Electrochemical energy storage facility fire monitoring system and method based on large model

Abstract

The application relates to the technical field of electrochemical energy storage safety management, and discloses an electrochemical energy storage facility fire supervision system and method based on a large model, which realizes high-precision reconstruction of a global three-dimensional temperature field and accurate identification of blind area hot spots under sparse measuring points, solves the problem of missed detection in the monitoring blind area; a lightweight large model embedded with an electrochemical mechanism of energy storage thermal runaway is taken as a core, sparse measuring point reconstruction loss constraint optimization training is matched, and the global temperature distribution of the energy storage cabin is restored under the condition of a small amount of temperature measuring hardware; in combination with a blind area hot spot feature library constructed by DBSCAN density clustering, hidden hot spots in the monitoring blind area such as the cell interlayer and the back of the air duct are accurately identified through cosine similarity matching, the missed detection rate of the electrochemical energy storage facility fire supervision blind area is reduced, the four-level risk is accurately divided through a fire risk grading large model, and the thermal spread sequence of the high-risk scene is deduced and accurate positioning early warning is output.

Description

Technical Field

[0001] This invention relates to the technical field of electrochemical energy storage safety management, specifically to a fire monitoring system and method for electrochemical energy storage facilities based on a large model. Background Technology

[0002] With the rapid construction of new power systems and the large-scale deployment of electrochemical energy storage power stations and containerized energy storage modules, fires caused by lithium battery thermal runaway are frequent. The enclosed structure of energy storage modules and the dense stacking of batteries significantly increases the speed of fire spread and the difficulty of handling. Large-scale models, with their nonlinear fitting and deep feature mining capabilities, have become the core technological support for intelligent monitoring of energy storage fires and a key foundation for the efficient operation of current energy storage safety supervision systems. The core of large-scale model-driven energy storage fire supervision relies on a closed-loop process of data acquisition, temperature field reconstruction, blind spot identification, and risk warning: the supervision system collects sparse temperature measurement data from multi-source monitoring terminals in the energy storage module to build a data foundation, trains a reconstruction model using historical thermal runaway data, reconstructs the temperature distribution across the entire energy storage module, and then determines the fire risk based on temperature anomaly characteristics. The reliable operation of this process depends on the core premise of high-precision reconstruction of the three-dimensional temperature field under sparse measurement points and complete identification of blind spots and hot spots. Existing technologies face significant technical barriers when addressing real-time fire monitoring scenarios for electrochemical energy storage facilities. First, due to the dense battery arrangement and hardware deployment costs, energy storage compartments can only accommodate a limited number of temperature monitoring points. This results in numerous temperature monitoring blind spots in areas such as the cell interlayer, the back of the air duct, and the bottom of the cables. Traditional interpolation algorithms cannot accurately reconstruct the temperature distribution in these blind spots. Second, existing monitoring systems suffer from inconsistent protocols among multi-source devices and misaligned spatiotemporal references. Temperature data, electrical data, and thermal imaging data cannot be aligned in both time and space, leading to coordinate deviations in subsequent temperature field reconstruction. Finally, traditional temperature reconstruction models do not incorporate the electrochemical mechanisms of thermal runaway in energy storage. With sparse monitoring points, the temperature reconstruction error in blind spots is large, hidden hotspots cannot be identified in advance, and risk classification relies solely on single-point temperature data, lacking support from the evolution characteristics of the entire temperature field. Ultimately, the delayed detection of thermal runaway in blind spots leads to the spread of fires, resulting in regulatory failure. This shows that the existing technology has the following main problems: 1. It lacks a unified spatiotemporal registration mechanism for multi-source heterogeneous data, and the misalignment of data benchmarks reduces the accuracy of supervision; 2. The accuracy of three-dimensional temperature field reconstruction under sparse measurement points is insufficient, and the identification of blind spot hotspots is lagging and the rate of missed detection is high; 3. The risk assessment dimension is single and does not combine the temperature evolution of the whole domain with the characteristics of blind spots, so it is impossible to achieve early warning and closed-loop supervision.

[0003] Chinese invention patent application CN119338200A, published on January 21, 2025, discloses a thermal runaway safety monitoring system for an electrochemical energy storage system in a power distribution network. The system includes a task classification module, a priority scoring module, a task allocation module, a resource locking module, a thermal runaway monitoring module, and an anomaly control module. By meticulously managing the urgency of tasks and resource requirements, and adjusting resource allocation and task priorities in a timely manner, the system significantly improves the efficiency and response speed of the energy storage system. It continuously monitors the dynamic matching of resource status and task requirements, effectively avoiding operational delays caused by resource shortages. It tracks node temperature and current parameters in real time, quickly identifying and intervening in potential thermal runaway risk nodes to avoid serious accidents. This optimization of resource and risk management not only enhances system safety but also ensures efficient energy utilization and system resilience to faults, significantly improving the overall stability and reliability of the system. However, the above technical solutions cannot achieve high-precision reconstruction of the three-dimensional temperature field of the energy storage system or full-area identification of blind zone fire hotspots under limited monitoring conditions. Summary of the Invention

[0004] (a) Technical problems to be solved

[0005] To address the shortcomings of existing fire monitoring systems for electrochemical energy storage facilities, such as the lack of a unified spatiotemporal registration mechanism for multi-source heterogeneous data, leading to misaligned data benchmarks and reduced monitoring accuracy; insufficient accuracy in 3D temperature field reconstruction under sparse monitoring points, resulting in delayed blind spot and hotspot identification and high false negative rates; and a single-dimensional risk assessment that fails to incorporate global temperature evolution and blind spot characteristics, hindering early warning and closed-loop monitoring, this system aims to achieve a unified data benchmark based on spatiotemporal registration of multi-source monitoring data. It also incorporates a fine-tuned large-scale model of energy storage thermal runaway mechanisms to achieve high-precision reconstruction of the global 3D temperature field under sparse monitoring points, coupled with a blind spot and hotspot feature database for accurate identification of hidden thermal anomalies. Furthermore, through hierarchical risk assessment and thermal propagation prediction based on global feature fusion, a closed-loop fire monitoring system is constructed to achieve early warning, adaptability to edge deployment, and effective protection of the operational safety of electrochemical energy storage facilities.

[0006] (II) Technical Solution

[0007] This invention is achieved through the following technical solution: a fire monitoring method for electrochemical energy storage facilities based on a large model, comprising the following steps:

[0008] S10: Collect multi-source monitoring data of the electrochemical energy storage facility. The multi-source monitoring data includes temperature of sparsely arranged fixed-point thermocouple measuring points, battery voltage, battery current, battery state of charge (SOC), local infrared thermal imaging images inside the cabin, ventilation and heat dissipation parameters, and ambient temperature and humidity. Perform spatiotemporal registration on the multi-source monitoring data and construct a three-dimensional spatial coordinate mapping relationship for energy storage. Generate a sparse measuring point temperature dataset and a three-dimensional geometric spatial correlation sample of the energy storage cabin suitable for fire monitoring of electrochemical energy storage facilities.

