As the global energy landscape shifts towards renewable sources, energy storage lithium battery systems have become indispensable for grid stability, peak shaving, and integrating intermittent power generation. However, the widespread deployment of these systems is accompanied by significant safety concerns, primarily due to the risk of thermal runaway in lithium-ion batteries. Thermal runaway can lead to fires, explosions, and substantial economic losses, as evidenced by numerous incidents worldwide. For instance, between 2017 and 2024, over 90 fire or explosion accidents were reported in energy storage lithium battery facilities, highlighting the critical need for robust safety measures. Traditional firefighting approaches often prove inadequate, as they address consequences rather than root causes, underscoring the importance of proactive risk management. In this context, safety risk early warning technologies have emerged as a pivotal solution, enabling the detection of potential hazards before they escalate into irreversible events. This article comprehensively reviews the safety risk sources, accident progression, and various early warning technologies for energy storage lithium battery systems, with a focus on their principles, comparative advantages, and practical engineering applications. By analyzing the entire accident development spectrum—from initial defects to thermal runaway—we aim to clarify the distinctions between “early warning” (targeting nascent risks) and “alarming” (triggered post-thermal runaway), thereby addressing market confusion and promoting standardized practices. Furthermore, we evaluate the effectiveness of these technologies through real-world data and case studies, emphasizing the role of data-driven approaches in enhancing the safety and reliability of energy storage lithium battery installations.
The safety of energy storage lithium battery systems is a multifaceted issue, rooted in the complex interplay of various risk sources. These risks can be broadly categorized into four main types: battery intrinsic defects, external severe impacts, operational environment degradation, and management system failures. Battery intrinsic defects arise from manufacturing imperfections, such as metal impurities, misaligned separators, or electrode damage, which can induce internal short circuits (ISC) and accelerate aging. External severe impacts include electrical shocks from insulation failures or external short circuits, mechanical shocks from handling or accidents, and thermal shocks from adjacent equipment fires, all of which can compromise battery integrity and trigger failures. Operational environment degradation refers to adverse conditions like high temperatures, which accelerate aging and increase thermal runaway risk; low temperatures, which promote lithium plating and ISC; and high humidity or dust, which degrade insulation and raise leakage current risks. Management system failures involve malfunctions in the Battery Management System (BMS), Energy Management System (EMS), or Power Conversion System (PCS), leading to issues like overcharging, over-discharging, or inadequate monitoring. For example, a faulty BMS sensor may fail to detect voltage anomalies, allowing a battery to operate beyond safe limits. According to industry analyses, integration and construction errors during the initial phases of system deployment account for a significant portion of incidents, with Balance of System (BOS) components and control systems being frequent failure points. Understanding these risk sources is crucial for developing effective early warning strategies, as they represent the genesis of potential accidents in energy storage lithium battery systems.
The progression of safety incidents in energy storage lithium battery systems follows a well-defined path, which can be mapped into distinct stages: the hidden risk stage, early fault stage, pre-thermal runaway stage, and thermal runaway stage. In the hidden risk stage, which spans years to months, issues like manufacturing defects or management system failures create latent hazards without immediate visible symptoms. For instance, a BMS communication error might lead to gradual capacity imbalance, but the battery’s external parameters remain within normal ranges. The early fault stage, lasting months to days, involves the emergence of abnormalities such as gas generation, swelling, micro-shorts, or lithium plating, often triggered by operational abuse or environmental stress. Here, the battery exhibits subtle changes in mechanical and electrochemical properties, detectable through advanced sensing techniques. The pre-thermal runaway stage occurs over hours to minutes, characterized by rapid gas production, valve venting, and a sharp drop in voltage, signaling imminent thermal runaway. Finally, the thermal runaway stage unfolds in seconds, with uncontrollable temperature rise, combustion, and potentially explosions. Each stage emits specific characteristic signals: for example, in the early fault stage, ultrasonic or stress sensors can detect internal changes, while in the pre-thermal runaway stage, gas or acoustic sensors provide timely alerts. This accident development map not only facilitates targeted monitoring but also clarifies the temporal window for intervention. Early warning technologies must align with these stages to be effective; for instance, methods focusing on the hidden risk stage can prevent accidents days in advance, whereas those activated in the pre-thermal runaway stage offer only minutes of lead time. Thus, a comprehensive approach that covers the entire spectrum is essential for safeguarding energy storage lithium battery systems.
