Air-cooled energy storage cabinet cluster level thermal runaway early warning method and system

Abstract

The application relates to the technical field of energy storage system safety monitoring, and particularly provides a method and system for early warning of thermal runaway of an air-cooled energy storage cabinet cluster. The method is characterized in that three types of functionally heterogeneous monitoring points are arranged in the cabinet to capture signals in different spatial dimensions. By establishing an adaptive benchmark for dynamically tracking the environment background and calculating multi-dimensional abnormal features such as the instantaneous difference, the change rate and the cumulative exposure of the gas concentration, the perception sensitivity of early weak signals under ventilation dilution is improved. The early warning trigger adopts a multi-level fusion logic, cross-verification is performed on the basis of multiple abnormal evidences, gradient response is performed from early risk to emergency stage, and a dual confirmation mechanism of spatial timing consistency confirmation and external disturbance exclusion confirmation is introduced, so that internal real faults and external environmental disturbances can be effectively distinguished, early warning is realized, and the false alarm rate is reduced.

Description

Technical Field

[0001] This application relates to the field of energy storage system safety monitoring technology, and provides a method and system for early warning of thermal runaway at the cluster level of air-cooled energy storage cabinets. Background Technology

[0002] With the expansion of new energy power generation, lithium-ion battery energy storage is widely used in power grids and industrial and commercial applications. Among them, air-cooled energy storage cabinets adopt a modular battery cluster design, with lithium iron phosphate batteries, which have a long cycle life, being the mainstream. However, batteries are at risk of thermal runaway under extreme conditions. Studies have shown that there is a precursor period of several minutes to tens of minutes before thermal runaway occurs, during which gases such as VOCs and CO are released sequentially, accompanied by a temperature rise. This provides a crucial window for early warning by monitoring signals such as gases and temperature.

[0003] Existing early warning technologies for thermal runaway in energy storage batteries mainly fall into three categories. The first is direct monitoring, which involves placing sensors inside the battery pack. This provides early warning but is costly and engineering-intensive. The second is a scheme that deploys gas and temperature sensors throughout the overall environment inside the energy storage cabinet. While simple to implement, the signals are easily diluted by ventilation, leading to detection delays. The third method involves fusing multi-parameter data and using algorithms for analysis. While this improves reliability, its monitoring layout often fails to fully consider the specific diffusion patterns of gases within the air-cooled cabinet.

[0004] However, the aforementioned technologies face three unresolved issues in practical applications of air-cooled commercial and industrial energy storage cabinets. First, continuous high-volume ventilation rapidly dilutes the trace amounts of gas released early in the battery's lifespan, resulting in extremely low overall concentrations within the cabinet and insufficient sensitivity of solutions based on fixed thresholds. Second, external contamination entering through ventilation or operational disturbances can cause changes in gas concentration within the cabinet; existing technologies cannot effectively distinguish between such external disturbances and genuine internal battery faults, leading to a high false alarm rate. Finally, existing fusion-based early warning strategies fail to utilize the temporal patterns of various parameter signals during the pre-thermal runaway phase and the spatial correlation characteristics of gas propagation within the cabinet, resulting in incomplete early warning logic and insufficient ability to reliably identify the source of anomalies under complex operating conditions. Summary of the Invention

[0005] This application addresses the problems of low sensitivity and high false alarm rate in existing thermal runaway warning systems for energy storage batteries under air-cooled conditions. One approach is to provide a cluster-level thermal runaway warning method for air-cooled energy storage cabinets, comprising the following steps:

[0006] Three types of monitoring points with heterogeneous functions are set up in the energy storage cabinet, including the first type of monitoring point for overall status judgment, the second type of monitoring point for early perception and joint verification, and the third type of monitoring point for external disturbance reference.

[0007] Simultaneously collect multi-parameter monitoring signals from the first type of monitoring points, the second type of monitoring points, and the third type of monitoring points;

[0008] Based on the collected gas concentration signals, an adaptive background benchmark for dynamically tracking the environmental background is established and maintained for each gas concentration signal;

[0009] Based on each gas concentration signal and its corresponding adaptive background benchmark, the first type of gas anomaly features, the second type of gas anomaly features, and the third type of gas anomaly features are calculated.

[0010] Based on the calculated first type of gas anomaly features, second type of gas anomaly features, third type of gas anomaly features, temperature anomaly features, and smoke signals, multi-source evidence fusion is performed within a sliding time window, and the first type of warning level trigger judgment corresponding to the early stage of the risk, the second type of warning level trigger judgment corresponding to the development stage of the risk, and the third type of warning level trigger judgment corresponding to the emergency stage of the risk are executed sequentially.

[0011] After any of the aforementioned warning levels is triggered, the abnormal signals that triggered the warning levels are sequentially subjected to spatial temporal consistency confirmation and external disturbance elimination confirmation.

[0012] After both the spatial temporal consistency confirmation and the external disturbance elimination confirmation are passed, an early warning signal corresponding to the current trigger level is output.

[0013] In one feasible implementation, three types of heterogeneous monitoring points are set up within the energy storage cabinet, including:

[0014] The first type of monitoring points are set in the exhaust collection area of ​​the energy storage cabinet;

[0015] The second type of monitoring point is set in the sensing area above the battery cluster;

[0016] The third type of monitoring point will be set in the air inlet area of ​​the energy storage cabinet;

[0017] The first type of monitoring point, the second type of monitoring point, and the third type of monitoring point are all equipped with gas sensors for monitoring the concentration of carbon monoxide, volatile organic compounds, and hydrogen. The first type of monitoring point and the second type of monitoring point are also equipped with temperature sensors and smoke sensors.

[0018] In one feasible implementation, based on the acquired gas concentration signals, an adaptive background benchmark for dynamically tracking the environmental background is established and maintained for each gas concentration signal, including:

[0019] The adaptive background benchmark is updated recursively using an exponential moving average algorithm.

[0020] When any of the aforementioned warning levels is triggered, the updates of the adaptive background baseline for all relevant gas concentration signals are immediately frozen;

[0021] Once the warning level is lifted, the adaptive background benchmark is reinitialized using the sampled value at the time of lifting, and its recursive update is resumed.

[0022] In one feasible implementation, based on each gas concentration signal and its corresponding adaptive background benchmark, the first type of gas anomaly features, the second type of gas anomaly features, and the third type of gas anomaly features are calculated, including:

[0023] Calculate the first type of gas anomaly feature, which is the gas concentration increment, and the gas concentration increment is the instantaneous difference between the current sampled value and the current adaptive background reference value;

[0024] Calculate the second type of gas anomaly characteristics, where the second type of gas anomaly characteristics are trend characteristics, and the trend characteristics are the rate of change of the first type of gas anomaly characteristics within a preset time period;

[0025] Calculate the third type of gas anomaly feature, which is a cumulative exposure feature, and the cumulative exposure feature is the sum of all positive first type of gas anomaly features within a preset sliding integral window.

[0026] In one feasible implementation, based on the calculated first type of gas anomaly features, second type of gas anomaly features, third type of gas anomaly features, temperature anomaly features, and smoke signals, multi-source evidence fusion is performed within a sliding time window, and the first type of early warning level trigger judgment corresponding to the early stage of risk is executed sequentially, including:

[0027] When the abnormal volatile organic compound concentration characteristics of both the first type of monitoring point and the second type of monitoring point meet their preset low-level threshold range.

[0028] Alternatively, the abnormal carbon monoxide concentration and abnormal hydrogen concentration characteristics of the first type of monitoring points and the second type of monitoring points both meet their preset low-level threshold range.

[0029] Alternatively, if smoke alarm signals are present at both the first type of monitoring point and the second type of monitoring point, the first type of warning level will be triggered.

[0030] In one feasible implementation, the step of performing multi-source evidence fusion within a sliding time window, and sequentially executing the second type of early warning level trigger judgment corresponding to the risk development stage and the third type of early warning level trigger judgment corresponding to the risk emergency stage, includes:

[0031] The corresponding level of warning is triggered based on whether, within the sliding time window, the first type of monitoring points and the second type of monitoring points possess at least three independent types of abnormal evidence selected from the following list:

[0032] (1) There is the same gas, and the concentration anomalies of the gas at both the first type of monitoring point and the second type of monitoring point meet the preset specific level threshold range;

[0033] (2) There is another gas, and the concentration anomalies of this gas at both the first type of monitoring point and the second type of monitoring point meet the preset specific level threshold range;

[0034] (3) All of them have smoke alarm signals;

[0035] (4) The temperature anomaly characteristics of the first type of monitoring point or the second type of monitoring point meet the preset specific level threshold range;

[0036] in,

[0037] The conditions for triggering the second type of warning level are: the abnormal gas concentration characteristics in evidence (1) meet the medium level threshold range, the abnormal gas concentration characteristics in evidence (2) meet the low level threshold range, and the abnormal temperature characteristics in evidence (4) meet the low level threshold range.

[0038] The conditions for triggering the third type of warning level are: the abnormal gas concentration feature in evidence (1) meets the high-level threshold range, the abnormal gas concentration feature in evidence (2) meets the medium-level threshold range, and the abnormal temperature feature in evidence (4) meets the high-level threshold range.

[0039] In one feasible implementation, after triggering any of the aforementioned warning levels, the abnormal signals that triggered the warning levels are sequentially subjected to spatial-temporal consistency verification, including:

[0040] Obtain the trigger gas on which the current warning level is based;

[0041] The moment when the trigger gas first meets its preset low-level threshold range at the second type of monitoring point, and the moment when the trigger gas first meets its preset low-level threshold range at the first type of monitoring point;

[0042] When the abnormal time of the trigger gas at the second type of monitoring point is earlier than the abnormal time at the first type of monitoring point, and the time difference is within a preset reasonable delay range, or when the abnormal time difference between the trigger gas at the first type of monitoring point and the second type of monitoring point is less than a preset synchronization threshold, the spatial temporal consistency is determined to be confirmed.

[0043] In one feasible implementation, after triggering any of the aforementioned warning levels, the abnormal signals that triggered the warning levels are sequentially subjected to external disturbance elimination confirmation, including:

[0044] Obtain the trigger gas on which the current warning level is based;

[0045] If the trigger gas does not meet its preset level threshold range at the third type of monitoring point, the external disturbance elimination confirmation is deemed successful.

[0046] If the trigger gas meets the preset low-level threshold range at the third type of monitoring point, it is further determined whether the abnormal time of the third type of monitoring point is earlier than the abnormal time of the first type of monitoring point or the second type of monitoring point, and whether the cumulative exposure characteristic difference of the trigger gas between the first type of monitoring point, the second type of monitoring point and the third type of monitoring point is less than the preset gradient threshold.

[0047] When the third type of monitoring point anomaly occurs earlier and the cumulative exposure feature difference is less than the gradient threshold, the current anomaly is determined to conform to the external disturbance pattern, and the warning output is suppressed or delayed for confirmation.

[0048] Another aspect of this application provides a thermal runaway early warning system at the cluster level for air-cooled energy storage cabinets, used to implement the thermal runaway early warning method at the cluster level for air-cooled energy storage cabinets as described in any of the above claims, including:

[0049] The heterogeneous monitoring layout module is configured to deploy three types of heterogeneous monitoring points, including a first type of monitoring point, a second type of monitoring point, and a third type of monitoring point, within the energy storage cabinet.

[0050] The signal acquisition and preprocessing module is configured to synchronously acquire multi-parameter monitoring signals from various monitoring points in the heterogeneous monitoring layout module, and to verify the validity of the acquired signals.

[0051] The adaptive reference processing module is configured to establish and maintain an adaptive background reference for each gas concentration signal based on the effective gas concentration signal output by the signal acquisition and preprocessing module, and to dynamically track the environmental background.

[0052] The anomaly feature calculation module is configured to calculate the first type of gas anomaly features, the second type of gas anomaly features, and the third type of gas anomaly features based on the background reference and gas concentration signals output by the adaptive reference processing module, and to calculate the temperature anomaly features.

[0053] The multi-level early warning triggering module is configured to perform multi-source evidence fusion within a sliding time window based on various abnormal features and smoke signals output by the abnormal feature calculation module, and sequentially execute the triggering judgment of the first type of early warning level, the second type of early warning level, and the third type of early warning level.

[0054] The early warning confirmation module is configured to, after the multi-level early warning triggering module triggers an early warning, sequentially perform spatial temporal consistency confirmation and external disturbance elimination confirmation on the abnormal signal that triggered the early warning, and output an early warning signal after both confirmations are passed.

[0055] In one feasible implementation, the heterogeneous monitoring layout module includes:

[0056] The first composite detector node, as the first type of monitoring point, is set in the exhaust air gathering area of ​​the energy storage cabinet and integrates sensors for carbon monoxide, volatile organic compounds, hydrogen, temperature and smoke.

