As the global energy landscape shifts toward decarbonization, the integration of renewable energy sources such as solar and wind has imposed unprecedented challenges on grid stability. The inherent intermittency and unpredictability of these sources necessitate efficient energy storage solutions. Among various technologies, electrochemical storage, particularly lithium-ion batteries, has emerged as a cornerstone for balancing supply and demand, enabling peak shaving, frequency regulation, and reliable power delivery. However, the performance, lifespan, and safety of lithium-ion batteries are profoundly sensitive to operating temperature. A well-designed thermal management system is therefore indispensable for any battery energy storage system (BESS). In this review, I aim to systematically explore the state-of-the-art thermal management technologies for BESS, focusing on air cooling and liquid cooling, along with their optimization strategies, thermal runaway prevention methods, and future directions. My objective is to provide a comprehensive reference for researchers and engineers working on safe and efficient BESS design.
The critical role of temperature in battery performance is well established. Lithium-ion batteries operate most efficiently within a narrow temperature window, typically between 25 °C and 40 °C. Within this range, electrochemical reactions are most active, achieving near 100% coulombic efficiency. Deviations from this range cause significant degradation. At low temperatures, electrolyte viscosity increases and reaction kinetics slow, leading to reduced capacity and power output; below -40 °C, the battery may become inoperable. Conversely, high temperatures accelerate aging mechanisms, increase internal resistance, and can trigger thermal runaway—a catastrophic chain reaction where the solid electrolyte interphase (SEI) dissolves, the anode reacts exothermically with the electrolyte, and the separator eventually fails, causing internal short circuits. Once initiated, thermal runaway is self-sustaining until all reactants are consumed. Hence, controlling both the maximum temperature and the temperature uniformity within a BESS is paramount, especially for large-scale installations where thousands of cells are densely packed. Unlike electric vehicle (EV) batteries, stationary BESS often operate in fixed enclosures with higher total energy content, making uniformity and long-term reliability even more critical. The following table summarizes the key thermal performance requirements for BESS, highlighting differences from EV applications.
| Author(s) | Cooling Method | Key Metrics | Distinction from EV |
|---|---|---|---|
| Various | Forced air cooling | Temperature difference, maximum temperature | Greater emphasis on overall uniformity; tighter tolerance for hot spots |
| Yang et al. | Forced air cooling | Average temperature, maximum temperature difference | Container-scale BESS allows adjustment of air supply strategies; larger spatial variability |
| Lin et al. | Forced air cooling | Maximum temperature difference, average temperature | Optimization is often geometry-specific; generalizability is limited |
| Ki et al. | Indirect liquid cooling | Temperature uniformity, maximum temperature | Traditional liquid cooling modules require high pumping power due to long serpentine channels; coolant temperature rise causes cell-to-cell deviation |
| Gu et al. | Indirect liquid cooling | Maximum temperature, maximum temperature difference | Complex environment and large number of cells necessitate cost-effective structural optimization |
| Cao et al. | — | Trade-off between prediction accuracy and computational cost | Full-scale BESS models must balance detail and efficiency |
| Wu et al. | Forced air cooling | Maximum average temperature, maximum average temperature difference, fan power consumption | Fan power in large BESS is substantial; must reduce power without sacrificing cooling performance |
| Meng et al. | Forced/natural air cooling | Maximum temperature, electricity cost | BESS charge/discharge power is often moderate; hybrid natural convection can reduce electricity cost |
Many studies have proposed mathematical models to describe battery thermal behavior. A common simplified lumped thermal model can be expressed as:
$$ \rho c_p \frac{\partial T}{\partial t} = \nabla \cdot (k \nabla T) + \dot{q} $$
where \(\rho\) is density, \(c_p\) is specific heat, \(T\) is temperature, \(t\) is time, \(k\) is thermal conductivity, and \(\dot{q}\) is the volumetric heat generation rate from the battery. The heat generation rate can be further decomposed into reversible and irreversible components, often expressed by the Bernardi equation:
$$ \dot{q} = I \left( E_{oc} – V \right) – I T \frac{\partial E_{oc}}{\partial T} $$
where \(I\) is current, \(E_{oc}\) is open-circuit voltage, \(V\) is terminal voltage, and the second term accounts for entropic heating. For a battery pack consisting of \(N\) cells, the temperature uniformity metric is often defined as the maximum temperature difference \(\Delta T_{max} = T_{max} – T_{min}\). Achieving \(\Delta T_{max} < 5 K\) is a common design target for BESS.
