Pareto-Based Power Management for Reconfigurable Multi-Point Multi-Power EV Charging Stations

Abstract

The increasing adoption of electric vehicles (EVs) is driving the need for more efficient, scalable, and flexible charging infrastructures. Among the most promising solutions are reconfigurable multi-point multi-power (MPMP) charging stations, which enable dynamic power allocation across multiple charging points operating at discrete power levels. This paper introduces a novel power management strategy for MPMP stations based on Pareto optimization, aiming to minimize the average charging time while ensuring fairness and efficiency. The method dynamically allocates power among charging points to minimize the average charging time across all connected EVs, while adhering to system constraints and the varying charging profiles required to preserve battery health. The proposed approach was validated through simulations in a dynamic scenario involving six EVs with heterogeneous battery capacities and charging profiles. Results demonstrated that the Pareto-based strategy achieved a significantly lower expected average charging time when compared to the first-come first-served strategy (FCFS).

1. Introduction

The transition to electric vehicles (EVs) represents one of the most profound transformations in contemporary mobility, redefining not only transportation systems but also reshaping paradigms in energy usage, sustainability, and future urban planning. With ongoing advances in battery technology, infrastructure expansion, and supportive regulatory frameworks, EVs are paving the way toward a cleaner, more efficient, and environmentally conscious mobility ecosystem.

Nevertheless, this transition is still underway, and several critical challenges remain before EVs can achieve widespread adoption. Among the most pressing is the development of a robust and accessible charging infrastructure capable of meeting growing user demand while coping with the technical constraints inherent to current EV charging systems. In response to the increasing deployment of EVs, substantial innovation has emerged in the design of charging stations (CSs), aiming to reduce charging times, alleviate stress on the electrical grid, and preserve battery health. Additionally, effective power management in stations hosting multiple charging points (CPs) is a key concern—particularly in high-traffic locations, such as highways [1]. Indeed, the limited driving range of EVs combined with relatively long charging times continues to hinder broader adoption.

At worst, traditional CSs consist of a single CP supplying a fixed power level to one vehicle at a time. Even when multiple CPs are available, they often deliver the same power level irrespective of the specific EV model or its charging needs. This static allocation strategy lacks the flexibility required to follow optimal charging profiles [2]. In practice, the available station power is usually divided equally among all active CPs, disregarding the varying demands of connected vehicles, which may require more or less than the nominal CP power. This approach leads to inefficiencies in both energy usage and infrastructure utilization, resulting in longer average charging times and higher operating costs for station owners.

To overcome these limitations, reconfigurable multi-point multi-power (MPMP) charging stations have been proposed as a viable solution. These architectures enable dynamic allocation of power among multiple CPs [3,4], allowing for optimal resource utilization tailored to each EV’s state of charge (SOC) and specific power requirements [5,6]. In fact, ultrafast charging does not follow constant or even constant current–constant voltage (CC-CV) profiles. To preserve battery longevity, a stepped power profile is commonly adopted [7], striking a balance between fast charging and battery health by maximizing the charging current without compromising the battery lifespan [8,9,10]. To implement such optimal profiles, it is essential to adopt flexible solutions that allow dynamic and demand-based power distribution across CPs, without the need to oversize the station’s electrical infrastructure. Within this framework, MPMP stations can reach optimal performance by continuously reallocating power based on real-time system status. In particular, surplus power from underutilized CPs can be redirected to those serving EVs with higher power acceptance, thus reducing average charging times and improving quality of service (QoS) for users [11]. A cornerstone of these reconfigurable systems is their modular architecture, composed of identical power converter modules interconnected to flexibly partition the total station capacity across multiple CPs [11]. This design promotes scalability for future demand scenarios and supports more resilient and adaptive infrastructures [12,13]. The proposed CS model adopts an MPMP configuration that enables concurrent charging of multiple EVs at different power levels through efficient internal power management. The station is built using isolated bidirectional DC-DC converters (IBDCs) based on a dual-active-bridge (DAB) topology [14]. These modules are arranged in an input-parallel/output-parallel (IPOP) configuration, with the input connected to a common DC bus and the outputs routed through a switching matrix. This matrix allows for parallel connection of modules to CPs, supporting reconfigurable power delivery tailored to individual EV demands. The bidirectional nature of DABs further enables integration with vehicle-to-grid (V2G) systems and local energy storage [15,16]. While DAB control strategies are well established, the novelty of this work lies in the power management algorithm specifically designed for MPMP stations. The goal is to minimize the average charging time in dynamic conditions, where not all CPs are occupied and station power availability varies. The proposed approach surpasses conventional power allocation strategies, such as uniform distribution, first-come first-served (FCFS), and SOC- or priority-based schemes [17], which may cause unacceptable discrepancies in the user experience [18].