[0057] The second composite detector node, as the second type of monitoring point, is set in the sensing area above the battery cluster and integrates sensors for carbon monoxide, volatile organic compounds, hydrogen, temperature and smoke.

[0058] The third composite detector node, as the third type of monitoring point, is set in the air inlet area of ​​the energy storage cabinet and integrates carbon monoxide, volatile organic compounds and hydrogen sensors.

[0059] The thermal runaway early warning method and system for air-cooled energy storage cabinets provided in this application enhance the ability to capture early signals under ventilation dilution by setting up three types of monitoring points with heterogeneous functions and combining adaptive benchmarks that dynamically track the environmental background with multi-dimensional anomaly feature calculations, thereby improving the timeliness of early warning. Secondly, it adopts a multi-level early warning triggering and multi-source evidence fusion strategy, and introduces a dual verification mechanism of spatial-temporal consistency confirmation and external disturbance exclusion. The former uses the physical laws of gas propagation from the battery cluster to the exhaust vent to verify the rationality of the anomaly source, while the latter uses the inlet monitoring point as a reference to effectively distinguish between internal gas generation and external pollution interference, thereby reducing the system's false alarm rate. Attached Figure Description

[0060] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the embodiments of the invention. It is obvious that the drawings described below are merely some embodiments of the invention, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort.

[0061] Figure 1This is a flowchart illustrating an exemplary embodiment of the thermal runaway early warning method for an air-cooled energy storage cabinet cluster.

[0062] Figure 2 This is a schematic diagram of the structure of a thermal runaway early warning system at the cluster level of an air-cooled energy storage cabinet, as shown in an exemplary embodiment of this application.

[0063] Figure 3 This is a schematic diagram illustrating the distribution of three types of monitoring points in an air-cooled energy storage cabinet, as shown in an exemplary embodiment of this application.

[0064] Figure label:

[0065] 100-System Module; 101-Heterogeneous Monitoring Layout Module; 102-Signal Acquisition and Preprocessing Module; 103-Adaptive Reference Processing Module; 104-Anomaly Feature Calculation Module; 105-Multi-level Early Warning Trigger Module; 106-Early Warning Confirmation Module;

[0066] 1011 - Category I monitoring point; 1012 - Category II monitoring point; 1013 - Category III monitoring point. Detailed Implementation

[0067] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that the embodiments of the invention will be more comprehensive and complete, and will fully convey the concept of the exemplary embodiments to those skilled in the art. The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments. In the following description, numerous specific details are provided to give a full understanding of how embodiments of the invention are carried out.

[0068] With the continuous expansion of new energy power generation and the increasing demand for grid peak shaving and frequency regulation, lithium-ion battery energy storage systems are widely used in grid-side, industrial and commercial, and distributed energy systems. Industrial and commercial energy storage systems typically employ a modular battery cluster structure, integrating battery modules into sealed or semi-sealed energy storage cabinets, and relying on forced air cooling or natural ventilation for battery thermal management. Lithium iron phosphate (LFP) batteries, due to their long cycle life and relatively good safety, have become the mainstream chemical system in the current industrial and commercial energy storage field.

[0069] However, lithium-ion batteries still pose a risk of thermal runaway under conditions such as overcharging, over-discharging, external short circuits, mechanical damage, or high-temperature environments. Once thermal runaway occurs, the internal temperature of the battery can rise rapidly to over 700°C within seconds to tens of seconds, accompanied by the release of a large amount of flammable gas. Therefore, early warning technology for thermal runaway is one of the requirements for the safety protection of energy storage systems.

[0070] Studies have shown that battery thermal runaway is a gradual evolutionary process. Before thermal runaway officially occurs, there is typically a precursor period lasting several minutes to tens of minutes, during which the following detectable physicochemical characteristics appear successively:

[0071] 1. Organic solvents (ethylene carbonate, dimethyl carbonate, etc.) in the electrolyte may volatilize or decompose due to increased temperature, releasing volatile organic compounds (VOCs).

[0072] 2. The decomposition of the SEI film on the negative electrode and the side reactions between the positive electrode material and the electrolyte release gases such as carbon monoxide (CO) and hydrogen (H2); the surface temperature of the battery rises slowly.

[0073] 3. As internal pressure accumulates, the pressure relief valve opens, accompanied by the ejection of smoke or aerosol particles.

[0074] The aforementioned precursor characteristics provide a signal window for early warning of thermal runaway, and joint monitoring based on multiple parameters such as gas concentration and temperature is considered an important technical approach.

[0075] Existing thermal runaway detection and early warning technologies are mainly divided into three categories:

[0076] The first category is battery internal monitoring technology, which directly senses the state by placing temperature, pressure, or fiber optic sensors inside the battery pack or the cell itself. This type of solution can detect internal temperature or pressure anomalies several minutes to more than ten minutes before thermal runaway occurs, providing an early warning response; however, it requires modification of the battery pack structure, which is difficult and costly in engineering, and the sensors are exposed to an electrochemical corrosion environment for a long time, posing challenges to reliability and lifespan, making it difficult to apply on a large scale in industrial and commercial energy storage.

[0077] The second category is overall environmental monitoring technology for energy storage cabinets, which involves installing gas sensors, smoke detectors, or temperature monitoring devices inside the cabinet. This solution is simple to implement and low in cost, and has already been applied in engineering projects. However, because the monitoring points are located at the top of the cabinet or in the overall environment, the sensors detect the average concentration after mixing and dilution, and only trigger an alarm when the anomaly develops to a significant stage, making it difficult to capture weak signals in the early stages. At the same time, relying on fixed threshold alarms cannot adapt to the long-term drift of the background gas concentration inside the cabinet with factors such as ambient temperature and humidity, and is prone to seasonal false alarms or missed alarms.

[0078] The third category is multi-sensor fusion early warning technology, which combines multiple parameters such as gas, temperature, and smoke for joint analysis. Some studies use machine learning methods for feature extraction and pattern recognition. However, existing solutions mostly monitor at a single point or the overall scale of the cabinet, and do not adequately consider the diffusion path, transmission timing, and spatial distribution of gas inside the battery cluster under air-cooled conditions. The sensor layout lacks a specific design for air-cooled conditions, which limits the actual early warning performance.

[0079] The existing technologies described above have the following three problems in the practical application of air-cooled industrial and commercial energy storage cabinets, which have not yet been effectively solved.

[0080] Question 1: The dilution effect of air-cooled ventilation leads to insufficient sensitivity in early warning.

[0081] Air-cooled energy storage cabinets rely on axial flow fans for continuous ventilation, resulting in a high air exchange rate. During the pre-thermal runaway phase, the gas release is extremely small (as low as a few milliliters per minute before the pressure relief valve opens). Under high-volume ventilation, this gas is rapidly diluted, keeping the overall gas concentration inside the cabinet at an consistently low level, below the alarm threshold of a fixed-threshold scheme, making it difficult for sensors to respond. Simultaneously, the fixed absolute threshold cannot distinguish between thermal runaway signals and normal fluctuations in background concentration, further reducing the reliability of early detection and leading to warning delays.

[0082] Question 2: The high false alarm rate is caused by confusion between external disturbances and internal anomalies.

[0083] The air inlet of the air-cooled cabinet is open to the outside, allowing external pollutants to enter the cabinet with the airflow. Operational procedures or changes in sealing can also cause instantaneous fluctuations in gas concentration. Current technology lacks a spatial discrimination mechanism for gas sources, making it impossible to distinguish whether the concentration increase originates from pre-thermal runaway gases inside the battery cluster or from external inputs, leading to frequent false alarms under external disturbances.

[0084] Question 3: The lack of synchronization of multiple parameters in time and the absence of spatial correlation lead to imperfect early warning logic.

[0085] During the prodromal phase of thermal runaway, the timing of various parameters is asynchronous: the release order of VOCs (volatile organic compounds), CO (carbon monoxide), and H2 (hydrogen) lags physically behind the temperature rise and smoke generation, with significant differences in timing and amplitude across different systems and failure modes. Existing multi-sensor fusion methods often employ a "multiple parameters exceeding thresholds simultaneously" or simple voting strategies, failing to utilize the temporal relationships of parameter anomalies for physical law verification, thus unable to distinguish between genuine thermal runaway and external disturbances. Furthermore, the spatial propagation time of gas from the release point to the detection point is not considered, and the signal sequence of sensors at different locations is not incorporated into the judgment logic, weakening the ability to identify the source of anomalies.

[0086] In summary, existing technologies for early warning of thermal runaway at the battery cluster scale in air-cooled industrial and commercial energy storage cabinets face three mutually restrictive problems: insufficient early sensitivity, high false alarm rate, and difficulty in identifying the source of the anomaly. Therefore, in order to solve the aforementioned problems, referring to... Figure 1 As shown, this application proposes a method for early warning of thermal runaway at the cluster level in an air-cooled energy storage cabinet, including the following steps:

[0087] S100: Three types of heterogeneous monitoring points are set up in the energy storage cabinet, including the first type of monitoring point for overall status judgment, the second type of monitoring point for early perception and joint verification, and the third type of monitoring point for external disturbance reference.

[0088] Specifically, for the battery cluster-level space of the air-cooled energy storage cabinet, according to the functional positioning of three types of monitoring points, the first, second, and third types of monitoring points are respectively deployed in the corresponding functional areas within the energy storage cabinet. The functions of the three types of monitoring points are independent and complementary to each other. The deployment area of ​​the first type of monitoring points must be able to cover the overall state of the battery cluster space of the energy storage cabinet; the deployment area of ​​the second type of monitoring points must be able to prioritize the detection of abnormal signals in the early stage of thermal runaway of the battery clusters; and the deployment area of ​​the third type of monitoring points must be able to reflect the relevant parameters of the external environment of the energy storage cabinet in real time.

[0089] In some embodiments, setting up three types of functionally heterogeneous monitoring points within the energy storage cabinet specifically includes the following steps:

[0090] S101: Set the first type of monitoring point in the exhaust collection area of ​​the energy storage cabinet.

[0091] For energy storage cabinets that use air cooling, the exhaust convergence area inside the cabinet is determined. This area is the convergence outlet of the airflow inside the energy storage cabinet and is the path through which all airflow exits the cabinet. It can reflect the overall cumulative state of parameters such as gas and temperature in the entire battery cluster space.

[0092] By setting the first type of monitoring points in the exhaust convergence area of ​​the energy storage cabinet, the first type of monitoring points can accurately collect the cumulative state of parameters such as gas concentration and temperature of the entire battery cluster space inside the cabinet. This realizes the functional positioning of the first type of monitoring points for overall state judgment. At the same time, the airflow convergence characteristics of the exhaust convergence area enable the first type of monitoring points to stably sense the overall parameter changes inside the cabinet, avoiding fluctuations in monitoring data caused by local airflow disturbances.

[0093] S102: Set the second type of monitoring point in the sensing area above the battery cluster.

[0094] Determine the installation location of the battery clusters inside the energy storage cabinet, and define a sensing area above the battery clusters. This area is located above the direct convection heat exchange zone of the battery modules. Gases released during the pre-thermal runaway phase of the batteries will first accumulate in this area after leaking from the top of the battery modules, and then diffuse towards the exhaust convergence area with the airflow inside the cabinet. Second-class monitoring points are set up in the sensing area above the battery clusters, which can preferentially contact the abnormal gases leaking from the battery modules without being disturbed by the airflow in other areas inside the cabinet.

[0095] By setting the second type of monitoring points in the sensing area above the battery cluster, the second type of monitoring points can detect the presence of abnormal gases before the abnormal gases released during the pre-thermal runaway period are diluted by a large amount of airflow in the cabinet. This improves the early warning system's ability to detect abnormal signals in the early stage of thermal runaway and can effectively eliminate false alarms caused by occasional abnormalities of single-point sensors.

[0096] S103: Set the third type of monitoring point in the air inlet area of ​​the energy storage cabinet.

[0097] The location of the air inlet of the energy storage cabinet is determined. This location is the path for outside air to enter the energy storage cabinet and can reflect the parameter status of the outside air in real time. A third type of monitoring point is set up in the air inlet area, which can come into contact with the outside air entering the cabinet and collect the real-time parameters of the outside air inlet without being disturbed by the airflow in the battery cluster area inside the cabinet.

[0098] By setting the third type of monitoring point in the air inlet area of ​​the energy storage cabinet, the third type of monitoring point can collect the parameter status of the external air in real time, providing an external environmental reference benchmark for subsequent external disturbance elimination and confirmation, enabling the early warning system to effectively distinguish whether the abnormal parameters inside the cabinet are caused by thermal runaway abnormalities inside the battery cluster or by disturbances caused by polluting gases in the external environment entering the cabinet with the air inlet.

