Optimal Sizing of a Battery-Supported Electric Vehicle Charging Hub with a Limited-Capacity Grid Connection
[ad_1]
1. Introduction
Recently, a large number of sites have been installed with a battery energy storage system (BESS) at DC charging stations. Projects and studies with a BESS at large AC charging hubs have been missing. These projects are, however, more complex in terms of determining the optimal sizing of the system, as well as operating the system in the most efficient manner. These systems often require a lower, but more continuous, power than high-power DC systems. A temporary reduced power does not always have to be problematic. Inverter dimensioning and smart operation play a large role in the efficiency of the system. These unique features make dimensioning and the optimisation of these systems a different problem.
1.1. Literature Review
1.2. Case Study
Smart-charging strategies are often thought of as a first potential solution to grid-limited CPs, such as load shifting by suspending charge sessions or scaling current delivery with total CP power demand. In a P&R charging hub, smart charging can offer an improved charging efficacy both for the users and for the network operator. However, to ensure the user experience is not negatively impacted in the coming years due to increased EV penetration, this pilot project sought to investigate the use of a BESS.
1.3. Contribution
To the authors’ knowledge, there have been no studies or projects besides the case study that supplied a charging hub using only a grid connection and BESS with the intention that the BESS limits grid loading during peak grid-load hours. This work uses measured data from the case study to address the oversized BESS and optimise the BESS and grid connection capacity using a variety of load profiles and 5 min time resolution. The model developed is a generalisable model of a BESS-supported type-2, level-2 charging hub, and is easily scalable for any number of CPs, grid connection capacity, BESS capacity, and load profile. The optimal system size was compared against the case study system for a number of BESS control strategies to form recommendations on sizing and control.
1.4. Structure
2. Optimisation Problem
The sizing of a BESS in a grid-limited AC charging hub should be large enough to aid in supplying demand but not oversized such that there is an excess of capacity. A BESS remains an expensive investment so the intention is to keep the storage capacity minimal.
The BESS control will have an effect on the optimisation process. In this optimisation problem, the standard ‘base-case’ control was used, which was derived from the case study. It is assumed that, in a newly installed and optimal system, the BESS would have individual phase power delivery, and, thus, phase imbalance and grid feedback would not be an issue. This means that the grid delivers up to the full connection capacity, and, if the BESS delivers power, it delivers only the remaining load above the maximum grid capacity.
where is the cost of the BESS installation, currently approximately 650 €/kWh [20,21]. is the cost of installation per grid connection capacity as presented in Equation (2) [18]. is the profit from the electricity sale, assumed to be 0.1 €/kWh. The battery investment is annualised by dividing by the expected system lifetime , 10 years as per the battery supplier capacity warranty. Similarly, the monthly loss of load, LL, is annualised by multiplying by 12. is the mean battery round-trip efficiency. Operational and maintenance costs of the BESS are not included in this objective function since it is assumed that they would remain similar for a BESS regardless of its capacity.
where is the total EV power demand at time , is the base load at time , is the power supplied by or delivered to the battery at time , and is the power supplied by the grid at time . The battery current convention employed is a negative battery power for discharging. The sum of power over the entire time-series is then multiplied by 1/12 to convert from the 5 min time step to hours. T is the total time period of 1 month. All other constraints, Equations (5)–(8), were internal to the system model and were handled during simulation runtime. These included the power balancing, the battery state-of-charge (SOC) limits, and the battery charge/discharge power limits.
Due to the stochastic nature of the model, a single month-long load profile was formed and repeatedly used for the simulations in the optimisation process:
-
A parent population of potential solutions was generated containing the decision variables , , and ;
-
A simulation was performed for a single potential solution, and the outputs LL and were retrieved;
-
Using these five decision variables, the objective function was evaluated and the results saved;
-
This process was repeated for all possible solutions in the population of the current generation;
-
A new parent population was created for the next generation, as the GA describes, allowing for crossover and mutation.
Additionally, the capacities were optimised for both a 5 MWh monthly load and a 6 MWh monthly load. All feasible solutions were plotted as the loss of potential load against the BESS capacity. These figures clearly illustrate the relationship between the BESS capacity and the loss of load.
3. Case Study
3.1. Model Development
The months of January–June 2022 were used to develop the model, with the month of July 2022 used to validate it. Individual charging sessions were identified and various session parameters determined, namely, the day of week, entry time, exit time, end of charging time, power delivery per time step, number of phases it is connected to, and the current per phase. The maximum charging power, total energy delivered, and connection and charging duration per charging event were deduced. The charging sessions were then filtered for charging duration and energy transfer, with limits of [0.5 h, 25 h) and [1 kWh, 80 kWh), respectively.
where is available power from the grid at time t and is the total base power at time t. The numerator consists of the current draw per phase for EV x at time t, and the denominator consists of the total current demand per phase at time t.
