A method and system for thermal runaway detection of an energy storage battery

Abstract

The present application belongs to the technical field of screening methods, and particularly relates to a method and system for detecting thermal runaway of energy storage batteries, parameter layer construction and accurate preprocessing of multi-source data, a dynamic and static dual-dimension fault feature parameter pool covering the whole cycle of thermal runaway is built, the risk identification data source system is improved, and a multi-level composite filtering and data verification mechanism is adopted to eliminate invalid data caused by working condition interference, equipment disturbance and sensor abnormality; multi-feature weighted fusion risk discrimination, for the implicit fault features of early thermal runaway multi-parameter weak collaborative deviation, the differentiated dynamic feature weight is configured through data training, the normal working condition disturbance and fault deviation are accurately distinguished, and a unified comprehensive risk coefficient is output through fusion calculation.

Description

Technical Field

[0001] This invention belongs to the technical field of screening methods, and particularly relates to a method and system for detecting thermal runaway of energy storage batteries. Background Technology

[0002] Currently, commercial energy storage battery BMS and supporting fire protection systems generally use single-parameter fixed threshold alarm logic for thermal runaway monitoring. They rely solely on instantaneous values ​​of single indicators such as cell temperature, individual cell voltage, and cabin gas concentration to determine exceedances, resulting in a limited monitoring scope. Existing monitoring systems only collect data from the battery casing surface and external module operation, failing to simultaneously capture early, subtle degradation characteristics within the cells. Furthermore, they do not distinguish between steady-state static parameters and dynamic features such as rate of change, limiting the coverage of monitoring data sources. During field operation, equipment vibration, charge / discharge rate switching, ventilation airflow disturbances, and changes in ambient temperature and humidity all cause slight parameter fluctuations. Fixed threshold mechanisms cannot distinguish between normal operating condition disturbances and actual fault deviations, easily generating numerous invalid alarms. Simultaneously, various sensors exhibit zero-point drift over long-term use, and as batteries age through cycles, their operating baseline parameters shift, making it impossible for static judgment thresholds to adapt synchronously. This can easily lead to missed fault detection in old cells and mixed old and new battery clusters.

[0003] Existing monitoring and judgment modes only identify out-of-limit values ​​at the instantaneous sampling point, lacking the ability to analyze time-series trends and perform multi-feature collaborative fusion for judgment. In the early stages of battery thermal runaway, only slight synchronous shifts in multiple parameters are observed; if a single indicator does not reach the alarm threshold, the potential hazard cannot be identified, lacking proactive prediction capabilities. Mainstream equipment has not established a battery lifecycle state linkage correction mechanism, nor has it formed a condition-linked anti-shake logic with PCS and thermal management equipment. Judgment standards cannot be dynamically adjusted according to battery health status and on-site operating conditions. Furthermore, traditional early warning algorithms are mostly deployed independently on BMS or fire control panels, with inconsistent communication protocols among devices, a lack of risk warning and graded handling logic, and a lag in the linkage between early warning signals and protective actions such as power outages, ventilation, and fire extinguishing. Moreover, there is a lack of a backend data review and iteration channel, and monitoring and judgment rules cannot be continuously optimized with long-term operating data. Overall, the accuracy and adaptability of thermal runaway monitoring are insufficient to meet the safe operation requirements of large-scale energy storage power stations. Summary of the Invention

[0004] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, embodiments of the present invention provide a method for detecting thermal runaway in an energy storage battery, the method comprising: The parameter hierarchical construction and multi-source data precise preprocessing are carried out by building a dynamic and static dual-dimensional fault characteristic parameter pool covering the entire thermal runaway cycle, improving the risk identification data source system, and adopting a multi-level composite filtering and data verification mechanism to eliminate invalid data caused by operating condition interference, equipment disturbance and sensor anomalies. Multi-feature weighted fusion risk discrimination targets the latent fault characteristics of weak collaborative shift of multiple parameters in early thermal runaway. By configuring differentiated dynamic feature weights through data training, it accurately distinguishes between normal operating condition disturbances and fault shifts, and calculates and outputs a unified comprehensive risk coefficient. The full life cycle adaptive threshold iteration relies on the battery aging status database to associate the cell health status and operating parameters. For battery aging and mixed new and old battery scenarios, the parameter baseline and alarm threshold are dynamically corrected. Combined with the periodic baseline iteration mechanism, the judgment deviation caused by aging and sensor drift is offset. Time-series trend extrapolation and hierarchical early warning and control: Based on continuous time-series operation data, fit the parameter change trend, predict the evolution of thermal runaway faults, and establish a multi-level risk classification system; The working condition linkage anti-shake correction is designed to identify disturbance scenarios in real time and adaptively adjust the judgment logic for complex working conditions such as energy storage system charging and discharging fluctuations, equipment start-up and shutdown, and environmental disturbances, and to filter out minor over-limits of non-fault parameters. The algorithm engineering implementation and closed-loop optimization are linked, and the early warning algorithm is adapted and embedded into the existing BMS and fire control panel. Multi-system communication links are built to realize multi-level risk linkage and response. At the same time, the model parameters are iteratively optimized through background data review.

