Adaptive Recombination-Based Control Strategy for Cell Balancing in Lithium-Ion Battery Packs: Modeling and Simulation

Abstract

This paper presents a novel adaptive cell recombination strategy for balancing lithium-ion battery packs, targeting electric vehicle (EV) applications. The proposed method dynamically adjusts the series–parallel configuration of individual cells based on instantaneous state of charge (SoC) and load demand, without relying on conventional DC-DC converters or passive components. A hardware-efficient switching topology using SPDT (Single Pole Double Throw) switches enables flexible recombination and fault isolation with minimal complexity. The control algorithm, implemented in MATLAB/Simulink, evaluates multiple cell-grouping configurations to optimize balancing speed, energy retention, and operational safety. Simulation results under charging, discharging, and resting conditions demonstrate up to 80% faster balancing compared to sequential methods, with significantly lower component count and minimal energy loss. Validation using Panasonic NCR18650PF cells confirms the model’s real-world applicability. The method offers a scalable, high-speed, and energy-efficient solution for integration into next-generation battery management systems (BMS), achieving performance gains typically reserved for more complex converter-based architectures.

1. Introduction

Lithium-ion batteries are foundational to modern energy storage systems, especially in electric vehicles, where performance, efficiency, and safety depend heavily on maintaining uniform charge distribution across cells in a battery pack. Variations in cell characteristics and operational stress cause imbalances in the state of charge over time, which reduces available capacity, accelerates aging, and can pose safety risks. To mitigate this, battery management systems employ cell-balancing techniques to equalize SoC across all cells [1].

Balancing methods are generally categorized as passive or active. Passive balancing dissipates excess charge from higher-SoC cells through resistive elements. While simple and low-cost, this method is energy-inefficient and unsuitable for high-capacity packs used in EVs. Several studies show that passive balancing alone fails to address capacity mismatches arising from differential aging in series-connected cells, which can exacerbate energy loss and reduce system longevity [2,3]. Active balancing methods redistribute energy between cells using converters, switched capacitors, or transformers. Though more energy-efficient, these solutions increase circuit complexity, component count, and control requirements [4], and often require additional components such as inductors, capacitors, or DC-DC converters, introducing thermal management and design challenges [5].

Several EV manufacturers implement large-scale lithium-ion battery packs consisting of hundreds of cells connected in series and parallel.

Table 1. Battery pack configurations and specifications across selected electric vehicle manufacturers.

Electric Vehicles	Tesla, Inc. [8,9,10,11]	BYD Co., Ltd. [12]	Lucid Group, Inc. [13,14,15]	Porsche AG [16]	Xiaomi Auto [17,18]
EV Model	Model S Plaid	Han EV	Air Dream Edition	Taycan Turbo S	SU7 Max
Introduced In	2021	2020	2021	2019	2024
EV Origin	USA	China	USA	Germany	China
Battery Voltage	407 V	570 V	800 V	800 V	727 V
Energy Capacity	100 kWh	88 kWh	120 kWh	93.4 kWh	101 kWh
Number of Cells	7920 Cells	178 Cells	6600 Cells	396 Cells	198 Cells
Number of Modules	5 Modules	1 Module	22 Modules	33 Modules	1 Module
Module Arrangement	22s72p	178s1p	10s30p	12s1p	198s1p
Series String	22 Cells	178 Cells	10 Cells	12 Cells	198 Cells
Parallel String	72 Series Strings	1 Series String	30 Series Strings	1 Series String	1 Series String
Cell Voltage	3.7 V	3.2 V	3.7 V	3.7 V	3.67 V
Cell Capacity	3.4 Ah	500 Ah	5 Ah	25 Ah	139 Ah
Cell Chemistry	Lithium-ion	Lithium Iron-Phosphate	Lithium-ion	Lithium-ion	Lithium Nickel-Manganese Cobalt-Oxide
Cell Model	Tesla 18650	Prismatic LFP	50G-2170	Pouch	Li-NMC
Cell Manufacturer	Panasonic	BYD	Samsung	Porsche	CATL
Cell Origin	Japan	China	South Korea	Germany	China

provides an overview of battery configurations in commercially available EVs, such as those by Tesla, BYD, etc. These configurations highlight the growing scale and complexity of modern EV battery systems, underscoring the need for scalable and efficient balancing solutions [6,7].

To address this need, researchers have developed numerous active balancing strategies, often using converter-based, modular, or algorithm-driven techniques. Table 2 compares several of these methods in terms of topology, balancing components, and operational trade-offs. Converter-based designs offer flexible energy redistribution but suffer from high component count and control overhead [19,20,21]. Switched-capacitor topologies provide simpler implementations but generally suffer from slow balancing rates, limited transfer capacity, and energy loss [4,22]. Transformer-based and multi-winding designs can handle higher power but are less scalable and more costly [23]. Consensus-based algorithms improve balancing efficiency but typically assume stable load conditions and require communication overhead [24]. Modular and multi-stage balancing topologies, although scalable, introduce additional weight, cost, and control complexity, as highlighted in comparative analyses [25,26].

Despite these developments, a number of limitations remain. Many existing approaches struggle to balance fast convergence, low hardware complexity, and adaptability to changing system states. For instance, Pham et al. proposed a push–pull converter design that provides effective charge redistribution, but introduces complex gate control and requires multiple isolation stages [27]. Transformer-based architecture offers high balancing accuracy, but suffers from limited scalability and increased system weight [23]. Switched-capacitor methods, while cost-effective, often result in slow energy transfer and poor convergence [22]. Additionally, maintaining consistency across cell behaviors becomes more challenging as batteries age or encounter operational anomalies, necessitating adaptive fault detection and recombination schemes [28]. Software-driven cell-balancing approaches have gained interest due to their hardware simplicity and ease of integration with existing BMS architectures [29], and recent developments in low-cost, efficient architectures validate the growing preference for hardware-light implementations that minimize thermal and spatial burdens [30].

Table 2. Comparison of active cell-balancing methods based on topology, components, and operational characteristics.