[0009] S20: Based on the sparse temperature measurement point dataset and the three-dimensional geometric space correlation samples of the energy storage chamber, and combined with the energy storage thermal runaway mechanism, the lightweight industry large model is fine-tuned in the domain to construct a three-dimensional temperature field global reconstruction large model suitable for fire monitoring of electrochemical energy storage facilities; at the same time, the DBSCAN density clustering algorithm is used to perform feature clustering analysis on the historical temperature blind zone missed samples, and a blind zone hotspot feature library suitable for fire monitoring of electrochemical energy storage facilities is established based on the clustering analysis results;

[0010] S30: Based on the three-dimensional temperature field global reconstruction model and the blind spot hotspot feature library, real-time multi-source monitoring data is input into the three-dimensional temperature field global reconstruction model, and the complete three-dimensional temperature field grid data of the energy storage compartment suitable for fire monitoring of electrochemical energy storage facilities is output; and the three-dimensional temperature field grid data is matched with the blind spot hotspot feature library to obtain the blind spot hotspot identification result; when the identification result is that there is a hidden hotspot in the temperature monitoring blind spot, the three-dimensional coordinates, temperature rise rate and temperature gradient of the blind spot hotspot used for fire hazard assessment are marked; when the identification result is that there is no hidden hotspot in the temperature monitoring blind spot, the monitoring status of no blind spot thermal anomaly is output;

[0011] S40: Based on the three-dimensional coordinates of the blind spot hotspot, the temperature rise rate, the temperature gradient, and the complete three-dimensional temperature field grid data of the energy storage chamber, extract the blind spot hotspot features and the spatiotemporal evolution features of the global temperature field and input them into the fire risk classification model to obtain the fire risk level determination result; when the determination result indicates that there is a high fire risk in the blind spot hotspot, calculate the heat spread time series prediction curve and output accurate spatial positioning and fire early warning signals for fire supervision of electrochemical energy storage facilities; when the determination result indicates that there is no high fire risk in the blind spot hotspot, output the conventional fire supervision status information of the electrochemical energy storage facility.

[0012] Preferably, in S10, the multi-source monitoring data is acquired through corresponding acquisition devices and uniformly accessed to the edge data acquisition platform; wherein the temperature of the fixed thermocouple measuring point is acquired by the distributed multi-channel thermocouple acquisition terminal, the battery voltage, battery current, and battery state of charge (SOC) are acquired by the energy storage battery management system (BMS), the local infrared thermal imaging image inside the cabin is acquired by the fixed infrared thermal imager deployed on the top of the cabin, the ventilation and heat dissipation parameters are acquired by the fan speed sensor and wind pressure transmitter of the energy storage cabin ventilation control system, and the ambient temperature and humidity are acquired by the digital temperature and humidity sensors deployed inside and outside the cabin; the edge data acquisition platform performs protocol parsing, outlier removal, and data temporary storage processing on the acquired data from each source, and outputs standardized pre-processed data to provide basic data support for fire monitoring of electrochemical energy storage facilities.

[0013] Preferably, the step S10, which involves uniformly performing spatiotemporal registration and constructing a three-dimensional spatial coordinate mapping relationship for energy storage, includes the following steps:

[0014] S101: Using the unified system clock of the edge data acquisition platform as the time reference, time registration is performed on multi-source monitoring data with different sampling frequencies using linear interpolation to generate synchronous multi-source data groups at the same time.

[0015] S102: Pre-store the three-dimensional point cloud model of the energy storage cabin obtained by three-dimensional laser scanning, assign a unique spatial identifier to each acquisition device, and bind the actual installation coordinates of each fixed-point thermocouple measuring point to the corresponding spatial position of the three-dimensional point cloud model.

[0016] S103: Camera calibration and homography matrix calculation are performed on the local infrared thermal images acquired by the fixed infrared thermal imager to establish the mapping relationship between the pixel coordinates of the infrared image and the coordinates of the three-dimensional point cloud, and to complete the spatial dimension registration.

[0017] S104: Outputs a sparse temperature dataset with spatiotemporal labels and a three-dimensional geometric spatial association sample of the energy storage compartment.

[0018] Preferably, the establishment of the blind spot hotspot feature database in S20 includes the following steps:

[0019] S201: Extract historical data of missed detection blind zone accidents, using three-dimensional coordinate position, temperature difference at measuring point, temperature rise gradient, SOC range and heat dissipation wind speed as multi-dimensional features;

[0020] S202: The DBSCAN density clustering algorithm is used, and the neighborhood radius and minimum number of points are set as clustering parameters to obtain multiple blind spot hotspot distribution clusters corresponding to different types of fire hazards;

[0021] S203: Integrate all blind spot hotspot distribution clusters to generate a standardized blind spot hotspot feature library suitable for fire monitoring of electrochemical energy storage facilities.

[0022] Preferably, the construction of the three-dimensional temperature field global reconstruction model in S20 includes the following steps:

[0023] S204: Collect finite element simulation data of full-domain temperature field of multi-capacity, multi-arrange energy storage compartments and sparse measurement point data from the field to construct a dual-source training set;

[0024] S205: Based on a general multimodal large model, it incorporates prior knowledge of battery heat transfer and cell thermal runaway electrochemistry for domain fine-tuning;

[0025] S206: Introducing sparse measurement point reconstruction loss constraints to complete model training, outputting a large-scale three-dimensional temperature field global reconstruction model suitable for fire monitoring of electrochemical energy storage facilities.

[0026] Preferably, obtaining the blind spot hotspot identification result in S30 includes the following steps:

[0027] S301: The reconstructed continuous three-dimensional temperature field grid data is divided into the visible area of ​​the battery body and the blind areas of the cell interlayer, back of the air duct and bottom of the cable that cannot be covered by the sensor. The temperature change rate and the second derivative of the spatial temperature gradient are extracted point by point from the grid of the blind area.

[0028] S302: Calculate the cosine similarity between the blind zone grid features and the blind zone hotspot feature library to obtain the matching degree;

[0029] S303: Preset matching degree threshold; if the matching degree is greater than or equal to the matching degree threshold, it is determined that there is a hidden hot spot with fire hazard in the corresponding blind zone grid, and an identification result of hidden hot spot with temperature monitoring blind zone is generated; if the matching degree is less than the matching degree threshold, it is determined that there is no fire thermal anomaly in the corresponding blind zone grid, and an identification result of no hidden hot spot with temperature monitoring blind zone is generated.

