A battery cabin fire automatic early warning method and system, electronic equipment and storage medium

Abstract

The application discloses a battery cabin fire automatic early warning method and system, electronic equipment and storage medium, which are used for realizing early and accurate early warning of battery cabin thermal runaway. The method first collects voltage time series of the battery monomer charging and discharging process, reconstructs the phase space through mutual information method and false nearest neighbor point method, constructs a recurrence plot matrix and extracts five-dimensional chaotic features such as recurrence rate and determination rate. Relying on the battery electrical connection and space thermal coupling topology, a graph convolution network is built to calculate the abnormal score of the battery monomer. Through the sliding window combined with the first and second derivative analysis, the real abnormal monomer is accurately identified. A local anomaly propagation network is built around the abnormal monomer, the average path length and clustering coefficient of the network are calculated, and the fault diffusion trend is quantified. Finally, according to the change rate threshold of the two topological parameters, the abnormal spread degree is judged, and the fire warning signal is triggered. The method can capture early implicit voltage anomalies of the battery, and effectively identify the fault deterioration and propagation situation.

Description

Technical Field

[0001] This invention belongs to the field of battery fire early warning technology, specifically relating to an automatic early warning method, system, electronic device and storage medium for battery compartment fire. Background Technology

[0002] Battery compartment fire early warning in electrochemical energy storage systems is one of the core technologies for ensuring energy storage safety. With the large-scale application of lithium-ion batteries in the energy storage field, fire accidents caused by battery thermal runaway are occurring frequently, placing higher demands on the reliability, timeliness, and accuracy of early warning technologies.

[0003] Existing early warning methods mainly fall into two categories. The first category relies on external sensing devices such as temperature sensors, gas detectors, or smoke sensors to assess fire risk by monitoring changes in environmental parameters inside the battery compartment. When a battery experiences thermal runaway, the heat and gases generated by its internal chemical reactions need to first diffuse to the outside of the battery casing before reaching the sensor installation location, a process with a significant physical propagation delay. Therefore, this type of method struggles to provide early warnings in the early stages of thermal runaway, often only responding after the battery has shown obvious abnormalities or even caught fire, leaving extremely limited time for proactive intervention.

[0004] The second type of method directly utilizes electrical parameters such as terminal voltage and current collected in real time by the battery management system. It identifies early faults by analyzing the consistency deviation of individual cell voltages, abnormal fluctuations in the rate of voltage change, or distortion characteristics of the charge-discharge curves. Compared to external sensor methods, this type of method can detect voltage anomalies caused by internal faults such as micro-short circuits earlier. However, existing electrical parameter early warning technologies generally have the following limitations: most methods are limited to the independent analysis of individual cell voltage data, or use simple statistical indicators such as range and variance to roughly characterize the overall voltage inconsistency of the battery pack, failing to fully utilize the rich correlation information contained between the multi-channel voltage time-series signals within the battery pack.

[0005] In reality, the individual battery cells within the battery compartment do not operate in isolation. Instead, they form electrical circuits through series or parallel connections and are thermally coupled via shared cooling ducts or liquid cooling pipes. When a micro-short circuit occurs in a single cell, the slow drop in its terminal voltage is not only directly reflected in its own timing waveform but also causes a redistribution of the circuit current through electrical connections, thereby generating weak, characteristic disturbances in the voltages of other cells. Simultaneously, the thermal management system also transfers localized heat to adjacent cells, leaving traceable traces in the voltage signals of those adjacent cells. This nonlinear dynamic coupling relationship between multiple cells is essentially an early sign of thermal runaway spreading from local to global, and is of significant value for achieving early warning.

[0006] However, existing methods generally lack effective means to explore such deep correlation structures. Most existing fault diagnosis methods assume that the state evolution of each battery cell is independent of each other and fail to establish a graph structure model that can characterize the anomaly propagation law. Summary of the Invention

[0007] In view of this, the present invention provides an automatic early warning method, system, electronic device and storage medium for battery compartment fire, in order to solve the problem that most existing fault diagnosis methods assume that the state evolution of each battery cell is independent and fail to establish a graph structure model that can characterize the law of anomaly propagation.

