The conditions for flipping the vehicle fleet to electricity are ripe in New Zealand: the electricity supply is largely renewable, policies incentivize the shift to electric vehicles (EVs), and customer interest in EVs is high. During this transformation, the home is expected to become the new fuel station. Integrating EV charging seamlessly into the electricity network is critical to guarantee that the vision of electric transport is delivered in a cost-efficient, equitable, and reliable supply. To understand customer behavior and network requirements for at-home charging, Vector, New Zealand’s largest electricity distribution company, has been carrying out an EV smart charging trial with 200 EV owners over the last three years. The trial established data-driven EV charging requirements that have been implemented into network planning processes. Furthermore, the trial demonstrates how smart charging can integrate EVs within existing network capacity while also delivering customer satisfaction. The trial provides a blueprint for how electricity networks can harness data, monitoring, and smart control to transition to the new energy future.
A growing list of EV models and new government rebates have recently accelerated EV adoption in New Zealand, with EV registrations (i.e., new and imports from overseas) in 2022 hitting 10% of the total light-duty vehicles registered in New Zealand during that year. At the heart of New Zealand’s vehicle fleet electrification is Auckland, which is home to a third of the New Zealand population and 45% of its EV fleet.
In 2021, the New Zealand government introduced an ambitious plan to transition 30% of the country’s light vehicle fleet to electricity by 2035. To achieve this, an incentive scheme has been introduced to subsidize low-emissions vehicles (including EVs) and penalize high-emissions ones. Within a year of the policy launch, EV numbers on the road had doubled. Customer research supports continued uptake growth, with an ever-growing percentage of people indicating a strong intention to purchase an EV as their next vehicle.
In terms of electricity generation, New Zealand is already largely renewable (about 85% annual production, mainly from hydro). With many more renewable developments in the pipeline, New Zealand is expected to have 90%–95% renewable generation within the next decade. Shifting the transport sector from oil to clean electricity is, therefore, one of the easier and more effective ways for New Zealand to meet its decarbonization commitments while also improving energy security and reducing local pollution levels.
Vector is New Zealand’s largest electricity distribution company, supplying Auckland’s 1.7 million residents. Auckland has been steadily growing for the last 20 years (∼2% per annum on average), and this trend is expected to continue. Although the bulk of Aucklanders live in urban areas, the region is large (i.e., low population density) and spread out, with sizable rural areas and hard-to-access native forests.
Unlike some cities of comparable size in other countries, Auckland does not have a mature or citywide public transport system, meaning that vehicle ownership is high. Approximately 1.3 million light-duty vehicles are in the region, serving a population of 1.7 million. The spread-out nature of the city, which is not helped by it being located on an isthmus, also means that Aucklanders rely on private transport.
The majority of Aucklanders who have EVs charge their vehicles at home. This situation has partly to do with cost as well as with Auckland having a very low rollout of public EV chargers. Additionally, customer research provides a strong indication that this trend for domestic charging will continue for at least the next five years.
Auckland homes are typically supplied via a single phase 60-A/240-V connection and use, on average, 8,000 kWh annually. The electricity network is winter peaking (June to August, typically between 6 p.m. and 9 p.m.) and has been designed for a typical peak demand [i.e., the after-diversity maximum demand (ADMD)] of 2.5 kW per individual connection point. The majority (about 80%–85%) of residential homes are not connected to the natural gas network, so heating is mainly electric, which explains the winter peak. Against this backdrop, residential EV charging (approximately 1.8 kW for level 1 and 7.2 kW for level 2 charging) represents a significant load. Therefore, successful vehicle electrification in New Zealand to achieve decarbonization goals will hinge on cost-efficient and reliable network integration.
Vector’s internal modeling indicates that unmanaged residential EV charging is likely to present significant capacity issues on the distribution network as EV density increases. In some situations, capacity constraints are predicted to occur on feeders at relatively low levels of EV penetration.
While modeling, we were uneasy with the fact that we were relying on inputs around charging diversity and the benefits of managed charging not sourced from real-world empirical studies but theoretical desktop exercises. This unease made us cautious as to the extent that we could rely on these to shape Vector’s asset and customer strategies, as it added more uncertainty. In addition, the majority of data we could source were from different cities abroad, meaning they were unlikely to represent the New Zealand situation or reflect the New Zealand consumer’s behavior and motivations.