[0030] Preferably, obtaining the fire risk level determination result in S40 includes the following steps:

[0031] S401: Obtain the highest temperature of the blind zone hotspot, the temperature difference between the blind zone hotspot and the surrounding batteries, the temperature rise rate of the blind zone, the thermal diffusion trend of the global temperature field, the battery SOC, the battery health status SOH, and the concentration of combustible gases in the cabin, as input features for the fire risk classification model.

[0032] S402: The fire risk classification model is used to perform classification calculations and output four fire risk levels: no abnormality, slight thermal distortion, thermal runaway precursor, and initial fire.

[0033] S403: If the output fire risk level is a precursor to thermal runaway or an initial fire, then a high fire risk is determined to exist in a blind spot hotspot, and a judgment result of a high fire risk in a blind spot hotspot is generated; if the output fire risk level is no abnormality or slight thermal distortion, then a high fire risk in a blind spot hotspot is determined to exist, and a judgment result of a high fire risk in a blind spot hotspot is generated.

[0034] Preferably, the execution process corresponding to different determination results in S40 is as follows:

[0035] When the judgment result indicates that there is a high fire risk in the blind spot hotspot, the 0-30min thermal spread time sequence prediction curve is derived based on the complete three-dimensional temperature field grid data of the energy storage compartment. Combined with the three-dimensional coordinates of the blind spot hotspot, the fire risk level and the predicted duration of thermal runaway, a standardized fire early warning signal is generated and output to support the fire emergency supervision and disposal of electrochemical energy storage facilities.

[0036] When the determination result is that there is no blind spot hotspot high fire risk, the current temperature field status of the whole area and the blind spot monitoring results are recorded, the routine fire supervision status information of electrochemical energy storage facilities is output, and real-time monitoring is continuously carried out.

[0037] A fire monitoring system for electrochemical energy storage facilities based on a large model, used to implement the aforementioned fire monitoring method for electrochemical energy storage facilities based on a large model, the system comprising:

[0038] Multi-source spatiotemporal registration and spatial mapping module: Collects multi-source monitoring data of electrochemical energy storage facilities. The multi-source monitoring data includes temperature of sparsely arranged fixed-point thermocouple measuring points, battery voltage, battery current, battery state of charge (SOC), local infrared thermal imaging images inside the cabin, ventilation and heat dissipation parameters, and ambient temperature and humidity. The module performs spatiotemporal registration on the multi-source monitoring data and constructs a three-dimensional spatial coordinate mapping relationship for energy storage. It generates a sparse measuring point temperature dataset and a three-dimensional geometric spatial association sample of the energy storage cabin, which are suitable for fire monitoring of electrochemical energy storage facilities.

[0039] The 3D temperature field reconstruction and blind spot hotspot mining module: Based on the sparse measurement point temperature dataset and the 3D geometric spatial correlation samples of the energy storage chamber, and combined with the energy storage thermal runaway mechanism, the module performs domain fine-tuning on the lightweight industry large model to construct a 3D temperature field global reconstruction large model suitable for fire monitoring of electrochemical energy storage facilities; simultaneously, it uses the DBSCAN density clustering algorithm to perform feature clustering analysis on historical temperature blind spot missed samples, and establishes a blind spot hotspot feature library suitable for fire monitoring of electrochemical energy storage facilities based on the clustering analysis results; outputs the blind spot hotspot identification results and executes corresponding operations according to the path;

[0040] Fire risk classification and early warning output module: Based on the three-dimensional temperature field global reconstruction model and the blind spot hotspot feature library, it outputs complete three-dimensional temperature field grid data of the energy storage compartment suitable for fire monitoring of electrochemical energy storage facilities, identifies hidden hotspots in temperature monitoring blind spots, classifies four levels of fire risk based on the spatiotemporal evolution characteristics of the global temperature field, calculates the heat spread time series prediction curve when high risk is determined, and outputs accurate spatial positioning and fire early warning signals for fire monitoring of electrochemical energy storage facilities.

[0041] (III) Beneficial Effects

[0042] This invention provides a fire monitoring system and method for electrochemical energy storage facilities based on a large-scale model. It has the following beneficial effects:

[0043] I. By constructing a unified spatiotemporal registration system for multi-source heterogeneous data, we can solidify the data accuracy foundation for fire monitoring of electrochemical energy storage facilities. Through the edge data acquisition platform, we can achieve standardized access to multi-protocol monitoring data. By combining a dual registration mechanism of linear interpolation time registration with a unified clock reference and three-dimensional point cloud spatial mapping, we can completely eliminate the temporal misalignment and spatial deviation of multi-source data, ensure the spatiotemporal reference of temperature measurement, electrical, and thermal imaging data, and provide reliable data support for subsequent three-dimensional temperature field reconstruction and blind spot hotspot location.

[0044] II. Achieving high-precision reconstruction of the full-domain three-dimensional temperature field and accurate identification of blind spot hotspots under sparse measurement points, solving the problem of missed detection in monitoring blind spots; taking a lightweight large model embedded with the electrochemical mechanism of thermal runaway in energy storage as the core, and combining it with sparse measurement point reconstruction loss constraint optimization training, the temperature distribution of the entire energy storage compartment can be restored under the condition of a small amount of temperature measurement hardware; combined with the blind spot hotspot feature library constructed by DBSCAN density clustering, hidden hotspots in monitoring blind spots such as cell sandwich and back of air duct are accurately identified by cosine similarity matching, reducing the blind spot missed detection rate of fire supervision of electrochemical energy storage facilities, capturing early signs of thermal runaway in advance, and at the same time, eliminating the need for dense sensor deployment, significantly reducing the system hardware deployment cost.

