Analysis of Liquid Cooling Performance of Honeycomb-Structured Automotive Power Batteries and Research on Machine Learning Algorithm Predictions

Abstract

To address the thermal management challenges of electric vehicle power batteries under complex operating conditions, this study proposes a biomimetic honeycomb-shaped liquid cooling plate and conducts a systematic analysis of its cooling performance along with machine learning-based prediction for CTP lithium iron phosphate battery packs. A fluid–solid coupling numerical model was developed using ANSYS Fluent, employing the control variable method to investigate the effects of coolant flow rate (0.2–4.2 m/s), coolant inlet temperature (5–32 °C), ambient temperature (15–39 °C), and battery heating power (1000–5500 W/m³) on the maximum battery temperature. Simulation results demonstrate that the honeycomb structure leverages its hexagonal channel geometry and large specific surface area to achieve rapid and uniform heat transfer, with no localized hot spots observed across all operating conditions. The maximum battery temperature exhibits a marginal decreasing trend as coolant flow rate increases, with 1.4 m/s approaching the optimal flow rate; it rises approximately linearly with elevated inlet temperature, ambient temperature, and heating power—each 3 °C increase in inlet or ambient temperature raises the maximum temperature by approximately 1.98 °C and 3 °C, respectively, while a 500 W/m³ increase in heating power corresponds to an approximately 2.8 °C rise. Under standard conditions (heating power: 3000 W/m³; inlet temperature ≤23 °C; ambient temperature ≤27 °C), the maximum battery temperature remains below 45 °C; high-heating (≥3500 W/m³) or high-temperature (≥30 °C) scenarios require coordinated control strategies. Furthermore, based on simulation data, seven machine learning models—BPNN, GA-BP, PSO-BP, SVM, RBFNN, RF, and LSTM—were constructed and evaluated for their performance in predicting the maximum temperature of battery packs. The results showed that the LSTM model achieved the highest prediction accuracy on the validation set, with RMSE, MAE, MAPE, and R² values of 0.8068, 0.6891, 1.5653%, and 0.9865, respectively, while models such as SVM and RBFNN exhibited severe overfitting. This study validated the engineering effectiveness of the honeycomb structure liquid cooling plate and identified LSTM as the optimal model for predicting battery pack maximum temperature, providing a theoretical foundation and data support for the structural design and intelligent control of power battery thermal management systems.

1. Introduction

In recent years, the global new energy vehicle industry has developed rapidly. Lithium-ion power batteries, featuring high energy density, long cycle life, and low self-discharge rate, have become the mainstream energy source for electric vehicles [1,2,3]. However, during charging and discharging, these batteries generate significant heat due to electrochemical polarization and Ohmic losses; inadequate heat dissipation can cause sharp temperature rises, accelerating capacity degradation, inducing lithium deposition, and even triggering thermal runaway [4,5]. Studies show that lithium iron phosphate batteries have an optimal operating temperature range of 25–45 °C, with the battery module temperature difference requiring strict control within 5 °C [6]. Thus, developing an efficient, uniform liquid cooling system and achieving precise battery temperature prediction are core technical challenges for ensuring the safe and reliable operation of power batteries.

Currently, liquid-cooled battery thermal management systems have become the mainstream technical solution due to their high heat transfer coefficient [7]. The flow channel topology of liquid cooling plates directly determines the distribution of coolant flow and heat transfer efficiency. Although traditional serpentine, parallel, and U-shaped channels feature simple structures, they suffer from two inherent drawbacks: poor temperature uniformity (local temperature differences often exceeding 5–8 °C, accelerating edge cell aging) and significantly increased pump power losses (particularly pronounced in high-speed flow regions, severely limiting vehicle range) [8,9]. To overcome this bottleneck, researchers worldwide have optimized flow channel geometries: Fan et al. [10] proposed a ribbed serpentine liquid cooling plate, finding that internal ribs and cross-flow improved battery temperature uniformity and identifying an optimal structure balancing heat dissipation and energy consumption; Liu et al. [11] designed a dual-layer counter-flow liquid cooling plate that maintained module temperature differences below 4.2 °C and reduced pressure drop by -22%; Yu [12] verified complex channel configurations’ effectiveness in suppressing temperature rise via comparisons with single-channel designs; Fu et al. [13] studied a honeycomb-shaped liquid cooling plate, revealing counter-flow between adjacent plates significantly enhanced temperature uniformity; Bidwaik et al. [14] developed a novel microchannel system enabling efficient heat dissipation at low flow rates with good temperature uniformity and extremely low pump consumption; Khan et al. [15] optimized a side channel for a 75 Ah prismatic battery using a dynamic ECM model, determining Z-flow with base cooling as optimal (35–40 K channel cooling, 10 mW pump consumption); Kwon et al. [16] proposed a milling-constrained topology optimization method for U-shaped/Z-shaped battery thermal management systems, improving flow/temperature uniformity while reducing weight and power consumption, and identifying the optimal Reynolds number; Fan et al. [17] employed a multi-objective genetic algorithm to optimize parallel microchannel distribution parameters, achieving a Pareto-optimal balance between thermal resistance and pressure drop; Liu et al. [18] designed a double-sided, mirror Y-shaped dendritic channel cold plate with arc-shaped protrusions, demonstrating higher thermal efficiency and reduced pressure drop, which—when combined with nanofluids and insulation layers—effectively controls temperature and temperature differences.

Inspired by nature’s efficient mass/heat transfer biological structures (e.g., leaf veins, blood vessels, honeycombs)—evolved for minimal resistance and optimal flow distribution—bionic design has driven liquid cooling channel innovation. Yang et al. [19] developed a spider-web-inspired bionic mesh channel (4.3 °C module temperature difference at 5 m/s); Xia et al. [20] created a tree root-derived dendritic fractal channel, highlighting branching level’s impact on flow uniformity; Mohammadian and Yan et al. [21] studied a leaf vein-bionic cooling plate, noting secondary vein distribution angle’s influence on local Nusselt numbers. However, most bionic designs only achieve geometric “similarity” and fail to integrate biological mass transfer physics into channel cross-sectional/branching parameter optimization. Biomedical fluid dynamics shows laminar flow branched channels require specific geometric optimization to minimize resistance and maximize distribution efficiency, but existing fractal channels neglect branch dimension–flow resistance quantitative relationships [22], leading to increased turbulent dissipation, nonlinear pressure drop (Re > 2300), and excessive pump losses. Du et al. [23] confirmed unoptimized branch dimensions cause flow separation/vortices at nodes, reducing local heat transfer coefficients; Zhao et al. [24] emphasized the need to shift from “morphological biomimicry” to “physical biomimicry”—translating biological energy minimization principles into quantitative channel design criteria. Dai et al. [25] proposed a multi-parallel serpentine (MPS) channel for electric commercial vehicle battery packs, achieving excellent cooling efficiency via CFD optimization and experiments. Overall, leveraging channel topology intrinsic symmetry to design structures with uniform flow distribution and low resistance is a key breakthrough for current technical bottlenecks.

Accurate dynamic temperature prediction and proactive thermal regulation are essential to guarantee battery system safety [26]. Although physical-model-based numerical simulations like finite element and finite volume methods can accurately characterize heat transfer, their heavy computation, high parameter dependence, and long calculation time fail to satisfy vehicle real-time thermal management demands [27].

Data-driven intelligent algorithms have been widely applied in battery temperature prediction for superior nonlinear fitting ability. Support vector machines perform well in small-sample regression; Zuo et al. [28] optimized LSSVM via FGRA for battery SOH estimation with error below 0.02; Azim and Razzaghi [29] adopted SVM for electric bus battery temperature prediction with satisfactory practical effects. Heydarian and Abdollahi [30] combined electro-thermal coupling models and improved SAC algorithm to realize coordinated optimization of battery fast charging and equalization. Wu et al. [31] integrated graph convolutional networks with impedance data for battery temperature field reconstruction with 1.2 °C average absolute error, yet it requires costly high-precision sensors.

Classic BPNN features simple structure but suffers from slow convergence, local optimum, and initial weight sensitivity. Wang et al. [32] established WOA-RBFNN for SOH estimation with error within 1%; Tang et al. [33] optimized BPNN parameters via genetic algorithm to boost prediction precision. Nevertheless, such optimization easily causes overfitting, leading to poorer generalization under insufficient samples and incomplete working conditions [34].

Equipped with gated structures, LSTM excels at mining time-series correlation and shows great superiority in thermal trend forecasting. Qi et al. [35] constructed EMD-LSTM-AM for battery temperature prediction with maximum error of 0.31 °C, suitable for thermal management and thermal runaway early warning. Currently, LSTM research in battery liquid cooling systems is still immature, and most studies regard it as a pure black-box model, lacking deep fusion between temporal features and physical parameters such as coolant flow velocity and inlet temperature [36].

Based on the above analysis, current power battery liquid cooling systems still have prominent defects in structural design and temperature prediction:

(1)

Deficiencies in bionic flow channel design: most studies only imitate biological geometric structures without introducing mass transfer optimization physical laws into channel design, which easily causes uneven coolant distribution and poor local heat dissipation effect.

(2)

Incomplete multi-physics coupling mechanism: electrochemical heat generation and fluid flow are modeled independently, ignoring the feedback effect of temperature on battery internal resistance, and cannot accurately predict nonlinear characteristics such as thermal saturation.

(3)

Contradiction between heat dissipation performance and energy consumption. Single-objective optimization fails to balance temperature uniformity and pump power loss, resulting in increased energy consumption with limited cooling improvement effect.

