Self-adaptive multi-parameter lithium ion battery thermal runaway early warning method and system

Abstract

The invention discloses a self-adaptive multi-parameter lithium ion battery thermal runaway early warning method and system, and relates to the technical field of energy storage system safety monitoring. Multi-parameter collaborative acquisition is realized by deploying temperature, voltage, internal stress and gas sensors; de-noising the signal by adopting self-adaptive weighted moving average filtering, and dynamically adjusting the length of a filtering window according to local abrupt change; constructing a comprehensive feature vector, carrying out standardization and principal component analysis dimension reduction, and extracting a key risk factor; generating a time-varying risk score R (t) through dynamic weighted fusion; the early warning threshold is adaptively corrected by combining the current SOC and SOH, and a three-level grading response mechanism is established: the sampling frequency is increased by primary early warning, liquid cooling and current limiting are started by intermediate early warning, a main circuit is cut off by advanced early warning, and a fire extinguishing system is triggered, so that closed-loop safety protection from sensing to disposal is realized.

Description

Technical Field

[0001] This application relates to the field of energy storage system safety monitoring technology, specifically to an adaptive multi-parameter lithium-ion battery thermal runaway early warning method and system. Background Technology

[0002] In current large-scale electrochemical energy storage systems, lithium-ion batteries are widely used in peak shaving and frequency regulation, renewable energy grid connection, and microgrid systems due to their high energy density and long cycle life. To ensure operational safety, existing technologies generally employ multi-level monitoring strategies based on BMS (Battery Management System), which collect parameters such as cell voltage, module temperature, and ambient temperature and humidity to achieve real-time sensing of battery status. Some systems introduce gas detection modules, using CO or hydrogen concentration as an early criterion for thermal runaway, and combining threshold comparison and logical judgment to implement alarm functions. In addition, some high-end applications are equipped with distributed fiber optic temperature measurement systems to obtain the battery surface temperature distribution; simultaneously, passive responses are implemented through fixed current limiting control, liquid cooling start-stop, and other means, initiating cooling or shutdown operations upon detecting anomalies. These technologies have improved the safety of energy storage systems to a certain extent and constitute the basic architecture of current battery safety management.

[0003] However, the aforementioned existing technologies still have significant drawbacks: on the one hand, relying solely on single or limited-dimensional parameters (such as voltage and temperature) makes it difficult to comprehensively capture the multimodal characteristics of the early stages of thermal runaway, resulting in delayed warning times; on the other hand, data processing often employs fixed window filtering and static threshold judgment, which cannot effectively suppress noise interference and has poor generalization ability under different SOC and SOH conditions, easily leading to false alarms or missed alarms. Furthermore, traditional early warning models lack the collection and fusion of key physical parameters (such as changes in internal stress), failing to reflect early mechanical degradation processes such as SEI membrane rupture and gas production expansion; and the fixed multi-level response mechanism is not dynamically correlated with the risk level, making precise intervention difficult. Summary of the Invention

[0004] This application provides an adaptive multi-parameter lithium-ion battery thermal runaway early warning method and system, addressing the technical problems of existing lithium-ion battery thermal runaway early warning technologies, such as high signal noise, single feature extraction, poor model generalization ability leading to low early identification sensitivity, high false alarm and false negative rates, and difficulty in adapting to different battery systems and operating conditions. This method can integrate multi-source heterogeneous information, possesses adaptive signal processing capabilities, supports dynamic threshold adjustment based on battery state, and achieves hierarchical closed-loop response, thereby improving the safety and reliability of energy storage systems under complex operating conditions.

[0005] To achieve the above objectives, the technical solution adopted in this application is as follows: Firstly, this application provides an adaptive multi-parameter lithium-ion battery thermal runaway early warning method, including: Real-time acquisition of multi-parameter state characteristics of the battery, including temperature, single cell voltage, internal stress changes, and CO and H2 concentrations inside the chamber; An adaptive weighted moving average filtering algorithm is used to denoise the multi-parameter state features. The weight coefficients of the adaptive weighted moving average filtering algorithm are dynamically adjusted according to the time decay factor and local signal fluctuations, and the length of the filtering window is automatically adjusted according to the degree of signal change. The filtered data is used to construct a comprehensive feature vector containing mean temperature, temperature variance, voltage change rate, internal stress amplitude, and gas concentration. The comprehensive feature vector is then standardized to eliminate the influence of dimensions. After standardization, principal component analysis (PCA) is used to reduce the dimensionality of the high-dimensional features. Based on the dimensionality-reduced comprehensive features, a dynamic weighted fusion mechanism is constructed to generate a comprehensive risk score R(t) that evolves over time. Based on the current battery state of charge (SOC) and state of health (SOH), the preset basic threshold is dynamically corrected to form a three-level warning threshold that changes with the battery operating conditions. When R(t) exceeds the first-level threshold, a primary warning is triggered and the sampling frequency is increased. When it exceeds the second-level threshold, an intermediate response is initiated to control the liquid cooling system and the charging and discharging current. When it exceeds the third-level threshold, an advanced response is executed, the main circuit is cut off, and the fire extinguishing device is activated.

[0006] As a further improvement of this application, the real-time acquisition of the multi-parameter state characteristics of the battery is obtained by building a multi-parameter data acquisition system that includes temperature, voltage, internal stress and gas concentration monitoring. Specifically, the battery surface temperature is monitored by distributed optical fiber temperature measurement, the voltage signal is acquired by isolated ADC, the battery internal stress signal is acquired by MEMS piezoresistive sensor, and the characteristic gas concentration is acquired by CO sensor and H2 sensor.

[0007] As a further improvement to this application, the step of using an adaptive weighted moving average filtering algorithm to denoise multi-parameter state features includes: An adaptive weighted moving average filtering algorithm is used to process the collected temperature, voltage, internal stress, and gas concentration data. The algorithm dynamically adjusts the weights to balance noise suppression and signal response speed requirements. Specifically, the processing involves:

[0008] in, The original data value at time i. N To adjust the sliding window size, These are the weighting coefficients. This is the filtered output value; Among them, weight The adaptive strategy is a hybrid weighting strategy, which dynamically allocates weights based on the signal's local stability and timeliness, specifically as follows:

[0009] in, This is the attenuation coefficient, with a value ranging from 0.1 to 0.3; It is a local mean; The sensitivity coefficient is denoted as .