Instead, this paper adopts a multi-objective decision-making framework grounded in Pareto optimization. This method evaluates trade-offs between conflicting goals, including average charging time, fairness, and energy efficiency, and identifies Pareto-optimal configurations, where no criterion can be improved without worsening another. This approach enables informed, goal-driven selection of the most appropriate power allocation strategy for each scenario. Efficient power management is indeed pivotal in MPMP stations. Advanced control paradigms, such as model predictive control (MPC) [19], AI-driven algorithms [20], and fuzzy logic [21], have been proposed to allocate power dynamically based on vehicle priority, target charging speed, and resource constraints [22]. These reconfigurable stations rely on both hardware and software adaptability to respond to real-time demand fluctuations. This study explores the potential of Pareto optimization to minimize the average charging time in MPMP charging systems. In the considered configuration, each CP operates at discrete power levels of 80 kW, 160 kW, or 240 kW via a dynamic switching matrix that enables real-time reconfiguration based on instantaneous demand and operational constraints [23].

Additionally, the link between the electrochemical behavior of next-generation battery materials and their suitability for integration with intelligent power distribution architectures should also be considered. In fact, emerging solid-state and hybrid battery chemistries offer enhanced thermal stability and response characteristics, which are directly relevant to dynamic load management [24]. Moreover, advances in electrode design and ionic conductivity can impact on the battery response times, thus enabling faster and more efficient power reallocation in real time [25,26].

The remainder of the paper is structured as follows: Section 1 outlines the proposed system architecture, Section 2 introduces the multi-objective function used for MPMP system management, Section 3 presents simulation results, and Section 4 concludes with key findings and future research directions.

2. Charging Station Architecture

Multi-port converters enable simultaneous charging at different power levels, offering enhanced flexibility in the management of EV charging infrastructure. The primary aim of this paper is to minimize the average charging time of EVs present at a charging station at any given moment by suitably reallocating power across individual charging points through a reconfigurable topological structure.

With the growing adoption of electric vehicles, the design and operational management of charging stations have become critical. Without any optimization approach, these stations could become critical bottlenecks in the whole charging infrastructure, particularly on high-traffic routes, thereby hindering the broader adoption of EVs, especially for long-distance travel where fast charging is essential.

The core issue lies in the significant heterogeneity among EVs regarding their fast-charging capabilities. Fast charging, typically utilized at motorway service stations, requires delivering different power levels depending on the vehicle model.

Table 1. Some EV models and corresponding charging characteristics.

Vehicle Model	Battery Capacity (kWh)	Max. Charging Power (kW)	Charging Voltage (V)	Estimated Time (min)	Real-World Avg. Time (min)
Audi e-tron GT	93.4	270	800	~12.4	~20–25
BMW iX	111.5	195	400	~20.6	~30–35
Fiat 500e	42	50	400	~30	~40–45
Ford Mustang Mach-E	98	150	400	~23.5	~38–45
Hyundai Ioniq 5	77.4	233	800	~11.9	~18–23
Kia EV6	77.4	233	800	~11.9	~18–23
Mazda MX-30	35.5	50	400	~25.5	~35–40
Mercedes EQS	107.8	200	400	~19.4	~28–35
Mini Cooper SE	32.6	50	400	~23.4	~30–35
Nissan Ariya	87	130	400	~24	~40–45
Peugeot e-208	50	50	400	~36	~45–50
Porsche Taycan	93.4	270	800	~12.4	~18–22
Renault Zoe (50 kWh)	52	46	400	~40.5	~50–60
Tesla Model 3	75	250	400	~11	~20–25
Tesla Model S	100	250	400	~14	~25–30
Volkswagen e-Up!	36.8	40	400	~33	~40–45