[0045] Third, by establishing a hierarchical risk assessment and closed-loop early warning mechanism that integrates full-domain features, the system balances fire supervision efficiency with scenario adaptability; it integrates blind spot hotspot characteristics, spatiotemporal evolution characteristics of the full-domain temperature field, and multi-dimensional operating parameters to achieve accurate classification of four levels of risk through a large-scale fire risk classification model; it extrapolates the heat spread sequence for high-risk scenarios and outputs precise location early warnings, while maintaining routine monitoring for low-risk scenarios, thus constructing a complete closed-loop supervision system; relying on edge-side lightweight model inference, it can be implemented and operated without additional computing hardware, adapting to various application scenarios such as containerized energy storage compartments and fixed energy storage power stations, thereby improving the scientific nature of fire supervision of electrochemical energy storage facilities. Attached Figure Description

[0046] Figure 1 A flowchart of the fire monitoring method for electrochemical energy storage facilities based on a large model provided by the present invention;

[0047] Figure 2 A schematic diagram of the modules of the fire monitoring system for electrochemical energy storage facilities based on a large model provided by the present invention. Detailed Implementation

[0048] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0049] An example of the fire monitoring system and method for electrochemical energy storage facilities based on a large model is as follows:

[0050] Example 1:

[0051] Please see Figures 1-2 A fire monitoring method for electrochemical energy storage facilities based on a large model includes the following steps:

[0052] S10: Collect multi-source monitoring data of the electrochemical energy storage facility. The multi-source monitoring data includes temperature of sparsely arranged fixed-point thermocouple measuring points, battery voltage, battery current, battery state of charge (SOC), local infrared thermal imaging images inside the cabin, ventilation and heat dissipation parameters, and ambient temperature and humidity. Perform spatiotemporal registration on the multi-source monitoring data and construct a three-dimensional spatial coordinate mapping relationship for energy storage. Generate a sparse measuring point temperature dataset and a three-dimensional geometric spatial correlation sample of the energy storage cabin suitable for fire monitoring of electrochemical energy storage facilities.

[0053] Specifically, the application scenario of this embodiment is a 1MW standard containerized electrochemical energy storage compartment, which has 8 battery clusters arranged inside, and is equipped with a battery management system (BMS), a ventilation and heat dissipation system and a fire alarm system; an edge computing gateway is deployed locally in the energy storage compartment, which undertakes the core functions of data acquisition, model inference and early warning output.

[0054] Specifically, in embodiment S10, the acquisition and preprocessing of multi-source monitoring data are as follows:

[0055] Twelve K-type armored thermocouples are sparsely arranged along the main channel of the battery clusters inside the compartment, connected to an 8-channel distributed thermocouple acquisition terminal with a sampling frequency of 1Hz, used to collect temperature data at fixed locations. Battery voltage, battery current, and battery state of charge (SOC) data are output by the battery management system (BMS) of the energy storage compartment via the Modbus protocol, with a sampling frequency of 0.5Hz. Two fixed infrared thermal imagers are deployed at diagonal positions on the top of the compartment, with a resolution of 384×288 and a frame rate of 1Hz, to acquire local thermal images of the front of the battery clusters. Ventilation and heat dissipation parameters are controlled by the ventilation system. The system's built-in fan speed sensor and air pressure transmitter collect data including fan operating frequency and duct inlet and outlet air pressure. Two digital temperature and humidity sensors are deployed both inside and outside the storage compartment to collect ambient temperature and humidity data. All data from these devices are uniformly connected to an edge data acquisition platform deployed locally in the energy storage compartment. The platform supports parsing multiple protocols including Modbus, RTSP, and MQTT, and simultaneously performs outlier removal and data standardization and temporary storage based on the Raida criterion, filters out invalid data with sudden changes, and outputs standardized preprocessed data to provide high-quality basic data for subsequent fire monitoring and analysis.

[0056] The steps involved in S10 to uniformly perform spatiotemporal registration and construct the three-dimensional spatial coordinate mapping relationship for energy storage include:

[0057] S101: Using the unified system clock of the edge data acquisition platform as the time reference, time registration is performed on multi-source monitoring data with different sampling frequencies using linear interpolation to generate synchronous multi-source data groups at the same time.

[0058] S102: Pre-store the three-dimensional point cloud model of the energy storage cabin obtained by three-dimensional laser scanning, assign a unique spatial identifier to each acquisition device, and bind the actual installation coordinates of each fixed-point thermocouple measuring point to the corresponding spatial position of the three-dimensional point cloud model.

[0059] S103: Camera calibration and homography matrix calculation are performed on the local infrared thermal images acquired by the fixed infrared thermal imager to establish the mapping relationship between the pixel coordinates of the infrared image and the coordinates of the three-dimensional point cloud, and to complete the spatial dimension registration.

[0060] S104: Outputs a sparse temperature dataset with spatiotemporal labels and a three-dimensional geometric spatial association sample of the energy storage compartment.

[0061] Specifically, in embodiments S101-S104, the time registration process uses the GPS synchronization system clock of the edge data acquisition platform as a unified time reference. For different sampling frequencies of 1Hz for thermocouples, 0.5Hz for BMS, and 1Hz for thermal imaging, linear interpolation is used to interpolate and complete the data at asynchronous times, generating a synchronous multi-source data group with one timestamp per second, ensuring that all data are fully aligned in the time dimension.

[0062] The spatial registration process involves pre-storing a high-precision 3D point cloud model of the energy storage cabin acquired using a handheld 3D laser scanner. The point cloud accuracy is 5mm, covering all cabin equipment such as battery clusters, air ducts, cables, and sensors. Each acquisition device is assigned a unique spatial ID, and the actual installation coordinates of the 12 thermocouples are bound to the corresponding positions in the 3D point cloud model. The fixed infrared thermal imager is calibrated using a Zhang camera, and the homography matrix is ​​calculated to establish a one-to-one mapping relationship between each pixel in the infrared image and the coordinates of the 3D point cloud, thus completing the spatial dimension registration.

[0063] The final output is a sparse temperature dataset with spatiotemporal labels and a three-dimensional geometric spatial correlation sample of the energy storage compartment, which serves as input data for subsequent temperature field reconstruction and blind zone identification.

[0064] By constructing a unified spatiotemporal registration system for multi-source heterogeneous data, a solid foundation for data accuracy in monitoring fires in electrochemical energy storage facilities is established. Standardized access to multi-protocol monitoring data is achieved through an edge data acquisition platform. Combined with a dual registration mechanism of linear interpolation time registration with a unified clock reference and spatial mapping of three-dimensional point clouds, the temporal misalignment and spatial deviation of multi-source data are completely eliminated, ensuring the spatiotemporal reference of temperature measurement, electrical, and thermal imaging data is unified, providing reliable data support for subsequent three-dimensional temperature field reconstruction and blind spot hotspot location.

[0065] S20: Based on the sparse temperature measurement point dataset and the three-dimensional geometric space correlation samples of the energy storage chamber, and combined with the energy storage thermal runaway mechanism, the lightweight industry large model is fine-tuned in the domain to construct a three-dimensional temperature field global reconstruction large model suitable for fire monitoring of electrochemical energy storage facilities; at the same time, the DBSCAN density clustering algorithm is used to perform feature clustering analysis on the historical temperature blind zone missed samples, and a blind zone hotspot feature library suitable for fire monitoring of electrochemical energy storage facilities is established based on the clustering analysis results.