[0010] Window size N Expressed as:

[0011] in, This is the threshold for sudden changes, which is set according to different battery systems and operating conditions.

[0012] As a further improvement to this application, the step of constructing a comprehensive feature vector from the filtered data, including mean temperature, temperature variance, voltage change rate, internal stress amplitude, and gas concentration, includes: By combining the filtered temperature, voltage, internal stress, and gas concentration signals, a comprehensive feature vector is constructed:

[0013] in, The average temperature within the window. The temperature variance within the window. This represents the rate of change of voltage sag. For battery internal stress, For gas concentration, This is a comprehensive feature vector.

[0014] As a further improvement to this application, the standardization process for the comprehensive feature vector to eliminate the influence of dimensions includes:

[0015] in, The filtered first i 1 eigenvector For feature vectors The average value under normal battery operating conditions; For feature vectors Standard deviation under normal battery operating conditions; This is the standardized feature vector.

[0016] As a further improvement to this application, the dimensionality reduction of high-dimensional features after the standardization process using principal component analysis (PCA) includes:

[0017] in, The principal component matrix is ​​obtained from eigenvalue decomposition. k×m, k The dimension after dimensionality reduction; The standardized feature vector has the following dimensions. m×1 , m The total number of features; The feature vectors after dimensionality reduction. k×1 , k m .

[0018] As a further improvement to this application, the step of constructing a dynamic weighted fusion mechanism based on the dimensionality-reduced comprehensive features to generate a comprehensive risk score R(t) that evolves over time includes:

[0019] in, For the dimensionality reduction of the first i Each principal component characteristic; The weighting coefficients are dynamic and adjusted according to different operating conditions and battery types to meet the requirements. ; To integrate risk scoring.

[0020] As a further improvement to this application, the three-level early warning threshold that varies with battery operating conditions is based on a dynamic risk score. and warning threshold Specifically: when At this time, as a primary warning, the system sampling frequency is increased; when At this time, a medium-level warning is issued, the liquid cooling system operates at full power and the charging and discharging current is limited; when At that time, a high-level warning is issued, the main circuit is cut off, and the fire extinguishing system is triggered.

[0021] As a further improvement to this application, the preset basic threshold is dynamically corrected based on the current battery state of charge (SOC) and state of health (SOH) to obtain the warning threshold. Specifically:

[0022] in, Basic threshold; This is the SOC correction factor; The correction factor for the state of health (SOH); This represents the current health status, with a value ranging from 0.8 to 1.0.

[0023] Secondly, this application provides an adaptive multi-parameter lithium-ion battery thermal runaway early warning system, based on the aforementioned adaptive multi-parameter lithium-ion battery thermal runaway early warning method, comprising: The multi-parameter characteristic signal acquisition module is used to acquire multi-parameter state characteristics of the battery in real time, including temperature, single cell voltage, internal stress change and CO and H2 concentration in the chamber. The adaptive moving average filtering module is used to denoise multi-parameter state features using an adaptive weighted moving average filtering algorithm. The weight coefficients of the adaptive weighted moving average filtering algorithm are dynamically adjusted according to the time decay factor and local signal fluctuations, and the length of the filtering window is automatically adjusted according to the degree of signal change. The multimodal feature fusion module is used to construct a comprehensive feature vector from the filtered data, which includes the mean temperature, temperature variance, voltage change rate, internal stress amplitude, and gas concentration. The comprehensive feature vector is standardized to eliminate the influence of dimensions. After standardization, principal component analysis (PCA) is used to reduce the dimensionality of the high-dimensional features. Based on the dimensionality-reduced comprehensive features, a dynamic weighted fusion mechanism is constructed to generate a comprehensive risk score R(t) that evolves over time. The graded early warning and response module is used to dynamically correct the preset basic thresholds based on the current battery state of charge (SOC) and state of health (SOH), forming a three-level early warning threshold that changes with the battery operating conditions. When R(t) exceeds the first-level threshold, a primary early warning is triggered and the sampling frequency is increased. When it exceeds the second-level threshold, an intermediate response is initiated to control the liquid cooling system and the charging and discharging current. When it exceeds the third-level threshold, a high-level response is executed, cutting off the main circuit and activating the fire extinguishing device.

[0024] Thirdly, this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement an adaptive multi-parameter lithium-ion battery thermal runaway early warning method.

[0025] Fourthly, this application provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements an adaptive multi-parameter lithium-ion battery thermal runaway early warning method.

[0026] Fifthly, this application provides a computer program product, which includes computer instructions that instruct a computer to execute an adaptive multi-parameter lithium-ion battery thermal runaway early warning method.

[0027] The beneficial effects of the technical solution proposed in this application are: This application comprehensively covers the entire thermal runaway evolution process by integrating four heterogeneous parameters: temperature, voltage, internal stress, and gas, significantly improving early warning capabilities. The adaptive weighted filtering algorithm employed balances noise suppression and transient response, effectively improving data quality. A highly robust risk scoring model is constructed through PCA dimensionality reduction and dynamic weighted fusion mechanisms, avoiding biases caused by manual weighting. Innovatively, SOC and SOH are introduced to correct the warning threshold online, enabling the system to automatically increase alarm sensitivity under high charge or aging conditions. A graded response mechanism achieves closed-loop control from passive alarm to active handling, allowing for differentiated responses at different times, ensuring safety while reducing losses from erroneous actions. The entire method requires no additional hardware investment, is highly compatible, and can be deployed in existing BMS systems, making it suitable for large-scale energy storage power stations, electric vehicles, and power system frequency regulation scenarios. Attached Figure Description

[0028] Figure 1 Flowchart of the adaptive multi-parameter lithium-ion battery thermal runaway early warning method provided in this application; Figure 2 This is a schematic diagram of the adaptive multi-parameter lithium-ion battery thermal runaway early warning system provided in this application; Figure 3 A schematic diagram of an electronic device provided in this application. Detailed Implementation

[0029] The embodiments of this application are described in detail below. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain this application, and should not be construed as limiting this application. The step numbers in the following embodiments are set only for ease of explanation, and there is no limitation on the order between the steps. The execution order of each step in the embodiments can be adaptively adjusted according to the understanding of those skilled in the art.

[0030] In the description of this application, unless otherwise expressly defined, terms such as "setup," "installation," and "connection" should be interpreted broadly, and those skilled in the art can reasonably determine the specific meaning of the above terms in this application in conjunction with the specific content of the technical solution.