At that stage, we set about developing a real-world empirical trial that could accomplish the following:
Participants in the trial were recruited through a social media advertising campaign, which ran for one week in May 2019. Potential candidates were profiled based on their responses to a pretrial survey, which covered different aspects including demographics, driving distance, and vehicle type. Given that Auckland’s EV owners were early adopters, we needed to take extra care to ensure the trial cohort was as representative of the population as possible with a suitable spread of vehicle types.
For safety reasons, it was decided that applicants whose houses did not have switchboards that met the newer compliance standards were excluded. Another prerequisite was the availability to access half-hourly smart meter data to support the analysis.
With the trialists confirmed (a total of 189), the installation of 7-kW Evnex EV chargers across the Auckland region began. Figure 1 shows an outdoor installation of this EV charger. Japanese EV imports, such as the popular Nissan Leaf (maximum charging capacity of 3.8 kW), made up almost 80% of vehicles in the trial. The charging provider Evnex was selected as the project’s equipment provider due to several key considerations. First, the company already had an established off-the-shelf charger that could receive and respond to management/control instructions, which meant that we did not have to do any product development as part of the project. Second, their chargers utilized open standards and communications protocols, meaning the integration into Vector’s systems (current and future) would be seamless. Finally, Evnex is based in New Zealand, which proved valuable for support and troubleshooting during the installation and trial process. A responsive on-the-ground resource helped ensure a good customer experience during both the setup and trial phases so that any equipment issues did not inadvertently negatively influence customer perceptions of managed charging.
Four main sources of data were collected during the trial:
EV customer behavior drives the requirements of the electricity network. A true customer-centered approach to EV network integration, therefore, needs to start with deep data-driven insights about how EVs are being used (both on and off the network) as well as the value and preference sets of those using them. These insights and preferences will inform the network planning process and ensure that the right balance of cost-efficiency and reliability is struck. In the face of heightened uncertainty due to new policies, technologies, and macroeconomic conditions, customer data can act like a compass for navigating foggy conditions.
When we started the trial, the common belief was that EVs were used for shorter trips than internal combustion engine (ICE) vehicles. The evidence contradicts this assumption. The weekly driving distance of an EV (200+ km) is comparable to that of an ICE vehicle. Higher range EVs cover slightly more distance per week, even if their owners live closer to the city center.
Customer behavior is never homogeneous, and critical infrastructure, like an electricity network, needs to understand the different needs and behaviors across the geographies it serves at an appropriate localized level. EV charging is no exception to this. Energy delivered by a charging session has a median of 6.9 kWh (∼41 km) and a long-tail distribution caused by some of the higher capacity vehicles. Most customers charge their EV several times a week, but only around 30% charge it every day. EV customers are, therefore, broadly speaking, either proactive or reactive, as they either top up regularly in small amounts or charge less frequently but in larger amounts. Drivers of low-range vehicles (<200 km) are much more likely to be proactive chargers who do “top-up” charges of only 1–3 kWh. With vehicle ranges increasing, we expect to see a trend toward more reactive charging. ADMD on the network is a function of the number of customers connected coincidentally and the duration of the connections. If reactive charging grows, both variables tend to neutralize each other, and the network impact remains similar, and, even if less customers are connected coincidentally, they are more likely to overlap in time due to longer charging times.
ADMD describes the peak contribution across a group by considering behavioral differences. It is a well-known and widely used metric for designing electric assets and has historically been established by measuring the aggregate demand across a group. Granular time series data, such as EV charger data and smart meter data, can compute ADMD as a function of group size to determine an ADMD curve. Interestingly, no single standardized methodology for calculating ADMD curves exists within the industry. Here, the ADMD curve is built (as a function of group size) by first taking 20 different samples; then calculating the corresponding ADMD values; and, finally, determining the 95th percentile value across the samples.
The resulting ADMD curve exponentially drops with the increasing number of EV chargers (Figure 2). As the number of EVs on a network asset is increased, it becomes statistically less likely that the customer behavior is aligned and that they charge at the same time. In other words, the behavior is diverse. At the level of a network with 100 EVs or more (e.g., large distribution transformers and assets at medium-/high-voltage levels), the curve has flattened out, and load is diversified. If a group of 100 EV chargers has an ADMD of 1 kW per EV, that rate would require 100 kW of network capacity. When the number of EVs is low—for example, a low-voltage (LV) network with 10 EVs (which is common in a rural area), the ADMD is about 3 kW per EV. To accommodate 10 EV chargers, an electricity network of 30-kW capacity is required. In other words, by lowering the number of chargers to 10 in the last example, the network capacity required is only reduced by a factor of three. Clearly, EV charging will affect the LV network proportionally more. An emerging issue is that the LV network has traditionally been built to “fit and forget,” with changes being driven by asset failure or customer complaints. To meet today’s customer expectations, this situation is no longer appropriate, and the industry will require increased visibility at the LV level to monitor and deliver the future needs of electricity customers.
