Energy storage system fire early warning and emergency disposal method and system based on digital twinning

Abstract

This invention discloses a method and system for fire early warning and emergency response in energy storage systems based on digital twins, belonging to the field of energy storage system safety and intelligent emergency response. The method includes: constructing a hybrid sensing network of distributed fiber optic sensors and micro IoT nodes to synchronously collect multi-dimensional data from the energy storage system; constructing a three-level, multi-scale, hierarchically coupled digital twin to simulate the coordinated response effect of electronic switches performing electrical disconnection and mechanical switches performing physical isolation; deriving the critical threshold and evolution path of thermal runaway, generating graded early warning commands; and executing the actions of electronic and mechanical switches according to risk levels. This invention employs a dual-drive logic of AI-driven predictive analysis and real-time data acquisition, combined with high-fidelity digital twin simulation and fusion algorithms, which can accurately capture the critical threshold and evolution path of thermal runaway, significantly reducing the probability of risk propagation and improving the predictability and reliability of early warning for energy storage systems.

Description

Technical Field

[0001] This invention relates to the field of energy storage system safety and intelligent emergency response, and in particular to a method and system for fire early warning and emergency response of energy storage systems based on digital twins. Background Technology

[0002] In the field of thermal runaway emergency response for energy storage systems, existing technologies have significant shortcomings and are unable to meet the safety requirements of large-scale energy storage power stations. Regarding early warning mechanisms, mainstream solutions rely on static threshold alarms, which cannot capture microscopic nonlinear changes such as SEI film decomposition and trace gas production in the early stages of thermal runaway. Furthermore, the monitoring network is sparse, with a missed detection rate of over 30% for local hotspots, and the lack of multi-parameter fusion analysis capabilities leads to a false alarm rate exceeding 15%. In terms of response mechanisms, mechanical fuses or single switches are commonly used, with response times of 100-500 milliseconds, far slower than the tens of milliseconds of thermal runaway propagation. The lack of coordination between electronic and mechanical switches often leads to cascading failures due to incomplete isolation, causing the fault to spread from a single battery pack to the entire energy storage unit. Fire suppression systems are mostly total flooding type, which is inefficient and prone to secondary disasters. In terms of system architecture, data acquisition, risk assessment, and response execution are isolated information silos, lacking closed-loop optimization supported by digital twins. Historical data is only used for traceability and cannot be used to build predictive models for self-iteration, resulting in stagnant protection capabilities. These defects result in a success rate of less than 60% and a warning delay of more than 30 seconds, which seriously restricts the large-scale application of lithium battery energy storage. Summary of the Invention

[0003] Purpose of the invention: In view of the above problems, the purpose of this invention is to provide a method and system for fire early warning and emergency response of energy storage systems based on digital twins.

[0004] Technical solution: One aspect of the present invention is a method for fire early warning and emergency response of energy storage systems based on digital twins, comprising the following steps:

[0005] Step 1: Construct a hybrid sensing network of distributed fiber optic sensors and micro IoT nodes to synchronously collect multi-dimensional data from the energy storage system; then, fuse the collected multi-dimensional data using a dynamic confidence fusion mechanism to obtain fused data.

[0006] Step 2: Based on the fused data, construct a three-level, multi-scale, hierarchically coupled digital twin. Use the digital twin to simulate the coordinated handling effect of electronic switches performing electrical disconnection and mechanical switches performing physical isolation under fault conditions. Dynamically correct the model parameters of each level in the digital twin according to the actual action data.

[0007] Step 3: Using the GNN-Transformer fusion algorithm, a hierarchical heat conduction-electrical parameter-microscopic feature coupled network is constructed. With the support of the digital twin, the critical threshold and evolution path of thermal runaway are deduced, and hierarchical early warning instructions are generated.

[0008] Step 4: Based on the graded early warning instructions, execute the actions of electronic and mechanical switches according to the risk level; receive the physical status judgment results in real time, and if an emergency physical risk exceeding the preset threshold is detected, directly trigger the emergency backup action; at the same time, coordinate with the precision fire protection system to perform targeted disposal, and the status feedback unit will transmit the action data back to the digital twin in real time to support closed-loop optimization and disposal verification, so as to realize multi-level safety protection of pre-disaster intervention and emergency backup.

[0009] Preferably, step 2 includes:

[0010] Step 21: Construct a digital twin with a three-level structure. This digital twin includes: using the battery cell as the smallest unit, constructing a battery cell-level model, and integrating micro-strain and local gas generation rate parameters to simulate the early internal reactions of thermal runaway; wherein, the mathematical formula for the battery cell-level model is:

[0011] ,

[0012] in, These are the corrected model parameters. These are the original parameters before correction, including the SEI film decomposition rate coefficient at the cell level and the thermal conductivity coefficient at the module level. For correction factor, This refers to the relative deviation of the parameters;

[0013] The module is used as a mid-level unit, composed of individual battery cells. A module-level model is constructed. By collecting the microscopic parameters of each battery cell within the module, and integrating macroscopic temperature field data and current distribution data at the module level, correlation analysis is performed on the temperature field and current distribution data to characterize the coupling and conduction laws of the module's thermal and electrical parameters. The mathematical formula for the module-level model is as follows:

[0014] ,

[0015] in, Indicates the density of the material. This represents the specific heat capacity at constant pressure. For the module temperature field, Represents the gradient or divergence operator. The thermal conductivity coefficient, For current density, For internal resistance, The convective heat transfer coefficient is... Indicates ambient temperature;

[0016] The energy storage compartment is used as the top-level unit, assembled from modules to construct a hierarchical model of the energy storage compartment. This model integrates the coupling and conduction laws of the thermal and electrical parameters of each module, simulating the heat diffusion path and fire suppression effects of the entire system, forming a complete physical digital mirror from microscopic reactions to the macroscopic system. The mathematical formulas for the hierarchical model of the energy storage compartment are as follows:

[0017] ,

[0018] in, The total heat capacity of the energy storage warehouse. The average temperature inside the cabin. Indicates ambient temperature. It is the aggregation of the heat generation power of each module. Indicates the number of modules. The total thermal resistance from the cabin to the environment; The heat transfer power of the fire protection system is expressed as:

[0019] ,

[0020] in, For gas flow rate, For specific heat capacity, Temperature of the injected gas;

[0021] Step 22: Receive real-time action data, including switch operation status and fire response effectiveness, and perform multi-dimensional comparison between the measured values ​​and the simulated values ​​of the digital twin. Quantify the differences using a relative deviation formula.