[0066] The establishment of the blind spot hotspot feature database in S20 includes the following steps:

[0067] S201: Extract historical data of missed detection blind zone accidents, using three-dimensional coordinate position, temperature difference at measuring point, temperature rise gradient, SOC range and heat dissipation wind speed as multi-dimensional features;

[0068] S202: The DBSCAN density clustering algorithm is used, and the neighborhood radius and minimum number of points are set as clustering parameters to obtain multiple blind spot hotspot distribution clusters corresponding to different types of fire hazards;

[0069] S203: Integrate all blind spot hotspot distribution clusters to generate a standardized blind spot hotspot feature library suitable for fire monitoring of electrochemical energy storage facilities.

[0070] Specifically, in embodiments S201-S203, historical data of 120 energy storage blind zone thermal runaway failures were extracted. Using three-dimensional coordinate position, temperature difference at measurement point, temperature rise gradient, SOC range, and heat dissipation wind speed as five-dimensional feature vectors, the DBSCAN density clustering algorithm was selected. The neighborhood radius was set to 0.8 and the minimum number of samples was 6. The blind zone hotspot distribution clusters were automatically divided into three categories: cell interlayer blind zone, air duct back blind zone, and cable bottom layer blind zone, totaling eight clusters corresponding to different fire hazard types. The critical temperature rise threshold, temperature gradient evolution curve, and typical operating condition range of each distribution cluster were statistically analyzed. All blind zone hotspot distribution clusters were integrated to generate a standardized blind zone hotspot feature library, which was stored in the edge local database.

[0071] The construction of the three-dimensional temperature field global reconstruction model in S20 includes the following steps:

[0072] S204: Collect finite element simulation data of full-domain temperature field of multi-capacity, multi-arrange energy storage compartments and sparse measurement point data from the field to construct a dual-source training set;

[0073] S205: Based on a general multimodal large model, it incorporates prior knowledge of battery heat transfer and cell thermal runaway electrochemistry for domain fine-tuning;

[0074] S206: Introducing sparse measurement point reconstruction loss constraints to complete model training, outputting a large-scale three-dimensional temperature field global reconstruction model suitable for fire monitoring of electrochemical energy storage facilities.

[0075] Specifically, in embodiments S204-S206, 500 sets of finite element simulation global temperature field data of energy storage cabins with 8 different capacities and battery arrangements were collected, and 3000 sets of on-site sparse measurement point measured data were combined to construct a dual-source training set. Based on a lightweight Transformer multimodal large model, electrochemical prior knowledge such as lithium battery heat generation formula, cabin convection heat transfer mechanism and thermal runaway chain reaction was embedded. A sparse measurement point reconstruction loss constraint term was added to the loss function to complete the domain fine-tuning training. The fine-tuned model was pruned using INT8 quantization, and a three-dimensional temperature field global reconstruction large model that can run on the edge gateway was output.

[0076] Achieving high-precision reconstruction of the full-domain three-dimensional temperature field and accurate identification of blind zone hotspots under sparse measurement points solves the problem of missed detection in monitoring blind zones. With a lightweight large model embedding the electrochemical mechanism of thermal runaway in energy storage as the core, and combined with sparse measurement point reconstruction loss constraint optimization training, the temperature distribution of the entire energy storage compartment can be restored under the condition of a small amount of temperature measurement hardware. Combined with the blind zone hotspot feature library constructed by DBSCAN density clustering, hidden hotspots in monitoring blind zones such as cell sandwich and back of air duct are accurately identified through cosine similarity matching, reducing the blind zone missed detection rate of fire monitoring of electrochemical energy storage facilities, capturing early signs of thermal runaway in advance, and significantly reducing the system hardware deployment cost without the need for dense sensor deployment.

[0077] S30: Based on the three-dimensional temperature field global reconstruction model and the blind spot hotspot feature library, real-time multi-source monitoring data is input into the three-dimensional temperature field global reconstruction model, and the complete three-dimensional temperature field grid data of the energy storage compartment suitable for fire monitoring of electrochemical energy storage facilities is output; and feature matching is performed between the three-dimensional temperature field grid data and the blind spot hotspot feature library to obtain the blind spot hotspot identification result; when the identification result is that there is a hidden hotspot in the temperature monitoring blind zone, the three-dimensional coordinates, temperature rise rate and temperature gradient of the blind spot hotspot used for fire hazard assessment are marked; when the identification result is that there is no hidden hotspot in the temperature monitoring blind zone, the monitoring status of no blind zone thermal anomaly is output.

[0078] The process of obtaining the blind spot hotspot identification result in S30 includes the following steps:

[0079] S301: The reconstructed continuous three-dimensional temperature field grid data is divided into the visible area of ​​the battery body and the blind areas of the cell interlayer, back of the air duct and bottom of the cable that cannot be covered by the sensor. The temperature change rate and the second derivative of the spatial temperature gradient are extracted point by point from the grid of the blind area.

[0080] S302: Calculate the cosine similarity between the blind zone grid features and the blind zone hotspot feature library to obtain the matching degree;

[0081] S303: Preset matching degree threshold; if the matching degree is greater than or equal to the matching degree threshold, it is determined that there is a hidden hot spot with fire hazard in the corresponding blind zone grid, and an identification result of hidden hot spot with temperature monitoring blind zone is generated; if the matching degree is less than the matching degree threshold, it is determined that there is no fire thermal anomaly in the corresponding blind zone grid, and an identification result of no hidden hot spot with temperature monitoring blind zone is generated.

[0082] Specifically, in embodiments S301-S303, the reconstructed continuous three-dimensional temperature field grid data is divided with a precision of 10cm into the visible area of ​​the battery body covered by sensor measurement points, and the blind areas of the cell interlayer, the back of the air duct, and the bottom layer of the cable without sensor coverage; the unit time temperature change rate and the second derivative of the spatial temperature gradient are extracted point by point for all blind area grids, and the cosine similarity is calculated one by one with the 8 feature clusters in the blind area hotspot feature library to obtain the matching degree;

[0083] The preset matching threshold is 0.85. If the matching degree is greater than or equal to 0.85, it is determined that there is a hidden hot spot of fire hazard in the corresponding blind zone grid, and an identification result of hidden hot spot of temperature monitoring blind zone is generated. If the matching degree is less than 0.85, it is determined that there is no fire thermal anomaly in the corresponding blind zone grid, and an identification result of no hidden hot spot of temperature monitoring blind zone is generated.

[0084] When the identification result indicates the existence of hidden hotspots, mark the three-dimensional spatial coordinates, 1-minute temperature rise rate, and spatial temperature gradient values ​​of all blind spot hotspots used for fire hazard assessment; when the identification result indicates no hidden hotspots, output the monitoring status of no blind spot thermal anomalies and continue real-time monitoring for the next cycle.