highlights the variability in maximum charging power across a selection of popular commercial EVs, underscoring the need for adaptive charging infrastructure, as also shown in Figure 1. This diversity introduces potential inefficiencies: if a high-power-capable vehicle is connected to a low-power outlet, charging is suboptimal. Conversely, connecting a low-power-capable EV to a high-power charger does not improve the charging time and may even accelerate battery degradation. Moreover, EVs exhibit non-linear charging profiles, especially under fast-charging conditions. While charging is often limited to a SOC range between 20% and 80% to avoid excessive battery wear, the power acceptance of the battery is not constant throughout this interval. Typically, power remains high in the mid-SOC range but tapers off significantly as SOC approaches 80%, due to battery management strategies designed to extend battery life and limit thermal stress. This variability demands real-time, dynamic power distribution strategies within charging stations to ensure efficient use of resources. Rather than viewing this variability as a limitation, this paper proposes a system that controls the dynamic nature of EV charging profiles. By continuously adapting the power supplied to each charging point based on the instantaneous charging demand and battery acceptance profile, it is possible to reduce average charging times while also minimizing unnecessary battery stress. Indeed, traditional charging stations often feature a fixed number of charging points delivering uniform distributed power levels. However, this static approach can lead to considerable inefficiencies.

Thus, a uniform power distribution among charging points is suboptimal, given the variability in maximum power acceptance across different EVs. To address this issue, this paper explores a modular yet discrete approach to configuring charging points within a station. Specifically, we propose a reconfigurable topology that allows for dynamic adjustment based on real-time demand.

The architecture of the considered charging station is shown in Figure 2. It is assumed that the input DC bus at 900 V is supplied by the medium-voltage (MV) grid through the cascaded connection of a step-down transformer and an active rectifier. The voltage level was chosen to ensure compatibility with the voltage and power requirements of all commercial EVs considered in this study. From this shared DC bus, a set of charging points can be connected.

The proposed modular converter architecture consists of m converter modules (CMs), each of which is composed of isolated bidirectional DC-DC converters based on DAB circuit configuration. Each CM can provide a maximum total power of 80 kW, and it can be individually connected to each of the p CPs by means of a suitable matrix of solid-state power contactors (SSPCs).

The choice of SSPCs for the switching matrix implementation relies on their enhanced characteristics with respect to mechanical circuit breakers: fast switching capability to achieve rapid and precise circuit interruption, also eliminating electrical arcs and reducing contact wear; bidirectionality, if required; integrated protection features (e.g., overcurrent, short-circuit, and thermal), and advanced diagnostics and communication interfaces. Moreover, power semiconductor devices based on wide bandgap (WBG) technology (e.g., SiC or GaN), or other innovative solutions (e.g., B-TRAN technology), can be selected for their higher efficiency (i.e., reduced losses thanks to lower on-state resistance and/or lower on-state voltage drop) and thermal performance. The thermal management design depends on the power/current to be handled—forced air cooling can be primarily considered for currents up to 400 A, while higher currents require liquid cooling and, therefore, lead to an increase in the system weight/volume.

The SSPC switching matrix allows to direct the proper power to each CP, depending on the power request of the EV that is currently using that CP. The switch matrix allows for obtaining an independent DC input for each CP, thus ensuring that the input voltage is compatible with the one of the connected EV battery pack, which depends on the inherent technology and SOC level.

The total number of dual-line switches needed is equal to p × m. These must be properly sized to carry the maximum possible current, which occurs when the connected CM delivers its maximum power (i.e., 80 kW) at the minimum output voltage (i.e., the minimum voltage level of the EV battery pack). The maximum power of each CP is equal to 240 kW, which can be achieved by connecting to the CP three CMs.

This dynamic configuration allows real-time power target distribution when vehicles reach charging points or even during the charging process itself. For instance, in scenarios where the number of EVs currently charging is lower than the number of available CPs, the total available power can be flexibly distributed to closely match the optimal charging profiles of each individual vehicle. This enables the system to maximize the efficiency of the charging process while aligning with the power acceptance characteristics of the batteries.