(4)

Imperfect data-driven prediction system. Existing prediction models suffer from overfitting, insufficient physical mechanism fusion, incomplete working condition samples, and lack of horizontal algorithm comparison, leading to poor generalization and low prediction accuracy under extreme conditions.

To address the above bottlenecks, this study explores bionic channel design and intelligent temperature prediction. A three-layer cooling channel is proposed based on the high space utilization, symmetrical structure, and balanced flow characteristics of hexagonal honeycomb structures, which eliminates flow dead zones and local overheating. A fluid–solid coupled thermal model is established in ANSYS 2022 Fluent to analyze the effects of flow velocity, inlet temperature, ambient temperature, and heat generation on battery temperature, and summarize its variation laws. The energy consumption of battery thermal management is significantly affected by high flow rates and dynamic operating conditions. By combining the LSTM prediction model with an adaptive pump control strategy, pumping power consumption can be further reduced while maintaining a safe temperature range, so as to improve the long-term energy efficiency of vehicle thermal management systems [37,38]. Seven mainstream prediction models are compared on multi-condition simulation data to distinguish their fitting accuracy and generalization performance and analyze the risk of overfitting. Integrating bionic structure design, multi-condition simulation and data-driven algorithm verification, this study resolves the contradictions between cooling uniformity and energy consumption, as well as between prediction accuracy and generalization ability, providing theoretical and engineering references for efficient and safe battery thermal management.

2. Battery Pack Model and Liquid Cooling Plate Structure Design

2.1. Basic Parameters of the Battery Pack for the Battery Liquid Cooling System

The liquid cooling system designed in this study is tailored for a 400 V voltage platform and an 80 kWh rated energy CTP (CellToPack) power battery pack used in a mid-size pure electric sedan. This battery pack employs a module-free, integrated architecture, eliminating the traditional layered module housings and redundant structural components of conventional power batteries. It utilizes highly consistent square lithium iron phosphate (LFP) cells as the smallest energy storage units. The overall structure adheres to the chassis-embedded flat layout design principle, balancing vehicle assembly accessibility, space utilization efficiency, and compatibility with the thermal management system. Both the core structure and electrical parameters are precisely calibrated for optimal liquid cooling efficiency and uniform temperature control, providing comprehensive and standardized baseline parameters for subsequent liquid cooling flow channel topology optimization, three-dimensional thermal simulation modeling, and thermal performance testing and validation.

The battery pack employs automotive-grade, mass-produced 120 Ah square aluminum case lithium iron phosphate power cells. These cells offer core advantages including excellent thermal stability, long cycle life, and controllable cost. Each cell has a nominal voltage of 3.2 V and standard dimensions of 25 mm (thickness) × 148 mm (width) × 100 mm (height), arranged in a vertical upright configuration. The cell bottom achieves full contact with the integrated liquid cooling plate via high-thermal-conductivity insulating gel, minimizing interfacial thermal resistance. Leveraging the broad temperature adaptability of lithium iron phosphate materials, the system effectively expands the temperature control tolerance range of the liquid cooling system, meeting cooling requirements under various operating conditions such as normal driving, high-speed acceleration, and DC fast charging.

To strictly align the vehicle’s rated voltage platform with its rated energy design specifications, the battery pack employs a standardized series-parallel configuration of 125 series and two parallel strings. Ten cells are arranged in a single row along the pack’s length, resulting in 25 rows totaling 250 cells. Theoretical calculations indicate a total rated voltage of 400 V and a total rated capacity of 240 Ah. The optimized symmetrical series-parallel configuration achieves zero deviation between voltage and capacity parameters. Additionally, the standardized cell arrangement effectively mitigates localized thermal accumulation, simplifies the design of uniform temperature control across the liquid cooling system, and ensures balanced temperature distribution throughout the battery pack under all operating conditions.

The battery cells are arranged in a matrix symmetric configuration along the length of the battery pack. Two parallel cells within the same series branch are positioned adjacent to each other, with a uniform 2.5 mm gap maintained between individual cells. This gap is filled with thermally conductive insulating gel, providing triple functionality: efficient heat transfer, electrical insulation, and mechanical cushioning. The electrical interconnections employ a thick, nickel-plated pure copper busbar manufactured via laser fully automated welding. Parallel branches enable direct shorting between cells of the same polarity, while series branches facilitate polarity reversal connections through bridging busbars. The contact resistance of the welded joints is controlled below 0.1 mΩ, significantly reducing Joule heat loss at connection points and minimizing additional thermal load on the liquid cooling system.

The battery pack features an ultra-thin, flat rectangular structure that fits perfectly into the central mounting space of a mid-size pure electric sedan chassis. With a total mass of approximately 552 kg and a system-level energy density of no less than 145 Wh/kg, the design ensures the maximum temperature variation across the cell stack remains within 5 °C. Key performance parameters are detailed in

Table 1. Core structure and electrical performance parameters of CTP battery pack.

Parameter Category	Core Performance Metrics	Parameter Declaration
Battery pack type	CTP modular lithium iron phosphate battery pack	400 V platform, with the sedan chassis-embedded
Cell specifications	120 Ah square LFP cell, 3.2 V per cell	Dimensions: 25 mm × 148 mm × 100 mm
Serial–parallel configuration	125S2P, totaling 250 sections	Rated voltage: 400 V; rated capacity: 240 Ah
Core performance	Rated energy: 80 kWh; energy density ≥145 Wh/kg	The temperature difference between cell electrodes is ≤5 °C, with bottom liquid cooling compatibility.

, with all structural and performance specifications meeting the high-efficiency heat exchange requirements of the liquid cooling system, as well as compliance with vehicle assembly and safety protection standards for passenger car power batteries.

2.2. Battery Pack Model

Based on the parameters of the CTP non-modular lithium iron phosphate battery pack, a simplified three-dimensional geometric model was constructed using SolidWorks 2024. To balance thermal simulation accuracy with computational efficiency, the model employs the “equivalent volume method” to simplify the cells into homogeneous rectangular prism units arranged in a “25 rows × 10 columns” matrix, precisely preserving their external dimensions, relative positions, and thermal property parameters (e.g., specific heat capacity, thermal conductivity, heat generation rate). Non-core thermal components such as battery pack busbars, reinforcing ribs, weight-reduction holes, and temperature sensors were appropriately omitted. This simplification strategy focuses on the primary heat transfer pathways, minimizing model geometric complexity while ensuring the integrity of critical thermal boundaries (cell liquid cooling plate contact area ≥98%). This approach significantly improves mesh quality and solution efficiency without substantially compromising temperature field simulation accuracy.

2.3. Design of Serpentine Liquid Cooling Plate

Inspired by the hexagonal structure of natural honeycombs, this study develops a bionic honeycomb liquid cooling plate. From the perspectives of geometric topology, structural mechanics, and fluid heat transfer, regular hexagons possess prominent comprehensive advantages, featuring high space-filling efficiency, stable mechanical bearing capacity, and uniform fluid flow distribution. offering three core scientific advantages: first, its spatial utilization rate reaches 90.6%, significantly higher than that of squares (78.5%) and triangles (60.1%), enabling a denser fluid channel network within the limited bottom space of battery packs and enhancing heat exchange efficiency per unit area; second, the geometric symmetry of hexagons ensures uniform fluid pressure distribution within channels, effectively suppressing “velocity dead zones” caused by corner vortices in traditional rectangular channels and reducing flow field heterogeneity; third, according to the principle of minimum surface energy, hexagonal structures exhibit the smallest specific surface area for their volume, minimizing fluid resistance while their regular geometry guides coolant to form stable “diversion–convergence” pathways, preventing localized turbulence dissipation and fulfilling the liquid cooling system’s dual requirements of “high heat transfer efficiency and low energy consumption.” Additionally, from a bioheat transfer analogy perspective, the fractal characteristics of the honeycomb structure align perfectly with the matrix-like heat distribution in power batteries, achieving precise correspondence between “heat generation units and cooling units” and providing a structural foundation for addressing the poor temperature uniformity issue in conventional liquid cooling plates.

The liquid cooling plate’s overall dimensions precisely match the mounting groove at the bottom of the CTP battery pack, measuring 1500 mm (length) × 685 mm (width) × 10 mm (thickness). The flow channel of the liquid cooling plate adopts a square configuration of 20 mm × 20 mm with a wall thickness of 2 mm, and the liquid cooling plate is fabricated from 6063 aluminum alloy. It features a three-tier flow channel architecture comprising a main channel, sub-channels, and honeycomb cells. The structural design’s scientific rigor is demonstrated in three aspects: first, the alignment between the flow channel topology and cell arrangement—the main channel (with a cross-section of 20 mm × 10 mm) extends along the battery pack’s length, while the 12 sub-channels (six on each side, spaced 125 mm apart) are perfectly aligned with the cell rows, ensuring precise correspondence between cooling and heating zones and eliminating temperature gradients caused by traditional “cross-row cooling” in conventional flow channels. Figure 1a physical reference diagram of honeycomb structure (sourced from public network general materials, only for structural morphology reference).