Balancing Methods	Switches	DC-DC Converters	Other Components	Description
Push–Pull Converter-Based [27]	2N	1	MOSFETs; Diodes; Capacitors; Resistors	Pull charge from high-potential cell and push to low potential cell one at a time with high frequency switching.
Reconfigurable Converter-Based [31]	2N	1	Transistors; Diodes; Capacitors; Inductors	Isolate high-potential cell/s to store charge in inductor and feed to isolated low potential cell/s with high frequency switching.
Bidirectional Cuk Converter-Based [32]	2N	N−1	MOSFETs; Diodes; Capacitors; Inductors	Individual balancing circuit in between every 2 cells. Each cell has access to 2 balancing circuits. High frequency switching.
Event-Triggered Consensus Algorithm [33]	2N−2	N−1	MOSFETs; Diodes; Capacitors; Inductors	Individual balancing circuit in between every 2 cells. Each cell has access to 2 balancing circuits. High frequency switching.
Modular Multilevel Series–Parallel Converter [34]	8N	1	Capacitors; Inductors	Uses 2 4-switch bridge rectifiers for each cell to change series–parallel combination with high frequency switching. Designed for grid storage.
Bidirectional Flyback Converter [35]	4N	2	MOSFETs; Diodes; Capacitors	Uses PWM control with high frequency switching at 2 flyback converters (forward and reverse), and bidirectional cell switches at both positive and negative ends of each cell.
Continuous Current Mode [36]	4N−4	0	N−1 Inductors	Each cell has access to 2 inductors as external storage topology.
Isolated DC-DC Converter [37]	4N	N	Diodes, Transformers	Isolates high-potential cells during charging to charge low-potential cells first. Designed for charging only.
Low-Voltage Output Regulation [38]	2N	N	Capacitors	Steps down all cells individually at different switching frequency to achieve balance.
Hybrid Duty Cycle Balancing [39]	3N	1+M 1	Converter Circuitry	Has 1 central DC-DC converter for the pack and 1 local DC-DC converter for each module.
Inductor-Based [40]	2N+2	0	N−1 Inductors	Targeted to reduce the number of switches, but added inductors as external storage topology. Each cell has access to 2 inductors.
Switched Supercapacitor-Based [41]	N	0	N Capacitors	On–off hysteresis control for supercapacitors as external storage topology connected with each individual cell.
Proposed Method	2N−2	0	None	Changes series–parallel combination of the sting to perform cell-balancing.

¹ M is the number of modules. Each module contains multiple cells in series. All modules in a pack contain equal numbers of cells.

Table 2. Comparison of active cell-balancing methods based on topology, components, and operational characteristics.

Balancing Methods	Switches	DC-DC Converters	Other Components	Description
Push–Pull Converter-Based [27]	2N	1	MOSFETs; Diodes; Capacitors; Resistors	Pull charge from high-potential cell and push to low potential cell one at a time with high frequency switching.
Reconfigurable Converter-Based [31]	2N	1	Transistors; Diodes; Capacitors; Inductors	Isolate high-potential cell/s to store charge in inductor and feed to isolated low potential cell/s with high frequency switching.
Bidirectional Cuk Converter-Based [32]	2N	N−1	MOSFETs; Diodes; Capacitors; Inductors	Individual balancing circuit in between every 2 cells. Each cell has access to 2 balancing circuits. High frequency switching.
Event-Triggered Consensus Algorithm [33]	2N−2	N−1	MOSFETs; Diodes; Capacitors; Inductors	Individual balancing circuit in between every 2 cells. Each cell has access to 2 balancing circuits. High frequency switching.
Modular Multilevel Series–Parallel Converter [34]	8N	1	Capacitors; Inductors	Uses 2 4-switch bridge rectifiers for each cell to change series–parallel combination with high frequency switching. Designed for grid storage.
Bidirectional Flyback Converter [35]	4N	2	MOSFETs; Diodes; Capacitors	Uses PWM control with high frequency switching at 2 flyback converters (forward and reverse), and bidirectional cell switches at both positive and negative ends of each cell.
Continuous Current Mode [36]	4N−4	0	N−1 Inductors	Each cell has access to 2 inductors as external storage topology.
Isolated DC-DC Converter [37]	4N	N	Diodes, Transformers	Isolates high-potential cells during charging to charge low-potential cells first. Designed for charging only.
Low-Voltage Output Regulation [38]	2N	N	Capacitors	Steps down all cells individually at different switching frequency to achieve balance.
Hybrid Duty Cycle Balancing [39]	3N	1+M 1	Converter Circuitry	Has 1 central DC-DC converter for the pack and 1 local DC-DC converter for each module.
Inductor-Based [40]	2N+2	0	N−1 Inductors	Targeted to reduce the number of switches, but added inductors as external storage topology. Each cell has access to 2 inductors.
Switched Supercapacitor-Based [41]	N	0	N Capacitors	On–off hysteresis control for supercapacitors as external storage topology connected with each individual cell.
Proposed Method	2N−2	0	None	Changes series–parallel combination of the sting to perform cell-balancing.

¹ M is the number of modules. Each module contains multiple cells in series. All modules in a pack contain equal numbers of cells.

Recent comparative reviews [1,7] and benchmarking studies [21] emphasize the need for intelligent, scalable strategies that can dynamically adapt to system-level changes in SoC, load demand, and fault conditions. Model-predictive strategies like those in [26] show performance improvements but face challenges in real-time deployment due to computational demands. For example, Vikhorev et al. [42] proposed a dynamic balancing mechanism for LiFePO4 battery packs incorporating BMSs with and without active balancers. While this demonstrates dynamic adaptability, it lacks the fine-grained recombination flexibility or scalability achieved by SPDT-based architectures. Xie et al. [43] implemented a hierarchical structure-based wireless active balancing approach to enhance system scalability and thermal efficiency. However, their solution introduces wireless networking complexity, which the present SPDT-based system avoids while offering similar performance gains. Breglio et al. [44] explored a model-based balancing algorithm using SoC observers and converters. However, the hardware overhead and centralized coordination required in their system contrast with the present study’s decentralized SPDT-switch-based topology.

This paper addresses these gaps by proposing a dynamically configurable and load-responsive cell-balancing strategy based on adaptive cell recombination. The proposed method utilizes a minimal set of SPDT switches to selectively group or isolate cells in series and parallel, based on their instantaneous SoC and load conditions. Unlike converter-based or consensus approaches, this eliminates the need for intermediate energy storage components or bidirectional energy pathways and significantly reduces hardware complexity.

The method achieves faster balancing with fewer switches, maintains system scalability, and dynamically adjusts balancing operations in response to load variation. It is simulated in MATLAB/Simulink across all operating conditions of resting, charging, and discharging, using both ideal and realistic cell models.

The contributions of this work are summarized as follows:

• Development of a dynamically configurable and load-responsive control logic that enables adaptive cell recombination without requiring external converters or passive dissipative elements;
• Design of a switching architecture using 2N−2 SPDT switches for full-cell accessibility and scalable implementation;
• Simulation of the proposed strategy in MATLAB/Simulink under resting, charging, and discharging conditions while balancing time comparison;
• Comparative analysis with state-of-the-art balancing strategies in terms of SoC uniformity and hardware complexity;
• Validation using a realistic Panasonic NCR18650PF lithium-ion cell model in simulation to assess discharge performance and confirm applicability in practical systems.