[0005] Furthermore, embodiments of the present invention also provide a thermal runaway detection system for energy storage batteries, characterized in that it includes: A processor; a machine-readable storage medium for storing machine-executable instructions of the processor; wherein the processor is configured to perform the above-described energy storage battery thermal runaway detection method by executing the machine-executable instructions.

[0006] In another aspect, embodiments of the present invention also provide a computer program product, the computer program product including machine-executable instructions, the machine-executable instructions being stored in a computer-readable storage medium, a processor of a computer device reading the machine-executable instructions from the computer-readable storage medium, the processor executing the machine-executable instructions, causing the computer device to execute the above-described energy storage battery thermal runaway detection method.

[0007] Based on the above, current commercial energy storage battery BMS and supporting fire protection systems generally adopt single-parameter fixed threshold alarm logic for thermal runaway monitoring. They rely solely on instantaneous values ​​of single indicators such as cell temperature, individual cell voltage, and cabin gas concentration to determine exceedances, resulting in a relatively limited monitoring dimension. Existing monitoring systems only collect data from the battery casing surface and external module operation, failing to simultaneously capture early, subtle degradation characteristics within the cells. Furthermore, they do not distinguish between steady-state static parameters and dynamic features such as rate of change, limiting the coverage of monitoring data sources. During field operation, equipment vibration, charge / discharge rate switching, ventilation airflow disturbances, and changes in ambient temperature and humidity all cause slight parameter fluctuations. Fixed threshold mechanisms cannot distinguish between normal operating condition disturbances and actual fault deviations, easily generating numerous invalid alarms. Simultaneously, various sensors exhibit zero-point drift over long-term use, and as batteries age through cycles, their operating baseline parameters shift, making it impossible for static judgment thresholds to adapt synchronously. This can easily lead to missed fault detection in old cells and mixed old and new battery clusters.

[0008] Existing monitoring and judgment modes only identify out-of-limit values ​​at the instantaneous sampling point, lacking the ability to analyze time-series trends and perform multi-feature collaborative fusion for judgment. In the early stages of battery thermal runaway, only slight synchronous shifts in multiple parameters are observed; if a single indicator does not reach the alarm threshold, the potential hazard cannot be identified, lacking proactive prediction capabilities. Mainstream equipment has not established a battery lifecycle state linkage correction mechanism, nor has it formed a condition-linked anti-shake logic with PCS and thermal management equipment. Judgment standards cannot be dynamically adjusted according to battery health status and on-site operating conditions. Furthermore, traditional early warning algorithms are mostly deployed independently on BMS or fire control panels, with inconsistent communication protocols among devices, a lack of risk warning and graded handling logic, and a lag in the linkage between early warning signals and protective actions such as power outages, ventilation, and fire extinguishing. Moreover, there is a lack of a backend data review and iteration channel, and monitoring and judgment rules cannot be continuously optimized with long-term operating data. Overall, the accuracy and adaptability of thermal runaway monitoring are insufficient to meet the safe operation requirements of large-scale energy storage power stations. Attached Figure Description

[0009] Figure 1 This is a schematic diagram of the execution flow of the thermal runaway detection method for energy storage batteries provided in this embodiment of the invention.

[0010] Figure 2 This is a schematic diagram of exemplary hardware and software components of the thermal runaway detection system for energy storage batteries provided in an embodiment of the present invention. Detailed Implementation

[0011] The present invention will now be described in detail with reference to the accompanying drawings. Figure 1 This is a flowchart illustrating a thermal runaway detection method for energy storage batteries according to an embodiment of the present invention. The following is a detailed description of the thermal runaway detection method for energy storage batteries.

[0012] Step S110: Parameter hierarchical construction and accurate preprocessing of multi-source data. By building a dynamic and static dual-dimensional fault characteristic parameter pool covering the entire thermal runaway cycle, the risk identification data source system is improved. A multi-level composite filtering and data verification mechanism is adopted to eliminate invalid data caused by operating condition interference, equipment disturbance and sensor anomalies. This implementation constructs a dynamic and static dual-dimensional fault feature parameter pool covering the entire lifecycle of battery thermal runaway operation, building a complete risk identification data source system. The parameter pool synchronously integrates static steady-state parameters and dynamic gradient change parameters, covering the data feature dimensions of the entire process of normal battery operation, state degradation, and thermal runaway development. The system uniformly performs multi-level composite filtering and data validity verification on the collected multi-source raw data. Abnormal data caused by equipment vibration interference, instantaneous charging and discharging disturbances, ventilation airflow fluctuations, and ambient temperature deviations during energy storage field operation are uniformly purified and removed. Invalid data generated by sensor acquisition anomalies is completely shielded, providing regular, unified, and standardized raw data input support for backend risk assessment.