In the following sections, I will delve into the two most widely adopted active cooling techniques—air cooling and liquid cooling—discussing their structural optimization, parametric tuning, and performance outcomes. I will then address thermal runaway prevention and suppression strategies, and finally outline future research directions.
2. Air Cooling Technology
Air cooling, due to its low cost, simple design, and ease of maintenance, remains the predominant thermal management method for many BESS, particularly those with moderate power densities. The basic principle involves using fans or blowers to drive air through the battery pack, removing heat via forced convection. However, the low thermal conductivity and specific heat capacity of air impose limitations on its cooling capacity, especially for high-rate applications. Consequently, optimizing airflow distribution is crucial to enhancing both the maximum temperature and temperature uniformity. In this section, I review recent advances in air cooling structure optimization, supply strategy adjustments, and enhanced cooling methods.
2.1 Structural Optimization of Air Cooling Systems
The geometry of air ducts, plenums, and guide vanes significantly affects the flow distribution across battery modules. Several researchers have explored novel duct configurations to improve uniformity. For instance, a “main duct + vertical riser” design has been proposed, where each riser corresponds to a battery rack and delivers air precisely to each battery box. By adding guide vanes at the inlet, duct corners, and riser outlets, and by reducing the outlet area, the coefficient of variation of air velocity was reduced by about 93%, effectively mitigating local hot spots. Another study employed a layered air duct system to achieve point-to-point air supply from the air conditioner to battery modules, and optimized the cross-sectional area of supply outlets to ensure temperature uniformity. This design reduced the airflow deviation between rows to only 0.011 m³/s, corresponding to a deviation rate of 7.8%. Similarly, adding guide vanes at the bottom duct to redirect airflow from the air conditioner was shown to reduce the temperature difference between battery surfaces to 4 K, significantly shrinking the high-temperature zones. Optimization of baffle dimensions and angles has also been reported to lower the system’s maximum temperature from 398.43 K to 328.41 K in one scenario, and from 367.90 K to 334.92 K in another, with average temperature reductions of 18 K and 8 K, respectively. The following table summarizes various system-level air cooling structural optimizations and their reported outcomes.
| Author(s) | Optimization Method | Metrics | Results |
|---|---|---|---|
| Lyu et al. | Changed fan baffle shape; reduced number of duct openings | Maximum temperature, maximum temperature difference | Maximum temperature reduced by 13.90 K; maximum temperature difference reduced by 10.50 K |
| Shen et al. | Added baffles at return air vent; altered inlet path | Maximum temperature, temperature difference | Temperature difference reduced; maximum temperature of battery pack in container BESS decreased |
| Wu et al. | Adjusted battery spacing distribution; installed tapered inlet/outlet ducts | Maximum temperature, maximum temperature difference, fan power | Maximum temperature dropped to 311.35 K; maximum temperature difference dropped to 0.30 K; power consumption reduced ~30% |
| Zhu et al. | Optimized supply air angle; optimized return air vent location | Maximum temperature, temperature difference | At 90° supply angle, maximum temperature reduced to 306.73 K, temperature difference to 3.46 K; optimal return vent location at Z = 0.85 m |
| Wang et al. | Composite duct with L-shaped baffle and orifice plate | Maximum temperature difference | Maximum temperature difference decreased from 17.36 K to 4.35 K; uniformity improved by 74.94% |
| Zhang et al. | Added baffles and heat exchangers at module bottom | Maximum temperature, maximum temperature difference | Maximum temperature difference reduced 1.30 K (10.9%); maximum temperature reduced 1.00 K (47.6%) |
| Huang et al. | Innovative configuration with batteries in series and air flowing parallel to batteries | Maximum temperature, temperature difference | Maximum temperature dropped from 315.45 K to 310.65 K; temperature difference from 287.45 K to 283.35 K |
| Ding et al. | Analyzed perforated deflectors; added guide vanes in overhead duct; used porous deflectors with varying porosity | Maximum temperature difference | Flow uniformity improved by 98.24% compared to non-perforated deflectors; maximum temperature on long and short flow sides reduced by 8.31 K and 5.13 K, respectively |
These improvements demonstrate that careful design of air passages can substantially lower both the peak temperature and the spatial temperature gradient, thereby extending battery life and reducing the risk of thermal runaway initiation. The underlying fluid dynamics can be described by the Navier-Stokes equations, and optimization often relies on computational fluid dynamics (CFD) simulations. For turbulent flow, the Reynolds-averaged Navier-Stokes (RANS) equations with turbulence models (e.g., k-epsilon) are commonly used:
$$ \frac{\partial (\rho u_j)}{\partial x_j} = 0 $$
$$ \frac{\partial (\rho u_i u_j)}{\partial x_j} = -\frac{\partial P}{\partial x_i} + \frac{\partial}{\partial x_j} \left[ (\mu + \mu_t) \left( \frac{\partial u_i}{\partial x_j} + \frac{\partial u_j}{\partial x_i} \right) \right] $$
where \(u\) is velocity, \(P\) is pressure, \(\mu\) is dynamic viscosity, and \(\mu_t\) is turbulent viscosity. These equations, coupled with the energy equation, enable prediction of temperature fields and guide structural optimization.