[0085] S40: Based on the three-dimensional coordinates of the blind spot hotspot, the temperature rise rate, the temperature gradient, and the complete three-dimensional temperature field grid data of the energy storage chamber, extract the blind spot hotspot features and the spatiotemporal evolution features of the global temperature field and input them into the fire risk classification model to obtain the fire risk level determination result; when the determination result indicates that there is a high fire risk in the blind spot hotspot, calculate the heat spread time series prediction curve and output accurate spatial positioning and fire early warning signals for fire supervision of electrochemical energy storage facilities; when the determination result indicates that there is no high fire risk in the blind spot hotspot, output the conventional fire supervision status information of the electrochemical energy storage facility.

[0086] The process of obtaining the fire risk level determination result in S40 includes the following steps:

[0087] S401: Obtain the highest temperature of the blind zone hotspot, the temperature difference between the blind zone hotspot and the surrounding batteries, the temperature rise rate of the blind zone, the thermal diffusion trend of the global temperature field, the battery SOC, the battery health status SOH, and the concentration of combustible gases in the cabin, as input features for the fire risk classification model.

[0088] S402: The fire risk classification model is used to perform classification calculations and output four fire risk levels: no abnormality, slight thermal distortion, thermal runaway precursor, and initial fire.

[0089] S403: If the output fire risk level is a precursor to thermal runaway or an initial fire, then a high fire risk is determined to exist in a blind spot hotspot, and a judgment result of a high fire risk in a blind spot hotspot is generated; if the output fire risk level is no abnormality or slight thermal distortion, then a high fire risk in a blind spot hotspot is determined to exist, and a judgment result of a high fire risk in a blind spot hotspot is generated.

[0090] Specifically, in embodiments S401-S403, seven features are extracted as input vectors for the fire risk classification model: the highest temperature of the blind zone hotspot, the temperature difference between the blind zone hotspot and the surrounding batteries, the temperature rise rate of the blind zone, the thermal diffusion trend of the overall temperature field, the battery SOC, the battery SOH, and the concentration of combustible gas in the cabin. The classification model is trained and used to perform classification calculations, outputting four levels of fire risk: no anomaly, slight thermal distortion, precursor to thermal runaway, and initial fire. If the output fire risk level is precursor to thermal runaway or initial fire, it is determined that there is a high fire risk in the blind zone hotspot, and a judgment result of "high fire risk in the blind zone hotspot" is generated. If the output fire risk level is no anomaly or slight thermal distortion, it is determined that there is no high fire risk in the blind zone hotspot, and a judgment result of "high fire risk in the blind zone hotspot" is generated.

[0091] The execution process corresponding to the different determination results in S40 is as follows:

[0092] When the judgment result indicates that there is a high fire risk in the blind spot hotspot, the 0-30min thermal spread time sequence prediction curve is derived based on the complete three-dimensional temperature field grid data of the energy storage compartment. Combined with the three-dimensional coordinates of the blind spot hotspot, the fire risk level and the predicted duration of thermal runaway, a standardized fire early warning signal is generated and output to support the fire emergency supervision and disposal of electrochemical energy storage facilities.

[0093] When the determination result is that there is no blind spot hotspot high fire risk, the current temperature field status of the whole area and the blind spot monitoring results are recorded, the routine fire supervision status information of electrochemical energy storage facilities is output, and real-time monitoring is continuously carried out.

[0094] Specifically, during the execution of the embodiment, when a high fire risk is determined to exist in a blind spot hotspot, a thermal spread time series prediction curve within 0-30 minutes is derived based on the complete three-dimensional temperature field grid data and heat conduction equation of the energy storage compartment. Combining the three-dimensional coordinates of the blind spot hotspot, the fire risk level, and the predicted duration of thermal runaway, a standardized fire early warning signal containing location code, risk level, and disposal suggestions is generated and simultaneously pushed to the site monitoring platform and the terminal of operation and maintenance personnel to support the fire emergency supervision and disposal of electrochemical energy storage facilities. When a high fire risk is determined not to exist in a blind spot hotspot, the current temperature field status of the entire area and the blind spot monitoring results are recorded, and the routine fire supervision status information of the electrochemical energy storage facility is output. The real-time monitoring frequency is maintained, and data collection and analysis continue for the next cycle.

[0095] By establishing a hierarchical risk assessment and closed-loop early warning mechanism that integrates full-domain features, the system balances fire monitoring efficiency with scenario adaptability. It integrates blind spot hotspot characteristics, spatiotemporal evolution characteristics of the full-domain temperature field, and multi-dimensional operating parameters to achieve precise classification of four risk levels through a large-scale fire risk classification model. For high-risk scenarios, it extrapolates the heat spread sequence and outputs precise location warnings, while maintaining routine monitoring for low-risk scenarios, thus constructing a complete closed-loop monitoring system. Relying on lightweight edge-side model inference, it can be deployed and operated without additional computing hardware, adapting to various application scenarios such as containerized energy storage units and fixed energy storage power stations, thereby improving the scientific nature of fire monitoring for electrochemical energy storage facilities.

[0096] For example, consider the specific scenario of thermal anomalies occurring in the dead zone of the battery cell interlayer in a commercial container energy storage compartment:

[0097] During the daily operation of the energy storage module, the system continuously collects multi-source monitoring data at a frequency of 1Hz, performs spatiotemporal registration, inputs a three-dimensional temperature field global reconstruction model, and generates a global three-dimensional temperature field mesh in real time.

[0098] At the 125th minute of operation, after reconstructing the temperature field, the system extracted features from the grid of the cell interlayer blind zone and calculated a matching degree of 0.91 with the blind zone hotspot feature library, which is greater than the matching degree threshold of 0.85. It was determined that there was a hidden hotspot in the temperature monitoring blind zone, and the three-dimensional coordinates of the hotspot were marked as the 5th cell interlayer of the 3rd battery cluster. The temperature rise rate in 1 minute was 1.2℃ / min, and the spatial temperature gradient was 0.8℃ / cm.

[0099] The system extracts blind zone hotspot features and global temperature field evolution features and inputs them into the fire risk classification model. The output risk level is a precursor to thermal runaway, which is judged as a high fire risk.

[0100] The system then simulates the 0-30 minute thermal runaway timeline prediction curve, predicting that thermal runaway will spread to adjacent battery clusters after 18 minutes. It generates a standardized fire warning signal that includes precise three-dimensional positioning, risk level, predicted thermal runaway time, and handling suggestions, and pushes it to the site monitoring center and the mobile terminals of maintenance personnel. At the same time, it links the ventilation system to increase the wind speed for auxiliary heat dissipation, leaving sufficient time for emergency response and preventing fire accidents.