2. Proposed Methodology

2.1. System Architecture

The proposed battery pack configuration consists of N lithium-ion cells connected via a switching matrix composed of SPDT switches. Each non-terminal cell is connected to its immediate neighbors (previous and next cell) through one upper and one lower switch (Su, Sl). Terminal cells have single-sided connections. The control system enables simplified switching to achieve dynamic series–parallel recombination.

As shown in Figure 1, the SPDT switch matrix allows for flexible reconnections between neighboring cells. The control surface generates switching signals based on each cell’s status, which are then applied through the switch network. The minimum number of switches required for N cells follows the expression in Equation (1). This hardware configuration avoids the use of DC-DC converters or energy storage components.

$$N_S=2N-2$$

(1)

2.2. Control Algorithm

The Battery Management System continuously monitors:

• State of charge of each cell: SoC_cell.
• Pack-level voltage (V_pack) and current (I_pack).
• Load power requirement (P_load).
• Battery phase (charging, discharging, resting).
• Cell fault status (binary flag).

Using these data, the BMS computes the average SoC, minimum required cell count for load, cell-specific balancing error, and the fault condition (based on the SoC_min threshold) using Equations (2)–(5).

$${SoC}_{avg}={\displaystyle \frac{1}{N}}{\sum }_{i=1}^N{SoC}_i$$

(2)

$${CC}_{min}=⌈{\displaystyle \frac{P_{Load}·n}{V_{pack}·I_{rated}}}⌉$$

(3)

$${Error}_i={SoC}_{avg}-{SoC}_i$$

(4)

$${Fault}_i=\left\{\begin{array}{c}1 if {SoC}_i<{SoC}_{min}\\ 0 if {SoC}_i\le {SoC}_{min}\end{array}\right.$$

(5)

Based on this evaluation, the BMS executes priority indexing. Cells are ordered by SoC deviation to optimize balancing efficiency. Priority logic is defined as prioritizing the lowest-SoC cell first during discharging and resting, as presented in Equation (6). In case of equal SoC error, selection is based on greater SoC gradient to neighbors, as presented in Equation (7). Terminal cells with equal error are deprioritized unless balancing is isolation-based, as presented in Equation (8).

$$\left[\begin{array}{c}{P}_1\\ {\begin{array}{c}:\\ P\end{array}}_n\end{array}\right]=\left[\begin{array}{c}{SoC}_{lowest}\\ {\begin{array}{c}:\\ SoC\end{array}}_{highest}\end{array}\right]$$

(6)

$$\left[\begin{array}{c}{SoC}_{low-1}\\ {SoC}_{low-2}\end{array}\right]=\left[\begin{array}{c}{P}_1\\ P_2\end{array}\right] :\left\{\left({SoC}_{low-1}-{SoC}_{prev}\right)>\left({SoC}_{low-2}-{SoC}_{next}\right)\right\}$$

(7)

$${SoC}_{low-n}=P_{last} :\left\{\left({SoC}_{low-n}-{SoC}_{prev}\right)=\left({SoC}_{low-n}-{SoC}_{next}\right)\right\}$$

(8)

The complete control flow is illustrated in Figure 2, and the iteration logic is shown in Figure 3.

2.3. Switching Strategy

Each non-terminal cell in the pack can assume up to 11 possible recombination states, depending on the conditions of adjacent cells and system requirements. Figure 4 illustrates the default configuration, where all switches are off and the cells are connected in series. This serves as the baseline mode against which all recombination scenarios are executed.

The switching logic used to enable various recombination states is defined in

Table 3. Logical cell combinations for any middle cell C_n.

Cell Combinations	Explanation
C(n−1), Cn, C(n+1)	Cn is in series with previous and next cell
(C(n−1) ǁ Cn), C(n+1)	Cn is in parallel with prev. cell, and together in series with next cell
C(n−1), (Cn ǁ C(n+1))	Cn is in parallel with next cell, and together in series with prev. cell
C(n−1) ǁ Cn ǁ C(n+1)	Cn is in parallel with previous and next cell
!C(n−1), Cn, C(n+1)	Cn is in series with next cell, previous cell is isolated
!C(n−1), (Cn ǁ C(n+1))	Cn is in parallel with next cell, previous cell is isolated
C(n−1), !Cn, C(n+1)	Cn is isolated, previous and next cells are in series
C(n−1), Cn, !C(n+1)	Cn is in series with previous cell, next cell is isolated
(C(n−1) ǁ Cn), !C(n+1)	Cn is in parallel with previous cell, next cell is isolated
!C(n−1), !Cn, C(n+1)	Cn and its previous cell both are isolated, leaving only its next cell
C(n−1), !Cn, !C(n+1)	Cn and its next cell are both isolated, leaving only its previous cell

. Each recombination pattern is determined by activating the appropriate upper and lower SPDT switches (Su and Sl) for a given cell. A schematic of how a middle cell (Cn) interfaces with its neighbors through switching logic is presented in Figure 5.

For a clearer understanding of how switching commands translate to actual recombination patterns,

Table 4. Switching commands for recombination in a 3-cell pack.

Cell-Combinations	Su1	Sl1	Su2	Sl2
C1, C2, C3	0	0	0	0
(C1 ǁ C2), C3	1	1	0	0
C1, (C2 ǁ C3)	0	0	1	1
C1 ǁ C2 ǁ C3	1	1	1	1
!C1, C2, C3	0	1	0	0
C1, !C2, C3	0	0	0	1
C1, C2, !C3	0	0	1	0
!C1, !C2, C3	0	1	0	1
C1, !C2, !C3	1	0	1	0
!C1, (C2 ǁ C3)	0	1	1	1
(C1 ǁ C2), !C3	1	1	1	0

presents the corresponding Su/Sl states required for a 3-cell battery pack. This example illustrates how individual switch activations create distinct electrical topologies within the proposed model. A visual summary of all 11 recombination modes applicable to each non-terminal cell is provided in Appendix A.1 as Figure A1. These configurations represent the logical foundation of the switching strategy employed in the proposed model.

To maximize flexibility in balancing operations, the control logic supports up to 81 valid recombination configurations for a 5-cell battery pack. These configurations are derived from combinations of serial, parallel, and isolation modes applied to each cell. A complete enumeration of these configurations is provided in Appendix A.2 as

Table A1. Complete set of recombination modes in a 5-cell pack under proposed logic.