3. CFD Analysis of Liquid Cooling for Battery Pack

3.1. Simulation Software Selection

To accurately investigate the flow heat transfer coupling characteristics of the honeycomb structure liquid cooling system and the temperature field distribution patterns of the battery pack, this study employed the Fluent module in ANSYS 2022R1 as the primary numerical simulation tool, leveraging its technical advantages and engineering applicability in multi-physics coupling simulations. In terms of numerical solvability, ANSYS Fluent’s Segregated Solver efficiently handles the coupling of laminar/transitional flow with solid domain heat conduction in liquid cooling systems. For the diverse operating conditions—including coolant flow rates (Re = 100–3000) and battery heat generation (3000–6000 W/m³)—the software’s standard k-ε turbulence model (with default wall functions) and solid domain heat conduction equations accurately capture secondary flow effects and temperature gradient variations within the flow channels. The discretization using the second-order upwind scheme achieves an optimal balance between computational accuracy and numerical stability. From the perspective of engineering compatibility, ANSYS Fluent supports seamless integration with SolidWorks geometric models, which preserves the hexagonal topology of the honeycomb channels and the matrix arrangement of the battery cells, thereby eliminating simulation deviations caused by the loss of geometric information.

3.2. Grid Division

The Mesh module on the ANSYS Workbench platform performs meshing for the coupled battery pack–liquid cooling plate model, adhering to the principle of “partition-based division and precise refinement” to balance computational accuracy in critical regions with overall efficiency. The model is first divided into three subdomains based on its structural characteristics: the battery cell domain, the honeycomb liquid cooling plate fluid domain, and the battery pack housing and busbar solid domain. Each subdomain employs a structured hexahedral mesh—a mesh type characterized by well-defined node connection rules and high element quality (orthogonal quality > 0.8), which reduces numerical dispersion errors and improves convergence speed by over 40% compared to unstructured meshes [39].

Differentiated mesh refinement is applied based on the specific grid accuracy requirements of each subdomain: The fluid domain of the cellular liquid-cooled plate, serving as the core heat exchange zone, employs boundary layer mesh refinement on the flow channel walls to accurately capture velocity and temperature gradients near these surfaces. The mesh element size is maintained between 0.5 and 3 mm, with a total of approximately 320,000 elements. Mesh quality validation results indicate an average orthogonal quality of 0.89, a minimum orthogonal quality of 0.72, and a maximum warpage angle < 15°, all meeting ANSYS Fluent’s mesh quality requirements (orthogonal quality > 0.6, warpage angle < 20°).

3.3. Determination of Physical Property Parameters of Lithium-Ion Batteries

3.3.1. Battery Heat Generation Rate

The heat generation characteristics of lithium-ion batteries serve as the core input for liquid cooling system simulations, with their accuracy directly determining the reliability of temperature field simulation results. This study focuses on 120 Ah square lithium iron phosphate (LFP) cells and employs the simplified heat generation model proposed by Bemadi et al. [40], which integrates both Ohmic losses and electrochemical polarization heat generation while eliminating complex entropy heat terms (with an error margin within 5%), making it more suitable for engineering simulations. The volumetric heat generation rate is expressed in Equation (1):

$$q={\displaystyle \frac{I}{V_b}}\left(I^2{R}_j+IT{\displaystyle \frac{\partial U}{\partial T}}\right)$$

(1)

In the formula: $U$—open circuit voltage, V; $I$—charge–discharge current, A; $V_b$—battery volume, $m^3$; $T$—temperature, °C; $R_j$—battery internal resistance, $\mathsf{\Omega }$.

This study integrated the battery heat generation mathematical model with the manufacturer-provided heat generation power meter. Through theoretical calculations, it determined the heat generation value of a 120 Ah square lithium iron phosphate battery under typical discharge conditions. The discharge rate refers to the ratio of the discharge current to the battery’s rated capacity; for example, at a 1 C rate, the battery discharges at 120 A and completes discharge in one hour, while a 0.5 C rate corresponds to 60 A, requiring two hours to discharge the full capacity. During the calculation, the heat generation model was first applied using discharge currents at different rates and measured internal resistance parameters to obtain the theoretical heat generation power. This value was then compared with the calibration values for the same conditions from the manufacturer’s heat generation power meter, showing a relative error of less than 2%, thereby verifying the reliability of the data.

The heat generation values of the battery cells during discharge at 0.5 C and 1 C rates were ultimately determined, with specific figures shown in

Table 2. Heat generation power of the battery at different charging rates.

Magnification	0.5 C	1 C
Heat generating power W/m3	2600	6000
Time/s	7200	5143

, providing precise thermal boundary conditions for the liquid cooling system simulation.

3.3.2. Calculation of the Specific Heat Capacity of Lithium Batteries

The specific heat capacity of lithium batteries is a critical parameter characterizing the cell’s heat absorption capability under temperature variations, directly influencing the accuracy of heat transfer rate calculations in temperature field simulations. This study employs the “component weighting method” to determine the specific heat capacity of a 120 Ah square lithium iron phosphate (LFP) cell. The method utilizes the mass ratios and specific heat capacities of each cell component, calculating the overall specific heat capacity through weighted summation as shown in Equation (2):

$$c_p={\displaystyle \frac{1}{m}}\sum c_i{m}_i$$

(2)

In the formula: m—quality of single battery, kg;

$c_i$—specific heat capacity of a material, J/(kg·K);

$m_i$—quality of corresponding materials, kg.

The square lithium iron phosphate battery consists of five core components: the cathode (LFP), anode (graphite), electrolyte, separator, and aluminum casing. The mass ratios and specific heat capacities of each component were determined through disassembly experiments and material analysis, with the specific parameters shown in Table 2. Substituting the data from the table into Equation (1) yields a specific heat capacity of 1020 J/(kg·°C) for this cell, which was adopted as the simulation input parameter.

3.3.3. Calculation of Lithium-Ion Battery Density

The density of lithium batteries is calculated as the ratio of their mass to volume, using the following formula:

$$\rho ={\displaystyle \frac{m}{v}}$$

(3)

In the formula: m—quality of single battery, kg;

$v$—volume of single cell, $m^3$.

By substituting the mass and volume of a 120 Ah square lithium iron phosphate battery cell into Equation (3), the final cell density is calculated as 2240 kg/m³. This parameter serves as the fundamental physical input for mass conservation and heat conduction calculations in subsequent thermal simulations.

3.3.4. Calculation of Thermal Conductivity of Lithium-Ion Batteries

Lithium-ion batteries feature an anisotropic layered composite structure, exhibiting significant variations in thermal conductivity across different directions, which directly influences the heat transfer pathways and temperature distribution within the cell. Based on measured data provided by battery manufacturers and considering the structural characteristics of square cells, this study determines the thermal conductivity coefficients for the three orthogonal directions of the cell as follows: ${\lambda }_x$ = 1.2, ${\lambda }_y$ = 15.1, ${\lambda }_z$ = 15.1.

3.4. Theoretical Model of CFD Analysis

The core principle of computational fluid dynamics (CFD) is to discretize continuous physical quantities in spatial and temporal dimensions into numerical variables at finite discrete nodes. In this study, the coolant flow velocity ranges from 0.2 to 4.2 m/s. A laminar flow model is adopted in this paper to accurately characterize the flow and heat transfer characteristics. For the flow characteristics of the honeycomb structure liquid cooling system in this study (coolant flow velocity: 0.2–4.2 m/s; corresponding Reynolds numbers Re = 100–3000; covering both laminar and transitional flow regimes), a turbulence model with strong adaptability and high computational efficiency must be selected. The standard k-e turbulence model has become the standard model for flow heat transfer coupling simulations in liquid cooling systems due to its advantages of rapid convergence, broad applicability, and manageable computational costs. By solving the transport equations for turbulent kinetic energy (k) and turbulent energy dissipation rate (e), this model quantifies turbulent pulsation effects in the flow field and accurately characterizes the coolant flow behavior within the honeycomb channels. The core transport equations are presented in Equations (4) and (5), respectively:

$${\displaystyle \frac{\partial (\rho k)}{\partial t}}+{\displaystyle \frac{\partial (\rho k{u}_i)}{\partial x_i}}={\displaystyle \frac{\partial }{\partial x_j}}[(\mu +{\displaystyle \frac{{\mu }_i}{{\sigma }_k}}){\displaystyle \frac{\partial k}{\partial x_j}}]+G_K-\rho \epsilon +S_k$$

(4)

$${\displaystyle \frac{\partial (\rho \epsilon )}{\partial t}}+{\displaystyle \frac{\partial (\rho \epsilon u_i)}{\partial x_i}}={\displaystyle \frac{\partial }{\partial x_j}}[(\mu +{\displaystyle \frac{{\mu }_i}{{\sigma }_{\epsilon }}}){\displaystyle \frac{\partial \epsilon }{\partial x_j}}]+G_{\sigma 1}{G}_K-C_{\sigma 2}\rho {\displaystyle \frac{{\epsilon }^2}{K}}+S_{\epsilon }$$

(5)

In the formula: $G_k$—turbulent kinetic energy generated by the average velocity gradient.

The calculation formula of $G_k$ is as follows:

$$G_k={\mu }_t({\displaystyle \frac{\partial u_i}{\partial x_j}}+{\displaystyle \frac{\partial u_j}{\partial x_i}}){\displaystyle \frac{\partial u_i}{\partial x_j}}$$

(6)

The turbulent viscosity is a function of ${\mu }_t$, $k$, and $\epsilon$, and the calculation formula is:

$${\mu }_t=\rho C_u{\displaystyle \frac{{\epsilon }^2}{k}}$$

(7)

In the formula: ${\sigma }_k$—Prandtl coefficient corresponding to turbulent kinetic energy;

${\sigma }_{\epsilon }$—Prandtl coefficient corresponding to turbulent energy dissipation rate.

In this paper, the parameters of Launder and Spalding [41] and other theoretical calculations and experimental verification are selected: $C_u$ = 0.09, ${\sigma }_k$ = 1.22, $C_{\sigma 1}$ = 1.44, $C_{\sigma 2}$ = 1.92.

The calculated Re values across all simulations ranged 200–1600, consistently below the transition threshold (2300), validating laminar dominance.