[0099] S104: Configure corresponding sensors for the three types of monitoring points. The first type of monitoring point, the second type of monitoring point, and the third type of monitoring point are all equipped with gas sensors for monitoring the concentration of carbon monoxide, volatile organic compounds, and hydrogen. The first type of monitoring point and the second type of monitoring point are also equipped with temperature sensors and smoke sensors.

[0100] Specifically, carbon monoxide, volatile organic compound (VOC), and hydrogen gas sensors are configured for all three types of monitoring points. These sensors are used to measure the concentrations of carbon monoxide, VOCs, and hydrogen at their respective monitoring points. Additionally, temperature and smoke sensors are provided for the first and second types of monitoring points. The temperature sensor monitors the ambient temperature at the corresponding point, and the smoke sensor monitors the smoke concentration. No temperature or smoke sensors are configured for the third type of monitoring point. All sensors are connected to the data acquisition and processing unit at their respective monitoring points to ensure signal acquisition and transmission at a uniform sampling frequency.

[0101] Carbon monoxide, volatile organic compounds, and hydrogen are all characteristic gases that are released prematurely during the pre-thermal runaway phase of lithium-ion batteries. By simultaneously configuring sensors for these three types of gases, the early warning system can capture the earliest gas release signals of thermal runaway. By additionally configuring temperature and smoke sensors at the first and second types of monitoring points, simultaneous monitoring of subsequent characteristic signals such as temperature increases and smoke release during the thermal runaway evolution process is achieved, covering the characteristic signals of the entire evolution process from the pre-thermal phase to the irreversible stage.

[0102] Furthermore, by not configuring temperature and smoke sensors for the third type of monitoring points, the hardware configuration of the third type of monitoring points is simplified while ensuring the external disturbance reference function. This reduces costs and avoids interference from changes in the external ambient temperature on temperature monitoring, allowing the third type of monitoring points to be used as a reference for judging external gas disturbances.

[0103] S200: Simultaneously collect multi-parameter monitoring signals from the first type of monitoring point, the second type of monitoring point, and the third type of monitoring point.

[0104] The collected multi-parameter monitoring signals include at least gas concentration signals, temperature signals, and smoke signals. Gas concentration signals are collected simultaneously at all three types of monitoring points, while temperature and smoke signals are also collected simultaneously at the first and second types of monitoring points.

[0105] By simultaneously acquiring multi-parameter monitoring signals from three types of monitoring points, the consistency of all monitoring data used for subsequent analysis and judgment in the time dimension was ensured, providing a unified time benchmark for subsequent time series analysis, feature calculation, and multi-source evidence fusion. At the same time, through the simultaneous acquisition of multi-parameter signals, comprehensive coverage of various physicochemical characteristic signals in the prodromal phase of thermal runaway was achieved.

[0106] S300: Based on the collected gas concentration signals, it establishes and maintains an adaptive background benchmark for dynamically tracking the environmental background for each gas concentration signal.

[0107] For each gas concentration signal at each monitoring point, a corresponding adaptive background benchmark is established. This adaptive background benchmark can dynamically track the changes in the environmental background value of the corresponding gas concentration according to changes in environmental conditions. Under normal monitoring conditions, the adaptive background benchmark corresponding to each gas concentration signal is continuously updated and maintained according to the preset update rules, so that the adaptive background benchmark can always accurately reflect the real-time environmental background level of the corresponding gas at the corresponding monitoring point and avoid judgment deviations caused by environmental background value drift.

[0108] Specifically, the update formula for the adaptive background benchmark is as follows:

[0109] ;

[0110] Where C(t) is the current sampled value, and C0(t) is the current baseline value. The smoothing coefficient satisfies 0 < α < 1; the smoothing coefficient With sampling period The correspondence between the smoothing time constant and the smoothing time constant is as follows:

[0111] ;

[0112] When the sampling frequency is 1Hz (i.e.) When =1s), the smoothing time constant The preferred value range is 600s to 1200s, corresponding to the smoothing coefficient. The value ranges from 1 / 1200 to 1 / 600.

[0113] By establishing and maintaining an adaptive background benchmark that dynamically tracks the environmental background for each gas concentration signal, subsequent anomaly judgments can be based on the degree of deviation of the gas concentration from the real-time environmental background, rather than on a fixed absolute concentration threshold. This fundamentally eliminates the interference of long-term drift of the gas concentration background value caused by factors such as environmental temperature, humidity, and seasonal changes on early warning judgments. It also solves the problem that existing fixed threshold schemes are prone to false alarms or missed alarms under background drift conditions, ensuring the accuracy and stability of early warning judgments under different environmental conditions.

[0114] Specifically, in some embodiments, establishing and maintaining an adaptive background benchmark for dynamically tracking the environmental background based on the acquired gas concentration signals further includes the following steps:

[0115] S301: The adaptive background benchmark is updated recursively using an exponential moving average algorithm.

[0116] S302: When any of the aforementioned warning levels is triggered, immediately freeze the update of the adaptive background baseline for all relevant gas concentration signals.

[0117] During the early warning triggering process, when it is determined that an early warning of any level 1, 2, or 3 has been triggered, a freeze command is immediately sent to the adaptive background benchmark update module to suspend the recursive update operation of the adaptive background benchmark corresponding to all gas concentration signals related to the triggering of the current early warning. This keeps the adaptive background benchmark values ​​of these gas channels at the values ​​at the time of the early warning trigger. During the duration of the early warning, no matter how the gas concentration sampling value changes, the corresponding adaptive background benchmark value remains frozen and no recursive update operation is performed.

[0118] This step effectively avoids the problem that, during the continuous rise of abnormal concentration, the adaptive background benchmark continues to rise along with the abnormal concentration, causing the deviation of gas concentration from the background benchmark to be gradually offset and the abnormal features to be gradually absorbed. It ensures that during the duration of the warning, the abnormal features can always reflect the true deviation from the environmental background level before the anomaly occurred, and will not decay due to the extension of the warning duration. This ensures the stability of the high-level warning trigger features and avoids the problem of the warning level falling back and the abnormal signal being masked due to the baseline rising.

[0119] S303: When the warning level is lifted, the adaptive background benchmark is reinitialized with the sampled value at the time of lifting and its recursive update is resumed.

[0120] Once the triggered warning level is confirmed to be lifted, the real-time valid sample value of the gas concentration signal corresponding to the warning lifting time is first obtained. This sample value is used as the new initial value of the adaptive background benchmark for the corresponding gas channel to complete the re-initialization of the adaptive background benchmark, replacing the benchmark value frozen at the warning triggering time. After the re-initialization is completed, the exponential moving average recursive update operation of the adaptive background benchmark for the corresponding gas channel is immediately resumed, and the normal monitoring state is entered. The adaptive background benchmark is continuously maintained according to the preset recursive rules.

[0121] By re-initializing and resuming recursive updates, the old baseline value, which was frozen at the time of the warning trigger and had deviated from the current real environmental background level, was avoided. This ensures that after the warning is lifted, the adaptive background baseline can quickly anchor to the current real environmental background level and restore normal dynamic tracking capabilities. The residual old baseline value will not cause deviations in subsequent abnormal feature calculations. This ensures that the system can quickly restore normal monitoring and warning capabilities after the warning is lifted, and avoids subsequent false alarms or missed alarms caused by unreasonable baseline values.

[0122] This embodiment establishes nine independent adaptive baselines based on an adaptive background benchmark, three for each of the three monitoring points. Each baseline operates independently, without interference, for the concentration values ​​of carbon monoxide, volatile organic compounds, and hydrogen. No adaptive baselines are established for the temperature and smoke channels; feature calculations are performed directly using absolute values ​​and increments. The adaptive baselines ensure that early warning judgments are always based on relative increments rather than absolute concentration values, eliminating interference from environmental background drift. Furthermore, the baseline freezing mechanism ensures that the early warning signal does not attenuate due to baseline following during the continuous development of the anomaly, effectively improving the reliability of early warnings during the thermal runaway progression phase.

[0123] S400: Based on each gas concentration signal and its corresponding adaptive background benchmark, calculate the first type of gas anomaly features, the second type of gas anomaly features, and the third type of gas anomaly features.

[0124] Within each sampling period, for each gas concentration signal, three types of gas anomaly features are calculated based on the real-time sampled value of the gas concentration signal and the corresponding adaptive background benchmark. The first type of gas anomaly feature, the second type of gas anomaly feature, and the third type of gas anomaly feature reflect the abnormal state of gas concentration from different signal dimensions. The three types of features complement each other and can cover different evolution patterns of gas concentration in the prodromal period of thermal runaway. After the calculation is completed, the three types of gas anomaly features are stored in the corresponding computing units for subsequent early warning trigger judgment.

[0125] By calculating three different types of gas anomaly features, comprehensive coverage of different evolution modes of gas concentration in the prodromal stage of thermal runaway is achieved. This avoids the shortcomings of single features being unable to adapt to different evolution paths of thermal runaway, enabling the early warning system to effectively perceive various types of gas concentration anomalies, such as explosive, gradual, and cumulative anomalies. At the same time, the calculation of anomaly features based on an adaptive background benchmark ensures that the anomaly features can accurately reflect the true deviation of gas concentration from the environmental background, thus improving the ability to identify early weak anomaly signals.

[0126] In some embodiments, the steps of calculating the first type of gas anomaly features, the second type of gas anomaly features, and the third type of gas anomaly features based on each gas concentration signal and its corresponding adaptive background benchmark include:

[0127] S401: Calculate the first type of gas anomaly feature, which is the gas concentration increment, and the gas concentration increment is the instantaneous difference between the current sampled value and the current adaptive background reference value.

[0128] Within each sampling period, for each gas concentration signal, the effective sampled value of that gas in the current sampling period and the corresponding current adaptive background reference value are acquired. The difference between the current sampled value and the current adaptive background reference value is calculated. This difference represents the first type of gas anomaly feature of that gas in the current sampling period, i.e., the gas concentration increment. The unit is ppm. The calculation formula is as follows:

[0129] .

[0130] After the calculation is completed, the gas concentration increment will be... The data is stored in the corresponding computing unit for subsequent early warning triggering judgment. If the calculated difference is negative, it indicates that the current gas concentration is lower than the environmental background level, and the gas concentration increment... Record it as 0 or keep a negative value, and do not participate in the statistics of positive outliers.

[0131] S402: Calculate the second type of gas anomaly characteristics, where the second type of gas anomaly characteristics are trend characteristics, and the trend characteristics are the rate of change of the first type of gas anomaly characteristics within a preset time period.

[0132] Within each sampling period, for each gas concentration signal, obtain the gas concentration increment for the current sampling period. And the gas concentration increment at the same sampling point before the preset time period. Calculate the difference between the two feature values, and then divide the difference by the preset duration of 60 seconds. The rate of change of the first type of gas anomaly characteristics within a preset time period is obtained; this rate of change is the trend characteristic. When the sampling period is 1 second, this difference is numerically equal to the rate of increase in ppm / min, calculated as follows:

[0133] .

[0134] The gas path exhibits the second type of gas anomaly characteristics in the current sampling period. After calculation, the trend characteristics will be... The data is stored in the corresponding computing unit for subsequent early warning triggering judgment. When the system runtime is less than the preset duration, trend characteristics are analyzed. Recorded as 0, it will not participate in the early warning trigger judgment.

[0135] S403: Calculate the third type of gas anomaly feature, which is a cumulative exposure feature, and the cumulative exposure feature is the sum of all positive first type of gas anomaly features within a preset sliding integral window.

[0136] Within each sampling period, for each gas concentration signal, a sliding integration window of a preset duration (preset duration 120s) is defined, ending at the current sampling time. First-type gas anomaly feature values ​​from all sampling periods within this sliding integration window are extracted. All positive feature values ​​greater than 0 are selected, and these positive feature values ​​are summed. The resulting sum represents the cumulative exposure feature of that gas signal in the current sampling period. It is defined as the cumulative value of the positive concentration increment within a 120-second sliding window ending at the current time, in ppm·s.

[0137] .

[0138] After the calculation is complete, the cumulative exposure features will be calculated. The data is stored in the corresponding computing unit for subsequent early warning triggering judgment. When the system runtime is less than the preset duration of the sliding integral window, the accumulated exposed features are processed. Recorded as 0, it will not participate in the early warning trigger judgment.

[0139] This embodiment calculates the cumulative sum of positive first-type gas anomaly features within a preset sliding integral window as the third-type gas anomaly feature, thereby achieving integral amplification of low-concentration, long-term gas anomalies. It can effectively identify very early leakage anomalies where the gas concentration remains at a low level but continuously deviates from the background value under the ventilation dilution effect. Even if the instantaneous difference feature of a single sample cannot exceed the warning threshold, the long-term cumulative sum can exceed the threshold, enabling the system to effectively capture the precursor signal of chronic leakage-type thermal runaway. This makes up for the shortcomings of the first two types of features in identifying continuous low-concentration anomalies under strong ventilation dilution scenarios.