#.	Cell Combinations	Cell Count	#.	Cell Combinations	Cell Count	#.	Cell Combinations	Cell Count
1	C1, C2, C3, C4, C5	5	28	C1, C2, (C3 ǁ C4), !C5	3	55	!C1, (C2 ǁ C3), !C4, C5	2
2	(C1 ǁ C2), C3, C4, C5	4	29	!C1, !C2, C3, C4, C5	3	56	!C1, (C2 ǁ C3), C4, !C5	2
3	C1, (C2 ǁ C3), C4, C5	4	30	!C1, C2, !C3, C4, C5	3	57	!C1, C2, (C3 ǁ C4), !C5	2
4	C1, C2, (C3 ǁ C4), C5	4	31	!C1, C2, C3, !C4, C5	3	58	C1, !C2, !C3, (C4 ǁ C5)	2
5	C1, C2, C3, (C4 ǁ C5)	4	32	!C1, C2, C3, C4, !C5	3	59	C1, !C2, (C3 ǁ C4), !C5	2
6	!C1, C2, C3, C4, C5	4	33	C1, !C2, !C3, C4, C5	3	60	(C1 ǁ C2), !C3, !C4, C5	2
7	C1, !C2, C3, C4, C5	4	34	C1, !C2, C3, !C4, C5	3	61	(C1 ǁ C2), !C3, C4, !C5	2
8	C1, C2, !C3, C4, C5	4	35	C1, !C2, C3, C4, !C5	3	62	(C1 ǁ C2), C3, !C4, !C5	2
9	C1, C2, C3, !C4, C5	4	36	C1, C2, !C3, !C4, C5	3	63	C1, (C2 ǁ C3), !C4, !C5	2
10	C1, C2, C3, C4, !C5	4	37	C1, C2, !C3, C4, !C5	3	64	!C1, !C2, !C3, C4, C5	2
11	(C1 ǁ C2), (C3 ǁ C4), C5	3	38	C1, C2, C3, !C4, !C5	3	65	!C1, !C2, C3, !C4, C5	2
12	(C1 ǁ C2), C3, (C4 ǁ C5)	3	39	(C1 ǁ C2 ǁ C3), (C4 ǁ C5)	2	66	!C1, !C2, C3, C4, !C5	2
13	C1, (C2 ǁ C3), (C4 ǁ C5)	3	40	(C1 ǁ C2), (C3 ǁ C4 ǁ C5)	2	67	C1, !C2, !C3, !C4, C5	2
14	(C1 ǁ C2 ǁ C3), C4, C5	3	41	(C1 ǁ C2 ǁ C3 ǁ C4), C5	2	68	C1, !C2, !C3, C4, !C5	2
15	C1, (C2 ǁ C3 ǁ C4), C5	3	42	C1, (C2 ǁ C3 ǁ C4 ǁ C5)	2	69	C1, C2, !C3, !C4, !C5	2
16	C1, C2, (C3 ǁ C4 ǁ C5)	3	43	!C1, (C2 ǁ C3), (C4 ǁ C5)	2	70	C1 ǁ C2 ǁ C3 ǁ C4 ǁ C5	1
17	!C1, (C2 ǁ C3), C4, C5	3	44	!C1, (C2 ǁ C3 ǁ C4), C5	2	71	!C1, (C2 ǁ C3 ǁ C4 ǁ C5)	1
18	!C1, C2, (C3 ǁ C4), C5	3	45	!C1, C2, (C3 ǁ C4 ǁ C5)	2	72	(C1 ǁ C2 ǁ C3 ǁ C4), !C5	1
19	!C1, C2, C3, (C4 ǁ C5)	3	46	C1, !C2, (C3 ǁ C4 ǁ C5)	2	73	!C1, !C2, (C3 ǁ C4 ǁ C5)	1
20	C1, !C2, (C3 ǁ C4), C5	3	47	(C1 ǁ C2), !C3, (C4 ǁ C5)	2	74	!C1, (C2 ǁ C3 ǁ C4), !C5	1
21	C1, !C2, C3, (C4 ǁ C5)	3	48	(C1 ǁ C2 ǁ C3), !C4, C5	2	75	(C1 ǁ C2 ǁ C3), !C4, !C5	1
22	(C1 ǁ C2), !C3, C4, C5	3	49	(C1 ǁ C2), (C3 ǁ C4), !C5	2	76	!C1, !C2, !C3, (C4 ǁ C5)	1
23	C1, C2, !C3, (C4 ǁ C5)	3	50	(C1 ǁ C2 ǁ C3), C4, !C5	2	77	!C1, !C2, (C3 ǁ C4), !C5	1
24	(C1 ǁ C2), C3, !C4, C5	3	51	C1, (C2 ǁ C3 ǁ C4), !C5	2	78	(C1 ǁ C2), !C3, !C4, !C5	1
25	C1, (C2 ǁ C3), !C4, C5	3	52	!C1, !C2, (C3 ǁ C4), C5	2	79	!C1, !C2, !C3, !C4, C5	1
26	(C1 ǁ C2), C3, C4, !C5	3	53	!C1, !C2, C3, (C4 ǁ C5)	2	80	!C1, !C2, !C3, C4, !C5	1
27	C1, (C2 ǁ C3), C4, !C5	3	54	!C1, C2, !C3, (C4 ǁ C5)	2	81	C1, !C2, !C3, !C4, !C5	1

. In the main body of this paper, representative examples are presented to demonstrate how the system dynamically selects optimal recombination patterns.

3. Simulation Setup and Performance Evaluation

3.1. Simulation Setup

To validate the performance of the proposed cell-balancing strategy, a five-cell lithium-ion battery pack was modeled and simulated using MATLAB/Simulink. The objective of the simulation was to evaluate the adaptive recombination logic under three distinct operating conditions: resting, charging, and discharging. The model does not utilize any passive or active balancing circuitry, and instead relies entirely on dynamic SPDT switching. A five-cell configuration was selected to examine the responsiveness of the recombination algorithm, the behavior of the switching logic, and the resulting balancing performance under controlled conditions. This configuration allows for a clear observation of the interaction dynamics, particularly in the case of the middle cell, while ensuring that both terminal and intermediate positions are represented. Furthermore, the selected configuration provides a feasible and scalable basis for future expansion to larger battery systems, acknowledging that the total number of cells in electric vehicle applications varies significantly between manufacturers.

The Simulink implementation of the full model is shown in Figure 6. Each cell is modeled as an ideal lithium-ion cell with a nominal voltage of 3.6 V and capacity of 100 mAh. The voltage profile shown in Figure 7 corresponds to the open-circuit voltage (OCV) behavior of the modeled cells during resting conditions. The SoC is obtained from the internal SoC tracking mechanism of the Simulink battery block, which uses a combination of coulomb counting and voltage-based estimation under the predefined cell parameters.