A similar approach applies when all charging points are engaged by EVs exhibiting diverse charging profiles, typically characterized by different maximum charging power limits and discrete power acceptance curves. In such cases, dynamic power reallocation can still be employed to optimize charging efficiency across the fleet. This involves adjusting the power supplied to each charging point in real time, based on vehicle-specific parameters such as the SOC, battery temperature, and the charging phase. However, it is evident that a perfect match between the supplied power and the instantaneous optimal charging power for each vehicle will not always be achievable. The ideal condition, where each EV receives exactly the maximum power it can accept at every time instant, would result in the theoretical minimum charging time for each vehicle. In practice, due to constraints such as finite station power capacity, different vehicle arrival times, and competing power demands, such an ideal distribution is rarely feasible. Therefore, it becomes necessary to adopt an optimization strategy that balances competing demands.

In this study, the proposed management strategy is designed to minimize the average charging time of all electric vehicles (EVs) present at the charging station at any given time. This objective is achieved by periodically evaluating the system states and vehicle parameters at defined sampling intervals, and by applying optimization criteria, such as minimizing the weighted sum of the remaining charging durations, within the power allocation algorithm.

Mathematically, the optimization problem can be formulated as:

$$\begin{array}{c}min\\ P\left(t\right)\end{array}\left({\displaystyle \frac{1}{N\left(t\right)}}{\sum }_{i=1}^{N\left(t\right)}{T}_{c,i}\left(t\right)\right),$$

(1)

where P(t) is the power allocation vector at time t, N(t) is the number of vehicles currently being charged, and T_c,i(t) is the estimated remaining charging time for the i-th vehicle. This formulation captures the time-varying and heterogeneous nature of the charging demand and provides a basis for implementing adaptive scheduling algorithms that react to system dynamics in real time. To effectively address this inherently multi-objective minimization problem, it is essential to adopt appropriate optimization criteria capable of balancing competing goals—such as minimizing individual vehicle charging times, maximizing infrastructure utilization, and ensuring fair power distribution across all charging points.

In this paper, the proposed strategy for managing charging point allocation is grounded in a multi-objective optimization framework based on Pareto efficiency. The aim is to identify power allocation solutions that offer an optimal trade-off among multiple performance metrics, without arbitrarily prioritizing one over the others. This approach avoids bias in decision-making and provides a set of non-dominated solutions, where no objective can be improved without degrading at least one of the others.

3. Pareto-Based Power Management Method

In the context of multi-port multi-power charging stations, where each charging point can operate at three discrete maximum power levels, 80 kW, 160 kW and 240 kW, the use of Pareto optimization presents a highly effective strategy for minimizing the average charging time across all connected EVs. Each charging point within the MPMP system is capable of reconfiguration via a dynamic switching matrix, which enables flexible allocation and redistribution of power resources based on current demand and system constraints. Therefore, among the conventional power management algorithms, the Pareto optimization framework appears to be the most appropriate solution for the proposed reconfigurable modular CS architecture. Indeed, the Pareto approach allows determining the optimal configuration of power levels among the various charging points in real time. Rather than aiming for a single “best” solution, Pareto optimization identifies a set of non-dominated solutions where any improvement in one objective (e.g., reducing the charging time for a specific vehicle) cannot occur without compromising another (e.g., increasing the charging time for another vehicle). This multi-objective approach is particularly well suited for MPMP systems, where balancing efficiency, fairness, and power availability is critical.

By applying this method, the system continuously evaluates all feasible switch configurations and power allocations, seeking the solution set that yields the lowest possible average charging time across the station. Once the Pareto front is established, a decision-making criterion, such as minimizing the overall queue time or prioritizing high-priority vehicles, can be applied to select the most appropriate configuration.

The Pareto-based minimization function is a scalar function that ranks candidate solutions based on their dominance relation with respect to others in the population. For a given solution x, the function evaluates the degree to which x is dominated by other solutions or dominates them. This evaluation can be expressed as:

$$P(x)={\sum }_{y\in \mathsf{{\rm P}}}D\left(x,y\right),$$

(2)

where P is the current population of solutions and D(x,y) is a dominance measure, such as binary dominance, crowding distance, or a more continuous metric like dominance depth or hypervolume contribution. The Pareto-based minimization function offers a robust mechanism for solving multi-objective optimization problems by incorporating dominance information directly into the optimization process.