[0013] Step S111: Build a two-layer fault characteristic parameter system, which is divided into two dimensions: static steady-state operation characteristics and dynamic change characteristics, to comprehensively cover the state data acquisition dimensions of the battery throughout the entire operation stage. This implementation establishes a two-layer fault characteristic parameter system, which is divided into a static steady-state operation characteristic dimension and a dynamic change characteristic dimension. These two types of characteristic dimensions work together to cover the state acquisition needs of the battery throughout its entire operating phase. The static steady-state characteristics characterize the battery's fixed operating state parameters at the current moment, while the dynamic change characteristics characterize the battery parameters' evolution over time. These two types of characteristics are independent yet complementary. During data acquisition, the system simultaneously reads parameter information from both dimensions, fully covering the battery's state performance under different operating conditions, including steady-state operation, transient fluctuations, slow degradation, and rapid anomalies, achieving full coverage of feature acquisition configuration across all battery operating conditions and states.

[0014] Step S112: The system performs multi-level standardized purification processing on the raw battery operation data collected on site, stabilizes high-frequency instantaneous jump signals caused by equipment vibration and instantaneous charging and discharging fluctuations, and corrects and calibrates steady-state data deviations caused by ventilation airflow and environmental temperature changes. This implementation method performs multi-level standardized purification processing on the raw battery operating data collected in real time on-site. For high-frequency pulse-like data jumps caused by equipment vibration and instantaneous charge / discharge rate fluctuations common in energy storage sites, a smoothing filter is used to correct the data and eliminate short-term abrupt interference. For steady-state baseline deviations caused by changes in cabin ventilation and slow shifts in ambient temperature and humidity, the system performs baseline calibration and offset correction to ensure a stable and consistent steady-state data benchmark. This step, without altering the original valid data variation patterns, completes the layered treatment of interference data from multiple scenarios, achieving standardized preprocessing of the raw data.

[0015] Step S113: Simultaneously set data validity verification rules to remove and shield invalid extreme value data caused by sensor drift, line signal interference, and signal disconnection anomalies, and select continuous, valid, and stable battery operation data as the basic input data for subsequent risk assessment.

[0016] This implementation method configures independent data validity verification rules to uniformly identify, eliminate, and shield invalid extreme value data caused by zero-point drift due to long-term sensor operation, line signal interference caused by the on-site electromagnetic environment, and data loss due to interrupted acquisition circuits and transient anomalies. The system performs validity determination on each set of sampled data, retaining continuous, stable, reasonably amplituded, and sequentially coherent valid data segments, while truncating and discarding abnormal and invalid data segments. Valid data after validity screening will be uniformly merged into the risk assessment data pool, serving as the basic input data for subsequent multi-feature fusion discrimination, threshold iteration, and trend inference.

[0017] Step S120: Multi-feature weighted fusion risk judgment. For the latent fault characteristics of weak collaborative shift of multiple parameters in early thermal runaway, differentiated dynamic feature weights are configured through data training to accurately distinguish between normal operating condition disturbances and fault shifts, and a unified comprehensive risk coefficient is output through fusion calculation. This implementation employs a multi-feature weighted fusion risk discrimination mechanism, adapting to the latent fault characteristics of the early thermal runaway stage of batteries, where only weak collaborative shifts in multiple parameters exist and no single parameter significantly exceeds limits. The system utilizes massive amounts of battery thermal runaway test samples and power plant normal operating condition samples to complete model training. Differentiated dynamic feature weights are configured based on the fault sensitivity characteristics of each parameter to distinguish between disturbances under normal operating conditions and fault-related parameter shifts. The system performs correlation and fusion calculations on the preprocessed multi-dimensional feature data, integrating multi-dimensional state information to generate a unified comprehensive risk coefficient for individual battery cells and battery modules, completing the quantitative discrimination output of abnormal battery states.

[0018] Step S121: The multi-dimensional feature collaborative association discrimination method is adopted to replace the traditional single-parameter independent judgment mode, and the model training is completed by relying on a large number of battery thermal runaway test samples and on-site normal working condition samples. This implementation method replaces the traditional single-parameter independent threshold judgment mode with a multi-dimensional feature collaborative correlation discrimination method, constructing a multi-feature joint discrimination model. During the model training phase, normal operation samples of batteries with different temperature ranges, different charge / discharge rates, and different aging degrees, as well as fault samples corresponding to various thermal runaway causes, are uniformly included to form a complete sample training dataset. Through iterative training with samples covering multiple scenarios and full coverage, the model establishes the distribution patterns of normal operating conditions and fault characteristics, enabling the model to adapt to complex field conditions in energy storage and achieve collaborative correlation recognition and state discrimination of multi-dimensional features.

[0019] Step S122: Based on the sensitivity of each feature parameter to battery thermal runaway fault and the difference in response timing, assign differentiated dynamic weights to different parameters so that the judgment weight of dynamic change features is higher than that of static numerical features. This implementation method assigns differentiated dynamic weighting coefficients to different types of characteristic parameters based on their sensitivity to battery thermal runaway faults and differences in response timing. The system sets high-weighting parameters for dynamic features that can reflect early fault ignition, such as temperature change rate, voltage decay rate, and gas concentration rise rate, while setting static numerical features such as conventional temperature, conventional voltage, and conventional gas concentration as auxiliary weighting parameters. The dynamic weights can be adaptively fine-tuned according to the battery's operating state and condition, always maintaining the dominant role of dynamic features in the risk assessment process.