3.5. Boundary Conditions and Solution Parameter Settings

To accurately simulate the fluid–solid coupled heat transfer process between the cellular structure liquid cooling system and the battery pack, the model’s boundary conditions and solution parameters were systematically configured based on actual operating conditions and simulation requirements, ensuring both physical accuracy and computational stability of the simulation results.

During the physical model activation phase, the energy equation is first activated to quantitatively analyze heat transfer between the fluid (coolant) and solids (cell, liquid cooling plate, housing). The standard k-ε turbulence model and incompressible fluid model are simultaneously activated to accommodate the coolant’s flow characteristics. For material property definitions, custom configurations have been implemented in the ANSYS Fluent material library, including the “battery” (cell with a specific heat capacity of 1020 J/(kg·°C), density of 2240 kg/m³, and anisotropic thermal conductivity), ensuring accurate characterization of thermodynamic behavior across all domains.

The boundary conditions are designed to focus on the core heat transfer and flow characteristics:

(1)

The heat source boundary employs a “constant heat generation rate,” uniformly distributing the cell’s heat generation power at the 0.5 C and 1 C rates specified in Section 3.3.1 across the cell domain to simulate heat generation characteristics under stable discharge conditions.

(2)

The coolant boundary is configured as “velocity inlet–pressure outlet,” with the inlet velocity ranging from 0.2 to 4.2 m/s (corresponding to different cooling intensities). The inlet temperature is adjusted according to the simulation conditions, while the outlet is set at atmospheric pressure (0 Pa gauge pressure) to ensure stable flow field.

(3)

For wall boundary treatment, the upper, lower, and side surfaces of the battery pack are defined as wall boundaries. Given that surface thermal radiation accounts for less than 5% of the total heat dissipation (at an ambient temperature of 25 °C), the radiation model is temporarily disabled. The wall surfaces are treated solely as natural convection boundaries, with a uniform convective heat transfer coefficient of 5 W/(m²·K) to account for convective heat exchange with the surrounding environment.

(4)

The fluid–solid interface (cell–liquid cooling plate, liquid cooling plate–shell) adheres to the no-slip boundary condition and employs a “coupled wall” approach to ensure efficient heat transfer between the fluid and solid, with interface thermal resistance neglected (corrected using thermal grease layer parameters).

The parameter configuration was optimized under the principle of “balancing accuracy and computational efficiency”. The time step was set to 0.01 s, with a total of 72,000 computational steps, corresponding to a simulation duration of 7200 s (2 h) to cover the stable discharge cycle of the battery pack. A maximum of 500 iterations per time step was specified to ensure adequate convergence of the flow and temperature fields. For numerical discretization, the momentum and energy equations were solved using the second-order upwind scheme.

During initialization, the initial temperature of the model was set to 27 °C (ambient temperature), and the initial fluid velocity was set to 0 m/s. The “hybrid initialization” function was adopted to rapidly establish the initial distributions of the flow and temperature fields. The convergence criteria were defined as follows: the residual of the energy equation was required to be below 1 × 10⁻⁶, while the residuals of all other governing equations were set below 1 × 10⁻⁴. Additionally, the maximum cell temperature and the temperature difference between the inlet and outlet of the liquid cooling plate were monitored as physical convergence indicators. The simulation was considered fully converged when the variations in these two parameters remained below 0.1 °C and 0.05 °C, respectively, over 500 consecutive time steps.

4. Design and Result Analysis of the Simulation Experiment for Liquid Cooling Effect of the Honeycomb Structure Liquid Cooling Plate

4.1. Simulation Experiment on the Performance of Honeycomb Structure Liquid Cooling Plates

Using the ANSYS Fluent platform, numerical simulations were conducted on the assembly model of a honeycomb structure liquid-cooling plate battery module. The parameters set included a coolant inlet velocity of 2.5 m/s, a battery heat generation power of 3000 W/m³, and an ambient temperature of 32 °C. The simulation compared the temperature field distributions under two operating conditions: coolant inlet temperatures of 20 °C and 27 °C. The simulation results are shown in Figure 2 and Figure 3.

4.2. Numerical Results and Discussion

Systematic analysis based on temperature field simulation results reveals that under both operating conditions, the battery pack temperature exhibits a gradient increase along the coolant flow direction: as coolant enters the honeycomb flow channels through the inlet, it absorbs heat generated by the cells, resulting in higher temperatures downstream compared to upstream regions—a pattern perfectly consistent with the visual characteristic shown in the cloud map: a gradual transition from blue (low-temperature zone) at the inlet to red (high-temperature zone) at the outlet. Notably, no pronounced localized dark red clusters are observed in the cloud map, indicating continuous and uniform temperature distribution across the honeycomb channels without the “edge hotspots” typically associated with conventional flow channels. This phenomenon stems from the hexagonal symmetrical structure of the honeycomb design—each cell receives uniform coolant flow, with over 95% of the channel walls in contact with the cells. The high specific surface area facilitates rapid heat transfer to the coolant, preventing localized temperature accumulation, which corroborates the cloud map’s observation of minimal temperature variations between adjacent cells and no significant temperature spikes.

Further analysis of the heat exchange details in the honeycomb flow channels reveals that the temperature map shows “temperatures in the channel-corresponding regions are significantly lower than those in the cell body,” while the channel wall temperatures are slightly higher than the internal coolant temperatures (the channel edges appear light red, while the center appears light blue). This characteristic directly reflects the fluid–solid coupled heat transfer process: heat generated by the cell is transferred to the channel walls via thermal conduction and then transferred to the coolant through convective heat transfer, with the walls serving as a thermal intermediary that creates a “cell-wall-c coolant” temperature gradient, demonstrating the honeycomb channels’ high-efficiency heat exchange capability. Comparing the temperature maps under two operating conditions shows that at an inlet temperature of 20 °C, the high-temperature region (red area) occupies a smaller proportion and is concentrated at the outlet edge; when the inlet temperature rises to 27 °C, the high-temperature region expands significantly and spreads toward the central area. This intuitively illustrates the impact of inlet temperature on cooling efficiency—lower-temperature coolant more effectively removes heat, suppressing the expansion of high-temperature zones, consistent with the observation that “the blue area is broader under low-inlet-temperature conditions” in the temperature map.

Based on the temperature quantification indicators and the extraction results from the cloud map data: under a 20 °C inlet condition, the maximum battery pack temperature reached 43.32 °C (at the outlet-side cells); and under a 27 °C inlet condition, the maximum temperature rose to 46.62 °C. The maximum temperatures under both conditions fall within the optimal operating range of 25–45 °C for lithium iron phosphate batteries, with the maximum temperature difference kept below 10 °C, meeting battery consistency requirements. Notably, the cloud map indicates that “temperature differences between cells covered by honeycomb flow channels are less than 3 °C,” significantly lower than those in non-flow channel areas (where the difference reaches 5 °C), further demonstrating that the honeycomb structure uniformly distributes coolant to minimize inter-cell temperature variations and prevent capacity degradation caused by temperature inconsistencies.

The comprehensive analysis of the temperature cloud map characteristics and quantitative data reveals that the thermal performance advantages of the honeycomb-structured liquid cooling plate manifest in two key aspects: first, the symmetrical layout of hexagonal flow channels and their large specific surface area enable rapid heat transfer and uniform distribution, as evidenced by the absence of localized hot spots and a gentle temperature gradient in the cloud map; second, its efficient flow field distribution ensures continuous heat exchange throughout the cooling liquid path, maintaining maximum temperatures within safe limits even at the high inlet temperature of 27 °C, thereby ensuring reliable thermal management for the CTP battery pack under all operating conditions.

5. Simulation Experiment and Result Analysis of Heat Transfer Characteristics of the Honeycomb Structure Liquid Cooling Plate System

5.1. Comparison of Performance of Honeycomb Structure Liquid Cooling Plates at Different Coolant Inlet Temperatures

5.1.1. Design of the Simulation Experiment Process

To systematically investigate the cooling performance of the honeycomb-structured liquid-cooling plate on CTP battery packs, this study employed the control variable method to design multi-scenario numerical simulations: with a fixed battery heating power of 3000 W/m³ and ambient temperature of 27 °C, the core variables included cooling liquid flow rate (0.2–4.2 m/s, step size 0.4 m/s) and inlet temperature (20 °C and 30 °C). A total of 22 simulation scenarios were established to focus on analyzing how these variables affect the maximum battery pack temperature.

5.1.2. Simulation Result Analysis

Figure 4 and Figure 5 show the temperature field maps of the battery pack under inlet temperatures of 20 °C and 30 °C, respectively, providing an intuitive visualization of the heat transfer efficiency and temperature distribution patterns of the honeycomb structure.

It can be seen from Figure 4 that the flow field and temperature field distribution differ significantly under different inlet flow velocities. At an inlet temperature of 20 °C, the temperature field increases gradually along the flow direction of the cooling liquid. The inlet side presents a blue low-temperature zone ranging from 35 °C to 38 °C, while the outlet side is a light red high-temperature zone of 42 °C to 43 °C, with no local deep red hot spots observed. When the flow velocity rises from 0.2 m/s to 1.4 m/s, the high-temperature area at the outlet shrinks obviously, and the maximum temperature drops from 43.25 °C to 42.37 °C, which verifies that the honeycomb flow channel can realize efficient heat transfer by evenly distributing cooling liquid. In addition, the decreasing range of the maximum temperature gradually becomes gentle with the further increase in flow velocity.