[0031] The embodiments of this application are described in detail below. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain this application, and should not be construed as limiting this application. The step numbers in the following embodiments are set only for ease of explanation, and there is no limitation on the order between the steps. The execution order of each step in the embodiments can be adaptively adjusted according to the understanding of those skilled in the art.

[0032] In the description of this application, unless otherwise expressly defined, terms such as "setup," "installation," and "connection" should be interpreted broadly, and those skilled in the art can reasonably determine the specific meaning of the above terms in this application in conjunction with the specific content of the technical solution.

[0033] Terminology Explanation: SOC (State of Charge): Refers to the percentage of a battery's remaining charge relative to its rated capacity, typically expressed as a percentage (0%–100%). In this application, SOC serves as a crucial input variable for dynamically adjusting the warning threshold. High SOC levels result in poor battery thermal stability, necessitating improved alarm sensitivity to prevent increased risk of thermal runaway due to heightened side reactions under full charge conditions.

[0034] SOH (State of Health): A technical indicator reflecting the degree of battery aging, generally calculated based on capacity decay or internal resistance increase, with a value ranging from 80% to 100%. This application introduces SOH into the early warning model to correct the early warning threshold, automatically enhancing monitoring sensitivity after battery aging and improving adaptability to deteriorating batteries.

[0035] BMS (Battery Management System): An electronic system used to collect key parameters of the battery pack in real time, such as voltage, current, and temperature, and to perform state estimation, equalization control, and fault diagnosis. In this solution, the BMS serves as the core platform for multi-parameter data acquisition and preliminary processing, supporting communication with the edge controller to achieve online identification of thermal runaway risks.

[0036] PCA (Principal Component Analysis) is a statistical dimensionality reduction method that extracts the principal components with the largest variance contribution by performing eigenvalue decomposition on the covariance matrix of eigenvectors, removing redundant information and retaining the main risk features. This application utilizes PCA to extract the first two principal components (PC1 and PC2) from the five-dimensional original features, corresponding to the "thermal-gas" and "electrical-mechanical" coupling failure paths, respectively, significantly improving the model's generalization ability and anti-interference performance.

[0037] DTS (Distributed Temperature Sensing): Based on the principle of fiber Raman scattering, this technology continuously measures the spatial temperature distribution through a single optical fiber, enabling full-time and spatial sensing of the temperature field on the surface of the battery module. This application uses DTS to replace traditional point-based temperature measurement, achieving a spatial resolution of 1 cm, effectively identifying local hot spots, and detecting signs of thermal runaway in advance.

[0038] MEMS (Micro-Electro-Mechanical Systems): Micro-electro-mechanical systems; integrated micro-devices that combine micro-sensors, actuators, and signal processing circuits. This application uses a MEMS piezoresistive sensor mounted on the surface of the battery casing to detect mechanical deformation caused by SEI film rupture, lithium plating, or gas generation, filling a gap in traditional monitoring in the mechanical dimension.

[0039] ADC (Analog-to-Digital Converter): An electronic device that converts analog signals (such as voltage and pressure) into digital signals for use by a processor. This application uses an isolated ADC (such as the TI ADS8588S), which has electrical isolation to prevent common-mode interference under high-voltage platforms and ensure voltage sampling accuracy and system safety.

[0040] R(t): Time-varying comprehensive risk score; a risk quantification index generated by weighted fusion of key features after principal component analysis, dynamically updated over time. This score serves as the basis for triggering graded responses, and can uniformly characterize the risk level of multi-source heterogeneous signals, realizing the transformation from multi-dimensional data to a single decision quantity.

[0041] Principal components are orthogonal characteristic directions obtained after PCA processing. The first principal component (PC1) is mainly driven by temperature variance and hydrogen concentration, reflecting the thermo-chemical failure path; the second principal component (PC2) consists of voltage change rate and internal stress, reflecting the electro-mechanical joint degradation process. The cumulative contribution of these two principal components is ≥85%, serving as the basic input for subsequent risk fusion.

[0042] Edge computing nodes, embedded computing devices deployed locally in energy storage power stations, possess the ability to independently complete data acquisition, signal processing, and risk assessment, enabling real-time early warning without relying on the cloud. Typical devices include the NXPS32K144 and Advantech UNO series industrial controllers, suitable for stable operation in complex electromagnetic environments.

[0043] OTA (Over-the-Air): Over-the-air upgrade, which remotely updates algorithm parameters and software logic in ECUs or edge controllers via wireless communication. This application supports pushing new warning model weights, correcting maps, or response strategies via OTA, enabling continuous system optimization and cross-project reuse, and reducing operation and maintenance costs.

[0044] PCS (Power Conversion System): A core power electronic device in an energy storage system, responsible for converting DC to AC power. In this application, when thermal runaway enters an advanced warning stage, the system will send a command to disconnect the PCS, achieving energy isolation and preventing the accident from escalating.

[0045] like Figure 1 As shown, the first objective of this application is to provide an adaptive multi-parameter lithium-ion battery thermal runaway early warning method. This method achieves early identification of thermal runaway by collecting multi-dimensional state data during battery operation and combining signal processing and risk assessment models. The method is described through steps S1 to S4, specifically including: S1, deploy temperature, voltage, internal stress and characteristic gas sensors in the battery module to collect real-time data on battery surface temperature, single cell voltage, internal stress and CO and H2 concentration in the chamber; Specifically, a multi-parameter data acquisition system was built, including monitoring of temperature, voltage, internal stress, and gas concentration. The battery surface temperature was monitored using distributed fiber optic temperature measurement, voltage signals were acquired using an isolated ADC, internal stress signals were acquired using a MEMS piezoresistive sensor, and characteristic gas concentrations were collected using CO and H2 sensors.

[0046] The internal stress refers to the shell deformation stress caused by factors such as gas generation inside the battery, as monitored by MEMS sensors.

[0047] S2 employs an adaptive weighted moving average filtering algorithm to denoise the signals of each channel. The weighting coefficients are dynamically adjusted based on the time decay factor and local signal fluctuations, and the length of the filtering window is automatically adjusted according to the degree of signal change. An adaptive weighted moving average filtering algorithm is used to process the collected temperature, voltage, internal stress, and gas concentration data. This algorithm can dynamically adjust the weights to balance the requirements of noise suppression and signal response speed. The specific processing method is as follows:

[0048] in, The original data value at time i. N To adjust the sliding window size, These are the weighting coefficients. This is the filtered output value.

