A lithium battery liquid nitrogen fire extinguishing scheme generation method and system based on fire alarm levels

Abstract

The application discloses a lithium battery liquid nitrogen fire extinguishing scheme generation method and system based on fire alarm grades, and belongs to the field of lithium battery energy storage fire prevention and control. In view of the defects of the existing liquid nitrogen fire extinguishing system, such as fixed threshold triggering, no hierarchical response, rigid spraying parameters, fire control lag and waste of fire extinguishing agent, the application constructs a four-layer architecture of data acquisition, model analysis, decision optimization and control execution. By fusing BIM structure, multiple sensors, BMS and historical fire data, a three-level fire grade differential simulation database is established by FDS; the fire spread, liquid nitrogen diffusion cooling and oxygen concentration replacement models are trained to determine the fire grade in real time, and the optimal liquid nitrogen spraying scheme is generated by NSGA multi-objective optimization; and the parameters are dynamically corrected according to the fire deviation, and the model is iteratively learned incrementally. The application realizes accurate identification of fire grades and self-adaptive regulation and control of liquid nitrogen spraying, shortens the response time, reduces the liquid nitrogen consumption, improves the system robustness, and is suitable for various energy storage cabin active fire prevention and explosion suppression scenes.

Description

Technical Field

[0001] This invention belongs to the field of fire prevention and control technology for lithium battery energy storage systems, specifically relating to a method and system for generating a liquid nitrogen fire extinguishing scheme for lithium batteries based on fire alarm levels. Background Technology

[0002] Lithium-ion batteries are widely used in energy storage power stations, new energy vehicles, and distributed energy storage scenarios due to their high energy density. However, lithium-ion batteries have a high probability of thermal runaway, and fires involving them are characterized by rapid combustion and explosion, strong chain propagation, high risk of reignition, and large-scale release of toxic and flammable gases. Traditional water-based and foam fire extinguishing agents are ineffective at suppressing deep thermal runaway in lithium-ion batteries, easily causing battery short circuits, equipment corrosion, and environmental pollution. Conventional gaseous fire extinguishing agents have insufficient cooling capacity and cannot quickly cut off the chain reaction of thermal runaway.

[0003] Liquid nitrogen, with its advantages of rapid cooling at ultra-low temperatures, inertization and asphyxiation, and lack of residue and pollution, has become the preferred extinguishing medium for lithium battery fires. Existing lithium battery liquid nitrogen fire extinguishing systems generally employ a crude control mode with fixed threshold triggering, fixed discharge duration, and fixed flow rate. This fails to differentiate between different stages of a fire, such as early warning, open flame development, and high-risk explosion, and lacks a dynamic control mechanism linked to fire alarm levels. Furthermore, it does not integrate the cabin's BIM structure, multi-sensor multi-dimensional features, and fire spread prediction, resulting in problems such as delayed response, excessive extinguishing agent waste, untimely fire control, poor adaptability, and inability to adapt to fire evolution. Consequently, it struggles to meet the engineering requirements for precise fire prevention and explosion suppression in large-scale lithium battery energy storage cabins. Summary of the Invention

[0004] This invention aims to overcome the shortcomings of existing technologies, such as insufficient extinguishing agent efficiency of lithium battery fire extinguishing agents, sluggish system response, rigid spray parameters, lack of graded differentiated control, weak adaptive capability, resource waste, and insufficient suppression of reignition. It provides a method and system for generating lithium battery liquid nitrogen fire extinguishing schemes based on fire alarm levels, realizing multi-source data fusion, intelligent fire level determination, accurate fire prediction, multi-objective optimization of liquid nitrogen spray parameters, and dynamic closed-loop correction.

[0005] To achieve the above objectives, the technical solution adopted in this invention is: a method for generating lithium battery liquid nitrogen fire extinguishing schemes based on fire alarm levels, comprising: multi-source heterogeneous data acquisition and preprocessing, FDS simulation and dedicated database construction, multi-prediction model training, generation and dynamic correction of graded fire extinguishing schemes, and a complete process of strategy self-evolution. Simultaneously, a four-layer intelligent system adapted to this method is provided, consisting of a data acquisition layer, a model analysis layer, a decision optimization layer, and a control execution layer working collaboratively to achieve intelligent operation throughout the entire process from perception, modeling, decision-making to execution.

[0006] This invention first collects multi-source data on BIM cabin structure, multi-sensor environment, BMS battery status, and historical fire cases, and performs hierarchical standardization and feature vectorization according to the three-level fire alarm level. Then, it uses FDS to perform differentiated simulations of different fire levels, and integrates measured data and simulation data to build a dedicated database. It then trains three intelligent models in sequence: fire spread, liquid nitrogen diffusion cooling, and oxygen concentration replacement. It integrates multiple parameters in real time to determine the fire level, and uses a multi-objective optimization algorithm to generate the optimal liquid nitrogen injection scheme matching the level. It sets deviation thresholds and time constraints to automatically reconstruct the fire scene and re-optimize injection parameters when the fire deviates or the level escalates. Finally, it uses incremental learning to achieve continuous iterative optimization of the model and fire extinguishing strategy.

[0007] The system's four-layer architecture each has its own function: the data acquisition layer is responsible for multi-source sensing, BIM modeling, data preprocessing, and 3D topology database construction; the model analysis layer integrates FDS simulation, level determination, and three major prediction models to complete the prediction of fire situation and fire extinguishing effect; the decision optimization layer generates Pareto optimal solutions based on multi-objective optimization algorithms and has real-time dynamic feedback correction capabilities; and the control execution layer is responsible for the start-up and shutdown of liquid nitrogen pipeline solenoid valves and wide-angle nozzles, as well as flow control, and provides a 3D visualization operation and maintenance interface.

