Energy storage battery thermal runaway early warning method and device based on multi-source information fusion

Abstract

The present application relates to the technical field of electrochemical energy storage safety, and particularly relates to a method and device for early warning of thermal runaway of energy storage batteries based on multi-source information fusion. The method comprises: constructing a multi-source parameter fusion feature parameter set, wherein the feature parameters in the multi-source parameter fusion feature parameter set include gas concentration, thermal parameters, electrical parameters and state parameters; constructing a fusion identification model based on the multi-source parameter fusion feature parameter set; and realizing early warning of thermal runaway of energy storage batteries based on multi-source information fusion based on the fusion identification model. By fusing four types of feature parameters, namely, gas, thermal, electrical and state parameters, a hierarchical early warning model is constructed, early identification and hierarchical response of thermal runaway precursors are realized, the false alarm rate is reduced, the warning reliability is improved, and the model is suitable for complex marine environments.

Description

Technical Field

[0001] This invention relates to the field of electrochemical energy storage safety technology, and in particular to an early warning method and device for thermal runaway of energy storage batteries based on multi-source information fusion. Background Technology

[0002] With the large-scale grid connection of new energy power generation systems, lithium-ion battery energy storage systems are increasingly widely used in grid frequency regulation, new energy consumption, and emergency power supply. However, when lithium-ion batteries experience thermal runaway, it is often accompanied by a sharp rise in temperature, release of flammable gases, and even fire and explosion, seriously threatening the safety of equipment and personnel.

[0003] Currently, the early warning methods commonly used in energy storage power stations are still mainly based on single temperature thresholds or smoke detection. Temperature detection is greatly affected by thermal inertia and sensor placement, making it difficult to effectively identify early signs of thermal runaway; smoke detection often lags behind gas release, failing to meet the need for "early warning." Some studies have attempted to introduce a single gas (such as carbon monoxide) as a warning signal, but this suffers from high false alarm rates and a lack of graded response under complex operating conditions.

[0004] Especially for offshore wind power platforms, their energy storage systems face harsh conditions such as high salt spray, high humidity, vibration and shock, limited ventilation, and remote monitoring. Existing early warning methods are insufficient to achieve the engineering goals of "early identification, low false alarms, and linkage." Therefore, there is an urgent need to develop an early warning method for thermal runaway based on multi-source information fusion, with hierarchical early warning capabilities, and applicable to the offshore environment. Summary of the Invention

[0005] To address the above problems, this invention provides a method and device for early warning of thermal runaway in energy storage batteries based on multi-source information fusion. The technical solution provided by this invention is as follows: This invention provides an early warning method for thermal runaway of energy storage batteries based on multi-source information fusion, the method comprising: A feature parameter set for multi-source parameter fusion is constructed, wherein the feature parameters in the feature parameter set for multi-source parameter fusion include gas concentration, thermal parameters, electrical parameters, and state parameters; A fusion recognition model is constructed based on the feature parameter set fused from multiple sources. Based on the aforementioned fusion identification model, early warning of thermal runaway in energy storage batteries is achieved through multi-source information fusion.

[0006] Furthermore, the gas concentration includes one or more of the following: hydrogen concentration, carbon monoxide concentration, methane concentration, ethane concentration, hydrogen fluoride concentration, and volatile organic compound concentration; and / or, Thermal parameters include cell surface temperature, battery pack internal temperature, ambient temperature, and cell surface temperature change rate; and / or, Electrical parameters include individual cell voltage, battery pack voltage, charging and discharging current, insulation resistance, and internal resistance variation; and / or, Status parameters include changes in battery mass, operating status of the liquid cooling system, status of the exhaust system, and smoke concentration.

[0007] Furthermore, based on the feature parameter set fused from multi-source parameters, a fusion recognition model is constructed, including: Each input feature parameter is defined as a low-risk fuzzy set, a medium-risk fuzzy set, and a high-risk fuzzy set, and the fuzzy risk membership degree is obtained by defuzzification using the centroid method; among them, the input feature parameters include gas concentration and thermal parameters; The input window length and prediction step size are set according to the thermal response characteristics of the equipment, and a time series sample dataset is constructed based on the feature parameter set fused from multiple sources. Based on the aforementioned time-series sample dataset, a long and short time-series neural network model is constructed. Based on the aforementioned long and short time-series neural network model, the probability and confidence level of thermal runaway are obtained; A fusion identification model is constructed by weighting and fusing the fuzzy risk membership degree, thermal runaway probability and confidence level.