2.2 Air Supply Strategy Adjustments
Beyond geometric modifications, controlling the manner in which air is supplied can yield significant benefits. One study developed a localized enhanced jet air duct region at the bottom of the battery box, where air is pressurized through a tapered duct and forced upward to cool central cells. At a 3C discharge rate, this strategy limited the maximum temperature differences during charging, resting, and discharging to 20.16 K, 7.16 K, and 22.81 K, respectively. Another innovative approach is dew-point evaporative cooling, which pre-cools ambient air before it enters the battery enclosure. The cooler consists of dry and wet channels: ambient air passes through the dry channel while water evaporates in the adjacent wet channel, cooling the dry channel. Part of the air is diverted to the wet channel to sustain evaporation, and the remaining product air is used for battery cooling. This technique can lower the supply air temperature to near the dew point of the inlet air, achieving average battery temperatures between 276.15 K and 286.75 K when ambient air is at 291.25 K. The following table summarizes different air supply optimization strategies.
| Author(s) | Optimization Method | Metrics | Results |
|---|---|---|---|
| Shi et al. | Periodically reversed supply direction to enhance airflow disturbance | Maximum temperature, temperature difference | Maximum temperature reduced from 311 K to 308 K; cell-to-cell temperature difference dropped by 3 K |
| Yang et al. | Changed fan working direction; proposed four ventilation schemes | Average temperature, maximum temperature difference | Optimized scheme reduced average temperature by 1.16% and maximum temperature difference by 54.36% compared to baseline |
| Lin et al. | Implemented floor supply + ceiling return; duct supply + ceiling return | Average temperature, maximum temperature difference | Maximum temperature difference reduced from 304.35 K to 276.65 K; average temperature reduced by 5.80 K |
3. Liquid Cooling Technology
For high-power BESS where air cooling is insufficient, liquid cooling provides superior heat transfer performance due to the higher thermal conductivity and specific heat of liquids. Liquid cooling systems can be categorized as direct (immersion) or indirect (cold plate) types. Indirect liquid cooling, where coolant flows through channels in a cold plate thermally coupled to the battery, is more common in BESS because it avoids electrical conductivity concerns and simplifies maintenance. In this section, I discuss the optimization of liquid cooling plate structures and the selection of coolants.
3.1 Liquid Cooling Plate Structural Optimization
The geometry of cooling channels within the cold plate is critical for achieving uniform temperature and low pressure drop. Various channel designs—such as serpentine, parallel, and manifold—have been investigated. One study optimized a dual-inlet/outlet ribbed channel; compared to a smooth channel, the ribbed design increased total heat transfer by 68% and enhanced the hydrothermal performance coefficient by 58%. Another work compared four different cold plate configurations (e.g., I-shaped, U-shaped, L-shaped, and m-shaped layouts) and found that the m-shaped arrangement provided the best trade-off between temperature uniformity and pumping power. For a 1C discharge rate, the optimal inlet temperature was approximately 291.90 K, with a minimum coolant flow rate of 6 L/h for 1C and 12 L/h for 2C. A high-efficiency liquid cooling module incorporating a manifold distributor connected to a multi-scale porous metal layer was also developed: at the same flow rate, this design improved temperature uniformity by 12.9% and lowered the maximum temperature by 7.4%. The following table presents several liquid cooling structural optimizations.