[0140] S500: Based on the calculated first type of gas anomaly features, second type of gas anomaly features, third type of gas anomaly features, temperature anomaly features, and smoke signals, multi-source evidence fusion is performed within a sliding time window, and the first type of warning level trigger judgment corresponding to the early stage of the risk, the second type of warning level trigger judgment corresponding to the development stage of the risk, and the third type of warning level trigger judgment corresponding to the emergency stage of the risk are executed sequentially.

[0141] Within each sampling period, all valid data within the current sliding time window are acquired, including the first type of gas anomaly features, the second type of gas anomaly features, the third type of gas anomaly features, as well as the temperature anomaly features calculated based on the acquired temperature signals and the acquired smoke signals. The above-mentioned anomaly signals of different dimensions are used as independent anomaly evidence, and multi-source evidence fusion processing is performed within the sliding time window. According to the order of risk level from low to high, the first type of warning level trigger judgment corresponding to the early stage of risk is executed first, the second type of warning level trigger judgment corresponding to the development stage of risk is executed, and finally the third type of warning level trigger judgment corresponding to the emergency stage of risk is executed. The trigger judgment of each warning level is based on the preset trigger rules of the corresponding level. Only when the trigger rules of the corresponding level are met is it determined that the warning level is triggered.

[0142] By fusing multi-source evidence within a sliding time window and combining three types of gas anomaly characteristics, temperature anomaly characteristics, and smoke signals for early warning trigger judgment, cross-validation of multi-dimensional anomaly evidence is achieved, improving the reliability of early warning trigger judgment and avoiding false triggers caused by single parameter anomalies. Through sequential trigger judgment of three levels of early warning, a refined graded response to the entire evolution process of thermal runaway from the early prodromal stage to the development stage and then to the emergency stage is achieved, breaking the limitation of the single threshold binary response in existing technologies. This enables maintenance personnel to take corresponding measures according to different risk levels, ensuring early warning sensitivity while reducing unnecessary emergency shutdown operations.

[0143] In some embodiments, step S500, which involves performing multi-source evidence fusion within a sliding time window based on the calculated first type of gas anomaly features, second type of gas anomaly features, third type of gas anomaly features, temperature anomaly features, and smoke signals, and then executing the first type of early warning level trigger judgment corresponding to the early stage of risk, further includes:

[0144] S501: Within the sliding time window, determine whether the abnormal volatile organic compound concentration characteristics of the first type of monitoring point and the second type of monitoring point meet their preset low-level threshold range. If they do, trigger the first type of warning level.

[0145] A preset sliding time window is defined. Within each sampling period, the first, second, and third types of abnormal characteristic data corresponding to volatile organic compounds (VOCs) at the first and second monitoring points are extracted within the sliding time window. It is then determined whether the abnormal VOC concentration characteristics at both points meet the preset low-level threshold range. That is, at least one of the three types of abnormal characteristics of VOCs at the two points continuously meets the corresponding low-level threshold requirement within the sliding time window. When the abnormal VOC concentration characteristics at both points meet the low-level threshold range, the first triggering condition for the first-level warning is met, and the first-level warning is triggered.

[0146] By using the requirement that the abnormal volatile organic compound (VOC) concentrations at both the first and second monitoring points meet the low-level threshold as one of the triggering conditions for the first-level early warning, this approach fully leverages the characteristic that VOCs are the earliest characteristic gases released during the pre-thermal runaway phase of lithium-ion batteries. Organic solvents in the electrolyte will volatilize first when the battery temperature rises abnormally, releasing VOCs before the pressure relief valve opens. Therefore, this condition enables effective detection of the earliest stage of thermal runaway, providing the system with the earliest warning trigger signal. Simultaneously, requiring that the abnormal VOC concentrations at both the first and second monitoring points meet the threshold effectively eliminates false triggers caused by occasional factors such as single-point sensor drift and local airflow disturbances, ensuring early warning sensitivity while improving the reliability of trigger judgment.

[0147] In some embodiments of this application, the low-level threshold range for volatile organic compounds (VOCs) is: gas concentration increment. The value range is 5ppm-30ppm; trend characteristics The value range is 5ppm / min-30ppm / min; cumulative exposure characteristics The value range is 500ppm·s-2000ppm·s.

[0148] S502: Within the sliding time window, determine whether the abnormal carbon monoxide concentration and abnormal hydrogen concentration characteristics of the first type of monitoring point and the second type of monitoring point meet their preset low-level threshold range. If they do, trigger the first type of warning level.

[0149] Within a preset sliding time window, extract the first, second, and third types of abnormal feature data corresponding to carbon monoxide and hydrogen gases at the first and second types of monitoring points. Determine whether the abnormal carbon monoxide and hydrogen concentration features at the two points both meet the preset low-level threshold range. That is, the abnormal carbon monoxide and hydrogen features at the first type of monitoring point both meet the low-level threshold, and the abnormal carbon monoxide and hydrogen features at the second type of monitoring point also both meet the low-level threshold. When all the above conditions are met, it is determined that the second triggering condition for the first type of warning level is met, and the first type of warning level is triggered.

[0150] By using the low-level threshold values ​​for both carbon monoxide and hydrogen concentrations at the first and second monitoring points as one of the triggering conditions for the first-level warning, the system leverages the characteristic properties of carbon monoxide and hydrogen as gases released during the decomposition of the SEI film inside the battery and the side reactions between the cathode material and the electrolyte. The simultaneous presence of these two gases indicates that electrochemical reactions related to thermal runaway have begun inside the battery. Compared to single-gas characteristics, the simultaneous fulfillment of the threshold requirements by both gases significantly reduces the risk of false triggering caused by noise and drift from single-gas sensors. Furthermore, requiring both gases at two monitoring points to meet the threshold values ​​further enhances the reliability of the trigger judgment, providing a second independent triggering basis for early warning of thermal runaway. This complements the triggering condition for volatile organic compounds, avoiding missed alarms caused by the failure of a single-gas triggering condition.

[0151] In some embodiments of this application, the low-level threshold range for carbon monoxide and hydrogen gases is: gas concentration increment The value range is 5ppm-30ppm; trend characteristics The value range is 5ppm / min-30ppm / min; cumulative exposure characteristics The value range is 500ppm·s-2000ppm·s.

[0152] Threshold ranges for carbon monoxide and hydrogen in gaseous gases: gas concentration increments The value range is 20ppm-60ppm; trend characteristics The value range is 15ppm / min-50ppm / min; cumulative exposure characteristics The value range is 1500ppm·s-5000ppm·s.

[0153] S503: Within the sliding time window, determine whether there is a smoke alarm signal at both the first type of monitoring point and the second type of monitoring point. If the condition is met, trigger the first type of warning level.

[0154] Within a preset sliding time window, smoke signals collected by smoke sensors at the first and second monitoring points are extracted. It is then determined whether the smoke signals at both points meet the preset smoke alarm threshold requirements. That is, the smoke signals at both points continuously reach the smoke alarm threshold within the sliding time window. When there are valid smoke alarm signals at both points, the third triggering condition for the first warning level is met, and the first warning level is triggered.

[0155] By using the presence of smoke alarm signals at both the first and second type of monitoring points as one of the triggering conditions for the first-level early warning, the system leverages the characteristic that smoke is a core feature signal when the battery pressure relief valve opens and aerosol particles are ejected. The appearance of a smoke signal indicates that the battery thermal runaway has progressed to the stage of pressure relief valve opening, which is an emergency early warning signal that requires immediate triggering. This condition provides a direct triggering basis for early warning in scenarios of rapid thermal runaway development, ensuring that the system can trigger an early warning immediately when the battery pressure relief valve opens, without any warning lag. At the same time, requiring the presence of smoke alarm signals at both points effectively eliminates false alarms caused by factors such as single-point smoke sensor failure and dust interference, improving the reliability of this triggering condition.

[0156] In some embodiments, the smoke alarm threshold range is 3000-5000 μg / m³. 3 .

[0157] Furthermore, based on the calculated first type of gas anomaly features, second type of gas anomaly features, third type of gas anomaly features, temperature anomaly features, and smoke signals, the steps of performing multi-source evidence fusion within a sliding time window and executing the second type of warning level trigger judgment corresponding to the risk development stage and the third type of warning level trigger judgment corresponding to the risk emergency stage also include:

[0158] S504: Define a sliding time window for multi-source evidence fusion and determine a list of four independent abnormal evidence categories for triggering the second and third warning levels, including:

[0159] (1) There is the same gas, and the concentration anomalies of the gas at both the first type of monitoring point and the second type of monitoring point meet the preset specific level threshold range;

[0160] (2) There is another gas, and the concentration anomalies of this gas at both the first type of monitoring point and the second type of monitoring point meet the preset specific level threshold range;

[0161] (3) All of them have smoke alarm signals;

[0162] (4) The temperature anomaly characteristics of the first type of monitoring point or the second type of monitoring point meet the preset specific level threshold range.

[0163] First, a preset sliding time window is defined as the time benchmark for triggering the judgment of the second and third warning levels. Within each sampling period, all valid monitoring data and abnormal feature data within the sliding time window are extracted and sorted into four independent types of abnormal evidence. The four types of evidence are independent of each other, and each type of evidence must continuously meet the corresponding validity requirements within the sliding time window to be counted as valid evidence.

[0164] S505: Within the sliding time window, the second type of early warning level trigger judgment corresponding to the risk development stage is executed. When there are at least three types of valid independent abnormal evidence within the sliding time window, and the abnormal gas concentration feature in evidence (1) meets the medium-level threshold range, the abnormal gas concentration feature in evidence (2) meets the low-level threshold range, and the abnormal temperature feature in evidence (4) meets the low-level threshold range, the second type of early warning level is triggered.

[0165] S506: Within the sliding time window, the third type of warning level trigger judgment corresponding to the emergency risk stage is executed. When there are at least three types of valid independent abnormal evidence within the sliding time window, and the abnormal gas concentration feature in evidence (1) meets the high-level threshold range, the abnormal gas concentration feature in evidence (2) meets the medium-level threshold range, and the abnormal temperature feature in evidence (4) meets the high-level threshold range, the third type of warning level is triggered.

[0166] In some embodiments of this application, the low-level threshold range for any gas is: gas concentration increment. The value range is 5ppm-30ppm; trend characteristics The value range is 5ppm / min-30ppm / min; cumulative exposure characteristics The value range is 500ppm·s-2000ppm·s.

[0167] Intermediate threshold range for any gas: gas concentration increment The value range is 20ppm-60ppm; trend characteristics The value range is 15ppm / min-50ppm / min; cumulative exposure characteristics The value range is 1500ppm·s-5000ppm·s.

[0168] High-level threshold range for any gas: gas concentration increment The value range is 40ppm-100ppm; trend characteristics The value range is 30ppm / min-80ppm / min; cumulative exposure characteristics The value range is 4000ppm·s-10000ppm·s.

[0169] Temperature anomaly characteristics include: temperature increment and rate of temperature rise Temperature increment The temperature rise rate is the difference between the current sampled temperature and the reference temperature reset during system initialization or after the most recent warning was cleared. : The average rate of change of temperature increment within a preset sliding window with the current time as the endpoint.

[0170] The low-level threshold range for temperature anomaly characteristics is: temperature increment. The value range is 3℃-5℃; temperature rise rate The value range is 2℃ / 2min-5℃ / 2min.

[0171] The high-level threshold range for temperature anomaly characteristics is: temperature increment. The value range is 5℃-8℃; temperature rise rate The value range is 5℃ / 2min-8℃ / 2min.

[0172] The smoke alarm threshold range is 3000 μg / m³ 3 -5000μg / m 3 .

[0173] Within each sampling period, the validity of four types of abnormal evidence within the sliding time window is statistically analyzed. First, the validity of the first type of evidence is determined, i.e., whether there is the same gas whose concentration anomaly characteristics at both the first and second type of monitoring points meet the medium / high level threshold range. Next, the validity of the second type of evidence is determined, i.e., whether there is another gas different from the first type of evidence whose concentration anomaly characteristics at both points meet the low / medium level threshold range. Then, the validity of the third type of evidence is determined, i.e. whether there are smoke alarm signals at both points. Finally, the validity of the fourth type of evidence is determined, i.e. whether at least one of the two points has temperature anomaly characteristics that meet the low / high level threshold range.

[0174] When the number of valid independent abnormal evidence obtained from statistics is no less than three types, and the requirements of the first type of evidence being at the medium-level threshold, the second type of evidence being at the low-level threshold, and the fourth type of evidence being at the low-level threshold are met simultaneously, the second-level warning level is determined to be triggered.

[0175] When the number of valid independent abnormal evidence obtained from statistics is no less than three categories, and the requirements of the first category of evidence being a high-level threshold, the second category of evidence being a medium-level threshold, and the fourth category of evidence being a high-level threshold are met simultaneously, the third-level warning level is determined to be triggered.

