© The Author(s), published by EDP Sciences, 2024

1 Introduction

As a signee to the 2015 Paris Agreement [1], the Japanese government has published and revised its carbon emissions reduction targets over the past few years, including the goal of achieving zero CO₂ emissions by 2050 [2]. Like many other countries, Japan aims to achieve these goals in part by decarbonizing the energy sector, as well as electrifying sectors such as transportation as part of the Green Growth Strategy [3].

As of April 2022, the country’s energy strategy towards 2030 is summarized in the 6th Strategic Energy Plan (6th SEP) [4]. Notably, the 6th SEP cites renewables as the top priority resource group, more than doubling solar PV and wind integration targets with respect to the 5th SEP, as well as introducing hydrogen/ammonia as a new resource to the planned energy mix for 2030 (Tab. 1).

Over the course of the 4th and 5th SEP, the foundation for integrating more variable renewable resources was laid through the introduction of an hour-ahead and renewable-friendly trading markets (which is to be continued more intensively over the timeframe of the 6th SEP), market access liberalization, as well as the vertical unbundling of transmission and distribution from generation [5, 6]. While the 6th SEP does not yet describe targets to move from more contract-focused to merit-order dispatch, the increased focus on (variable) renewables and market platform creation could suggest the emergence of a European-style system by 2030. The need for a system such as this, which is more flexible and competitive, has previously been discussed in the literature [7–11].

Japan aims for 20–30% of vehicles sold in 2030 to be battery-electric and around 3% fuel cell [12], with 100% of all vehicles produced in Japan by 2050 to be electrified [13]. With regards to specific drivetrains, the 2014 Strategic Roadmap for Hydrogen and Fuel Cells [14] aims for 800,000 fuel cell electric vehicles (FCEVs) by 2030. Additionally, 160 hydrogen refuelling stations (HRS) were planned for 2020, and 320 by 2025. These infrastructure developments are overseen by Japan H2 Mobility (JHyM), an organisation comprised of 11 companies [15]. For battery electric vehicles (BEVs), approximately 4.9 million BEVs are expected to be on the roads by 2030 [12, 16], with no specific targets for recharging infrastructure.

With this trend towards an electrified transport sector [17, 18], it becomes increasingly important to understand the hourly electricity demand curves of the various drivetrains and how they will impact the energy grid, in order to understand what would be the optimum maximum amount of renewables in the system and the optimal charging strategies [19, 20]. In this context, the CO₂ reduction potential of transport-integrated models has been extensively studied [21–25], together with the analysis of specific resource- and vehicle technology combinations [26, 27]. Studies have confirmed that electrified transport generally increases the penetration of intermittent renewable resources, [22, 24, 28–34], as well as helping to reduce the burden on the grid [34–39]. Specific drivetrains have also been found to support the integration of different resources more efficiently, with BEV electricity demand supporting solar PV integration particularly well [36, 37, 40–42], and FCEV electricity demand supporting wind more efficiently [21, 26, 35].

Refuelling/recharging infrastructure is an important demand-constraining factor, due to possible fuel/electricity supply bottlenecks. There may also be regional differences in these bottlenecks depending on charging/refuelling station density. Further, the various drivetrains require different infrastructure and supply chains. While BEV infrastructure connects relatively seamlessly to the electricity network, FCEV infrastructure requires separate conversion of electricity to hydrogen, and also transport and storage facilities (Lee et al., 2014; [43]. In contrast to BEV infrastructure, the hydrogen supply chain for FCEVs can be decoupled from the electricity supply system if the fuel is sourced through, for instance, steam reforming [44]. While this is currently a common hydrogen-generating process, it will not be further considered in this study, as the Japanese Basic Hydrogen Strategy lays out plans to shift toward a system of hydrogen production from electricity from around 2030 to reduce its carbon footprint and revitalise regional economies [16].

The positive impact of electrified transport fleets on renewable energy integration and reducing grid burdens is often conditional on the assumption of grid-optimised charge scheduling, especially – though not exclusively – for BEVs [34, 42, 45–48]. Studies that have scheduled electrified fleets non-optimally find that the outcome is less efficient, but overall still beneficial to decarbonisation goals [34, 38, 46]. One likely reason why unscheduled charging is modelled less frequently is that it is difficult to get real vehicle use data with which to generate realistic demand patterns from driver profiles [39]. This is especially true for BEVs, though studies in the past have used internal combustion engine driving data to estimate FCEV profiles [34].

Driver profiles and user data (including distance covered and amount of time spent charging/refuelling, as well as number of times spent charging/refuelling per week/day and variations between days) are currently still rare [17]. For instance, Taljegard et al. [47] collected data on BEVs in the Västraland region of Sweden, from which they identified 200 unique driver profiles. Also at the regional level, Iacobucci et al. [49, 50] used transport survey data for Tokyo city. Richards (2013) also highlighted how models treat driver profiles computationally to estimate transport demand: as it is computationally expensive to recalculate the charge and driving status for each vehicle every hour, some studies have aggregated the vehicle fleet into one entity, so that there is effectively only one driver profile and one vehicle [51, 52]. Similarly, Tarroja et al. [34], aggregated a representative dataset of Californian vehicle travel data to represent the fleet of BEVs. As a result of this general lack of data and heavy reliance on assumptions, demand profiles for alternative fuel vehicles tend to be quite different between studies, even for the same region (see, for instance, the BEV profiles estimated for California by [34, 53, 54]. In Komiyama and Fujii’s [30, 45] Japanese transport-integrated modelling study, the demand for electric vehicles is assumed to remain constant every day of the year and charging only takes place overnight (between 00:00 and 08:00) to minimise the computational cost. A different modelling approach has been pursued by Iacobucci et al. [49, 50], in looking at an autonomous, shared vehicle fleet in the Tokyo area. The role of a “profile” to make charging decisions is limited in this approach, as the vehicle can generally charge any time of the day in between trips. The authors model the battery charge levels of individual vehicles and regulate charging requests by, for instance, optimising for solar rooftop and wind generation availability [50]; or by requesting charging when an individual vehicle values electricity higher than its current price (the perceived value is inversely related to battery depletion levels) and by disconnecting from the charger at the price equilibrium point [49].