[0049] Weight The adaptive strategy is a hybrid weighting strategy, which dynamically allocates weights based on the signal's local stability and timeliness, expressed by the formula:

[0050] in, The attenuation coefficient is determined by timeliness, meaning newer data carries greater weight, and its value ranges from 0.1 to 0.3. The data is local mean, with priority given to abrupt changes in response; that is, the greater the signal fluctuation, the greater the data weight. The sensitivity coefficient is 0.3.

[0051] Regarding window size N The window size can be dynamically adjusted to balance smoothness and real-time performance. When the signal changes suddenly and drastically (potentially abnormally), the calculation window shrinks for a faster response; when the signal is stable, the window expands to better smooth noise. Window size N The formula is expressed as:

[0052] in, This is the threshold for sudden change, which can be set according to different battery systems and operating conditions. If a sudden change in signal is detected, that is... Then reduce the window size. N For fast tracking; otherwise, increase the window size. N To enhance smoothness.

[0053] In the above scheme, the adaptive weighted moving average filtering algorithm effectively suppresses signal noise, improves the accuracy and stability of data acquisition, and reduces the impact of signal interference on the early warning system.

[0054] S3 constructs a comprehensive feature vector from the filtered data, including mean temperature, temperature variance, voltage change rate, internal stress amplitude, and gas concentration, and performs standardization to eliminate the influence of dimensions; Principal component analysis (PCA) is used to reduce the dimensionality of high-dimensional features and extract the key principal components with the highest contribution rate as risk-sensitive indicators. By combining the filtered temperature, voltage, internal stress, and gas concentration signals, a comprehensive feature vector is constructed:

[0055] in, The average temperature within the window. The temperature variance within the window. This represents the rate of change of voltage sag. For battery internal stress, For gas concentration, This is a comprehensive feature vector.

[0056] Then the eigenvalues ​​are standardized to eliminate the influence of dimensions:

[0057] in, The filtered first i 1 eigenvector For feature vectors The average value under normal battery operating conditions; For feature vectors Standard deviation under normal battery operating conditions; This is the standardized feature vector.

[0058] Perform PCA dimensionality reduction to remove redundant features that do not change significantly:

[0059] in, The principal component matrix obtained by eigenvalue decomposition ( k×m, k (Dimensionality after dimensionality reduction) The standardized feature vector has the following dimensions. m×1 ( m (total number of features); The eigenvectors after dimensionality reduction ( k×1 ,generally k m ).

[0060] Perform weighted fusion:

[0061] in, For the dimensionality reduction of the first i Each principal component characteristic; The weighting coefficient is dynamic and can be adjusted according to different operating conditions and battery types to meet the requirements. ; To integrate risk scoring.

[0062] The above scheme employs standardized dimensionality reduction and dynamic weighted fusion technology to comprehensively capture early signs of thermal runaway, effectively integrate multi-dimensional signal features, and improve the comprehensiveness and generalization ability of thermal runaway risk assessment.

[0063] S4. Based on the dimensionality-reduced features, a dynamic weighted fusion mechanism is constructed to generate a comprehensive risk score R(t) that evolves over time. According to the current battery state of charge (SOC) and state of health (SOH), the preset basic threshold is dynamically corrected to form a three-level early warning threshold that changes with the battery operating conditions. When R(t) exceeds the first-level threshold, a primary early warning is triggered and the sampling frequency is increased. When it exceeds the second-level threshold, an intermediate response is initiated to control the liquid cooling system and the charging and discharging current. When it exceeds the third-level threshold, a high-level response is executed, the main circuit is cut off, and the fire extinguishing device is activated.

[0064] Specifically, the tiered early warning method is based on dynamic risk scoring. and warning threshold Precise intervention is achieved through a three-tiered response: when When it is a primary warning, the system sampling frequency is increased; when At this time, a medium-level warning is issued, the liquid cooling system operates at full power and the charging and discharging current is limited; when At that time, a high-level warning is issued, the main circuit is cut off, and the fire extinguishing system is triggered.

[0065] Warning threshold The formula is:

[0066] in, Basic threshold; This is the SOC correction factor, for full charge time. =0.1, at low SOC =-0.05; The correction factor for the state of health (SOH) of the aged battery. =0.15; This represents the current health status, with a value ranging from 0.8 to 1.0.

[0067] By using dynamic threshold correction and a three-level response mechanism, precise graded intervention based on real-time risk scoring is achieved, improving the adaptability of the early warning strategy to different battery states and operating conditions and the accuracy of early warning.

[0068] In existing technologies, traditional lithium-ion battery thermal runaway early warning mainly relies on voltage and temperature information collected by the BMS (Battery Management System) and determines whether an anomaly has occurred by setting a fixed threshold. However, this method struggles to capture weak signals in the early stages of thermal runaway and is susceptible to noise interference leading to false alarms. To improve sensing capabilities, some systems have introduced gas detection, but the issues of multi-parameter collaborative analysis and dynamic response remain unresolved. Therefore, this application proposes a four-dimensional sensing system integrating physical, chemical, and electrical characteristics, combined with intelligent signal processing and hierarchical response logic, to upgrade from "passive alarm" to "active prevention and control."

[0069] The specific implementation is as follows: A standard 20-foot containerized battery compartment is selected in the lithium iron phosphate energy storage system. Each battery cluster is equipped with a distributed fiber optic temperature measurement chain (model: YOKOGAWA DTSX2), deployed along the side wall of the cell, with a spatial resolution of 1 cm and a sampling period of 1 Hz. Voltage is acquired by the BMS through an isolated ADC (TI ADS8588S, 16-bit precision), with an update frequency of 1 kHz. Internal stress monitoring uses a MEMS piezoresistive sensor (TE Connectivity MS5839-02-BA), mounted in the middle of the square aluminum-cased battery, and the output analog signal is conditioned by an operational amplifier before being connected to the main control MCU. Gas detection uses electrochemical H2 and CO sensors (Figaro TGS series), installed below the vents on the top of the compartment, with detection limits of 5 ppm and 10 ppm, respectively. All sensor data is transmitted to the edge computing node (NXP S32K144 MCU) via a CAN bus, running an embedded Linux system to execute subsequent processing. In practical applications, the fiber optic can be replaced with an infrared thermal imaging array, and the gas sensor can be integrated with a multi-component detection module; this embodiment does not limit this.