[0176] This embodiment sets at least three independent abnormal evidences that simultaneously meet the requirements as triggering conditions for the second / third level of warning. Combined with cross-validation of multi-source evidence, it significantly improves the reliability of warning triggering during the thermal runaway development stage and avoids high-level false alarms caused by a single parameter anomaly.

[0177] The second warning level corresponds to the risk development stage of thermal runaway, where internal side reactions intensify, multiple characteristic gases are released simultaneously, and the temperature begins to rise significantly. This is achieved by requiring the first gas to meet a medium-level threshold, the second gas to meet a low-level threshold, and the temperature to meet a low-level threshold. The third warning level corresponds to the emergency stage where thermal runaway is about to occur or has already occurred. At this stage, the internal thermal runaway reaction is violent, the concentration of characteristic gases increases sharply, the temperature rises rapidly, and even smoke is released. By requiring at least three types of evidence to be met simultaneously, the system accurately matches the characteristic evolution patterns of the thermal runaway development stage / emergency stage, forming a complete three-level triggering system from the early stage to the development stage and then to the emergency stage, achieving a gradient response to the thermal runaway process.

[0178] In some embodiments of this application, before including any anomalous feature in the statistics of valid evidence, its persistence in meeting the threshold condition must be verified to filter out occasional noise spikes. The stability determination parameters in this embodiment are as follows:

[0179] Gas low-level or medium-level threshold anomalies: Statistical window 20s, minimum effective percentage 70%, that is, at least 14 sampling points in the past 20s (20 sampling points) continuously meet the corresponding threshold conditions to be considered as stable and valid evidence.

[0180] High-level gas threshold anomaly: Statistical window of 10 seconds, minimum effective percentage of 70%, meaning at least 7 sampling points have consistently met the high-level threshold conditions in the past 10 seconds. The shorter statistical window for high-level thresholds is used to accelerate response in emergency situations.

[0181] Temperature anomaly: Statistical window 20s, minimum effective percentage 80%, meaning at least 16 sampling points in the past 20s must meet the temperature threshold condition. The higher effective percentage requirement is because temperature sensors are more sensitive to airflow disturbances; stricter stability requirements effectively filter out temperature misjudgments caused by brief airflow fluctuations.

[0182] S600: After triggering any of the aforementioned warning levels, the abnormal signals that triggered the warning levels are sequentially confirmed for spatial temporal consistency and external disturbance elimination.

[0183] Once the trigger layer judgment is completed and it is determined that a warning of any level of the first, second, or third category has been triggered, the warning signal is not output directly. Instead, two layers of confirmation operations are performed on the abnormal signal that triggered the current warning level. First, spatial temporal consistency confirmation is performed. After the spatial temporal consistency confirmation is completed, external disturbance elimination confirmation is performed. Only after the two layers of confirmation operations are completed in sequence can the subsequent warning output stage be entered.

[0184] In some embodiments, the steps of sequentially verifying the spatial temporal consistency and eliminating external disturbances of the abnormal signal that triggers the warning level specifically include:

[0185] S601: Obtain the triggering gas on which the current warning level is based.

[0186] Once the trigger layer judgment is completed and the warning of any level is determined, all relevant data for the current warning level are retrieved, and the type of gas on which the current warning is based is identified, i.e., the triggering gas.

[0187] If the current warning is triggered by the abnormal characteristics of a single gas meeting the corresponding triggering conditions, then that gas is the triggering gas for this warning; if the current warning is triggered by the abnormal characteristics of multiple gases jointly meeting the corresponding triggering conditions, then all gases involved in triggering the warning are triggering gases for this warning, and the subsequent spatial temporal consistency confirmation process needs to be performed independently for each triggering gas.

[0188] S602: Obtain the moment when the trigger gas first meets its preset low-level threshold range at the second type of monitoring point, and the moment when the trigger gas first meets its preset low-level threshold range at the first type of monitoring point.

[0189] For each identified trigger gas, the historical monitoring data, anomaly characteristic calculation data, and threshold judgment data corresponding to that gas are traced back to determine the moment when the anomaly characteristic of the trigger gas at the second type of monitoring point first passes the preset stability verification rule and meets the preset low-level threshold range. This moment is recorded as the anomaly moment at the second type of monitoring point. At the same time, the moment when the anomaly characteristic of the trigger gas at the first type of monitoring point first passes the preset stability verification rule and meets the preset low-level threshold range is recorded as the anomaly moment at the first type of monitoring point.

[0190] Both timestamps use a unified sampling timestamp system. The second type of monitoring points are located in the sensing area above the battery clusters, while the first type of monitoring points are located in the exhaust convergence area of ​​the energy storage cabinet. The preset low-level threshold range corresponds to the threshold interval for early-stage thermal runaway anomalies, and the preset stability verification rules are verification rules that require the corresponding anomaly evidence to meet the minimum effective percentage requirement within the statistical time window.

[0191] S603: When the abnormal time of the trigger gas at the second type of monitoring point is earlier than the abnormal time at the first type of monitoring point, and the time difference is within a preset reasonable delay range, or when the abnormal time difference between the trigger gas at the first type of monitoring point and the second type of monitoring point is less than a preset synchronization threshold, it is determined that the spatial temporal consistency confirmation is passed.

[0192] Specifically, based on the time obtained in step S602, a spatial temporal consistency verification judgment is made on the trigger gas.

[0193] The type of gas used to trigger the current warning is defined as the triggering gas. The moment when the trigger gas at a second-class monitoring point first meets its preset low-level threshold range; defined The moment when the trigger gas at the first type of monitoring point first meets its preset low-level threshold range.

[0194] For each trigger gas, calculate the time difference between the abnormal time of the first type of monitoring point and the abnormal time of the second type of monitoring point, and perform a judgment on the two timing modes based on this:

[0195] 1. Determine if it is a normal transmission pattern:

[0196] If the anomaly time of the second type of monitoring point is earlier than the anomaly time of the first type of monitoring point, and the time difference between the two satisfies:

[0197] ;

[0198] Then it is determined that the trigger gas conforms to the normal propagation mode.

[0199] 2. If the normal transmission pattern is not met, determine whether it is a strong ventilation synchronous mode:

[0200] If the absolute time difference between the time of anomaly at the first type of monitoring point and the time of anomaly at the second type of monitoring point satisfies:

[0201] ;

[0202] in To synchronize the determination threshold, the value range is 10 to 20 seconds, which is suitable for strong ventilation or rapid gas mixing conditions.

[0203] If the triggering gas meets either of the two timing modes mentioned above, then the triggering gas is deemed to have passed the spatial timing consistency confirmation. If the current warning is triggered jointly by multiple triggering gases, then each triggering gas must pass the spatial timing consistency confirmation individually before the overall spatial timing consistency confirmation can be determined to be successful. If any triggering gas does not meet either of the two timing modes mentioned above, then the spatial timing consistency confirmation is deemed to have failed.

[0204] This embodiment verifies the timing of abnormal trigger gas occurrences using two timing modes that conform to the physical laws of gas propagation under air-cooled conditions. This enables the determination of the physical rationality of the abnormal signal source. From the perspective of gas spatial propagation laws, it verifies whether the abnormal signal originates from actual thermal runaway release within the battery cluster, effectively filtering out abnormal situations that do not conform to the physical laws of gas propagation. The strong ventilation synchronization mode is adapted to scenarios where rapid mixing of gases inside the cabinet under high wind speed and strong ventilation conditions leads to near-synchronous detection of abnormalities at two monitoring points. This avoids incorrectly judging the method as failing due to synchronous anomalies under strong ventilation conditions, ensuring the adaptability of this method under different air-cooled conditions. At the same time, for scenarios where multiple trigger gases trigger together, all trigger gases are required to pass timing verification, filtering out false trigger signals that do not conform to the physical laws of thermal runaway gas propagation.

[0205] S700: After both the spatial temporal consistency confirmation and the external disturbance elimination confirmation are passed, output a warning signal corresponding to the current trigger level.

[0206] Once the abnormal signal triggering the current warning level has passed spatial-temporal consistency confirmation and external disturbance elimination confirmation, a corresponding warning signal is generated according to the determined warning level and output to the battery management system, energy management system, or host computer platform according to the preset communication protocol. Simultaneously, all relevant data triggering the warning, intermediate calculation processes, and confirmation results are output synchronously. If the abnormal signal fails any confirmation, no warning signal is output, and continuous tracking and monitoring continue.

[0207] In some embodiments, after both the spatial temporal consistency confirmation and the external disturbance elimination confirmation are passed, a warning signal corresponding to the current trigger level is output. The specific steps include:

[0208] S701: Obtain the trigger gas on which the current warning level is based.

[0209] After confirming spatial temporal consistency, retrieve the relevant data that triggered the current warning level to determine the type of gas on which the warning was triggered, i.e., the triggering gas. If the warning is triggered by a single gas, then that gas is the triggering gas; if it is triggered by multiple gases in combination, then all the gases involved in the triggering are triggering gases, and the subsequent external disturbance elimination confirmation process must be performed independently for each triggering gas.

[0210] S702: If the trigger gas does not meet the preset level threshold range at the third type of monitoring point, it is determined that the external disturbance elimination confirmation has been passed.

[0211] For each trigger gas, determine whether its abnormal characteristics at the third monitoring point (located in the air inlet area of ​​the energy storage cabinet) meet the preset low-level threshold range (consistent with the trigger warning threshold range). If not, the external disturbance elimination for that trigger gas is considered successful. If multiple trigger gases trigger together, all trigger gases must meet this condition for overall confirmation to be successful.

[0212] S703: If the trigger gas meets the preset low-level threshold range at the third type of monitoring point, then it is further determined whether the abnormal time of the third type of monitoring point is earlier than the abnormal time of the first type of monitoring point or the second type of monitoring point, and whether the cumulative exposure characteristic difference of the trigger gas between the first type of monitoring point, the second type of monitoring point and the third type of monitoring point is less than the preset gradient threshold.

[0213] S704: When the third type of monitoring point anomaly is earlier and the cumulative exposure feature difference is less than the gradient threshold, the current anomaly is determined to conform to the external disturbance mode, and the warning output is suppressed or delayed for confirmation.

[0214] Specifically, to eliminate false triggering caused by external gas pollution or airflow disturbances, a third type of monitoring point is introduced as an external environmental reference point. Definition The moment when the trigger gas at the third type of monitoring point first crosses the low-level threshold for stability confirmation; defined. This represents the maximum cumulative exposure of the gas triggered at point P within the current sliding time window.

[0215] First, conduct an effectiveness check on the third type of monitoring points: If If the gas does not exist (i.e., the third-category monitoring point has not yet passed the preset low-level threshold range), then it is determined that the external disturbance has been eliminated and the gas can pass directly. If it exists, the following two conditions will be combined for determination. When the following conditions are met simultaneously for the same trigger gas, it is determined to be an external disturbance mode:

[0216] (1) The anomalies at the third type of monitoring points appeared earlier than those at the first or second type of monitoring points, that is:

[0217] ;

[0218] (2) The difference in cumulative exposure between the first or second type of monitoring sites and the third type of monitoring sites is not significant, that is, the following conditions are met simultaneously:

[0219] ;

[0220] ;

[0221] in, The differential judgment threshold is set to 60ppm·s-120ppm·s.

[0222] When an alert is triggered by multiple gases in combination, the above-mentioned determination is made for each triggering gas separately.

[0223] If any trigger gas is determined to be in an external disturbance mode, a delayed confirmation process is executed on the current warning output, with the delay duration ranging from 30 to 60 seconds. During the delay period, if the trigger gas at the third-category monitoring point does not meet the low-level threshold conditions, the external disturbance confirmation is deemed invalid, and the gas re-enters the spatial-temporal consistency confirmation process. If the external disturbance determination conditions still hold after the delay period expires, the current warning output is directly suppressed. If all trigger gases are determined to be in an external disturbance mode, the current warning output is directly suppressed. The suppression state will continue until the disturbance determination conditions are no longer met.

[0224] The thermal runaway early warning method for air-cooled energy storage cabinets provided in this application, by setting up three types of functionally heterogeneous monitoring points and calculating gas anomaly characteristics based on an adaptive background benchmark, can effectively capture weak anomaly signals generated in the precursor period of thermal runaway before the effects of air-cooled ventilation dilution occur, thus improving the sensitivity of early warning. Simultaneously, by using a dedicated third type of monitoring point for external disturbance reference and introducing disturbance elimination confirmation and spatial-temporal consistency verification steps after triggering the warning, it can effectively distinguish between genuine internal anomalies of the battery cluster and external environmental interference or sensor noise, significantly reducing the system's false alarm rate while ensuring high sensitivity.