Among transport-integrated energy system modelling studies there is a focus on the regional (sub-grid) level [55–57], or the grid is ignored when trying to integrate transport demand [45]. While Taljegard et al. [47] do consider the grid and transport demand, their study only consists of two regions. Further, studies either assess the impact of integrating one-vehicle technology [41, 42, 45, 47, 49, 50, 58] or the comparative performance of different technologies on specific outcomes [28, 34], while few studies model more than one technology in the system concurrently [55, 56]. Despite the lack of concurrent transport-integrated energy system modelling studies, the value of mixed-fleet studies in identifying synergistic technology mixes [27, 34, 38, 59], as well as challenges due to interactions between components [21] has been recognised.

Based on these gaps, this study has two objectives. First, to estimate the transport demand curves for BEVs and FCEVs in Japan, based on the 2030 target fleet size and infrastructure. Second, to estimate the impact on the energy mix that will result when multi-drivetrain transport demand is integrated into a country-level energy system with a transmission grid. The overall contribution of the paper resides in the integration of a system that can accurately reproduce the electricity production of intermittent sources, with both a transmission grid that can meet the demands of energy of the different regions of Japan and also of BEVs and FCEVs. While some simplifications exist at different levels within he model, the authors hope that EnSym will eventually be able to provide policymakers with an easy-to-use tool that can help see the outcomes of different strategies and policies.

2 Methodology

2.1 Model overview

EnSym was programmed in VBA by the authors, with the data being compiled in Python 3 and stored in an SQL database (see Knuepfer et al. [8]). The model simulates both traditional and transport-based hourly electricity demand, as well as all major electricity supply and storage sources. It can also adopt different dispatch hierarchies and take into account inter-regional electricity transmission limitations in the target country. The model so far has only been applied to the case of Japan, although it should be noted that it simulates the sub-division of regions and major metropolitan areas (MMAs). All prefectures in Japan are included in the model except Okinawa, an island archipelago to the south of the country that is not connected to the main grid.

Once the dispatch hierarchy, the installed capacity of each resource in each region and the target share of each resource in the energy mix have been set, the main model workflow basically consists of three steps. First, the hourly potential generation capacity and electricity demand for each region are estimated. Second, based on these estimates, each region sets its own hourly resource generation targets to meet the annual desired target share set by the user of the software. Third, electricity is traded between the regions to fulfil electricity supply targets and avoid blackouts via the transmission grid. Grid trading and balancing are conducted taking into account spatial proximity between demand and supply as well as the direction of flow, with the model not allowing for looping of transmission volumes. For further details on EnSym, see Knuepfer et al. [8].

In this study, the authors expanded the EnSym model to include electrified transport sector electricity demand, which was estimated using Python 3 and C++. EnSym thus improves on previous combined transport and energy system models which did not include transmission grid constraints (e.g. [45, 55–57]. While Taljegard et al. [47] do include both, these authors only consider one type of drivetrain (BEVs) and their system is subdivided into only two regions. Table 2 highlights the operational parameters on which EnSym improves on from those in the existing literature.

2.2 Electricity supply

2.2.1 Installed capacity targets

On the supply side, all major electricity-generating and storing resources relevant to the year 2030 are simulated, including utility-scale batteries and pumped hydro storage. The Ministry of Economy, Trade and Industry (METI) has outlined a target of installed generation capacities for 2030 for most resources at the country level [60]. This includes 88 GW solar PV, 1.4–1.6 GW geothermal, 13.3–15.3 GW wind, 7.2 GW of biomass, and focused growth of small- to mid-scale conventional hydropower plants from 9.7 GW to 10.94–11.65 GW. The installed capacity of thermal power will be reduced while maintaining an “appropriate thermal portfolio” [60]. Given the unchanged generation share target for nuclear power between the 5th and 6th SEPs, the authors assume that the target capacity for this resource will remain at 38 GW [61]. Where ranges were published, the authors chose the mean value.

Among the planned resource capacities, the authors chose to slightly alter the installed capacity of nuclear power and solar PV. Of the planned 38 GW nuclear power, only 33.2 GW are likely going to still be operational, with no additional new plants having made enough progress to expect them to be operational by 2030 (see the Appendix). The planned 88 GW of solar PV capacity by 2030 seem to be an underestimation, based on the most recent information on installed and approved solar PV capacities under the FiT scheme, which is 139.12 GW [62]. Therefore, the authors amended the target solar PV capacity to 139.12 GW for 2030. The wind resource includes both on- and offshore capacity, in line with the government target of 10 GW [60]. As no capacity targets for LNG, coal and oil were published as of April 2022, these resource capacities were assumed to remain unchanged from the 2021 estimate (see Knuepfer et al. [8]), which is a conservative estimate, given the planned reduction in their generation target shares.

The regional pattern of the distribution of resource capacities remains mostly unchanged from Knuepfer et al. [8] and, in the case of offshore wind, Chen et al. [63]. Any increase or decrease in capacity at the national level was distributed proportionally amongst regions based on this (see Appendix for regional changes). Table 3 summarises the resource distribution by region as projected for 2030.

Further, utility-scale batteries were also expected to be installed [64], with the capacity assumed to remain unchanged between 2020 and 2030, based on the lack of more detailed information available (Tab. 4).