[0020] Step S123: The system distinguishes between normal operating disturbances and fault parameter deviations in real time. It marks the operating condition disturbances caused by charge / discharge rate switching, ambient temperature fluctuations, and small parameter fluctuations caused by equipment start-up and shutdown. It performs correlation analysis on data of synchronous, unidirectional, and continuous deviations of multiple parameters. Based on the fusion calculation of preprocessed multi-dimensional data, it outputs the comprehensive risk coefficient of battery cells and modules.

[0021] This implementation method identifies routine operating condition disturbances and faulty parameter deviations in the energy storage system in real time during operation. Short-term parameter fluctuations caused by charge / discharge rate switching, minor ambient temperature fluctuations, and routine equipment start-ups and shutdowns are marked as operating condition disturbances and are not used as the basis for fault determination. The system performs correlation analysis on parameter deviations that are synchronous, consistent in direction, and persistent, and combines the dynamic weights corresponding to each feature to complete multi-dimensional data fusion calculations, ultimately outputting a comprehensive fault risk coefficient that characterizes the safety status of individual cells and battery modules.

[0022] Step S130: Full life cycle adaptive threshold iteration. Relying on the battery aging status database to associate cell health status and operating parameters, for battery aging and mixed old and new battery scenarios, the parameter baseline and alarm threshold are dynamically corrected. Combined with the periodic baseline iteration mechanism, the judgment deviation caused by aging and sensor drift is offset. This implementation establishes an adaptive threshold iteration mechanism for the entire battery lifecycle. It leverages a battery aging state database to dynamically correlate and match cell health status with operational judgment parameters, adapting to scenarios of gradual battery aging and mixed clusters of new and old batteries. The system dynamically adjusts the baseline values ​​of each characteristic parameter and the alarm judgment threshold based on the degree of cell aging, implementing differentiated judgment standard configurations for cells in different health states. The system also synchronously configures a periodic baseline iteration update mechanism, continuously updating the normal fluctuation range of parameters based on steady-state battery operation data. This real-time calibration addresses baseline offset issues caused by long-term battery aging and sensor drift, maintaining dynamic adaptation of the judgment benchmark.

[0023] Step S131: Construct a database of battery aging status throughout its entire life cycle, continuously collect and record operating data such as the number of cycles, SOH health, static internal resistance, and normal operating temperature range for each cell, and establish a linkage and adaptation mechanism between battery health status and judgment parameters. This implementation constructs a battery lifecycle aging status database. The database continuously collects and records the cumulative cycle count, real-time state of health (SOH), static internal resistance parameters, and normal operating temperature range for each individual cell throughout its lifecycle. The database establishes an independent status profile for each cell, continuously updating cell aging evolution data to form a distribution pattern of aging characteristics specific to each cell. The system links and binds the cell health status data stored in the database with a risk assessment model, achieving a one-to-one correspondence between battery health status and assessment rules.

[0024] Step S132: Based on the real-time aging status of the battery cells, the system differentiates the baseline and alarm threshold of each characteristic parameter, adjusts the allowable range of parameter fluctuations and fault identification sensitivity for old battery cells with high cycle count and low SOH, and sets the standard judgment threshold for new battery cells. This implementation method differentiates the baseline and alarm thresholds for each characteristic parameter based on the real-time aging status of the battery cells. For older cells with high cumulative cycle counts, significant SOH decay, and obvious internal resistance deviations, the allowable range of normal parameter fluctuations is adjusted, and the sensitivity for identifying abnormal changes is improved. For brand-new cells and cells in excellent health, the system retains the factory standard judgment thresholds and baseline parameters. Through batch-based and state-based differentiated threshold configuration, targeted and adaptable state discrimination configuration is achieved for the mixed operation of new and old battery clusters.

[0025] Step S133: The system sets up a monthly baseline iteration update mechanism to continuously collect operating data under steady-state battery conditions, dynamically update the normal fluctuation range of each parameter, and calibrate the baseline offset caused by battery aging and sensor temperature drift in real time.

[0026] This implementation method establishes a monthly baseline iteration update mechanism. Within a fixed monthly iteration cycle, the system automatically filters valid data under stable battery operating conditions and statistically analyzes the normal distribution range and baseline center value of each characteristic parameter. Based on the steady-state data, the system updates the normal fluctuation range and baseline of each parameter, simultaneously correcting baseline offsets caused by long-term temperature drift of sensors and aging drift of battery internal resistance. This ensures that the risk assessment benchmark can continuously and adaptively update along with the battery aging process and the operating status of the sensing device, maintaining the real-time effectiveness of the assessment benchmark.