[0008] The technical solution adopted in this invention is as follows: An automatic early warning method for battery compartment fire, characterized in that it includes: Step S1: During the charging or discharging operation of the battery management system, the terminal voltage value of each battery cell is continuously collected at a preset sampling period to form a voltage time sequence of each battery cell. Step S2: Perform phase space reconstruction to generate an embedded trajectory matrix for the voltage time series of each battery cell. Specifically, this includes: for the voltage time series of each battery cell, calculating the mutual information function using the mutual information method, and taking the delay value when the mutual information function first reaches a local minimum as the delay time for phase space reconstruction; gradually increasing the embedding dimension starting from the delay time being equal to one sampling period, calculating the proportion of false nearest neighbors under each embedding dimension, and determining the embedding dimension as the final embedding dimension when the proportion first drops below a preset threshold; based on the determined delay time and embedding dimension, combining each data point in the voltage time series with several subsequent data points of delay intervals to form a high-dimensional phase point, and arranging all high-dimensional phase points in chronological order to form the embedded trajectory matrix.

[0009] The calculation of the Euclidean distance between any two phase points in the embedded trajectory matrix and the binarization based on the recursion threshold to obtain a recursive graph matrix specifically includes: taking each row of the embedded trajectory matrix as a phase point, and sequentially calculating the Euclidean distance between each phase point and all other phase points to obtain a distance matrix; calculating the average value of all non-zero distances in the distance matrix, and using a preset ratio of the average value as the recursion threshold; setting the elements in the distance matrix that are less than or equal to the recursion threshold to the value 1, and setting the elements that are greater than the recursion threshold to the value 0, to obtain a recursive graph matrix composed of values ​​1 and 0.

[0010] Extract a recursive quantization feature vector from the recursive graph matrix, containing recursion rate, determination rate, laminar flow rate, recursion entropy, and capture time; specifically including: Calculate the recursion rate, which is equal to the number of elements with a value of 1 in the recursion graph matrix divided by the total number of elements in the recursion graph matrix; To calculate the certainty rate, find all diagonal segments consisting of consecutive 1s in the recursive graph matrix, count the total number of 1s on diagonal segments whose length is greater than or equal to a preset minimum length, and divide the total number by the total number of 1s in the recursive graph matrix to obtain the certainty rate. To calculate the laminar flow rate, find all vertical line segments consisting of consecutive values ​​1 in the recursive graph matrix, count the total number of values ​​1 on vertical line segments whose length is greater than or equal to a preset minimum length, and divide the total number by the total number of values ​​1 in the recursive graph matrix to obtain the laminar flow rate. To calculate the recursive entropy, the probability of occurrence of diagonal segments of different lengths in the recursive graph matrix is ​​counted. For each probability of occurrence, the logarithm to the base of the natural constant is taken and multiplied by the corresponding probability of occurrence. Then, all the products are added together and the negative value is taken to obtain the recursive entropy. To calculate the capture time, multiply the length of each diagonal segment in the recursive graph matrix by its probability of occurrence, and then sum all the products to obtain the capture time. The recursion rate, determination rate, laminar flow rate, recursion entropy, and capture time are combined in sequence into a five-dimensional vector, which serves as the recursive quantization feature vector for each battery cell.

[0011] Step S3: Organize the recursively quantized feature vectors of all battery cells into a multi-channel feature tensor according to the spatial arrangement position within the battery compartment. Input the multi-channel feature tensor into the graph convolutional network model and output the anomaly score of each battery cell. Step S3 specifically includes: Obtain the spatial arrangement coordinates of all battery cells in the battery compartment. Based on the series and parallel electrical connection relationship and spatial adjacency relationship of the battery cells, construct an undirected graph structure with each battery cell as a node and electrical connection or spatial distance less than a preset threshold as edges. The recursive quantization feature vector of each battery cell is used as the initial feature of the corresponding node. The initial features of all nodes are stacked into a two-dimensional matrix according to the arrangement order of the nodes in the battery compartment. The two-dimensional matrix is ​​used as a multi-channel feature tensor. Construct a two-layer graph convolutional network and input the multi-channel feature tensor and the adjacency matrix of the undirected graph structure into the two-layer graph convolutional network; The initial features of each node are aggregated with the features of its neighboring nodes through the first layer of graph convolution, and the hidden features are output. The hidden features are aggregated again by a second layer of graph convolution, and a one-dimensional scalar value for each node is output. The one-dimensional scalar value is mapped to the interval between 0 and 1 using the Sigmoid function to obtain the anomaly score for each battery cell.