As previously noted, long-range EVs charge less frequently, but they charge longer. Therefore, the ADMD curve for short- and long-range EVs looks identical. The lower frequency pulls down the ADMD, but this is offset by the longer charging times with increased coincidence.
EV integration will also affect network voltage if clustered. To understand the potential impacts, the ADMD curve can be used as an input for power flow simulations. Let us consider a typical LV feeder with 18 homes with one EV per home. Given that most homes in Auckland have more than two vehicles, clusters of 18 vehicles on an LV feeder will occur in the near future if they are not already a reality. After a power flow analysis for different conductor types and sizes (copper of 25 mm2 as well as all-aluminum conductors of 74 mm2 and 122 m2) was run, the end-of-line voltage did not drop by more than 10% of the nominal voltage (i.e., 0.9 per unit), which is commonly used as the standard in many countries. New Zealand voltage regulations, though, prescribe a voltage limit of ±6% of the nominal voltage. In this instance, the end-of-line voltage would be breached at the end of the two skinnier conductor types. This case study also highlights the practical application of the ADMD curve for network modeling.
EV customer data also illustrate the high variability of demand load profiles during weekdays (Figure 3). Critical electricity infrastructure is not developed for the average case but to deliver to more extreme instances that are vital for the security of supply. The percentile values of the load curve depict the difference between median and different percentiles. The load profiles also show a noticeable step around 9 p.m. Customer engagement confirmed that this is caused by the participants’ use of the EV’s onboard timer, which schedules charging to specified hours of the day. If the vehicle is plugged in outside of these times, the charging is delayed until the start time specified. Survey results confirmed that 38% of customers set the onboard timer and that an additional 22% reported physically plugging in their vehicle to begin charging after 9 p.m., which is commonly considered the end of the peak period at the networkwide level.
While 60% of the trial participants are on an electricity retail tariff that incentivizes off-peak charging, this fact did not define their actual behavior. Interestingly, a participant’s awareness of whether they are on an off-peak plan or not is a much better predictor of off-peak charging. This finding indicates that general engagement with electricity matters can steer participants toward off-peak charging without financial remuneration. These kinds of “hidden insights” can only be gathered through examining data that highlight the importance of using trials, such as these, to identify the range of EV driver behaviors and equate these to real-world network impacts.
The wealth of data produced during our trial identified three typical EV charging segments based on plug-in time and connection duration:
Most notably, the onboard timer is used almost exclusively during the after-work sessions. This explains why the electricity load profile peaks around 9 p.m. (Figure 3) even if most EV customers plug in after work between 5 p.m. and 6 p.m. (Figure 4).
A common concern expressed by the industry is related to the potential impacts of the mass alignment of EV charging and resulting peaks during extreme weather events, natural disasters, or special dates, such as the night before or after a public holiday. During the three years of the trial, some initial real-world insights were created:
Combined with smart meter data, EV charger data provide further behavioral insights. EV charging typically increases total electricity consumption by ∼20%. With full electrification, this number will rise, as most of our trialists owned only one EV, whereas the average household in Auckland has more than two vehicles. The difference between the load duration curve with and without EV charging is less pronounced on each end (Figure 5). The convergence of demand during peak time (left side of Figure 5) illustrates that EV charging does not peak coincidentally with other household demand peaks due to the use of the onboard timer.
The EV charging data have helped establish an understanding of the different customer behavior types and determine the network integration requirements. This understanding provides a vital input to the network planning process by allowing us to identify where and when additional capacity is needed.
An analysis of charging behaviors shows that EVs are plugged in for a significantly longer duration than they charge. In fact, around two thirds of the charging sessions are longer than 8 h, but active charging only lasts for 4 h. Figure 6 shows the distribution of all EV charging sessions (y-axis) against charging time (x-axis). A charging session is defined as the period when the EV is physically connected to the charger unit. This figure shows that load shifting flexibility overnight is available, as there are long periods when the vehicle is just passively parked with a full battery. The potential for smart charging to optimize EV network integration while still fully charging the EV is there.