Early warning technologies for energy storage lithium battery systems can be classified into two main categories: those based on external signal sensing and those leveraging operational data analysis. External signal sensing technologies monitor physical or chemical changes in the battery or its environment, such as particles, gases, temperature, stress, or sound. For instance, particle signal-based warning detects submicron particles released during thermal runaway initiation using light scattering or resistive techniques, but it is prone to environmental interference and typically provides alerts only in the pre-thermal runaway stage. Gas signal-based warning, often targeting hydrogen (H₂) emitted early in thermal runaway, employs sensors like electrochemical or semiconductor types, with studies showing it can offer warnings up to 145 seconds before thermal runaway. However, its effectiveness is limited by gas diffusion delays and sensor durability. Thermal signal-based warning relies on temperature sensors (e.g., thermocouples, RTDs, or NTC thermistors) to monitor surface temperature and rate of rise, but it primarily serves as an alarm due to its late response. Acoustic signal-based warning uses ultrasonic sensors to detect internal changes like gas bubbles or cracks, offering early detection in the hidden risk stage, but it faces challenges in cost and installation complexity. Stress signal-based warning measures mechanical forces from swelling or impacts using piezoelectric or strain sensors, providing high sensitivity to early faults but suffering from environmental susceptibility and high costs. In contrast, operational data analysis technologies utilize voltage, current, and temperature data from the BMS to identify anomalies through model-based, statistical, or machine learning algorithms. For example, equivalent circuit models (ECM) can estimate state of charge (SOC) and internal resistance to detect ISC, while statistical methods like correlation analysis or entropy measures identify outliers. Machine learning approaches, such as artificial neural networks (ANN) or long short-term memory (LSTM) networks, enable predictive analytics by learning from historical data, potentially covering all accident stages. The table below summarizes the key characteristics of these technologies:
| Technology Type | Signals Monitored | Applicable Stages | Advantages | Disadvantages |
|---|---|---|---|---|
| Particle Sensing | Micro-particles, aerosols | Pre-thermal runaway | Fast response to early emissions | High false alarms, environment-dependent |
| Gas Sensing | H₂, CO, other gases | Pre-thermal runaway | Early detection, cost-effective | Diffusion delays, sensor aging |
| Thermal Sensing | Temperature, rate of change | Thermal runaway | Direct measurement, simple | Late warning, limited proactivity |
| Acoustic Sensing | Ultrasonic waves, sound waves | Hidden risk, early fault | High sensitivity to internal changes | High cost, complex installation |
| Stress Sensing | Mechanical force, strain | Early fault | Precise detection of swelling | Environmental interference, expensive |
| Operational Data Analysis | Voltage, current, temperature | All stages | Comprehensive coverage, predictive | High computational demand, data quality reliance |
Mathematically, many operational data analysis methods rely on battery models to quantify risks. For instance, the equivalent circuit model (ECM) is commonly used to describe battery dynamics. The voltage output \( V(t) \) can be expressed as:
$$ V(t) = OCV(SOC) – I(t)R_0 – \sum_{i=1}^{n} I(t) * R_i \left(1 – e^{-t/(R_i C_i)}\right) $$
where \( OCV(SOC) \) is the open-circuit voltage as a function of state of charge, \( I(t) \) is the current, \( R_0 \) is the internal resistance, and \( R_i \) and \( C_i \) are the resistance and capacitance of the RC loops, respectively. By estimating parameters like \( R_0 \) or detecting residuals, anomalies such as ISC can be identified early. Similarly, machine learning algorithms use historical data to predict faults; for example, a support vector machine (SVM) classifier might minimize the risk function:
$$ \min_{w,b} \frac{1}{2} \|w\|^2 + C \sum_{i=1}^{m} \max(0, 1 – y_i (w^T x_i + b)) $$
where \( w \) and \( b \) are model parameters, \( C \) is a regularization constant, \( x_i \) represents input features (e.g., voltage sequences), and \( y_i \) denotes fault labels. Such data-driven approaches are particularly effective for energy storage lithium battery systems, as they can integrate multiple data streams to provide holistic risk assessments.
