In modern power systems, the integration of high-proportion renewable energy sources has made large-scale energy storage a critical component for ensuring grid stability and reliability. Among various technologies, lithium-ion batteries, particularly LiFePO4-based energy storage lithium battery systems, have gained prominence due to their rapid response capabilities and flexible deployment. However, the operational safety of these energy storage lithium battery installations is often compromised by thermal runaway (TR) events, which can escalate into cascading failures within densely packed modules. This phenomenon poses significant risks to energy storage lithium battery infrastructure, leading to potential system-wide disasters. Understanding the triggering mechanisms and propagation dynamics of thermal runaway in energy storage lithium battery configurations is essential for developing effective safety protocols and fault mitigation strategies.
This study focuses on the thermal runaway behavior of 280 Ah LiFePO4 energy storage lithium battery cells, modules, and clusters, which are representative of real-world grid-scale deployments. Through controlled thermal abuse experiments, we systematically capture key parameters such as voltage drop thresholds, temperature rise rates, and peak temperatures during thermal runaway events. The experimental data provide a foundation for constructing high-fidelity multiphysics models that simulate the electro-thermal-chemical coupling in energy storage lithium battery systems. By integrating a forgetting factor recursive least squares (FFRLS) parameter identification method, we establish dynamic mappings between the state of charge (SOC) and thermal runaway parameters, enabling accurate predictions of thermal propagation paths under varying operational conditions.
The thermal runaway process in energy storage lithium battery systems is governed by a series of exothermic side reactions, which can be described using Arrhenius-based kinetics. The primary reactions include solid-electrolyte interphase (SEI) decomposition, anode-electrolyte reactions, cathode-electrolyte reactions, and electrolyte decomposition. The heat generation from these reactions is quantified as follows:
$$ Q_{\text{SEI}} = H_{\text{SEI}} \cdot W_{\text{SEI}} \cdot R_{\text{SEI}} $$
$$ Q_{\text{ne}} = H_{\text{ne}} \cdot W_{\text{ne}} \cdot R_{\text{ne}} $$
$$ Q_{\text{pe}} = H_{\text{pe}} \cdot W_{\text{pe}} \cdot R_{\text{pe}} $$
$$ Q_{\text{ele}} = H_{\text{ele}} \cdot W_{\text{ele}} \cdot R_{\text{ele}} $$
$$ \Sigma Q_{\text{side-tot}} = Q_{\text{SEI}} + Q_{\text{ne}} + Q_{\text{pe}} + Q_{\text{ele}} $$
where $Q$ represents the heat generation rate per unit volume (W/m³), $H$ denotes the heat of reaction per unit mass (J/kg), $W$ is the mass concentration of reactants (kg/m³), and $R$ is the reaction rate (s⁻¹). The reaction rates follow the Arrhenius equation:
$$ R_i = A_i \cdot \exp\left(-\frac{E_{a,i}}{RT}\right) \cdot c^m $$
where $A_i$ is the pre-exponential factor, $E_{a,i}$ is the activation energy, $R$ is the universal gas constant, $T$ is temperature, $c$ is the concentration of unstable lithium, and $m$ is the reaction order.
To address the nonlinear dependencies of thermal runaway parameters on SOC, we employ the FFRLS algorithm for dynamic parameter identification. The algorithm updates parameter estimates recursively:
$$ \hat{\theta}(k+1) = \hat{\theta}(k) + K(k+1) \cdot e(k+1) $$
$$ K(k+1) = \frac{P(k) \cdot \phi(k+1)}{\lambda + \phi^T(k+1) \cdot P(k) \cdot \phi(k+1)} $$
$$ P(k+1) = \frac{1}{\lambda} \left[ P(k) – K(k+1) \cdot \phi^T(k+1) \cdot P(k) \right] $$
where $\hat{\theta}$ is the parameter vector, $K$ is the correction gain, $P$ is the covariance matrix, $\phi$ is the input vector, $\lambda$ is the forgetting factor (set to 0.98), and $e$ is the prediction error. This approach allows us to model the SOC-dependent activation energy as:
$$ E_{a,i}(\text{SOC}) = k_{i,0} + k_{i,1} \cdot \exp(k_{i,2} \cdot \text{SOC}) $$
where $k_{i,0}$, $k_{i,1}$, and $k_{i,2}$ are identified coefficients.
Experimental results for 280 Ah energy storage lithium battery cells at different SOC levels are summarized in the table below, highlighting key thermal runaway characteristics:
| Parameter | 25% SOC | 50% SOC | 75% SOC | 100% SOC |
|---|---|---|---|---|
| Voltage Drop Time (s) | 1496 | 1430 | 984 | 327 |
| TR Trigger Time (s) | 2683 | 1698 | 1010 | 469 |
| Peak Temperature (°C) | 236.8 | 238.4 | 350.9 | 378.3 |
| Max Heating Rate (°C/s) | 2.2 | 3.9 | 16.4 | 26.5 |
The data indicate that higher SOC levels lead to earlier thermal runaway triggering and more severe temperature excursions, emphasizing the need for SOC-aware safety management in energy storage lithium battery systems.