[0027] Step S140: Time-series trend extrapolation and hierarchical early warning and control. Based on continuous time-series operation data, fit the parameter change trend, predict the early signs of thermal runaway failure, and establish a multi-level risk classification system. This implementation establishes a time-series trend prediction and hierarchical early warning and control mechanism. It leverages continuous time-series battery operation data to build a parameter change trend fitting capability, enabling proactive analysis of battery parameter evolution patterns. By continuously tracking the time-series changes of multi-dimensional parameters, the system predicts the evolution trend of battery thermal runaway faults, overcoming the limitations of instantaneous sampling judgment. Based on parameter degradation characteristics, the system establishes a multi-level risk classification system, standardizing the classification of different degrees of abnormal battery states and matching corresponding early warning output commands, thus expanding the application of battery faults from instantaneous judgment to trend prediction and hierarchical control.

[0028] Step S141: Continuously collect continuous time-series data of various battery operating parameters, calculate the change gradient and growth rate slope of each parameter in real time, and fit the continuous change law of the parameters and predict the evolution trend of subsequent operating parameters through linear fitting and time-series trend extrapolation algorithm. This implementation method continuously collects continuous time-series data of various battery operating parameters, constructs a long-term time-series data sequence, and calculates the gradient and rate of increase of each parameter per unit time in real time. The system uses linear fitting and time-series trend extrapolation algorithms to perform continuous curve fitting on discrete sampled data, restores the true change law of parameters, and extrapolates the parameter change trend in the near future based on historical time-series data, thereby capturing the evolution trend of continuous parameter deterioration in advance and realizing the early identification of abnormal battery conditions.

[0029] Step S142: The system combines the duration of parameter degradation, degradation rate, and multi-feature collaborative abnormal state to classify the battery operation risk into levels, and successively defines four risk levels: normal operation, minor hidden danger, general fault, and critical thermal runaway. For different risk levels, four types of standardized output instructions are configured: prompt, warning, alarm, and emergency linkage, so as to realize the pre-level hierarchical control of battery thermal runaway risk.

[0030] This implementation method combines the duration of parameter degradation, the rate of parameter degradation, and the comprehensive state of multi-feature coordinated anomalies to hierarchically classify battery operation risks, setting four risk levels: normal operation, minor hidden dangers, general faults, and critical thermal runaway. The system configures corresponding standardized output commands for each risk level, with four execution modes: notification push, early warning, alarm output, and emergency linkage control. Based on the real-time risk level, it outputs corresponding control commands to achieve refined, proactive, and hierarchical safety management of battery thermal runaway risks.

[0031] Step S150: Operating condition linkage anti-shake correction, for complex operating conditions such as energy storage system charging and discharging fluctuations, equipment start-up and shutdown, environmental disturbances, etc., the disturbance scenario is identified in real time and the judgment logic is adaptively adjusted to filter out minor over-limits of non-fault parameters. This implementation establishes a condition-linked anti-shake correction mechanism to address complex field conditions such as fluctuations in the charge / discharge rate of the energy storage system, equipment start-up and shutdown switching, environmental temperature and humidity disturbances, and changes in cabin ventilation. It enables real-time identification of disturbance scenarios and adaptive adjustment of the judgment logic. When the system identifies a transient process of a disturbance, it proactively adjusts its judgment strategy, filtering out minor parameter exceedances caused by non-fault reasons. After the condition stabilizes, it restores the normal judgment logic, ensuring that the risk assessment process adapts to complex dynamic field conditions and maintains the condition-adaptive capability of the judgment logic.

[0032] Step S151: Establish an operating condition linkage adaptive correction mechanism, connect with the operating data of energy storage PCS equipment, thermal management equipment and environmental sensing equipment in real time, and dynamically identify operating condition disturbance scenarios such as sudden changes in charging and discharging power, fan start and stop, fluctuations in ambient temperature and humidity, and switching of cabin ventilation. This implementation constructs a condition-linked adaptive correction mechanism. The system connects in real-time with the energy storage PCS converter, thermal management equipment, and environmental sensing equipment to obtain real-time operating data, simultaneously acquiring the system's power operating status, heat dissipation equipment operating status, and environmental parameter status. Based on multi-source operating condition data, the system dynamically identifies various operating condition disturbance scenarios such as sudden changes in charging and discharging power, start and stop of cooling fans, sudden changes in ambient temperature and humidity, and switching of cabin ventilation. It completes real-time marking and classification of disturbance scenarios, providing operating condition basis for subsequent anti-shake judgment logic switching.

[0033] Step S152: When the system detects a momentary fluctuation in the operating condition, it automatically activates the anti-shake judgment logic, extends the parameter observation and judgment time, and reduces the judgment weight of momentary fluctuation data. This implementation automatically activates anti-jitter judgment logic when the system detects instantaneous fluctuations in operating conditions. This extends the effective duration of parameter observation and judgment, reduces the weight of instantaneous fluctuation data in the risk assessment process, and prevents temporary parameters caused by instantaneous operating condition fluctuations from exceeding limits and participating in fault determination. The system maintains anti-jitter judgment mode throughout the disturbance duration, accumulating only persistent and stable parameter anomalies while ignoring short-term transient fluctuations, thus achieving adaptive optimization of the judgment logic under disturbed operating conditions.