The simulation environment was configured to assess how SoC deviation evolves over time as different recombination strategies are applied. The pack operates in either one-cell-at-a-time or all-cells-at-once modes. This allows for comparative analysis of balancing performance, speed, and energy efficiency.

To examine current behavior under varying cell counts, a comparative analysis of pack current versus voltage during constant power discharge is presented in Figure 8. The figure illustrates how the current dynamically changes with respect to reduced active cell count, highlighting how load distribution adapts based on the number of active series-connected cells. The simulation assumes a rated current limit of 5 A and a load demand of 50 W, as outlined in

Table 5. Simulation parameters used for evaluating the proposed 5-cell BMS model under various operating modes.

Parameters	Values
Switch Type	SPDT
Cell Type	lithium-ion
Cell Capacity	100 mAh
Max. Cell Current	400 mA
Number of Cells	5
Number of Switches	8
Nominal Cell Voltage	3.6 V
Nominal Pack Voltage	18 V
DC Source	24 V
DC Load	1 W

.

Using these values, the minimum number of active cells required to support the load can be determined based on Equation (2). For instance, when cell voltage is 3.6 V, the required minimum number of active cells becomes 3, and if the voltage drops to 2.8 V, the required number of active cells increases to 4, as per Equation (2). This demonstrates how CC_min dynamically changes depending on instantaneous cell voltage. All simulation parameters used across the scenarios are listed in Table 5. These include simulation step size, capacity, voltage range, power demand, switching interval, and balancing threshold. Recombination topologies based on varying active cell counts (from five to one) were tested. Examples of these topologies, including isolation and parallel pairing, are shown in Figure 9. Each configuration is used under different balancing phases, triggered by control logic depending on SoC deviation and fault conditions.

The simulations also rely on previously defined control parameters from Equations (1)–(5), introduced in earlier sections. In this stage, balancing convergence is monitored using a threshold-based approach, where the simulation is terminated when maximum SoC deviation (∆SoC) drops below a preset threshold, as defined in Equations (9) and (10).

$$△SoC={SoC}_{max}-{SoC}_{min}$$

(9)

$$△SoC<0.01 (i.e.,1\% deviation)$$

(10)

This balancing criterion ensures fairness across cell states without causing frequent switching near convergence.

3.2. Balancing in Resting Condition

The resting condition represents a battery pack state in which no current flows into or out of the system—effectively simulating periods of vehicle dormancy or standby phases in battery-driven systems. Although no external current is present, internal imbalances persist and may worsen over time due to intrinsic differences in cell characteristics. The objective of this simulation is to evaluate how the proposed recombination-based cell-balancing strategy performs in equalizing the state of charge (SoC) under purely idle conditions, without the influence of charging or discharging currents.

To assess the effectiveness of the control logic, three test scenarios were designed with varying degrees of initial SoC deviation, reflecting real-world cell inconsistencies. These distributions are detailed in

Table A2. Initial SoC values used in three test cases to simulate varying levels of imbalance under resting conditions.

Test Cases	Cell-1	Cell-2	Cell-3	Cell-4	Cell-5	Initial ∆SoC
1	15%	98%	96%	49%	80%	83%
2	66%	71%	26%	71%	57%	45%
3	46%	53%	48%	50%	50%	7%

, categorized as mild (7%), moderate (45%), and severe (83%) imbalances. For each scenario, simulations were carried out using two distinct balancing approaches: one-cell-at-a-time recombination and all-cells-at-once recombination. The simulation was terminated once the SoC deviation between the most and least charged cells met the balancing convergence threshold, defined by Equation (10) as ∆SoC < 0.01, ensuring uniform charge distribution across the pack.

The results of the simulation are presented in

Table A3. Final SoC deviation, SoC average, and balancing time for one-cell-at-a-time and all-cells-at-once recombination strategies under resting conditions.

Test Cases	Initial ∆SoC	Initial SoCavg	One-Cell-at-a-Time			All-Cells-at-a-Time
			Final ∆SoC	Final SoCavg	Balancing Time	Final ∆SoC	Final SoCavg	Balancing Time
1	83%	67.6%	0.00%	67.6%	470 s	0.87%	67.6%	160 s
2	45%	58.2%	0.83%	58.2%	100 s	0.83%	58.2%	100 s
3	7%	49.4%	0.50%	49.4%	60 s	0.87%	49.4%	40 s

, which compares balancing time, final ∆SoC, and SoC_avg across the three test cases and two strategies. The accompanying plots in Figure 10 provide a visual representation of SoC convergence over time. Figure 10a, Figure 10c, and Figure 10e correspond to the one-cell-at-a-time method, while Figure 10b, Figure 10d, and Figure 10f depict the all-cells-at-once results for Test Cases 1 through 3, respectively.

A clear performance advantage is evident for the all-cells-at-once strategy. In Test Case 3, where initial deviation was as high as 83%, the balancing time was reduced by approximately 80% compared to the sequential approach. Even in lower-deviation scenarios, the multi-cell recombination configuration consistently outperformed the one-cell-at-a-time method in terms of speed, without introducing additional energy loss. This is reflected by the stability of the SoC_avg between the initial and final states, confirming that the algorithm does not sacrifice stored energy for speed. Additionally, none of the strategies led to over-balancing or SoC inversion, which indicates the reliability of the algorithm even under aggressive recombination schedules.

The simulation confirms that the proposed architecture is capable of achieving full SoC equalization autonomously in resting states, making it suitable for embedded energy maintenance during idle periods. This advantage is particularly valuable in electric vehicle systems, where passive drain or slight manufacturing variance can affect performance even when the vehicle is not in active operation.

3.3. Balancing in Charging Conditions

This simulation scenario investigates the system’s behavior when the battery pack is subjected to a constant current charging condition. Charging events are typically rapid and highly dynamic, especially in electric vehicles, and may exacerbate SoC divergence if the charging current is distributed unevenly across imbalanced cells. Therefore, the purpose of this simulation is to evaluate the proposed control strategy’s ability to manage cell balancing during active energy input.

Three test cases with distinct SoC distributions were analyzed, consistent with the resting condition tests to maintain comparative integrity. The initial SoC values used in these simulations are presented in

Table A4. Initial SoC distributions for three test cases representing varying imbalance levels under charging conditions.