[0012] Step S4: Select the battery cell with the highest anomaly score as the candidate anomaly cell, perform a sliding window analysis of the recursion rate on the voltage time series of the candidate anomaly cell to generate a recursion rate change curve, calculate the first and second derivatives of the recursion rate change curve in the current window, and mark the candidate anomaly cell as a confirmed anomaly cell when the first derivative is greater than the positive threshold and the second derivative is greater than the acceleration threshold. Step S4 specifically includes: Step S41: Obtain the original voltage timing sequence of the candidate abnormal unit, set a sliding window of fixed length, the width of the sliding window is a preset window width range, and the sliding step size is a preset step size; Step S42: Move the sliding window sequentially to each position of the original voltage time series, repeat step S2 for the voltage subsequence within the window at each window position, obtain the recursive quantization feature vector corresponding to the window subsequence, and extract the recursion rate from the recursive quantization feature vector; Step S43: Arrange the recursion rate values ​​calculated for all window positions in chronological order of the window center time to form a recursion rate change curve; Step S44: Perform local polynomial smoothing on the recursion rate change curve to obtain a smoothed recursion rate curve; Step S45: On the smoothed recursion rate curve, for the current window position, take the recursion rate value one step before and after the current window position, calculate the difference between the previous recursion rate value and the next recursion rate value, divide it by twice the step size, and use the calculation result as the first derivative of the current window position. Step S46: Repeat the operation of step S45 on the obtained first derivative sequence to obtain the second derivative sequence. Set the positive threshold of the first derivative and the acceleration threshold of the second derivative. When the first derivative at the current window position is greater than the positive threshold and the second derivative at the current window position is greater than the acceleration threshold, mark the candidate abnormal entity as a confirmed abnormal entity.

[0013] Step S5: Using the confirmed abnormal cell as the center, and combining the recursive quantized feature vectors of adjacent battery cells that have electrical connection and thermal coupling relationship with it, construct a local anomaly propagation network, and calculate the average path length and clustering coefficient of the local anomaly propagation network. Step S5 specifically includes: Using the confirmed abnormal cell as the central node, all battery cells that are directly connected in series or in parallel with the central node are obtained from the electrical topology of the battery management system and used as the first-layer neighbor nodes. Obtain the flow direction of the cooling medium in the battery compartment thermal management system, and designate the battery cells located downstream of the central node and separated from the central node by no more than two battery cells as the second-layer neighbor nodes. The central node, the first-layer neighbor nodes, and the second-layer neighbor nodes together constitute the node set of the local anomaly propagation network; For each pair of nodes in the node set, calculate the cosine similarity between the recursively quantized feature vectors of the two nodes. When the cosine similarity is greater than a preset similarity threshold, establish an edge between the pair of nodes and use the cosine similarity as the weight of the edge. Calculate the average path length of the local anomaly propagation network. The average path length is equal to the sum of the shortest path lengths between all node pairs divided by the total number of node pairs. The shortest path length between node pairs is defined as the minimum of the sum of the reciprocals of the weights of each edge on the path. To calculate the clustering coefficient of the local anomaly propagation network, for each node, the actual number of edges between all neighboring nodes is divided by the maximum possible number of edges between the neighboring nodes to obtain the local clustering coefficient of the node. Then, the arithmetic mean of the local clustering coefficients of all nodes is taken to obtain the clustering coefficient of the local anomaly propagation network.

[0014] Step S6: When the rate of decrease of the average path length exceeds the path decrease threshold and the rate of increase of the clustering coefficient exceeds the cluster increase threshold, an early warning signal is generated.