In this study, systematic simulations are carried out within the full flow velocity range of 0.2–4.2 m/s. For the convenience of intuitively showing the variation law, four representative working conditions are selected at equal intervals for comparative analysis in Figure 4.

As shown in Figure 5, the overall temperature rises significantly under the 30 °C inlet condition: at a flow rate of 0.2 m/s, the red high-temperature zone (49–50 °C) covers most of the battery pack; when the flow rate increases to 1.4 m/s, the high-temperature zone shrinks markedly, with the maximum temperature dropping to 48.98 °C—still higher than that observed at the same flow rate under the 20 °C inlet condition—indicating that inlet temperature exerts a far greater influence on cooling performance than flow rate.

5.1.3. Flow Rate–Maximum Temperature Variation Pattern

Based on 22 sets of simulation data, the “line graph showing the maximum battery pack temperature versus coolant flow rate” (Figure 6) was plotted, clearly illustrating the quantitative relationships among the variables.

Observation and analysis of Figure 6 reveal two patterns:

(1)

Diminishing marginal effect: At both inlet temperatures, the maximum battery pack temperature exhibits a “rapid decline followed by gradual stabilization” trend as flow velocity increases. At an inlet temperature of 20 °C, increasing the flow velocity from 0.2 m/s to 1.4 m/s reduces the maximum temperature by 0.29 °C (42.61 °C → 42.32 °C); further increase to 3.8 m/s results in only a 0.10 °C decrease (42.32 °C → 42.22 °C). At 30 °C inlet temperature, the corresponding temperature reductions are 0.28 °C (49.19 °C → 48.91 °C) and 0.07 °C (48.91 °C → 48.84 °C). This indicates that 1.4 m/s approaches the optimal flow velocity; further increases yield minimal temperature reduction benefits but elevate pump power losses.

(2)

Dominance of inlet temperature: At the same flow rate, the maximum temperature under a 30 °C inlet condition is approximately 6.6 °C higher than that under a 20 °C condition (e.g., 48.91 °C vs. 42.32 °C at a flow rate of 1.4 m/s). This indicates that reducing the inlet temperature is the primary means of controlling battery temperature, while flow rate serves only an auxiliary regulatory role.

In summary, under conditions of a 20 °C inlet temperature and a flow velocity of 1.4 m/s, the honeycomb structure liquid-cooled plate keeps the battery pack’s maximum temperature below 43 °C, fully meeting the optimal operating temperature range of 25–45 °C for lithium iron phosphate batteries while ensuring excellent temperature uniformity. This provides reliable technical support for comprehensive thermal management of CTP battery packs under all operating conditions.

5.2. Comparison of Performance of Honeycomb Structure Liquid Cooling Plates at Different Coolant Flow Rates

5.2.1. Simulation Experiment Process Design

To investigate the impact of the coupling effect between coolant inlet temperature and battery heating power on the cooling performance of the honeycomb structure liquid-cooling plate, this study designed simulation scenarios using the control variable method: with a fixed coolant flow rate of 3 m/s, ambient temperature of 27 °C, coolant inlet temperature (5–32 °C, step size 3 °C), and battery heating power (3000 W/m³, 6000 W/m³) as variables, a total of 20 simulation scenarios were established. This comprehensive analysis elucidates the influence pattern of the two-factor coupling on the maximum battery pack temperature, providing data support for optimizing thermal management strategies under extreme operating conditions.

5.2.2. Simulation Result Analysis

Figure 7 and Figure 8 show the temperature field maps of the battery pack corresponding to different coolant inlet temperatures under heating power conditions of 3000 W/m³ and 6000 W/m³, respectively, intuitively illustrating the coupled effects of inlet temperature and heating power on temperature distribution.

As shown in Figure 7, under the conventional heating condition of 3000 W/m³, the battery pack temperature exhibits a gradient increase along the coolant flow direction. The inlet side appears as a blue low-temperature zone, while the outlet side shows a red high-temperature zone, with no localized hot spots, demonstrating the uniform heat transfer advantage of the honeycomb structure. At an inlet temperature of 5 °C, the battery pack appears predominantly red-yellow with a maximum temperature of 33.39 °C. As the inlet temperature rises to 14 °C, 23 °C, and 32 °C, the overall temperature field increases progressively, and the red high-temperature region continues to expand, with maximum temperatures reaching 38.61 °C, 44.19 °C, and 50.23 °C respectively, clearly illustrating the temperature elevation effect caused by increased inlet temperature.

As shown in Figure 8, under the high heat generation condition of 6000 W/m³, the overall temperature field increases significantly: at an inlet temperature of 5 °C, the battery pack reaches a maximum temperature of 50.47 °C, with the red high-temperature zone covering most of the area; when the inlet temperature rises to 32 °C, the maximum temperature surges to 70.12 °C, and the battery pack appears deep red, far exceeding the safe operating temperature for lithium batteries. Under the same inlet temperature, the proportion of the high-temperature zone and the temperature gradient under the high heat generation condition are significantly higher than under conventional conditions, demonstrating the amplifying effect of heat generation power on cooling performance.

5.2.3. Variation Pattern of Coolant Inlet Temperature Versus Maximum Temperature

Based on simulation experimental data, the maximum temperatures of the battery pack under various operating conditions were recorded and plotted as the variation curve shown in Figure 9.

Observation and analysis of Figure 9 reveal:

(1)

Linear regulation pattern of inlet temperature

Under both heating power conditions, the maximum battery pack temperature exhibited an approximately linear increase with rising coolant inlet temperature (Figure 9). At a heating power of 3000 W/m³, the inlet temperature rose from 5 °C to 32 °C, and the maximum temperature correspondingly increased linearly from 33.39 °C to 50.23 °C—a temperature rise of approximately 1.98 °C per 3 °C increase. Under the higher heating condition of 6000 W/m³, the maximum temperature increased from 50.47 °C to 70.12 °C, with a battery temperature rise of about 2.18 °C per 3 °C increase in inlet temperature. This linear relationship demonstrated high statistical significance (R² > 0.99) across both power levels, indicating that the inlet temperature exerts a robust and predictable linear regulatory effect on the battery’s thermal state.

(2)

Amplification effect of thermal power

A comparison of the thermal response curves under two heating power conditions reveals that, at the same inlet temperature, the maximum temperature of the battery under the 6000 W/m³ condition is 16.98–19.89 °C higher than under the 3000 W/m³ condition. Notably, this temperature difference increases with rising inlet temperature: it is 17.08 °C at 5 °C and expands to 19.89 °C at 32 °C. This phenomenon demonstrates the significant amplifying effect of heating power on the temperature control efficacy—under high heat generation conditions, reducing inlet temperature brings more prominent temperature control gains, indicating that high-temperature-rise working conditions are more sensitive to coolant temperature. In the thermal management of new energy vehicles, priority shall be given to the impact of battery heat generation power on battery thermal performance.

(3)

Operating condition compatibility and implications for system thermal management

From an engineering application perspective, under conventional heating conditions of 3000 W/m³, maintaining the coolant inlet temperature below 23 °C ensures the battery pack’s maximum temperature remains below 45 °C, fully meeting the optimal operating temperature requirements for lithium iron phosphate batteries. When the inlet temperature rises to 32 °C, the maximum temperature reaches 50.23 °C, exceeding the safe operating range. In contrast, under high-heating conditions of 6000 W/m³, even with an inlet temperature as low as 5 °C, the battery’s maximum temperature still reaches 50.47 °C—significantly surpassing the 45 °C upper limit. This demonstrates that merely reducing the inlet temperature cannot mitigate thermal runaway risks during high-rate discharge or fast charging; it requires comprehensive, coordinated control measures such as increasing coolant flow rate and optimizing flow channel topology.

The coolant inlet temperature and battery heating power exhibit a significant coupled regulatory effect on the thermal state of the battery pack: the inlet temperature serves as the key variable determining the steady-state temperature level, while the heating power dictates the complexity level of the thermal management system. The honeycomb-shaped liquid-cooling plate demonstrates excellent temperature control performance under conventional heating conditions, but faces challenges under extreme high-heating scenarios, necessitating the implementation of a multi-parameter collaborative strategy. These findings provide a quantitative theoretical foundation and engineering guidance for designing thermal management systems for power batteries operating under all conditions, highlighting the potential of bionic flow channel structures in intelligent thermal regulation.

5.3. Comparison of the Performance of Honeycomb Structure Liquid Cooling Plates at Different Environmental Temperatures

5.3.1. Simulation Experiment Process Design

To investigate the influence of ambient temperature on the cooling performance of honeycomb-structured liquid-cooled plates, this study employed the control variable method to design multi-condition numerical simulations. With a fixed battery heating power of 3000 W/m³, two comparative conditions were established: one with a coolant flow rate of 3 m/s and the other with 1 m/s. Under each condition, the coolant inlet temperature varied synchronously with the ambient temperature (at 15 °C, 18 °C, 21 °C, 24 °C, 27 °C, 30 °C, 33 °C, 36 °C, and 39 °C), resulting in a total of 18 simulation scenarios. This comprehensive analysis examines the combined effects of ambient temperature and flow rate on the maximum battery pack temperature, providing data support for optimizing thermal management strategies in diverse climatic environments.

5.3.2. Simulation Result Analysis

Figure 10 and Figure 11 show the temperature field contour maps of the battery pack under flow velocities of 3 m/s and 1 m/s, respectively, corresponding to different ambient temperatures, intuitively illustrating how ambient temperature affects the temperature distribution.