[0008] Specifically, this includes: a method for generating a lithium battery liquid nitrogen fire extinguishing scheme based on fire alarm levels, comprising the following steps:

[0009] S101. Multi-source heterogeneous data acquisition and preprocessing: Collect BIM structural data of lithium battery energy storage compartment, environmental sensor data inside the compartment, BMS operation data of battery management system and historical fire case data of lithium battery; classify and standardize heterogeneous data according to three fire alarm levels: Level I warning, Level II open flame and Level III explosion risk, and vectorize the discrete features.

[0010] S102, Simulation Modeling and Dedicated Database Construction: The BIM structural model of the lithium battery energy storage compartment is imported into the FDS fire dynamics simulation platform. Differentiated fire and liquid nitrogen injection numerical simulations are performed for Level I warning, Level II open flame, and Level III explosion risks, respectively, to generate a fire feature vector dataset with fire level labels. The fire feature vector dataset is then fused with the multi-source data preprocessed in step S1 to construct a dedicated database for lithium battery liquid nitrogen fire extinguishing.

[0011] S103, Multi-intelligent prediction model training: Based on the lithium battery liquid nitrogen fire extinguishing dedicated database, fire spread prediction model, liquid nitrogen diffusion cooling model, and oxygen concentration replacement model embedded with fire alarm level features are constructed respectively.

[0012] S104. Generation and Dynamic Correction of Graded Fire Extinguishing Plans: Real-time acquisition of multi-source sensor data from the lithium battery energy storage compartment, integrating temperature gradient, smoke density, combustible gas concentration, and thermal radiation flux parameters to determine the current fire alarm level; inputting the fire level and real-time fire characteristics into three prediction models, employing a multi-objective optimization algorithm to generate a Pareto-optimal liquid nitrogen fire extinguishing spray plan based on fire level-differentiated constraints; real-time comparison of sensor-measured data with model prediction data, when the deviation between fire spread parameters and predicted values ​​exceeds a preset deviation threshold or the fire alarm level is upgraded, and the oxygen concentration fails to drop to the target oxygen concentration threshold for the corresponding fire alarm level within a preset response time, the current fire status is redefined and the liquid nitrogen spray parameters are updated, achieving dynamic adaptive adjustment of the fire extinguishing plan.

[0013] In step S101, the cabin environment sensing data includes one or more of the following: infrared thermal imager temperature field data, smoke detector smoke density data, CO / H2 / VOCs multi-gas concentration data, temperature and humidity data, and wind speed data; the BIM structural data includes: battery cluster spatial coordinates, ventilation opening locations, obstacle distribution, fireproof partition parameters, cable layout, and cabin airtightness parameters; the BMS operating data includes: battery voltage, current, SOC gradient, and coordinates of the initial position of thermal runaway.

[0014] The hierarchical standardization process described in step S101 specifically includes:

[0015] Enhance the annotation of local temperature abrupt changes and early smoke diffusion characteristics in Level I warning data;

[0016] Enhance the labeling of combustible / toxic gas concentrations, heat radiation flux, and open flame development characteristics for Class II open flame data;

[0017] Enhance the labeling of pressure fluctuations, combustible gas enrichment, rapid flame propagation, and deflagration risk characteristics for Level III explosion risk data;

[0018] Unique thermal coding is used to vectorize discrete tags for battery chemistry system, cabin airtightness level, and fire alarm level.

[0019] Step S102, the FDS differential simulation settings include: configuring different heat release rates, flame propagation speeds and explosion overpressure peaks according to the fire level; setting corresponding heat release parameters, flame propagation parameters, liquid nitrogen injection parameters and mesh generation accuracy according to different fire alarm levels, and outputting spatiotemporal evolution data of temperature field, oxygen concentration field, liquid nitrogen coverage and temperature drop rate under different levels.

[0020] In step S103, the fire spread prediction model uses a hierarchical adaptive LSTM network. The input includes the fire alarm level encoding vector, the battery cluster BIM spatial topology, the initial thermal runaway coordinates, the ventilation intensity, and the battery SOC gradient. The output is a hierarchical fire spread path, propagation speed, and thermal distribution. The liquid nitrogen diffusion cooling model uses a random forest or XGBoost regression algorithm. The input includes the injection port coordinates, injection flow rate, fire alarm level, ambient temperature, and obstacle density. The output is the cooling radius, minimum temperature, and temperature drop rate. The oxygen concentration replacement model uses a hierarchical SVR algorithm. The input includes the nitrogen injection rate, initial oxygen concentration, fire alarm level, cabin leakage rate, and ventilation volume. The output is the time it takes for the oxygen concentration to drop to the safe threshold and the asphyxiation area coverage.

[0021] The fire alarm level determination model described in step S104 is based on one or more parameters among temperature change rate, smoke density, combustible gas concentration, and thermal radiation flux; based on the multi-parameter fusion results, it outputs Level I warning, Level II open flame, and Level III explosion risk and their corresponding confidence levels, with Level I confidence level ≥ 0.85 and Level III confidence level ≥ 0.95.

[0022] In step S104, the multi-objective optimization algorithm takes minimizing liquid nitrogen consumption, minimizing fire extinguishing time, and maximizing oxygen suppression efficiency as its objective functions; the optimization variables include nozzle combination sequence, single nozzle spray flow rate, spray duration, spray pressure, and nozzle start-stop sequence.

[0023] In step S104, differentiated constraints and spray modes are set according to the fire alarm level: Level I warning corresponds to lower liquid nitrogen usage constraints and shorter response time constraints, and a local fixed-point spray mode is adopted; Level II open flame corresponds to medium liquid nitrogen usage constraints and extinguishing time constraints, and a total flooding or area coverage spray mode is adopted; Level III explosion risk corresponds to enhanced explosion suppression constraints and target oxygen concentration constraints, and a backup nozzle is activated or the spray intensity is increased.

[0024] The dynamic correction process in step S104 includes: collecting multi-source sensor data according to a preset sampling period and comparing it with the model prediction value; when the oxygen concentration fails to drop to the corresponding level threshold within the preset response time, the fire spread deviation exceeds the preset deviation threshold, or the fire level changes, the fire reconstruction module is called to redetermine the current fire status, and multi-objective optimization is re-executed to update the number of nozzles opened, the spray flow rate, and the spray duration.