[0225] Based on the above embodiments, the following is a detailed implementation process of the thermal runaway early warning method for air-cooled energy storage cabinet clusters provided in this application.

[0226] Each threshold is within the range defined by the threshold definition in the above embodiments. Based on the operating conditions of this embodiment, the following specific values ​​are selected:

[0227] Low-level gas threshold (corresponding to early, weak anomalies): Take 15 ppm, r C Take 10 ppm / min, AUC 120 Take 800 ppm·s;

[0228] Gas grade threshold (corresponding to significant mid-term anomalies): Take 35 ppm, r C Take 25 ppm / min, AUC 120 Take 2500 ppm·s;

[0229] High-level gas threshold (corresponding to an emergency phase anomaly): Take 70 ppm, r C Take 50 ppm / min, AUC 120 Take 6000 ppm·s;

[0230] Low temperature threshold: Take 4℃, r T Set the temperature to 3℃ / 2min; High-level temperature threshold: Take 6℃, r T Take 6℃ for 2 minutes.

[0231] Taking a typical early thermal runaway scenario as an example: At time t0, overcharging causes slow electrolyte evaporation in a battery cluster, and the VOC concentration at the second monitoring point begins to rise continuously at a rate of approximately 9 ppm / min. At t0+120s, the 120-s integration window completely covers the abnormal rise for the first time, and the AUC of the VOC at the second monitoring point... 120 The calculated value was approximately 1089 ppm·s, exceeding the low-level threshold of 800 ppm·s. Subsequently, the system entered a 20-second stability verification window for the VOC at the second type of monitoring points, continuously evaluating the AUC at each subsequent sampling point. 120 Whether it remains above the threshold; due to continuous VOC release, AUC 120 The concentration remained above 800 ppm·s for the next 20 seconds, with an effective proportion of over 85%, and the stability verification was passed. The abnormal VOC characteristics of the second type of monitoring point were determined to be valid evidence, and the completion time was approximately t0+137s.

[0232] Meanwhile, at the first type of monitoring points, the VOC concentration was partially diluted by the ventilation airflow within the cabinet as the gas diffused from the top of the battery module to the exhaust convergence area, resulting in an actual concentration increase rate approximately 80% (about 7 ppm / min) of that at the second type of monitoring points. Trend characteristics r C The VOC concentration was approximately 7 ppm / min, failing to reach the low-level threshold of 10 ppm / min; however, after accumulation within a 120-second integration window, the AUC of VOC at the first-class monitoring site was... 120At t0+120s, the concentration reached approximately 871 ppm·s, exceeding the low-level threshold of 800 ppm·s. After confirmation within a 20-s stability verification window, the VOC anomaly characteristics at the first-category monitoring site were also deemed valid evidence. This process also confirms the AUC... 120 In scenarios where "concentration increases slowly and the increase is relatively low after dilution," compared to ΔC and r... C Stronger early perception capabilities.

[0233] Around t0+140s, the AUC of VOC at the first and second monitoring points 120 All features have completed low-level threshold stability verification, the judgment condition is met, and a level one warning is triggered. The system immediately freezes all nine baselines to prevent the continuously rising abnormal concentration from being absorbed by the baseline recursion, and enters the confirmation layer to further verify the spatial temporal sequence and disturbance source of this trigger.

[0234] Continuing the aforementioned scenario, the anomaly persisted. At t0+400s, the characteristics of each channel evolved as follows: VOC concentrations at both the first and second monitoring points continued to rise, with ΔC_VOC reaching approximately 60ppm, exceeding the intermediate-level threshold ΔC of 35ppm. AUC... 120 The VOC level was approximately 4300 ppm·s, exceeding the intermediate threshold of 2500 ppm·s. VOC, as gas A, met the intermediate threshold condition and was considered Type I evidence. As internal side reactions intensified, CO gas began to be released around t0+180s, reaching approximately 40 ppm ΔC_CO at t0+400s, exceeding the low-level threshold by 15 ppm. Both Type I and Type II monitoring points passed stability verification and were considered Type II evidence. Simultaneously, battery heat generation caused the cluster temperature to continuously rise, with ΔT approximately 5°C, exceeding the low-level temperature threshold of 4°C. T The temperature rises by approximately 3°C / 2 minutes, reaching the lower limit of the low-level threshold, classifying the temperature anomaly as Category III evidence. If all three types of independent evidence simultaneously pass stability verification within a 60-second sliding time window, the judgment condition is met, and the system upgrades the warning level from Level I to Level II, notifying maintenance personnel to perform power-off and other related operations on the warning battery cluster.

[0235] Continuing with the above scenario, at t0+600s, both VOC and CO concentrations rapidly increased: ΔC_VOC reached approximately 90ppm, exceeding the high-level threshold of 70ppm; ΔC_CO reached approximately 80ppm, also exceeding the high-level threshold of 70ppm; both AUC 120 Both significantly exceeded the high-level threshold of 6000 ppm·s, and were classified as Category I and Category II evidence, respectively. The temperature increment ΔT was approximately 10℃, r T Approximately 7℃ / 2min, exceeding the high-level temperature threshold (ΔT=6℃, r T=6℃ / 2min), which is considered the third type of evidence. At the same time, if the smoke sensor output signal at the first or second type of monitoring point exceeds the alarm threshold, it is considered the fourth type of evidence. If all four types of evidence are met, far exceeding the lower limit of the three types required for judgment, the system issues an emergency shutdown command with the highest priority and activates the fire linkage interface.

[0236] The three-tiered triggering mechanism enables the energy storage operation and maintenance team to respond in stages: Level 1 alerts (approximately t0+140s) trigger manual on-site inspections, at which point the batteries are still in their early stages, providing ample intervention window; Level 2 alerts (approximately t0+420s) trigger remote isolation of affected battery clusters to prevent the spread of anomalies; Level 3 alerts (approximately t0+620s) trigger emergency shutdowns and fire suppression coordination. The complete response window is approximately 8 minutes, providing sufficient time margin for on-site handling. This tiered response strategy ensures safety while avoiding frequent and unnecessary shutdowns due to false alarms, thus improving the availability of the energy storage system.

[0237] Furthermore, in actual industrial and commercial operating environments, besides battery thermal runaway, the following scenarios may also lead to abnormal gas levels inside the cabinet: external vehicle exhaust fumes entering during ventilation, organic solvent evaporation from adjacent equipment, and emissions from surrounding industries. These external disturbances are difficult to distinguish from internal thermal runaway signals on a single-point sensor basis, which is the root cause of the high false alarm rate in existing technologies. This embodiment adds a confirmation layer after the trigger layer passes, based on spatial temporal logic and third-type monitoring point disturbance references. This layer performs spatial tracing and physical rationality verification of the anomaly source, including: spatial temporal consistency confirmation and external disturbance exclusion confirmation.

[0238] like Figure 1 As shown, the spatial temporal consistency confirmation utilizes the physical law of the propagation of gas released from the battery cluster from the second type of monitoring point to the first type of monitoring point with the airflow of the cold air. By examining the temporal relationship between the occurrence of anomalies at the second type of monitoring point and the first type of monitoring point, it is determined whether the source of the anomaly conforms to the intra-cluster propagation mode.

[0239] In practice, the moment when the trigger gas first passes the low-level threshold stability confirmation at the second type of monitoring point is recorded as follows: The moment when the stability of the first type of monitoring point is confirmed by the low-level threshold is This invention defines the following two scenarios for determining whether a decision is successful:

[0240] Scenario 1 – Normal Transmission Pattern: The anomaly at the second type of monitoring point appears before that at the first type of monitoring point, and the time difference is within a reasonable range, i.e., 5s ≤ ( - The minimum time for gas to travel from the top of the battery module to the exhaust vent is 5 seconds (under strong ventilation conditions). The maximum time is 120 seconds (the longest reasonable delay under weak ventilation or extremely low gas concentration conditions). In this specific implementation case, the measured travel time of gas from the second monitoring point to the first monitoring point in the cabinet at the rated wind speed of the axial fan is approximately 12 to 35 seconds, which is within a reasonable range.

[0241] Scenario Two – Strong Ventilation Synchronous Mode: When the fan is running at high speed, the gas inside the cabinet mixes rapidly, and the second-type monitoring point and the first-type monitoring point may show abnormalities almost synchronously. The judgment condition in this case is | - |≤ In this embodiment, =15s. This avoids the situation where, under high wind speed conditions, the occurrence of synchronization is incorrectly judged as a failure of spatial temporal consistency.

[0242] When an alert is triggered by multiple gases, each triggering gas must be independently assessed for spatial-temporal consistency. Only after all gases pass the assessment is the overall spatial-temporal consistency considered acceptable. If any triggering gas fails to meet the conditions for spatial-temporal consistency confirmation, the trigger is deemed not to conform to the intra-cluster propagation law, and the alert output is suppressed. This situation usually indicates sensor failure or algorithm mis-triggered alert.

[0243] External disturbance elimination confirmation utilizes the air intake reference at the third-category monitoring point to determine the direction of the triggering gas source. The prerequisite for external disturbance elimination confirmation is that a valid anomaly exists at the third-category monitoring point within the current time window. If the third type of monitoring point is not abnormal, it is directly determined that the external disturbance is not valid, and the external disturbance elimination confirmation is automatically passed.

[0244] When a valid anomaly is found at a third-category monitoring point, the following joint judgment is made:

[0245] Condition (1) - Timing judgment: The anomaly at the third type of monitoring point occurs at least 10 seconds earlier than that at the first or second type of monitoring point, which satisfies the condition. ≤ -10s or ≤ -10s. The physical meaning of this condition is: if the abnormal gas comes from the external air intake, the third type of monitoring point (located at the air intake) should detect the abnormality earlier than the sensor inside the cabinet; the minimum time difference of 10s is to filter out the timing jitter caused by the difference in sensor response time.

[0246] Condition (2) – Concentration gradient judgment: AUC between the first type of monitoring point, the second type of monitoring point, and the third type of monitoring point 120 The differences are all insignificant, i.e. and .

[0247] In this example, δ = 90 ppm·s is used. The physical meaning of this condition is: if the anomaly comes from inlet air disturbance, the cumulative exposure at the three monitoring points (Type I, Type II, and Type III) should be similar (the disturbing gas is evenly distributed); while if it comes from thermal runaway within the cluster, the cumulative exposure at the two monitoring points (Type II and Type I) should be higher than that at the Type III monitoring point (internal sources accumulate more inside the cabinet).

[0248] Taking a typical false alarm scenario due to external disturbance as an example: The energy storage cabinet is installed in a factory workshop. VOC gas generated by the spraying operation in the workshop enters the air inlet of the energy storage cabinet through the ventilation system. The VOC concentration at the third monitoring point rises first. C ^L is approximately t B Subsequently, the VOC concentrations at the first and second monitoring points inside the cabinet also increased, with the abnormality occurring approximately 30 seconds later than at the third monitoring point (meeting condition (1)). C ^L≤t_B^L -10 s and t C ^L≤t A ^L -10 s). Meanwhile, the AUC of VOCs at the three monitoring points (Category I, Category II, and Category III) was... 120 The maximum difference is within 60 ppm·s and less than δ=90 ppm·s (satisfying condition (2): the cumulative exposure at the three points is similar, which is consistent with the characteristic of uniform distribution of external disturbance gas with the inlet air).

[0249] If both conditions for confirming the elimination of external disturbances are met simultaneously, the current VOC anomaly is determined to be an external disturbance. Since VOC is the only gas triggering the first-level warning and is identified as an external disturbance, the system performs a 45-second delay confirmation on the current warning output. During this delay, spraying operations cease, and the VOC levels at the third-level monitoring points gradually return to normal. Within a 20-second statistical window, the low-level threshold stability condition is no longer met, the external disturbance determination fails, and the system re-performs the spatial-temporal consistency confirmation process for VOCs. Because the VOC levels at the first and second-level monitoring points also decrease synchronously with the dissipation of the external disturbance, the temporal judgment condition is also not met, ultimately suppressing the warning output and effectively preventing false alarms caused by the entry of external industrial VOCs.

[0250] If the external VOC disturbance persists during the delay period, and the concentrations at the three monitoring points (Category I, Category II, and Category III) remain evenly distributed, and both conditions for eliminating the external disturbance are still met after the delay period, the system will continue to suppress the warning output until the disturbance conditions are no longer met.

[0251] Spatial-temporal consistency verification and external disturbance exclusion verification constitute a dual verification mechanism: spatial-temporal consistency verification verifies the physical rationality of the anomaly source from the perspective of spatial propagation laws, while external disturbance exclusion verification excludes external disturbance sources from the perspective of air intake reference. Both layers of verification are indispensable, jointly ensuring that warning signals are output only in cases of genuine intra-cluster anomalies.