[0101] Example 2:

[0102] Please see Figures 1-2 A fire monitoring system for electrochemical energy storage facilities based on a large model is provided to implement the fire monitoring method for electrochemical energy storage facilities based on a large model. The system includes:

[0103] Multi-source spatiotemporal registration and spatial mapping module: Collects multi-source monitoring data of electrochemical energy storage facilities. The multi-source monitoring data includes temperature of sparsely arranged fixed-point thermocouple measuring points, battery voltage, battery current, battery state of charge (SOC), local infrared thermal imaging images inside the cabin, ventilation and heat dissipation parameters, and ambient temperature and humidity. The module performs spatiotemporal registration on the multi-source monitoring data and constructs a three-dimensional spatial coordinate mapping relationship for energy storage. It generates a sparse measuring point temperature dataset and a three-dimensional geometric spatial association sample of the energy storage cabin, which are suitable for fire monitoring of electrochemical energy storage facilities.

[0104] The 3D temperature field reconstruction and blind spot hotspot mining module: Based on the sparse measurement point temperature dataset and the 3D geometric spatial correlation samples of the energy storage chamber, and combined with the energy storage thermal runaway mechanism, the module performs domain fine-tuning on the lightweight industry large model to construct a 3D temperature field global reconstruction large model suitable for fire monitoring of electrochemical energy storage facilities; simultaneously, it uses the DBSCAN density clustering algorithm to perform feature clustering analysis on historical temperature blind spot missed samples, and establishes a blind spot hotspot feature library suitable for fire monitoring of electrochemical energy storage facilities based on the clustering analysis results; outputs the blind spot hotspot identification results and executes corresponding operations according to the path;

[0105] Fire risk classification and early warning output module: Based on the three-dimensional temperature field global reconstruction model and the blind spot hotspot feature library, it outputs complete three-dimensional temperature field grid data of the energy storage compartment suitable for fire monitoring of electrochemical energy storage facilities, identifies hidden hotspots in temperature monitoring blind spots, classifies four levels of fire risk based on the spatiotemporal evolution characteristics of the global temperature field, calculates the heat spread time series prediction curve when high risk is determined, and outputs accurate spatial positioning and fire early warning signals for fire monitoring of electrochemical energy storage facilities.

[0106] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A fire monitoring method for electrochemical energy storage facilities based on a large model, characterized in that, Includes the following steps: S10: Collect multi-source monitoring data of the electrochemical energy storage facility. The multi-source monitoring data includes temperature of sparsely arranged fixed-point thermocouple measuring points, battery voltage, battery current, battery state of charge (SOC), local infrared thermal imaging images inside the cabin, ventilation and heat dissipation parameters, and ambient temperature and humidity. Perform spatiotemporal registration on the multi-source monitoring data and construct a three-dimensional spatial coordinate mapping relationship for energy storage. Generate a sparse measuring point temperature dataset and a three-dimensional geometric spatial correlation sample of the energy storage cabin suitable for fire monitoring of electrochemical energy storage facilities. S20: Based on the sparse temperature measurement point dataset and the three-dimensional geometric space correlation samples of the energy storage chamber, and combined with the energy storage thermal runaway mechanism, the lightweight industry large model is fine-tuned in the domain to construct a three-dimensional temperature field global reconstruction large model suitable for fire monitoring of electrochemical energy storage facilities; at the same time, the DBSCAN density clustering algorithm is used to perform feature clustering analysis on the historical temperature blind zone missed samples, and a blind zone hotspot feature library suitable for fire monitoring of electrochemical energy storage facilities is established based on the clustering analysis results; S30: Based on the three-dimensional temperature field global reconstruction model and the blind spot hotspot feature library, real-time multi-source monitoring data is input into the three-dimensional temperature field global reconstruction model, and the complete three-dimensional temperature field grid data of the energy storage compartment suitable for fire monitoring of electrochemical energy storage facilities is output; and the three-dimensional temperature field grid data is matched with the blind spot hotspot feature library to obtain the blind spot hotspot identification result; when the identification result is that there is a hidden hotspot in the temperature monitoring blind spot, the three-dimensional coordinates, temperature rise rate and temperature gradient of the blind spot hotspot used for fire hazard assessment are marked; when the identification result is that there is no hidden hotspot in the temperature monitoring blind spot, the monitoring status of no blind spot thermal anomaly is output; S40: Based on the three-dimensional coordinates of the blind spot hotspot, the temperature rise rate, the temperature gradient, and the complete three-dimensional temperature field grid data of the energy storage compartment, extract the blind spot hotspot features and the spatiotemporal evolution features of the global temperature field and input them into the fire risk classification model to obtain the fire risk level determination result. When the judgment result indicates that there is a high fire risk in the blind spot hotspot, the heat spread time series prediction curve is calculated, and the precise spatial positioning and fire early warning signal for fire supervision of electrochemical energy storage facilities are output. When the determination result is that there are no blind spots or hotspots with high fire risk, the system outputs the conventional fire monitoring status information of the electrochemical energy storage facility.

2. The fire monitoring method for electrochemical energy storage facilities based on a large model according to claim 1, characterized in that, In S10, multi-source monitoring data is acquired through corresponding acquisition devices and uniformly accessed by the edge data acquisition platform. Specifically, the temperature of the fixed thermocouple measuring point is acquired by the distributed multi-channel thermocouple acquisition terminal, the battery voltage, battery current, and battery state of charge (SOC) are acquired by the energy storage battery management system (BMS), the local infrared thermal imaging image inside the cabin is acquired by the fixed infrared thermal imager deployed on the top of the cabin, the ventilation and heat dissipation parameters are acquired by the fan speed sensor and wind pressure transmitter of the energy storage cabin ventilation control system, and the ambient temperature and humidity are acquired by the digital temperature and humidity sensors deployed inside and outside the cabin. The edge data acquisition platform performs protocol parsing, outlier removal, and data temporary storage processing on each acquired data source and outputs standardized preprocessed data.

3. The fire monitoring method for electrochemical energy storage facilities based on a large model according to claim 2, characterized in that, The steps involved in S10 to uniformly perform spatiotemporal registration and construct the three-dimensional spatial coordinate mapping relationship for energy storage include: S101: Using the unified system clock of the edge data acquisition platform as the time reference, time registration is performed on multi-source monitoring data with different sampling frequencies using linear interpolation to generate synchronous multi-source data groups at the same time. S102: Pre-store the three-dimensional point cloud model of the energy storage cabin obtained by three-dimensional laser scanning, assign a unique spatial identifier to each acquisition device, and bind the actual installation coordinates of each fixed-point thermocouple measuring point to the corresponding spatial position of the three-dimensional point cloud model. S103: Camera calibration and homography matrix calculation are performed on the local infrared thermal images acquired by the fixed infrared thermal imager to establish the mapping relationship between the pixel coordinates of the infrared image and the coordinates of the three-dimensional point cloud, and to complete the spatial dimension registration. S104: Outputs a sparse temperature dataset with spatiotemporal labels and a three-dimensional geometric spatial association sample of the energy storage compartment.