2.2.2 Electricity generation potential

EnSym estimates potential electricity generation for each resource on an hourly basis throughout an entire year. The hourly electricity generation potential for solar PV and wind was calculated using meteorological data from one meteorological station per prefecture [65]; see Appendix for a list of the stations used). For thermal resources, the authors assume that no maintenance is required throughout the year -no fluctuation in the generation potential- and that the generation potential depends only on the capacity factor. Both pumped hydro and utility-scale batteries are determined to be driven by variable renewables, that is, wind and solar PV (which allows the authors to categorise these two types of storage as renewable). All pumped hydro stations and utility-scale batteries are assumed to have the same capacity throughout the country (see Knuepfer et al. [8].

2.3 Traditional electricity demand

EnSym considers both hourly traditional as well as transportation electricity demand from BEVs and FCEVs. The total annual traditional electricity demand for 2030 was estimated at 1065 TWh [3]. The hourly traditional demand curve for each prefecture was estimated by the authors using the following procedure: First, the annual expected electricity demand for 2030 [3] was distributed over each prefecture and month based on monthly electricity demand estimates from 2017 [66]. The resulting monthly electricity demand estimates for 2030 by prefecture were distributed over each hour of the month by using the hourly demand data for the Kanto region [67] as a representative weighting factor (see Knuepfer et al. [8].

2.4 Transport electricity demand

2.4.1 BEV charging infrastructure

BEV charging infrastructure is divided into private and public chargers. As official data [68] only provides information as to the number of chargers available in each prefecture and whether they are slow- or fast-charging, all public charging stations were assumed to have two dispensers (i.e. two cars can charge simultaneously). Private chargers were assumed to make up 10.6% of all chargers [69] and to be exclusively installed in single-family homes, with one charger per home and per car (among those that have private chargers).

The total BEV charging infrastructure distribution in 2018 [68] is also assumed to be representative of 2030. To distribute the chargers between each MMA and the rest of each region, the distribution of the human population in and outside of an MMA [70] was used as a ratio. While the share of private chargers per region is assumed to be fixed at 10.6%, it is also assumed to be unequally distributed between MMAs and the rest of each region, as most private chargers as of 2016 are estimated to be installed in detached houses, which are more often situated outside of the MMAs [69, 71].

It is further assumed that chargers are not capable of bi-directional flow; that is, electricity generation from vehicle to grid is not possible (V2G). The final distribution of the approximately 76,155 BEV chargers expected by 2030 is summarized in Table 5.

2.4.2 FCEV refuelling infrastructure

The number of Hydrogen Refuelling Stations (HRSs) per city is provided by CEV [72]. While a minimum of 320 HRS is expected to exist by 2030 in Japan [16], this work uses only the 135 HRS available in June 2020 (see Tab. 6).

The authors decided to not increase the number of HRS in 2030 based on 3 assumptions. First, in contrast with BEV chargers, the current HRS distribution is less likely to be geographically representative of the 2030 distribution due to a different expansion strategy [73]. Second, in contrast to BEV chargers, the number of dispensers per HRS can vary more. While the currently installed HRS can be assumed to have only one dispenser each (based on their hourly dispensing volume), the number of dispensers per site will likely increase as the fleet size grows, which further decreases the representativeness of the current HRS infrastructure for extrapolation to 2030. Lastly, the authors argue that the simulation can run smoothly with the current number of HRS and fleet size, as the current ratio of vehicles per HRS would be 5926 which – while above the current ratio of combustion engine vehicles to petrol stations in Japan (roughly 2033 vehicles per station [74]) – is still likely within the norm of a developed country in general (e.g. Germany’s ca. 4558 vehicles per station [75]).

HRS infrastructure is exclusively public, with only the hourly dispensing volume known, but not the number of dispensers per HRS. To estimate the latter, the authors used the HRS standardization chart defined by H₂ MOBILITY Germany (see Tab. 7).

It is assumed that each HRS provides H₂ at each FCEV’s optimal pressure (350 or 700 bar) in all cases and that it is exclusively generated through electrolysis using an Alkaline Electrolyser (Tab. 8).

2.4.3 The vehicle fleet

The entire fleet of alternative fuel vehicles (AFVs) is assumed to be comprised of two BEV models (Nissan Leaf II and Tesla Model 3) and one FCEV model (Toyota Mirai). The modelling of the drivetrain specifications is described in the Appendix).

The AFV fleet was distributed between MMAs and the rest of each region according to the distribution of the private chargers and the ratio between public chargers in metropolitan areas and the countryside [68]. The relation between infrastructure availability and AFV distribution in turn influences refuelling/charging decisions and therefore transport demand curves in each region. The total number of BEVs and FCEVs as of 2018 and that expected by 2030 is summarized in Table 9 (note that this does not include diesel and various hybrid vehicles).

The number and distribution of AFVs by region and MMA is estimated based on information on such vehicles purchased with subsidies [12, 68] and it is assumed that the concentration of refuelling/recharging infrastructure is related to vehicle purchase decisions in each region and MMA. For details regarding the estimation, please see the Appendix. The resulting number of BEVs and FCEVs is provided in Table 10.

Assumptions regarding the characteristics of the vehicle fleet include that:

• BEV (dis-)charge behaviour is constant, irrespective of the charge level (that is, the battery does not (dis-)charge faster or slower when it is already relatively full or empty, which is a conservative estimate);

• There is no battery degradation in the fleet;

• There are only two BEV and one FCEV vehicle technologies on the market (which also means that the assumptions about their efficiency will be outdated (and hence conservative) in a decade;

• No other mode of transport places further electricity demands on the grid (such as non-passenger vehicles or hybrid cars);

• All drivers behave in accordance with the average energy consumption per 100 km of their respective vehicle.