[0008] Furthermore, the Mamdani fuzzy inference method is used to define each input parameter as a low-risk fuzzy set, a medium-risk fuzzy set, and a high-risk fuzzy set to obtain a fuzzy rule base; among which, the input parameters include gas concentration and thermal parameters; The fuzzy risk membership degree is obtained by defuzzifying the fuzzy rule base using the centroid method. R fuzzy ;in, When one or more of the input feature parameters are high-risk, the fuzzy risk membership degree... R fuzzy High risk; When none of the input feature parameters are high-risk, and one or more of the input feature parameters are medium-risk, then the fuzzy risk membership degree... R fuzzy Medium risk; When all input feature parameters are low risk, the fuzzy risk membership degree R fuzzy Low risk.

[0009] Furthermore, when the gas concentration is greater than or equal to the first threshold, the gas concentration is considered high-risk. When the gas concentration is less than the first threshold and greater than or equal to the second threshold, the gas concentration is considered medium risk. When the gas concentration is less than the second threshold, the gas concentration is considered low risk. When a thermal parameter is greater than or equal to the third threshold, the thermal parameter is considered high-risk. When a thermal parameter is less than the third threshold and greater than or equal to the fourth threshold, the thermal parameter is considered to be of medium risk. When the thermal parameter is less than the fourth threshold, the thermal parameter is considered low-risk.

[0010] Furthermore, the construction of the time-series sample dataset includes: Based on the thermal response characteristics of the equipment, the input window length and prediction step size are set. Based on the feature parameter set of multi-source parameter fusion, a continuous time series is extracted as the input sequence through a sliding window method. The time series sample dataset is generated in chronological order. The input sequence includes gas concentration, thermal parameters, electrical parameters and state parameters within a preset time period.

[0011] Furthermore, based on the aforementioned time-series sample dataset, a long and short time-series neural network model is constructed, including: The initial model of the long and short time series neural network model is trained based on the time series sample dataset; The initial model of the trained long and short time series neural network model was validated based on historical thermal runaway test data, and the long and short time series neural network model was obtained.

[0012] Furthermore, the expression for the fusion recognition model is:

[0013] In the formula, R is the comprehensive risk index; α , β For dynamic weighting coefficients, satisfying α + β =1; R fuzzy To define the membership degree of fuzzy risk; P nn This represents the probability of thermal runaway. C nn , where is the confidence level.

[0014] Furthermore, based on the aforementioned fusion recognition model, early warning of thermal runaway in energy storage batteries is achieved through multi-source information fusion, including: When the concentration of one of the gases rises by more than 30% above the baseline value or the temperature rise rate of the cell surface exceeds the set threshold, a level one warning is triggered, which includes increasing the sampling frequency, recording abnormal data, and remote alarm. When the characteristic parameters fused from two or more multi-source parameters simultaneously meet the first-level warning conditions and the comprehensive risk index exceeds the fifth threshold, a second-level warning is triggered, and power-limited operation, charging and discharging are stopped, and the explosion-proof ventilation system is put on standby. When the probability of thermal runaway is higher than the set value and the comprehensive risk index exceeds the sixth threshold, a level 3 warning is triggered, pack-level fire suppression spraying is executed, audible and visual alarms are activated, and the compartment-level total flooding fire suppression is initiated.

[0015] A multi-source information fusion-based early warning device for thermal runaway of energy storage batteries is also provided, the device comprising: The first construction unit is used to construct a feature parameter set for multi-source parameter fusion, wherein the feature parameters for multi-source parameter fusion in the feature parameter set include gas concentration, thermal parameters, electrical parameters and state parameters; The second building unit is used to construct a fusion recognition model based on the feature parameter set fused from multiple sources. The early warning unit is used to realize early warning of thermal runaway of energy storage battery based on the fusion identification model and multi-source information fusion.

[0016] The technical effects and advantages of this invention are as follows: This invention, centered on the core objectives of "early identification, low false alarm, interoperability, and adaptability to the marine environment," constructs a complete early warning technology system for thermal runaway in energy storage batteries. Its key technologies are reflected in the following four aspects: 1) Multi-source parameter fusion. A feature parameter set is constructed for four categories of parameters: gas, thermal, electrical, and state, overcoming the limitations of single-parameter early warning and providing a complete information foundation for subsequent fusion identification. 2) Near-source detection and multi-level deployment. Gas and temperature sensors are deployed 5-10 cm from the outlet of the battery module cell pressure relief valve to achieve near-source detection within the pack; simultaneously, cabin-level detectors are deployed on the top of the energy storage compartment and at the air duct outlet, forming a "near-source + global" dual-layer detection architecture to improve early signal capture capabilities. 3) Fuzzy logic and neural network hybrid identification model. A fuzzy logic layer is used to quantify expert rules and threshold boundaries, while a lightweight temporal neural network layer is used to learn and classify the nonlinear temporal features of multi-source parameters. A comprehensive risk index is obtained through weighted fusion. R 4) Three-level early warning and graded linkage mechanism. Based on the changing trends of the comprehensive risk index and key parameters, a three-level early warning system is set up: attention level, risk level, and critical level, to achieve graded response and avoid the crude mode of "triggering a full response".

