Battery cluster level fire-fighting linkage control method and system based on BMS

Abstract

The invention discloses a battery cluster level fire-fighting linkage control method and system based on a BMS, and the method comprises the steps: carrying out the thermal runaway prediction through a multi-source feature fusion network according to the real-time voltage, temperature and gas concentration data of a battery cluster, and outputting a thermal runaway prediction level; on the basis of the thermal runaway prediction level, a dynamic priority mapping algorithm is utilized, and a control instruction set containing execution priority and spatial positioning is output according to the battery cluster topological structure; according to the control instruction set, a dual-channel redundant signal verification mechanism is adopted to execute a fire-fighting action, and a real-time action feedback signal is output; and on the basis of the real-time action feedback signal and the residual state data of the battery cluster, updating the parameter weight of the multi-source feature fusion network through a reinforcement learning model so as to be used for fire-fighting linkage control in the next period, and forming a closed-loop control link. According to the embodiment of the invention, battery cluster-level fire-fighting linkage control can be realized, so that the safety protection capability of a battery system is improved.

Description

Technical Field

[0001] The present invention belongs to the field of fire protection technology, and particularly relates to a battery cluster-level fire linkage control method and system based on BMS. Background Art

[0002] With the wide application of intelligent battery technology, especially in electric vehicles, power station energy storage, and large-scale energy storage systems, the battery cluster, as the core energy storage unit, its safety has increasingly become the focus of attention. During the operation of the battery cluster, thermal runaway may be triggered due to reasons such as overcharging, short circuit, and temperature rise, causing serious fire and explosion accidents, posing a huge threat to equipment safety and personnel life.

[0003] The traditional Battery Management System (BMS) is mainly responsible for monitoring the battery state, controlling charging and discharging, and performing basic protection. However, in the face of sudden thermal runaway events, the traditional system has deficiencies in reaction speed and linkage control ability, and it is difficult to achieve rapid and effective fire warning and linkage fire extinguishing measures. In addition, existing technologies mostly use a single information source for monitoring, lacking the ability of multi-source information fusion, and it is difficult to accurately judge the thermal runaway level and its diffusion trend. Summary of the Invention

[0004] The purpose of the present invention is to provide a battery cluster-level fire linkage control method and system based on BMS to solve the deficiencies in the prior art, and to be able to achieve battery cluster-level fire linkage control to enhance the safety protection ability of the battery system.

[0005] An embodiment of the present application provides a battery cluster-level fire linkage control method based on BMS, and the method includes: According to the real-time voltage, temperature, and gas concentration data of the battery cluster, perform thermal runaway prediction through a multi-source feature fusion network, and the multi-source feature fusion network fuses the SOC gradient change feature and the SOH attenuation curve feature to output the thermal runaway prediction level; Based on the thermal runaway prediction level, use the dynamic priority mapping algorithm to synergistically plan the aerosol spraying range and the thermal runaway isolation area according to the battery cluster topology, and output a control instruction set including execution priority and spatial positioning; According to the control instruction set, drive the aerosol spraying fire extinguishing device and the thermal runaway isolation device through a pulse trigger circuit powered by a redundant power supply, execute the fire protection action using a dual-channel redundant signal verification mechanism, and output a real-time action feedback signal; Based on the real-time action feedback signal and the battery cluster residual state data, update the parameter weights of the multi-source feature fusion network through a reinforcement learning model for the next cycle of fire linkage control, forming a closed-loop control link.

[0006] Optionally, based on the real-time voltage, temperature, and gas concentration data of the battery cluster, thermal runaway prediction is performed through a multi-source feature fusion network. The multi-source feature fusion network fuses the SOC gradient change feature and the SOH decay curve feature, and outputs a thermal runaway prediction level, including: Collect the time-series fluctuation data of the voltages of each cell in the battery cluster, the spatial distribution heat map of the temperature sensors, and the gas concentration change rate, perform timestamp synchronization alignment, and generate a cross-dimensional original monitoring matrix; Perform outlier correction on the original monitoring matrix, extract the gradient change feature of the SOC using the sliding window difference method, and at the same time fit the SOH decay curve by combining cyclic aging experiment data to generate a SOC-SOH joint degradation feature vector; Construct a multi-source feature fusion network, input the SOC-SOH joint degradation feature vector and the real-time temperature and gas concentration data into a bidirectional gated recurrent unit, dynamically allocate the weight coefficients of the voltage, temperature, and gas three channels through a feature-level attention mechanism, and output the fused thermal runaway feature code; Input the thermal runaway feature code into a pre-trained multi-level classifier, divide the warning level according to the threshold boundary of historical thermal runaway cases, generate a thermal runaway prediction level with confidence and mark the spatio-temporal position label.

[0007] Optionally, based on the thermal runaway prediction level, use a dynamic priority mapping algorithm to perform collaborative planning on the aerosol injection range and the thermal runaway isolation area according to the battery cluster topology structure, and output a control instruction set including execution priority and spatial positioning, including: Analyze the spatio-temporal label of the thermal runaway prediction level, construct an adjacency matrix according to the physical topology structure of the battery cluster, simulate the maximum energy diffusion range of the thermal runaway propagation path, and generate a potential risk propagation map; In the potential risk propagation map, with the battery cell with an urgent prediction level as the center, calculate the aerosol injection coverage radius and the safe isolation distance between adjacent clusters to generate the initial control area boundary; Use the dynamic priority mapping algorithm to perform weighted optimization on the initial control area in combination with the parallel string relationship and heat dissipation channel distribution of the battery cluster; Divide the concentric circle action scope of the aerosol injection according to the weighted optimization result, and mark the deployment coordinates of the thermal runaway isolation device to generate a three-dimensional space control instruction draft; Perform conflict detection on the three-dimensional space control instruction draft, eliminate the overlapping area of the device action range, and output the final control instruction set including execution priority and spatial positioning.

[0008] Optionally, according to the control instruction set, the aerosol jet fire extinguishing device and the thermal runaway isolation device are driven by a pulse trigger circuit powered by a redundant power source, and a dual-channel redundant signal verification mechanism is adopted to execute fire protection actions and output real-time action feedback signals, including: Activate the emergency power supply module of the BMS, use a capacitor energy storage type pulse generator to generate a high-voltage trigger signal, encode the control instruction set into two modulation waveforms with a 90° phase difference, and transmit them to the execution terminal through independent channels; Deploy a dual-channel redundant verification module at the aerosol jet device end to demodulate and cross-compare the two modulation waveforms, generate an execution permission signal when the instruction consistency exceeds the threshold, otherwise trigger the self-check loop; Drive the high-speed solenoid valve according to the execution permission signal, adjust the jet angle and dosage according to the spatial positioning parameters of the control instruction set, and synchronously start the telescopic diversion cover of the thermal runaway isolation device to form a physical barrier; Capture the diffusion form of the fire extinguishing agent through an infrared camera, combine the air flow intensity feedback by the pressure sensor, generate an action execution effect evaluation matrix as a real-time action feedback signal; Perform residual calculation on the real-time action feedback signal and the expected control target. If the deviation exceeds the safety threshold, trigger secondary injection, and at the same time update the fault code to the BMS log system.

[0009] Optionally, based on the real-time action feedback signal and the residual state data of the battery cluster, update the parameter weights of the multi-source feature fusion network through a reinforcement learning model for the next cycle of fire protection linkage control, forming a closed-loop control link, including: Integrate the real-time action feedback signal and the residual state data of the battery cluster to construct a multi-dimensional reinforcement learning state vector; Design a reward function based on policy gradient, where the positive reward includes the thermal runaway suppression efficiency and the resource consumption ratio, and the negative penalty includes the number of misoperations and the device loss value, generating a dynamic reward score; Use the asynchronous advantage actor-critic algorithm to update the parameters of the multi-source feature fusion network. Among them, the actor network optimizes the feature weight allocation strategy, and the critic network corrects the thermal runaway level prediction deviation; Encrypt and synchronize the updated network parameters to all BMS nodes, and reset the fire protection linkage control cycle counter to form a closed-loop control link from state perception to policy optimization.

[0010] Another embodiment of the present application provides a battery cluster-level fire protection linkage control system based on the BMS. The system includes: A prediction module, configured to perform thermal runaway prediction through a multi-source feature fusion network based on the real-time voltage, temperature, and gas concentration data of the battery cluster. The multi-source feature fusion network fuses the SOC gradient change feature and the SOH decay curve feature, and outputs a thermal runaway prediction level; A planning module, configured to perform collaborative planning on the aerosol spraying range and the thermal runaway isolation area based on the thermal runaway prediction level, using a dynamic priority mapping algorithm and according to the battery cluster topology structure, and output a control instruction set including execution priorities and spatial positioning; An execution module, configured to drive an aerosol spraying fire extinguishing device and a thermal runaway isolation device through a pulse trigger circuit powered by a redundant power source according to the control instruction set, perform fire fighting actions using a dual-channel redundant signal verification mechanism, and output a real-time action feedback signal; An update module, configured to update the parameter weights of the multi-source feature fusion network based on the real-time action feedback signal and the battery cluster residual state data through a reinforcement learning model for the next cycle of fire fighting linkage control, forming a closed-loop control link.

[0011] Another embodiment of the present application provides a storage medium, in which a computer program is stored. The computer program is configured to execute the method described in any one of the above when running.

[0012] Another embodiment of the present application provides an electronic device, including a memory and a processor. A computer program is stored in the memory, and the processor is configured to run the computer program to execute the method described in any one of the above.

[0013] Compared with the prior art, a battery cluster-level fire fighting linkage control method based on BMS provided by the present invention performs thermal runaway prediction through a multi-source feature fusion network based on the real-time voltage, temperature, and gas concentration data of the battery cluster, and outputs a thermal runaway prediction level; based on the thermal runaway prediction level, using a dynamic priority mapping algorithm and according to the battery cluster topology structure, outputs a control instruction set including execution priorities and spatial positioning; according to the control instruction set, performs fire fighting actions using a dual-channel redundant signal verification mechanism and outputs a real-time action feedback signal; based on the real-time action feedback signal and the battery cluster residual state data, updates the parameter weights of the multi-source feature fusion network through a reinforcement learning model for the next cycle of fire fighting linkage control, forming a closed-loop control link, thereby enabling battery cluster-level fire fighting linkage control to improve the safety protection ability of the battery system. Description of the Drawings

[0014] Figure 1 It is a hardware structure block diagram of a computer terminal of a battery cluster-level fire fighting linkage control method based on BMS provided by an embodiment of the present invention; Figure 2Schematic flowchart of a battery cluster-level fire linkage control method provided by an embodiment of the present invention; Figure 3 Schematic structural diagram of a battery cluster-level fire linkage control system based on BMS provided by an embodiment of the present invention. Detailed implementation manners

[0015] The embodiments described below by referring to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0016] An embodiment of the present invention first provides a battery cluster-level fire linkage control method based on BMS. This method can be applied to electronic devices, such as computer terminals, specifically, ordinary computers, etc.