4. The fire monitoring method for electrochemical energy storage facilities based on a large model according to claim 1, characterized in that, The establishment of the blind spot hotspot feature database in S20 includes the following steps: S201: Extract historical data of missed detection blind zone accidents, using three-dimensional coordinate position, temperature difference at measuring point, temperature rise gradient, SOC range and heat dissipation wind speed as multi-dimensional features; S202: The DBSCAN density clustering algorithm is used, and the neighborhood radius and minimum number of points are set as clustering parameters to obtain multiple blind spot hotspot distribution clusters corresponding to different types of fire hazards; S203: Integrate all blind spot hotspot distribution clusters to generate a standardized blind spot hotspot feature library suitable for fire monitoring of electrochemical energy storage facilities.

5. The fire monitoring method for electrochemical energy storage facilities based on a large model according to claim 1, characterized in that, The construction of the three-dimensional temperature field global reconstruction model in S20 includes the following steps: S204: Collect finite element simulation data of full-domain temperature field of multi-capacity, multi-arrange energy storage compartments and sparse measurement point data from the field to construct a dual-source training set; S205: Based on a general multimodal large model, it incorporates prior knowledge of battery heat transfer and cell thermal runaway electrochemistry for domain fine-tuning; S206: Introducing sparse measurement point reconstruction loss constraints to complete model training, outputting a large-scale three-dimensional temperature field global reconstruction model suitable for fire monitoring of electrochemical energy storage facilities.

6. The fire monitoring method for electrochemical energy storage facilities based on a large model according to claim 1, characterized in that, The process of obtaining the blind spot hotspot identification result in S30 includes the following steps: S301: The reconstructed continuous three-dimensional temperature field grid data is divided into the visible area of ​​the battery body and the blind areas of the cell interlayer, back of the air duct and bottom of the cable that cannot be covered by the sensor. The temperature change rate and the second derivative of the spatial temperature gradient are extracted point by point from the grid of the blind area. S302: Calculate the cosine similarity between the blind zone grid features and the blind zone hotspot feature library to obtain the matching degree; S303: Preset matching degree threshold; if the matching degree is greater than or equal to the matching degree threshold, it is determined that there is a hidden hot spot with fire hazard in the corresponding blind zone grid, and an identification result of hidden hot spot with temperature monitoring blind zone is generated; if the matching degree is less than the matching degree threshold, it is determined that there is no fire thermal anomaly in the corresponding blind zone grid, and an identification result of no hidden hot spot with temperature monitoring blind zone is generated.

7. The fire monitoring method for electrochemical energy storage facilities based on a large model according to claim 1, characterized in that, The process of obtaining the fire risk level determination result in S40 includes the following steps: S401: Obtain the highest temperature of the blind zone hotspot, the temperature difference between the blind zone hotspot and the surrounding batteries, the temperature rise rate of the blind zone, the thermal diffusion trend of the global temperature field, the battery SOC, the battery health status SOH, and the concentration of combustible gases in the cabin, as input features for the fire risk classification model. S402: The fire risk classification model is used to perform classification calculations and output four fire risk levels: no abnormality, slight thermal distortion, thermal runaway precursor, and initial fire. S403: If the output fire risk level is a precursor to thermal runaway or an initial fire, then a high fire risk is determined to exist in a blind spot hotspot, and a judgment result of a high fire risk in a blind spot hotspot is generated; if the output fire risk level is no abnormality or slight thermal distortion, then a high fire risk in a blind spot hotspot is determined to exist, and a judgment result of a high fire risk in a blind spot hotspot is generated.

8. The fire monitoring method for electrochemical energy storage facilities based on a large model according to claim 1, characterized in that, The execution process corresponding to the different determination results in S40 is as follows: When the judgment result indicates that there is a high fire risk in the blind spot hotspot, the 0-30min thermal spread time sequence prediction curve is derived based on the complete three-dimensional temperature field grid data of the energy storage compartment. Combined with the three-dimensional coordinates of the blind spot hotspot, the fire risk level and the predicted duration of thermal runaway, a standardized fire early warning signal is generated and output to support the fire emergency supervision and disposal of electrochemical energy storage facilities. When the determination result is that there is no blind spot hotspot high fire risk, the current temperature field status of the whole area and the blind spot monitoring results are recorded, the routine fire supervision status information of electrochemical energy storage facilities is output, and real-time monitoring is continuously carried out.

9. A fire monitoring system for electrochemical energy storage facilities based on a large model, characterized in that, For implementing the fire monitoring method for electrochemical energy storage facilities based on a large model as described in any one of claims 1-8, the system comprises: Multi-source spatiotemporal registration and spatial mapping module: Collects multi-source monitoring data of electrochemical energy storage facilities. The multi-source monitoring data includes temperature of sparsely arranged fixed-point thermocouple measuring points, battery voltage, battery current, battery state of charge (SOC), local infrared thermal imaging images inside the cabin, ventilation and heat dissipation parameters, and ambient temperature and humidity. The module performs spatiotemporal registration on the multi-source monitoring data and constructs a three-dimensional spatial coordinate mapping relationship for energy storage. It generates a sparse measuring point temperature dataset and a three-dimensional geometric spatial association sample of the energy storage cabin, which are suitable for fire monitoring of electrochemical energy storage facilities. The 3D temperature field reconstruction and blind spot hotspot mining module: Based on the sparse measurement point temperature dataset and the 3D geometric spatial correlation samples of the energy storage chamber, and combined with the energy storage thermal runaway mechanism, the module performs domain fine-tuning on the lightweight industry large model to construct a 3D temperature field global reconstruction large model suitable for fire monitoring of electrochemical energy storage facilities; simultaneously, it uses the DBSCAN density clustering algorithm to perform feature clustering analysis on historical temperature blind spot missed samples, and establishes a blind spot hotspot feature library suitable for fire monitoring of electrochemical energy storage facilities based on the clustering analysis results; outputs the blind spot hotspot identification results and executes corresponding operations according to the path; Fire risk classification and early warning output module: Based on the three-dimensional temperature field global reconstruction model and the blind spot hotspot feature library, it outputs complete three-dimensional temperature field grid data of the energy storage compartment suitable for fire monitoring of electrochemical energy storage facilities, identifies hidden hotspots in temperature monitoring blind spots, classifies four levels of fire risk based on the spatiotemporal evolution characteristics of the global temperature field, calculates the heat spread time series prediction curve when high risk is determined, and outputs accurate spatial positioning and fire early warning signals for fire monitoring of electrochemical energy storage facilities.

Images