Test Cases	Cell-1	Cell-2	Cell-3	Cell-4	Cell-5	Initial ∆SoC
1	18%	44%	77%	78%	80%	62%
2	25%	37%	42%	49%	56%	31%
3	45%	46%	48%	50%	51%	6%

. Each case was simulated using both the one-cell-at-a-time and the all-cells-at-once recombination strategies. During charging, the algorithm prioritizes the charging of low-SoC cells by recombining them in parallel to increase their charging rate, while temporarily bypassing high-SoC cells to prevent overcharging. The simulation was concluded when ∆SoC fell below the convergence threshold defined earlier in Equation (10).

The simulation results, summarized in

Table A5. Final SoC deviation, average SoC, and time to convergence for both recombination strategies during charging simulations.

Test Cases	Initial ∆SoC	Initial SoCavg	One-Cell-at-a-Time			All-Cells-at-a-Time
			Final ∆SoC	Final SoCavg	Balancing Time	Final ∆SoC	Final SoCavg	Balancing Time
1	62%	59.4%	0.42%	80.2%	12 min	0.99%	66.6%	250 s
2	31%	41.8%	0.67%	55.7%	8 min	0.98%	48.2%	220 s
3	6%	48.0%	0.97%	52.3%	150 s	0.97%	52.3%	150 s

, include the final SoC_avg, final ∆SoC, and the total time required to reach convergence for each strategy. The balancing progress is visually illustrated in Figure 11, where the left column (Figure 11a, Figure 11c, Figure 11e) represents the one-cell-at-a-time recombination method, and the right column (Figure 11b, Figure 11d, Figure 11f) shows results using all-cells-at-once recombination across Test Cases 1 to 3, respectively.

The findings clearly indicate that all-cells-at-once recombination outperforms the sequential method across all test cases, particularly under larger initial SoC deviations. In Test Case 3, the most imbalanced scenario, convergence was achieved in approximately 40% less time using the multi-cell strategy. Additionally, the SoC_avg across all test cases remained stable throughout the process, confirming that the algorithm efficiently distributes the charging energy without skewing the overall energy profile of the pack.

Notably, the algorithm demonstrated dynamic adaptability in regulating the recombination topology in real time as cells approached similar charge levels. High-SoC cells were temporarily isolated until their neighbors caught up, after which they were reintegrated into the charging path. This behavior reduced the risk of overcharging and highlights the control logic’s ability to enforce charging fairness without the need for additional hardware such as voltage clamps or bypass circuitry.

These results validate the practical viability of the proposed switching architecture under charging conditions. Its capability to reconfigure cell topology in real time ensures faster balancing while maintaining energy integrity, making it a suitable candidate for implementation in smart BMS platforms for EVs and other high-performance battery systems.

3.4. Balance in Discharging Conditions

The discharging condition scenario simulates the battery pack supplying power at a constant load, which represents the most critical operational phase for electric vehicles and high-drain battery applications. In this state, cell-level SoC imbalances are most likely to escalate due to varying discharge rates, especially in the absence of active control. The objective of this simulation is to examine the effectiveness of the proposed recombination-based balancing logic during load operation, where maintaining system stability and power continuity is essential.

To maintain a uniform test design, the same three SoC deviation scenarios used in previous simulations were applied. The initial SoC values for Test Cases 1, 2, and 3 under discharging conditions are presented in

Table A6. Initial SoC distributions for three test cases under constant power discharge conditions.

Test Cases	Cell-1	Cell-2	Cell-3	Cell-4	Cell-5	Initial ∆SoC
1	14%	37%	53%	55%	96%	82%
2	39%	44%	53%	63%	71%	32%
3	45%	49%	50%	53%	55%	10%

. The balancing control logic seeks to discharge higher-SoC cells more aggressively by selectively isolating or bypassing lower-SoC cells, preserving their charge until parity is restored. Both one-cell-at-a-time and all-cells-at-once recombination strategies were executed, and the system was monitored until the convergence threshold of ∆SoC < 0.01, defined in Equation (10), was satisfied.

The final SoC values, average SoC, and balancing durations are summarized in

Table A7. Final SoC deviation, average SoC, and balancing duration for each recombination strategy under discharging conditions.

Test Cases	Initial ∆SoC	Initial SoCavg	One-Cell-at-a-Time			All-Cells-at-a-Time
			Final ∆SoC	Final SoCavg	Balancing Time	Final ∆SoC	Final SoCavg	Balancing Time
1	82%	51.0%	0.12%	13.7%	40 min	0.87%	47.7%	220 s
2	32%	54.0%	0.79%	39.5%	16 min	0.95%	51.3%	180 s
3	10%	50.4%	0.85%	46.6%	320 s	0.94%	48.6%	120 s

. Visual comparisons of SoC balancing trajectories are provided in Figure 12, where Figure 12a, Figure 12c, and Figure 12e show the one-cell-at-a-time strategy, and Figure 12b, Figure 12d, and Figure 12f illustrate the all-cells-at-once method across the three test scenarios, respectively.

The simulation results highlight several important observations. First, while both strategies succeeded in achieving SoC convergence, the all-cells-at-once approach consistently required less simulation time, demonstrating its superiority in dynamic load conditions. In Test Case 2, for example, the balancing process was completed nearly 45% faster compared to the sequential method. This time saving becomes particularly advantageous in high-power applications, where energy optimization is critical.

Second, in contrast to resting and charging modes, the average SoC at the end of the discharging simulations is significantly lower than the initial SoC_avg. This outcome reflects the energy expended by the pack during the load cycle and further validates the realism of the model under active discharge conditions.

Third, the control algorithm adeptly avoided excessive stress on weaker cells by temporarily removing them from the discharge path, allowing stronger cells to support the load until SoC levels equilibrated. This not only extends the operational lifespan of the pack but also contributes to safer discharge behavior without requiring additional load-sharing circuits.

Overall, the results under discharging conditions confirm that the proposed adaptive recombination strategy provides reliable, efficient, and safe balancing behavior, even in the most demanding operational scenarios. Its dynamic responsiveness, hardware simplicity, and balancing speed offer strong justification for future embedded implementation in EV battery management systems.

4. Model Validation Using Panasonic NCR18650PF Cell Parameters

To evaluate the applicability of the proposed balancing strategy under practical operating conditions, the ideal five-cell configuration was replaced with Panasonic NCR18650PF lithium-ion cells. This specific cell model is available as a predefined component in MATLAB/Simulink’s Battery Block Parameterization Manager and incorporates a second-order equivalent circuit representation. It accounts for key electrochemical behaviors such as temperature dependence, internal resistance, and dynamic SoC–voltage characteristics, enabling more realistic discharge simulations. In contrast, the earlier simulations used idealized cells with linear voltage-to-SoC relationships and no internal losses. These ideal models were intentionally used first to isolate and verify the correctness of the recombination control logic under simplified conditions, before introducing the parametric nonlinearities associated with real-world cell behavior.