[0022] ,

[0023] in, The relative deviation of the core parameters These are measured values. The simulated value of the digital twin;

[0024] Step 23: Make a judgment based on the magnitude of the deviation. If If the value is greater than 1%, a correction process will be triggered, indicating that the model parameters at each level in the digital twin or the sensors need to be adjusted or calibrated.

[0025] In the correction process, the mathematical formulas of the battery cell hierarchical model are applied to dynamically update the model parameters of each level, and the model parameters are adjusted by scaling the deviation ratio and smoothing.

[0026] Step 24: Based on the corrected digital twin, combined with the real-time acquired data, real-time twin data is obtained through iterative calculation using an electrochemical-thermal coupling model.

[0027] Preferably, step 3 includes:

[0028] Step 31: Construct a hierarchical heat conduction-electrical parameter-microscopic feature coupling network, including a feature extraction layer and a cross-feature temporal fusion layer. The feature extraction layer aggregates microscopic features, electrical parameter features and macroscopic features, and the cross-feature temporal fusion layer fuses all parameters to obtain a fused feature vector.

[0029] Step 32: By learning from historical data and real-time twin data, the risk level is assessed using a risk quantification formula. The formula is as follows:

[0030] ,

[0031] in, As a comprehensive risk index, This is the normalized value of the local H2 / CO gas production rate. This is the normalized value of the dynamic internal resistance increase. This is the normalized value for module temperature. This is the normalized value for the number of abnormal individuals. This represents the normalized value of the micro-strain.

[0032] Step 33, calculate the critical threshold for thermal runaway. The calculation formula is as follows:

[0033] ,

[0034] in, This is the critical threshold for thermal runaway. Weights are assigned to historical data. This represents the average thermal runaway threshold under similar operating conditions over a month. The current parameter trend value calculated for real-time twin data;

[0035] Step 34: GNN is responsible for constructing the spatial topology of the energy storage system. In this spatial topology, the battery cells are used as nodes, and the thermal-electric connection relationship of the cells in the module is used as edges. By aggregating the features of adjacent nodes, the transmission of abnormal parameters in space is simulated to obtain the spatial diffusion chain.

[0036] The historical sequence of each parameter is analyzed using a self-attention mechanism, and the time evolution curve is predicted by Transformer.

[0037] Step 35: Couple the spatial diffusion chain and temporal evolution curve into the digital twin and perform high-fidelity simulation. Using the current fused feature vector as the initial state, the simulation is performed step by step in the twin sandbox. At each step, the parameters of each node are updated, and it is checked whether any issues are triggered. After the iteration is complete, a complete path with key spatiotemporal nodes marked is output.

[0038] Step 36: Finally, based on the safety operation standards and graded disposal requirements of the energy storage system, the comprehensive risk index is divided into three risk ranges: low-risk scenario, medium-risk scenario, and high-risk scenario. In the low-risk scenario, an electronic switch trigger command is issued; in the medium-risk scenario, an electronic switch priority and mechanical switch supplementary trigger command is issued; in the high-risk scenario, an electronic switch and mechanical switch synchronous trigger command is issued.

[0039] Preferably, step 4 includes:

[0040] The urgency score is calculated based on the magnitude of parameter exceedance and the abnormal spread rate, using the following formula:

[0041] ,

[0042] in, Rate the urgency level. For real-time voltage, The safe threshold for voltage, This represents the voltage normalized reference value. For real-time voltage, This is the safe threshold for voltage. This represents the current number of abnormal individuals. This represents the number of abnormal individuals at the previous sampling time. The sampling time interval, This is a normalization coefficient used to map rate values ​​to a reasonable fractional range;

[0043] when When an emergency is identified, emergency response is immediately triggered, and the risk type, target handling area, action sequence parameters, and response logic are simultaneously defined.

[0044] During the switching action, the operating parameters of the two types of electronic switches and mechanical switches are collected in real time, and the battery temperature and voltage change data of the target area after the treatment are collected simultaneously and transmitted back to the digital twin in real time via industrial Ethernet.

[0045] Preferably, step 1 includes:

[0046] Distributed fiber optic sensors are embedded along the length of the battery module to collect the continuous temperature field and micro-strain distribution on the battery surface in real time.

[0047] The micro IoT node integrates a voltage sensor, a dynamic internal resistance monitoring module, and a gas sensor to collect data on battery cell voltage, dynamic internal resistance, and local gas generation rate at a synchronous frequency.

[0048] The two types of sensors align their data using timestamp synchronization technology to generate multi-dimensional data.

[0049] Preprocess the collected multi-dimensional data;

[0050] Based on the sensor health index and environmental interference coefficient, the fusion weights of multi-source data are dynamically adjusted, using the following formula:

[0051] ,

[0052] In the formula, For the first Sensor-like fusion weights, For sensor health index, Environmental interference coefficient;

[0053] Multi-dimensional data is merged using fusion weights to obtain fused data.

[0054] Another aspect of the present invention provides a fire early warning and emergency response system for energy storage systems based on digital twins, comprising:

[0055] The data acquisition module is used to collect real-time data on the operating conditions of the energy storage unit, electrical parameters of electronic switches, physical operation data of mechanical switches, and ambient temperature and humidity data. The acquired data is preprocessed and then transmitted in two paths: one path to the digital twin construction module and the other path directly to the physical execution module.

[0056] The digital twin construction module constructs a digital twin of the energy storage system based on preprocessed data. It maintains dynamic synchronization between the physical system and the digital twin through real-time data fusion and obtains simulation data through the digital twin simulation, providing a simulation verification environment for the AI ​​inference module.

[0057] The AI ​​simulation module, based on simulation data and real-time acquired data, uses the GNN-Transformer fusion algorithm to predict thermal runaway in advance; and according to the prediction results and real-time physical conditions, the risk is divided into low, medium and high levels, and corresponding graded instructions for coordinated action of electronic switches and mechanical switches are generated.

[0058] The physical execution module synchronously acquires hierarchical instructions and real-time physical condition judgment results, drives electronic switches to cut off electrical circuits, controls mechanical switches to perform physical isolation, and links with the precision fire protection system for targeted handling; at the same time, it transmits the effectiveness of the dual switch actions and the operating status of fire protection equipment back to the data acquisition module and the digital twin construction module in real time.