2.4.4 Driver profiles

To estimate the transport demand curve, as a first step some characteristics of vehicle drivers were considered (driving distance, timing, and charging/refuelling needs). The rules according to which drivers decide whether, when and how far to travel, as well as basic decision-making during the refuelling/charging process, are defined as the “driver profile”. Each parameter in the driver profile is constrained by boundary values and sub-divided into weighted range groups based on assumptions by the authors or previously observed patterns for each drivetrain (for details, please see the Appendix). Each vehicle is randomly assigned into one of the weighted range groups (based on [69], and within the assigned range group they are again randomly assigned a daily driving distance value. To further reduce the level of synchronicity between the vehicles in the fleet, every vehicle is randomly assigned an initial tank/battery fill level at the beginning of the simulation.

Two driver profile patterns were defined for each technology, namely “commuter” and “leisure”. The commuter profile is relatively well established in terms of driver behaviour, with assumptions regarding commuting periods being based on Zhang et al. (2020), and the interval between commutes being approximated from the day-time BEV charging duration given in Pareschi et al. [76].

For the case of the leisure profile Taljegard et al. [47] observed that driver behaviour throughout the day is relatively random: to represent the demand pattern observed from 429 GPS-tracked petrol cars, Taljegard et al. [47] extracted 200 unique driver profiles. Based on this, it can be assumed, that the 24 leisure profiles generated in the current study are unlikely to be fully representative (just as the commuter profile does not realistically cover the full bandwidth of actual commuter driving patterns). However, both profiles can be useful to broadly represent distinctly different types of drivers using a relatively computationally efficient method to do so and thus diversify the fleet.

2.4.5 Infrastructure use rules

As alternative fuel vehicles move through the day and use fuel (either electricity or electrolysed hydrogen), they eventually will request to connect with a charger or HRS. Apart from driving and resting-related rules, this simulation also sets five basic rules that govern the use of infrastructure and its constraints (for detailed definitions of the parameters, see the Appendix):

• First, the rules for refuelling/ charging describe the general process and duration of the connection.

• Second, infrastructure preference describes the rules by which types of chargers (home, public) or HRS will be preferred by a driver.

• Third, the queueing rule describes the decision-making procedure that vehicles undergo when their request to connect to refuelling/charging infrastructure is unsuccessful. The outcome of this procedure is that only those that most urgently need it will queue for charging/refuelling, so that the simulation completely avoids empty tanks/batteries (as at a specified depth of discharge drivers refuse to move further), and that it generates more variability in movement patterns, as drivers can forego or delay their planned activities to queue.

• Fourth, the reconnection intervals rule defines how long a vehicle needs to wait after disconnecting from refuelling/charging infrastructure before it can reconnect again. The purpose of this is to ensure that, in case of infrastructure bottlenecks, all vehicles wanting to connect get a similar chance to do so.

• Lastly, the movement delay rule defines how vehicles adapt their movement schedule (consisting of driving, parking and home hours) if an activity overlaps with queueing (which takes priority).

2.5 Transmission

EnSym assumes regional electricity transmission constraints for the year 2030 that are based on the grid connections that are outlined in OCCTO [61]. Additionally, a fixed 5% heat loss is assumed for each hourly transmission volume. Figure 1 provides an overview of the transmission constraints by region (See Knuepfer et al. [8].

Figure 1 Capacity of each major transmission line in the grid, with DC lines marked by triangles and AC lines without (Source: [61]. The dates inside the brackets indicate when the increased capacity should be available.

2.6 Scenario setup

To meet the first objective of this study, the authors generated the annual electricity demand curves for each vehicle drivetrain. To meet the second objective of this study, a business-as-usual (BAU) base case for the 2030 energy mix was simulated without electricity demand from transport and assuming a merit order dispatch of renewable resources, given the focus of prioritizing and maximizing renewable generation stated in the 6th SEP. Further, a transport-integrated scenario was defined to estimate the change in the energy mix resulting from the integration of transport demand.

The total estimated electricity demand for 2030 according to METI [3] is 1065 TWh, including transport-derived electricity demand. To estimate the expected electricity demand for the BAU (which does not include electricity demand for transport), the transport demand share therefore needs to be deducted. To do so, the authors assumed that the transport demand estimated through the present model (based on the 2030 fleet and infrastructure) is going to be representative of the 2030 transport electricity demand, and simply deducted it from the overall consumption for the case of the BAU.

The scenario setup is summarised in Table 11.

3 Results

3.1 Transport demand estimation

3.1.1 Vehicle fleet characteristics

The two driver profile patterns defined in the methods section (“commuter” and “leisure”) were applied to both FCEVs and BEVs. The model determined the number of unique subsets possible for each technology taking into account the movement plan based on the flexibility of driving hours, which gave a number of possible unique profiles of N = 24 for each driver type and N = 48 for each technology. This number was used for the FCEV fleet, though for the BEV fleet, an additional movement plan was generated for each subset, resulting in N = 50. This additional movement plan for each subset is a duplicate in terms of its planned movement schedule. However, the assigned daily driving distance and battery charge state at the beginning of the simulation are still unique to each individual driver. The additional movement plans were added to illustrate how the movement plans of the BEV fleet should be more diversified than that of the FCEV fleet (which would be otherwise identical). This can be observed when comparing the BEV (Fig. 2b) and FCEV leisure driver movement plans (Fig. 3b).

Figure 2 Combined planned movement of BEV commuter (a) and leisure (b) subsets throughout the day.

Figure 3 Combined planned movement of FCEV commuter (a) and leisure (b) subsets throughout the day.

Figures 2 and 3 provide an overview of the number of subsets for each driver profile and technology, which plan to move at a given hour of the day. Note that the number of subsets in Figures 2 and 3 adds to twice the number of subsets for each driver type, as each subset should drive twice a day.