[0017] This invention is applicable to the identification and graded early warning of thermal runaway risks in energy storage compartments under high humidity, high salt spray, space-constrained, and remote monitoring scenarios. By integrating four types of characteristic parameters—gas, thermal, electrical, and state—this invention constructs a graded early warning model, enabling early identification and graded response to precursors of thermal runaway, reducing false alarm rates, improving early warning reliability, and adapting to complex marine environments.

[0018] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures pointed out in the description and the drawings. Attached Figure Description

[0019] Figure 1 This is a flowchart of an early warning method for thermal runaway of energy storage batteries based on multi-source information fusion, provided in an embodiment of this application. Figure 2 This is a schematic diagram of the structure of the multi-source parameter fusion early warning device provided in the embodiments of this application; Figure 3 This application provides a schematic diagram of long and short temporal neural networks in its embodiments; Figure 4 A flowchart illustrating the three-level early warning logic for energy storage thermal runaway provided in this application embodiment; Figure 5 A diagram of an early warning device for thermal runaway of an energy storage battery based on multi-source information fusion, provided in an embodiment of this application; Figure 6 Electronic device structural block diagram according to embodiments of this application. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Definitions of abbreviations and key terms: BMS: Battery Management System, responsible for battery status monitoring, safety protection, and equalization management.

[0022] Modele: A battery module is a standardized intermediate unit with a unified electrical interface, formed by integrating multiple battery cells through series and parallel connections and equipped with structural components, connectors, and management devices.

[0023] Pack: A battery pack is composed of one or more modules and is equipped with a battery management system, a thermal management system, and a structural housing.

[0024] VOC: Volatile Organic Compounds, organic gases released during battery thermal runaway.

[0025] SOC: State of Charge, which reflects the percentage of remaining battery capacity.

[0026] R The comprehensive risk index, a multi-source fusion early warning index constructed in this invention, is used to characterize the risk level of thermal runaway.

[0027] LSTM: Long Short-Term Memory, a deep learning model suitable for processing time-series data.

[0028] GRU: Gated Recurrent Unit, a simplified variant of LSTM.

[0029] To address the shortcomings of existing technologies, this invention discloses an early warning method for thermal runaway in energy storage batteries based on multi-source information fusion, such as... Figure 1 and Figure 2 As shown, the method includes, Step 1: Construct a feature parameter set for multi-source parameter fusion. The feature parameters in the multi-source parameter fusion feature parameter set include gas concentration, thermal parameters, electrical parameters, and state parameters; the gas concentration includes one or more of the following: hydrogen concentration, carbon monoxide concentration, methane concentration, ethane concentration, hydrogen fluoride concentration, and volatile organic compound concentration; and / or, Thermal parameters include cell surface temperature, battery pack internal temperature, ambient temperature, and cell surface temperature rise rate; and / or, Electrical parameters include individual cell voltage, battery pack voltage, charging and discharging current, insulation resistance, and internal resistance variation; and / or, Status parameters include changes in battery mass, operating status of the liquid cooling system, status of the exhaust system, and smoke concentration.

[0030] Step 2: Construct a fusion recognition model based on the feature parameter set fused from multiple sources; Step 3: Based on the fusion identification model, realize early warning of thermal runaway of energy storage battery through multi-source information fusion.

[0031] In one specific embodiment of the present invention, combined with Figure 2 The steps for constructing a feature parameter set for multi-source parameter fusion are as follows: Miniature gas sensor arrays and patch thermocouples are arranged near the cell pressure relief valve or exhaust path of each battery module in the battery pack to collect characteristic gas concentrations, cell surface temperature, internal temperature of the battery pack, and electrical parameters during the initial stage of thermal runaway. A cabin-level gas detector, smoke detector, and temperature detector are installed on the top of the energy storage compartment and at the air duct outlet to collect ambient temperature and status parameters.