[0059] Beneficial effects: Compared with the prior art, the significant advantages of this invention are:

[0060] 1. This invention adopts a dual-drive logic of AI-driven anticipatory prediction and real-time direct judgment based on data acquisition. Combined with digital twin high-fidelity simulation and improved fusion algorithm, compared with existing technologies that lack early warning mechanisms, it can accurately capture the critical threshold and evolution path of thermal runaway in advance, significantly reduce the probability of risk spread, reserve sufficient response time for emergency response, and greatly improve the anticipatory nature and reliability of early warning for energy storage systems.

[0061] 2. This invention proposes an electronic-mechanical switch dual redundancy collaborative mechanism. Through risk-level triggering, precise timing control and CANopen bus synchronization technology, compared with the traditional single switch solution, it not only avoids the problem of electronic switches being prone to failure under long-term pressure, but also makes up for the defect of mechanical switch response lag. It achieves a dual barrier of millisecond-level electrical disconnection and second-level physical isolation, significantly improving the speed and thoroughness of thermal runaway handling.

[0062] 3. This invention constructs a closed-loop optimization mechanism based on data acquisition, twin modeling, AI inference, and execution feedback. Combining dynamic confidence fusion data preprocessing technology with targeted and precise fire-fighting linkage strategies, it can dynamically adapt to changes in the operating conditions of energy storage systems, effectively avoiding misjudgments and omissions caused by data distortion. At the same time, it upgrades traditional post-disaster fire fighting to pre-disaster intervention, emergency replenishment, and multi-dimensional prevention and control strategies, significantly reducing the operational risks and equipment wear of energy storage systems, balancing safety and economy, and providing reliable guarantees for the stable operation of large-scale energy storage power stations. Attached Figure Description

[0063] Figure 1 This is a flowchart of the method described in this invention;

[0064] Figure 2 This is a flowchart of the method described in this invention;

[0065] Figure 3 A flowchart for building a digital twin;

[0066] Figure 4 The results show a comparison between the invention and traditional methods in terms of early warning and response time;

[0067] Figure 5 The results show the reliability comparison of energy storage systems of different sizes;

[0068] Figure 6 This is a structural framework diagram of the system described in this invention. Detailed Implementation

[0069] The embodiments of the present invention will be further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are merely illustrative of the present invention and not intended to limit the scope of the invention. Furthermore, it should be noted that, for ease of description, the accompanying drawings show only the parts relevant to the embodiments of the present invention, and not all structures.

[0070] In the following description, specific details such as target system architecture and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods are omitted so as not to obscure the description of this application with unnecessary detail.

[0071] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0072] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0073] Furthermore, in the description of this application and the appended claims, the terms "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0074] References to "one embodiment" or "some embodiments" in this specification mean that one or more embodiments of this application include the target features, structures, or characteristics described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized.

[0075] Example 1

[0076] The fire early warning and emergency response method for energy storage systems based on digital twins described in this embodiment refers to an energy storage system whose core consists of individual battery cells and battery modules, serving as the main carrier for energy storage. These cells are also the core objects for system status monitoring, with their operating parameters such as voltage and temperature being collected in real time. Under different risk scenarios, corresponding protection mechanisms ensure its operational safety. This invention's fire early warning and emergency response method for energy storage systems based on digital twins combines... Figure 1 and Figure 2 As shown, the specific steps include:

[0077] Step 1: Construct a hybrid sensing network of distributed fiber optic sensors and micro IoT nodes to synchronously collect multi-dimensional data from the energy storage system; then, fuse the collected multi-dimensional data using a dynamic confidence fusion mechanism to obtain fused data.

[0078] Further, step 1 includes:

[0079] Distributed fiber optic sensors are embedded along the length of the battery module to collect the continuous temperature field and micro-strain distribution on the battery surface in real time.

[0080] The micro IoT node integrates a voltage sensor, a dynamic internal resistance monitoring module, and a gas sensor to collect data on battery cell voltage, dynamic internal resistance, and local gas generation rate at a synchronous frequency.

[0081] The two types of sensors align their data using timestamp synchronization technology to generate multi-dimensional data.

[0082] Preprocess the collected multi-dimensional data;

[0083] Based on the sensor health index and environmental interference coefficient, the fusion weights of multi-source data are dynamically adjusted, using the following formula:

[0084] ,

[0085] In the formula, For the first Sensor-like fusion weights, For sensor health index, Environmental interference coefficient;

[0086] Multi-dimensional data is merged using fusion weights to obtain fused data.

[0087] In one example, distributed fiber optic sensors are embedded along the length of the battery module, with one monitoring point per meter. These sensors collect the continuous temperature field and micro-strain distribution on the battery surface in real time, capturing the trend of global parameter changes at the module level. Micro IoT nodes are evenly distributed at a density of one per 2-3 batteries. These nodes integrate high-precision voltage sensors, dynamic internal resistance monitoring modules, and micro gas sensors. They collect core parameters such as battery cell voltage, dynamic internal resistance, and local H2 / CO gas production rate at millisecond-level synchronization frequency. The two types of sensors achieve data alignment through timestamp synchronization technology, forming a multi-dimensional data matrix of global field distribution data and single-point high-frequency parameter data. This provides a full-dimensional, blind-spot-free parameter sensing foundation for the dual-redundancy collaborative mechanism of electro-mechanical switches.

[0088] The data preprocessing stage focuses on dynamic confidence fusion, ensuring data reliability through both physical constraint verification and dynamic weight allocation: First, outlier identification is performed based on the battery physical constraint model, with clearly defined parameter threshold ranges. For example, data with voltage exceeding ±20% of the rated value, temperature exceeding the 80℃ thermal runaway threshold, or local gas generation rate exceeding 50ppm / min are considered invalid outliers and directly removed. Transient jump data caused by electromagnetic interference and vibration are smoothed using sliding window mean filtering. The specific calculation formula is as follows:

[0089] ,

[0090] In the formula, For the smoothed data of the i-th sampling point, The data is the j-th original sampling point; a fixed window size is used to ensure the consistency and repeatability of the data smoothing effect.

[0091] Subsequently, a dynamic confidence fusion mechanism is introduced. Based on the sensor health index and environmental interference coefficient, the fusion weights of multi-source data are dynamically adjusted. Sensor data with high health and low interference receive a higher weight ratio, ensuring hierarchical optimization of data reliability. The formula first calculates the health-interference ratio of each sensor, then normalizes it, ultimately achieving a quantitative allocation where sensor data with high health and low interference receive a higher weight ratio, thus realizing hierarchical fusion of multi-source data for reliability.

[0092] Step 2: Based on the fused data, construct a three-level, multi-scale, hierarchically coupled digital twin. Use the digital twin to simulate the coordinated handling effect of electronic switches performing electrical disconnection and mechanical switches performing physical isolation under fault conditions. Dynamically correct the model parameters of each level in the digital twin according to the actual action data.