As shown in Figure 10, under the high-flow-rate condition of 3 m/s, the battery pack temperature exhibits a gradient increase along the coolant flow direction: the inlet side shows a blue low-temperature zone, while the outlet side displays a red high-temperature zone with no localized hot spots, demonstrating the uniform heat transfer advantage of the honeycomb structure. At an ambient temperature of 15 °C, the maximum battery pack temperature is only 31.98 °C, appearing predominantly red-yellow. As the ambient temperature rises to 24 °C, 33 °C, and 39 °C, the overall temperature field increases progressively, with the red high-temperature zone continuously expanding; the maximum temperatures reach 40.98 °C, 49.98 °C, and 55.98 °C respectively, clearly illustrating the temperature elevation effect of rising ambient conditions on the battery.

As shown in Figure 11, under the low-flow-rate condition of 1 m/s, the overall temperature field is significantly higher than that under the 3 m/s condition at the same ambient temperature: at an ambient temperature of 15 °C, the maximum temperature reaches 32.08 °C, with a larger proportion of the red high-temperature zone; when the ambient temperature rises to 39 °C, the maximum temperature surges to 56.08 °C, and the battery pack appears entirely dark red, far exceeding the safe operating temperature range for lithium iron phosphate batteries. Under identical ambient temperatures, the proportion of high-temperature areas and the temperature gradient under the low-flow-rate condition are significantly greater than those under the high-flow-rate condition, demonstrating the auxiliary regulatory effect of flow rate on cooling performance.

5.3.3. Variation Patterns of Maximum Battery Pack Temperature Under Different Ambient Temperatures

Based on simulation data, the variation curves of the battery pack’s maximum temperature under different operating conditions are presented in Figure 12.

Observation and analysis of Figure 12 reveal the following:

(1)

The linear dominant effect of ambient temperature

Under constant battery heating power conditions (3000 W/m³), the maximum battery temperature exhibits a strictly linear positive correlation with the ambient temperature. At both coolant flow rates (3 m/s and 1 m/s), a 3 °C increase in ambient temperature corresponds to a simultaneous rise of approximately 3 °C in the maximum battery temperature, with a linear slope approaching 1; this linear relationship remains largely unaffected by flow rate variations. This phenomenon can be attributed to the synchronized setting of coolant inlet temperature and ambient temperature in the experimental design: as the ambient temperature rises, the coolant inlet temperature increases proportionally, maintaining a nearly constant heat transfer temperature difference between the coolant and the battery. Under these conditions, increasing the flow rate only modestly enhances the convective heat transfer coefficient and cannot offset the overall temperature rise caused by elevated ambient temperature. These results demonstrate that ambient temperature exerts a “rigid” regulatory effect on the battery’s thermal state, with its influence far exceeding that of secondary parameters such as flow rate.

(2)

Operating condition compatibility and safe operation limits

From an engineering application perspective, when the flow rate is 3 m/s and the ambient temperature is ≤27 °C, the maximum battery pack temperature can be maintained below 45 °C, fully satisfying the optimal operating temperature range of 25–45 °C for lithium batteries. Once the ambient temperature rises to 30 °C or higher, the maximum temperature exceeds the safety limit. Under a flow rate of 1 m/s, the maximum battery temperature at the same ambient temperature is slightly higher than that under the 3 m/s condition (by approximately 0.1–0.2 °C on average), and the temperature control disadvantage becomes more pronounced in high-temperature environments (≥30 °C); however, the overall behavior still follows the same linear pattern.

(3)

The advantage of temperature uniformity in honeycomb structures

Notably, under all ambient temperature and flow rate conditions, the battery pack exhibited no localized hot spots, demonstrating excellent temperature distribution uniformity. This confirms the stable heat transfer capability of the honeycomb structure liquid cooling plate across a wide ambient temperature range—the hexagonal flow channel array employs a flow self-balancing mechanism that effectively eliminates the common “edge flow velocity dead zones” and “heat transfer blind spots” found in traditional channels, ensuring consistent temperature distribution throughout all operating conditions. This feature is crucial for preventing temperature differentials from accelerating battery aging and enhancing the system’s cycle life.

Comprehensive analysis shows that ambient temperature is the major external factor affecting the heat dissipation performance of the honeycomb liquid cooling plate, while coolant flow rate only acts as a fine-tuning parameter with limited effect. Under normal ambient temperature (≤27 °C), the cooling plate can keep battery temperature within the optimal operating range independently. When the temperature rises to 30 °C and above, thermal management relying solely on coolant flow velocity fails to satisfy temperature control demands of power batteries, and multi-mode collaborative cooling strategies are required.

5.4. Comparison of Thermal Power Emission and Honeycomb Structure Liquid Cooling Plate Performance Across Different Battery Packs

5.4.1. Simulation Experiment Process Design

To investigate the influence of battery heating power on the cooling performance of the honeycomb structure liquid-cooling plate, this study employed the control variable method to design multi-condition numerical simulations. With a fixed coolant flow rate of 3 m/s and coolant inlet temperature of 25 °C, two comparative ambient temperature conditions were established: one at 30 °C and the other at 38 °C. Under each condition, the battery heating power was increased incrementally from 1000 W/m³ to 5500 W/m³ in 500 W/m³ steps, resulting in a total of 20 simulation conditions. This comprehensive analysis examines the impact of the coupling between heating power and ambient temperature on the maximum battery pack temperature, providing data support for optimizing thermal management strategies under various load conditions.

5.4.2. Simulation Result Analysis

Figure 13 and Figure 14 show the temperature field cloud diagrams of battery packs under operating conditions with ambient temperatures of 30 °C and 38 °C, respectively, visually illustrating how heating power affects temperature distribution.

As shown in Figure 13, at an ambient temperature of 30 °C, the battery pack temperature exhibits a gradient increase along the coolant flow direction, with the inlet side representing the blue low-temperature zone and the outlet side the red high-temperature zone, without any localized hot spots, demonstrating the uniform heat transfer advantage of the honeycomb structure. At a heating power of 1000 W/m³, the maximum temperature is only 33.29 °C, with the overall temperature distribution appearing red-yellow. When the heating power increases to 2500 W/m³, 4000 W/m³, and 5500 W/m³, the temperature field rises progressively, and the red high-temperature region continues to expand, with maximum temperatures reaching 41.78 °C, 50.26 °C, and 58.74 °C respectively, clearly illustrating how increased heating power elevates battery temperature.

As shown in Figure 14, under the high environmental temperature of 38 °C, the overall temperature field is significantly higher than that under the 30 °C condition at the same heating power: at a heating power of 1000 W/m³, the maximum temperature reaches 38.03 °C, with a larger proportion of the red high-temperature zone; when the heating power increases to 5500 W/m³, the maximum temperature surges to 62.96 °C, and the battery pack appears entirely dark red, far exceeding the safe operating temperature of lithium iron phosphate batteries. At the same heating power, the proportion of the high-temperature zone and the temperature gradient under the high-temperature condition are significantly greater than those under the low-temperature condition, demonstrating the amplifying effect of environmental temperature on cooling performance.

5.4.3. Heat Generation Power of Different Battery Packs—Relationship Between Maximum Battery Pack Temperature and Heat Generation Power

Based on the simulation data, the maximum temperature of the battery pack under various operating conditions was recorded and plotted as the variation curve shown in Figure 15.

Observation and analysis of Figure 15 reveal:

(1)

The linear dominant effect of heating power

Under constant coolant flow rate (3 m/s) and inlet temperature (25 °C), the maximum battery temperature exhibits a linear positive correlation with the battery heating power. At both ambient temperature conditions (30 °C and 38 °C), a 500 W/m³ increase in heating power corresponds to a simultaneous rise in battery temperature of approximately 2.8 °C, with a linear correlation coefficient R² > 0.999. The fundamental reason for this phenomenon is that the battery heating power serves as the sole direct source of internal heat generation within the system, and its linear increase directly translates into a temperature rise within the battery itself; whereas the ambient temperature acts merely as a constant external offset superimposed on the temperature response curve, without altering the sensitivity of the temperature rise to the heating power.

(2)

Operating Condition Compatibility and Safe Operation Limits

From an engineering application perspective, when the ambient temperature is 30 °C, maintaining a heating power of no more than 3000 W/m³ keeps the battery pack’s maximum temperature below 45 °C, fully complying with the optimal operating temperature range of 25–45 °C for lithium batteries. Once the heating power exceeds 3500 W/m³, the maximum temperature surpasses the safety limit. When the ambient temperature rises to 38 °C, the safe operating range narrows to a heating power of ≤2500 W/m³. This indicates that high-temperature environments significantly reduce the load capacity of liquid cooling systems; under high heating power conditions, complementary strategies such as increasing flow rate, optimizing flow channel design, or employing pre-cooling must be employed.

(3)

The load-adaptive advantage of the honeycomb structure

Notably, under all heating power levels and both ambient temperature conditions, the battery pack temperature exhibited no localized hot spots, maintaining a highly uniform distribution. This demonstrates the stable heat transfer capability of the honeycomb structure liquid cooling plate across a wide load range—its hexagonal channel array utilizes geometric symmetry and a self-balancing flow mechanism to effectively prevent the “local overheating zones” and “heat transfer saturation” commonly observed in traditional serpentine or rectangular channels at high power. This characteristic holds significant engineering value for mitigating thermal shock during transient high-power discharge scenarios in electric vehicles, such as rapid acceleration or fast charging.

Based on the above analysis, the battery heating power has been identified as the dominant load variable affecting the cooling performance of the cellular structure liquid cooling plate, while ambient temperature only provides a linear translational offset. Under conventional heating power (≤3000 W/m³) and ambient temperature (≤30 °C) conditions, the liquid cooling plate independently achieves efficient thermal management; however, under extreme conditions of high heating (≥3500 W/m³) or high temperatures (≥38 °C), its performance limitations become apparent, necessitating the implementation of a multi-parameter coordinated control strategy.