[0017] The following takes running on a computer terminal as an example to describe it in detail. Figure 1 Hardware structure block diagram of a computer terminal of a battery cluster-level fire linkage control method based on BMS provided by an embodiment of the present invention. As Figure 1 shown, this computer device includes a processor, a memory, and a network interface connected through a system bus. Among them, the memory can include a non-volatile storage medium and an internal memory.

[0018] The non-volatile storage medium can store an operating system and a computer program. This computer program includes program instructions. When the program instructions are executed, the processor can execute any battery cluster-level fire linkage control method based on BMS.

[0019] The processor is used to provide computing and control capabilities to support the operation of the entire computer device.

[0020] The internal memory provides an environment for the operation of the computer program in the non-volatile storage medium. When the computer program is executed by the processor, the processor can execute any battery cluster-level fire linkage control method based on BMS.

[0021] This network interface is used for network communication, such as sending assigned tasks, etc. Those skilled in the art can understand that Figure 1 the structure shown in

[0022] It should be understood that the processor may be a Central Processing Unit (CPU), and the processor may also be other general-purpose processors, Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field-Programmable Gate Arrays (FPGAs) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0023] See Figure 2 , an embodiment of the present invention provides a battery cluster-level fire linkage control method based on a BMS, which may include the following steps: S201, according to the real-time voltage, temperature and gas concentration data of the battery cluster, perform thermal runaway prediction through a multi-source feature fusion network. The multi-source feature fusion network fuses the SOC gradient change feature and the SOH attenuation curve feature, and outputs a thermal runaway prediction level; specifically, it may include: S2011, collect the time-series fluctuation data of the voltages of each single cell in the battery cluster, the spatial distribution heat map of the temperature sensors and the gas concentration change rate, perform timestamp synchronization alignment, and generate a cross-dimensional original monitoring matrix; This step constructs a full-dimensional state monitoring basic data set for the battery cluster through high-precision data acquisition and multi-source heterogeneous data alignment technology, providing a standardized input for subsequent feature fusion.

[0024] ‌Multi-source data acquisition system deployment‌: Voltage time-series fluctuation data acquisition‌: The battery management unit of the BMS is configured with a 16-bit precision ADC (Analog-to-Digital Converter) to collect the voltages of each single cell in the battery cluster in real time at a sampling rate of 1 kHz. For example, the battery cluster of a certain energy storage power station contains 96 series-connected single cells (nominal voltage 3.2V), and each BMS node (such as TI BQ76952 chip) is responsible for monitoring 12 single cells, and the voltage data is uploaded to the main control unit through the CAN bus (baud rate 500 kbps).

[0025] Thermal Map Generation of Temperature Spatial Distribution: Eight DS18B20 digital temperature sensors (accuracy ±0.5°C) are evenly deployed in each module of the battery cluster in a 4×2 grid layout. The master control unit constructs a two-dimensional temperature field based on the sensor coordinates and generates a real-time thermal map with a resolution of 0.1°C through bilinear interpolation algorithm. For example, the temperature interpolation result in the central area (X = 2.5, Y = 1.5) of a certain module is 45.3°C.

[0026] Monitoring of Gas Concentration Change Rate: An NDIR (Non-Dispersive Infrared) gas sensor (such as Figaro TGS823) is installed at the top of the battery cluster to collect the concentrations of CO, H2, and VOCs every 5 seconds. The gas concentration change rate (ΔC / Δt) is calculated by the backward difference method. For example, if the CO concentration is 50 ppm at t = 10:00 and 65 ppm at t = 10:05, then ΔC / Δt = (65 - 50) / 300s = 0.05 ppm / s.

[0027] Timestamp Synchronization and Alignment Technology Global Clock Synchronization: The master control unit obtains the PPS (Pulses Per Second) signal through a GPS module (such as U-blox NEO-M8N) as the reference clock source (error < 1 μs). Each BMS and sensor node uses the NTPv4 (Network Time Protocol) protocol to synchronize the local clock to ensure that the timestamp alignment accuracy of the entire system is < 10 ms.

[0028] Cross-Dimensional Data Alignment: The original data streams of voltage (1 kHz), temperature (1 Hz), and gas (0.2 Hz) are interpolated to a unified time axis (such as 1-second granularity) according to timestamps. For example, the sampling point of voltage data at t = 10:00:00.500 seconds is linearly interpolated to match the timestamp of t = 10:00:00 of temperature data.

[0029] Construction of Original Monitoring Matrix: The matrix dimension is [time step × feature dimension]. For example, a 10-minute monitoring window (600 seconds) corresponds to 600 rows, and each row contains 96 single-cell voltages, 64 temperature points (8 sensors × 8 interpolation points), 3 gas concentrations and their change rates, totaling 164 columns. The data is stored in Float32 format, occupying approximately 393 KB of memory (600×164×4≈393KB).

[0030] Application Example: In a certain energy storage project for cascade utilization, local overheating occurred in battery cluster numbered B-07 (sensor T23 reported 48°C), and at the same time, the voltage of cell V33 dropped suddenly by 0.15V (from 3.05V to 2.90V). The main control unit aligned the voltage, temperature, and gas data at t = 14:23:05 seconds, and the eigenvalues corresponding to the corresponding row in the original monitoring matrix were: voltage column V33 = 2.90V, temperature column T23 = 48.0°C, CO concentration = 120 ppm (ΔC / Δt = 0.2 ppm / s), which was marked as a potential thermal runaway event.

[0031] S2012. Perform outlier correction on the original monitoring matrix, extract the gradient change characteristics of SOC using the sliding window difference method, and at the same time fit the SOH decay curve in combination with the cyclic aging experiment data to generate the SOC-SOH joint degradation feature vector; This step quantifies the dynamic coupling relationship between the battery health state and the state of charge through data cleaning and joint degradation feature modeling, providing key degradation indicators for thermal runaway prediction.

[0032] Outlier Correction Strategy: Voltage Anomaly Detection: Remove voltage outliers based on the 3σ principle (three times the standard deviation). For example, the normal fluctuation range of the voltage of a certain cell is 2.8V - 3.6V (μ = 3.2V, σ = 0.1V). If a sampling value is 2.5V (exceeding μ - 3σ = 2.9V), it is replaced with the median value of the 10-second sliding window (such as 3.15V).

[0033] Temperature Spatial Smoothing: Perform median filtering on the mutation points (adjacent temperature difference > 5°C) in the heat map. For example, the temperature of a certain point is 50°C, and the temperatures of its surrounding 8-neighborhood are [45, 46, 48, 47, 52, 49, 46, 44], and the median value is 47°C, then the temperature of this point is corrected to 47°C.

[0034] Gas Data Compensation: For the response delay of gas sensors (such as a 20-second lag for CO sensors), use the ARIMA (Autoregressive Integrated Moving Average) model to predict the real-time value. For example, the actual reading at time t is 80 ppm, and the model prediction value is 95 ppm, then the prediction value is used as the standard.

[0035] SOC Gradient Feature Extraction: Sliding window difference method: Taking 30 seconds as the window (30 voltage sampling points), calculate the SOC change rate ΔSOC / Δt. The SOC estimation adopts the ampere-hour integration method: SOC(t)=SOC(t0)+∑(I×Δt) / Capacity, where I is the current (from the Hall sensor, with an accuracy of ±1A), and Capacity is the rated capacity (such as 280Ah). For example, if the average current within the window is -50A (discharging), then ΔSOC = (-50A×30s) / 280Ah≈-0.0446 (i.e., the SOC drops by 4.46%).

[0036] Gradient change feature encoding: Input the ΔSOC sequence (such as 20 ΔSOC values within a 600-second window) into a one-dimensional convolutional layer (convolution kernel size = 5, stride = 5) to extract local gradient patterns. For example, the convolutional output features are [0.12, -0.08, 0.05], indicating the trend that the SOC first drops rapidly and then rebounds.

[0037] SOH decay curve fitting: Cyclic aging experiment data: Through laboratory accelerated aging tests (such as 1000 charge-discharge cycles at 0.5C), record the capacity decay curve. Adopt cubic polynomial fitting: SOH(n)=a×n³ + b×n² + c×n + d, where n is the number of cycles. For example, for a certain battery, when n = 500, SOH = 92%, and the fitting parameters are a = -1.2e-7, b = 0.0003, c = -0.021, d = 100.

[0038] Real-time SOH estimation: Calculate the theoretical SOH based on the current number of cycles (from the BMS log) and the fitting curve, and then compare it with the actual capacity (calibrated through full charge and discharge) to generate a residual compensation term. For example, if the theoretical SOH = 88% and the measured capacity is 85% of the rated capacity, then the compensation term is -3%.

[0039] Joint feature vector generation: Concatenate the SOC gradient feature (3D) and the SOH decay feature (4D: theoretical SOH, compensation value, number of cycles, capacity residual) into a 7D vector, and normalize it as the input. For example, the vector values are [0.12, -0.08, 0.05, 0.85, -0.03, 620, -0.02].

[0040] Application Example: In a scenario of second-life utilization of a power battery, the SOC of battery cluster C-12 drops suddenly from 65% to 58% within 30 seconds (ΔSOC = -7%), exceeding the normal decay rate (expected ΔSOC ≈ -2%). The sliding window difference method detects an abnormal gradient pattern (convolutional features = [-0.25, 0.1, 0.05]). Combining with the SOH fitting results (theoretical SOH = 76%, measured = 70%), a joint feature vector [-0.25, 0.1, 0.05, 0.76, -0.06, 800, -0.07] is generated, triggering a thermal runaway warning.

[0041] S2013. Construct a multi-source feature fusion network. Input the SOC-SOH joint degradation feature vector, real-time temperature, and gas concentration data into a bidirectional gated recurrent unit. Dynamically allocate the weight coefficients of the voltage, temperature, and gas three channels through a feature-level attention mechanism, and output the fused thermal runaway feature encoding. This step realizes the adaptive fusion of multi-source heterogeneous features through a deep learning model, captures the spatio-temporal propagation pattern of thermal runaway, and improves the warning accuracy.