In the Simulink model shown earlier in Figure 6, all five ideal cells used in previous simulations were replaced with NCR18650PF cells. The predefined model was selected directly from the Simulink parameterization interface. The model includes empirical data based on manufacturer specifications and is preconfigured to reflect thermal characteristics, internal resistance, and SoC–voltage dynamics under standard operating conditions. The manufacturer’s datasheet was used to cross-reference the cell’s nominal characteristics, such as capacity, voltage, and current ratings, which are summarized in

Table 6. Electrical specifications of the Panasonic NCR18650PF lithium-ion cell used for simulation.

Parameters	Values
Cell Type	Lithium-ion
Cell Capacity	2700 mAh
Nominal Cell Voltage	3.6 V
Nominal Cell Current	1375 mA
Part Number	NCR18650PF
Manufacturer	Panasonic

. These parameters differ from the ideal cell values used in Table 5, introducing real-world complexity into the simulation environment.

Due to the temperature-dependent behavior of the selected cell model and the absence of active thermal control in this study, the simulation temperature was fixed at 25 °C. To verify the accuracy of the Simulink cell model, a discharge simulation at 25 °C was performed and qualitatively compared with the discharge curve provided in the Panasonic NCR18650PF datasheet. Although only the simulated profile is presented in Figure 13 due to copyright limitations regarding the datasheet figures and information, the simulated discharge characteristics closely follow the overall behavior and trends reported in the manufacturer’s data [45]. This consistency supports the model’s fidelity and indicates that no voltage offset correction was required.

To test the balancing logic under the Panasonic cell parameters, three different SoC distributions were created to simulate resting, charging, and discharging scenarios. The initial SoC values for each test case are presented in

Table A8. Initial SoC values used for simulation with Panasonic NCR18650PF cells across three operating conditions.

Test Cases	Operating Condition	Cell-1	Cell-2	Cell-3	Cell-4	Cell-5	Initial ∆SoC
1	Resting	50%	73%	54%	54%	36%	37%
2	Charging	47%	49%	57%	61%	63%	16%
3	Discharging	34%	44%	47%	64%	65%	31%

. All other simulation parameters, such as balancing threshold (∆SoC < 1%), control intervals, and logic flow, remained unchanged from the previous simulations using ideal cells. The final simulation results are summarized in

Table A9. Simulation results of cell balancing using Panasonic NCR18650PF cells under resting, charging, and discharging scenarios.

Test Cases	Initial ∆SoC	Initial SoCavg	One-Cell-at-a-Time			All-Cells-at-a-Time
			Final ∆SoC	Final SoCavg	Balancing Time	Final ∆SoC	Final SoCavg	Balancing Time
1	37%	53.4%	0.70%	53.4%	80 min	0.87%	53.4%	75 min
2	16%	55.4%	0.86%	62.4%	115 min	0.92%	59.1%	60 min
3	31%	50.8%	0.97%	34.7%	8.25 h	0.90%	47.4%	105 min

, showing initial and final SoC deviation (∆SoC), average SoC (SoC_avg), and total balancing time for both one-cell-at-a-time and all-cells-at-once strategies. The SoC evolution plots corresponding to these test cases are illustrated in Figure 14.

Across all operating conditions, the proposed strategy successfully achieved SoC balancing within the defined threshold. However, the overall balancing time was noticeably longer than in simulations using 100 mAh ideal cells. This increase can be attributed to the larger capacity (2700 mAh) and slower response characteristics of the Panasonic cells. Nonetheless, the recombination logic demonstrated the same behavioral patterns—such as midpoint convergence during resting, upward convergence during charging, and downward convergence during discharging—as previously observed with ideal cells.

Notably, in all test cases, the all-cells-at-once strategy outperformed the sequential method in terms of speed, reaffirming earlier conclusions. The Panasonic cell validation thus confirms that the proposed model can adapt to real-world battery behavior while maintaining reliable and efficient balancing control.

5. Comparative Analysis

To establish the broader relevance of the proposed adaptive recombination strategy, three tiers of comparison have been made: (1) balancing behavior compared to established methods from the literature, (2) topology-level differences in hardware complexity, and (3) balancing performance under varying cell connection schemes. These comparisons underscore the model’s flexibility, minimal hardware overhead, and superior balancing speed across different operating conditions.

The first comparison highlights how the proposed model aligns with or outperforms existing strategies under resting, charging, and discharging conditions. As shown in

Table 7. Functional comparison of SoC balancing behavior across existing methods and the proposed model under resting, charging, and discharging conditions.

Balancing Models	Resting Condition	Charging Condition	Discharging Condition
Push–Pull Converter-Based Model [5]	Final SoCavg = Initial SoCavg	Final SoCavg = 100% SoC	Final SoCavg = 0% SoC
Reconfigurable Converter-Based Model [6]	Initial SoCavg > Final SoCavg Final SoCavg > 0% SoC	N/A	Initial SoCavg > Final SoCavg Final SoCavg > 0% SoC
Bidirectional Cuk Converter-Based Model [22]	Final SoCavg = Initial SoCavg	N/A	Initial SoCavg > Final SoCavg Final SoCavg > 0% SoC
Event-Triggered Consensus Algorithm [27]	Initial SoCavg > Final SoCavg Final SoCavg > 0% SoC	Initial SoCavg < Final SoCavg Final SoCavg < 100% SoC	Initial SoCavg > Final SoCavg Final SoCavg > 0% SoC
Low-Voltage Output Regulation [36]	N/A	N/A	Initial SoCavg > Final SoCavg Final SoCavg > 0% SoC
Hybrid Duty Cycle Balancing Model [39]	N/A	Initial SoCavg < Final SoCavg Final SoCavg < 100% SoC	Initial SoCavg > Final SoCavg Final SoCavg > 0% SoC
Inductor-Based Model [40]	Initial SoCavg > Final SoCavg Final SoCavg > 0% SoC	Initial SoCavg < Final SoCavg Final SoCavg < 100% SoC	Initial SoCavg > Final SoCavg Final SoCavg > 0% SoC
Proposed Model	Final SoCavg = Initial SoCavg	Initial SoCavg < Final SoCavg Final SoCavg < 100% SoC	Initial SoCavg > Final SoCavg Final SoCavg > 0% SoC

, the balancing efficiency of the proposed model meets or exceeds that of push–pull converters, bidirectional Cuk converters, and other consensus- or inductor-based algorithms. Particularly at resting condition, the final average SoC (SoC_avg) equals the initial SoC_avg, indicating zero energy loss during balancing, an important marker of efficiency. During charging and discharging, SoC_avg adjusts in expected directions while staying within safe thresholds, affirming control stability and robustness.