[0025] It also includes a strategy self-evolution step: recording the parameters of the liquid nitrogen injection scheme and the actual fire extinguishing effect under each fire alarm level, using incremental learning to continuously iterate and optimize the internal parameters of the fire spread prediction model, the liquid nitrogen diffusion cooling model, and the oxygen concentration replacement model, and solidifying the optimal injection strategy for different cabin layouts and battery types.

[0026] A lithium battery liquid nitrogen fire extinguishing scheme generation system based on fire alarm level, implementing the method described in any one of the above, includes: a data acquisition layer, a model analysis layer, a decision optimization layer, and a control execution layer with hierarchical communication connection;

[0027] The data acquisition layer is used to collect BIM structural data of lithium battery energy storage compartment, multi-sensor environmental data, BMS battery operation data and historical fire case data, complete data cleaning, hierarchical standardization and feature vectorization, and construct a three-dimensional topology database.

[0028] The model analysis layer includes an FDS numerical simulation module, a fire alarm level determination model, a fire spread prediction model, a liquid nitrogen diffusion cooling model, and an oxygen concentration replacement model.

[0029] The decision optimization layer includes a built-in NSGA-II / NSGA-III multi-objective optimization module, a dynamic feedback correction module, and a solution library storage module.

[0030] The control execution layer includes a liquid nitrogen storage tank, an electromagnetic control valve, multiple sets of wide-angle spray nozzles, a three-dimensional visualization terminal, and an infrared thermal imaging monitoring unit. It is used to receive decision commands to control the start and stop of the nozzles and adjust the flow rate, and to display the liquid nitrogen coverage area, temperature field, and oxygen concentration gradient in real time using three-dimensional visualization.

[0031] Furthermore, the data acquisition layer includes an infrared thermal imager, a smoke detector, a CO / H2 / VOCs multi-gas sensor, a temperature, humidity and anemometer, a BIM modeling unit, a BMS data interaction unit, and a historical fire case database.

[0032] Furthermore, the system adopts an edge-localized deployment architecture with a built-in lightweight model inference engine, eliminating the need to rely on cloud computing power; the system response time is reduced by more than 40%, liquid nitrogen consumption is reduced by 25% to 30%, and the overall deployment and maintenance costs are reduced by 20%.

[0033] Furthermore, the dynamic feedback correction module presets a 20% fire spread deviation threshold and a 10-second oxygen concentration response time limit; when the correction is triggered, it automatically opens backup spray nozzles and increases the spray flow rate to block the chain reaction of lithium battery thermal runaway and the risk of reignition.

[0034] Beneficial effects

[0035] Compared with existing traditional lithium battery fire extinguishing technology and fixed-mode liquid nitrogen fire extinguishing systems, this invention has the following outstanding advantages:

[0036] 1) Achieve accurate fire level identification and rapid response

[0037] This invention integrates multi-dimensional features such as temperature gradient, smoke density, combustible gas concentration, and thermal radiation flux to establish a three-level fire alarm determination mechanism: Level I early warning, Level II open flame, and Level III explosion risk. It combines BIM topology and LSTM time-series prediction to predict the fire spread path in advance. Compared to traditional fixed threshold triggering methods, the system response time is reduced by more than 40%, and the efficiency of fire extinguishing plan generation is improved by approximately 50%. It can achieve accurate identification and early intervention in the early stages of lithium battery thermal runaway, preventing chain-like thermal runaway spread.

[0038] 2) Significantly saves liquid nitrogen medium and reduces the risk of equipment damage.

[0039] This invention introduces the NSGA-II / NSGA-III multi-objective optimization algorithm to differentiate the constraints on liquid nitrogen usage, extinguishing duration, and asphyxiation efficiency according to different fire levels, and adaptively optimizes the number of nozzles, spray flow rate, and spray duration. In practical applications, liquid nitrogen consumption is reduced by 25% to 30%, eliminating the waste of extinguishing agent caused by traditional fixed-flow, long-duration spraying; at the same time, it avoids excessive liquid nitrogen cryogenic shock, reducing the risk of cryogenic damage to energy storage equipment, battery clusters, and pipelines.

[0040] 3) The system's adaptive capability and anti-interference robustness are significantly improved.

[0041] This invention employs a dual feedback mechanism: a 20% fire spread deviation threshold and a 10-second oxygen concentration attainment time limit. If the fire deviates from the prediction or the fire severity escalates, the system automatically reconstructs the fire scene and re-optimizes the spraying strategy. The system is compatible with different lithium battery chemistry systems, energy storage compartment layouts, and sealed operating conditions, exhibiting strong generalization and adaptability. Its robustness against environmental interference and sudden fire changes is improved by approximately 35%, effectively addressing the re-ignition and deflagration of lithium battery fires.

[0042] 4) Full-process visualized intelligent decision-making enables a proactive transformation in prevention and control.

[0043] This invention integrates fire spread prediction, FDS liquid nitrogen diffusion simulation, multi-objective optimization decision-making, and a 3D visualization platform. It displays the temperature field, oxygen concentration field, liquid nitrogen coverage area, and the entire fire suppression process in real time, providing maintenance personnel with intuitive and visual decision-making support. This promotes the upgrading of lithium battery fire suppression systems from traditional passive post-fire suppression to a proactive prevention and control mode characterized by graded early warning, precise explosion suppression, and full controllability.

[0044] 5) Reduce deployment and maintenance costs, and adapt to lightweight edge deployment.