[0042] Multi-source Feature Fusion Network Architecture: Input Layer Design: The network accepts three types of inputs: Degradation Features: A 7-dimensional SOC-SOH joint vector (see the previous step), which is mapped to 16 dimensions by a fully connected layer; Temperature Data: A 64-dimensional heat map is flattened into a vector and compressed to 16 dimensions by a CNN (convolution kernel 3×3, stride 2); Gas Data: Three gas concentrations and change rates (6 dimensions), which are mapped to 8 dimensions by a fully connected layer.

[0043] Bidirectional GRU (Gated Recurrent Unit): Each channel (degradation, temperature, gas) independently inputs a bidirectional GRU layer (number of hidden units = 32) to capture the temporal dependence relationship. For example, the output dimension of the degradation channel GRU is 64 (32 for each direction), the temperature channel outputs 64, and the gas channel outputs 32.

[0044] Attention Mechanism: Feature-level Attention: Generate weight coefficients for each channel. Calculation method: Weight = Softmax(MLP(GRU output)), where MLP (Multi-Layer Perceptron) contains 16 hidden units. For example, the weight of the degradation channel = 0.6, temperature = 0.3, gas = 0.1.

[0045] Spatio-temporal Attention: Within the temperature channel, allocate weights to the spatial positions of the heat map. For example, the weight of the central region = 0.8, the edge = 0.2.

[0046] Generation of Thermal Runaway Feature Encoding: Weighted Fusion: The outputs of each channel GRU are multiplied by their weights and then concatenated. For example, for degradation, 64 × 0.6 = 38.4 dimensions, for temperature, 64 × 0.3 = 19.2 dimensions, for gas, 32 × 0.1 = 3.2 dimensions, and the total dimension is 60.8 → rounded up to 61 dimensions.

[0047] Feature Compression: Generate high-dimensional encodings through a fully connected layer (61 → 256 dimensions) and use the ReLU activation function. For example, at a certain moment, the encoding values are [0.8, -0.2, 1.5, ..., 0.3] (256 dimensions).

[0048] Normalization: Use Layer Normalization to standardize the encodings to prevent gradient explosion.

[0049] Application Example: In a thermal runaway event, the temperature channel detects that the temperature at the center of the module rises to 60°C (weight = 0.8), the CO concentration in the gas channel reaches 200 ppm (weight = 0.15), and the SOC gradient shows a sharp drop (weight = 0.7). After feature fusion, the spatio-temporal attention weight in the high-temperature area is increased to 0.9, and finally, the thermal runaway feature encoding highlights the collaborative anomaly between temperature and SOC.

[0050] In S2014, input the thermal runaway feature encoding into a pre-trained multi-level classifier, divide the warning level according to the threshold boundary of historical thermal runaway cases, generate a thermal runaway prediction level with confidence and mark the spatio-temporal position label.

[0051] In this step, through classification decision-making and confidence evaluation, map the abstract features into actionable warning signals and locate the source of risk.

[0052] Multi-level Classifier Design: Hierarchical Division: The warning level is divided into 3 levels: Level 1 (Low Risk): The overlap degree between the feature encoding and the normal working condition > 70%, confidence < 60%; Level 2 (Medium Risk): The encoding deviates from the normal but does not reach the thermal runaway threshold, confidence 60% - 85%; Level 3 (High Risk): The encoding exceeds the threshold of historical cases, confidence > 85%.

[0053] Classifier Structure: Use a cascade of random forest (200 trees) and support vector machine (SVM): The First Level: The random forest screens suspected cases (recall rate > 95%); The Second Level: SVM (kernel function = RBF, C = 1.0) performs precise classification and outputs Level 1 - 3.

[0054] Confidence calculation: Based on the classification probability (the proportion of random forest votes) and the distance from the feature encoding to the decision boundary (SVM Margin). For example, for a certain sample, the random forest vote rate = 92% (Level 3), SVM Margin = 1.5 (> threshold 1.2), then the final confidence = 90%.

[0055] Space-time position marking: Risk source location: Based on the area with the highest weight in the temperature heat map (such as X = 3, Y = 2) and the position of the SOC abnormal monomer (such as V45), mark the risk source coordinates as three-dimensional space (cluster number + module row and column number + monomer index).

[0056] Propagation range prediction: Based on the diffusion speed of historical cases (such as 0.5 m / s), estimate the module range that may be affected within the next 5 minutes. For example, if the current risk source is located in module M5, it is predicted that M4, M6, and M7 will be affected.

[0057] Application example: In a certain energy storage power station, the central temperature of module M3 in battery cluster D-09 reaches 58°C. The SOC gradient feature encoding triggers the random forest Level 3 classification (vote rate 88%), and SVM determines it as Level 3 (Margin = 1.8), with a confidence of 91%. The system marks the risk source as module M3 (coordinates X = 2, Y = 1) and warns that adjacent modules M2 and M4 may be affected within 3 minutes.

[0058] This step collects the voltage, temperature, and gas concentration data of the battery cluster in real time, combines the gradient change characteristics of SOC (state of charge, reflecting abnormal battery charge and discharge) and the decay curve characteristics of SOH (state of health, characterizing battery aging degree), and uses a multi-source feature fusion network for data fusion. The network uses a bidirectional gated recurrent unit (BiGRU) to extract temporal correlations, and dynamically assigns weights to different data channels through an attention mechanism. Finally, it outputs the thermal runaway prediction level (such as low risk, medium risk, high risk) through a multi-level classifier, and marks specific space-time position labels, solving the problem of misjudgment or missed judgment caused by traditional thermal runaway early warning relying on a single parameter (such as temperature). By fusing multi-dimensional data, the prediction accuracy is improved, the local or overall thermal runaway risk of the battery cluster is identified in advance, providing a decision basis for subsequent precise fire linkage.

[0059] S202, based on the thermal runaway prediction level, use the dynamic priority mapping algorithm to synergistically plan the aerosol spraying range and the thermal runaway isolation area according to the battery cluster topology, and output a control instruction set including execution priority and spatial positioning; specifically, it may include: In 2021, analyze the spatio-temporal tags of the thermal runaway prediction level, construct an adjacency matrix based on the physical topology of the battery cluster, simulate the maximum energy diffusion range of the thermal runaway propagation path, and generate a potential risk propagation map; In this step, by analyzing the spatio-temporal attributes of the thermal runaway prediction results and combining the physical connection relationships of the battery cluster, a mathematical model of thermal runaway propagation is constructed to accurately predict the risk diffusion range, providing a spatial decision-making basis for subsequent fire extinguishing and isolation.

[0060] Spatio-temporal tag parsing technology: Spatio-temporal tag structure: Extract spatio-temporal tags from the thermal runaway prediction level output in the above step. The tag contains four parts of information: Timestamp: The moment when the thermal runaway prediction is triggered (e.g., 2024-07-15 14:23:05.500); Spatial coordinates: The three-dimensional position encoding of the risk source, such as battery cluster number C-07, module number M3 (row number 2, column number 3), cell index V45; Risk level: Level 1 / 2 / 3 (e.g., Level 3); Confidence level: The percentage of the credibility of the prediction result (e.g., 91%).

[0061] Tag parsing process: The main control unit matches the tag fields through regular expressions. For example, by parsing the string "C-07_M3(2,3)_V45_L3_91%", the cluster number C-07, module M3 coordinates (X = 2, Y = 3), cell V45, risk level Level 3, and confidence level 91% are extracted.

[0062] Battery cluster topology modeling: Definition of physical topology structure: The battery cluster is composed of multiple modules connected in parallel or in series, and each module contains several single cells. For example, a certain energy storage battery cluster is composed of 6 modules connected in parallel (16 single cells in series in each group), with a total voltage of 16 × 3.2V = 51.2V and a capacity of 280Ah × 6 = 1680Ah.

[0063] Construction of adjacency matrix: According to the electrical connection relationship between modules (such as parallel strings) and physical spacing (such as module interval of 0.5 meters), an undirected graph adjacency matrix is constructed. The matrix element A[i][j] represents the connection strength between module i and module j. If directly connected in parallel, A[i][j] = 1; if the interval exceeds 1 meter, A[i][j] = 0.2. For example, module M3 is adjacent to and connected in parallel with M2 and M4, so A[3][2] = A[3][4] = 1; it is 1.2 meters away from M1, so A[3][1] = 0.2.

[0064] Thermal runaway propagation simulation algorithm: Energy Diffusion Model: Assume that the total energy Q released during thermal runaway is determined by the cell capacity (such as 3.2V / 280Ah) and SOC. The calculation formula is Q = 3.2V × 280Ah × SOC × 3600s (unit: joules). For example, if the SOC of a certain cell is 80%, then Q = 3.2 × 280 × 0.8 × 3600 ≈ 2.3×10^6 J.

[0065] Propagation Path Simulation: The improved Dijkstra algorithm is used. Starting from the risk source cell, the energy propagation path is calculated along the weights of the adjacency matrix (electrical connection strength + heat dissipation condition). For example, when thermal runaway occurs in module M3, the energy diffuses to M2 and M4 through the parallel copper busbar (weight 0.9), and at the same time radiates to the adjacent module M5 through the air (weight 0.1).

[0066] Risk Propagation Map Generation: The propagation path is superimposed on the 3D model of the battery cluster, and the risk intensity is marked with a color gradient. For example, the red area (energy > 1×10^6 J) covers modules M3, M2, and M4, and the orange area (energy > 5×10^5 J) spreads to M5.

[0067] Application Example: In a certain energy storage power station, the cell V45 of module M3 in battery cluster C-07 triggers a Level 3 warning (confidence level 91%). After parsing the spatio-temporal tags, the main control unit loads the topological data of C-07: 6 parallel modules with a module spacing of 0.5 meters. Through simulation using the Dijkstra algorithm, it is found that the thermal runaway energy spreads to M2 and M4 within 5 seconds and affects M5 within 10 seconds. The generated risk map shows that the energy in the core area of M3 reaches 2.3×10^6 J, that of M2 / M4 is 1.8×10^6 J, and that of M5 is 0.9×10^6 J.

[0068] In S2022, in the potential risk propagation map, with the battery cell with a predicted level of emergency as the center, calculate the aerosol spray coverage radius and the safety isolation distance between adjacent clusters to generate the boundary of the initial control area; This step is based on the quantitative data of the risk propagation map and combines the performance parameters of the fire extinguishing device to dynamically delimit the initial action range of aerosol spraying and isolation, ensuring rapid suppression of thermal runaway and prevention of cross-cluster spread.

[0069] Aerosol Spray Parameter Calculation: Coverage Radius Modeling: The relationship between the coverage radius R (in meters) of an aerosol injection device (such as the HATS-200 model), the injection pressure P (in kPa), and the injection angle θ (in degrees) is R = 0.12 × P × sinθ. For example, when P = 500 kPa and θ = 60°, R = 0.12 × 500 × sin60° ≈ 0.12 × 500 × 0.866 ≈ 52 meters (the actual value needs to be scaled according to the battery cluster size, for example, the actual effective radius is 5.2 meters).