The architecture used for the smart charging trial was composed of three main parts: the distribution energy resource management system (DERMS), the EV charging manager (aggregator), and the EV charger itself. DERMS decides on the optimal EV charging schedule based on network asset information, smart meter consumption data, and EV charging information. Via a web-based application programming interface, the optimal schedule for each EV charger is sent to the EV charging manager, who sends the schedule via cellular networks to the charger using the Open Charge Point Protocol (OCPP). The charger sends loading and connection information back to the EV charging manager, who feeds the information through to DERMS via an application programming interface. When considering how a full solution may be implemented, this architecture is scalable if 1) chargers have communication capability based on an international standard (e.g., OCPP) and 2) a “last resort” EV charging manager is available to provide aggregation for communicable EV chargers that do not provide aggregation services.
A custom-built research DERMS was used so that the control logic and scheduling algorithm could be customized to test tradeoffs. Two scheduling algorithms were designed with the same objective of integrating EVs into a residential network without expanding capacity, which relies on whole-house demand (i.e., EV charging and smart meter data). However, the control logics of both algorithms differ as follows:
The operation of the algorithms is exemplified in Figure 7 for a winter day, which is the peak season in Auckland due to heating loads. (As previously mentioned, the vast majority of homes in Auckland heat with electricity.) In Figure 7, the average household’s demand profile is shown in dark blue. The light blue horizontal line indicates the typical distribution feeder design limit (on a per-home basis), and the shaded red areas depict the periods where demand is above the design limit (i.e., an overload).
Throughout our smart charging period (2 blocks of 2.5 months), we were able to demonstrate that smart charging was able to integrate EV charging without breaching network security limits while keeping customer satisfaction (collected via regular surveys) at similar levels to when their EV charging session was not being managed (Figure 8). The satisfaction rate was higher than 90% for Static (at the same level as fir uncontrolled), while Dynamic was just under 90%, an insignificant difference that we were unable to relate to any difference in customer experience.
A more detailed look at how the trial results differentiated based on the algorithms themselves provides insight for logic development. The Dynamic algorithm performs well during our typical peak times from 5 p.m. to 9 p.m., as it continuously adjusts charging. However, it is less well equipped to deal with the sudden spike created by customers using the EV onboard timer, as it needs 20 min to react to and then make the best decision for the next 20-min window. The Static algorithm, with its day-ahead forecast, decides in one view what is best across the full window from 5 p.m. to midnight. It is able to capture the customers who use the onboard timer as the default choice much better than the Dynamic algorithm. However, it struggles with unpredicted demands and, as a result, still allows some breaches of the network security level during the period from 5 p.m. to 9 p.m. As previously described, we deliberately developed a custom-built DERMS for this project to depict those tradeoffs.
Despite the success of both control algorithms, it is still important to consider if simple timer-based charging management or pricing could also deliver the same outcomes. While customer satisfaction was essentially identical to smart charging when timer-based charging management was used for a fixed window every day, the network benefit deteriorated. Blunt timer-based control led to a synchronization of the load after the management period, which created a secondary peak that will increase network capacity requirements (Figure 9).
Likewise, time-of-use (ToU) tariffs are unlikely to be a panacea. ToU tariffs offer considerably lower network prices after a certain time (commonly 9 p.m. in New Zealand) to incentivize EV customers to charge out of peak. However, the physics and loading of the network are not considered dynamically, which, like the simple timer-based charging, will create a synchronization of the load and secondary peak as confirmed by the trial. The synchronization or destruction of natural diversity means that the secondary peak will quickly exceed the primary peak. This additional secondary peak is of most concern with respect to the LV network, where there is very little diversity of customer type, and, even without EVs, the load around 9 p.m. or 10 p.m. is still relatively high. As EV penetration increases, the secondary peaks will become a major concern, as flipping the fleet to electric transport will typically mean that more EVs than homes are connected. (In our case, we expect roughly twice as many vehicles as homes.) An active distribution management system (“smart charging”) is, therefore, a superior solution, as it accomplishes the following:
As previously mentioned, our trial also indicated that the ToU price plan itself does not actually incentivize delayed charging. It is the knowledge of peak and off-peak times and the level of engagement with energy a person has that drives behavior. In fact, all EV customers who clearly knew what type of price plan they were on charged off peak, even if they were not on a ToU price plan. Those who did not know what type of price plan they had tended to charge during the network peak. This finding suggests that the engagement level, not the pricing plan, drives off-peak charging—a point that may often be missed or not considered in analyses around the effectiveness of ToU pricing to change behavior. For an electricity network company, this highlights the importance of strong customer engagement and related messaging to achieve the smooth integration of EVs into the network.