[0070] In terms of working principle, the system constructs a unified risk input framework through multi-source data fusion, and then generates a quantifiable and comparable comprehensive risk score R(t) through adaptive filtering, standardization, PCA dimensionality reduction and dynamic weighted fusion. The system also dynamically adjusts the warning threshold based on the current SOC and SOH of the battery to achieve personalized judgment. Finally, the closed-loop control is completed through a three-level response mechanism.

[0071] Using this method, effective early warning can be achieved within 3 minutes before thermal runaway occurs, which is 1 to 2 minutes earlier than the traditional single temperature method; the comprehensive risk scoring mechanism significantly reduces the false alarm rate, with the measured false alarm rate dropping to below 5%; dynamic threshold correction reduces the false alarm rate of aging batteries by 40%, improving the robustness and engineering applicability of the system.

[0072] As an optional solution, the temperature is acquired through a distributed fiber optic temperature measurement system, continuously deployed along the battery surface, achieving a spatial resolution of 1 cm and supporting thermal distribution imaging. In existing technologies, point-type temperature sensors (such as thermocouples or thermistors) can only reflect local temperatures, failing to locate hot spots and exhibiting blind spots. Therefore, this application employs a Raman scattering-type distributed fiber optic temperature measurement system, using polyimide-coated single-mode fiber (temperature resistant up to 300℃), deployed in a serpentine pattern along the sidewall of the battery module, acquiring a temperature point every 1 cm, covering the entire internal structure of the enclosure. The demodulator uses a Yokogawa DTSX2, with a measurement range of -40 to 150℃ and an accuracy of ±1℃. Data is uploaded to a local server via TCP / IP protocol. In practical applications, armored or low-smoke halogen-free sheathed fiber optics can be used, or a Fluke 544 infrared thermal imager can be used instead; this application does not limit the specific type. This system enables continuous temperature measurement throughout all time and space, accurately identifying local temperature rise trends and avoiding missed detections due to improper sensor placement. Significantly enhances thermal field reconfiguration capabilities, supports thermal runaway path tracing and fault source identification, and improves system transparency.

[0073] As an optional solution, an isolated ADC is used to acquire the voltage of individual battery cells, with a sampling accuracy of no less than 16 bits and strong anti-common-mode interference capability. To prevent common-mode interference and ground loop noise under high-voltage platforms from affecting the accuracy of voltage acquisition, this application uses the isolated ADC chip ADS8588S (Texas Instruments), which has 16-bit resolution, 8-channel synchronous sampling capability, and a sampling rate of 200kSPS. A dual-redundant design of optocoupler isolation and magnetic isolation is adopted between the power supply and the digital interface, with an isolation withstand voltage ≥2.5kV. The voltage probe uses a differential input mode, and an RC filter network (cutoff frequency 10kHz) is added at the front end to suppress high-frequency noise. The sampled data is transmitted to the main control unit through the SPI interface, and the timestamp is aligned with the temperature data. In practical applications, Σ-Δ modulator architectures such as ADI AD7403 can also be used; this application does not limit this approach. This acquisition method effectively suppresses electromagnetic interference under high-voltage environments, ensuring the authenticity of the voltage signal, and is particularly suitable for 1500V DC systems. It improves the detection sensitivity of abnormal behaviors such as voltage drops and dips, and supports early short-circuit identification.

[0074] As an optional solution, a MEMS piezoresistive sensor is mounted on the surface of the battery casing to monitor casing deformation caused by internal gas generation or SEI film rupture. The battery swells in the early stages of thermal runaway to release mechanical stress. Therefore, a TE MS5839-02-BA MEMS piezoresistive sensor is selected, with a range of 0–2 MPa, a sensitivity of 1 mV / kPa, a supply voltage of 5V, and an analog signal output. The sensor is bonded to the center of the long side of the square battery cell using thermally conductive adhesive to ensure good contact. The signal is conditioned by an AD620 instrumentation amplifier and then sent to the MCU ADC channel, with a sampling frequency set to 100Hz. When a stress surge exceeding 0.5 MPa / s is detected, it is determined to be a precursor to an internal pressure rise. In practical applications, a fiber Bragg grating (FBG) sensor or a capacitive micro-deformation sensor can be used instead; this embodiment does not limit the choice. This structure achieves non-invasive online monitoring of the battery's internal mechanical behavior, filling the gap in traditional monitoring in the mechanical dimension. It can issue an early warning 1–2 minutes before the temperature rises, greatly improving safety.

[0075] As an optional solution, CO and H2 gas sensors are installed on the top of the energy storage compartment, with detection limits below 10 ppm and response times ≤30 s. Hydrogen is one of the main products of lithium-ion battery electrolyte decomposition, and its concentration increase is an important chemical indicator of thermal runaway. A TGS2616 H2 sensor and a TGS2109 CO sensor are installed on the top of the battery compartment, with measurement ranges of 0–1000 ppm and 0–500 ppm respectively, built-in temperature compensation function, and output as an analog voltage signal. The sampling frequency is set to 1 Hz, and it is connected to the central controller via an RS485 bus. When the H2 concentration exceeds 100 ppm for 10 seconds, it is considered a level one gas anomaly. In practical applications, the gas detection module can be expanded to a four-in-one or multi-gas analyzer to support the detection of more characteristic gases such as CH4 and C2H4; this embodiment does not limit this. This gas monitoring scheme establishes a response mechanism for typical gas production behaviors in the mid-to-late stages of thermal runaway and provides a third-party verification method independent of thermoelectric parameters.

[0076] As an optional solution, this mechanism implements a differentiated early warning strategy based on battery status, avoiding false alarms caused by a "one-size-fits-all" approach. This significantly improves the adaptability and accuracy of the early warning strategy under complex operating conditions. The three-level response corresponds to: increasing the sampling frequency, activating liquid cooling and current limiting, and cutting off the circuit and triggering the fire suppression system. To achieve precise intervention, this application sets up a three-stage response mechanism: Primary warning: When At the same time, the data sampling frequency was increased from 1Hz to 10Hz to enhance the sensing capability; Intermediate warning: When At this time, the liquid cooling system is started and operated at full power, and the charging current is limited to 30% of the rated value; Advanced warning: When At that time, the DC contactor is disconnected, the PCS is shut down, and the heptafluoropropane fire extinguishing device is activated in conjunction with it.