[0070] Dose Requirement Estimation: The amount of fire extinguishing agent D (in grams) is positively correlated with the risk level, and the formula is D = Level × BaseDose, where Level 3 corresponds to BaseDose = 200 g. For example, when there is a Level 3 warning, D = 3 × 200 = 600 g.

[0071] Dynamic Radius Adjustment: If multiple risk sources are adjacent (spacing < 2R), then the coverage areas are merged. For example, the distance between the risk sources of module M3 and M2 is 4 meters (R = 5 meters), and after merging, the coverage radius expands to 7 meters.

[0072] Calculation of Safety Isolation Distance: Thermal Runaway Propagation Speed: According to historical data, the thermal runaway propagation speed in air is V_spread = 0.5 m / s, and in conductors (such as copper bars) it is V_spread = 2 m / s.

[0073] Safety Distance Formula: The safety isolation distance L = Max(V_spread × t_response, L_min), where t_response is the system response time (such as 5 seconds), and L_min is the physical isolation margin of the equipment (such as 0.5 meters). For example, in the air propagation direction, L = 0.5 × 5 + 0.5 = 3.0 meters, and in the copper bar direction, L = 2 × 5 + 0.5 = 10.5 meters.

[0074] Cross-Cluster Isolation Strategy: If the distance between adjacent battery clusters is less than L, then isolation devices need to be deployed between the clusters. For example, the distance between cluster C-07 and C-08 is 8 meters (less than 10.5 meters in the copper bar direction), and a diversion cover isolation needs to be activated between the clusters.

[0075] Generation of Initial Control Area: Geometric Boundary Definition: Taking the coordinates of a single risk source (such as M3(X = 2, Y = 3)) as the center, draw an aerosol coverage circle (radius 5.2 meters) and a safety isolation rectangle (length = 2L, width = 2L).

[0076] 3D Space Mapping: Map the two-dimensional geometric boundary to the three-dimensional coordinate system (X, Y, Z) of the battery cluster, where the Z-axis represents the module layer height (e.g., 0.2 meters per layer). For example, for module M3, Z = 1.2 meters (6th layer), and its control area is a cylinder (radius 5.2 meters, height 0.2 meters).

[0077] Boundary Data Encapsulation: The initial control area is stored in JSON format, containing fields such as center coordinates, radius, isolation distance, timestamp, etc. For example: { "cluster": "C-07", "center": {"X": 2, "Y": 3, "Z": 1.2}, "radius": 5.2, "isolation_distance": {"air": 3.0, "conductor": 10.5}, "timestamp": "2024-07-15 14:23:05.500" }

[0078] Application Example: For the thermal runaway event of module M3 in cluster C-07, the aerosol injection radius is set to 5.2 meters (covering M3, M2, M4), and the safety isolation distance is set to 3 meters in the air direction and 10.5 meters in the copper busbar direction. The initial control area includes a cylinder (radius 5.2 meters, Z = 1.2 meters) and a cuboid isolation area (length 21 meters, width 6 meters, covering the channel between clusters C-07 and C-08).

[0079] In S2023, use the dynamic priority mapping algorithm, combine the parallel string relationship of the battery cluster and the heat dissipation channel distribution, and perform weighted optimization on the initial control area; In this step, through the dynamic priority algorithm, comprehensively consider the electrical topology and thermodynamic characteristics, optimize the allocation strategy of fire extinguishing and isolation resources, improve the fire protection efficiency and reduce the risk of misoperation.

[0080] Parallel String Relationship Modeling: String Weight Allocation: The electrical influence weight of parallel modules W_electric = 1 / (1 + Distance × R_parallel), where Distance is the module spacing (meters) and R_parallel is the parallel resistance (e.g., 0.01 Ω). For example, the distance between module M3 and M2 is 0.5 meters, then W_electric = 1 / (1 + 0.5 × 0.01) = 0.995.

[0081] Fault propagation probability: According to Ohm's law, the fault current \(I_{fault}\) is inversely proportional to the parallel resistance, and the propagation probability \(P_{propagate}=I_{fault} / I_{max}\). For example, if \(I_{max} = 100A\) and \(I_{fault}=80A\), then \(P_{propagate}=0.8\).

[0082] Heat dissipation channel evaluation: Heat dissipation coefficient calculation: The heat dissipation capacity \(H\) (W / ℃) is related to the fan wind speed \(V\) (m / s) and the heat sink area \(A\) (m²) by \(H = 15×V×A^{0.5}\). For example, when the wind speed is \(2m / s\) and the area is \(0.5m²\), then \(H = 15×2×0.707≈21.2W / ℃\).

[0083] Thermal resistance influence factor: Define the thermal resistance factor \(K_{thermal}=1 / (1 + H×\Delta t)\), where \(\Delta t\) is the duration of thermal runaway (seconds). For example, when \(H = 21.2W / ℃\) and \(\Delta t = 10\) seconds, then \(K_{thermal}=1 / (1 + 21.2×10)=0.0045\).

[0084] Dynamic priority mapping algorithm: Weighted optimization objective: The priority Score=\(\alpha×W_{electric}+\beta×P_{propagate}+\gamma×(1 - K_{thermal})\), where \(\alpha\), \(\beta\), \(\gamma\) are weight coefficients (by default, \(\alpha = 0.6\), \(\beta = 0.3\), \(\gamma = 0.1\)). For example, for a certain module, \(W_{electric}=0.995\), \(P_{propagate}=0.8\), \(K_{thermal}=0.0045\), then Score=\(0.6×0.995 + 0.3×0.8+0.1×(1 - 0.0045)=0.597+0.24+0.0995≈0.936\).

[0085] Real-time adjustment mechanism: If the temperature rise rate of a certain module>\(5℃ / s\) (detected by the heat map), then automatically increase its \(\gamma\) weight to \(0.3\) and decrease \(\alpha\) to \(0.4\). For example, Score=\(0.4×0.995 + 0.3×0.8 + 0.3×0.995≈0.398+0.24+0.299≈0.937\).

[0086] Optimization result output: Sort by Score from high to low, and give priority to performing fire protection actions in the high Score area. For example, for module M3, Score = 0.936 (the highest), for M2, Score = 0.892, for M4, Score = 0.885, and for M5, Score = 0.752.

[0087] Application Example: During the optimization of cluster C-07, module M3 was listed as the highest priority with Score = 0.936 because it was directly connected to the fault source and had poor heat dissipation (H = 15 W / ℃); module M5 was adjusted to secondary processing with Score = 0.752 because it was at a relatively far distance and had good heat dissipation (H = 30 W / ℃). The algorithm changed the evenly distributed aerosol injection dose to M3: 400 g, M2: 300 g, M4: 300 g, M5: 100 g.

[0088] In S2024, divide the concentric circle action scope of aerosol injection according to the weighted optimization result, mark the deployment coordinates of the thermal runaway isolation device, and generate a draft three-dimensional space control instruction; This step maps the optimized priority to an executable space control instruction, and through hierarchical injection and precise positioning, realizes the coordinated operation of fire extinguishing and isolation.

[0089] Division of Concentric Circle Action Scope: Hierarchy Definition: Divide the control area into three layers according to the priority Score: Core Layer (Score ≥ 0.9): Radius 3 meters, injection dose ratio 60% (e.g., M3: 400 g); Middle Layer (0.7 ≤ Score < 0.9): Radius 5 meters, dose ratio 30% (e.g., M2: 300 g); Peripheral Layer (Score < 0.7): Radius 7 meters, dose ratio 10% (e.g., M5: 100 g).

[0090] Dose Gradient Allocation: Allocate the dose proportionally according to the module Score within the same layer. For example, in the middle layer, M2 Score = 0.892, M4 Score = 0.885, the total dose is 600 g × 30% = 180 g, M2 dose = 0.892 / (0.892 + 0.885) × 180 ≈ 90 g, M4 ≈ 90 g.

[0091] Isolation Device Deployment Strategy: Coordinate Calculation: The deployment position of the isolation device (such as a telescopic deflector) needs to meet two conditions: Located at the boundary of the inter-cluster safety isolation distance (such as 3 meters in the air direction); avoid heat dissipation channels and maintenance entrances.

[0092] Three-Dimensional Coordinate Marking: For example, between clusters C-07 and C-08, the four anchor point coordinates for deploying the deflector are (10, 3, 1.2), (10, 5, 1.2), (12, 5, 1.2), (12, 3, 1.2), forming a rectangular isolation area.

[0093] Generation of Draft Control Instruction: Instruction Structure: Each instruction contains the device type (aerosol jet / isolation device), spatial coordinates, action parameters, and execution priority (1 - 100). For example: { "device": "Aerosol Jet #23", "type": "Fire Extinguishing", "coordinates": {"X": 2, "Y": 3, "Z": 1.2}, "radius": 3.0, "dose": 400, "priority": 95 }, { "device": "Deflector #07", "type": "Isolation", "coordinates": {"X": 10, "Y": 3, "Z": 1.2}, {"X": 10, "Y": 5, "Z": 1.2}, {"X": 12, "Y": 5, "Z": 1.2}, {"X": 12, "Y": 3, "Z": 1.2} , "action": "extend", "priority": 90 }。

[0094] Time Synchronization Marker: All instructions are appended with a unified timestamp (e.g., 14:23:06.200) to ensure coordinated execution.

[0095] Application Example: In the draft control instructions generated by Cluster C - 07, the aerosol jet #23 sprays 400g of fire extinguishing agent at coordinates (2, 3, 1.2) (priority 95), and the deflector #07 expands at coordinates (10, 3, 1.2) - (12, 5, 1.2) (priority 90), covering the safety isolation area in the direction of the busbar.

[0096] In S2025, conflict detection is performed on the draft three - dimensional space control instructions, the overlapping areas of device action ranges are eliminated, and the final control instruction set containing execution priority and spatial positioning is output.

[0097] This step ensures that the actions of the fire extinguishing and isolation devices do not interfere with each other through spatial conflict detection and priority arbitration, and optimizes the execution order to maximize fire fighting efficiency.

[0098] ‌Conflict detection algorithm‌: ‌Spatial hash grid division‌: Divide the three-dimensional space of the battery cluster into cubic grids of 0.5 m × 0.5 m × 0.2 m, and record the device action status (idle / occupied) for each grid.