[0045] This invention is based on a pre-built 3D topology library using BIM and trains a lightweight model using historical cases, reducing reliance on high-end computing hardware. It supports localized deployment at the edge, eliminating cloud transmission latency and enabling a closed-loop process of perception, modeling, decision-making, and execution on-site. The overall system deployment cost is reduced by approximately 20%, and subsequent incremental learning allows for automatic model iteration, eliminating the need for repeated manual parameter adjustments and significantly reducing maintenance and upgrade workload. Attached Figure Description

[0046] Fig. 1 This is a schematic diagram of the method flow of the present invention;

[0047] Fig. 2 This is a block diagram of the system structure of the present invention. Detailed Implementation

[0048] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. However, it should be understood that the specific embodiments described herein are merely illustrative and are not intended to limit the scope of the invention.

[0049] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of this invention is for the purpose of describing particular embodiments only and is not intended to limit the invention.

[0050] like Figs. 1-2 As shown, a method and system for generating liquid nitrogen fire extinguishing schemes for lithium batteries based on fire alarm levels is presented. The system is divided into an initialization phase, a real-time response phase, a dynamic correction phase, and a strategy self-evolution phase. The overall system architecture consists of four layers: a data acquisition layer, a model analysis layer, a decision optimization layer, and a control execution layer. Relying on BIM modeling, FDS numerical simulation, multi-sensor data fusion, machine learning models, and NSGA-II / NSGA-III multi-objective optimization algorithms, the system dynamically and adaptively generates liquid nitrogen spray fire extinguishing schemes based on three fire alarm levels: Level I warning, Level II open flame, and Level III explosion risk.

[0051] Overall implementation process of the method

[0052] This invention relates to a method for generating a lithium battery liquid nitrogen fire extinguishing scheme based on fire alarm levels, comprising the following steps:

[0053] Step S101: Collect structural data, environmental data, and on-site sensor data of the lithium battery energy storage compartment.

[0054] Using web crawling technology, historical fire case data of lithium battery energy storage compartments were systematically obtained from public fire databases and industry reports. Key fields were labeled according to fire alarm level (Level I warning / Level II open flame / Level III explosion risk), including fire location coordinates, thermal runaway triggering conditions, fire spread path time sequence data, extinguishing agent type, liquid nitrogen dosage and extinguishing efficiency, etc.

[0055] The collected heterogeneous data is subjected to hierarchical standardization processing:

[0056] Level I warning data focuses on the characteristics of temperature abrupt changes and early smoke diffusion patterns;

[0057] Enhanced correlation analysis between peak CO concentration and thermal radiation flux using Class II open flame data;

[0058] Level III explosion risk data includes the detection of pressure fluctuations and flame propagation acceleration.

[0059] One-hot encoding is used to vectorize discrete features, such as battery chemistry system, cabin airtightness level, and fire alarm level labels, to improve the model's compatibility with level features.

[0060] Meanwhile, high-precision temperature and humidity sensors and anemometers are used to monitor the microenvironment parameters inside the cabin in real time at a sampling frequency of 1Hz; infrared thermal imagers are used to dynamically capture the temperature field distribution on the battery surface, and multiple gas sensors are used to simultaneously collect the concentration gradients of CO, H2, and VOCs; the energy storage cabin design drawings are imported into the BIM modeling software Revit to extract three-dimensional topological data, including battery cluster spatial coordinates, ventilation system flow parameters, fire resistance limits of fire-resistant partitions, obstacle distribution matrix, cable type and voltage level, etc., providing multi-dimensional structured input for the subsequent generation of liquid nitrogen injection strategies according to fire alarm levels.

[0061] Step S102: Based on the structural data of the lithium battery energy storage compartment, use FDS to simulate fires of different fire alarm levels to form a fire feature vector database with level labels, and combine it with the multi-source data collected in S101 to construct a lithium battery liquid nitrogen fire extinguishing database.

[0062] The BIM structural model of the lithium battery energy storage compartment was imported into the FDS (FireDynamicsSimulator) numerical simulation platform, and differentiated three-dimensional simulation models were constructed according to fire alarm levels I / II / III:

[0063] Level I warning scenario: Set a low heat release rate curve with a peak value of 300kW / m² and a duration of 60s to simulate early thermal runaway;

[0064] Level II open flame scenario: using standard heat release rate, peak value 1.2MW / m², lasting 120s, corresponding to full fire development;

[0065] Level III explosion risk scenario: Integrated overpressure release model, peak pressure 0.3MPa, flame velocity ≥3m / s.

[0066] Optimize physical parameters for different levels:

[0067] Level I warning focuses on simulating the effect of local ventilation with a wind speed of 1.5 m / s to enhance smoke diffusion; Level III explosion risk simulates the coupling of shock wave and turbulence with a turbulence intensity ≥10%.

[0068] Mesh generation employs an unstructured variable-scale strategy:

[0069] In the Level I warning area, the fire source grid is densified to 0.05m×0.05m×0.05m to focus on early hotspots;

[0070] The grid for Level II / III fire source areas is refined to 0.03m×0.03m×0.03m to capture details of deflagration;

[0071] The uniform grid size for the liquid nitrogen diffusion zone is 0.08m × 0.08m × 0.08m, and the global maximum grid size is ≤0.25m.

[0072] In FDS simulations, monitoring networks are arranged according to levels:

[0073] For Level I warning, temperature slices are set between battery clusters at 0.3m intervals to monitor local temperature rise; for Level III explosion risk, gas concentration monitoring lines are added, with H2 / CO points placed every 0.2m.

[0074] In the simulation, liquid nitrogen operating parameters are set according to different levels:

[0075] Level I: Low-flow laminar flow model is adopted, with a flow rate of 5-10 L / min and a droplet size of 30-100 μm per minute (DPM).

[0076] Level III: High flow rate turbulence model is enabled, with a flow rate of 15-20 L / min, a particle size of 100-180 μm, and a Realizable k-ε model with a turbulence intensity of 5%.

[0077] Liquid nitrogen property parameter settings:

[0078] Liquid nitrogen has a density of 808 kg / m³ (liquid at -196℃) and 1.25 kg / m³ (gaseous), a latent heat of vaporization of 199 kJ / kg, and a dynamic viscosity of 1.6 × 10⁻⁻⁻⁶. 5 Pa・s.