3.2. Model Validation
4. Control Scenarios
The installed battery system round-trip efficiency was low due to the conversion losses across the inverter at low charging and discharging powers. Additionally, the battery made frequent and small discharge/charge cycles. Finally, there was no consideration for the power imported from the grid during peak grid-load hours. Given the current state of the power grid, it is logical to limit the power drawn from the grid during the peak grid-load hours of approximately 17:00 to 20:00. The following control scenarios were therefore decided upon:
-
The base case in which the grid supplies all load up to the maximum capacity. The battery supplies the remaining load above the maximum grid capacity.
-
During the peak grid-load hours of 17:00–20:00, all load is supplied by the battery. If there is no EV load, the grid will supply the base load. If the battery is drained, the grid will supply the load.
-
Charging/discharging the battery deadband of 10 kW and 15 kW. If the EV load is above this deadband, the battery supplies the entire load.
-
The combination of limited peak hour power draw and battery charge/discharge power deadband of both 10 kW and 15 kW.
-
Potential load not delivered (lost load);
-
BESS round-trip efficiency;
-
Energy drawn from the grid during peak hours;
-
Percentage of total load supplied by the BESS;
-
Percentage of users still charging at the end of their session.
5. Results
5.1. Optimal System Sizing
The optimal system sizing, regardless of the BESS cost, delivers a much better quality of service to system users when compared to the case study system, for a reduced annual investment. For the highest BESS cost of €750/kWh, the loss of load was reduced from 6.5% of the total load to 1.5%, and the annual investment remained comparable. The predicted 2030 BESS cost of €250/kWh resulted in a loss of potential load of less than 1% of the total load, and the annual investment fell by 36%.
The relationship between LL and annual investment is linear for all load profiles and grid connection capacities. The gradient differs for load profiles but appears to be consistent across grid connection capacities. These figures show the BESS capacity that is required for each grid connection capacity to ensure no loss of load.
For a 5 MWh monthly load and a 3 × 50 A grid connection, a BESS capacity of approximately 60 kWh is required to ensure no loss of load. For a 6 MWh monthly load and a 3 × 50 A grid connection, a BESS capacity of approximately 70 kWh is required to ensure no loss of load. For a 7 MWh monthly load and a 3 × 80 A grid connection, a BESS capacity of approximately 180 kWh is required to ensure no loss of load.
5.2. Comparative Analysis of Optimal Sizing and Case Study Sizing
With the optimal sizing, an imposed battery deadband resulted in an increased grid import during peak hours, with respect to both the optimal base case and case study sizing. Above the deadband, the battery delivered the full load; therefore, at the end of the day, the battery was more depleted with respect to the base case. This is consistent with the P&R usage pattern which tends towards a high EV load in the morning and early afternoon due to commuters. With the optimal grid connection capacity, the high battery-charging power could fully recharge the battery in the three-hour window. Furthermore, the low battery utilisation in the optimal base case means the battery is not often recharged during these peak hours, hence the decrease with respect to the case study sizing.
6. Discussion
The model presented in this study is for the charging of EVs in a charging hub with a stationary BESS and grid connection. The model is easily scalable for any number of CPs, BESS capacity, grid connection capacity, and load profile. The charge session data used in these simulations were measured at a P&R charging hub. Given the appropriate data, for example, from a workplace charging hub or shopping centre charging hub, the model is easily transferable.
Therefore, one must consider what the goals of such a solution like a BESS are. A limited loss of load, limited grid interaction, and high BESS round-trip efficiency are all considered in this study.
The choice of load profile used in solving the optimisation problem had a large effect on the outcome. The simulated load profile used in the optimisation was chosen over other ~7 MWh profiles because it included a high demand day—a peak power demand of 81 kW which lasted over 3 h. This high demand day served to stress-test the sizing and ensures the optimal sizing is capable of serving future loads.
The control method used in solving the optimisation problem also had a large effect. For instance, if the system was optimised using the 15 kW PHBDB control strategy, the BESS would inevitably require a larger energy storage capacity to satisfy the constraint represented by Equation (4), the volume of potential load lost. Furthermore, these are only a selection of specific, yet limited, control strategies that were intended to address specific performance metrics. The optimal power dispatch and charge session scheduling which would result in an improved system performance were outside the scope of this study.