[0077] This mechanism achieves closed-loop management of the entire process from "monitoring → early warning → response". In practical applications, it can be integrated with fire-fighting linkage control systems, supporting remote manual confirmation before execution of shutdown operations; this application embodiment does not limit this. The tiered response mechanism ensures both a safety baseline and avoids economic losses caused by emergency shutdowns. It achieves an optimal balance between safety and economy, complying with power system operation standards.

[0078] Example The following provides a detailed description of the method of this application using specific embodiments: An adaptive multi-parameter lithium-ion battery thermal runaway early warning method includes the following steps: multi-parameter feature signal acquisition, adaptive moving average filtering, multi-modal feature fusion, and hierarchical early warning and response. Multi-parameter feature signal acquisition is achieved by deploying multiple sensors at key locations within the battery module. Distributed fiber optic temperature sensors are installed on the battery surface for temperature monitoring; battery voltage is acquired via an isolated ADC; MEMS piezoresistive sensors are mounted on the battery casing surface to monitor internal stress; and CO and H2 sensors are installed on the top of the battery cluster for gas concentration monitoring.

[0079] System initialization, settings =5、 =10, Attenuation coefficient =0.2, mutation threshold =2. The temperature data sequence for a certain period is: {35.8, 36.2, 35.5, 36.9, 36.1, 35.7, 37.8, 36.6, 35.3, 36.7}. An adaptive moving average filter is applied to the temperature signal, based on the window size. N formula:

[0080] A mutation was detected at k=7, i.e., |37.8-35.7|=2.1≥ =2, therefore the window is reduced to... =5, the window contains the last 5 data points (i=3~7): T(3)=35.5, T(4)=36.9, T(5)=36.1, T(6)=35.7, T(7)=37.8, and the local mean is obtained. =36.4.

[0081] Calculate weights according to the hybrid strategy Value:

[0082] in =0.2, a=0.3, =36.4, k=7. Find w(3)=0.3537, w(4)=0.4772, w(5)=0.6149, w(6)=0.6767, w(7)=0.7042.

[0083] Weighted average, calculate the filtered output :

[0084] Seeking =36.49, meaning that the original data of 37.8 is output as a filtered value of 36.49 after weighted averaging, which effectively suppresses temperature noise.

[0085] 3. Perform multimodal feature fusion. During battery thermal runaway, the average surface temperature (°C) of the battery at a certain time period is {309.98, 310.8, 311.72, 313.72, 317.24}, the temperature variance is {0.2856, 3.008, 4.0016, 11.9216, 33.1824}, the battery voltage (V) is {3.052, 2.999, 2.913, 2.913, 2.820, 2.691}, the hydrogen concentration (ppm) in the energy storage chamber is {68.875, 395.0625, 395.0625, 871.5625, 1062.5}, and the battery internal stress (MPa) data is {0.52, 1.23, 1.28, 2.53, 4.02}.

[0086] Then, based on the comprehensive feature vector formula:

[0087] Seeking

[0088] The features are standardized using historical mean and standard deviation to obtain the standardized feature vector matrix. :

[0089] The rows represent the standardized values ​​of mean temperature, temperature variance, voltage change rate, battery internal stress, and hydrogen concentration, respectively.

[0090] Based on the standardized eigenvectors, the covariance matrix is ​​calculated and eigenvalue decomposition is performed. Since the eigenvalue decomposition shows that the second feature (temperature variance) and the fifth feature (hydrogen concentration) have the largest variances, these two features are selected as principal components. (Eigenvectors after dimensionality reduction are shown.) for:

[0091] The first row shows the temperature variance characteristics, and the second row shows the hydrogen concentration characteristics, preserving the main variance of the original data.

[0092] Calculate the fusion risk score based on the provided PCA-reduced feature vectors and weighted fusion formula. The weighted fusion formula is:

[0093] Therefore, the fusion risk score at 5 time points for:

[0094] Set early warning thresholds :

[0095] For a fully charged new battery, the basic threshold , , , , Seeking .

[0096] Comparative risk scores With warning threshold ,Discover This triggers an advanced alarm, cuts off the main circuit, and activates the fire suppression system.

[0097] like Figure 2 As shown, this application also provides an adaptive multi-parameter lithium-ion battery thermal runaway early warning system, comprising: Multi-parameter feature signal acquisition is used to collect multi-parameter state features of the battery in real time. These multi-parameter state features include temperature, single cell voltage, internal stress changes, and CO and H2 concentrations inside the chamber. Adaptive moving average filtering is used to denoise multi-parameter state features using an adaptive weighted moving average filtering algorithm. The weight coefficients of the adaptive weighted moving average filtering algorithm are dynamically adjusted according to the time decay factor and local signal fluctuations, and the length of the filtering window is automatically adjusted according to the degree of signal change. Multimodal feature fusion is used to construct a comprehensive feature vector from the filtered data, which includes mean temperature, temperature variance, voltage change rate, internal stress amplitude, and gas concentration. The comprehensive feature vector is then standardized to eliminate the influence of dimensions. After standardization, principal component analysis (PCA) is used to reduce the dimensionality of the high-dimensional features. Based on the dimensionality-reduced comprehensive features, a dynamic weighted fusion mechanism is constructed to generate a comprehensive risk score R(t) that evolves over time. The graded early warning and response system is used to dynamically correct the preset basic thresholds based on the current battery state of charge (SOC) and state of health (SOH), forming a three-level early warning threshold that changes with the battery operating conditions. When R(t) exceeds the first-level threshold, a primary early warning is triggered and the sampling frequency is increased. When it exceeds the second-level threshold, an intermediate response is initiated to control the liquid cooling system and the charging and discharging current. When it exceeds the third-level threshold, a high-level response is executed, cutting off the main circuit and activating the fire extinguishing device.

[0098] like Figure 3 As shown, a third objective of this application embodiment is to provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the aforementioned adaptive multi-parameter lithium-ion battery thermal runaway early warning method. It also includes a communication interface and a bus.