[0079] Boundary condition settings:

[0080] The nitrogen injection port is set as a velocity inlet with a flow rate of 5-25 m / s; the vent is set as a pressure outlet with a static pressure of 0 Pa.

[0081] The simulation outputs the time for oxygen concentration to drop to the threshold (12%) at each level, the coverage radius of the low-temperature area (≤-50℃), and the nitrogen gas integral cloud map. Based on the simulation results, spatiotemporal evolution data with level labels are extracted, including timestamps, coordinates, temperature, O2 / CO concentration, liquid nitrogen coverage, and fire alarm level. Finally, it is fused with multi-source data from S101 to construct a complete lithium battery liquid nitrogen fire extinguishing database with a total data volume of ≥100,000 records.

[0082] Step S103: Based on the lithium battery liquid nitrogen fire extinguishing database, construct a fire spread prediction model, a liquid nitrogen diffusion cooling model, and an oxygen concentration replacement model that incorporate fire alarm levels.

[0083] The training data required for the fire spread prediction model, liquid nitrogen diffusion cooling model, and oxygen concentration replacement model all come from the lithium battery liquid nitrogen fire extinguishing database. This database integrates historical thermal runaway cases, real-time monitoring data from multiple sensors (temperature, smoke, H2 / CO concentration), and the spatiotemporal evolution sequence of fire output from FDS simulation.

[0084] (1) Fire spread prediction model

[0085] The system employs a hierarchical adaptive LSTM architecture, with input features including: fire alarm level (I / II / III level encoding vectors), battery cluster spatial topology (based on BIM coordinate encoding), initial thermal runaway location (x,y,z), ventilation intensity (m³ / h), and SOC gradient (%).

[0086] The training data were taken from samples of each level in the database: 500 cases for Level I, 300 cases for Level II, and 200 cases for Level III. The model output is a graded heat map of fire spread. The Level I warning predicts the local spread path 30 seconds in advance, and the Level III explosion risk corrects the direction of shock wave propagation in real time.

[0087] (2) Liquid nitrogen diffusion cooling model

[0088] Based on the FDS liquid nitrogen diffusion behavior, the cooling effect is predicted by combining machine learning regression algorithm; random forest / XGBoost regression is used to establish a hierarchical input-output mapping.

[0089] Input: Injection port coordinates, flow rate, fire alarm level, ambient temperature, obstacle density;

[0090] Outputs: Cooling radius, minimum temperature, temperature drop rate.

[0091] The model is based on the heat conduction equation and the Euler-Lagrange method for the liquid nitrogen phase change heat source term. The objective function is to minimize the mean square error.

[0092] Key physical formulas include:

[0093] Heat conduction equation (used to simulate the evolution of the temperature field):

[0094]

[0095] in, It's temperature. It is the thermal diffusivity. It is a liquid nitrogen phase change heat source; the latent heat of vaporization of liquid nitrogen is... =199kJ / kg.

[0096] Liquid nitrogen phase change heat source term (Euler-Lagrange method):

[0097]

[0098] in, It is mass flow rate =808kg / m 3 (Liquid density). It is the droplet volume.

[0099] Random Forest / XGBoost Regression Model:

[0100]

[0101] The objective function is to minimize the mean squared error: .

[0102] The training data covers FDS samples of all levels in the database, enabling rapid prediction of Level I localized jetting, Level II total flooding, and Level III intensified asphyxiation.

[0103] (3) Oxygen concentration replacement model

[0104] The graded SVR algorithm is adopted, and the input parameters include: nitrogen injection rate (kg / s), initial oxygen concentration (21%), fire alarm level, cabin airtightness (leakage rate %), and ventilation volume (m³ / h).

[0105] Output: Time (s) for oxygen concentration to drop to the threshold and coverage of the asphyxiation area (%).

[0106] SVR algorithm:

[0107]

[0108] in, These are input features (such as nitrogen injection rate and initial oxygen concentration). It is the output oxygen concentration. and These are hyperparameters (optimized through training with a database).

[0109] The model is trained using gas diffusion data of various levels in the database and dynamically adapts to different levels of fire suppression requirements: Level I oxygen concentration threshold 15%, Level III 12%.

[0110] All models are assigned a confidence score: Level I ≥ 0.85, Level III ≥ 0.95, to ensure a strong correlation between the spraying strategy and the fire alarm level.

[0111] Step S104: Based on the fire spread prediction model, liquid nitrogen diffusion cooling model, and oxygen concentration replacement model, and combined with a multi-objective optimization algorithm, generate the optimal fire extinguishing scheme set and dynamically adjust it according to sensor data.

[0112] After the model training is completed, it is executed in three parts: scheme generation, dynamic adjustment, and strategy evolution, based on the current fire alarm level and real-time fire feature vectors (thermal runaway coordinates, temperature gradient, smoke concentration, and battery SOC state).

[0113] ①Solution generation stage

[0114] Input features into the three major models according to their levels:

[0115] Level I warning: Invoke the fire spread prediction model and output the local spread path;

[0116] Level II open flame: Superimposed liquid nitrogen diffusion cooling model, output liquid nitrogen coverage area;

[0117] Level III Explosion Risk: Full model activation, superimposed oxygen concentration replacement model, output asphyxiation inhibition efficiency.

[0118] The NSGA-III multi-objective optimization algorithm is adopted, and differentiated objective functions are set according to the levels:

[0119] Level I warning corresponds to a lower liquid nitrogen usage constraint and a shorter response time constraint, and adopts a local fixed-point spray mode; Level II open flame corresponds to a medium liquid nitrogen usage constraint and extinguishing time constraint, and adopts a total flooding or area coverage spray mode.