[0015] An automatic early warning system for battery compartment fires includes: The voltage timing acquisition module continuously acquires the terminal voltage value of each battery cell at a preset sampling period during the charging or discharging operation of the battery management system, forming a voltage timing sequence for each battery cell. The phase space reconstruction and recursive graph generation module performs phase space reconstruction on the voltage time sequence of each battery cell to generate an embedded trajectory matrix, calculates the Euclidean distance between any two phase points in the embedded trajectory matrix, and obtains the recursive graph matrix by binarization according to the recursive threshold. The recursive quantization feature extraction module extracts a recursive quantization feature vector from the recursive graph matrix, which includes recursion rate, determination rate, laminar flow rate, recursion entropy, and capture time. The graph convolutional anomaly detection module is used to organize the recursively quantized feature vectors of all battery cells into a multi-channel feature tensor according to the spatial arrangement position within the battery compartment. The multi-channel feature tensor is then input into the graph convolutional network model to output the anomaly score for each battery cell. The recursion rate evolution analysis module selects the battery cell with the highest anomaly score as a candidate anomaly cell, performs a sliding window analysis of the recursion rate on the voltage time series of the candidate anomaly cell to generate a recursion rate change curve, calculates the first and second derivatives of the recursion rate change curve within the current window, and marks the candidate anomaly cell as a confirmed anomaly cell when the first derivative is greater than a positive threshold and the second derivative is greater than an acceleration threshold. The local anomaly propagation network module, centered on the confirmed abnormal cell, combines the recursive quantized feature vectors of adjacent battery cells that have electrical connections and thermal coupling relationships with it to construct a local anomaly propagation network, and calculates the average path length and clustering coefficient of the local anomaly propagation network. The early warning generation module generates an early warning signal when the rate of decrease of the average path length exceeds the path decrease threshold and the rate of increase of the clustering coefficient exceeds the cluster increase threshold.

[0016] An electronic device includes a processor, a memory, and a device stored in the memory and capable of being used in various applications. The computer program running on the processor executes an automatic early warning method for battery compartment fire.

[0017] A computer-readable storage medium storing a computer program that, when executed by a processor, implements an automatic early warning method for a battery compartment fire.

[0018] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are: 1. In this invention, recursive quantization analysis is performed directly using existing voltage time-series data from the battery management system. This allows for the capture of implicit nonlinear dynamic characteristics in the voltage time-series signal at an early stage of micro-short circuits within the battery. By performing phase space reconstruction and recursive graph generation on the voltage time-series sequence, recursive quantization features such as recursion rate, determinism rate, and laminar flow rate are extracted. This allows for the earlier detection of abnormal signs compared to traditional passive early warning methods based on temperature or gas sensors, thus gaining a valuable time window for proactive intervention. 2. In this invention, a graph network structure is constructed with each battery cell as a node and electrical connections and thermal coupling relationships as edges. The recursive quantization features of adjacent nodes are automatically aggregated using a graph convolutional network, enabling effective mining of nonlinear dynamic coupling relationships between multiple cells within the battery pack. When a battery experiences a micro-short circuit, its characteristic perturbation is propagated and amplified through the neighbor aggregation mechanism of graph convolution, significantly improving the sensitivity and accuracy of anomaly identification. Attached Figure Description

[0019] The present invention will be described by way of example and with reference to the accompanying drawings, wherein: Figure 1 This is a schematic diagram of the process structure of the present invention; Figure 2 This is a schematic diagram of the process structure of step S3 of the present invention; Figure 3 This is a schematic diagram of the process structure of step S4 of the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0021] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0022] It should be noted that, unless otherwise specified, the embodiments and features described in this invention can be combined with each other.

[0023] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0024] In this invention, unless otherwise explicitly specified and limited, "above" or "below" the second feature can include direct contact between the first and second features, or contact between the first and second features through another feature between them. Furthermore, "above," "over," and "on top" of the second feature includes the first feature directly above or diagonally above the second feature, or simply indicates that the first feature is at a higher horizontal level than the second feature. "Below," "below," and "under" the second feature includes the first feature directly below or diagonally below the second feature, or simply indicates that the first feature is at a lower horizontal level than the second feature.

[0025] It should be noted that, unless otherwise specified, the embodiments and features described in this invention can be combined with each other.