For the BEV fleet (i.e. both commuters and leisure drivers) it can be observed that all vehicles are expected to park between 11:00 and 13:00 on weekdays, (and in the case of leisure drivers, also weekend) and that the entire BEV fleet (all four subset groups) is also expected to be at home between 22:00 and 05:00 h. This generally agrees with the localised charging peak around midday for BEVs that was observed by Pareschi et al. [76]. For the BEV fleet, most of the morning charging should arise from the leisure fleet, and the midday charging from the commuter fleet. Alterations are possible, as vehicles can shift their movement plans during the simulation based on the movement delay parameter (as described in the methodology). For FCEVs, the planned movement curves are similar, though less nuanced, due to the lesser number of driver profiles (N = 48 instead of 50). Refuelling timing is not as predictable as in the BEV fleet, due to shorter connection times with a HRS.

Apart from driving times, the second component of diversifying fleet behaviour is the randomisation of the other parameters shown in Tables A8 and A9 in the Appendix, such as for example the daily driving distances for each technology, see Figure 4. The line in each box shows the median daily driving distance for each vehicle model group and the “X” is the mean daily driving distance. Half of all observations fall within the boxes, with individual points outside showing outliers. It can be seen that 75% of all commuters drove less than 70 km/day, with a mean distance of 40–60 km/day. The leisure subsets saw greater variation in their 50% range, with the Nissan Leaf drivers highly concentrated in the 20–40 km/day range, while Tesla drivers had the most spread-out driving distances at 20 to around 100 km/day. This difference between technologies is also influenced by the specific maximum ranges.

Figure 4 Distribution of daily driving distance by car model for the commuter (a) and leisure(b) subset.

3.1.2 Estimating transport demand

The 2030 mixed technology fleet would be comprised of 4.9 million BEVs and 800,000 FCEVs, which would account for an estimated annual electricity demand of 55 TWh and 0.51 TWh respectively, which would mean that transport would comprise 5.21% of the total estimated electricity demand by that date (See Tab. 12).

The hourly demand curves for each vehicle technology would present no seasonal variations and are assumed to be repetitive on a weekly basis. Regional differences are based on the number of cars, as well as the density of charging/refuelling stations (see Fig. 5 for BEVs and Fig. 6 for FCEVs).

Figure 5 Weekly demand due to BEV region (2nd–8th January).

Figure 6 Weekly demand due to FCEV region (2nd–8th January).

FCEV demand peaks in the early morning (around 8:00 h), followed by a sharp drop as drivers begin their daily activities. It picks up again at the beginning of the afternoon, but is overall relatively flat throughout the day, with less pronounced diurnal variation than the BEV demand curves. This is partially due to the shorter refuelling duration, and partially due to the lower-than-expected HRS:FCEV ratio (the number of HRS was not increased between 2020 and 2030, as discussed in the methodology section). Hence, available dispensers are constantly occupied, even during the night. Lastly, the FCEV demand curve is also shaped by the fact that H₂ is assumed to be electrolysed on-site at the HRS, rather than being stored and distributed from centralized storage facilities or pipelines, which would result in a much more variable demand curve than the one observed in Figure 7.

Figure 7 Weekday (a) and weekend (b) FCEV demand for electrolysis directly at the station.

BEV charging on the other hand requires at least one -usually several- hours of parking. The flat demand overnight (Fig. 8) indicates that the number of drivers connecting during the night is steady (i.e. mainly – but not exclusively – due to those owning home chargers, who have guaranteed access). Further, as most of the BEV fleet is scheduled to be parked between 11:00 and 13:00 h on weekdays, this leads to a sustained midday peak in demand. The regional differences in electricity demand on weekdays and weekends derive in part from chance (as driver profile parameters are randomly generated), and in part from how urbanized the region is (i.e. a difference in the number of detached houses, home chargers, and vehicle density compared to the number of chargers).

Figure 8 BEV demand on Tuesday, 3rd (a) and Sunday, 8th (b) of January.

3.2 The BAU energy mix

Deducting the transport-derived electricity demand estimated in the previous section (55.81 TWh) for each hour from the total estimated electricity demand for 2030 (1065 TWh), the total BAU electricity demand was determined at 1009.19 TWh. Figure 9 shows the estimated BAU energy mix without transport demand, which indicates that several resources are likely to fall short of the 6th SEP target shares set for 2030 (see Fig. 9). For instance, solar PV only reached a 10.2% share (target: 14–16%), biomass 4.2% (target: 5%), wind 2.8% (target: 5%), geothermal 0.9% (target: 1%) and hydro 8% (target: 11%). These deficits were balanced by increased generation shares from oil (2.8% instead of 2%) and particularly LNG (30.5% instead of 20%).

Figure 9 The BAU energy mix (%) for 2030 without transport demand.

The energy mix varies each hour and day, particularly given the availability of variable renewable resources. For instance, despite the 9th of August being one of the days of the year with the highest electricity demands in Japan, PV production exceeds the planned 14–16% (see Fig. 10 and Tab. 13).

Figure 10 Hourly generation by resource (MWh) on the 9th of August.

3.3 The transport-integrated energy mix

For this scenario, total demand was estimated at 1065 TWh, i.e. including 55.81 TWh of transport demand. The resulting energy mix is summarized in Figure 11.

Figure 11 The energy mix (%) for 2030 including transport demand.

It can be seen that while the generation shares of coal, batteries, geothermal and solar PV remained basically the same or increased slightly, the shares of all other resources decreased, except for oil, which increased from 2.8% to 4% for 2030. The reduction in generation shares for most resources is partially due to the changes in demand between regions due to the introduction of transport demand. In regions where demand has increased significantly, resource capacities would need to be increased to maintain their generation shares with respect to the BAU. For instance, the energy mix on the 9th of August shows that while the shares of biomass, geothermal, solar PV, wind and LNG decreased, their overall generation volume increased with respect to the BAU scenario (see Fig. 12 and Tab. 14).