In engineering applications, the implementation of early warning technologies for energy storage lithium battery systems is guided by international and national standards to ensure safety and interoperability. Key standards include UL 1973 and UL 9540 in North America, which focus on battery safety and system-level hazards, and IEC 62619 and IEC 62933-5-2 in Europe, addressing industrial battery safety and grid integration. In China, standards such as GB/T 36276 for lithium batteries used in energy storage and GB/T 42288 for safety regulations provide a framework for risk monitoring and management. Recent updates, like GB/T 44767 for safety monitoring information systems, emphasize the need for active early warning capabilities, including data analysis and anomaly diagnosis. Statistically, an analysis of deployed systems shows that thermal and gas sensing technologies are universally adopted due to their direct relevance to thermal runaway, accounting for 100% of installations. In contrast, ultrasonic and stress sensing are rarely used (0% adoption) because of high costs and technical complexities. Particle sensing sees moderate adoption (around 20%), reflecting a market interest in early detection despite limitations, while operational data analysis is claimed by many but often lacks verification, leading to potential misrepresentation. For instance, some vendors market basic threshold-based monitoring as “machine learning,” causing confusion. A survey of over 4 GWh of energy storage lithium battery capacity revealed that BMS failures constitute the most frequent risk source (74.8% of incidents), with issues like sensor failures and均衡 malfunctions being predominant. This underscores the importance of robust data analytics; in one case study from Shandong, China, operational data analysis detected voltage inconsistencies and temperature anomalies days before failures, enabling preventive maintenance. The table below illustrates the distribution of risk events identified in practical deployments:
| Risk Category | Specific Failure | Percentage of Total Incidents |
|---|---|---|
| BMS Failures | NTC Sensor Failure | 30.6% |
| BMS Failures | 均衡 Failure | 18.1% |
| BMS Failures | Voltage Sensor Failure | 4.1% |
| Cooling System Failures | Liquid Cooling Fault | 11.3% |
| Cooling System Failures | Air Cooling Fault | 1.8% |
| Connection Issues | BCU Wiring Looseness | 9.6% |
| Battery Abnormalities | Internal Short Circuit | 0.3% |
| Battery Abnormalities | Abnormal Aging | 2.2% |
These findings highlight that operational data analysis can effectively identify risks at their origin, such as BMS issues, preventing escalation to battery-level faults. For example, in a 100 MW/200 MWh energy storage lithium battery plant, early warnings based on voltage divergence and temperature spikes allowed for timely repairs, reducing the probability of thermal runaway by over 50% in monitored systems. This demonstrates the critical role of full-stage early warning technologies in enhancing the safety and operational efficiency of energy storage lithium battery systems.
In conclusion, safety risk early warning is paramount for the sustainable deployment of energy storage lithium battery systems, given their susceptibility to thermal runaway and associated hazards. Through a detailed examination of risk sources, accident progression, and warning technologies, it is evident that a multi-stage approach is essential. External signal-based methods, such as gas or thermal sensing, offer valuable alerts but are largely reactive, limited to later stages like pre-thermal runaway. In contrast, operational data analysis technologies, encompassing model-based, statistical, and machine learning algorithms, provide comprehensive coverage across all stages—from hidden risks to thermal runaway—enabling proactive risk mitigation. The superiority of data-driven approaches is reinforced by engineering applications, where they have significantly reduced incident probabilities by addressing root causes like BMS failures. However, market challenges persist, including the misrepresentation of alarm technologies as early warning solutions and a lack of standardization in algorithm validation. To address this, we recommend stricter adherence to international standards, enhanced transparency in machine learning implementations, and the development of unified evaluation frameworks for energy storage lithium battery safety. Future efforts should focus on advancing intelligent algorithms, improving sensor integration, and fostering industry-wide collaboration to ensure that early warning systems not only meet regulatory requirements but also deliver reliable performance in real-world scenarios. By prioritizing these measures, stakeholders can bolster the safety and reliability of energy storage lithium battery systems, supporting the global transition to clean energy.