[0099] The aforementioned adaptive multi-parameter lithium-ion battery thermal runaway early warning method includes: S1 collects multi-parameter state characteristics of the battery in real time, including temperature, single cell voltage, internal stress changes, and CO and H2 concentrations in the chamber. S2, an adaptive weighted moving average filtering algorithm is used to denoise the multi-parameter state features. The weight coefficients of the adaptive weighted moving average filtering algorithm are dynamically adjusted according to the time decay factor and local signal fluctuations, and the length of the filtering window is automatically adjusted according to the degree of signal change. S3 constructs a comprehensive feature vector from the filtered data, including mean temperature, temperature variance, voltage change rate, internal stress amplitude, and gas concentration. The comprehensive feature vector is then standardized to eliminate the influence of dimensions. After standardization, principal component analysis (PCA) is used to reduce the dimensionality of the high-dimensional features. Based on the dimensionality-reduced comprehensive features, a dynamic weighted fusion mechanism is constructed to generate a comprehensive risk score R(t) that evolves over time. S4 dynamically corrects the preset basic threshold based on the current battery state of charge (SOC) and state of health (SOH), forming a three-level warning threshold that changes with the battery operating conditions. When R(t) exceeds the first-level threshold, a primary warning is triggered and the sampling frequency is increased. When it exceeds the second-level threshold, an intermediate response is initiated to control the liquid cooling system and the charging and discharging current. When it exceeds the third-level threshold, an advanced response is executed, cutting off the main circuit and activating the fire extinguishing device.

[0100] The fourth objective of this application is to provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned adaptive multi-parameter lithium-ion battery thermal runaway early warning method.

[0101] The aforementioned adaptive multi-parameter lithium-ion battery thermal runaway early warning method includes: S1 collects multi-parameter state characteristics of the battery in real time, including temperature, single cell voltage, internal stress changes, and CO and H2 concentrations in the chamber. S2, an adaptive weighted moving average filtering algorithm is used to denoise the multi-parameter state features. The weight coefficients of the adaptive weighted moving average filtering algorithm are dynamically adjusted according to the time decay factor and local signal fluctuations, and the length of the filtering window is automatically adjusted according to the degree of signal change. S3 constructs a comprehensive feature vector from the filtered data, including mean temperature, temperature variance, voltage change rate, internal stress amplitude, and gas concentration. The comprehensive feature vector is then standardized to eliminate the influence of dimensions. After standardization, principal component analysis (PCA) is used to reduce the dimensionality of the high-dimensional features. Based on the dimensionality-reduced comprehensive features, a dynamic weighted fusion mechanism is constructed to generate a comprehensive risk score R(t) that evolves over time. S4 dynamically corrects the preset basic threshold based on the current battery state of charge (SOC) and state of health (SOH), forming a three-level warning threshold that changes with the battery operating conditions. When R(t) exceeds the first-level threshold, a primary warning is triggered and the sampling frequency is increased. When it exceeds the second-level threshold, an intermediate response is initiated to control the liquid cooling system and the charging and discharging current. When it exceeds the third-level threshold, an advanced response is executed, cutting off the main circuit and activating the fire extinguishing device.

[0102] A fifth objective of this application is to provide a computer program product comprising computer instructions that instruct a computer to execute the aforementioned adaptive multi-parameter lithium-ion battery thermal runaway early warning method.

[0103] The aforementioned adaptive multi-parameter lithium-ion battery thermal runaway early warning method includes: S1 collects multi-parameter state characteristics of the battery in real time, including temperature, single cell voltage, internal stress changes, and CO and H2 concentrations in the chamber. S2, an adaptive weighted moving average filtering algorithm is used to denoise the multi-parameter state features. The weight coefficients of the adaptive weighted moving average filtering algorithm are dynamically adjusted according to the time decay factor and local signal fluctuations, and the length of the filtering window is automatically adjusted according to the degree of signal change. S3 constructs a comprehensive feature vector from the filtered data, including mean temperature, temperature variance, voltage change rate, internal stress amplitude, and gas concentration. The comprehensive feature vector is then standardized to eliminate the influence of dimensions. After standardization, principal component analysis (PCA) is used to reduce the dimensionality of the high-dimensional features. Based on the dimensionality-reduced comprehensive features, a dynamic weighted fusion mechanism is constructed to generate a comprehensive risk score R(t) that evolves over time. S4 dynamically corrects the preset basic threshold based on the current battery state of charge (SOC) and state of health (SOH), forming a three-level warning threshold that changes with the battery operating conditions. When R(t) exceeds the first-level threshold, a primary warning is triggered and the sampling frequency is increased. When it exceeds the second-level threshold, an intermediate response is initiated to control the liquid cooling system and the charging and discharging current. When it exceeds the third-level threshold, an advanced response is executed, cutting off the main circuit and activating the fire extinguishing device.

[0104] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0105] These computer program instructions can also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0106] This application may take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application may take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, readable storage media, optical storage, etc.) containing computer-usable program code.

[0107] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0108] Obviously, the described embodiments are only some, not all, of the embodiments in this application. All other embodiments obtained by those skilled in the art based on the embodiments in this application without inventive effort should fall within the scope of protection of this application.

[0109] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application and not to limit them. Although this application has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation methods of this application. Any modifications or equivalent substitutions that do not depart from the spirit and scope of this application should be covered within the protection scope of this application.

Claims

1. A self-adapting multi-parameter lithium-ion battery thermal runaway early warning method, characterized in that, include: Real-time acquisition of multi-parameter state characteristics of the battery, including temperature, single cell voltage, internal stress, and CO and H2 concentrations inside the chamber; An adaptive weighted moving average filtering algorithm is used to denoise the multi-parameter state features. The weight coefficients of the adaptive weighted moving average filtering algorithm are dynamically adjusted according to the time decay factor and local signal fluctuations, and the length of the filtering window is automatically adjusted according to the degree of signal change. The filtered data is used to construct a comprehensive feature vector containing mean temperature, temperature variance, voltage change rate, internal stress amplitude, and gas concentration. The comprehensive feature vector is then standardized to eliminate the influence of dimensions. After standardization, principal component analysis (PCA) is used to reduce the dimensionality of the high-dimensional features. Based on the dimensionality-reduced comprehensive features, a dynamic weighted fusion mechanism is constructed to generate a comprehensive risk score R(t) that evolves over time. Based on the current battery state of charge (SOC) and state of health (SOH), the preset basic threshold is dynamically corrected to form a three-level early warning threshold that changes with the battery operating conditions. When R(t) exceeds the first-level threshold, a primary warning is triggered and the sampling frequency is increased. When it exceeds the second-level threshold, an intermediate response is initiated to control the liquid cooling system and the charging and discharging current. When it exceeds the third-level threshold, an advanced response is executed, the main circuit is cut off, and the fire extinguishing device is activated.