[0093] Specifically, the constructed three-level, multi-scale, layered, coupled digital twin provides a high-fidelity sandbox environment for AI simulation and the dual-redundancy collaborative mechanism of electro-mechanical switches. Battery cells, modules, and energy storage chambers are hierarchically aggregated layer by layer. The battery cell, as the smallest unit, corresponds to the microscale, integrating micro-strain, local gas generation rate, and other micro-parameters to simulate early internal reactions of thermal runaway, such as SEI film decomposition and electrode expansion. The module, a mid-level unit, is composed of multiple battery cells, integrating macro-temperature field, current distribution, and other mid-level data to characterize the coupling and conduction laws of the module's thermal-electrical parameters. The energy storage chamber, a top-level unit, is composed of multiple modules, corresponding to the system scale, and is used to simulate the thermal diffusion path and fire suppression effects of the entire system, forming a complete physical digital mirror from microscopic reactions to the macroscopic system.

[0094] Combination Figure 3 As shown, step 2 further includes:

[0095] Step 21: Construct a digital twin with a three-level structure. This digital twin includes: using the battery cell as the smallest unit, constructing a battery cell-level model, and integrating micro-strain and local gas generation rate parameters to simulate the early internal reactions of thermal runaway; wherein, the mathematical formula for the battery cell-level model is:

[0096] ,

[0097] in, These are the corrected model parameters. These are the original parameters before correction, including the SEI film decomposition rate coefficient at the cell level and the thermal conductivity coefficient at the module level. This is a correction factor, with a value ranging from 0.05 to 0.1. This formula addresses the relative deviation of parameters, enhances the mapping accuracy between physical and digital quantities, optimizes the training samples and feature weights of the AI ​​inference algorithm, and improves the foresight and accuracy of risk prediction. Through a closed loop of feedback, analysis, correction, and optimization, it continuously reduces the error between the twin and the physical system, consolidates the reliability of AI inference, provides dynamically adaptable technical support for the pre-disaster intervention strategy of electro-mechanical dual-switch collaboration, and ensures the long-term stable operation of fire early warning and emergency response of energy storage systems.

[0098] The module is used as a mid-level unit, composed of individual battery cells. A module-level model is constructed. By collecting the microscopic parameters of each battery cell within the module, and integrating macroscopic temperature field data and current distribution data at the module level, correlation analysis is performed on the temperature field and current distribution data to characterize the coupling and conduction laws of the module's thermal and electrical parameters. The mathematical formula for the module-level model is as follows:

[0099] ,

[0100] in, Indicates the density of the material. This represents the specific heat capacity at constant pressure. For the module temperature field, Represents the gradient or divergence operator. The thermal conductivity coefficient, For current density, For internal resistance, The convective heat transfer coefficient is... Indicates ambient temperature.

[0101] The energy storage compartment is used as the top-level unit, assembled from modules to construct a hierarchical model of the energy storage compartment. This model integrates the coupling and conduction laws of the thermal and electrical parameters of each module, simulating the heat diffusion path and fire suppression effects of the entire system, forming a complete physical digital mirror from microscopic reactions to the macroscopic system. The mathematical formulas for the hierarchical model of the energy storage compartment are as follows:

[0102] ,

[0103] in, The total heat capacity of the energy storage warehouse. The average temperature inside the cabin. Indicates ambient temperature. It is the aggregation of the heat generation power of each module. Indicates the number of modules. The total thermal resistance from the cabin to the environment; The heat transfer power of the fire protection system is expressed as:

[0104] ,

[0105] in, For gas flow rate, For specific heat capacity, This refers to the temperature of the injected gas.

[0106] Specifically, the battery cell is the smallest unit. Microscopic parameters such as micro-strain and local gas generation rate are integrated and combined with a battery physical constraint model. The collected microscopic parameters are mapped to the electrochemical and physical changes within the battery. Ultimately, these early internal reactions of thermal runaway are reproduced in a digital twin, forming a physical digital mirror at the battery cell level. Simulating early internal reactions of thermal runaway, such as SEI film decomposition and electrode expansion, the microscopic reaction simulation results of the digital twin serve as input data for AI inference. This data is used to analyze the evolution of parameters over time, assisting AI in predicting the development trend of thermal runaway and determining the risk level. Simultaneously, these simulation results are also transmitted to the physical execution module, providing real-time risk status information for the emergency backup action of the electronic-mechanical switch dual-redundancy collaborative mechanism. This forms the basis for the precise triggering of subsequent response actions.

[0107] Specifically, taking the module as the middle layer, macroscopic temperature field and current distribution data are integrated. By collecting microscopic parameters of each battery cell within the module, such as microstrain, local H2 / CO gas production rate, cell voltage, and dynamic internal resistance, microstrain is used to characterize the expansion of the battery cell shell and internal structural deformation, local H2 / CO gas production rate is used to characterize electrolyte decomposition and early gas evolution characteristics, and cell voltage and dynamic internal resistance are used to characterize electrical anomalies and polarization changes of the battery cell. At the same time, macroscopic temperature field data and current distribution data at the module level are integrated. Based on the middle-scale modeling of the digital twin, the temperature field and current distribution data are correlated and analyzed. Since thermal and electrical parameters have mutual influence, the coupling and transmission law of thermal and electrical parameters of the module can be characterized. This can serve as the core basis for subsequent risk management, to predict the spread trend of module runaway, and to assist AI in inferring and judging the risk level.

[0108] Specifically, taking the energy storage compartment as the top layer, the thermal-electric coupling conduction laws of each module are integrated. Combined with the physical layout of the energy storage compartment, the temperature field and thermal runaway trend data at the module level are aggregated upwards. Through top-level modeling of the digital twin, the direction and rate of heat diffusion in the entire energy storage compartment, as well as the path of thermal runaway from a single module to other modules and the entire compartment, are simulated to simulate the system-level thermal diffusion path. The action of the fire protection system is reproduced in the digital twin. Combined with the thermal diffusion path in the energy storage compartment, the temperature drop trend in the compartment after the implementation of fire protection measures, the blocking effect of thermal runaway propagation, and the degree of inhibition of electrolyte decomposition gas production by nitrogen are simulated, thereby obtaining the fire protection effect and forming a complete physical digital mirror from microscopic reaction to macroscopic system. Meanwhile, the standardized full data is stored in a structured manner according to the physical entity level and time series. A physical level index is established in the spatial dimension to support fast query by location. In the time dimension, time series data is stored at 1-second intervals. The second-level sampling data of the past 24 hours and the hour-level sampling data of the past 30 days are retained and associated with environmental parameters to provide a complete dataset for AI to infer and analyze the evolution of parameters over time.