Therefore, with regards to the application considered, the optimization problem was formulated using a Pareto-based minimization function, which considers a vector of objective functions:

$$F\left(P\left(t\right)\right)=\left[f_1\left(P\left(t\right)\right), f_2\left(P\left(t\right)\right), f_3\left(P\left(t\right)\right),\dots ,f_k\left(P\left(t\right)\right)\right],$$

(3)

where P(t) is the vector of power allocation at time t, and each f_k represents a distinctive objective, such as:

-

The average remaining charging time of EV at the station,

-

The standard deviation of power distribution,

-

The difference between requested and allocated power in order to minimize the mismatch from requested optimal profile.

The optimization seeks to find power allocation vectors P^*(t) such that:

$$P^{*}\left(t\right)\in \mathsf{{\rm P}}=\{P\mid \nexists P\prime :f_i\left(P\prime \right)\le f_i\left(P\right) \forall i, and \exists j : f_j\left(P\prime \right)<f_j\left(P\right),$$

(4)

where P denotes the Pareto front, the set of all non-dominated solutions.

By employing a Pareto-front-based approach, the system can dynamically choose an optimal power distribution that reflects the current charging station conditions, the state-of-charge of connected EVs, their power acceptance characteristics, and operational constraints, such as total available power or thermal limitations of the infrastructure.

To ensure a balanced charging experience among all connected EVs, the selection of the optimal solution from the Pareto front is guided by a fairness criterion based on the standard deviation of charging times. Specifically, among the non-dominated solutions forming the Pareto front, the final selected power allocation vector P^*(t) must satisfy the following condition: the standard deviation of the individual remaining charging times must not exceed ±5% of the average remaining charging time across all EVs currently being charged.

Formally, let:

-

$T_{c,i}\left(t\right)$ be the estimated remaining charging time for vehicle i at time t,

-

${\mu }_T\left(t\right)={\displaystyle \frac{1}{N\left(t\right)}}{\sum }_{i=1}^{N\left(t\right)}{T}_{c,i}\left(t\right)$ be the average remaing charging time,

-

${\sigma }_T\left(t\right)=\sqrt{{\displaystyle \frac{1}{N\left(t\right)}}{\sum }_{i=1}^{N\left(t\right)}{\left(T_{c,i}\left(t\right)-{\mu }_T\left(t\right)\right)}^2}$ be the standard deviation.

The selection criterion becomes:

$${\sigma }_T\left(t\right)\le 0.05 {\mu }_T$$

(5)

Only those Pareto-optimal solutions that meet this constraint are considered eligible. Among these, the final choice may be refined based on secondary objectives, such as minimal cumulative battery stress or minimal deviation from vehicle-specific optimal charging profiles. This criterion ensures that no individual EV is penalized, promoting fairness and user satisfaction while still leveraging the benefits of dynamic power allocation.

4. Simulation Results

To validate the effectiveness of the proposed Pareto-based power management strategy, a series of simulations were conducted using a MATLAB/Simulink (R2024a) environment. The simulation model replicates the behavior of a reconfigurable MPMP charging station, incorporating the modular architecture described in Section 2. The system consists of eight converter modules and eight charging points, where each CM is capable of delivering up to 80 kW. All the simulations have been performed with reference to a fixed number of electric vehicles (six), which start the charging process simultaneously. Battery capacities and SOC levels were vehicle specific, while maximum charging power was set to 2C for all the vehicles. All the vehicles were characterized by multi-step charging profiles that constrained the acceptable C-rate to decreasing levels as the SOC increased. A vehicle was considered fully charged when the correspondent SOC exceeded 80%.

Firstly, a simulation was carried out with reference to the vehicle set of

Table 2. Characteristics and SOC at the initial time of the EVs used to simulate the first considered charging scenario.

EVs	Initial SOC (%)	Battery Capacity (kWh)	Max. Charging Power (kW)
EV 1	10	120	240
EV 2	20	104	208
EV 3	30	88	176
EV 4	40	72	144
EV 5	50	56	112
EV 6	60	40	80