[0099] ‌Action area projection‌: Project the aerosol spraying range (cylinder) and the isolation device range (cuboid) onto the grid. For example, the cylinder of aerosol sprayer #23 covers grid G(2-5, 3-8, 1.2), and the cuboid of deflector #07 covers grid G(10-12, 3-5, 1.2).

[0100] ‌Overlap area detection‌: If there is an intersection in the action grids of two devices, it is determined as a conflict. For example, assume that the coverage grid G(10-13, 3-5, 1.2) of sprayer #24 overlaps with the grid G(10-12, 3-5, 1.2) of deflector #07, then a conflict is triggered.

[0101] ‌Conflict resolution strategy‌: ‌Priority arbitration‌: Devices with high priority are executed first, and devices with low priority adjust parameters (such as reducing the radius, delaying execution). For example, when aerosol sprayer #24 (priority 85) conflicts with deflector #07 (priority 90), the deflector is executed first, and the radius of sprayer #24 is reduced from 5 m to 4 m.

[0102] ‌Spatial parameter optimization‌: Use the gradient descent method to iteratively adjust the device coordinates or range until the overlap is eliminated. For example, adjust the X coordinate of deflector #07 from 10-12 to 10-11.5 to avoid the coverage area of sprayer #24.

[0103] ‌Final instruction set generation‌: ‌Execution sequence arrangement‌: Sort in descending order of priority and add time delays (such as high-priority devices execute immediately, and low-priority devices are delayed by 0.5 seconds). For example: {"device": "Aerosol sprayer #23", "delay_ms": 0}, {"device": "Deflector #07", "delay_ms": 500}, {"device": "Aerosol sprayer #24", "delay_ms": 1000} 。

[0104] ​Logging and Rollback Mechanism: If the conflict cannot be completely eliminated (such as physical space limitations), record the fault code (such as ERR_Conflict_Unresolved) and trigger manual intervention.

[0105] Based on the high-risk battery positions marked according to the prediction level, combined with the physical topology of the battery cluster (such as parallel string relationships, heat dissipation channel distributions), the dynamic priority mapping algorithm simulates the thermal runaway propagation path, calculates the aerosol injection range (coverage radius) and isolation area (safety distance), and performs spatial optimization based on the battery cluster layout. The algorithm generates three-dimensional control instructions, including injection angle, dosage, and deployment coordinates of isolation devices, through weighted optimization (such as preferentially protecting the core strings and avoiding heat dissipation blind spots), while solving the problem of equipment action conflicts, avoiding resource waste or response delays caused by the "one-size-fits-all" traditional fire protection strategy, and precisely suppressing the spread of thermal runaway through dynamic priority adjustment and spatial collaborative planning, maximizing the fire extinguishing efficiency and reducing the impact on normal batteries.

[0106] S203, According to the control instruction set, drive the aerosol fire extinguishing device and the thermal runaway isolation device through a pulse trigger circuit powered by the redundant power supply, and execute the fire protection action using a dual-channel redundant signal verification mechanism, and output a real-time action feedback signal; specifically, it can include: S2031, Activate the emergency power supply module of the BMS, use a capacitor energy storage type pulse generator to generate a high-voltage trigger signal, encode the control instruction set into two modulation waveforms with a 90° phase difference, and transmit them to the execution terminal through independent channels; This step realizes emergency power supply through capacitor energy storage technology, uses phase difference modulation to ensure the reliability and anti-interference of signal transmission, and provides high-precision trigger instructions for the fire protection execution terminal.

[0107] Emergency Power Supply Module Activation Logic: The BMS is built-in with a super capacitor bank (parameters: 6 capacitors of 2.7V / 100F in series, total capacity 16.2V / 16.7F). When the main power supply fails, the capacitor bank discharges through the MOSFET switch (model IRF3205), outputs a voltage of 12V - 16.2V, and the continuous power supply time ≥ 30 seconds.

[0108] Pulse Generator Design: The capacitor energy is converted into high-voltage pulses through a boost circuit (topology: Boost Converter). The key parameters include a pulse width of 1ms, a peak voltage of 300V, and a repetition frequency of 10Hz. For example, triggering the aerosol injection device requires 3 consecutive pulses (pulse width 1ms, interval 50ms).

[0109] Control Instruction Encoding and Modulation: Two-channel Phase Difference Modulation: Convert the control instruction set (in JSON format) into a binary data stream, and adopt QPSK (Quadrature Phase Shift Keying) modulation. The phase difference between the two carriers is 90° (0° for the I channel and 90° for the Q channel). For example, the instruction "spray dose 400g" corresponds to the binary "010001101101", which generates an I-channel waveform (frequency 10kHz, amplitude 5V) and a Q-channel waveform (frequency 10kHz, amplitude 5V, phase delay 1 / 4 cycle) after modulation.

[0110] Independent Channel Transmission: The I-channel signal passes through the CAN bus (twisted pair at the physical layer, impedance 120Ω), and the Q-channel passes through the RS-485 bus (differential signal, shielded twisted pair) to avoid signal loss caused by electromagnetic interference.

[0111] Receiving Signals at the Execution Terminal Signal Demodulation Circuit: Deploy an AD8302 phase detection chip at the execution terminal to perform coherent demodulation on the two signals and restore the original binary instructions. For example, when the I-channel signal is 5V@0° and the Q-channel signal is 5V@90°, AD8302 outputs the logic level "11", corresponding to the highest priority of the spray instruction.

[0112] Application Example: When a thermal runaway occurs in a certain energy storage power station, the BMS emergency power supply module is activated, and the capacitor bank discharges 16.2V and boosts to 300V to drive the pulse generator. After the control instruction "deflector coordinates (10, 3, 1.2)" is modulated by QPSK, the I-channel signal is transmitted through the CAN bus, and the Q-channel is transmitted through the RS-485. After demodulation at the execution terminal, the instruction is restored, and the error rate <0.1%.

[0113] S2032: Deploy a dual-channel redundancy check module at the aerosol spraying device end to demodulate and cross-compare the two modulated waveforms. When the instruction consistency exceeds the threshold, generate an execution permission signal; otherwise, trigger the self-check loop; This step ensures the integrity of the instruction through dual-channel redundancy check, combines the self-check mechanism to prevent misoperation, and improves the fault tolerance of the fire protection system.

[0114] Redundancy Check Module Design Dual-channel Data Analysis: The I-channel and Q-channel signals are received through the CAN controller and RS-485 interface of the STM32F407 respectively, parsed into binary instructions and stored in the buffer area (Buffer1 and Buffer2, each with a capacity of 1KB).

[0115] Cross - comparison algorithm: Compare the contents of Buffer1 and Buffer2 bit by bit by byte, and calculate the consistency ratio Consistency = (number of matching bytes / total number of bytes) × 100%. For example, the consistency between the content of Buffer1 "010001101101" and Buffer2 "010001101101" is 100%; if Buffer2 is "010001101100", then Consistency = 11 / 12 × 100% ≈ 91.7%.

[0116] Execute permission logic: Threshold determination: Set the consistency threshold Consistency_Threshold = 95%. When Consistency ≥ 95%, trigger the execution permission signal (high level 5V); otherwise, trigger the self - check loop. For example, if Consistency = 91.7%, send a self - check request (error code E01) to the BMS.

[0117] Self - check loop process: Self - check includes signal path detection (such as testing the CAN bus terminal resistance of 120Ω), power supply voltage detection (required ≥ 11V), and solenoid valve impedance detection (normal range 20 - 30Ω). If the self - check passes, resend the instruction; if it fails, record the fault log (such as "ERR_CAN_BUS_OFF").

[0118] Anti - interference enhancement technology: Hamming code error correction: Add Hamming Code during the instruction encoding stage. Add 3 parity bits for every 4 - bit data, which can correct 1 - bit errors. For example, the original data "0100" is encoded as "0100011", and the receiving end detects and corrects single - bit flips through the parity bits.

[0119] Application example: During a certain fire trigger, 2 bytes of the I - channel instruction are lost due to electromagnetic interference (Consistency = 83.3%). The redundant check module triggers self - check, discovers that the CAN bus impedance is abnormal (measured 60Ω), switches to the backup channel and then resends the instruction, taking 120ms to complete error correction.

[0120] S2033, drive the high - speed solenoid valve according to the execution permission signal, adjust the injection angle and dose according to the spatial positioning parameters of the control instruction set, and synchronously start the telescopic diversion cover of the thermal runaway isolation device to form a physical barrier; This step realizes the coordinated action of the directional injection of the fire extinguishing agent and physical isolation through a high - precision actuator, ensuring the timeliness and spatial coverage of thermal runaway suppression.

[0121] High - speed solenoid valve control technology: Drive Circuit Design: The solenoid valve (model SMC VQD212) is driven by an H-bridge circuit, and a PWM signal is input (frequency 1 kHz, duty cycle 10% - 90% to adjust the flow rate). For example, a duty cycle of 50% corresponds to a valve opening of 50% and a flow rate of 20 L / min; the injection dose Dose = flow rate × time. If the injection time is 2 seconds, then Dose = 20 L / min × 2 / 60 ≈ 0.67 L.

[0122] Spatial Positioning Parameter Analysis: Extract three-dimensional coordinates (X, Y, Z) and injection angles (Azimuth θ, Elevation φ) from the control instruction. For example, the coordinates (2, 3, 1.2) correspond to module M3, θ = 45° (horizontal direction), φ = 30° (vertical elevation angle), and the nozzle direction is adjusted by a stepper motor (28BYJ - 48, step angle 5.625°).

[0123] Thermal Runaway Isolation Device Action: Retractable Deflector Control: The deflector is driven by a cylinder (air pressure 0.6 MPa), with an expansion speed of 0.5 m / s. After expansion, it forms a U-shaped barrier (size 1.5 m × 1 m × 0.5 m). For example, the deflector deployed at the coordinates (10, 3, 1.2) is fully expanded within 500 ms to isolate the heat conduction path between clusters.

[0124] Synchronous Timing Control: The start signals of the solenoid valve and the deflector are synchronized by an FPGA (model Xilinx Spartan - 6), with an error < 1 ms. For example, the solenoid valve is opened at t = 0 ms, and the deflector is started at t = 10 ms to ensure the timing match between the fire extinguishing agent injection and the formation of the physical barrier.

[0125] Application Example: For the thermal runaway event of module M3, the solenoid valve adjusts the nozzle direction according to the instruction (θ = 45°, φ = 30°) and injects 400 g of KL - 5 type aerosol fire extinguishing agent (flow rate 25 L / min × 1.6 seconds); at the same time, the deflector is deployed at the coordinates (10, 3, 1.2) to isolate the heat radiation channel between clusters C - 07 and C - 08.