Additionally, smart charging via the network’s DERMS guarantees a coordinated and holistic approach across other DERs. The most important resource for Vector is currently electric hot water, which is actively managed today. Clearly, hot water and EV load need to be coordinated dynamically to achieve network outcomes while not negatively affecting customer outcomes. Static ToU tariffs optimized by local house-level algorithms would likely schedule all smart loads to come online at the same time, creating new peaks and reinforcement needs.
For customers, smart charging will result in lower network reinforcement and constraint management costs over time, ultimately keeping network prices affordable. As previously discussed, only customers participating in a dynamic DERMS can genuinely claim to deliver this full benefit across different network levels. The high acceptability of smart charging was best exemplified by the fact that the share of customers who believed they would participate in smart charging without additional compensation increased from 72% before the trial to 85% at the end of the trial.
The EV charging trial uncovered customer behavior and demonstrated smart charging in action, thereby allowing the network impacts and benefits to be discovered. Our key learnings presented in this article can be summarized as follows:
In the face of growing uncertainty, the customer data and learnings provide a compass for navigating the energy transition. The trial findings have cemented or accelerated many actions across our work program, in particular, the following:
The electrification of transport is a massive societal and technology shift that will improve energy security and air quality as well as mitigate climate change. Smart electricity network integration is key to ensure long-term cost-efficiency but relies on customer acceptability as well as increased network and customer visibility.
We would like to acknowledge the contributions of the project team, particularly Louise Murphy and Leon Hayward (project managers), Chris Franks (data management), and Jacques de la Bat (load flow analysis). Also, we are grateful for the financial contribution received from EECA’s low-emission vehicle fund for the chargers installed in Waiheke.
“AGL electric vehicle orchestration trial,” Australian Renewable Energy Agency, Canberra, ACT, Australia, 2023. Accessed: Jun. 1, 2023. [Online] . Available: https://arena.gov.au/projects/agl-electric-vehicle-orchestration-trial/
“Jemena dynamic electric vehicle charging trial,” Jemena, Melbourne, VIC, Australia. Accessed: Jun. 1, 2023. [Online] . Available: https://arena.gov.au/projects/jemena-dynamic-electric-vehicle-charging-trial/
“Understanding the home charging behaviour of EV customers,” Wellington Electricity, Jul. 2018. Accessed: Jun. 1, 2023. [Online] . Available: https://www.welectricity.co.nz/disclosures/pricing/evtrial/document/159
“EV SmartCharge Queensland program,” Energex, Newstead, QLD, Australia. Accessed: Jun. 1, 2023. [Online] . Available: https://www.energex.com.au/home/control-your-energy/smarter-energy/electric-vehicles/ev-insights
S. Heinen and P. Richards, “Towards customer-centric energy utilities - A granular data-driven bottom-up approach to understanding energy customer trends,” Electricity J., vol. 33, no. 9, Nov. 2020, Art. no. 106836, doi: 10.1016/j.tej.2020.106836.
B. Anderson, R. Parker, D. Myall, H. Moller, and M. Jack, “Will flipping the fleet F**k the grid?” in Proc. 7th IAEE Asia-Oceania Conf., Energy Transition, Auckland, New Zealand, Feb. 2020. [Online] . Available: https://www.researchgate.net/publication/339291656_Will_flipping_the_fleet_fk_the_grid
Steve Heinen is with Vector Ltd., Auckland, 1023, New Zealand.
Andre Botha is with Vector Ltd., Auckland, 1023, New Zealand.
Duncan Head is with Vector Ltd., Auckland, 1023, New Zealand.
Rafferty Parker is with Vector Ltd., Auckland, 1023, New Zealand.
Pieter Richards is with Vector Ltd., Auckland, 1023, New Zealand.
Digital Object Identifier 10.1109/MPE.2023.3308250
Date of current version: 19 October 2023
1540-7977/23©2023IEEE