| Author(s) | Optimization Method | Metrics | Results |
|---|---|---|---|
| Ashkboos et al. | Studied dual-inlet/outlet ribbed channels with four different longitudinal rib configurations | Maximum temperature | Total heat transfer increased by 68%; hydrothermal performance coefficient improved by 58% |
| Xu et al. | Compared four cold plate structures; m-shaped arrangement selected | Temperature difference | Optimal inlet temperature ~291.90 K; for 1C/2C, minimum flow rates of 6/12 L/h |
| Gu et al. | Designed liquid cooling pipe with increased channel count and reduced pipe cross-section | Maximum temperature, maximum temperature difference | Optimized pack maximum temperature reduced to 289.65 K; liquid line maximum temperature difference reduced to 298.15 K |
| Ki et al. | Developed energy-efficient liquid cooling module with manifold distributor and multi-scale porous metal layer | Maximum temperature, temperature difference | At same flow rate, temperature difference improved 12.9%; maximum temperature lowered 7.4% |
3.2 Selection of Coolant
The choice of coolant significantly impacts the thermal management performance. While water-glycol mixtures and dielectric oils are conventional, supercritical carbon dioxide (sCO₂) has emerged as a promising candidate due to its exceptional thermophysical properties near the critical point. Near the pseudo-critical temperature, sCO₂ exhibits a specific heat capacity that can be several orders of magnitude higher than that of conventional coolants, along with high thermal conductivity. A comparative study between sCO₂, mineral oil, and a commercial dielectric fluid (AmpCool) showed that sCO₂ reduced the peak temperature by 61% and 53%, and reduced the temperature spread by 61% and 56%, respectively, compared to mineral oil and AmpCool. Moreover, sCO₂ is environmentally benign and low-cost. The thermophysical properties of sCO₂—such as density, specific heat, viscosity, and thermal conductivity—vary dramatically with temperature and pressure near the critical point (31.1 °C, 7.38 MPa). This behavior can be modeled using equations of state, for example, the Peng-Robinson equation:
$$ P = \frac{RT}{(V_m – b)} – \frac{a(T)}{V_m(V_m + b) + b(V_m – b)} $$
where \(V_m\) is molar volume, \(R\) is the gas constant, and \(a(T)\) and \(b\) are substance-specific parameters. The high heat capacity near the critical point allows sCO₂ to absorb large amounts of heat with minimal temperature rise, making it highly effective for high-heat-flux BESS applications.
4. Thermal Runaway Prevention and Suppression
Thermal runaway is the most critical safety hazard in BESS, especially in large-scale installations where a single cell failure can propagate to neighboring cells, leading to catastrophic fires or explosions. Effective thermal management plays a dual role in both preventing and suppressing thermal runaway. Prevention relies on early detection and intervention, while suppression focuses on slowing or arresting propagation once it begins.
Enhancing heat dissipation is a primary method for suppression. Both air cooling and liquid cooling can significantly extend the time before thermal runaway occurs in adjacent cells. For instance, increasing the airflow rate or using liquid cooling can maintain cell temperatures below the onset temperature of exothermic reactions. However, prevention requires more sophisticated strategies, including electro-thermal modeling and real-time temperature monitoring.
Electro-thermal models accurately simulate battery thermal behavior, enabling predictive control. These models range from lumped parameter models to multi-node distributed models. A comparison between model predictions and field data from an operational BESS showed that the overall temperature difference across the container surface was within 2 K, and the cell-to-cell difference was within 1 K, demonstrating excellent agreement. Another study developed a multi-node electro-thermal model (MNETM) that matched finite element results with a maximum error less than 2 K and a root-mean-square error less than 1 K over the entire cycle. The following table summarizes key electro-thermal models developed for large-scale BESS.