[0032] For example, in view of the characteristics of high humidity, high salt spray and limited space of offshore platforms, a two-layer deployment scheme of "pack-level near source detection + cabin-level global detection" is adopted: (1) Pack-level near source detection: micro gas sensor arrays (including CO, H2 and CH4 sensors) and patch thermocouples are arranged near the cell pressure relief valve or exhaust path of each battery module in the battery pack to capture characteristic gases and local temperature rise signals in the early stage of thermal runaway; (2) Cabin-level global detection: cabin-level gas detectors, smoke detectors and temperature detectors are arranged on the top of the energy storage cabin and the air duct outlet for global signal monitoring and redundancy verification.

[0033] All sensor signals are synchronized to the central controller via the data acquisition module. The normal sampling frequency is 1Hz, which automatically increases to 10Hz after a Level 1 warning. The acquired data is processed using a Kalman filter algorithm for noise reduction to improve data accuracy and reliability.

[0034] This invention employs a near-source detection and multi-level deployment scheme. A miniature gas sensor array and patch thermocouples are arranged within a 5-10cm range of the battery module's pressure relief valve outlet to achieve near-source detection within the pack. Simultaneously, cabin-level gas detectors, smoke detectors, and temperature detectors are arranged on the top of the energy storage compartment and at the air duct outlet, forming a dual-layer detection architecture of "near-source + global".

[0035] A multi-source parameter fusion early warning parameter system was constructed. Four types of information—gas concentration (H2, CO, CH4, C2H6, HF, VOC), thermal parameters (cell temperature, pack temperature, ambient temperature, temperature rise rate), electrical parameters (cell voltage, pack voltage, charging / discharging current, insulation resistance), and state parameters (mass change, liquid cooling status, exhaust status, smoke detection status)—were combined into a unified feature space, serving as the basic input for subsequent fusion and identification.

[0036] In a specific embodiment of the present invention, for step 2, constructing a fusion recognition model based on the feature parameter set fused from multiple sources includes: Step 201: Define each input parameter as a low-risk fuzzy set, a medium-risk fuzzy set, and a high-risk fuzzy set, and obtain the fuzzy risk membership degree by defuzzifying using the centroid method; wherein, the input parameters include gas concentration and thermal parameters, specifically: Step 2011: Using the Mamdani fuzzy inference method, each input parameter is defined as a low-risk fuzzy set, a medium-risk fuzzy set, and a high-risk fuzzy set to obtain a fuzzy rule base; among which, the input parameters include gas concentration and thermal parameters; when the gas concentration is greater than or equal to the first threshold, the gas concentration is considered high-risk. When the gas concentration is less than the first threshold and greater than or equal to the second threshold, the gas concentration is considered medium risk. When the gas concentration is less than the second threshold, the gas concentration is considered low risk. When a thermal parameter is greater than or equal to the third threshold, the thermal parameter is considered high-risk. When a thermal parameter is less than the third threshold and greater than or equal to the fourth threshold, the thermal parameter is considered to be of medium risk. When the thermal parameter is less than the fourth threshold, the thermal parameter is considered low-risk.

[0037] Step 2012: Calculate the fuzzy risk membership degree by defuzzifying the fuzzy rule base using the centroid method. R fuzzy ;in, When one or more of the input parameters are high-risk, the fuzzy risk membership degree... R fuzzy High risk; When none of the input parameters are high-risk, and one or more of the input parameters are medium-risk, then the fuzzy risk membership degree... R fuzzy Medium risk; When all input parameters are low risk, the fuzzy risk membership degree is... R fuzzy Low risk.

[0038] For example, the fuzzy logic layer is used to transform expert experience and engineering thresholds into a computable mathematical model, ensuring the interpretability and auditability of the decision-making process. Three fuzzy sets are defined for each input parameter: low risk (L), medium risk (M), and high risk (H). Taking CO concentration as an example, its membership function adopts a combination of trapezoidal and triangular forms. Based on the thermal runaway evolution mechanism and expert experience, a fuzzy rule base is constructed using the Mamdani fuzzy inference method. The matching degree of the rule antecedents is calculated through an AND operation (taking the smaller value), and the outputs of each rule are aggregated through an OR operation (taking the larger value). Finally, the fuzzy risk membership degree is obtained by defuzzifying using the centroid method. R fuzzyTypical rules are as follows: 1) Rule 1: If CO concentration is high risk AND temperature rise rate is medium risk, then thermal runaway risk is high risk; 2) Rule 2: If CO concentration is high risk AND H2 concentration is high risk, then thermal runaway risk is high risk; 3) Rule 3: If CO concentration is medium risk AND temperature rise rate is low risk, then thermal runaway risk is medium risk; 4) Rule 4: If single-cell voltage deviation is high risk AND temperature rise rate is medium risk, then thermal runaway risk is high risk; 5) Rule 5: If all parameters are low risk, then thermal runaway risk is low risk. For example, fuzzy risk membership is a qualitative level: low risk (0-0.3, can be 0.2), medium risk (0.4-0.7, can be 0.5), high risk (0.8-1.0, can be 0.9). Step 202: Set the input window length and prediction step size based on the device's thermal response characteristics. Construct a time-series sample dataset based on the feature parameter set fused from multiple sources, including: Based on the thermal response characteristics of the equipment, the input window length and prediction step size are set. Based on the feature parameter set of multi-source parameter fusion, a continuous time series is extracted as the input sequence through a sliding window method. The time series sample dataset is generated in chronological order. The input sequence includes gas concentration, thermal parameters, electrical parameters and state parameters within a preset time period.