Figure 12 Hourly generation by resource (MWh) on the 9th of August.

Figure A1 EnSym balancing workflow for one hour and one resource in one region.

Regions with a relatively larger increase in demand in the transport-integrated scenario also have a relatively larger installed capacity of oil (Kanto and Kansai alone have over half of the installed oil generation capacity of Japan as of 2022). This suggests that the increase in oil generation was likely due to local requirements in only a couple of regions (most demand increase comes from the Kanto and Kansai vehicle fleets), which could not be met by more renewable resources due to insufficient local capacity and constraints in the transmission grid. This is also likely the reason why LNG generates an approximately 10% larger share in the annual energy mix than planned for in the 6th SEP.

3.4 Comparative global warming potential

As the target of the Japanese energy strategy is to have zero CO₂ emissions by 2050 [2], it is relevant to understand the global warming potential of the energy mix. Table 15 shows the global warming potential for utility-scale batteries [77], offshore wind [78] and all other major resources [79].

Table 16 provides an overview of the annual generation volume and share, as well as the global warming potential of each resource. While the total generation volume increased by 5.3% between the BAU and the transport-integrated scenario, emissions increased by 3.7%. The largest sources of emissions were coal (44.64% and 45.41% of total emissions for each scenario, respectively), LNG (37.73% and 37.78% of emissions respectively) and conventional hydropower (15.09% and 14.23% of emissions respectively). Of the total emissions increase from the BAU to the transport-integrated scenario, 96.4% were caused by coal and LNG and the other 3.6% were caused by nuclear, biomass, geothermal, oil, offshore wind and solar PV. The remaining resources did not increase their output. Lastly, the BAU overall performed slightly better (99.55% demand coverage) than the transport-integrated scenario (99.31% demand coverage), suggesting that the planned future installed capacity volume is neither sufficient nor is it distributed optimally to meet the increase in demand from electrified transport. At the same time, the increase in the generation of all renewable sources (except onshore wind and conventional hydro) suggests that electrified transport is generally compatible with renewables, which are prioritized in the 6th SEP.

4 Discussion

The electricity demand curves for BEV and FCEV vehicles were found to be distinct from each other, which mainly stemmed from different refuelling/charging patterns. Therefore, when studying the impact of transport demand integration into the energy system, it is important to estimate the total electricity demand curve as a composite of each technology’s individual demand. Compared to previous literature, the BEV electricity demand curve estimated in this study is highly day (rather than night-time) focused, due to the low availability of home chargers (10.6% of all chargers). This illustrates that the type of infrastructure available (public, workplace, private), as well as the rules around how it can be used, significantly influence the shape of the transport electricity demand curve. As some of these rules are based on legislation rather than infrastructure setup or technology, the demand curves may also differ between regions with similar infrastructure setups.

While the FCEV demand curve estimated in this study has some limitations, its morning and evening focus is likely to remain a prominent feature that sets it apart from the BEV demand curve. This also highlights that a composite transport demand curve of these two types of technologies would likely have several localised maxima throughout the day, which may in part exacerbate the current electricity demand curve at some times (e.g. around midday and the early evening), and even it out at others (e.g. the morning, when FCEVs are refuelling on the way to work). The case of the FCEV electricity demand curve also highlights the importance of explicit infrastructure modelling: the demand curve estimated in this study is based on the assumption that the HRS infrastructure electrolyses on-site and therefore FCEV demand for H₂ directly translates into instantaneous electricity demand (i.e. the production of H₂ using off-peak electricity and its storage for later use is not considered by the simulation). While this assumption is not unrealistic, it leaves out one of the major anticipated benefits of any H₂ network, which is the increased demand elasticity created by H₂ storage, which should be tackled by future research and improvements to the model.

The infrastructure setup for both BEVs and FCEVs can be expected to significantly influence the shape of the transport electricity demand curve. Based on this, policy-based infrastructure development targets for both HRS and BEV chargers would be useful. However, in the case of Japan, BEV charging infrastructure development (both public and private) does not seem to have policy targets. The authors therefore recommend that transport policy should determine infrastructure development targets for both private and public BEV charger development throughout the country.

To estimate the expected electricity demand for the BAU, the transport demand share was deducted from the total estimated electricity demand (1065 TWh) estimated by METI [3]. While the government of Japan has published estimates for the number of cars and infrastructure facilities of both BEVs and FCEVs by 2030, estimated electricity demand values are less conclusive. Data related to electricity consumption often focuses on targets for the fuel economy of electrified vehicles [12, 80]. One transport electricity demand estimate for 2030 could be identified [81]: p. 73), at 190 GWh for the entire electrified fleet over a whole year. Given that, among the electrified fleet, 4.9 million BEVs are expected and given that one full charge of, for example, a Tesla model 3 requires 82 kWh, the estimated value of 190 GWh would be equivalent to less than half a full charge per BEV over the entire year. This seems highly unlikely and therefore the authors refrained from using this value in the present simulation. Instead, the authors assumed that the transport demand estimated through the model (based on the 2030 fleet and infrastructure) was representative of the 2030 transport electricity demand. It is obvious that more clarity is needed regarding how the electricity demand of various sectors (particularly transport) will be estimated in the future, given that research such as that outlined in the present paper relies heavily on such assumptions. While the authors acknowledge the limitations of the assumptions they followed when estimating the overall distribution of the transport and total electricity demand, it is also outside the scope of this paper to arrive at more accurate projections.