[0120] Level III explosion risk corresponds to enhanced explosion suppression constraints and target oxygen concentration constraints, requiring the activation of backup nozzles or increased injection intensity. Optimization variables include: nozzle combination sequence, single nozzle flow rate (5-20 L / min), injection duration (10-30 s), and nozzle start-up and shutdown sequence. The optimal solution set is obtained through Pareto front search, and the final scheme is selected according to the level weights: Level I prioritizes economy, while Level III prioritizes fire control speed and explosion suppression effect.

[0121] The multi-objective objective function aims to minimize liquid nitrogen consumption, minimize fire extinguishing time, and maximize oxygen suppression efficiency. It searches for Pareto optimal solutions based on non-dominated sorting and congestion distance, and incorporates constraints such as response delay, flow range, and duration range.

[0122] The NSGA-III multi-objective optimization algorithm is used to generate the optimal fire suppression plan. Its core formulas include:

[0123] Multi-objective function (minimize extinguishing time, liquid nitrogen consumption, etc.):

[0124]

[0125] in, These are decision variables (such as nitrogen injection port layout and flow rate). It refers to oxygen suppression efficiency.

[0126] Pareto optimality search (based on non-dominated sorting and crowding distance):

[0127]

[0128] Constraint handling (such as response latency limits):

[0129]

[0130] ② Dynamic adjustment phase

[0131] The system interacts in real time with a multi-source sensor network (infrared thermal imager, gas sensor, BMS unit), collecting data every 5 seconds and comparing it with the model predictions.

[0132] Determine the triggering correction condition:

[0133] 1) The actual rate of fire spread exceeded the predicted rate by 20%;

[0134] 2) Automatic upgrade of fire alarm level (e.g., from Level II to Level III);

[0135] 3) The oxygen concentration did not drop to the corresponding threshold level within 10 seconds.

[0136] Execution after triggering:

[0137] The lightweight FDS module can be used to quickly reconstruct the fire scene, with a calculation time of ≤2 seconds.

[0138] The NSGA-III algorithm was re-run according to the new level, and the injection parameters were re-optimized under the constraints of remaining liquid nitrogen reserves and allowable time.

[0139] Issue control commands to adjust the execution equipment: Level I: only activate local nozzles at a low flow rate of 5-10 L / min; Level III: add backup nozzles and increase the flow rate to 15-20 L / min.

[0140] ③ Strategy Evolution Stage

[0141] The system is equipped with an adaptive iteration module to record the scheme parameters (e.g., 3-5 spray points for Level I, ≥10 for Level III) and the actual fire extinguishing effect under each fire alarm level. Through incremental learning, the system continuously feeds back to the database, iteratively updates the internal parameters of each model, strengthens the shock wave suppression and rapid inertia logic for Level III explosion risk cases, and realizes fully automatic level response and iterative optimization of fire extinguishing strategies.

[0142] This invention discloses a lithium battery liquid nitrogen fire extinguishing scheme generation system based on fire alarm level. It consists of four parts connected from top to bottom: a data acquisition layer, a model analysis layer, a decision optimization layer, and a control execution layer. The system achieves intelligent generation and automatic execution of fire extinguishing schemes through multi-source data fusion and dynamic feedback mechanisms.

[0143] (1) Data Acquisition Layer

[0144] It integrates temperature sensors, infrared thermal imagers, smoke detectors, CO / H2 / VOCs gas concentration sensors, and a battery management system (BMS) to collect real-time data on voltage, current, and SOC. Simultaneously, it accesses the BIM model to extract the cabin geometry, including battery cluster layout, ventilation vent locations, and obstacle distribution, constructing a three-dimensional topology database. It is responsible for multi-source data acquisition, cleaning, hierarchical standardization, and feature vectorization, providing structured input for the upper-level model.

[0145] (2) Model Analysis Layer

[0146] Based on historical fire cases and real-time sensor data, a fire spread prediction model, a liquid nitrogen diffusion cooling model, and an oxygen concentration replacement model are trained and built-in. At the same time, a fire alarm level determination model is built-in. The input features include: temperature gradient ΔT≥10℃ / min, smoke density≥15%obs / m, combustible gas concentration H2≥1%LEL, and thermal radiation flux≥10kW / m². The output is Level I warning / Level II open flame / Level III explosion risk and confidence level, and the corresponding liquid nitrogen injection mode is automatically matched (Level I local injection, Level II total flooding injection, Level III enhanced asphyxiation injection).

[0147] (3) Decision optimization layer

[0148] Built-in NSGA-II / NSGA-III multi-objective optimization algorithm, with fire extinguishing time, liquid nitrogen consumption, and oxygen suppression efficiency as multiple objectives, optimizes nozzle position, flow rate, and duration; built-in dynamic feedback mechanism: if the actual fire spread rate exceeds the predicted value by 20%, the alarm level is automatically upgraded and the spraying scheme parameters are re-optimized.

[0149] (4) Control Execution Layer

[0150] It includes liquid nitrogen storage tanks, pipeline solenoid valves, wide-angle spray nozzles, and a 3D visualization terminal; it receives the optimal solution output from the decision optimization layer, controls the start and stop of solenoid valves and nozzles, and adjusts the flow rate; at the same time, it displays the liquid nitrogen coverage effect, temperature field distribution, oxygen concentration gradient, and the entire process of fire dissipation in real time through a 3D visualization interface, providing intuitive decision-making basis for operation and maintenance personnel.

[0151] 3. Specific implementation of the overall system workflow

[0152] 1) Initialization phase: Load the lithium battery compartment BIM model, label the battery clusters, ventilation openings, and sensor locations; pre-train the fire spread and liquid nitrogen diffusion model, and generate a basic fire extinguishing scheme library;

[0153] 2) Real-time response phase: Sensor data is input into the fire level determination model, triggering a Level II or higher alarm and immediately activating the scheme optimization module; FDS simulation of liquid nitrogen coverage effect is called, and Pareto optimal solution set is output in combination with NSGA-II / NSGA-III algorithm; solenoid valves are controlled to start and stop nozzles according to the optimized sequence, and infrared thermal imager monitors the fire dissipation in real time;

[0154] 3) Dynamic correction phase: The model prediction value is compared with the sensor measured data in real time. If the oxygen concentration does not drop to the threshold within 10 seconds, the backup nozzle is automatically opened. The data of each fire extinguishing process is transmitted back to the database to continuously iterate and optimize the model parameters.