[0039] Step 203: Based on the aforementioned time-series sample dataset, construct a long and short-term neural network model, including: The initial model of the long and short time series neural network model is trained based on the time series sample dataset; The initial model of the trained long and short time series neural network model was validated based on historical thermal runaway test data, and the long and short time series neural network model was obtained.

[0040] Step 204: Based on the aforementioned long and short time-series neural network model, obtain the probability and confidence level of thermal runaway; For example, such as Figure 3 As shown, the neural network layer is used to learn the nonlinear temporal features of multi-source parameters, improving the recognition accuracy under complex working conditions. The network adopts a two-layer LSTM structure. The network parameters are configured as follows: 1) Input layer, time window T=10 steps, feature dimension N=12; 2) LSTM layer 1: 64 hidden units, dropout=0.2, returns the sequence; 3) LSTM layer 2: 64 hidden units, dropout=0.2, returns the final state; 4) Fully connected layer: 64→32, ReLU activation; 5) Output layer: 2 neurons, Sigmoid activation, outputs the probability of thermal runaway. P nn With confidence level Cnn The network employs a strategy that combines offline pre-training with online fine-tuning: supervised training is conducted using historical thermal runaway test data and normal operating condition data; during actual operation, an active learning strategy is used to add high-confidence normal samples and manually confirmed abnormal samples to the incremental training set, and the model is fine-tuned periodically.

[0041] Step 205: Weight and fuse the fuzzy risk membership degree, thermal runaway probability, and confidence level to construct a fusion identification model. The expression of the fusion identification model is:

[0042] In the formula, R is the comprehensive risk index; α , β For dynamic weighting coefficients, satisfying α + β =1; R fuzzy To define the membership degree of fuzzy risk; P nn This represents the probability of thermal runaway. C nn , where is the confidence level.

[0043] For example, the weighting coefficients are adaptively adjusted based on data quality and operating conditions: when data quality is high and operating conditions are typical, the neural network weights... β A higher value, such as 0.6, improves accuracy; when the data is noisy or missing, or under extreme conditions, the fuzzy logic weight... α A higher value, such as 0.8, is used to ensure stability; in case of sensor failure or communication interruption, the system relies entirely on fuzzy logic. α =1.0, β =0.0, maintain basic early warning function.

[0044] This invention constructs a hybrid recognition model combining fuzzy logic and neural networks. A fuzzy logic layer quantifies expert rules and threshold boundaries, while a lightweight temporal neural network layer learns and classifies the nonlinear temporal features of multi-source parameters. A comprehensive risk index is obtained through weighted fusion. R This serves as a unified decision-making basis for tiered early warning systems.

[0045] In a specific embodiment of the present invention, step 3: realizing early warning of thermal runaway of energy storage battery based on the fusion identification model, includes: When the concentration of one of the characteristic gases rises by more than 30% above the baseline value or the temperature rise rate of the cell surface exceeds the set threshold, a first-level warning is triggered, which includes increasing the sampling frequency, recording abnormal data, and remote alarm. When the characteristic parameters fused from two or more multi-source parameters simultaneously meet the first-level warning conditions and the comprehensive risk index R exceeds the fifth threshold, a second-level warning is triggered, and power-limited operation, charging and discharging are stopped, and the explosion-proof ventilation system is put on standby. When the probability of thermal runaway is higher than the set value and the comprehensive risk index exceeds the sixth threshold, a level 3 warning is triggered, pack-level fire suppression spraying is executed, audible and visual alarms are activated, and the compartment-level total flooding fire suppression is initiated.