[0109] Step 22: Receive real-time action data, including switch operation status and fire response effectiveness, and perform multi-dimensional comparison between the measured values ​​and the simulated values ​​of the digital twin. Quantify the differences using a relative deviation formula.

[0110] ,

[0111] in, The relative deviation of the core parameters These are measured values. The simulated value of the digital twin;

[0112] Step 23: Make a judgment based on the magnitude of the deviation. If If the deviation exceeds 1%, a correction process will be triggered, indicating that the model parameters at each level in the digital twin or the sensors need to be adjusted to control the core parameter deviation within 1% and provide reliable digital twin support for the coordinated action of electronic-mechanical switches.

[0113] In the correction process, the mathematical formulas of the battery cell hierarchical model are applied to dynamically update the model parameters of each level, and the model parameters are adjusted by scaling the deviation ratio and smoothing.

[0114] Step 24: Based on the corrected digital twin, combined with the real-time acquired data, real-time twin data is obtained through iterative calculation using an electrochemical-thermal coupling model.

[0115] Building upon this foundation, a high-fidelity simulation sandbox is constructed to simulate thermal conduction and electrical parameter coupling relationships between batteries. It embeds emergency response logic for millisecond-level electronic switch cutoff and second-level mechanical switch isolation, reproducing the state and simulating parameter changes and response effects, providing an interactive and simulable experimental environment for AI simulation. Finally, three core data types are extracted from structured storage: real-time status, historical data from similar operating conditions over the past 30 days, and coupled network structure. This allows AI to simulate thermal runaway thresholds and evolution paths 3-5 minutes in advance and generate graded collaborative instructions for electronic-mechanical switches.

[0116] Step 3: Using the GNN-Transformer fusion algorithm, a hierarchical coupled network of heat conduction, electrical parameters, and microscopic features is constructed. With the support of the digital twin, the critical threshold and evolution path of thermal runaway are deduced, and hierarchical early warning instructions are generated.

[0117] Furthermore, step 3 includes:

[0118] Step 31: Construct a hierarchical heat conduction-electrical parameter-microscopic feature coupling network, including a feature extraction layer and a cross-feature temporal fusion layer. The feature extraction layer aggregates microscopic features, electrical parameter features, and macroscopic features, and the cross-feature temporal fusion layer fuses all parameters to obtain a fused feature vector.

[0119] Specifically, relying on an improved GNN-Transformer fusion algorithm, a hierarchical thermal conduction-electrical parameter-microscopic feature coupling network is innovatively constructed. This network is based on a hierarchical, cross-feature coupling structure built on the improved GNN-Transformer fusion algorithm. The hierarchical feature extraction layer corresponds to the three-level structure of the digital twin: battery cell-module-energy storage compartment. The battery cell is used as a microscopic node, aggregating its micro-strain, gas production rate, and other microscopic features. The module is used as a mid-level node, aggregating mid-level thermal-electrical parameters such as cell features, module temperature, and dynamic internal resistance. The energy storage compartment is used as a system-level node, aggregating macroscopic parameters such as module features and the number of abnormal cells. Ultimately, this achieves hierarchical feature association from micro to macro. The cross-feature temporal fusion layer introduces the Transformer's attention mechanism to capture the evolution of the above-mentioned hierarchical features over time, achieving cross-dimensional coupling of thermal conduction features, electrical parameter features, and microscopic features. The input to this network is the collected and preprocessed feature parameters of each level, including microscopic features: the normalized value of the local H2 / CO gas production rate. Microstrain normalized value Electrical parameter characteristics: Normalized values ​​of local H2 / CO production rates Macroscopic characteristics: Normalized module temperature T, normalized abnormal unit number The network outputs a fusion feature vector that integrates hierarchical correlation and temporal patterns. This vector is directly used for subsequent risk level quantification and calculation of thermal runaway critical threshold formulas, accurately capturing the entire chain of local anomalies from microscopic reactions to macroscopic diffusion.

[0120] Step 32: By learning from historical data and real-time twin data, the risk level is assessed using a risk quantification formula. The formula is as follows:

[0121] ,

[0122] in, As a comprehensive risk index, This is the normalized value of the local H2 / CO gas production rate. This is the normalized value of the dynamic internal resistance increase. This is the normalized value for module temperature. This is the normalized value for the number of abnormal individuals. This is the normalized value for microstrain.

[0123] Based on this comprehensive risk index, the critical threshold and evolution path of thermal runaway can be predicted in advance. The preprocessed multi-dimensional normalized parameters are fed into a hierarchical thermal conduction-electrical parameter-micro feature coupling network constructed by an improved GNN-Transformer fusion algorithm. This network learns historical data and real-time twin data of similar operating conditions over the past 30 days. By linking cross-level features of battery cells-modules-energy storage compartments through GNN and capturing the temporal evolution of features through Transformer, the network accurately extracts the full-link pattern of local anomalies from microscopic reaction to macroscopic diffusion.

[0124] Step 33, calculate the critical threshold for thermal runaway. The calculation formula is as follows:

[0125] ,

[0126] in, This is the critical threshold for thermal runaway. Weights are assigned to historical data. This represents the average thermal runaway threshold under similar operating conditions over a month. The current parameter trend value calculated for real-time twin data.

[0127] First, the comprehensive risk index is calculated using the risk level quantification formula to clarify the current risk level. Then, the thermal runaway critical threshold is obtained using the thermal runaway critical threshold formula to clarify the distance between the current state and the critical state. Based on this, the process of simulating parameters approaching the critical threshold is simulated. Finally, the complete process path from the current state to thermal runaway is simulated in advance. This path will serve as the basis for generating graded early warning instructions that support the coordination of electromechanical switches.

[0128] Step 34: GNN is responsible for constructing the spatial topology of the energy storage system. In this spatial topology, the battery cells are used as nodes, and the thermal-electric connection relationship of the cells in the module is used as edges. By aggregating the features of adjacent nodes, the transmission of abnormal parameters in space is simulated to obtain the spatial diffusion chain.

[0129] The historical sequence of each parameter is analyzed using a self-attention mechanism, and the time evolution curve is predicted by Transformer.