The demand for electricity incurred by 5.7 million BEVs and FCEVs (9.25% of all passenger vehicles expected to be on the roads of Japan by 2030) was thus estimated to comprise around 5.21% of total electricity demand by 2030. It can therefore be expected that as the electrification of the transport sector progresses, its demand share will become a significant factor when balancing the energy system. The addition of transport demand was most beneficial for solar PV and offshore wind, as well as oil, while all other resources either maintained their target shares or reduced them slightly. Two significant factors contributing to this were the country-wide spatial resource distribution and insufficient resource capacity to further increase generation. Spatial resource distribution favoured oil in particular, as regions with oil thermal generating capacity tend to be ones that also have MMAs, where the increase in demand due to the transport component was most pronounced. As electrified transport becomes more widely adopted and less concentrated in MMAs this effect might decrease.

EnSym necessarily relies on simplifying assumptions in some key areas. First, thermal power plants are assumed to run without interruption (i.e. no maintenance schedule is considered). Further, no thermal resource was permitted to deposit excess generation in storage (either pumped hydro or batteries). Additionally, among RES-E, only solar PV and wind resources were permitted to be deposited in storage, as geothermal and biomass capacities are highly controllable and for the most part deployed at low shares. As a result of such limitations, the available storage capacity is under-utilised. Further, all pumped hydro stations and batteries are assumed to have the same operational parameters, though this is likely to be quite varied in reality.

Second, the driver profiles (commuter and leisure) significantly rely on assumptions, as the amount of observed data is very limited. More data observations are necessary to derive a clearer picture of the different types of driving and charging behaviour, particularly for technologies with lengthy charging times. In the future, a self-driving fleet component should also be considered, as well as more profiles with more instantiations compared to the current 50.

Third, the FCEV electricity demand curve is limited by the assumption that no new infrastructure would be built between May 2021 and 2030, which means that in the simulation the HRS are constantly used, even in the middle of the night, leading to a relatively uniform curve. However, a HRS can comprise one dispenser (all current HRSs in Japan have one dispenser), or up to four, and it is likely that in the future more dispensers will be installed. Further, it is a simplification to assume that HRS infrastructure only works with on-site generation, rather than also (or even mainly) having a centralized distribution network.

Fourth, the BEV charger fleet is assumed not to be capable of vehicle-to-grid (V2G), despite this being anticipated as an important demand-side management mechanism by 2030. In terms of the impact of V2G on the current results, the authors expect that including V2G would increase the share of renewable-derived electricity from storage and decrease oil and LNG generation during the day. Further, V2G would likely increase the average number of hours per day spent connected to a charger and may increase the share of evening charging.

As a next step, vehicle to grid (V2G) and a centralized H₂ storage and distribution network should be added to the model. Further, it would be useful to differentiate private solar PV rooftops from other solar PV installations, collect observed data on driving and charging patterns of AFV owners (specific for Japan) and add partial grid outage scenarios to the simulation, addressing the potential for disaster resilience through combined energy and transport systems. Also, exploring how the relationship between energy system security, renewable integration and an AFV fleet changes as that fleet grows larger would be interesting. This would be an important benchmarking task to identify at which developmental stages renewable technology installation targets, storage, infrastructure and vehicle fleet development need to coalesce to maintain supply security while increasingly electrifying and decarbonizing the energy sector. At a more fundamental level, it is important to improve the understanding of how transport demand behaves based on different types of infrastructure and rule setups, as well as from the perspectives of a differentiated set of users, especially as new ones (e.g. autonomous vehicles) arise.

5 Conclusion and policy implications

The main conclusions of this work can be summarized in three points. First, the variation in electricity demand for transport throughout the day is influenced by physical infrastructure, the rules by which it can be used, technology, and how its drivers use it. Therefore, no single technology is representative of the transport electricity demand curve. To apply transport electricity demand curves from one administrative region to another, the related infrastructure, rules and driver profiles should be known, though there is scarce literature and data to inform models on this. Further, this highlights the need for more data collection that can provide specific information on these components and their daily fluctuations and geographic considerations.

Second, this study found that there is potential for symbiotic co-development between alternative fuel vehicle (AFV) infrastructure and RES-E, which is an important contribution to existing literature. However, to fully harness this potential it may be beneficial for policymakers to not only see the energy, transport and hydrogen strategies as related but instead to fully integrate them through infrastructure and resource development plans and pathways that incorporate two or all three of these strategies.

Third, this study has expanded the modelling literature by improving the EnSym model to simultaneously integrate a country-level energy system with a transmission grid and multi-technology transport sector demand. However, while the analysis conducted provides new insights into energy and transport sector developments, there are several model limitations and parameter functionality requirements that should be tackled to further improve the quality of the evaluations. It would be particularly interesting to extend the model to consider other types of vehicles, such as for example an intermediate type of vehicle, such as a utility type of FCEV. Also, the authors would like to consider transforming the current model into a Monte Carlo simulation, which should be possible given that each simulation run only takes a handful of minutes.

Acknowledgments

A part of the present work was performed as a part of activities of the Research Institute of Sustainable Future Society, Waseda Research Institute for Science and Engineering, Waseda University.

Funding

The work was financially supported by Japan Science and Technology Agency (JST) as a part of the Belmont Forum, Re-Energize DR3 project, Grant Number JPMJBF2005, and by Japan Society of Promotion of Science (JSPS) KAKENHI Grant Number JP20KK0107 and JP19K24677.

Conflicts of interest

The authors declare that all the work described in this manuscript is purely academic and that there are no conflicts of interest of any kind.

References

Appendix

Nuclear power target capacity adjustment

The nuclear capacity in the model was reduced from the expected 38.04 [4] to 33.2 GW. The reported installed nuclear capacity in 2018 [61] includes 4.4 GW installed at Fukushima Daiichi and Daini, which have since been completely scrapped [82]. A further 810 MW capacity has been removed, as is not listed in the annual status update provided by the Japan Atomic Industrial Forum [82], and it is therefore unclear where that capacity would come from. Further, 8.2 GW of TEPCO-owned capacity is assigned to Hokuriku, as this is the region where it is located in. While 6.5 GW of Kansai’s capacity is located in Hokuriku, these power stations connect to the Kansai regional grid directly and are therefore assigned to Kansai. Table A1 provides an overview of all nuclear power plants considered.