in order to highlight how the proposed approach allocated the CPs’ available power. The Pareto front was updated at a fixed rate based on the chosen charging process observation time T_O. T_O was set to 2 min. Lower values of T_O would allow for the optimization to perform more effectively but, on the other side, would significantly burden the simulation. The Pareto front at the start of the simulation is reported in Figure 3. The font links the average charging time (horizontal axis) and the normalized standard deviation with respect to the corresponding average charging time (vertical axis). While charging time varied between around 33 and 45 min, the standard deviation varied between around 32% and 1%. By selecting the solution that exhibited a standard deviation of 5%, an average charging time of around 42 min was predicted. It should be noted that the Pareto front was rebuilt as the charging process proceeded. This led to a finalized charging time that could be significantly different from the initial predicted one. The distributions of the charging powers’ allocations corresponding to the Pareto front of Figure 3 are reported in Figure 4. As expected, as the average charging time increased (which corresponded to a decrease in the standard deviations), charging powers decreased at different rates to minimize differences between the charging times of the vehicles set.

Time behaviors of the charging powers and of the batteries’ SOC in the whole process are reported, respectively, in Figure 5 and Figure 6. Figure 5 highlights how the Pareto approach and the vehicles’ power profiles acted simultaneously to limit the charging power reserved to each vehicle. In particular, vehicle 1 was charged initially at 160 kW in order to allocate an acceptable charging power to the other vehicles. The charging power of vehicle 1 was ramped up to 240 kW only after vehicles 5 and 6 had completed their charging process. Then, it was ramped down to lower levels as the power profile capped the charging power.

A second set of simulations was implemented to benchmark the proposed method against a first-come first-served (FCFS) strategy. For each strategy, 100,000 simulations were performed with randomized vehicle sets. Each set was based on six vehicles, with SOC levels randomized in the range 10–60% and batteries capacities randomized in the range 40–120 kW. In each simulation, the performance of the strategy was evaluated by computing for each vehicle the difference between the resulting charging time and the ideal one that would be achieved through a virtual charging point able to supply the maximum charging power compatible with the vehicle profile.

The charging time differences were then synthetized with their average and charging deviation values normalized with respect to the ideal average charging time. Figure 7 and Figure 8 report the distributions of the two performance indexes for the Pareto strategy. The corresponding distributions obtained for the FCFS strategy are shown in Figure 9 and Figure 10. It can be deduced that the Pareto strategy was able to significantly reduce the expected charging time since it shifted the peak of both the distributions to lower values. In particular, if 90% of the population was considered, the PO strategy gave an average charging time error under 54% and a charging time deviation error under 45%, versus 57% and 63% obtained, respectively, by the FCFS strategy.

5. Conclusions

This paper introduced a novel power management method for multi-point multi-power (MPMP) electric vehicle charging stations based on Pareto optimization. The proposed framework leveraged a reconfigurable modular topology, employing dual-active-bridge (DAB) converters and a dynamic switching matrix to flexibly allocate power across charging points operating at discrete levels. By incorporating a multi-objective optimization approach, the system dynamically balanced several competing goals: minimizing average charging time, ensuring fair resource distribution, minimizing power mismatch with battery acceptance profiles, and reducing battery stress.

The simulation results strongly validated the effectiveness of the proposed Pareto-based optimization strategy for power management in modular, reconfigurable MPMP charging infrastructures. Unlike traditional static allocation methods, the dynamic and multi-objective approach adopted in this work enabled significant improvements in charging performance by intelligently distributing available power among multiple charging points.

The results demonstrated that even under constrained conditions, such as a fixed 80 kW per converter module, the system achieved an optimized average charging time of 42.5 min with minimal deviation across vehicles. This represented a significant improvement over the baseline strategy, which resulted in an average charging time of 46.8 min under constant charging power conditions.

The benefits became even more pronounced when the time-varying nature of battery charging behavior was taken into account, and the optimization process was applied iteratively at a sampling interval of 2 min. Through adaptive reallocation of power resources in response to real-time vehicle demands and battery characteristics, the system achieved a substantial reduction in charging time, bringing all vehicles to at least 80% SOC in just 29.4 min. This corresponded to a performance gain of over 30% compared to conventional strategies.

These findings confirmed the suitability of the proposed Pareto-based method for real-world applications, particularly in scenarios involving multiple EVs with heterogeneous charging requirements. Furthermore, the modularity of the charging infrastructure and the scalability of the algorithm make this solution highly adaptable to future high-demand charging stations. The approach not only enhanced efficiency but also contributed to extending battery life by respecting dynamic charging profiles, thus aligning technical performance with long-term sustainability.