[0130] Step 35: Couple the spatial diffusion chain and temporal evolution curve into the digital twin and perform high-fidelity simulation. Using the current fused feature vector as the initial state, the simulation is performed step by step in the twin sandbox. At each step, the parameters of each node are updated, and it is checked whether any issues are triggered. After the iteration is completed, a complete path with key spatiotemporal nodes marked is output.

[0131] If touched The current moment of impact and the corresponding spatial node are immediately marked as critical risk nodes, the risk propagation path is output, and the result is sent to the subsequent risk classification and disposal decision-making steps to trigger the corresponding switch control command in step 36.

[0132] Step 36: Finally, based on the safety operation standards and graded disposal requirements of the energy storage system, the comprehensive risk index is divided into three risk ranges: low-risk scenario, medium-risk scenario, and high-risk scenario. In the low-risk scenario, an electronic switch trigger command is issued; in the medium-risk scenario, an electronic switch priority and mechanical switch supplementary trigger command is issued; in the high-risk scenario, an electronic switch and mechanical switch synchronous trigger command is issued.

[0133] In one example, the comprehensive risk index The value range is 0 to 1. Based on the safety operation standards and graded disposal requirements of energy storage systems, The 0-1 range is divided into three risk ranges. For low-risk scenarios, For medium-risk scenarios, For high-risk scenarios, different levels generate different priority triggering commands for electronic switches and coordinated supplementary warning commands for mechanical switches. For example, in low-risk scenarios, only electronic switches are triggered to achieve millisecond-level electrical circuit disconnection, quickly curbing the risk of instantaneous short circuits. In high-risk scenarios, electronic switches and mechanical switches work together in layers, with mechanical switches simultaneously completing second-level physical isolation, forming a dual-redundancy barrier of rapid electrical disconnection and complete physical isolation.

[0134] Step 4: Based on the graded early warning instructions, execute the actions of electronic and mechanical switches according to the risk level; receive the physical status judgment results in real time, and if an emergency physical risk exceeding the preset threshold is detected, directly trigger the emergency backup action; at the same time, coordinate with the precision fire protection system to perform targeted disposal, and the status feedback unit will transmit the action data back to the digital twin in real time to support closed-loop optimization and disposal verification, so as to realize multi-level safety protection of pre-disaster intervention and emergency backup.

[0135] Furthermore, step 4 includes:

[0136] The urgency score is calculated based on the magnitude of parameter exceedance and the abnormal spread rate, using the following formula:

[0137] ,

[0138] in, Rate the urgency level. For real-time voltage, The safe threshold for voltage, This represents the voltage normalized reference value. For real-time voltage, This is the safe threshold for voltage. This represents the current number of abnormal individuals. This represents the number of abnormal individuals at the previous sampling time. The sampling time interval, This is a normalization coefficient used to map rate values ​​to a reasonable fractional range;

[0139] when When an emergency is identified, emergency response is immediately triggered, and the risk type, target handling area, action sequence parameters, and response logic are simultaneously defined.

[0140] During the switching action, the operating parameters of the two types of electronic switches and mechanical switches are collected in real time, and the battery temperature and voltage change data of the target area after the treatment are collected simultaneously and transmitted back to the digital twin in real time via industrial Ethernet.

[0141] Specifically, by receiving drive signals through dual channels simultaneously, the system firstly receives hierarchical instructions issued by the AI ​​simulation steps, and secondly receives real-time physical situation judgment results. The switch-coordinated control unit prioritizes the analysis of the physical urgency level to clarify the risk type, target disposal area, action sequence, and emergency backup logic.

[0142] For low-risk scenarios, only the electronic switch is triggered: IGBT high-frequency electronic switches are used to cut off the electrical circuit of the target area within 10-20 milliseconds, quickly suppressing abnormal current fluctuations and avoiding local energy accumulation; at this time, the mechanical switch remains on standby, reducing unnecessary mechanical movement losses and extending equipment life.

[0143] For medium-risk scenarios, an electronic-first, mechanical-complementary coordinated mode is activated: the electronic switch acts 50-100 milliseconds before the mechanical switch, quickly cutting off the electrical circuit and initially blocking the energy source of thermal runaway; then the mechanical switch performs second-level physical isolation, completely disconnecting the target area from the main circuit through mechanical contact disconnection, while triggering the contact sealing mechanism to prevent the electric arc from igniting the battery's volatile gases. This timing design of electrical disconnection first and physical isolation later avoids the problem of electronic switches being prone to failure due to long-term high voltage, and also solves the risk of risk spread caused by the slow response of mechanical switches.

[0144] For high-risk scenarios, a mechanism is triggered to activate the synchronous linkage of dual switches and fire protection coordination: the electronic switch and the mechanical switch achieve action synchronization in less than 50 milliseconds through a dedicated synchronous signal bus. While the electronic switch cuts off the circuit, the mechanical switch instantly completes physical isolation, forming an electrical-physical dual barrier; the synchronous linkage precision fire protection system injects inert gas into the target area to suppress the decomposition of electrolyte and further block the thermal runaway chain.

[0145] During the switching action, the operating parameters of both types of switches are collected in real time, including the on / off state and voltage drop of electronic switches, and the contact position and breaking time of mechanical switches. Simultaneously, the battery temperature and voltage change data of the target area after the action are collected and transmitted back to the digital twin in real time via industrial Ethernet. On the one hand, the transmitted data is used to verify the action effect. If the action is found to be substandard, a secondary collaborative action is triggered. On the other hand, the data serves as the basis for optimizing the digital twin and AI inference algorithms, continuously iterating parameters such as switch collaboration timing and action thresholds to ensure more accurate subsequent actions.

[0146] To further demonstrate the effectiveness and significance of the digital twin-based fire early warning and emergency response method for energy storage systems described in this invention, this invention is compared with traditional methods. Here, traditional methods refer to existing fire early warning and response methods for energy storage systems that rely on static threshold alarms, single switch / mechanical fuse isolation, and lack digital twin closed-loop simulation and collaborative control.

[0147] Figure 4 This invention demonstrates its advantages in early warning and response. Compared to traditional methods, the early warning and response times are shorter, indicating that the improved GNN-Transformer fusion algorithm combined with a high-fidelity digital twin sandbox can detect thermal runaway trends in advance, significantly reducing response time. Traditional methods suffer from delayed early warning and poor coordination in response, resulting in long response times and high risks of thermal runaway propagation. In contrast, this invention achieves high efficiency in emergency response through AI-driven advanced simulation and electro-mechanical dual-switch hierarchical coordination, directly demonstrating the technical effectiveness in improving the efficiency of thermal runaway response in energy storage systems.