[0126] S2034, capture the diffusion pattern of the fire extinguishing agent through an infrared camera, combine the air flow intensity feedback from the pressure sensor, generate an action execution effect evaluation matrix, and use it as a real-time action feedback signal; This step quantifies the execution effect of the fire fighting action through multi-sensor fusion technology, providing data support for closed-loop control.

[0127] Fire Extinguishing Agent Diffusion Monitoring: Infrared imaging technology: The FLIR A315 infrared camera (resolution 320×240, temperature measurement range -20°C to 150°C) captures images at a frame rate of 30 fps, and identifies the area covered by the fire extinguishing agent through temperature gradients. For example, when the temperature difference ΔT ≥ 5°C in the aerosol diffusion area (ambient 25°C → fire extinguishing area 20°C), it is determined to be effectively covered.

[0128] Image processing algorithm: Use the OpenCV library to perform binary processing on the infrared image (threshold ΔT ≥ 5°C), and calculate the coverage rate Coverage = number of valid pixels / total number of pixels × 100%. For example, if 5000 pixels in the image meet the standard and the total number of pixels is 76800, then Coverage ≈ 6.5%.

[0129] Airflow intensity feedback: Pressure sensor deployment: The MPX5700AP differential pressure sensor (range 0 - 700 kPa) is installed at the nozzle outlet to measure the real-time pressure P (unit: kPa). The formula for calculating the airflow intensity Q (L / min) is: Q = K × √(P / ρ), where K = 0.07 (calibration coefficient) and ρ = 1.2 kg / m³ (air density). For example, when P = 500 kPa, Q = 0.07 × √(500 / 1.2) ≈ 0.07 × 20.4 ≈ 1.43 L / s = 85.8 L / min.

[0130] Evaluation matrix generation: Matrix data structure: The evaluation matrix is a 5×5 table, including timestamp, coverage rate, airflow intensity, temperature reduction rate, and device status code. For example: Timestamp: 2024-07-15 14:23:07.500; Coverage: 65%; Flow: 85.8 L / min; Temp_Reduction: 12°C; Status_Code: 0x0000 (normal).

[0131] Application example: After a certain injection, the infrared image shows that the coverage rate is 65% (target 70%), the pressure sensor feedbacks an airflow intensity of 85.8 L / min (target 90 L / min), and the evaluation matrix marks the status code 0x0001 (insufficient flow), triggering subsequent optimization strategies.

[0132] In S2035, calculate the residual between the real-time action feedback signal and the expected control target. If the deviation exceeds the safety threshold, trigger a secondary injection, and at the same time update the fault code to the BMS log system.

[0133] This step dynamically adjusts the fire protection strategy through residual analysis and realizes system self-optimization by combining with the fault log.

[0134] Residual calculation logic: Deviation quantification method: Define the coverage deviation ΔC = |actual coverage rate - target coverage rate|, and the flow deviation ΔQ = |actual flow - target flow|. For example, if the target coverage rate is 70% and the actual is 65%, then ΔC = 5%; if the target flow is 90 L / min and the actual is 85.8 L / min, then ΔQ = 4.2 L / min.

[0135] Safety threshold setting: ΔC_max = 10%, ΔQ_max = 15 L / min. If ΔC > 10% or ΔQ > 15 L / min, it is determined that the action fails and a secondary injection is triggered.

[0136] Secondary injection strategy: Incremental adjustment rule: Secondary injection dose = original dose × (1 + ΔC / target coverage rate). For example, for the first injection of 400 g (ΔC = 5%), the secondary dose = 400 × (1 + 5 / 70) = 400 × 1.071 ≈ 428 g.

[0137] Execution priority improvement: The priority of the secondary injection instruction is increased from 80 to 95 to preempt system resources to ensure timely response.

[0138] Fault log management: Log format specification: Log entries include timestamp, device ID, fault type, and handling measures. For example: 2024-07-15 14:23:08.200 | Aerosol injector #23 | Low coverage rate (65% vs 70%) | Secondary injection 428 g.

[0139] Encryption and synchronization mechanism: After the log is encrypted by AES-256, it is synchronized to all BMS nodes through the LoRa wireless module (frequency band 433 MHz, rate 1 kbps).

[0140] Application example: An evaluation after a fire extinguishing shows that ΔC = 12% (exceeding the threshold of 10%), the system automatically triggers a secondary injection of 428 g, the coverage rate is increased, and at the same time, the fault log "ERR_COVERAGE_UNDER" is recorded.

[0141] Generate a high-voltage pulse signal using the emergency power supply (capacitor energy storage) of the BMS, encode the control instruction into two modulation waveforms with a 90° phase difference, and transmit them to the execution terminal through independent channels. The dual-channel redundancy check module demodulates and cross-verifies the signals. After ensuring the instruction consistency, trigger the aerosol injection (adjust the dose and angle) and the isolation device (physically isolate the telescopic deflector). At the same time, use the infrared camera and pressure sensor to real-time feedback the fire extinguishing agent diffusion effect and air flow intensity, generate an action execution evaluation matrix, and solve the problem of misoperation caused by signal interference or equipment failure in traditional fire fighting execution through the redundant signal verification and real-time feedback mechanism, improving the reliability of fire fighting actions and the traceability of execution effects.

[0142] S204. Based on the real-time action feedback signal and the battery cluster residual state data, update the parameter weights of the multi-source feature fusion network through the reinforcement learning model for the next cycle of fire fighting linkage control, forming a closed-loop control link. Specifically, it may include: S2041. Integrate the real-time action feedback signal and the battery cluster residual state data to construct a multi-dimensional reinforcement learning state vector; This step constructs the state space of reinforcement learning through multi-source data fusion, providing comprehensive environmental perception input for policy optimization.

[0143] ‌Data integration logic‌: ‌Real-time action feedback signal‌: Includes the percentage of aerosol injection coverage (e.g., 65%), the air flow intensity feedback by the pressure sensor (e.g., 85.8L / min), the temperature suppression rate identified by the infrared camera (e.g., ΔT = 12℃), and the device status code (e.g., 0x0000 indicates normal).

[0144] ‌Battery cluster residual state data‌: Collect the single-cell voltage after fire extinguishing (e.g., 3.2V / cell), the remaining capacity (SOC = 45%), the health status (SOH = 78%), and the inter-cluster insulation impedance (e.g., ≥100MΩ).

[0145] ‌Normalization processing‌: All data is uniformly mapped to the range [0,1]. For example, for the voltage range of 2.5V - 4.2V, 3.2V corresponds to the normalized value (3.2 - 2.5) / (4.2 - 2.5) = 0.41; the coverage rate of 65% corresponds to 0.65.

[0146] ‌State vector construction method‌: ‌Dimension Definition‌: The state vector is 14-dimensional, including: Action feedback data (4D): coverage rate, air flow intensity, temperature suppression rate, device status code; Battery residual data (6D): average cell voltage, SOC, SOH, insulation impedance, maximum temperature difference within the cluster, gas residual concentration; Spatiotemporal label (4D): X / Y / Z coordinates of the thermal runaway cluster, time stamp difference (second difference between the current time and the event trigger time).

[0147] ‌Example‌: In a certain case, the state vector is [0.65, 0.86, 0.48, 0.0, 0.41, 0.45, 0.78, 0.92, 0.15, 0.03, 2.1, 3.5, 1.2, 120], corresponding to the above 14 dimensions respectively.

[0148] ‌Application Example‌: After a certain energy storage system is extinguished, the "temperature suppression rate" in the state vector is 0.48 (corresponding to ΔT = 12℃ / 25℃ ambient temperature), the "maximum temperature difference within the cluster" is 0.15 (corresponding to 4.5℃ / 30℃ full scale), and the device status code is 0x0000 (normalized to 0.0), which is used for subsequent reinforcement learning training.

[0149] ‌S2042, design a reward function based on policy gradient. Among them, the positive reward includes the thermal runaway suppression efficiency and the resource consumption ratio, and the negative penalty includes the number of misoperations and the device loss value, generating a dynamic reward score; This step guides the reinforcement learning model to optimize decisions through a quantitative reward mechanism, balancing the fire extinguishing effect and cost loss.

[0150] ‌Positive Reward Calculation Rule‌: ‌Thermal Runaway Suppression Efficiency (R1)‌: Defined as (actual suppression time / theoretical fastest suppression time) × 100%. For example, if the theoretical fastest suppression time is 5 seconds and the actual time taken is 6 seconds, then R1 = 5 / 6 × 100 ≈ 83.3%, corresponding to a reward value of +8.3.

[0151] ‌Resource Consumption Ratio (R2)‌: Calculate the ratio of the amount of fire extinguishing agent used to the theoretical minimum amount. For example, if the theoretical minimum amount is 300g and the actual amount used is 400g, then R2 = 300 / 400 = 0.75, and the reward value is +7.5.

[0152] ‌Negative Penalty Calculation Rule‌: ‌Number of Misoperations (P1)‌: Record the number of times the fire protection device is triggered unnecessarily. For example, if it is mis-triggered once, the penalty value is -5.0.

[0153] Device loss value (P2): Calculated based on the number of solenoid valve operations (service life of 100,000 times) and the mechanical wear of the flow deflector (service life of 5,000 times). For example, in a certain operation, the solenoid valve life consumption is 0.01% (1 / 10,000), and the flow deflector life consumption is 0.02% (1 / 5,000), then P2 = -(0.01 + 0.02) × 100 = -3.0.

[0154] Dynamic reward scoring formula: Total reward R_total = α × (R1 + R2) + β × (P1 + P2), where α = 0.6 (effect weight) and β = 0.4 (loss weight).

[0155] Example: If R1 = 8.3, R2 = 7.5, P1 = -5.0, and P2 = -3.0, then R_total = 0.6 × (8.3 + 7.5) + 0.4 × (-5.0 - 3.0) = 0.6 × 15.8 + 0.4 × (-8.0) = 9.48 - 3.2 = 6.28.

[0156] Application example: In a certain fire extinguishing operation, the flow deflector life consumption is 0.05%, but due to the thermal runaway suppression efficiency reaching 95%, the total reward is 6.28; while in another case, due to mis-triggering, the reward is -2.1, and the model will automatically reduce the misjudgment probability for similar scenarios.

[0157] In S2043, the asynchronous advantage actor-critic algorithm is used to update the parameters of the multi-source feature fusion network. Among them, the actor network optimizes the feature weight allocation strategy, and the critic network corrects the prediction deviation of the thermal runaway level. This step realizes the efficient update of the model parameters through a distributed reinforcement learning framework, improving the accuracy of thermal runaway prediction.