A second dimension of evaluation centers on hardware simplicity and flexibility.

Table 8. Comparison of proposed model with existing balancing models in 3-cell configuration, showing achievable cell combinations, switching requirements, and hardware complexity.

Balancing Models	Possible Combinations	Switch Required	DC-DC Conversion	High Freq. Switching
Isolated DC-DC Converter-Based Charging Model [32]	6	12	3	Yes
Reconfigurable Converter-Based Model [6]	6	6	1	Yes
Modular Multilevel Series–Parallel Converter-Based Model [28]	11	24	1	Yes
Proposed Model	11	4	0	No

summarizes the comparison between the proposed model and related topologies that use series–parallel reconfiguration and cell isolation. The proposed model achieves the highest number of possible cell combinations (11) with the lowest switch count (4), while eliminating the need for DC-DC conversion or high-frequency switching. This translates to fewer components, reduced cost, and better scalability compared to isolated DC-DC, modular multilevel, or reconfigurable converter-based designs.

Finally, the model was benchmarked under different connection schemes to assess how much time can be saved using the all-cells-at-once recombination strategy, particularly under larger SoC deviations. The results are shown in

Table 9. Balancing time comparison between one-cell-at-a-time and all-cells-at-once connection schemes for both ideal and Panasonic NCR18650PF cells.

Experimental Cells	Operating Condition	One-Cell-at-a-Time Connection Scheme	All-Cells-at-a-Time Connection Scheme	Reduction Rate
Ideal Cell	Resting	100 s	100 s	0.0%
	Charging	8 min	220 s	54.2%
	Discharging	16 min	180 s	81.3%
Panasonic Cell	Resting	80 min	75 min	6.3%
	Charging	115 min	60 min	47.8%
	Discharging	8.25 h	105 min	78.8%

. For both ideal and Panasonic cell simulations, the all-cells-at-once approach consistently outperformed the sequential connection strategy. Under discharging conditions, balancing time was reduced by up to 81.3% for ideal cells and 78.8% for real Panasonic cells. The charging condition also saw notable improvements (~50%), while the resting condition showed limited time savings, indicating that the advantage of full parallel connection is most significant when energy flow is active.

These comparative outcomes provide compelling justification for the proposed method’s deployment in practical systems. It balances technical rigor with real-world adaptability, minimizing hardware, maximizing flexibility, and accelerating convergence, all while maintaining safety and efficiency under diverse cell behaviors and system states. While sequential recombination strategies serve as a practical baseline, the proposed method also outperforms more complex approaches including converter-based and consensus balancing models. It achieves convergence with significantly fewer components, eliminates high-frequency switching, and avoids energy loss, as confirmed in both ideal and real-cell simulations. These technological improvements support not only theoretical merit but also practical deployment in embedded BMS applications.

In comparison with state-of-the-art balancing approaches, the proposed strategy achieves faster convergence and improved energy efficiency while avoiding complex or costly hardware components such as transformers or bidirectional converters. By leveraging a control-driven SPDT switching logic with minimal interconnection, the method outperforms traditional converter-based and consensus strategies in terms of scalability, simplicity, and balancing performance. The advantages are both qualitatively and quantitatively summarized in Table 2, Table 7 and Table 8.

6. Discussion and Conclusions

This paper presents a novel control-based cell-balancing strategy using adaptive recombination logic enabled by SPDT switching. The unique contribution of this work lies in its recombination-centric balancing method requiring no external power converters, being implemented entirely through control logic and low-switch-count hardware. Unlike conventional active or passive balancing approaches, the proposed approach allows the dynamic series–parallel reconfiguration of lithium-ion cells based on SoC deviation, fault conditions, and power demand, without relying on external DC-DC converters, passive elements, or high-frequency switching.

Through MATLAB/Simulink simulation using both ideal and realistic Panasonic NCR18650PF cell models, the proposed model demonstrated robust balancing performance under resting, charging, and discharging conditions. The recombination logic consistently achieved SoC convergence within the 1% threshold across all scenarios. The all-cells-at-once strategy outperformed the traditional one-cell-at-a-time approach, reducing balancing time by up to 81% in some cases. Simulation results also confirmed negligible energy loss during resting-state balancing and validated the model’s flexibility across varying SoC deviations and cell behaviors. Beyond hardware simplicity, the proposed approach demonstrates measurable improvements in balancing speed, energy retention, and scalability. These benefits were consistently observed under multiple operating conditions and test scenarios, confirming that the design addresses performance, not just minimalism. Its full-cell recombination control enables fast convergence without introducing switching overhead or energy inefficiency, which marks a clear advancement over existing converter-heavy balancing techniques.

This study focused on cell-balancing dynamics using simulated models and did not include thermal, aging, or safety modeling. The fixed operating temperature of 25 °C does not reflect the dynamic thermal environment of real battery packs. Additionally, the Panasonic NCR18650PF validation was conducted in a virtual environment, and real hardware performance may vary due to parasitic elements, wiring losses, and latency in control signal propagation. This study was conducted using identical lithium-ion cells to validate the baseline performance of the proposed strategy under idealized and controlled conditions.

Although the proposed strategy aims to minimize SoC disparity within each recombination group, small voltage differences may still lead to charge redistribution when cells are connected in parallel. While this effect is short-lived and limited in current under controlled switching, its potential impact on long-term cycle life in aged or mismatched cells requires further investigation. This remains a subject for future experimental validation and lifetime modeling.

Future work will explore the integration of this control strategy into embedded BMS hardware platforms, including microcontroller-based implementations and real-world switching. Additional studies will evaluate the algorithm’s performance under thermal stress, aging degradation, and dynamic load profiles. Hardware-in-the-loop (HIL) simulation and scaled lab prototypes are also planned to bridge the gap between simulation and physical deployment. Future work is also planned to extend the framework to accommodate non-identical cells with varying capacities, degradation levels, and open-circuit voltage profiles, to assess the robustness of the recombination control logic in battery reuse and second-life applications.

The outcomes of this research suggest significant potential for real-world application in electric vehicle battery management systems, particularly where fast and energy-efficient balancing is critical. The proposed strategy’s ability to reduce balancing time by up to 81% with minimal hardware requirements makes it suitable for embedded implementation in both new and repurposed battery packs. Its scalability and control-based design also allow flexible integration into modular BMS platforms. As EV adoption and battery reuse initiatives expand globally, such adaptive and efficient balancing solutions can contribute meaningfully to improving battery lifespan, system safety, and energy utilization.