[0046] For example, such as Figure 4 As shown, three warning levels are set based on the comprehensive risk index R and the changing trends of key parameters. 1) Level 1 Warning (Attention Level): Triggered when the concentration of a single characteristic gas rises by more than 30% above the baseline value or the temperature rise rate exceeds the set threshold. The system automatically increases the sampling frequency, records abnormal data, issues remote alarms, and retains a manual confirmation window. 2) Level 2 Warning (Risk Level): Triggered when two or more parameters (such as rising CO concentration + abnormal temperature rise rate on the cell surface) simultaneously meet the warning conditions, and the comprehensive risk index R exceeds the first threshold (0.5). The system automatically performs power-limited operation, cuts off power to unnecessary loads, stops charging and discharging the energy storage system, and puts the explosion-proof ventilation system on standby. 3) Level 3 Warning (Critical Level): Triggered when the probability of thermal runaway is determined to be higher than the set value by the comprehensive assessment of multiple source parameters (gas, temperature, voltage, smoke detection), and R exceeds the second threshold (0.7). The system automatically starts Pack-level fire suppression spraying, links audible and visual alarms, remotely sends emergency signals, and initiates compartment-level total flooding fire preparation.

[0047] Output the warning level, trigger parameters, key characteristic values, confidence level and suggested actions; send the warning results to the energy storage compartment controller EMU and the site-level EMS in a standardized interface format to achieve linkage with power control, fire protection system and ventilation system.

[0048] Preferably, the gas sensor array uses an electrochemical sensor for CO and H2 detection, a non-dispersive infrared sensor for CO2 and CH4 detection, and a semiconductor sensor for VOC detection. The sensor housing has a protection rating of not less than IP65 and is resistant to salt spray and moisture.

[0049] Preferably, the near-source gas sensor is arranged within 5-10 cm of the battery module near the cell pressure relief valve or exhaust path to ensure that the characteristic gas can be captured in the early stage of thermal runaway.

[0050] Preferably, after the thermal runaway early warning model is triggered at level three, it automatically records multi-source data for 30 seconds before and after the warning to form a traceable evidence package for post-event analysis and model optimization.

[0051] This invention establishes a three-level early warning and tiered linkage mechanism. Based on the comprehensive risk index R and the changing trends of key parameters, three early warning levels are set. Level 1 early warning: triggered by a single anomaly, executing high sampling and remote alarm; Level 2 early warning: triggered by two or more types of parameter anomalies and... R If the first threshold is exceeded, power limiting, power outage preparation, and explosion-proof ventilation standby will be implemented; Level 3 warning: multi-source comprehensive judgment and R If the second threshold is exceeded, execute Pack-level fire suppression spraying and total flooding fire suppression linkage.

[0052] This invention establishes a traceable evidence chain and an engineering acceptance mechanism. During the early warning process, trigger parameters, key feature values, confidence levels, and suggested actions are automatically recorded, forming a traceable evidence package that includes the entire chain of "input data—fusion results—early warning level—linked output," which is used for post-event analysis and model optimization, meeting the requirements of network security and operation and maintenance auditing.

[0053] This invention, centered on the core objectives of "early identification, low false alarm, interoperability, and adaptability to the marine environment," constructs a complete early warning technology system for thermal runaway in energy storage batteries. Its key technologies are reflected in the following four aspects: 1) Multi-source parameter fusion. A feature parameter set is constructed for four categories of parameters: gas, thermal, electrical, and state, overcoming the limitations of single-parameter early warning and providing a complete information foundation for subsequent fusion identification. 2) Near-source detection and multi-level deployment. Gas and temperature sensors are deployed 5-10 cm from the outlet of the battery module cell pressure relief valve to achieve near-source detection within the pack; simultaneously, cabin-level detectors are deployed on the top of the energy storage compartment and at the air duct outlet, forming a "near-source + global" dual-layer detection architecture to improve early signal capture capabilities. 3) Fuzzy logic and neural network hybrid identification model. A fuzzy logic layer is used to quantify expert rules and threshold boundaries, while a lightweight temporal neural network layer is used to learn and classify the nonlinear temporal features of multi-source parameters. A comprehensive risk index is obtained through weighted fusion. R 4) Three-level early warning and graded linkage mechanism. Based on the changing trends of the comprehensive risk index and key parameters, a three-level early warning system is set up: attention level, risk level, and critical level, to achieve graded response and avoid the crude mode of "triggering a full response".

[0054] This invention also provides an early warning device for thermal runaway of energy storage batteries based on multi-source information fusion, such as... Figure 5 As shown, the device includes, The first construction unit is used to construct a feature parameter set for multi-source parameter fusion, wherein the feature parameters for multi-source parameter fusion in the feature parameter set include gas concentration, thermal parameters, electrical parameters and state parameters; The second building unit is used to construct a fusion recognition model based on the feature parameter set fused from multiple sources. The early warning unit is used to realize early warning of thermal runaway of energy storage battery based on the fusion identification model and multi-source information fusion.