[0155] 4. Implementation of technical effects

[0156] This invention achieves in-depth extraction of high-dimensional features of lithium battery fires through BIM structured modeling, FDS simulation, and multi-dimensional data fusion; relying on hierarchical machine learning models and multi-objective optimization, it breaks away from the traditional fixed threshold and manual experience judgment mode.

[0157] Actual implementation results:

[0158] The system response time is reduced by more than 40%, and the efficiency of fire extinguishing plan generation is improved by about 50%.

[0159] Reduce liquid nitrogen usage by 25%-30% to avoid over-spraying and causing low-temperature damage to equipment;

[0160] Adaptive robustness and anti-interference performance are improved by about 35%, making it suitable for different lithium battery types and energy storage compartment layouts;

[0161] It supports edge deployment, eliminating cloud computing latency and reducing overall deployment and maintenance costs by approximately 20%.

[0162] This will enable a shift from "passive response and firefighting" to "proactive, tiered prevention and control, precise suppression of explosions and prevention of reignition."

[0163] 5. Specific implementation of alternative technical solutions

[0164] 1) Model algorithm replacement: Hierarchical adaptive LSTM can be replaced with GRU or TCN time series models; Random Forest / XGBoost can be replaced with LightGBM or CatBoost; SVR can be replaced with BP neural network; NSGA series multi-objective optimization can be replaced with particle swarm optimization, ant colony optimization, or simulated annealing algorithm, while still achieving hierarchical scheme generation and parameter optimization.

[0165] 2) Hardware component replacement: Ordinary wide-angle nozzles can be replaced with omnidirectional adjustable nozzles and dot matrix fine mist nozzles; the sensing system can be equipped with flame image recognition and lidar temperature measurement modules to enrich the dimensions of input features.

[0166] 3) Simulation tool replacement: FDS can be replaced with fire fluid simulation software such as Pyrosim and FireFOAM to complete numerical simulation of level 3 fire scenarios and liquid nitrogen diffusion.

[0167] 4) Deployment architecture replacement: Edge local decision-making can be extended to edge + cloud collaboration, with real-time fire control at the edge and global model updates for multiple site cases gathered in the cloud, further improving generalization capabilities.

[0168] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions or improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for generating a lithium battery liquid nitrogen fire extinguishing scheme based on fire alarm level, characterized in that, Includes the following steps: S101. Multi-source heterogeneous data acquisition and preprocessing: Collect BIM structural data of lithium battery energy storage compartment, environmental sensor data inside the compartment, BMS operation data of battery management system and historical fire case data of lithium battery; classify and standardize heterogeneous data according to three fire alarm levels: Level I warning, Level II open flame and Level III explosion risk, and vectorize the discrete features. S102. Simulation Modeling and Dedicated Database Construction: Import the BIM structural model of the lithium battery energy storage compartment into the FDS fire dynamics simulation platform, and conduct differentiated fire and liquid nitrogen injection numerical simulations for Level I warning, Level II open flame, and Level III explosion risks, respectively, to generate a fire feature vector dataset with fire level labels. By integrating the fire feature vector dataset with the multi-source data preprocessed in step S101, a dedicated database for lithium battery liquid nitrogen fire extinguishing is constructed. S103, Multi-intelligent prediction model training: Based on the lithium battery liquid nitrogen fire extinguishing dedicated database, fire spread prediction model, liquid nitrogen diffusion cooling model, and oxygen concentration replacement model embedded with fire alarm level features are constructed respectively. S104. Generation and Dynamic Correction of Graded Fire Extinguishing Plans: Real-time acquisition of multi-source sensor data from the lithium battery energy storage compartment, integrating temperature gradient, smoke density, combustible gas concentration, and thermal radiation flux parameters to determine the current fire alarm level; inputting the fire level and real-time fire characteristics into three prediction models, employing a multi-objective optimization algorithm to generate a Pareto-optimal liquid nitrogen fire extinguishing spray plan based on fire level-differentiated constraints; real-time comparison of sensor-measured data with model prediction data, when the deviation between fire spread parameters and predicted values ​​exceeds a preset deviation threshold, or when the fire alarm level is upgraded, or the oxygen concentration fails to drop to the target oxygen concentration threshold for the corresponding fire alarm level within a preset response time, the current fire status is redefined and the liquid nitrogen spray parameters are updated, achieving dynamic adaptive adjustment of the fire extinguishing plan.

2. The method for generating a lithium battery liquid nitrogen fire extinguishing scheme based on fire alarm level according to claim 1, characterized in that: In step S101, the cabin environment sensing data includes one or more of the following: infrared thermal imager temperature field data, smoke detector smoke density data, CO / H2 / VOCs multi-gas concentration data, temperature and humidity data, and wind speed data. The BIM structural data includes: battery cluster spatial coordinates, ventilation opening locations, obstacle distribution, fire separation parameters, cable layout, and cabin airtightness parameters; the BMS operational data includes: battery voltage, current, SOC gradient, and coordinates of the initial location of thermal runaway.

3. The method for generating a lithium battery liquid nitrogen fire extinguishing scheme based on fire alarm level according to claim 1, characterized in that, The hierarchical standardization process described in step S101 specifically includes: Enhance the annotation of local temperature abrupt changes and early smoke diffusion characteristics in Level I warning data; Enhance the labeling of combustible / toxic gas concentrations, heat radiation flux, and open flame development characteristics for Class II open flame data; Enhance the labeling of pressure fluctuations, combustible gas enrichment, rapid flame propagation, and deflagration risk characteristics for Level III explosion risk data; employ unique thermal coding to vectorize the discrete labels of battery chemical system, cabin airtightness level, and fire alarm level.