Building on the experimental insights, we develop a multiphysics model for energy storage lithium battery modules, comprising 52 cells in a 4×13 series configuration. The model incorporates thermal propagation constraints, including anisotropic heat transfer due to the layered structure of battery cores. The energy conservation equation is expressed as:
$$ \rho C_p \frac{\partial T}{\partial t} = \lambda \nabla^2 T + \frac{\Sigma Q_{\text{side-tot}}}{V_{\text{cell}}} – \frac{Q_{\text{diss-tot}}}{V_{\text{batt}}} $$
where $\rho$ is density, $C_p$ is specific heat capacity, $\lambda$ is thermal conductivity, $V_{\text{cell}}$ is the cell volume, and $V_{\text{batt}}$ is the battery volume. Dissipation terms account for convection and radiation:
$$ Q_{\text{diss-tot}} = h A_{\text{batt}} (T – T_{\text{amb}}) + \epsilon \sigma_B A_{\text{batt}} (T^4 – T_{\text{amb}}^4) $$
with $h$ as the convection coefficient, $A_{\text{batt}}$ as the surface area, $\epsilon$ as emissivity, and $\sigma_B$ as the Stefan-Boltzmann constant.
Simulation results reveal distinct thermal propagation paths in energy storage lithium battery modules. At high SOC (75% and 100%), thermal runaway propagates rapidly along the inter-layer direction (y-axis) due to higher radial thermal conductivity ($\lambda_y = \lambda_z = 18$ W/m·K) compared to the through-layer direction ($\lambda_x = 1.5$ W/m·K). The time intervals between successive cell failures are summarized below:
| Propagation Direction | 100% SOC Interval (s) | 75% SOC Interval (s) |
|---|---|---|
| X-axis (Through-layer) | 73 | 98.2 |
| Y-axis (Inter-layer) | 66 | 89.5 |
At lower SOC levels (25% and 50%), thermal runaway propagation is suppressed, indicating a critical SOC threshold for cascading failures in energy storage lithium battery arrays.
For fault localization in energy storage lithium battery clusters, we propose a hybrid Genetic Algorithm-Grey Wolf Optimizer (GA-GWO) enhanced Time Difference of Arrival (TDOA) method. The TDOA algorithm leverages the time differences of thermal wave arrivals at multiple sensors to locate the fault source. The fundamental equation is:
$$ \Delta t = \frac{d_A – d_B}{v} $$
where $\Delta t$ is the time difference, $d_A$ and $d_B$ are distances from the fault to sensors A and B, and $v$ is the thermal wave speed. The fault location lies on a hyperbola defined by:
$$ \frac{(x – x_A)^2}{a^2} – \frac{(y – y_B)^2}{b^2} = 1 $$
with parameters $a$ and $b$ related to $\Delta t$ and $v$.
The GA-GWO fusion optimizes the inversion process by combining global search capabilities of GA with the leadership hierarchy of GWO. The fitness function for localization is:
$$ \text{Fitness}(t_i) = \frac{1}{n} \sum_{j=1}^{n} (t_{1j} – t_{i,j})^2 $$
where $t_{1j}$ is the time difference between the first and j-th sensors in the reference data, and $t_{i,j}$ is the corresponding value for the i-th fault candidate. The algorithm dynamically adjusts parameters to minimize this fitness value, ensuring robust localization across varying SOC conditions.
We evaluate the localization performance using a directional dual-threshold error metric (DD-TEM), defined as:
$$ \text{DD}(p, a) = \sum_{i \in \{x,y,z\}} \frac{|p_i – a_i|}{\tau_i} + \alpha \cdot \prod_{i: |p_i – a_i| > \tau_i} \lambda $$
where $p$ is the predicted position, $a$ is the actual position, $\tau_i$ are direction thresholds, $\alpha$ is a scaling factor, and $\lambda$ is a penalty term for cross-layer errors. A DD error below 0.03 indicates accurate localization.
Comparative analysis of localization methods for energy storage lithium battery clusters demonstrates the superiority of the GA-GWO TDOA approach:
| Localization Method | Average DD Error (25% SOC) | Average DD Error (100% SOC) |
|---|---|---|
| TDOA Direct | 0.127 | 0.088 |
| PSO-Based | 0.053 | 0.048 |
| Single GWO | 0.079 | 0.057 |
| GA-GWO TDOA | 0.027 | 0.019 |
The GA-GWO method achieves over 94% accuracy in fault localization across all SOC levels, with minimal cross-layer misjudgments, making it highly suitable for real-world energy storage lithium battery monitoring systems.
In conclusion, this study presents a comprehensive framework for modeling thermal runaway propagation and enabling precise fault localization in LiFePO4 energy storage lithium battery systems. The integration of experimental data, multiphysics simulations, and advanced optimization algorithms provides a robust foundation for enhancing the safety and reliability of grid-scale energy storage lithium battery deployments. Future work will focus on refining thermal management strategies and extending the methodology to other battery chemistries and configurations.