[0252] Another aspect of this application provides a thermal runaway early warning system at the cluster level for air-cooled energy storage cabinets, used to implement the aforementioned thermal runaway early warning method at the cluster level for air-cooled energy storage cabinets, referring to... Figure 2 and Figure 3 As shown, system module 100 includes: heterogeneous monitoring layout module 101, signal acquisition and preprocessing module 102, adaptive reference processing module 103, abnormal feature calculation module 104, multi-level early warning triggering module 105, and early warning confirmation module 106.

[0253] The heterogeneous monitoring layout module 101 is set within the battery cluster-level space of the target air-cooled energy storage cabinet, and is equipped with three types of monitoring points with different functional orientations and non-overlapping deployment areas. Figure 3 In the diagram, point A is marked as the first type of monitoring point 1011, point B is marked as the second type of monitoring point 1012, and point C is marked as the third type of monitoring point 1013.

[0254] The first type of monitoring point 1011 is located in the exhaust convergence area of ​​the energy storage cabinet. This area is the convergence outlet for all airflow within the cabinet and can reflect the overall cumulative state of parameters within the entire battery cluster space. The second type of monitoring point 1012 is located in the sensing area above the battery cluster. This area is situated above the direct convection heat transfer zone of the battery module and is the initial accumulation area for gases released during the pre-thermal runaway phase. The third type of monitoring point 1013 is located in the air inlet area of ​​the energy storage cabinet and can reflect the parameter status of the incoming external air in real time, serving as a reference benchmark for external environmental disturbances.

[0255] After the three types of monitoring points are deployed, a hardware connection and signal transmission link are established with the signal acquisition and preprocessing module 102 to ensure synchronous signal transmission.

[0256] The signal acquisition and preprocessing module 102 is connected to the heterogeneous monitoring layout module 101 to synchronously acquire and preprocess monitoring signals.

[0257] The signal acquisition and preprocessing module 102 first synchronously acquires multi-parameter signals from three types of monitoring points at a unified sampling frequency. The acquired signals include at least the gas concentration signals of carbon monoxide, volatile organic compounds, and hydrogen, as well as temperature and smoke signals. Gas concentration signals are acquired from all three types of points; temperature and smoke signals are also acquired from the first and second types of points. Synchronization verification is performed during acquisition to ensure a unified time reference.

[0258] Secondly, range overload detection and sensor fault detection are performed sequentially on the raw signal. Range overload detection determines whether the sampled value exceeds the upper limit of the sensor's rated range. Sensor fault detection determines the sensor's operating status flag. Invalid sampling points are replaced with the previous valid sampled value. When a sensor has more than a preset number of consecutive invalid sampling points, a sensor fault alarm is reported, and this channel does not participate in subsequent processes during the fault period.

[0259] The adaptive reference processing module 103 is used to receive the effective gas concentration signal output by the signal acquisition and preprocessing module 102, and to establish and maintain an adaptive background reference for each gas concentration signal.

[0260] The adaptive reference processing module 103 is used to establish an adaptive background reference for each gas concentration signal at each location, and uses an exponential moving average algorithm for recursive updates. Under normal monitoring conditions, each sampling period calculates and updates the current reference value based on the current sampled value, the reference value of the previous period, and a preset smoothing coefficient, so that it slowly tracks the long-term changes in the environmental background.

[0261] Upon receiving the warning trigger signal sent by the multi-level warning trigger module 105, the recursive update operation of the adaptive background reference of all gas channels related to the warning is immediately suspended, so that the reference value remains at the value at the time of the warning trigger and remains frozen during the duration of the warning.

[0262] Upon receiving the warning cancellation signal, the system acquires the real-time valid sample value of the gas concentration signal corresponding to the warning cancellation time, uses this value as the new initial value, and reinitializes the corresponding adaptive background reference. After completion, the recursive update operation of the channel reference is immediately resumed.

[0263] The anomaly feature calculation module 104 is used to receive the gas concentration signal and temperature signal output by the signal acquisition and preprocessing module 102, as well as the adaptive background reference value output by the adaptive reference processing module 103, and calculate various anomaly features.

[0264] The anomaly feature calculation module 104 is used in each sampling period to calculate three types of gas anomaly features for each gas concentration signal, combining its real-time sampled value with an adaptive background benchmark. These include: First type of gas anomaly feature: the instantaneous difference between the current sampled value and the current adaptive background benchmark value, reflecting the instantaneous deviation of the concentration; Second type of gas anomaly feature: the rate of change of the first type of feature within a preset time period, reflecting the upward trend and rate of the anomaly; Third type of gas anomaly feature: the cumulative sum of all positive first type of gas anomaly features within a preset sliding integral window ending at the current time, reflecting the cumulative exposure degree of the anomaly. When the system runtime is insufficient, the corresponding feature value is recorded as 0.

[0265] Simultaneously, it is also used to calculate two types of temperature anomaly characteristics for the temperature signals collected from the first and second types of monitoring points, including: the first type of temperature anomaly characteristic: the difference (temperature increment) between the current temperature sampling value and the reference temperature reset at the time of system initialization or after the most recent warning was lifted; the second type of temperature anomaly characteristic: the average rate of change (temperature rise rate) of the temperature increment within a preset sliding window with the current time as the endpoint; the temperature signals from the third type of monitoring points are not included in the calculation.

[0266] The multi-level early warning triggering module 105 is used to receive the abnormal features output by the abnormal feature calculation module 104 and the smoke signal output by the signal acquisition and preprocessing module 102, perform multi-source evidence fusion within the sliding time window, and execute graded triggering judgment.

[0267] The multi-level early warning triggering module 105 is used to verify the persistence of any abnormal feature or smoke signal in meeting a threshold condition before classifying it as valid evidence. Specifically, within a preset statistical time window ending at the current time, it determines whether the proportion of the number of sampling points meeting the condition to the total number of sampling points is not less than a preset minimum effective proportion threshold. Only signals that meet this requirement can be counted as valid abnormal evidence.

[0268] Furthermore, based on a predetermined sliding time window, all valid abnormal evidence (including gas anomalies, temperature anomalies, and smoke signals) within the window are extracted in each sampling period, and abnormal signals from different dimensions are cross-validated and fused as independent evidence.

[0269] According to the risk level from low to high, the trigger judgments for the first, second, and third warning levels are executed sequentially. After any level of warning is triggered, a warning trigger signal is immediately sent to the adaptive baseline processing module 103, and relevant data is sent to the warning confirmation module 106 to initiate the confirmation process.

[0270] The early warning confirmation module 106 is used to confirm the abnormal signal after the multi-level early warning triggering module 105 triggers the early warning, and output the early warning signal after the confirmation is successful.

[0271] Specifically, upon receiving the warning trigger signal, the warning confirmation module 106 first performs this confirmation. It obtains the moment when the trigger gas first meets the low-level threshold range at both the second-class and first-class monitoring points, and judges the signal according to two preset valid timing patterns based on this timing relationship. Only signals conforming to the valid timing patterns can pass. If a signal fails, the warning process is terminated.

[0272] This confirmation is performed after the spatial temporal consistency is verified. Based on the abnormal state of the trigger gas at the third type of monitoring point, a judgment is made according to the preset disturbance discrimination logic. Only signals whose external disturbance modes have been excluded can pass. If it is determined to be an external disturbance, the warning output suppression or delay confirmation is performed according to the preset rules.

[0273] Once the abnormal signal triggering the warning passes the above two confirmations in sequence, a corresponding warning signal is generated based on the warning level determined by the multi-level warning triggering module 105, and output to the battery management system, energy management system, or host computer platform according to the preset communication protocol. Simultaneously, all relevant data, intermediate processes, and confirmation results are output. If any confirmation fails, no warning signal is output, and continuous monitoring continues.

[0274] This embodiment primarily addresses three key issues in thermal runaway early warning systems for air-cooled energy storage cabinets and proposes improved solutions. First, early warning sensitivity is insufficient. Due to the dilution effect of air cooling ventilation, sensors typically only detect the average concentration of the mixed gas, resulting in weak signals that are easily affected by background drift, leading to alarm delays. Second, the false alarm rate is high. Existing methods struggle to distinguish whether an increase in concentration originates from internal battery gas generation or external environmental interference. Often, increasing the threshold is used to reduce false alarms, but this further reduces sensitivity. Third, the early warning logic is incomplete. When fusing multiple sensors, the temporal relationships of parameter anomalies and the spatial propagation patterns of gas within the cabinet are not fully considered, resulting in a lack of physical basis for early warning judgments.

[0275] To address the aforementioned issues, the air-cooled energy storage cabinet cluster-level thermal runaway early warning system provided in this embodiment has been optimized in terms of monitoring layout, signal processing, and early warning logic. By placing monitoring points near the gas release source close to the battery cluster, combined with adaptive background subtraction and multi-dimensional feature extraction, the ability to capture early, weak signals is improved. Simultaneously, utilizing a multi-level triggering and dual-confirmation mechanism, stability verification and anomaly source identification are performed before triggering the early warning, effectively distinguishing between internal thermal runaway precursors and external interference, reducing false alarms. Furthermore, the system incorporates the signal timing relationships of sensors at different locations into the analysis based on the spatiotemporal laws of gas diffusion, making the early warning logic more consistent with physical reality and improving reliability under complex operating conditions.

[0276] In some embodiments of this application, reference is made to Figure 3 As shown, the heterogeneous monitoring layout module 101 specifically includes: a first composite detector node, a second composite detector node, and a third composite detector node.

[0277] Among them, the first composite detector node, as the first type of monitoring point 1011, is set in the exhaust gathering area of ​​the energy storage cabinet and integrates sensors for carbon monoxide, volatile organic compounds, hydrogen, temperature and smoke.

[0278] The first composite detector node is the hardware carrier of the first type of monitoring point 1011 in the heterogeneous monitoring layout module 101. It adopts an integrated composite detector structure, with a built-in microcontroller processing unit and five types of sensors, namely carbon monoxide gas sensor, volatile organic compound gas sensor, hydrogen gas sensor, temperature sensor and smoke sensor. The built-in microcontroller processing unit has multiple analog signal acquisition interfaces and digital communication interfaces, which can uniformly collect real-time monitoring data of the above five sensors in each sampling period, and at the same time detect the working status of each sensor in real time. The collected measurement data and the corresponding sensor working status flag are encapsulated into a standard communication message and output to the outside through a preset communication link.

[0279] The first composite detector node is installed in the exhaust convergence area of ​​the energy storage cabinet, specifically on the airflow path of the cabinet's exhaust outlet. This ensures that the sensor's sensing surface can fully contact the airflow converging and exiting the cabinet, enabling complete acquisition of the overall cumulative state of gas, temperature, and smoke parameters within the entire battery cluster space. This node achieves data interconnection with the signal acquisition and preprocessing module through a pre-set communication link, synchronously uploading all acquired data to the signal acquisition and preprocessing module 102.

[0280] The second composite detector node, serving as the second type of monitoring point 1012, is located in the sensing area above the battery cluster and integrates sensors for carbon monoxide, volatile organic compounds, hydrogen, temperature, and smoke.

[0281] The second composite detector node is the hardware carrier of the second type of monitoring point 1012 in the heterogeneous monitoring layout module 101. It adopts the same integrated composite detector structure as the first composite detector node, with a built-in microcontroller processing unit and five types of sensors, namely carbon monoxide gas sensor, volatile organic compound gas sensor, hydrogen gas sensor, temperature sensor and smoke sensor. Its hardware configuration, data acquisition logic and communication message format are completely consistent with the first composite detector node, ensuring that the data collected by the two nodes are completely comparable and time-synchronized.

[0282] The second composite detector node is installed in the sensing area above the battery cluster, specifically above the direct convection heat exchange zone of the battery modules at the top of the battery cluster. It maintains a preset, reasonable distance from the top of the battery cluster to ensure that abnormal gas leaking from the battery modules can first contact the sensor surface of this node, without excessive interference from airflow in other areas of the cabinet. This allows it to detect weak abnormal signals in the lead-up to thermal runaway before the abnormal gas is significantly diluted by ventilation airflow. This node interconnects with the signal acquisition and preprocessing module via a preset communication link, synchronously uploading all acquired data to the signal acquisition and preprocessing module 102, maintaining time synchronization with the data from the first composite detector node.

[0283] The third composite detector node, serving as the third type of monitoring point 1013, is located in the air inlet area of ​​the energy storage cabinet and integrates sensors for carbon monoxide, volatile organic compounds, and hydrogen.

[0284] The third composite detector node is the hardware carrier of the third type of monitoring point 1013 in the heterogeneous monitoring layout module 101. It adopts an integrated composite detector structure, with a built-in microcontroller processing unit and three types of gas sensors: carbon monoxide gas sensor, volatile organic compound gas sensor, and hydrogen gas sensor. The model, range, and resolution of the three types of gas sensors are completely consistent with the corresponding gas sensors in the first and second composite detector nodes, ensuring that the gas concentration monitoring data of the three nodes are completely comparable. The hardware configuration, data acquisition logic, and communication message format of the built-in microcontroller processing unit are consistent with those of the first and second composite detector nodes, ensuring that the data acquisition time of the three nodes is completely synchronized.