| Author(s) | Results |
|---|---|
| Cao et al. | Model predicted temperature within 2 K on container surface and within 1 K between cells when compared with field data from a SGCCESS BESS; good consistency |
| Tao et al. | Battery modules installed at dimensionless position 0.19 along z-axis; temperature monitored by thermocouple 2 matched simulated maximum temperature closely in all four cases; temperature difference <1 K even in worst case |
| Liu et al. | Internal temperature estimation with mean absolute error <0.8 K; mean relative error <2%; root-mean-square error for validation conditions ranged 0.0833–0.5705 K |
| Pan et al. | Multi-node electro-thermal model (MNETM) showed strong agreement with finite element model; maximum error <2 K, root-mean-square error <1 K over entire cycle |
Real-time temperature monitoring combined with intelligent control can prevent overheating. For example, a hybrid approach integrating electro-thermal models with long short-term memory (LSTM) neural networks has been proposed for early thermal warning. The system can detect abnormal temperature diffusion patterns and issue alerts before critical thresholds are reached. The control strategy can then adjust cooling parameters, such as fan speed or coolant flow rate, to mitigate the hot spot. The following table presents several control strategies applied to large BESS.
| Author(s) | Strategy | Results |
|---|---|---|
| He et al. | Observor-based control combined with reciprocating cooling flow; derived from reduced-order model | Rapidly reduced cell-to-cell temperature non-uniformity from 4.20 K to 1.00 K upon detection of temperature rise |
| Mesgarpour et al. | Pattern-based neural networks (PBANNs) integrating physics-informed neural networks with visual tracking | Computation time reduced by 65.41% for fast response; by distributing cooling tubes uniformly, average temperature of battery and PCM reduced by 25.30% |
| Liu et al. | Adaptive model predictive control based on intelligent neural network for J-type air-cooled BESS | Control method could timely adjust cooling and maintain temperature uniformity within 1.33 K |
It is important to note that prevention also involves limiting the charge/discharge rate (C-rate) during normal operation, as high C-rates generate excessive heat that can overwhelm the cooling system. Moreover, thermal runaway suppression can be enhanced by using flame-retardant electrolytes or phase change materials that absorb heat and delay propagation. However, these materials add cost and complexity, and their integration with active cooling systems remains an active research area.
5. Future Outlook
As BESS scales continue to grow and applications diversify, the demands on thermal management will become more stringent. Several future research directions merit attention.
First, novel cooling technologies such as mist cooling, spray cooling, and dew-point evaporative cooling offer the potential for higher heat transfer coefficients and lower energy consumption compared to conventional air or liquid cooling. However, their practical deployment in large BESS requires further investigation into reliability, maintenance, and cost. For instance, droplet-based cooling can achieve high heat fluxes but may introduce issues of electrical insulation and corrosion.
Second, multi-dimensional evaluation frameworks should be established to assess thermal management systems not only on cooling performance but also on energy efficiency, cost, weight, volume, and safety. A comprehensive objective function might be formulated as:
$$ \text{Cost Function} = w_1 \Delta T_{max} + w_2 T_{max} + w_3 P_{fan/pump} + w_4 \text{Cost}_{material} + w_5 \text{Risk}_{TR} $$
where \(w_i\) are weighting factors determined by application priorities. Such a framework would enable systematic optimization across conflicting criteria.
Third, advanced control algorithms, including model predictive control (MPC) and reinforcement learning, can dynamically adjust cooling parameters in real time based on load forecasts and battery state of health. These methods can significantly reduce energy consumption while maintaining safe temperatures.
Fourth, the integration of phase change materials (PCMs) and heat pipes with active cooling systems is promising. Hybrid systems can handle peak heat loads passively, reducing the required capacity of active cooling components. However, the low thermal conductivity of many PCMs remains a challenge; embedded metal foams or graphite matrices can enhance conductivity but add mass and cost.
Finally, the development of standardized testing protocols and simulation benchmarks for BESS thermal management would accelerate the comparison and adoption of new technologies. The use of digital twins—virtual replicas of physical BESS that integrate real-time sensor data and physics-based models—offers a powerful platform for design optimization and predictive maintenance.
In conclusion, the thermal management of battery energy storage systems is a multifaceted engineering challenge that directly impacts safety, performance, and economics. Air cooling and liquid cooling remain the most mature technologies, with substantial gains achievable through geometric optimization and smart control. Emerging coolants like sCO₂ and advanced control strategies promise further improvements. However, the ultimate goal—safe, efficient, and long-lasting BESS operation—requires a holistic approach that balances thermal, electrical, and economic trade-offs. I hope this review provides a valuable roadmap for researchers and practitioners striving to achieve that goal.