4. The method for generating a lithium battery liquid nitrogen fire extinguishing scheme based on fire alarm level according to claim 1, characterized in that, Step S102, the FDS differential simulation settings include: configuring different heat release rates, flame propagation speeds and explosion overpressure peaks according to the fire level; setting corresponding heat release parameters, flame propagation parameters, liquid nitrogen injection parameters and mesh generation accuracy according to different fire alarm levels, and outputting spatiotemporal evolution data of temperature field, oxygen concentration field, liquid nitrogen coverage and temperature drop rate under different levels.

5. The method for generating a lithium battery liquid nitrogen fire extinguishing scheme based on fire alarm level according to claim 1, characterized in that, In step S103, the fire spread prediction model uses a hierarchical adaptive LSTM network. The input includes the fire alarm level encoding vector, the battery cluster BIM spatial topology, the initial thermal runaway coordinates, the ventilation intensity, and the battery SOC gradient. The output is a hierarchical fire spread path, propagation speed, and thermal distribution. The liquid nitrogen diffusion cooling model uses a random forest or XGBoost regression algorithm. The input includes the injection port coordinates, injection flow rate, fire alarm level, ambient temperature, and obstacle density. The output is the cooling radius, minimum temperature, and temperature drop rate. The oxygen concentration replacement model uses a hierarchical SVR algorithm. The input includes the nitrogen injection rate, initial oxygen concentration, fire alarm level, cabin leakage rate, and ventilation volume. The output is the time it takes for the oxygen concentration to drop to the safe threshold and the asphyxiation area coverage.

6. The method for generating a lithium battery liquid nitrogen fire extinguishing scheme based on fire alarm level according to claim 1, characterized in that, The fire alarm level determination model described in step S104 is based on one or more parameters among temperature change rate, smoke density, combustible gas concentration, and thermal radiation flux; based on the multi-parameter fusion results, it outputs Level I warning, Level II open flame, and Level III explosion risk and their corresponding confidence levels, with Level I confidence level ≥ 0.85 and Level III confidence level ≥ 0.

95.

7. The method for generating a lithium battery liquid nitrogen fire extinguishing scheme based on fire alarm level according to claim 1, characterized in that, In step S104, the multi-objective optimization algorithm takes minimizing liquid nitrogen consumption, minimizing fire extinguishing time, and maximizing oxygen suppression efficiency as its objective functions; the optimization variables include nozzle combination sequence, single nozzle spray flow rate, spray duration, spray pressure, and nozzle start-stop sequence.

8. The method for generating a lithium battery liquid nitrogen fire extinguishing scheme based on fire alarm level according to claim 1, characterized in that, In step S104, differentiated constraints and spraying modes are set according to the fire alarm level: Level I warning corresponds to lower liquid nitrogen usage constraints and shorter response time constraints, and adopts a local fixed-point spraying mode; Level II open flame corresponds to moderate liquid nitrogen usage and extinguishing time constraints, employing total flooding or area coverage spraying modes; Level III explosion risk corresponds to enhanced explosion suppression constraints and target oxygen concentration constraints, activating backup nozzles or increasing spray intensity.

9. The method for generating a lithium battery liquid nitrogen fire extinguishing scheme based on fire alarm level according to claim 1, characterized in that, The dynamic correction process in step S104 includes: collecting multi-source sensor data according to a preset sampling period and comparing it with the model prediction value; when the oxygen concentration fails to drop to the corresponding level threshold within the preset response time, the fire spread deviation exceeds the preset deviation threshold, or the fire level changes, the fire reconstruction module is called to redetermine the current fire status, and multi-objective optimization is re-executed to update the number of nozzles opened, the spray flow rate, and the spray duration.

10. A method for generating a lithium battery liquid nitrogen fire extinguishing scheme based on fire alarm level according to claim 1, characterized in that, It also includes a strategy self-evolution step: recording the parameters of the liquid nitrogen injection scheme and the actual fire extinguishing effect under each fire alarm level, using incremental learning to continuously iterate and optimize the internal parameters of the fire spread prediction model, the liquid nitrogen diffusion cooling model, and the oxygen concentration replacement model, and solidifying the optimal injection strategy for different cabin layouts and battery types.

11. A lithium battery liquid nitrogen fire extinguishing scheme generation system based on fire alarm level, characterized in that, The method according to any one of claims 1 to 10 includes: a data acquisition layer, a model analysis layer, a decision optimization layer, and a control execution layer with hierarchical communication connections; The data acquisition layer is used to collect BIM structural data of lithium battery energy storage compartment, multi-sensor environmental data, BMS battery operation data and historical fire case data, complete data cleaning, hierarchical standardization and feature vectorization, and construct a three-dimensional topology database. The model analysis layer includes a built-in fire dynamics simulation module, a fire alarm level determination model, a fire spread prediction model, a liquid nitrogen diffusion cooling model, and an oxygen concentration replacement model. The decision optimization layer includes a built-in NSGA-II / NSGA-III multi-objective optimization module, a dynamic feedback correction module, and a solution library storage module. The control execution layer includes a liquid nitrogen storage tank, an electromagnetic control valve, multiple sets of wide-angle spray nozzles, a three-dimensional visualization terminal, and an infrared thermal imaging monitoring unit. It is used to receive decision commands to control the start and stop of the nozzles and adjust the flow rate, and to display the liquid nitrogen coverage area, temperature field, and oxygen concentration gradient in real time using three-dimensional visualization.

12. A lithium battery liquid nitrogen fire extinguishing scheme generation system based on fire alarm level according to claim 11, characterized in that, The data acquisition layer includes an infrared thermal imager, a smoke detector, a CO / H2 / VOCs multi-gas sensor, a temperature, humidity and anemometer, a BIM modeling unit, a BMS data interaction unit, and a historical fire case database.

Images