The control strategy that limited grid interaction during peak grid-load hours yielded the most desirable results with the optimal sizing. The BESS round-trip efficiency was increased with respect to both the optimal sizing base case and the case study sizing, to 79%. Energy losses were kept low since the load was mostly supplied via the grid connection; the battery supplied only 12% of the load. The grid interaction during peak evening load hours was reduced to 110 kWh, compared to 713 kWh for the optimal sizing base case. Finally, there was no loss of potential load.
If the battery were to be used for grid ancillary services, such as frequency response and voltage control, then an additional revenue would be available for the battery, and the optimisation problem would be reformed. The optimal sizing would likely tend towards a larger battery to benefit from the ancillary service revenue whilst still maintaining the security of supply for the P&R users.
Dynamic charging tariffs are thought to be a good method for incentivising users to charge their EVs at low grid-load times and reduce the disruption to the power grid. This would have little effect in a P&R since the intended user groups associated with a P&R charging hub, namely, commuters and visitors, are not as flexible in their arrival and connection time as resident CP users.
Vehicle-to-Grid (V2G) is another rapidly progressing technology. During times of high electricity price, the EV can act as a battery and deliver power to a household when connected and laying idle on the driveway. In an urban neighbourhood that relies on public CPs, a fleet of EVs could be used to reduce evening peak residential loads behind the substation. V2G may be feasible in a P&R charging hub but only for specific users who meet certain criteria, such as commuters who park for the full working day. However, transferring energy from one commuter to another commuter may result in unsatisfied users. How V2G would be implemented in a P&R charging hub is yet unknown.
Whilst DC fast chargers are becoming more prevalent, their installation at a P&R is not necessary. They are suited for rapid turnover charge sessions, such as along motorways or in taxi ranks, or for high-battery-capacity vehicles, such as at bus depots or for heavy goods vehicles. Typical user connection durations are multiple hours at P&Rs. The measured data indicated the average connection duration to be 13 h. Therefore, level-2 charging will remain applicable for coming years.
Larger EV batteries are, of course, to be expected in the coming years; battery capacities greater than 100 kWh are already on the market. This study assumed a maximum charging demand of 75 kWh since the largest measured charge session was 68.9 kWh. It is hard to predict how larger EV battery capacities will affect charging behaviour since it is so highly dependent on social demographics, the availability of charging infrastructure, social and cultural norms, and personal preference. Considering the price of BEVs with large capacities and the rate at which EVs are penetrating the car fleet, it will be many years before such large-capacity BEVs are the norm.
Finally, it is clear that a multi-objective optimisation is required, in which grid interaction during peak hours is minimised, as well as the annual system cost. The intention of this system is to reduce grid loading for large charging hubs, especially during peak grid hours. Thus, the BESS should be adequately sized and appropriately controlled to service all EV users whilst maintaining a high round-trip efficiency and keeping grid interaction to a minimum. This could best be integrated using electricity pricing, such that the BESS is prioritised during times of high electricity price and the grid is prioritised during times of low electricity price.
Perhaps the most practical recommendation is that the battery be installed with individual phase control or to ensure an energy contract with the distribution network operator to allow power flow back to the grid. These design considerations will allow for power to be delivered individually and unevenly on separate phases.
7. Conclusions
This study used measured data from an installed EV charging hub with an on-site stationary battery (336 kWh/250 kW) and limited capacity grid connection (17.4 kW) to develop and validate a computer model in Python. A genetic algorithm was used to minimise the annual costs of the system by optimising the battery energy storage capacity and the grid connection capacity for a monthly load of 7 MWh. Three different battery costs were evaluated; the approximate current cost of €750/kWh, the expected 2030 cost of €250/kWh, and the middle €500/kWh. The optimal sizing, with the €500/kWh cost, a 55.4 kW grid connection, and a 69 kWh/45 kW battery, was then assessed using a variety of simple control strategies; namely, limiting grid power draw during peak evening grid-load hours, and implementing a battery charge/discharge deadband, and comparing this against the case study sizing. The limited peak hour grid interaction control strategy was determined to perform best with the optimal sizing.
The feasible solutions to the optimisation problem for three load profiles, 5 MWh, 6 MWh, and 7 MWh, were plotted as the battery capacity against the loss of potential load. These figures illustrated what battery capacity was required at each grid connection capacity to ensure no loss of potential load.
Finally, the limitations of this study were addressed and ideas for future work were presented.
[ad_2]