[0148] Figure 5This invention demonstrates its reliability advantages. Compared to traditional methods, the reliability degradation of this invention is more gradual in energy storage systems of different scales. This indicates that the three-level, multi-scale, hierarchically coupled digital twin combined with the electro-mechanical dual-switch redundancy coordination mechanism can adapt to system expansion and reduce the cascading effects of single-point failures. Traditional methods, lacking hierarchical coupling and closed-loop optimization, experience a sharp drop in reliability as the system scales up. In contrast, this invention maintains high reliability through full-dimensional parameter sensing and dynamic precision calibration, directly confirming the technical effectiveness of the patent in ensuring the stable operation of large-scale energy storage power stations.

[0149] The improvement effect of the method of the present invention is further illustrated by several key evaluation indicators, the results of which are shown in Table 1. Specifically, the early warning time of thermal runaway is used to characterize the system's ability to perceive the risk of thermal runaway in advance. The larger this indicator is, the more timely the warning is, allowing more time for subsequent emergency response. The local hot spot missed detection rate is used to characterize the system's ability to identify local abnormal heat sources, and is used for monitoring coverage and anomaly detection capabilities. The false alarm rate of the warning is used to characterize the accuracy of the system's risk judgment. The lower this indicator is, the more invalid alarms the system can reduce, and the more reliable the warning is. The multi-parameter fusion data distortion rate is used to characterize the fidelity of multi-source sensor data after fusion processing, which is more conducive to subsequent digital twin modeling and AI inference. As shown in Table 1, the method of the present invention is superior to the traditional method in all the above indicators, further demonstrating its good technical effect in terms of early warning, monitoring accuracy, and data fusion reliability.

[0150] Table 1 Comparison results of multiple key evaluation indicators

[0151]

[0152] Example 2

[0153] Combination Figure 6 As shown in this embodiment, a fire early warning and emergency response system for energy storage systems based on digital twins includes:

[0154] The data acquisition module is used to collect real-time data on the operating conditions of the energy storage unit, electrical parameters of electronic switches, physical operation data of mechanical switches, and ambient temperature and humidity data. The acquired data is preprocessed and then transmitted in two paths: one path to the digital twin construction module and the other path directly to the physical execution module.

[0155] The digital twin construction module constructs a digital twin of the energy storage system based on preprocessed data. It maintains dynamic synchronization between the physical system and the digital twin through real-time data fusion and obtains simulation data through the digital twin simulation, providing a simulation verification environment for the AI ​​inference module.

[0156] The AI ​​simulation module, based on simulation data and real-time acquired data, uses the GNN-Transformer fusion algorithm to predict thermal runaway in advance; and according to the prediction results and real-time physical conditions, the risk is divided into low, medium and high levels, and corresponding graded instructions for coordinated action of electronic switches and mechanical switches are generated.

[0157] The physical execution module synchronously acquires hierarchical instructions and real-time physical condition judgment results, drives electronic switches to cut off electrical circuits, controls mechanical switches to perform physical isolation, and links with the precision fire protection system for targeted handling; at the same time, it transmits the effectiveness of the dual switch actions and the operating status of fire protection equipment back to the data acquisition module and the digital twin construction module in real time.

[0158] The above embodiments are only for illustrating the technical concept and features of the present invention, and are intended to enable those skilled in the art to understand the content of the present invention and implement it accordingly. They should not be construed as limiting the scope of protection of the present invention. All equivalent transformations or modifications made in accordance with the spirit and essence of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A digital-twin-based energy storage system fire warning and emergency disposal method, characterized in that, Includes the following steps: Step 1: Construct a hybrid sensing network of distributed fiber optic sensors and micro IoT nodes to synchronously collect multi-dimensional data from the energy storage system. The collected multi-dimensional data is fused using a dynamic confidence fusion mechanism to obtain fused data; Step 2: Based on the fused data, construct a three-level, multi-scale, hierarchically coupled digital twin. Use the digital twin to simulate the coordinated handling effect of electronic switches performing electrical disconnection and mechanical switches performing physical isolation under fault conditions. Dynamically correct the model parameters of each level in the digital twin according to the actual action data. Step 3: Using the GNN-Transformer fusion algorithm, a hierarchical heat conduction-electrical parameter-microscopic feature coupled network is constructed. With the support of the digital twin, the critical threshold and evolution path of thermal runaway are deduced, and hierarchical early warning instructions are generated. Step 4: According to the graded early warning instructions, execute the actions of electronic and mechanical switches according to the risk level; receive the physical status judgment results in real time, and if an emergency physical risk exceeding the preset threshold is detected, directly trigger the emergency backup action; at the same time, synchronize with the precision fire protection system to perform targeted disposal, and the status feedback unit will transmit the action data back to the digital twin in real time.

2. The digital-twin-based energy storage system fire warning and emergency disposal method according to claim 1, characterized in that, Step 2 includes: Step 21: Construct a digital twin with a three-level structure. This digital twin includes: using the battery cell as the smallest unit, constructing a battery cell-level model, and integrating micro-strain and local gas generation rate parameters to simulate the early internal reactions of thermal runaway; wherein, the mathematical formula for the battery cell-level model is: ， wherein, are the corrected model parameters, are the original parameters before correction, including the SEI film decomposition rate coefficient at the battery monomer level and the heat conduction coefficient at the module level, are the correction coefficients, is the relative deviation of the parameters; The module is used as a mid-level unit, composed of individual battery cells. A module-level model is constructed. By collecting the microscopic parameters of each battery cell within the module, and integrating macroscopic temperature field data and current distribution data at the module level, correlation analysis is performed on the temperature field and current distribution data to characterize the coupling and conduction laws of the module's thermal and electrical parameters. The mathematical formula for the module-level model is as follows: ， in, Indicates the density of the material. This represents the specific heat capacity at constant pressure. For the module temperature field, Represents the gradient or divergence operator. The thermal conductivity coefficient, For current density, For internal resistance, The convective heat transfer coefficient is... Indicates ambient temperature; The energy storage compartment is used as the top-level unit, assembled from modules to construct a hierarchical model of the energy storage compartment. This model integrates the coupling and conduction laws of the thermal and electrical parameters of each module, simulating the heat diffusion path and fire suppression effects of the entire system, forming a complete physical digital mirror from microscopic reactions to the macroscopic system. The mathematical formulas for the hierarchical model of the energy storage compartment are as follows: ， in, The total heat capacity of the energy storage warehouse. The average temperature inside the cabin. Indicates ambient temperature. It is the aggregation of the heat generation power of each module. Indicates the number of modules. The total thermal resistance from the cabin to the environment; The heat transfer power of the fire protection system is expressed as: ， in, For gas flow rate, For specific heat capacity, Temperature of the injected gas; Step 22: Receive real-time action data, including switch operation status and fire response effectiveness, and perform multi-dimensional comparison between the measured values ​​and the simulated values ​​of the digital twin. Quantify the differences using a relative deviation formula. ， in, The relative deviation of the core parameters These are measured values. The simulated value of the digital twin; Step 23: Make a judgment based on the magnitude of the deviation. If If the value is greater than 1%, a correction process will be triggered, indicating that the model parameters at each level in the digital twin or the sensors need to be adjusted or calibrated. In the correction process, the mathematical formulas of the battery cell hierarchical model are applied to dynamically update the model parameters of each level, and the model parameters are adjusted by scaling the deviation ratio and smoothing. Step 24: Based on the corrected digital twin, combined with the real-time acquired data, real-time twin data is obtained through iterative calculation using an electrochemical-thermal coupling model.

