger batteries enable SAEVs to act simultaneously as mobile energy user and storage. But with current battery costs and static electricity pricing, fleet operators are not incentivized to adopt LR vehicles. Decreasing battery costs (and increasing battery capacity) and instituting dynamic electricity pricing are two keys to ensure SAEV fleets can adapt to the needs of the grid. Finally, EV charging management is a combination of demand side management and energy storage, because both charging schedule and charging intensity can be controlled. Due to this unique characteristic, SAEV charging can be managed to effectively absorb PV generation. Results from this study suggest that pairing SAEV service with PV generation can result in self consumption rates up to 99%. The findings are consistent with existing literatures on privately-owned EV-PV coupling, demonstrating the potential for this pairing. While this study provides a first examination into opportunities and challenges of SC management with a SAEV fleet, there are several limitations. First, the SAEV simulation does not take into account dynamic congestion (travel speeds are dictated by time of day), and added congestion caused by SC-induced VMT is ignored. Second, the RTP scenario samples pricing profiles from historic data, and does not consider the feedback impact of SAEV charging demand on RTP. Additionally, the charging stations are generated on a as-need basis within the study area, and the spatial distribution of charging stations are independent from electricity pricing scenarios. Lastly, spatial imbalances of LMPs and local transmission capacities are not modeled here. Author contribution statement Tony Zhang developed the theoretical smart charging framework, performed the analytic calculations, and executed the numerical simulations under the supervision of T. Donna Chen. Both Tony Zhang and T. Donna Chen contributed to the final version of the manuscript. Acknowledgment The authors would like to thank Brice Nichols from Puget Sound Regional Council for providing regional travel demand data. Appendix A. Supplementary material Supplementary data associated with this article can be found, in the online version, at https://doi.org/10.1016/j.trd.2019.11.013. References Aluisio, B., Conserva, A., Dicorato, M., Forte, G., Trovato, M., 2017. Optimal operation planning of v2g-equipped microgrid in the presence of ev aggregator. Electr. Power Syst. Res. 152, 295–305. Anastasiadis, A.G., Kondylis, G.P., Vokas, G.A., Konstantinopoulos, S.A., Salame, C.-T., Polyzakis, A., Tsatsakis, K., 2017a. Economic benefits from the coordinated control of distributed energy resources and different charging technologies of electric vehicles in a smart microgrid. Energy Procedia 119, 417–425. Anastasiadis, A.G., Vokas, G.A., Konstantinopoulos, S.A., Kondylis, G.P., Khalil, T., Polyzakis, A., Tsatsakis, K., 2017b. Wind generation and electric vehicles coordination in microgrids for peak shaving purposes. Energy Procedia 119, 407–416. Bauer, G.S., Greenblatt, J.B., Gerke, B.F., 2018. Cost, energy, and environmental impact of automated electric taxi fleets in Manhattan. Environ. Sci. Technol. 52 (8), 4920–4928. Behboodi, S., Crawford, C., Djilali, N., Chassin, D.P., 2016. Integration of price-driven demand response using plug-in electric vehicles in smart grids. In: 2016 IEEE Canadian Conference on Electrical and Computer Engineering (CCECE), pp. 1–5. Bhatti, A.R., Salam, Z., Ashique, R.H., 2016. Electric vehicle charging using photovoltaic based microgrid for remote islands. Energy Procedia 103, 213–218. Brownell, C., Kornhauser, A., 2014. A driverless alternative. Transp. Res. Rec. J. Transp. Res. Board 2416, 73–81. Chen, T.D., Kockelman, K.M., Hanna, J.P., 2016. Operations of a shared, autonomous, electric vehicle fleet: Implications of vehicle and charging infrastructure decisions. Transp. Res. Part A: Policy Pract. 94, 243–254. ColumbiaGrid, 2018. Planning and expansion overview. https://www.columbiagrid.org. Puget Sound Regional Council, 2017. Activity-based travel model: Soundcast. https://www.psrc.org/activity-based-travel-model-soundcast. Seattle City Light, 2018. 2018 rate information. http://www.seattle.gov/light/Rates/. Department of Energy Office of Energy Efficiency and Renewable Energy, 2018. Fuel economy estimates. Fagnant, D.J., Kockelman, K., 2015. Preparing a nation for autonomous vehicles: opportunities, barriers and policy recommendations. Transp. Res. Part A: Policy Pract. 77, 167–181. Fagnant, D.J., Kockelman, K.M., 2014. The travel and environmental implications of shared autonomous vehicles, using agent-based model scenarios. Transp. Res. Part C: Emerg. Technol. 40, 1–13.