2. The adaptive multi-parameter lithium-ion battery thermal runaway early warning method of claim 1, wherein, The real-time acquisition of the battery's multi-parameter state characteristics is achieved by building a multi-parameter data acquisition system that includes monitoring temperature, voltage, internal stress, and gas concentration. Specifically, the battery surface temperature is monitored by distributed optical fiber temperature measurement, voltage signals are acquired by an isolated ADC, internal stress signals are acquired by a MEMS piezoresistive sensor, and characteristic gas concentrations are acquired using CO and H2 sensors.

3. The adaptive multi-parameter lithium-ion battery thermal runaway early warning method of claim 1, wherein, The step of using an adaptive weighted moving average filtering algorithm to denoise multi-parameter state features includes: An adaptive weighted moving average filtering algorithm is used to process the collected temperature, voltage, internal stress, and gas concentration data. The algorithm dynamically adjusts the weights to balance noise suppression and signal response speed requirements. Specifically, the processing involves: wherein, is the original data value at the i-th moment, N is the size of the sliding window, is the weight coefficient, is the filtered output value; where the weight The adaptive strategy is a hybrid weight strategy, which dynamically allocates weights according to the local stability and timeliness of the signals. Specifically, wherein, is an attenuation coefficient, and has a value ranging from 0.1 to 0.3; is a local mean value; is a sensitivity coefficient; Window size N Expressed as: wherein, is a mutation threshold, which is set according to different battery systems and working conditions.

4. The adaptive multi-parameter lithium-ion battery thermal runaway early warning method according to claim 1, characterized in that, The process of constructing a comprehensive feature vector from the filtered data, including mean temperature, temperature variance, voltage change rate, internal stress amplitude, and gas concentration, includes: By combining the filtered temperature, voltage, internal stress, and gas concentration signals, a comprehensive feature vector is constructed: in, The average temperature within the window. The temperature variance within the window. This represents the rate of change of voltage sag. For battery internal stress, For gas concentration, This is a comprehensive feature vector.

5. The adaptive multi-parameter lithium-ion battery thermal runaway early warning method according to claim 1, characterized in that, The standardization process for the comprehensive feature vector to eliminate the influence of dimensions includes: in, The filtered first i 1 eigenvector For feature vectors The average value under normal battery operating conditions; For feature vectors Standard deviation under normal battery operating conditions; This is the standardized feature vector.

6. The adaptive multi-parameter lithium-ion battery thermal runaway early warning method according to claim 1, characterized in that, The standardization process followed by principal component analysis (PCA) is used to reduce the dimensionality of high-dimensional features, including: in, The principal component matrix is ​​obtained from eigenvalue decomposition. k×m, k The dimension after dimensionality reduction; The standardized feature vector has the following dimensions. m×1 , m The total number of features; The feature vectors after dimensionality reduction. k×1 , k m .

7. The adaptive multi-parameter lithium-ion battery thermal runaway early warning method according to claim 1, characterized in that, The dynamic weighted fusion mechanism constructed based on the comprehensive features after dimensionality reduction generates a comprehensive risk score R(t) that evolves over time, including: in, For the dimensionality reduction of the first i Each principal component characteristic; The weighting coefficients are dynamic and adjusted according to different operating conditions and battery types to meet the requirements. ; To integrate risk scoring.

8. The adaptive multi-parameter lithium-ion battery thermal runaway early warning method according to claim 1, characterized in that, The three-level early warning threshold that varies with battery operating conditions is based on dynamic risk scoring. and warning threshold Specifically: when At this time, as a primary warning, the system sampling frequency is increased; when At this time, a medium-level warning is issued, the liquid cooling system operates at full power and the charging and discharging current is limited; when At that time, a high-level warning is issued, the main circuit is cut off, and the fire extinguishing system is triggered.

9. An adaptive multi-parameter lithium-ion battery thermal runaway early warning method according to claim 8, characterized in that, The system dynamically adjusts the preset base threshold based on the current battery state of charge (SOC) and state of health (SOH) to obtain the warning threshold. Specifically: in, Basic threshold; This is the SOC correction factor; The correction factor for the state of health (SOH); This represents the current health status, with a value ranging from 0.8 to 1.

0.

10. An adaptive multi-parameter lithium-ion battery thermal runaway early warning system, characterized in that, include: The multi-parameter characteristic signal acquisition module is used to acquire multi-parameter state characteristics of the battery in real time, including temperature, single cell voltage, internal stress change and CO and H2 concentration in the chamber. The adaptive moving average filtering module is used to denoise multi-parameter state features using an adaptive weighted moving average filtering algorithm. The weight coefficients of the adaptive weighted moving average filtering algorithm are dynamically adjusted according to the time decay factor and local signal fluctuations, and the length of the filtering window is automatically adjusted according to the degree of signal change. The multimodal feature fusion module is used to construct a comprehensive feature vector from the filtered data, which includes the mean temperature, temperature variance, voltage change rate, internal stress amplitude, and gas concentration. The comprehensive feature vector is standardized to eliminate the influence of dimensions. After standardization, principal component analysis (PCA) is used to reduce the dimensionality of the high-dimensional features. Based on the dimensionality-reduced comprehensive features, a dynamic weighted fusion mechanism is constructed to generate a comprehensive risk score R(t) that evolves over time. The graded early warning and response module is used to dynamically correct the preset basic threshold based on the current battery state of charge (SOC) and state of health (SOH), forming a three-level early warning threshold that changes with the battery operating conditions. When R(t) exceeds the first-level threshold, a primary warning is triggered and the sampling frequency is increased. When it exceeds the second-level threshold, an intermediate response is initiated to control the liquid cooling system and the charging and discharging current. When it exceeds the third-level threshold, an advanced response is executed, the main circuit is cut off, and the fire extinguishing device is activated.

11. An electronic device, characterized in that, The method includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the adaptive multi-parameter lithium-ion battery thermal runaway early warning method according to any one of claims 1-9.

12. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the adaptive multi-parameter lithium-ion battery thermal runaway early warning method according to any one of claims 1-9.

13. A computer program product, said computer program product comprising computer instructions, characterized in that, The computer instructions instruct the computer to execute the adaptive multi-parameter lithium-ion battery thermal runaway early warning method according to any one of claims 1-9.

Images