3. The method for fire early warning and emergency response of energy storage systems based on digital twins according to claim 1, characterized in that, Step 3 includes: Step 31: Construct a hierarchical heat conduction-electrical parameter-microscopic feature coupling network, including a feature extraction layer and a cross-feature temporal fusion layer. The feature extraction layer aggregates microscopic features, electrical parameter features and macroscopic features, and the cross-feature temporal fusion layer fuses all parameters to obtain a fused feature vector. Step 32: By learning from historical data and real-time twin data, the risk level is assessed using a risk quantification formula. The formula is as follows: ， in, As a comprehensive risk index, This is the normalized value of the local H2 / CO gas production rate. This is the normalized value of the dynamic internal resistance increase. This is the normalized value for module temperature. This is the normalized value for the number of abnormal individuals. This represents the normalized value of the micro-strain. Step 33, calculate the critical threshold for thermal runaway. The calculation formula is as follows: ， in, This is the critical threshold for thermal runaway. Weights are based on historical data. This represents the average thermal runaway threshold under similar operating conditions over a month. The current parameter trend value calculated for real-time twin data; Step 34: GNN is responsible for constructing the spatial topology of the energy storage system. In this spatial topology, the battery cells are used as nodes, and the thermal-electric connection relationship of the cells in the module is used as edges. By aggregating the features of adjacent nodes, the transmission of abnormal parameters in space is simulated to obtain the spatial diffusion chain. The historical sequence of each parameter is analyzed using a self-attention mechanism, and the time evolution curve is predicted by Transformer. Step 35: Couple the spatial diffusion chain and temporal evolution curve into the digital twin and perform high-fidelity simulation. Using the current fused feature vector as the initial state, the simulation is performed step by step in the twin sandbox. At each step, the parameters of each node are updated, and it is checked whether any issues are triggered. After the iteration is complete, a complete path with key spatiotemporal nodes marked is output. Step 36: Finally, based on the safety operation standards and graded disposal requirements of the energy storage system, the comprehensive risk index is divided into three risk ranges: low-risk scenario, medium-risk scenario, and high-risk scenario. In the low-risk scenario, an electronic switch trigger command is issued; in the medium-risk scenario, an electronic switch priority and mechanical switch supplementary trigger command is issued; in the high-risk scenario, an electronic switch and mechanical switch synchronous trigger command is issued.

4. The fire early warning and emergency response method for energy storage systems based on digital twins according to claim 3, characterized in that, Step 4 includes: The urgency score is calculated based on the magnitude of parameter exceedance and the abnormal spread rate, using the following formula: ， in, Rate the urgency level. For real-time voltage, The safe threshold for voltage, This represents the voltage normalized reference value. For real-time voltage, This is the safe threshold for voltage. This represents the current number of abnormal individuals. This represents the number of abnormal individuals at the previous sampling time. The sampling time interval, This is a normalization coefficient used to map rate values ​​to a reasonable fractional range; when When an emergency is identified, emergency response is immediately triggered, and the risk type, target handling area, action sequence parameters, and response logic are simultaneously defined. During the switching action, the operating parameters of the two types of electronic switches and mechanical switches are collected in real time, and the battery temperature and voltage change data of the target area after the treatment are collected simultaneously and transmitted back to the digital twin in real time via industrial Ethernet.

5. The method for fire early warning and emergency response of energy storage systems based on digital twins according to claim 1, characterized in that, Step 1 includes: Distributed fiber optic sensors are embedded along the length of the battery module to collect the continuous temperature field and micro-strain distribution on the battery surface in real time. The micro IoT node integrates a voltage sensor, a dynamic internal resistance monitoring module, and a gas sensor to collect data on battery cell voltage, dynamic internal resistance, and local gas generation rate at a synchronous frequency. The two types of sensors align their data using timestamp synchronization technology to generate multi-dimensional data. Preprocess the collected multi-dimensional data; Based on the sensor health index and environmental interference coefficient, the fusion weights of multi-source data are dynamically adjusted, using the following formula: ， In the formula, For the first Sensor-like fusion weights, For sensor health index, Environmental interference coefficient; Multi-dimensional data is merged using fusion weights to obtain fused data.

6. A fire early warning and emergency response system for energy storage systems based on digital twins, characterized in that, include: The data acquisition module is used to collect real-time data on the operating conditions of the energy storage unit, electrical parameters of electronic switches, physical operation data of mechanical switches, and ambient temperature and humidity data. The acquired data is preprocessed and then transmitted in two paths: one path to the digital twin construction module and the other path directly to the physical execution module. The digital twin construction module constructs a digital twin of the energy storage system based on preprocessed data. It maintains dynamic synchronization between the physical system and the digital twin through real-time data fusion and obtains simulation data through the digital twin simulation, providing a simulation verification environment for the AI ​​inference module. The AI ​​simulation module, based on simulation data and real-time acquired data, uses the GNN-Transformer fusion algorithm to predict thermal runaway in advance; and according to the prediction results and real-time physical conditions, the risk is divided into low, medium and high levels, and corresponding graded instructions for coordinated action of electronic switches and mechanical switches are generated. The physical execution module synchronously acquires hierarchical instructions and real-time physical condition judgment results, drives electronic switches to cut off electrical circuits, controls mechanical switches to perform physical isolation, and links with the precision fire protection system for targeted handling; at the same time, it transmits the effectiveness of the dual switch actions and the operating status of fire protection equipment back to the data acquisition module and the digital twin construction module in real time.

Images