5.5. Comprehensive Analysis and Safe Operating Range Summary of Multi-Parameter Coupling Conditions

Based on the multi-condition simulation results of coolant flow velocity, inlet temperature, ambient temperature, and battery heat generation power, the influence weight of each parameter on the battery thermal state is clarified as follows: battery heat generation power > ambient temperature > coolant inlet temperature > coolant flow velocity. Heat generation power and ambient temperature are the core dominant parameters, which present rigid regulation characteristics on battery temperature rise. Coolant inlet temperature possesses stable linear temperature control capability, while flow velocity only acts as an auxiliary fine-tuning parameter. When the flow velocity exceeds 1.4 m/s, the heat transfer gradually reaches saturation with negligible temperature control improvement, resulting in an insignificant temperature difference between the 1 m/s and 3 m/s working conditions.

Combined with the optimal operating temperature range of 25–45°C for lithium iron phosphate batteries, the optimal safe operating conditions of the honeycomb liquid cooling system in this study are determined as follows: heat generation power ≤ 3000 W/m³, ambient temperature ≤ 27 °C, coolant inlet temperature ≤ 23 °C, and flow velocity of 1.4–3.0 m/s. This parameter matching scheme achieves a balance between heat dissipation performance and energy consumption economy. Safety boundary analysis indicates that battery temperature will exceed the safe operating range when the heat generation power surpasses 5500 W/m³ or the ambient temperature rises above 39 °C. Enhanced active cooling or reduced battery heat generation is therefore required to ensure safe temperature control.

6. Prediction of the Maximum Temperature of Battery Packs Based on Machine Learning

6.1. Data Acquisition and Preprocessing

Power batteries serve as the core energy units in electric vehicles, with their operating temperature directly determining battery safety, lifespan, charge/discharge performance, and overall vehicle reliability. Battery packs are highly susceptible to localized overheating, uneven temperature distribution, and increased thermal runaway risks under conditions such as fast charging, high-rate discharge, continuous hill climbing, and high-temperature environments. The maximum battery pack temperature stands as one of the most critical and intuitive indicators of thermal risk. Consequently, research on predicting maximum battery pack temperatures holds significant theoretical and practical value. Four key factors influencing maximum battery pack temperature—including coolant flow rate, coolant inlet temperature, ambient temperature, and battery heating power—are identified. Using simulation software and the variable-control method, multiple parameter scenarios were established to obtain maximum temperature data, enabling analysis of how different parameters affect temperature fluctuations. The study yielded 97 experimental datasets through ANSYS finite element simulations.

To eliminate dimensional and magnitude differences between input and output variables, all sample data undergo normalization preprocessing, uniformly mapping the feature values to the [0, 1] interval using the following normalization formula:

$$\mathrm{z} = {\displaystyle \frac{\mathrm{y} - {\mathrm{y}}_{\min }}{{\mathrm{y}}_{\max }- {\mathrm{y}}_{\min }}}$$

(8)

where z denotes the new value obtained by normalizing the original data, with its value range strictly constrained to the interval [0, 1]; y represents the original sample data to be normalized; $y_{\mathit{min}}$ corresponds to the minimum value among all training set samples of the variable; and $y_{\mathit{max}}$ corresponds to the maximum value among all training set samples of the variable.

The collected sample data were proportionally divided into a training set and a test set. Specifically, 72 sample sets were selected to construct the training set for fitting and training the neural network model, while the remaining 25 sample sets constituted the test set to independently evaluate the model’s generalization performance and prediction accuracy.

6.2. Performance Comparison of Machine Learning Algorithms for Maximum Temperature Prediction of Battery Packs

6.2.1. Prediction Based on Traditional Machine Learning Models

BPNN (Back Propagation Neural Network) [42], PSO (particle swarm optimization)-BPNN, GA (genetic algorithm)-BPNN, SVM (support vector machine) [43], RBFNN (Radial Basis Function Neural Network) [44], and RF (Random Forest) [45], as traditional machine learning models, have been widely applied across various industries. All the models take the same four characteristic variables—coolant flow rate, coolant inlet temperature, ambient temperature, and battery heating power—as inputs, and the maximum battery pack temperature as the output variable. The specific prediction process and hyperparameter selection methods for each model follow those described in reference [42,43,44,45]. They are not repeated here.

6.2.2. LSTM-Based Prediction

LSTM (Long Short-Term Memory), as the core enhanced model of recurrent neural networks (RNN), employs gating mechanisms (forget gate, input gate, output gate) and temporal gradient optimization strategies during training. By leveraging long-term memory storage of cell states and gating-based information filtering, combined with temporal data reconstruction and normalization preprocessing techniques, LSTM effectively addresses the gradient disappearance issue inherent in traditional RNNs. It accurately captures both short-term and long-term dependencies in data, demonstrating exceptional modeling and fitting capabilities for complex time-series datasets characterized by nonlinearity, non-stationarity, and temporal features. This makes LSTM a robust and efficient solution for tasks such as time-series prediction and sequence analysis.

(1)

Data preprocessing

The input layer selects the values of calorific value, ambient temperature, inlet temperature, and inlet flow velocity, while the output layer provides the maximum temperature of the battery pack. The data undergoes dimensionless processing and is substituted into Formula (8), with both input and output variables constrained to the interval [0, 1] to prevent model training instability and slower convergence caused by dimensional discrepancies.

(2)

Time-series data reconstruction

LSTM is suitable for modeling time-series data, requiring the raw data to be structured in a three-dimensional format of “number of samples × time steps × number of features” to meet the network’s input requirements.

(3)

Model structure and gate mechanism

The LSTM network architecture primarily consists of an input layer, LSTM hidden layers, and a fully connected output layer, as illustrated in Figure 16. Its core mechanism regulates information propagation through a forgetting gate, an input gate, and an output gate, with the internal structure detailed in Figure 17.

Within each LSTM layer, every neuron utilizes two normalization activation functions—Sigmoid and tanh—to process information. The variables are clearly defined: the neuron’s input at time t is X_t, its hidden state is h_t, and its output state is C_t. The information processing workflow is as follows: First, the neuron receives the previous time’s output h_t−1 and the current time’s input X_t; these are activated by the Sigmoid function to produce an input vector f_t. Since the Sigmoid function outputs values between 0 and 1, ft serves as the weight coefficient for the forgetting gate—where 0 corresponds to complete discarding of the information and 1 to its full retention. Finally, ft is element-wise multiplied with the cell state C_t−1. The specific expression for ft is:

$$f_t=\sigma \left(W_f\cdot \left[h_{t-1},X_t\right]+b_f\right)$$

(9)

In the formula: $W_f$ is the weight vector; $b_f$ is the bias vector; and σ denotes the Sigmoid activation function. The input gate acquires input information in the same manner as the forget gate, as shown in the following equation:

$$i_t=\sigma \left(W_i\cdot \left[h_{t-1},X_t\right]+b_i\right)$$

(10)

$$C_t=tanh\left(W_c\cdot \left[h_{t-1},X_t\right]+b_c\right)$$

(11)

In the formula: $W_i$ and $W_c$ are weight vectors; $b_i$ and $b_c$ are bias vectors. Furthermore, the neuron state at the current time step is updated as follows:

$$C_t=f_t\cdot C_{t-1}+i_t\cdot C_t$$

(12)

The output gate control is denoted as:

$$O_t=\sigma \left(W_o\cdot \left[h_{t-1},X_t\right]+b_o\right)$$

(13)

The final output is determined jointly by the output gate and the neuronal state:

$$h_t=O_t\cdot tanh\left(C_t\right)$$

(14)

(4)

Model Training and Prediction

Using the mean square error (MSE) as the loss function, the expression is as follows:

$$L={\displaystyle \frac{1}{m}}{\sum }_{i=1}^m{\left(y_i-\widehat{y_i}\right)}^2$$

(15)

where $y_i$ represents the observed value; $\widehat{y_i}$ denotes the predictive value; i denotes the sample index; and m denotes the number of samples.

The Adam optimizer iteratively updates the network weights and biases. The dataset is divided into a training set and a test set, with 72 samples used for model training and 25 samples for test validation. Using calorific value, ambient temperature, inlet temperature, and inlet flow rate as inputs, and the maximum battery pack temperature as the output, the LSTM model is trained. The trained network is then employed for predictions, as illustrated in the Figure 18 below.

6.3. Comparison of Prediction Results

The maximum temperature of the battery pack was predicted via BPNN, GA-BPNN, PSO-BPNN, SVM, RBFNN, RF, and LSTM. The known data samples were divided into training and validation sets. The fitting results for the training samples and the prediction results for the validation samples for each method are presented in Figure 19.

As shown in Figure 19a, except for the RF model, which exhibits significant deviations within local intervals and weak capability in capturing temperature abrupt changes, other models closely approximate the actual values, making direct comparison of their performance through this graph challenging. Figure 19b demonstrates that, in terms of overall trend fitting accuracy, the LSTM model’s prediction curve shows the highest degree of alignment with actual values, precisely capturing the global patterns of temperature sequences—including the rapid ascent from points 1 to 8, the sharp decline from points 8 to 10, the secondary rebound from points 10 to 17, and the stable fluctuation range from points 17 to 25. The prediction deviations at critical features such as peaks (point 8) and troughs (point 10) are minimal, highlighting its exceptional temporal feature learning ability. Traditional machine learning models (SVM, RF, RBF) exhibit notable limitations: the RBF model’s curve deviates significantly from actual values, showing severe underfitting during the high-temperature phase (points 1–8), with maximum prediction deviations exceeding 15 °C, failing to capture the nonlinear upward trend; the SVM model demonstrates insufficient accuracy during the rising phase, consistently producing lower-than-actual predictions; while the RF model performs adequately in the stable phase (points 17–25), its weak tracking capability during the volatile period (points 1–10) and pronounced curve smoothing characteristics make it incapable of reproducing temperature abrupt changes. A noteworthy anomaly is that traditional BP neural networks demonstrate significantly higher prediction accuracy across a large number of sample points compared to the GA-BP and PSO-BP models enhanced with intelligent optimization algorithms—a finding that contradicts the conventional belief that optimization algorithms improve neural network performance.