[0285] The third composite detector node is installed in the air inlet area of ​​the energy storage cabinet, specifically inside the air inlet grille, on the necessary path for outside air to enter the cabinet. This ensures that the sensor's sensing surface can fully contact the incoming outside air, accurately collecting the real-time gas concentration status of the incoming air without interference from airflow in the battery cluster area inside the cabinet. This node achieves data interconnection with the signal acquisition and preprocessing module 102 via a pre-set communication link, synchronously uploading the collected gas concentration data to the signal acquisition and preprocessing module 102, maintaining time synchronization with the data from the first and second composite detector nodes. This node does not have temperature or smoke sensors; it only collects three types of gas concentration data, specifically for reference judgment of external environmental disturbances.

[0286] Continue to refer to Figure 3 As shown, this embodiment takes a commercial or industrial energy storage cabinet with a rated capacity of 193kWh and forced ventilation using a top axial flow fan as an example. The cabinet integrates one battery cluster, and the battery chemistry is lithium iron phosphate (LFP). A set of three sensor nodes is configured for this battery cluster to form the cluster-level multi-point monitoring system of this invention.

[0287] This invention utilizes a microcontroller and sensors to construct a composite thermal runaway detector. Each composite detector incorporates five types of sensors: a CO sensor (range 0–5000ppm, resolution 1ppm), a VOC sensor (range 0–1000ppm, resolution 1ppm), an H2 sensor (range 0–2000ppm, resolution 1ppm), an NTC thermistor temperature sensor (range -20–80℃, accuracy ±1℃), and a smoke sensor (range 0–10000µg / m³). The built-in microcontroller is a 32-bit microcontroller with a main frequency of no less than 72MHz, multiple ADC inputs, and a CAN communication interface. It uniformly collects real-time monitoring data from the five sensors in each sampling cycle, while simultaneously monitoring the operating status of each sensor in real time. After data acquisition, the measurement data from each sensor, along with the corresponding operating status flags, is encapsulated into a standard CAN message and output externally for direct use by the edge computing unit in the validity verification stage.

[0288] Three composite detector nodes are installed at their respective monitoring locations. Each node is interconnected with the embedded edge computing unit inside the cabinet via a CAN bus, forming a local sensing network within the cabinet. The CAN message transmission period for each node is uniformly set to 1 second (i.e., a sampling frequency of 1 Hz). The anomaly feature calculation module 104 uses an embedded processor with a main frequency of no less than 200 MHz and RAM of no less than 512 KB. After receiving messages from the three nodes in each cycle, it completes real-time early warning calculations according to the thermal runaway early warning method provided in this application for air-cooled energy storage cabinet clusters. This includes 15-channel signal validity verification, 9 EMA baseline recursion, 33 anomaly feature calculations, and multi-level trigger logic judgments. The early warning results are then reported to the energy storage system BMS / EMS via the CAN bus or Ethernet interface. The BMS / EMS then executes corresponding protection actions or maintenance alarms based on the early warning level.

[0289] In this embodiment, the node's built-in microcontroller is a 32-bit microcontroller with a main frequency of no less than 72MHz, multiple ADC inputs, and a CAN communication interface, meeting the real-time requirement of completing the acquisition and message encapsulation of 5 sensor signals within a 1s sampling period. The anomaly feature calculation module 104 has a main frequency of no less than 200MHz and RAM of no less than 512KB, meeting the real-time requirements of completing the validity verification of 15 signals, the recursion of 9 EMA baselines, the calculation of 33 anomaly features, and the multi-level trigger logic judgment within a 1s sampling period.

[0290] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the disclosure in the specification and the embodiments. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein.

Claims

1. A method for early warning of thermal runaway at the cluster level in an air-cooled energy storage cabinet, characterized in that, Includes the following steps: Three types of monitoring points with heterogeneous functions are set up in the energy storage cabinet, including the first type of monitoring point for overall status judgment, the second type of monitoring point for early perception and joint verification, and the third type of monitoring point for external disturbance reference. Among them, the first type of monitoring points are set in the exhaust air collection area of ​​the energy storage cabinet; The second type of monitoring point is set in the sensing area above the battery cluster; The third type of monitoring point is set in the air inlet area of ​​the energy storage cabinet; The first type of monitoring point, the second type of monitoring point and the third type of monitoring point are all equipped with gas sensors for monitoring the concentration of carbon monoxide, volatile organic compounds and hydrogen. The first type of monitoring point and the second type of monitoring point are also equipped with temperature sensors and smoke sensors. Simultaneously collect multi-parameter monitoring signals from the first type of monitoring points, the second type of monitoring points, and the third type of monitoring points; Based on the collected gas concentration signals, an adaptive background benchmark for dynamically tracking the environmental background is established and maintained for each gas concentration signal; Based on each gas concentration signal and its corresponding adaptive background benchmark, a first type of gas anomaly feature is calculated. The first type of gas anomaly feature is the gas concentration increment, which is the instantaneous difference between the current sampled value and the current adaptive background benchmark value. Calculate the second type of gas anomaly characteristics, which are trend characteristics, and the trend characteristics are the rate of change of the first type of gas anomaly characteristics within a preset time period; Calculate the third type of gas anomaly feature, which is a cumulative exposure feature, and the cumulative exposure feature is the sum of all positive first type of gas anomaly features within a preset sliding integral window; Based on the calculated first type of gas anomaly features, second type of gas anomaly features, third type of gas anomaly features, temperature anomaly features, and smoke signals, multi-source evidence fusion is performed within a sliding time window, and the first type of warning level trigger judgment corresponding to the early stage of the risk, the second type of warning level trigger judgment corresponding to the development stage of the risk, and the third type of warning level trigger judgment corresponding to the emergency stage of the risk are executed sequentially. After any of the aforementioned warning levels is triggered, the abnormal signals that triggered the warning levels are sequentially subjected to spatial temporal consistency confirmation and external disturbance elimination confirmation. The timing consistency verification includes the following steps: Obtain the trigger gas on which the current warning level is based; The moment when the trigger gas first meets its preset low-level threshold range at the second type of monitoring point, and the moment when the trigger gas first meets its preset low-level threshold range at the first type of monitoring point; When the abnormal time of the trigger gas at the second type of monitoring point is earlier than the abnormal time at the first type of monitoring point, and the time difference is within a preset reasonable delay range, or when the abnormal time difference between the trigger gas at the first type of monitoring point and the second type of monitoring point is less than a preset synchronization threshold, the spatial temporal consistency is determined to be confirmed. The confirmation of external disturbance elimination includes the following steps: Obtain the trigger gas on which the current warning level is based; If the trigger gas does not meet its preset level threshold range at the third type of monitoring point, the external disturbance elimination confirmation is deemed successful. If the trigger gas meets its preset low-level threshold range at the third type of monitoring point, it is further determined whether the abnormal time of the third type of monitoring point is earlier than the abnormal time of the first type of monitoring point or the second type of monitoring point, and whether the cumulative exposure characteristic difference of the trigger gas between the first type of monitoring point, the second type of monitoring point and the third type of monitoring point is less than a preset gradient threshold. When the third type of monitoring point anomaly is earlier and the cumulative exposure feature difference is less than the gradient threshold, the current anomaly is determined to conform to the external disturbance mode, and the early warning output is suppressed or delayed for confirmation. After both the spatial temporal consistency confirmation and the external disturbance elimination confirmation are passed, an early warning signal corresponding to the current trigger level is output.

2. The method for early warning of thermal runaway at the cluster level of air-cooled energy storage cabinet according to claim 1, characterized in that, Based on the collected gas concentration signals, an adaptive background benchmark for dynamically tracking the environmental background is established and maintained for each gas concentration signal, including: The adaptive background benchmark is updated recursively using an exponential moving average algorithm. When any of the aforementioned warning levels is triggered, the updates of the adaptive background baseline for all relevant gas concentration signals are immediately frozen; Once the warning level is lifted, the adaptive background benchmark is reinitialized using the sampled value at the time of lifting, and its recursive update is resumed.

3. The method for early warning of thermal runaway at the cluster level of air-cooled energy storage cabinet according to claim 1, characterized in that, Based on the calculated first type of gas anomaly features, second type of gas anomaly features, third type of gas anomaly features, temperature anomaly features, and smoke signals, multi-source evidence fusion is performed within a sliding time window, and the first type of early warning level trigger judgment corresponding to the early stage of risk is executed sequentially, including: When the abnormal volatile organic compound concentration characteristics of both the first type of monitoring point and the second type of monitoring point meet their preset low-level threshold range. Alternatively, the abnormal carbon monoxide concentration and abnormal hydrogen concentration characteristics of the first type of monitoring points and the second type of monitoring points both meet their preset low-level threshold range. Alternatively, if smoke alarm signals are present at both the first type of monitoring point and the second type of monitoring point, the first type of warning level will be triggered.

4. The method for early warning of thermal runaway at the cluster level of air-cooled energy storage cabinet according to claim 1, characterized in that, The process of fusing multi-source evidence within a sliding time window, and sequentially executing the second type of early warning level trigger judgment corresponding to the risk development stage and the third type of early warning level trigger judgment corresponding to the risk emergency stage, includes: The corresponding level of warning is triggered based on whether, within the sliding time window, the first type of monitoring points and the second type of monitoring points possess at least three independent types of abnormal evidence selected from the following list: (1) There is the same gas, and the concentration anomalies of the gas at both the first type of monitoring point and the second type of monitoring point meet the preset specific level threshold range; (2) There is another gas, and the concentration anomalies of this gas at both the first type of monitoring point and the second type of monitoring point meet the preset specific level threshold range; (3) All of them have smoke alarm signals; (4) The temperature anomaly characteristics of the first type of monitoring point or the second type of monitoring point meet the preset specific level threshold range; in, The conditions for triggering the second type of warning level are: the abnormal gas concentration characteristics in evidence (1) meet the medium level threshold range, the abnormal gas concentration characteristics in evidence (2) meet the low level threshold range, and the abnormal temperature characteristics in evidence (4) meet the low level threshold range. The conditions for triggering the third type of warning level are: the abnormal gas concentration feature in evidence (1) meets the high-level threshold range, the abnormal gas concentration feature in evidence (2) meets the medium-level threshold range, and the abnormal temperature feature in evidence (4) meets the high-level threshold range.

5. A thermal runaway early warning system for air-cooled energy storage cabinet clusters, used to implement the thermal runaway early warning method for air-cooled energy storage cabinet clusters as described in any one of claims 1-4, characterized in that, include: The heterogeneous monitoring layout module is configured to deploy three types of heterogeneous monitoring points, including a first type of monitoring point, a second type of monitoring point, and a third type of monitoring point, within the energy storage cabinet. The signal acquisition and preprocessing module is configured to synchronously acquire multi-parameter monitoring signals from various monitoring points in the heterogeneous monitoring layout module, and to verify the validity of the acquired signals. The adaptive reference processing module is configured to establish and maintain an adaptive background reference for each gas concentration signal based on the effective gas concentration signal output by the signal acquisition and preprocessing module, and to dynamically track the environmental background. The anomaly feature calculation module is configured to calculate the first type of gas anomaly features, the second type of gas anomaly features, and the third type of gas anomaly features based on the background reference and gas concentration signals output by the adaptive reference processing module, and to calculate the temperature anomaly features. The multi-level early warning triggering module is configured to perform multi-source evidence fusion within a sliding time window based on various abnormal features and smoke signals output by the abnormal feature calculation module, and sequentially execute the triggering judgment of the first type of early warning level, the second type of early warning level, and the third type of early warning level. The early warning confirmation module is configured to, after the multi-level early warning triggering module triggers an early warning, sequentially perform spatial temporal consistency confirmation and external disturbance elimination confirmation on the abnormal signal that triggered the early warning, and output an early warning signal after both confirmations are passed.

6. The thermal runaway early warning system at the cluster level of air-cooled energy storage cabinet according to claim 5, characterized in that, The heterogeneous monitoring layout module includes: The first composite detector node, as the first type of monitoring point, is set in the exhaust air gathering area of ​​the energy storage cabinet and integrates sensors for carbon monoxide, volatile organic compounds, hydrogen, temperature and smoke. The second composite detector node, as the second type of monitoring point, is set in the sensing area above the battery cluster and integrates sensors for carbon monoxide, volatile organic compounds, hydrogen, temperature and smoke. The third composite detector node, as the third type of monitoring point, is set in the air inlet area of ​​the energy storage cabinet and integrates carbon monoxide, volatile organic compounds and hydrogen sensors.

Images