[0026] Example 1 like Figures 1-3 As shown in the figure, an automatic early warning method for battery compartment fire is disclosed in an embodiment of the present invention, characterized in that it includes: Step S1: During the charging or discharging operation of the battery management system, the terminal voltage value of each battery cell is continuously collected at a preset sampling period to form a voltage timing sequence for each battery cell; the preset sampling period is set to... Engineering standard values This value is determined based on the fact that voltage fluctuations in the early stages of thermal runaway in lithium iron phosphate batteries are mainly concentrated in the 0.1-10Hz frequency band, which can completely capture abnormal signals while also taking into account the BMS calculation load; the total number of sampling points within a single continuous sampling period is Common values (Corresponding to a 50-second sampling duration); Targeting the first [item] inside the battery compartment Each battery cell ( , (This refers to the total number of battery cells; a standard passenger vehicle battery pack typically contains 96-192 cells). Voltage timing sequences were continuously collected. ,in Indicates the first The first battery cell The terminal voltage value at each sampling time, in units of Example: If a commercial vehicle's battery compartment contains 192 lithium iron phosphate battery cells, with a sampling period of 100ms and 500 consecutive samplings, then each cell generates a voltage time sequence of length 500, completely recording the dynamic voltage changes during the battery's charging and discharging process within 50 seconds.

[0027] Step S2: Perform phase space reconstruction on the voltage time series of each battery cell to generate the embedded trajectory matrix. Specifically, this includes: calculating the mutual information function using the mutual information method for the voltage time series of each battery cell. The mutual information function is used to measure the degree of information correlation between data points at different times in time series data. The calculation formula is: In the formula: The delay time is The mutual information value at time, in nats (nat); for The marginal probability density of the voltage value at a given time is the probability that the voltage value falls within a certain interval. For delay Marginal probability density of voltage values ​​at time points; for Time and The joint probability density of the voltage values ​​at time points; the delay value at which the mutual information function first reaches a local minimum is taken as the delay time for phase space reconstruction. The unit is the sampling period. At this point, the data points at the two time points have minimal information redundancy and retain sufficient dynamic correlation; Example: when the sampling period Calculated The time-interaction information is initially minimal, meaning the delay time is determined to be 300ms.

[0028] The embedding dimension is gradually increased starting from a delay time equal to one sampling period. The initial embedding dimension is taken as Calculate the proportion of false nearest neighbors for each embedding dimension. False nearest neighbor points are points that are very close in low-dimensional space but are actually far apart in high-dimensional space. The formula for calculating the ratio is: In the formula: For the embedding dimension is The proportion of false nearest neighbors at any given time; a preset proportion threshold (valued at 2%) is set, and when the proportion first drops below the preset threshold, the embedding dimension is determined as the final embedding dimension. At this point, the high-dimensional chaotic trajectory is fully unfolded, and false nearest neighbor points have essentially disappeared; Example: The proportions of false nearest neighbors were 12%, 7%, 3.5%, and 1.8% respectively, and the first time it fell below the 2% threshold... That is, the embedding dimension is determined to be 5.

[0029] Based on the determined delay time and embedding dimension Each data point in the voltage time series is linked to its subsequent... Data points with multiple delay intervals are combined to form a high-dimensional phase point. The expression for a single high-dimensional phase point is: In the formula: For the first A high-dimensional phase point; For delay The voltage value at each moment; all high-dimensional phase points are arranged vertically in chronological order to form an embedded trajectory matrix. ,in The matrix represents the total number of valid phase points. Each row corresponds to a high-dimensional phase point, and each column corresponds to a voltage value for a delay dimension. Example: When... , , At that time, the total number of effective phase points The embedded trajectory matrix has a dimension of .

[0030] Calculating the Euclidean distance between any two phase points in the embedded trajectory matrix and obtaining the recursive graph matrix by binarization according to the recursive threshold specifically includes: treating each row of the embedded trajectory matrix as an independent phase point. , ( ), and calculate the Euclidean distance between each phase point and all other phase points in turn. The Euclidean distance is used to measure the spatial distance between two high-dimensional phase points in the phase space. The calculation formula is: In the formula: For phase point and The Euclidean distance between them; For phase point The The numerical values ​​of each dimension; traverse all phase pairs to generate Distance matrix diagonal elements , representing the distance between the phase point and itself.

[0031] Calculate the average of all non-zero distances in the distance matrix. To remove items with zero distance on the diagonal to avoid setting the threshold too low, the calculation formula is as follows: In the formula: The average value is the sum of all non-zero distances; a preset ratio of this average value is used as the recursive threshold. Preset proportional coefficient The standard value for engineering projects is 0.2, that is... The adaptive threshold can adapt to the voltage fluctuation range under different battery charging and discharging conditions, avoiding the distortion of recursive features caused by a fixed threshold.