[0034] Step S153: After the energy storage system's operating conditions return to stability and parameter fluctuations return to the normal range, the system automatically switches back to the normal judgment logic to complete the adaptive correction of parameter judgment under complex operating conditions.

[0035] This implementation method monitors the operating status of the energy storage system in real time. Once the charging and discharging power, equipment operating status, and environmental parameters have stabilized and the parameter fluctuations have returned to the normal range, the system automatically exits the anti-shake judgment mode and resumes the normal accurate judgment logic. The system completes the adaptive correction process for a single operating condition disturbance, waits for the next operating condition disturbance identification trigger, and continuously loops to achieve dynamic judgment adaptation under complex operating conditions, ensuring that the judgment logic under different operating conditions is stable and controllable.

[0036] Step S160: Algorithm engineering implementation and closed-loop optimization linkage, adapting and embedding the early warning algorithm into the existing BMS and fire control panel, building multi-system communication links to realize multi-level risk linkage and response, and at the same time optimizing model parameters through background data review and iteration.

[0037] This implementation method achieves the engineering implementation and closed-loop optimization configuration of the early warning algorithm, modularly embedding the core algorithms of multi-feature fusion, adaptive threshold iteration, and time-series trend prediction into the existing energy storage BMS main control equipment and fire protection main control equipment. The system establishes standardized multi-system communication links to achieve coordinated response to multiple levels of risks. Relying on its backend data accumulation capabilities, the system continuously reviews operational, early warning, and fault data, periodically iteratively optimizing model feature weights and threshold judgment rules, forming a complete closed-loop application system encompassing algorithm deployment, operational application, data review, and iterative upgrades.

[0038] Step S161: Modularly integrate the multi-feature fusion discrimination, adaptive threshold iteration, and time-series trend prediction algorithms into the existing energy storage BMS main control and fire protection host equipment, and complete the algorithm adaptation and deployment without changing the original hardware acquisition architecture. This implementation method embeds multi-feature fusion discrimination algorithm, adaptive threshold iteration algorithm, and time-series trend prediction algorithm into the existing energy storage BMS main control and energy storage fire protection host in a modular software form. The overall deployment process does not require replacement or modification of the original field sensor hardware acquisition architecture and is fully compatible with the existing hardware acquisition link and sampling method. Each algorithm module runs independently and works in concert to complete data preprocessing, risk fusion discrimination, threshold adaptive update, trend prediction, and hierarchical output functions, thereby realizing the software upgrade and expansion of the original equipment functions.

[0039] Step S162: The system unifies the data interaction protocol of BMS, energy storage fire protection system, thermal management system and PCS converter, opens up the real-time data communication link of multiple devices, and realizes bidirectional communication of operating condition data, battery status data, risk warning data and control commands; This implementation method unifies the data interaction and communication protocol between the BMS, energy storage fire protection system, thermal management system, and PCS converter, standardizes the data interaction format, data transmission cycle, and data interaction address of each system, and establishes a real-time bidirectional communication link between multiple devices. Operating condition data, battery status data, risk warning output data, and equipment control command data of each system can be exchanged in real time, achieving data synchronization, status synchronization, and command synchronization among multiple devices and systems, providing a communication foundation for coordinated control.

[0040] Step S163: Based on the four risk levels, the system matches corresponding maintenance prompts, temperature control interventions, emergency power outages, forced ventilation, and fire alarm activation with linked control logic. The backend system continuously accumulates equipment operation data, early warning records, and fault cases, and periodically iterates and optimizes feature weights and threshold judgment rules to complete the closed-loop iterative upgrade of the algorithm.

[0041] This implementation method configures corresponding linkage control execution logic according to four preset risk levels. Under normal conditions, the system maintains regular operation; for minor hidden dangers, it outputs maintenance and troubleshooting prompts; for general faults, it initiates thermal management temperature control intervention; and in critical thermal runaway states, it executes emergency power cut-off, forced ventilation, and fire suppression activation operations. The background system continuously accumulates equipment operation data, early warning records, and fault sample data during the power plant's operation, and regularly updates model feature weights and optimizes threshold rules, realizing a closed-loop upgrade process of long-term autonomous iteration of the algorithm model and continuous adaptation to changes in on-site operating conditions.

[0042] Based on the same inventive concept, please refer to Figure 2 The diagram shows a schematic block diagram of an energy storage battery thermal runaway detection system 100 provided in this application embodiment for performing the above-described energy storage battery thermal runaway detection method. The energy storage battery thermal runaway detection system 100 may include a communication unit 110, a machine-readable storage medium 120, and a processor 130.

[0043] In this embodiment, both the machine-readable storage medium 120 and the processor 130 are located within the energy storage battery thermal runaway detection system 100 and are separately configured. However, it should be understood that the machine-readable storage medium 120 may also be independent of the energy storage battery thermal runaway detection system 100 and may be accessed by the processor 130 via a bus interface. Alternatively, the machine-readable storage medium 120 may also be integrated into the processor 130 and may communicate with external systems via the communication unit 110.