To further analyze Figure 19, the residual values of all methods at each sample point are calculated, and the corresponding residual plots are presented in Figure 20.

It can be seen from Figure 20a that the residuals of each sample in the training set range from −10 to 7 overall. The RF model exhibits the most drastic residual fluctuations with obvious peaks and troughs, and its maximum residual is close to ±10, indicating poor fitting stability on the training set and susceptibility to extreme errors. The residuals of other models including SVM, BP, GA-BP, PSO-BP, RBF, and LSTM are mostly within ±3 with slight fluctuations and satisfactory stability. The fitting accuracy of the adopted LSTM model on training samples is basically equivalent to that of other comparison models, and its prediction performance is even inferior to some other models at partial sample points. It can be observed from Figure 20b that the residuals of all samples in the validation set are distributed within the range of −20 to 5, and the overall error is obviously larger than that of the training set. The RBF model shows the most intense residual fluctuation, and its residual drops rapidly from 0 to −18 in the first ten samples, resulting in severe prediction underestimation, which reveals its poor generalization ability for new datasets. The SVM model ranks second in error fluctuation, with its residuals decreasing continuously in the first eight samples and reaching a maximum negative residual of approximately −13, presenting obvious generalization limitations. In general, the LSTM model achieves remarkably higher prediction accuracy than other comparative models.

To provide a more intuitive comparison of fitting accuracy and prediction accuracy, the four parameters—RMSE, MAE, MAPE (%), and R²—were evaluated separately, with the results presented in the table below.

As shown in

Table 3. Parameter values of the fitted sample.

Model	RMSE	MAE	MAPE(%)	R2
SVM	0.8215	0.6135	1.3852	0.9907
BP	1.1241	0.9326	2.0868	0.9825
GA-BP	0.7318	0.5363	1.2302	0.9926
PSO-BP	0.4787	0.3238	0.6903	0.9968
RF	2.7495	1.7786	3.8093	0.8953
RBF	0.8776	0.7129	1.5763	0.9893
LSTM	1.0661	0.9133	1.9536	0.9843

, the RMSE, MAE, MAPE (%), and R² results for the seven models demonstrate that, with the exception of the Random Forest (RF) model—which exhibits relatively lower fitting accuracy—the remaining six models all exhibit high fitting performance, consistent with the visual analysis conclusions presented in Figure 19. Quantitatively, the fitting accuracy ranking of the models is: RF < BP < LSTM < RBF < SVM < GA-BP < PSO-BP. Notably, the Long Short-Term Memory (LSTM) model’s accuracy is not optimal, ranking only in the middle tier; however, its performance gap with the best model (PSO-BP) is not statistically significant. Additionally, the traditional Back Propagation Neural Network (BP) performs worse than the genetically algorithm-optimized BP (GA-BP) and particle swarm optimization-optimized BP (PSO-BP) models, confirming that genetic algorithms (GA) and particle swarm optimization (PSO) enhance the performance of the traditional BP model, effectively improving its fitting capability.

However, as evidenced by the prediction metric data of validation samples presented in

Table 4. Parameter values of the validation sample.

Model	RMSE	MAE	MAPE(%)	R2
SVM	4.9137	3.2088	6.1244	0.4998
BP	1.8474	1.6624	3.5765	0.9293
GA-BP	3.4836	2.3424	4.5110	0.7486
PSO-BP	3.8410	2.9706	5.8779	0.6943
RF	2.5820	1.8098	3.7935	0.8619
RBF	6.8672	4.3751	8.2828	0.0230
LSTM	0.8068	0.6891	1.5653	0.9865

, the seven models exhibit markedly divergent performance: only the LSTM model maintains high prediction accuracy, while the other models show a sharp decline in performance. Notably, the radial basis function (RBF) and support vector machine (SVM) models produce severely distorted predictions that deviate significantly from actual values; all other models demonstrate varying degrees of accuracy degradation compared to the training phase. This phenomenon indicates that the six models except LSTM suffer from significant overfitting—their training process leads to excessive reliance on local features of the training data, resulting in poor generalization capabilities and an inability to adapt to subsequent data variations in validation samples. Crucially, the BP model outperforms both GA-BP and PSO-BP models in validation samples—a counterintuitive result that further confirms overfitting: while algorithmic optimization enhances model fit, it simultaneously increases dependence on local training features, compromising generalization performance. The prediction accuracy ranking across validation samples is as follows: RBF < SVM < PSO-BP < GA-BP < RF < BP < LSTM.

In summary, the LSTM model demonstrates stable and excellent accuracy performance on both training and validation datasets, combining strong fitting capability with robust generalization performance. It effectively adapts to the dynamic temperature patterns of electric vehicle battery packs, providing reliable technical support for temperature prediction in this context and offering substantial practical value.

7. Conclusions

This paper takes CTP lithium-ion power battery packs as the research object, designs a bionic honeycomb liquid cooling plate, and establishes a fluid–structure coupling numerical model via ANSYS Fluent. The effects of coolant flow velocity, inlet temperature, ambient temperature, and heat generation power on heat dissipation performance are studied. Meanwhile, seven machine learning models are compared for temperature prediction, and an integrated thermal management system combining structural optimization, working condition analysis and intelligent prediction is established. The main conclusions are drawn as follows:

• The honeycomb liquid cooling plate achieves excellent heat dissipation and temperature uniformity. Compared with traditional rectangular and serpentine channels, hexagonal honeycomb channels realize uniform flow distribution with low flow resistance, which effectively eliminates flow dead zones and local heat accumulation. Under conventional conditions, no obvious hot spots appear inside the battery pack, and the overall maximum temperature difference is controlled within 5 °C. It can steadily keep the battery temperature within the optimal range of 25–45 °C, showing better comprehensive thermal performance than traditional cooling structures.
• The quantitative variation rules and safe operating thresholds of key thermal management parameters are clarified. The cooling effect gradually weakens with the rise in flow velocity, and the optimal flow velocity is confirmed as 1.4 m/s. Further increasing the flow velocity brings little temperature drop but greatly raises energy consumption. Battery temperature is approximately linearly positively correlated with coolant inlet temperature, ambient temperature, and heat generation power. Under the conventional heat generation of 3000 W/m³, the battery can work safely when the coolant inlet temperature is below 23 °C and the ambient temperature is below 27 °C. Multi-parameter cooperative regulation is required under high heat generation and high ambient temperature conditions.
• The coupling mechanism of multiple parameters and working condition adaptability are summarized. Coolant inlet temperature and ambient temperature are the dominant temperature control factors, flow velocity plays only an auxiliary role, and battery heat generation power is the core internal factor causing temperature rise. High heat generation conditions present obvious thermal amplification effect with higher temperature rise. The honeycomb structure is applicable to a wide range of working conditions and maintains favorable temperature uniformity under diverse environments and loads.
• LSTM is selected as the optimal model for predicting the maximum battery temperature. Verified by simulation data, PSO-BPNN and GA-BPNN are prone to overfitting with poor practical prediction performance. The LSTM model can accurately capture the nonlinear and time-varying characteristics of battery temperature, and possesses higher prediction accuracy and stronger generalization ability than other algorithms, which is more suitable for real-time engineering temperature prediction.
• An integrated battery thermal management scheme is constructed. Combined with bionic cooling structure, quantified temperature control thresholds, and high-precision intelligent prediction model, this scheme realizes high-efficiency heat dissipation, low energy consumption, and reliable thermal safety operation of CTP battery packs. It can provide theoretical basis and technical support for battery thermal management design, parameter matching, and intelligent early warning of electric vehicles under fast charging, high-rate discharge and high-temperature operating scenarios.

Research Limitations

• The simulation model is properly simplified without considering battery thermal expansion and gas–liquid two-phase flow of coolant, leading to minor temperature deviations under extreme working conditions.
• All research results are obtained based on numerical simulation without prototype tests. Key conclusions such as optimal flow velocity and safe temperature control boundaries need further verification through bench experiments.
• The machine learning models are mainly trained using steady-state data, and their prediction accuracy needs to be optimized under transient fast charging and dynamic variable load conditions.
• The model neglects the electro-thermal coupling characteristics inside the battery, adopts the equivalent volume method and the assumption of constant heat source, and fails to reflect the dynamic feedback of temperature on battery internal resistance and heat generation rate, which may introduce simulation errors under working conditions such as high rate and large temperature difference.

Future Research Directions

• Establish an electro-thermal-fluid-structure multi-field coupling model to analyze the influence of battery deformation and thermal expansion on heat dissipation performance.
• Combine nanofluids and phase change materials with honeycomb channels to further improve heat dissipation capacity under high heat generation conditions.
• Integrate LSTM with reinforcement learning to develop adaptive cooperative control strategies for coolant flow velocity and temperature.
• Adopt temperature-dependent thermophysical parameters to improve prediction accuracy in both high- and low-temperature environments.