Geothermal regional capacity development projects leading up to 2030

By 2030, five geothermal plants totalling 66.1 MW are scheduled to connect to the energy grid (Tab. A2). Given the 500 MW already installed in 2022, the total, regionally distinguishable installed capacity for geothermal is 566.1 MW. The remaining 933.9 MW scheduled to go online by 2030 were distributed over the regions proportionately.

Solar PV and wind data measurement stations

Table A3 lists the weather stations used to measure solar PV, onshore wind, and offshore wind potential generation [65].

BEV charging infrastructure makeup and distribution

The total BEV charging infrastructure by prefecture for 2030 , including public and private chargers, is given as:(A1)

Where denotes each prefecture.

The regional distribution of BEV charging infrastructure by region is given by the sum of chargers in each prefecture that constitutes a region:(A2)

It was assumed that the current infrastructure distribution will be representative of 2030, based on the data provided by NEV [68].

The number of chargers located in each city is given as:(A3)

Where denotes the human population of a location. The population by city of each city is provided by IPSS [70]. The total number of chargers located in an MMA is given by the sum of the chargers located in the cities that comprise each MMA.(A4)

Lastly, the number of private chargers in each prefecture is given as:(A5)

where denotes a constant for the share of private chargers, which takes a value of 10.6% [69].

The number of private chargers located in each city is given as:(A6)

where denotes detached households, which are provided at the city-level by Stat [71].

The number of private chargers located in an MMA is given as the sum of private chargers located in the cities comprising each MMA:(A7)

Drivetrain specifications for each car model

Tables A4 and A5 summarize the parameters used for the representative BEV and FCEV vehicles, respectively.

AFV fleet distribution by region and MMA

The number and distribution of BEVs and FCEVs purchased with subsidies as of 2018 are provided by NEV [68]. The distribution factor of AFVs purchased through the subsidy scheme by prefecture is given as:(A8)

The distribution factor provided by equation (A8) is assumed to be representative of the entire fleet for each of these technologies, and it is also assumed that demographic changes will not affect the average number of vehicles per 100 households in each prefecture. Provided these assumptions hold, the distribution of the 2018 fleet for each AFV technology by prefecture can be estimated by:(A9)

where denotes the total number of AFVs in each fleet, which is provided by METI [12].

Based on the 2018 fleet of each technology, the 2030 fleets by prefecture can be estimated as:(A10)

To estimate the distribution of vehicles between MMAs and the rest of each region, it is assumed that the distribution of each technology’s refuelling/charging infrastructure is also representative of the vehicle distribution. The number of AFVs of each technology in each city is given as:(A11)

Where denotes the number of chargers or HRS for each technology.

The number of AFVs of each technology in each MMA by 2030 is given by the sum of the number of cars for each technology in the cities comprising the MMA:(A12)

Driver behaviour parameters

The daily driving distance is constrained by:(A13)

where denotes the minimum daily driving distance , which takes a value of 10, and denotes the driving duration , which takes a value of 2 for both the commuter and leisure profiles in this study. Lastly, denotes the typical speed limit on Japanese highways, which takes a value of 90.

The actual daily distance is estimated considering two things: first, a weighted driving range distribution was created (Tab. A6), accounting for the driving distance distributions observed by Sun [69].

The driver profile

Table A7 provides the driver profile generated for each driver, together with a description of each variable and its permissible range and unit.

Tables A8 and A9 provide the setups for each driver type (“commuter” and “leisure”) adopted in the present study.

Infrastructure use rules

Note that the definitions for queueing and movement delay apply to BEVs and FCEVs equally (Tab. A10).

Simulation workflow of EnSym

The main simulation workflow of EnSym is summarised in Figure A1 (see [8], which indicates the balancing procedure for each resource, in each region, for each hour. The workflow can be divided into the following tasks:

1. Regional estimation of available capacity (or whether a capacity deficit exists at a given hour) by resource,

2. Hourly resource target setting.

3. Estimation of excess capacity or deficits in a given hour that can be traded between regions depending on grid constraints.

4. Aggregation and tabulation of results for all regions.

All Tables

All Figures

	Figure 1 Capacity of each major transmission line in the grid, with DC lines marked by triangles and AC lines without (Source: [61]. The dates inside the brackets indicate when the increased capacity should be available.
In the text

	Figure 2 Combined planned movement of BEV commuter (a) and leisure (b) subsets throughout the day.
In the text

	Figure 3 Combined planned movement of FCEV commuter (a) and leisure (b) subsets throughout the day.
In the text

	Figure 4 Distribution of daily driving distance by car model for the commuter (a) and leisure(b) subset.
In the text

	Figure 5 Weekly demand due to BEV region (2nd–8th January).
In the text

	Figure 6 Weekly demand due to FCEV region (2nd–8th January).
In the text

	Figure 7 Weekday (a) and weekend (b) FCEV demand for electrolysis directly at the station.
In the text

	Figure 8 BEV demand on Tuesday, 3rd (a) and Sunday, 8th (b) of January.
In the text

	Figure 9 The BAU energy mix (%) for 2030 without transport demand.
In the text

	Figure 10 Hourly generation by resource (MWh) on the 9th of August.
In the text

	Figure 11 The energy mix (%) for 2030 including transport demand.
In the text

	Figure 12 Hourly generation by resource (MWh) on the 9th of August.
In the text

	Figure A1 EnSym balancing workflow for one hour and one resource in one region.
In the text