[0055] Regarding the apparatus in the above embodiments, the specific manner in which each unit performs its operation has been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0056] Based on the above disclosure, the present invention also provides an electronic device. For example... Figure 6 As shown, the electronic device of this disclosure includes at least one processor electrically connected to the present invention and at least one memory electrically connected to the processor, wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the method steps as executed by the controller above.

[0057] An embodiment of the present invention also provides a storable medium storing computer instructions, which, when executed by a processor, are specifically executed according to the steps in the method described in the above embodiment.

[0058] An embodiment of the present invention also provides a computer program product, including computer instructions, which, when executed by a processor, specifically follow the steps in the method described in the above embodiment.

[0059] The technical solution of the present invention will be further described below with reference to specific embodiments.

[0060] This embodiment takes the energy storage cabin of an offshore wind power platform as the application scenario. The energy storage system consists of 4 battery clusters, each cluster containing 4 battery packs. Each battery pack consists of 8 battery modules connected in series, but each battery module consists of 13 314Ah lithium iron phosphate cells connected in series.

[0061] (1) Sensor deployment. Install a miniature gas sensor array (including CO, H2, CH4 sensors) and a patch thermocouple at 8cm from the outlet of the cell pressure relief valve of each battery module; arrange cabin-level CO detectors, smoke detectors and temperature detectors at the four corners of the top of the energy storage compartment; connect all sensor signals to the central controller, with a normal sampling frequency of 1Hz, which is increased to 10Hz after the first-level warning.

[0062] (2) Multi-source parameter acquisition and fusion. Continuously acquire data from various sensors and extract characteristic parameters: CO concentration change rate, H2 concentration peak value, temperature rise rate, voltage consistency deviation, liquid cooling system operating status, etc. Input the fuzzy logic-neural network fusion model and output a comprehensive risk index. R .

[0063] (3) Early warning and linkage. In an overcharge test, the CO concentration began to rise continuously 7 minutes before thermal runaway occurred, triggering a level one early warning. The system increased the sampling frequency and issued a remote alarm. 4 minutes later, the CO concentration continued to rise, and the temperature rise rate exceeded 0.5℃ / s. The voltage fluctuated, triggering a level two early warning. The system automatically limited power and started explosion-proof ventilation. 1 minute before thermal runaway, the multi-source parameters were comprehensively judged to trigger a level three early warning. Pack-level perfluorohexanone spraying was automatically started, successfully suppressing the spread of thermal runaway.

[0064] (4) Effectiveness verification. Test records show that the method of the present invention identifies risks about 7 minutes earlier than the traditional temperature threshold warning. There were no missed or false alarms throughout the process, and the key equipment in the energy storage cabin was not damaged, which verifies the effectiveness of the method.

[0065] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for early warning of thermal runaway in energy storage batteries based on multi-source information fusion, characterized in that, The method includes, A feature parameter set for multi-source parameter fusion is constructed, wherein the feature parameters in the feature parameter set for multi-source parameter fusion include gas concentration, thermal parameters, electrical parameters, and state parameters; A fusion recognition model is constructed based on the feature parameter set fused from multiple sources. Based on the aforementioned fusion identification model, early warning of thermal runaway in energy storage batteries is achieved through multi-source information fusion.

2. The method for early warning of thermal runaway of energy storage batteries based on multi-source information fusion according to claim 1, characterized in that, Gas concentrations include one or more of the following: hydrogen concentration, carbon monoxide concentration, methane concentration, ethane concentration, hydrogen fluoride concentration, and volatile organic compound concentration; and / or, Thermal parameters include cell surface temperature, battery pack internal temperature, ambient temperature, and cell surface temperature change rate; and / or, Electrical parameters include individual cell voltage, battery pack voltage, charging and discharging current, insulation resistance, and internal resistance variation; and / or, Status parameters include changes in battery mass, operating status of the liquid cooling system, status of the exhaust system, and smoke concentration.

3. The method for early warning of thermal runaway of energy storage batteries based on multi-source information fusion according to claim 1, characterized in that, Based on the feature parameter set fused from multiple sources, a fusion recognition model is constructed, including: Each input feature parameter is defined as a low-risk fuzzy set, a medium-risk fuzzy set, and a high-risk fuzzy set, and the fuzzy risk membership degree is obtained by defuzzification using the centroid method; among them, the input feature parameters include gas concentration and thermal parameters; The input window length and prediction step size are set according to the thermal response characteristics of the equipment, and a time series sample dataset is constructed based on the feature parameter set fused from multiple sources. Based on the aforementioned time-series sample dataset, a long and short time-series neural network model is constructed. Based on the aforementioned long and short time-series neural network model, the probability and confidence level of thermal runaway are obtained; A fusion identification model is constructed by weighting and fusing the fuzzy risk membership degree, thermal runaway probability and confidence level.