[0044] The processor 130 is the control center of the energy storage battery thermal runaway detection system 100. It connects various parts of the system via various interfaces and lines, and performs various functions and processes data by running or executing software programs and / or modules stored in the machine-readable storage medium 120 and calling data stored in the machine-readable storage medium 120, thereby providing overall monitoring of the energy storage battery thermal runaway detection system 100. Optionally, the processor 130 may include one or more processors; for example, the processor 130 may integrate an application processor and a modem processor, wherein the application processor mainly handles the operating system, user interface, and applications, and the modem processor mainly handles wireless communication. It is understood that the modem processor may not be integrated into the processor. The machine-readable storage medium 120 stores machine-executable instructions for implementing the scheme of this application, and the processor 130 executes the machine-executable instructions stored in the machine-readable storage medium 120 to implement the energy storage battery thermal runaway detection method provided in the aforementioned method embodiments.

[0045] It should be noted that, in order to simplify the description of the present invention and thus help to understand one or more embodiments of the invention, multiple features may sometimes be grouped into one embodiment, drawing or description thereof in the foregoing description of the embodiments of the present invention.

[0046] The embodiments of this application have been described above with reference to the accompanying drawings. Unless otherwise specified, the embodiments and features in the embodiments of this application can be combined with each other. This application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

Claims

1. A method for detecting thermal runaway in an energy storage battery, characterized in that: Includes the following steps: The parameter hierarchical construction and multi-source data precise preprocessing are carried out by building a dynamic and static dual-dimensional fault characteristic parameter pool covering the entire thermal runaway cycle, improving the risk identification data source system, and adopting a multi-level composite filtering and data verification mechanism to eliminate invalid data caused by operating condition interference, equipment disturbance and sensor anomalies. Multi-feature weighted fusion risk discrimination targets the latent fault characteristics of weak collaborative shift of multiple parameters in early thermal runaway. By configuring differentiated dynamic feature weights through data training, it accurately distinguishes between normal operating condition disturbances and fault shifts, and calculates and outputs a unified comprehensive risk coefficient. The full life cycle adaptive threshold iteration relies on the battery aging status database to associate the cell health status and operating parameters. For battery aging and mixed new and old battery scenarios, the parameter baseline and alarm threshold are dynamically corrected. Combined with the periodic baseline iteration mechanism, the judgment deviation caused by aging and sensor drift is offset. The time-series trend prediction and hierarchical early warning and control system, based on the fitting of parameter change trends from continuous time-series operation data, predicts the early signs of thermal runaway failures, abandons the traditional passive instantaneous value judgment method, and establishes a multi-level risk classification system; The working condition linkage anti-shake correction is designed to identify disturbance scenarios in real time and adaptively adjust the judgment logic for complex working conditions such as energy storage system charging and discharging fluctuations, equipment start-up and shutdown, and environmental disturbances, and to filter out minor over-limits of non-fault parameters. The algorithm engineering implementation and closed-loop optimization are linked, and the early warning algorithm is adapted and embedded into the existing BMS and fire control panel. Multi-system communication links are built to realize multi-level risk linkage and response. At the same time, the model parameters are iteratively optimized through background data review.

2. The method for detecting thermal runaway of an energy storage battery according to claim 1, characterized in that: The hierarchical parameter construction and precise preprocessing of multi-source data improve the risk identification data source system by building a dynamic and static dual-dimensional fault characteristic parameter pool covering the entire thermal runaway cycle. Furthermore, a multi-level composite filtering and data verification mechanism is employed to eliminate invalid data caused by operating condition interference, equipment disturbances, and sensor anomalies, including: A two-layer fault characteristic parameter system is established, which is divided into two dimensions: static steady-state operation characteristics and dynamic change characteristics, comprehensively covering the state data acquisition dimensions of the entire battery operation stage. The system performs multi-level standardized purification processing on the raw battery operation data collected on site, stabilizes high-frequency instantaneous jump signals caused by equipment vibration and instantaneous charging and discharging fluctuations, and corrects and calibrates steady-state data deviations caused by ventilation airflow and environmental temperature changes. At the same time, data validity verification rules are set to remove and shield invalid extreme value data caused by sensor drift, line signal interference, and signal disconnection, and to select continuous, valid, and stable battery operation data as the basic input data for subsequent risk assessment.

3. The method for detecting thermal runaway of an energy storage battery according to claim 1, characterized in that: The multi-feature weighted fusion risk discrimination, targeting the latent fault characteristics of weak coordinated shifts in multiple parameters during early thermal runaway, uses data training to configure differentiated dynamic feature weights to accurately distinguish between normal operating condition disturbances and fault shifts, and calculates and outputs a unified comprehensive risk coefficient, including: A multi-dimensional feature collaborative association discrimination method is adopted to replace the traditional single-parameter independent judgment mode, and the model training is completed based on a large number of battery thermal runaway test samples and normal on-site working condition samples. Based on the sensitivity of each feature parameter to battery thermal runaway faults and the difference in response timing, differentiated dynamic weights are assigned to different parameters, so that the judgment weight of dynamic change features is higher than that of static numerical features. The system distinguishes between normal operating disturbances and fault parameter deviations in real time. It marks the operating condition disturbances caused by charge / discharge rate switching, ambient temperature fluctuations, and small parameter fluctuations caused by equipment start-up and shutdown. It performs correlation analysis on data with synchronous, unidirectional, and continuous deviations of multiple parameters. Based on the fusion calculation of preprocessed multi-dimensional data, it outputs the comprehensive risk coefficient of battery cells and modules.