[0158] Asynchronous Advantage Actor-Critic (A3C) architecture: Actor network: The input is a 14-dimensional state vector, and the output is the weight adjustment amounts (Δw1, Δw2, Δw3) of the voltage, temperature, and gas three channels in the multi-source feature fusion network. For example, the original weights are [0.3, 0.5, 0.2], and the actor network suggests adjusting to [+0.02, -0.01, +0.03].

[0159] Critic network: The same state vector is input, and the output is the deviation between the predicted value of the value function V(s) and the actual reward. For example, if the predicted V(s) = 6.5 and the actual R_total = 6.28, then the deviation δ = 6.28 - 6.5 = -0.22.

[0160] Asynchronous Update Mechanism: Deploy 8 parallel worker threads, each thread interacts with the environment independently, and synchronizes gradients to the global network regularly. For example, after Worker1 trains 10 batches locally (batch_size = 256), it uploads the gradients to update the global model.

[0161] Parameter Update Process: Advantage Value Calculation: A(s,a) = R_total + γ × V(s') - V(s), where γ = 0.99 is the discount factor and s' is the next state. For example, if V(s) = 6.5, V(s') = 7.0, and R_total = 6.28, then A = 6.28 + 0.99 × 7.0 - 6.5 = 6.28 + 6.93 - 6.5 = 6.71.

[0162] Policy Gradient Update: The actor network adjusts the weight allocation policy according to the advantage value, the learning rate is set to 0.001, and the RMSprop optimizer is used (decay factor ρ = 0.9).

[0163] Value Function Correction: The critic network uses δ = -0.22 as the supervision signal and updates the parameters using the mean squared error (MSE) loss function.

[0164] Application Example: In a certain training, the actor network increased the gas channel weight from 0.2 to 0.23, and the critic network corrected the prediction deviation from -0.22 to +0.05, resulting in a 12% increase in the confidence level of the subsequent thermal runaway prediction level.

[0165] S2044, Encrypt and synchronize the updated network parameters to all BMS nodes, and reset the fire linkage control cycle counter to form a closed-loop control link from state perception to policy optimization.

[0166] This step realizes the cooperation of the distributed system through a secure communication protocol to ensure policy consistency and complete closed-loop control.

[0167] Parameter Encryption and Synchronization Mechanism: Encryption Algorithm: Use AES-256 (Advanced Encryption Standard, 256-bit key) to encrypt the updated network parameters. For example, the parameter file (size 512KB) is encrypted in CBC (Cipher Block Chaining) mode, and the initialization vector (IV) is a 16-byte random number.

[0168] Synchronization Protocol: Use the MQTT (Message Queuing Telemetry Transport) protocol to publish encrypted parameters with the topic " / BMSCluster / ModelUpdate". Each BMS node subscribes to this topic and decrypts the data through the Hardware Security Module (HSM) after receiving it.

[0169] Version Control: Increment the version number each time there is an update (e.g., v1.2.3 → v1.2.4), and record the hash value (SHA-256) to prevent tampering.

[0170] Control Cycle Reset Logic: Counter Initialization: The fire linkage control cycle counter starts from 0, and the maximum number of cycles is set to 1000 times (about 30 days). After reaching this limit, a forced full-system model update is triggered.

[0171] Event-Triggered Reset: After each successful execution of a fire action, the counter is reset to 0; if there are no thermal runaway events for 10 consecutive cycles, a lightweight model fine-tuning is triggered (only update the critic network).

[0172] Application Example: In a certain model update, the encrypted parameter file (version v2.1.9) is broadcast to 56 BMS nodes through MQTT. The decryption takes 15 ms, and the system completes full-node synchronization within 200 ms. After the counter is reset, it enters the first cycle of monitoring.

[0173] Integrate fire action feedback data (such as the coverage of fire extinguishing agent, isolation effect) and the remaining state of the battery cluster (such as voltage recovery, temperature drop rate) to construct a reinforcement learning state vector. Adopt the Asynchronous Advantage Actor-Critic (A3C) algorithm. The actor network optimizes the feature weight allocation strategy of the multi-source feature fusion network, the critic network corrects the thermal runaway prediction deviation, and the reward function synthesizes the thermal runaway suppression efficiency, resource consumption ratio, and equipment loss value. Dynamically update the network parameters and encrypt and synchronize them to all BMS nodes to achieve a closed-loop control from data perception to policy optimization. By continuously learning historical fire cases and real-time environmental changes, adaptively improve the accuracy of thermal runaway prediction and the efficiency of fire linkage, and enhance the robustness and self-adaptability of the system during long-term operation.

[0174] It can be seen that, based on the real-time voltage, temperature, and gas concentration data of the battery cluster, thermal runaway prediction is performed through a multi-source feature fusion network, and a thermal runaway prediction level is output; based on the thermal runaway prediction level, using a dynamic priority mapping algorithm, according to the battery cluster topology, a control instruction set including execution priority and spatial positioning is output; according to the control instruction set, a dual-channel redundant signal verification mechanism is adopted to execute fire-fighting actions and output real-time action feedback signals; based on the real-time action feedback signals and the residual state data of the battery cluster, the parameter weights of the multi-source feature fusion network are updated through a reinforcement learning model for the next cycle of fire-fighting linkage control, forming a closed-loop control link, thereby enabling battery cluster-level fire-fighting linkage control to improve the safety protection ability of the battery system.

[0175] Another embodiment of the present invention provides a battery cluster-level fire-fighting linkage control system based on a BMS. Refer to Figure 3 , the system may include: A prediction module 301, configured to perform thermal runaway prediction through a multi-source feature fusion network according to the real-time voltage, temperature, and gas concentration data of the battery cluster. The multi-source feature fusion network fuses the SOC gradient change feature and the SOH attenuation curve feature, and outputs a thermal runaway prediction level; A planning module 302, configured to, based on the thermal runaway prediction level, use a dynamic priority mapping algorithm, and according to the battery cluster topology, perform collaborative planning on the aerosol spraying range and the thermal runaway isolation area, and output a control instruction set including execution priority and spatial positioning; An execution module 303, configured to, according to the control instruction set, drive an aerosol jet fire extinguishing device and a thermal runaway isolation device through a pulse trigger circuit powered by a redundant power source, adopt a dual-channel redundant signal verification mechanism to execute fire-fighting actions, and output real-time action feedback signals; An update module 304, configured to, based on the real-time action feedback signals and the residual state data of the battery cluster, update the parameter weights of the multi-source feature fusion network through a reinforcement learning model for the next cycle of fire-fighting linkage control, forming a closed-loop control link.

[0176] It can be seen that according to the real-time voltage, temperature and gas concentration data of the battery cluster, thermal runaway prediction is carried out through a multi-source feature fusion network, and the thermal runaway prediction level is output; based on the thermal runaway prediction level, using the dynamic priority mapping algorithm, according to the battery cluster topology structure, a control instruction set including execution priority and spatial positioning is output; according to the control instruction set, a dual-channel redundant signal verification mechanism is adopted to execute the fire-fighting action, and a real-time action feedback signal is output; based on the real-time action feedback signal and the residual state data of the battery cluster, the parameter weights of the multi-source feature fusion network are updated through a reinforcement learning model for the next cycle of fire-fighting linkage control, forming a closed-loop control link, so as to realize the battery cluster-level fire-fighting linkage control and improve the safety protection ability of the battery system.

[0177] An embodiment of the present invention further provides a storage medium, in which a computer program is stored, and wherein the computer program is set to execute the steps in any one of the above method embodiments when running.

[0178] Specifically, in this embodiment, the above storage medium can be set to store a computer program for executing the following steps: S201, according to the real-time voltage, temperature and gas concentration data of the battery cluster, carry out thermal runaway prediction through a multi-source feature fusion network, the multi-source feature fusion network fuses the SOC gradient change feature and the SOH attenuation curve feature, and outputs the thermal runaway prediction level; S202, based on the thermal runaway prediction level, using the dynamic priority mapping algorithm, according to the battery cluster topology structure, carry out collaborative planning on the aerosol spraying range and the thermal runaway isolation area, and output a control instruction set including execution priority and spatial positioning; S203, according to the control instruction set, drive the aerosol fire extinguishing device and the thermal runaway isolation device through a pulse trigger circuit powered by a redundant power source, adopt a dual-channel redundant signal verification mechanism to execute the fire-fighting action, and output a real-time action feedback signal; S204, based on the real-time action feedback signal and the residual state data of the battery cluster, update the parameter weights of the multi-source feature fusion network through a reinforcement learning model for the next cycle of fire-fighting linkage control, forming a closed-loop control link.

[0179] It can be seen that according to the real-time voltage, temperature and gas concentration data of the battery cluster, thermal runaway prediction is carried out through a multi-source feature fusion network, and a thermal runaway prediction level is output; based on the thermal runaway prediction level, using a dynamic priority mapping algorithm, according to the battery cluster topology, a control instruction set including execution priority and spatial positioning is output; according to the control instruction set, a fire-fighting action is executed by adopting a dual-channel redundant signal verification mechanism, and a real-time action feedback signal is output; based on the real-time action feedback signal and the residual state data of the battery cluster, the parameter weights of the multi-source feature fusion network are updated through a reinforcement learning model for the next cycle of fire-fighting linkage control, forming a closed-loop control link, so as to realize the fire-fighting linkage control at the battery cluster level and improve the safety protection ability of the battery system.

[0180] An embodiment of the present invention further provides an electronic device, including a memory and a processor, wherein a computer program is stored in the memory, and the processor is configured to run the computer program to execute the steps in any one of the above method embodiments.

[0181] Specifically, the above electronic device may further include a transmission device and an input / output device, wherein the transmission device is connected to the above processor, and the input / output device is connected to the above processor.

[0182] Specifically, in this embodiment, the above processor may be configured to execute the following steps through a computer program: S201, according to the real-time voltage, temperature and gas concentration data of the battery cluster, perform thermal runaway prediction through a multi-source feature fusion network, and the multi-source feature fusion network fuses the SOC gradient change feature and the SOH attenuation curve feature, and outputs a thermal runaway prediction level; S202, based on the thermal runaway prediction level, using a dynamic priority mapping algorithm, according to the battery cluster topology, synergistically plan the aerosol spraying range and the thermal runaway isolation area, and output a control instruction set including execution priority and spatial positioning; S203, according to the control instruction set, drive the aerosol fire extinguishing device and the thermal runaway isolation device through a pulse trigger circuit powered by a redundant power source, execute a fire-fighting action by adopting a dual-channel redundant signal verification mechanism, and output a real-time action feedback signal; S204, based on the real-time action feedback signal and the residual state data of the battery cluster, update the parameter weights of the multi-source feature fusion network through a reinforcement learning model for the next cycle of fire-fighting linkage control, forming a closed-loop control link.