4. The method for early warning of thermal runaway of energy storage batteries based on multi-source information fusion according to claim 3, characterized in that, The Mamdani fuzzy inference method is used to define each input parameter as a low-risk fuzzy set, a medium-risk fuzzy set, and a high-risk fuzzy set to obtain a fuzzy rule base; among which, the input parameters include gas concentration and thermal parameters; The fuzzy risk membership degree is obtained by defuzzifying the fuzzy rule base using the centroid method. R fuzzy ;in, When one or more of the input feature parameters are high-risk, the fuzzy risk membership degree... R fuzzy High risk; When none of the input feature parameters are high-risk, and one or more of the input feature parameters are medium-risk, then the fuzzy risk membership degree... R fuzzy Medium risk; When all input feature parameters are low risk, the fuzzy risk membership degree R fuzzy Low risk.

5. The method for early warning of thermal runaway of energy storage batteries based on multi-source information fusion according to claim 4, characterized in that, When the gas concentration is greater than or equal to the first threshold, the gas concentration is considered high-risk. When the gas concentration is less than the first threshold and greater than or equal to the second threshold, the gas concentration is considered medium risk. When the gas concentration is less than the second threshold, the gas concentration is considered low risk. When a thermal parameter is greater than or equal to the third threshold, the thermal parameter is considered high-risk. When a thermal parameter is less than the third threshold and greater than or equal to the fourth threshold, the thermal parameter is considered to be of medium risk. When the thermal parameter is less than the fourth threshold, the thermal parameter is considered low-risk.

6. The method for early warning of thermal runaway of energy storage batteries based on multi-source information fusion according to claim 3, characterized in that, The construction of the time-series sample dataset includes: Based on the thermal response characteristics of the equipment, the input window length and prediction step size are set. Based on the feature parameter set of multi-source parameter fusion, a continuous time series is extracted as the input sequence through a sliding window method. The time series sample dataset is generated in chronological order. The input sequence includes gas concentration, thermal parameters, electrical parameters and state parameters within a preset time period.

7. The method for early warning of thermal runaway of energy storage batteries based on multi-source information fusion according to claim 6, characterized in that, Based on the aforementioned time-series sample dataset, a long and short time-series neural network model is constructed, including: The initial model of the long and short time series neural network model is trained based on the time series sample dataset; The initial model of the trained long and short time series neural network model was validated based on historical thermal runaway test data, and the long and short time series neural network model was obtained.

8. The method for early warning of thermal runaway of energy storage batteries based on multi-source information fusion according to claim 6, characterized in that, The expression for the fusion recognition model is: In the formula, R is the comprehensive risk index; α , β For dynamic weighting coefficients, satisfying α + β =1; R fuzzy To define the membership degree of fuzzy risk; P nn This represents the probability of thermal runaway. C nn , where is the confidence level.

9. The method for early warning of thermal runaway of energy storage batteries based on multi-source information fusion according to claim 8, characterized in that, Based on the aforementioned fusion recognition model, early warning of thermal runaway in energy storage batteries is achieved through multi-source information fusion, including: When the concentration of one of the gases rises by more than 30% above the baseline value or the temperature rise rate of the cell surface exceeds the set threshold, a level one warning is triggered, which includes increasing the sampling frequency, recording abnormal data, and remote alarm. When the characteristic parameters fused from two or more multi-source parameters simultaneously meet the first-level warning conditions and the comprehensive risk index exceeds the fifth threshold, a second-level warning is triggered, and power-limited operation, charging and discharging are stopped, and the explosion-proof ventilation system is put on standby. When the probability of thermal runaway is higher than the set value and the comprehensive risk index exceeds the sixth threshold, a level 3 warning is triggered, pack-level fire suppression spraying is executed, audible and visual alarms are activated, and the compartment-level total flooding fire suppression is initiated.

10. An early warning device for thermal runaway of an energy storage battery based on multi-source information fusion, characterized in that, The device includes, The first construction unit is used to construct a feature parameter set for multi-source parameter fusion, wherein the feature parameters for multi-source parameter fusion in the feature parameter set include gas concentration, thermal parameters, electrical parameters and state parameters; The second building unit is used to construct a fusion recognition model based on the feature parameter set fused from multiple sources. The early warning unit is used to realize early warning of thermal runaway of energy storage battery based on the fusion identification model and multi-source information fusion.

Images