4. The method for detecting thermal runaway of an energy storage battery according to claim 1, characterized in that: The full lifecycle adaptive threshold iteration relies on a battery aging status database that correlates cell health status and operating parameters. For scenarios involving battery aging and mixed installation of new and old batteries, it dynamically adjusts the parameter baseline and alarm thresholds. Combined with a periodic baseline iteration mechanism, it offsets judgment biases caused by aging and sensor drift, including: Construct a database of battery aging status throughout its entire life cycle, continuously collect and record operational data such as cycle count, SOH health, static internal resistance, and normal operating temperature range for each cell, and establish a linkage and adaptation mechanism between battery health status and judgment parameters. Based on the real-time aging status of the battery cells, the system differentiates the baseline and alarm threshold of each characteristic parameter, adjusts the allowable range of parameter fluctuations and fault identification sensitivity for old battery cells with high cycle count and low SOH, and sets standard judgment thresholds for new battery cells. The system is set up with a monthly baseline iteration update mechanism to continuously collect operating data under steady-state battery conditions, dynamically update the normal fluctuation range of each parameter, and calibrate the baseline offset caused by battery aging and sensor temperature drift in real time.

5. The method for detecting thermal runaway of an energy storage battery according to claim 1, characterized in that: The aforementioned time-series trend extrapolation and hierarchical early warning and control system, based on fitting parameter change trends from continuous time-series operational data, predicts the early signs of thermal runaway fault evolution. It abandons the traditional passive instantaneous value judgment method and establishes a multi-level risk classification system, including: Continuously collect continuous time-series data of various battery operating parameters, calculate the change gradient and growth rate slope of each parameter in real time, and fit the continuous change law of parameters and predict the evolution trend of subsequent operating parameters through linear fitting and time-series trend extrapolation algorithms. The system combines the duration of parameter degradation, degradation rate, and multi-feature collaborative abnormal states to classify battery operation risks into levels, namely normal operation, minor hidden dangers, general faults, and critical thermal runaway. For different risk levels, it configures four types of standardized output instructions: prompts, warnings, alarms, and emergency linkage, to achieve pre-level hierarchical control of battery thermal runaway risks.

6. The method for detecting thermal runaway of an energy storage battery according to claim 1, characterized in that: The aforementioned condition-linked anti-shake correction addresses complex operating conditions such as energy storage system charging and discharging fluctuations, equipment start-up and shutdown, and environmental disturbances. It identifies disturbance scenarios in real time and adaptively adjusts the judgment logic to filter out minor deviations in non-fault parameters, including: Establish an adaptive correction mechanism for operating conditions, connect with the operating data of energy storage PCS equipment, thermal management equipment and environmental sensing equipment in real time, and dynamically identify operating condition disturbance scenarios such as sudden changes in charging and discharging power, fan start and stop, fluctuations in ambient temperature and humidity, and switching of cabin ventilation. When the system detects a momentary fluctuation in operating conditions, it automatically activates the anti-shake judgment logic, extends the parameter observation and judgment time, and reduces the judgment weight of momentary fluctuation data. Once the operating conditions of the energy storage system return to stability and parameter fluctuations return to the normal range, the system will automatically switch back to the normal judgment logic and complete the adaptive correction of parameter judgment under complex operating conditions.

7. The method for detecting thermal runaway of an energy storage battery according to claim 1, characterized in that: The algorithm engineering implementation and closed-loop optimization are linked, adapting the early warning algorithm into existing BMS and fire control panels, constructing multi-system communication links to achieve multi-level risk linkage and response, and simultaneously optimizing model parameters through background data review and iteration, including: The multi-feature fusion discrimination, adaptive threshold iteration, and time-series trend prediction algorithms are modularly embedded into the existing energy storage BMS main control and fire protection host equipment, and the algorithm adaptation and deployment are completed without changing the original hardware acquisition architecture. The system unifies the data interaction protocol of BMS, energy storage fire protection system, thermal management system, and PCS converter, and opens up real-time data communication links between multiple devices to achieve bidirectional communication of operating condition data, battery status data, risk warning data and control commands; Based on four risk levels, the system matches corresponding maintenance prompts, temperature control interventions, emergency power outages, forced ventilation, and fire alarm activation with linked control logic. The backend system continuously accumulates equipment operation data, early warning records, and fault cases, periodically iterating and optimizing feature weights and threshold judgment rules to complete the closed-loop iterative upgrade of the algorithm.

8. A thermal runaway detection system for energy storage batteries, characterized in that, include: processor; A machine-readable storage medium for storing machine-executable instructions of the processor; The processor is configured to execute the thermal runaway detection method for energy storage batteries according to any one of claims 1 to 7 by executing the machine-executable instructions.

Images