[0183] It can be seen that according to the real-time voltage, temperature and gas concentration data of the battery cluster, thermal runaway prediction is carried out through a multi-source feature fusion network, and the thermal runaway prediction level is output; based on the thermal runaway prediction level, using the dynamic priority mapping algorithm, according to the battery cluster topology structure, a control instruction set including execution priority and spatial positioning is output; according to the control instruction set, a fire-fighting action is executed by adopting a dual-channel redundant signal verification mechanism, and a real-time action feedback signal is output; based on the real-time action feedback signal and the remaining state data of the battery cluster, the parameter weights of the multi-source feature fusion network are updated through a reinforcement learning model for the next cycle of fire-fighting linkage control, forming a closed-loop control link, so as to realize the fire-fighting linkage control at the battery cluster level and improve the safety protection ability of the battery system.

[0184] The structure, features and effects of the present invention have been described in detail based on the embodiments shown in the drawings. The above is only the preferred embodiment of the present invention, but the present invention is not limited to the scope defined by the drawings. Any changes made according to the concept of the present invention, or equivalent embodiments modified into equivalent changes, still within the spirit covered by the description and drawings, shall be within the protection scope of the present invention.

Claims

1. A battery cluster-level fire linkage control method based on BMS, characterized in that, The method includes: Based on the real-time voltage, temperature, and gas concentration data of the battery cluster, thermal runaway prediction is performed through a multi-source feature fusion network. The multi-source feature fusion network fuses the SOC gradient change feature and the SOH decay curve feature, and outputs a thermal runaway prediction level. Based on the thermal runaway prediction level, using a dynamic priority mapping algorithm, according to the battery cluster topology, collaborative planning is performed on the aerosol injection range and the thermal runaway isolation area, and a control instruction set including execution priority and spatial positioning is output. According to the control instruction set, the aerosol injection fire extinguishing device and the thermal runaway isolation device are driven by a pulse trigger circuit powered by the redundant power supply, and the fire fighting action is executed by a dual-channel redundant signal verification mechanism, and a real-time action feedback signal is output. Based on the real-time action feedback signal and the battery cluster residual state data, the parameter weights of the multi-source feature fusion network are updated through a reinforcement learning model for the next cycle of fire fighting linkage control, forming a closed-loop control link.

2. The method according to claim 1, wherein The thermal runaway prediction through the multi-source feature fusion network based on the real-time voltage, temperature, and gas concentration data of the battery cluster. The multi-source feature fusion network fuses the SOC gradient change feature and the SOH decay curve feature, and outputs a thermal runaway prediction level, includes: Collect the time-series fluctuation data of the voltages of each single battery in the battery cluster, the spatial distribution heat map of the temperature sensors, and the gas concentration change rate, perform timestamp synchronization alignment, and generate a cross-dimensional original monitoring matrix. Perform outlier correction on the original monitoring matrix, extract the SOC gradient change feature using the sliding window difference method, and at the same time fit the SOH decay curve in combination with the cyclic aging experiment data to generate a SOC-SOH joint degradation feature vector. Construct a multi-source feature fusion network, input the SOC-SOH joint degradation feature vector and the real-time temperature and gas concentration data into a bidirectional gated recurrent unit, and dynamically allocate the weight coefficients of the voltage, temperature, and gas three channels through a feature-level attention mechanism, and output a fused thermal runaway feature encoding. Input the thermal runaway feature encoding into a pre-trained multi-level classifier, divide the warning level according to the threshold boundary of historical thermal runaway cases, and generate a thermal runaway prediction level with confidence and mark the spatio-temporal position label.

3. The method according to claim 2, characterized in that, The collaborative planning of the aerosol injection range and the thermal runaway isolation area based on the thermal runaway prediction level using a dynamic priority mapping algorithm according to the battery cluster topology, and outputting a control instruction set including execution priority and spatial positioning, includes: Analyze the spatio-temporal label of the thermal runaway prediction level, construct an adjacency matrix according to the physical topology of the battery cluster, simulate the maximum energy diffusion range of the thermal runaway propagation path, and generate a potential risk propagation map. In the potential risk propagation map, with the battery cell with an urgent prediction level as the center, calculate the aerosol injection coverage radius and the safe isolation distance between adjacent clusters, and generate an initial control area boundary. Use the dynamic priority mapping algorithm to perform weighted optimization on the initial control area in combination with the parallel string relationship and the heat dissipation channel distribution of the battery cluster. Divide the concentric circle action scope of aerosol injection according to the weighted optimization result, mark the deployment coordinates of the thermal runaway isolation device, and generate a draft three-dimensional space control instruction; Perform conflict detection on the draft three-dimensional space control instruction, eliminate the overlapping area of the device action scope, and output the final control instruction set including the execution priority and spatial positioning.

4. The method according to claim 3, characterized in that According to the control instruction set, drive the aerosol injection fire extinguishing device and the thermal runaway isolation device through a pulse trigger circuit powered by a redundant power supply, and execute the fire fighting action by adopting a dual-channel redundant signal verification mechanism, and output a real-time action feedback signal, including: Activate the emergency power supply module of the BMS, generate a high-voltage trigger signal by using a capacitor energy storage type pulse generator, encode the control instruction set into two modulation waveforms with a 90° phase difference, and transmit them to the execution terminal through independent channels; Deploy a dual-channel redundant verification module at the aerosol injection device end, demodulate and cross-compare the two modulation waveforms, generate an execution permission signal when the instruction consistency exceeds the threshold, otherwise trigger the self-check loop; Drive the high-speed solenoid valve according to the execution permission signal, adjust the injection angle and dose according to the spatial positioning parameters of the control instruction set, and synchronously start the telescopic diversion cover of the thermal runaway isolation device to form a physical barrier; Capture the diffusion form of the fire extinguishing agent through an infrared camera, combine the air flow intensity feedback by the pressure sensor, and generate an action execution effect evaluation matrix as the real-time action feedback signal; Perform residual calculation on the real-time action feedback signal and the expected control target. If the deviation exceeds the safety threshold, trigger secondary injection, and at the same time update the fault code to the BMS log system.

5. The method according to claim 4, characterized in that, Based on the real-time action feedback signal and the battery cluster residual state data, update the parameter weights of the multi-source feature fusion network through a reinforcement learning model for the next cycle of fire fighting linkage control, and form a closed-loop control link, including: Integrate the real-time action feedback signal and the battery cluster residual state data to construct a multi-dimensional reinforcement learning state vector; Design a reward function based on policy gradient, where the positive reward includes the thermal runaway suppression efficiency and the resource consumption ratio, and the negative penalty includes the number of misoperations and the device loss value, and generate a dynamic reward score; Adopt the asynchronous advantage actor-critic algorithm to update the parameters of the multi-source feature fusion network. Among them, the actor network optimizes the feature weight allocation strategy, and the critic network corrects the thermal runaway level prediction deviation; Encrypt and synchronize the updated network parameters to all BMS nodes, and reset the fire fighting linkage control cycle counter to form a closed-loop control link from state perception to policy optimization.

6. A battery cluster-level fire linkage control system based on BMS, characterized in that, The system includes: A prediction module for performing thermal runaway prediction through a multi-source feature fusion network according to the real-time voltage, temperature and gas concentration data of the battery cluster. The multi-source feature fusion network fuses the SOC gradient change feature and the SOH decay curve feature, and outputs the thermal runaway prediction level; A planning module for synergistically planning the aerosol injection range and the thermal runaway isolation area based on the thermal runaway prediction level, using a dynamic priority mapping algorithm according to the battery cluster topology, and outputting a control instruction set including the execution priority and spatial positioning; An execution module, configured to drive an aerosol jet fire extinguishing device and a thermal runaway isolation device through a pulse trigger circuit powered by a redundant power supply according to the control instruction set, execute a fire fighting action by adopting a dual-channel redundant signal verification mechanism, and output a real-time action feedback signal; An update module, configured to update the parameter weights of the multi-source feature fusion network based on the real-time action feedback signal and the remaining state data of the battery cluster through a reinforcement learning model for the next-cycle fire fighting linkage control, so as to form a closed-loop control link.

7. The system according to claim 6, characterized in that, The prediction module is specifically configured to: Collect the time-series fluctuation data of the voltages of each cell of the battery cluster, the spatial distribution thermal map of the temperature sensors, and the gas concentration change rate, perform timestamp synchronization alignment, and generate a cross-dimensional original monitoring matrix; Correct the outliers in the original monitoring matrix, adopt a sliding window difference method to extract the gradient change features of the SOC, and simultaneously fit the SOH decay curve by combining the cyclic aging experiment data to generate a SOC-SOH joint degradation feature vector; Construct a multi-source feature fusion network, input the SOC-SOH joint degradation feature vector and the real-time temperature and gas concentration data into a bidirectional gated recurrent unit, and dynamically allocate the weight coefficients of the voltage, temperature, and gas three channels through a feature-level attention mechanism to output a fused thermal runaway feature code; Input the thermal runaway feature code into a pre-trained multi-level classifier, divide the warning levels according to the threshold boundaries of historical thermal runaway cases, generate a thermal runaway prediction level with confidence and mark the spatio-temporal position label.

8. The system according to claim 7, wherein The planning module is specifically configured to: Analyze the spatio-temporal labels of the thermal runaway prediction level, construct an adjacency matrix according to the physical topology structure of the battery cluster, simulate the maximum energy diffusion range of the thermal runaway propagation path, and generate a potential risk propagation map; In the potential risk propagation map, taking the battery cell with an urgent prediction level as the center, calculate the aerosol jet coverage radius and the safety isolation distance between adjacent clusters to generate the boundary of the initial control area; Adopt a dynamic priority mapping algorithm, and combine the parallel string relationship of the battery cluster and the distribution of the heat dissipation channels to perform weighted optimization on the initial control area; Divide the concentric circle action scope of the aerosol jet according to the weighted optimization result, and mark the deployment coordinates of the thermal runaway isolation device to generate a draft three-dimensional space control instruction; Perform conflict detection on the draft three-dimensional space control instruction, eliminate the overlapping area of the device action range, and output a final control instruction set including the execution priority and the spatial positioning.

9. A storage medium, characterized in that, A computer program is stored in the storage medium, wherein the computer program is set to execute the method according to any one of claims 1-5 when running.

10. An electronic device, comprising a memory and a processor, characterized in that, A computer program is stored in the memory, and the processor is set to run the computer program to execute the method according to any one of claims 1-5.

Images