14

Transportation Research Part D 78 (2020) 102184

T.Z. Zhang and T.D. Chen

Fattori, F., Anglani, N., Muliere, G., 2014. Combining photovoltaic energy with electric vehicles, smart charging and vehicle-to-grid. Sol. Energy 110, 438–451. Forrest, K.E., Tarroja, B., Zhang, L., Shaffer, B., Samuelsen, S., 2016. Charging a renewable future: the impact of electric vehicle charging intelligence on energy storage requirements to meet renewable portfolio standards. J. Power Sources 336, 63–74. Fulton, L., Mason, J., Meroux, D., 2017. Three revolutions in urban transportation. Gawron, J.H., Keoleian, G.A., De Kleine, R.D., Wallington, T.J., Kim, H.C., 2018. Life cycle assessment of connected and automated vehicles: sensing and computing subsystem and vehicle level effects. Environ. Sci. Technol. 52 (5), 3249–3256. Ghofrani, M., Arabali, A., Ghayekhloo, M., 2014. Optimal charging/discharging of grid-enabled electric vehicles for predictability enhancement of pv generation. Electr. Power Syst. Res. 117, 134–142. Graabak, I., Wu, Q., Warland, L., Liu, Z., 2016. Optimal planning of the nordic transmission system with 100. Howell, D., Cunningham, B., Duong, T., Faguy, P., 2016. Overview of the doe vto advanced battery r&d program. Jian, L., Zheng, Y., Shao, Z., 2017. High efficient valley-filling strategy for centralized coordinated charging of large-scale electric vehicles. Appl. Energy 186, 46–55. Kara, E.C., Macdonald, J.S., Black, D., Bérges, M., Hug, G., Kiliccote, S., 2015. Estimating the benefits of electric vehicle smart charging at non-residential locations: a data-driven approach. Appl. Energy 155, 515–525. Karali, N., Gopal, A.R., Steward, D., Connelly, E., Hodge, C., 2017. Vehicle-grid integration, a global overview of opportunities and issues. Kavousi-Fard, A., Khodaei, A., 2016. Efficient integration of plug-in electric vehicles via reconfigurable microgrids. Energy 111, 653–663. Langton, A. Crisostomo, N., 2014. Vehicle - grid integration. california public utilities commission. www.cpuc.ca.gov/WorkArea/DownloadAsset.aspx?id=7744. Loeb, B., Kockelman, K.M., Liu, J., 2018. Shared autonomous electric vehicle (saev) operations across the Austin, Texas network with charging infrastructure decisions. Transp. Res. Part C: Emerg. Technol. 89, 222–233. Luthander, R., Widén, J., Nilsson, D., Palm, J., 2015. Photovoltaic self-consumption in buildings: a review. Appl. Energy 142, 80–94. Martinez, L.M., Viegas, J.M., 2017. Assessing the impacts of deploying a shared self-driving urban mobility system: an agent-based model applied to the city of Lisbon, Portugal. Int. J. Transp. Sci. Technol. 6 (1), 13–27. Moon, S.-K., Kim, J.-O., 2017. Balanced charging strategies for electric vehicles on power systems. Appl. Energy 189, 44–54. Mortaz, E., Valenzuela, J., 2017. Microgrid energy scheduling using storage from electric vehicles. Electr. Power Syst. Res. 143, 554–562. National Renewable Energy Laboratory, 2018. Solar integration data sets. https://www.nrel.gov/grid/solar-integration-data.html. Pratt, R., Kintner-Meyer, M., Balducci, P., Sanquist, T., Gerkensmeyer, C., Schneider, K., Katipamula, S., Secrest, T., 2010. The smart grid: An estimation of the energy and CO2 benefits. Schneider, S.J., Bearman, R., McDermott, H., Xu, X., Benner, S., Huber, K., 2011. An assessment of the price impacts of electric vehicles on the PJM market, a joint study by PJM and better place. Shaheen, S., Cohen, A., Roberts, J., 2006. Carsharing in north america: market growth, current developments, and future potential. Transp. Res. Rec. J. Transp. Res. Board 1986, 116–124. Shoup, D.C., 2005. The High Cost of Free Parking. Planners Press, American Planning Association. Smith, M., Castellano, J., 2015. Costs associated with non-residential electric vehicle supply equipment. Sperling, D., 2018. Three Revolutions. Island Press. U.S Department of Energy, 2018. Time based rate programs. https://www.smartgrid.gov. van der Kam, M., van Sark, W., 2015. Smart charging of electric vehicles with photovoltaic power and vehicle-to-grid technology in a microgrid; a case study. Appl. Energy 152, 20–30. Zhang, K., Xu, L., Ouyang, M., Wang, H., Lu, L., Li, J., Li, Z., 2014. Optimal decentralized valley-filling charging strategy for electric vehicles. Energy Convers. Manage. 78, 537–550.

15