[0032] Elements in the distance matrix that are less than or equal to the recursive threshold are set to 1, representing that the trajectories of the two phase points recursively overlap in phase space, meaning the system returns to a similar state at different times; elements that are greater than the recursive threshold are set to 0, representing that the trajectories of the two phase points do not overlap; finally, the matrix consists of elements with values ​​of 1 and 0. Recursive graph matrix The recursion graph matrix can intuitively reflect the chaotic recursive characteristics of the battery voltage timing signal. Early thermal runaway will cause significant changes in the recursion mode.

[0033] The recursive quantization feature vector, comprising recursion rate, determination rate, laminar flow rate, recursion entropy, and capture time, is extracted from the recursive graph matrix. These five features can quantify the chaotic and anomalous characteristics of battery voltage timing from different dimensions; specifically including: The recurrence rate (RR) is calculated. The recurrence rate characterizes the overall repetition of the time series trajectory and reflects the strength of the regularity of battery voltage fluctuations. The calculation formula is as follows: In the formula: The recursion rate is equal to the number of elements with a value of 1 in the recursion graph matrix divided by the total number of elements in the recursion graph matrix. An abnormally high recursion rate indicates that the battery voltage fluctuation has changed from a random chaotic state to a regular abnormal state, which is a typical characteristic of the early stage of thermal runaway.

[0034] Calculate the Deterministic Estimation Rate (DET), which characterizes the deterministic periodicity of the time series and reflects the strength of the periodicity of abnormal battery voltage fluctuations; find all diagonal segments consisting of consecutive 1s in the recursive graph matrix, and set a preset minimum length threshold. (The standard value for engineering is 3), and the statistical length is greater than or equal to... The total number of 1s on the diagonal segment The certainty rate is obtained by dividing the total number by the total number of 1s in the recursion graph matrix. The calculation formula is as follows: In the formula: For certainty; For length ≥ The total number of 1s on the diagonal segment; the higher the certainty rate, the stronger the periodicity of abnormal fluctuations in battery voltage and the more obvious the fault trend.

[0035] Calculate the laminar flow rate (LAM), which characterizes the steady-state stagnation of the time series and reflects the degree of sustained steady-state abnormal shift in battery voltage. Find all vertical line segments consisting of consecutive 1s in the recursive graph matrix, and similarly... The minimum length threshold is used to count lengths greater than or equal to... The total number of 1s on the vertical line segment Dividing the total number by the total number of 1s in the recursive graph matrix yields the laminar flow rate, calculated using the following formula: In the formula: Laminar flow rate; For length ≥ The total number of 1s on the vertical line segment; an abnormal increase in laminar flow rate indicates an irreversible electrochemical change inside the battery, and a continuous abnormal voltage shift is an important precursor to impending thermal runaway.

[0036] Calculate the recursion entropy (ENT), which characterizes the complexity of temporal recursion features and reflects the degree of chaos in the battery's operating state; statistically analyze the probability of occurrence of diagonal segments of different lengths in the recursion graph matrix, assuming the length is 1. The probability of a diagonal segment appearing is That is, the length is The number of line segments divided by the total length is ≥ The total number of line segments, with the probability of each occurrence taken as a natural constant. The recursive entropy is calculated by multiplying the base logarithm by the corresponding probability of occurrence, summing all the products, and taking the negative value. The formula is as follows: In the formula: For recursive entropy; For length is The probability of the occurrence of the diagonal segment; a sudden change in entropy value indicates a drastic change in the complexity of the battery's operating state, and the emergence of latent faults.

[0037] The capture time (TT) is calculated. The capture time characterizes the average duration of the time-series recursive trajectory, reflecting the duration of the abnormal battery state. The capture time is obtained by multiplying the length of each diagonal segment in the recursive graph matrix by its probability of occurrence, and then summing all the products. The calculation formula is as follows: In the formula: The capture time is the time taken to capture the battery. A longer capture time indicates that the abnormal battery condition lasts longer and the fault is in a continuous deterioration process.

[0038] The recursion rate, determination rate, laminar flow rate, recursion entropy, and capture time are combined sequentially into a five-dimensional vector, which serves as the recursive quantization feature vector for each battery cell. This vector fully represents the first... Chaotic anomaly characteristics of individual cell voltage timing signals.

[0039] Step S3: Organize the recursively quantized feature vectors of all battery cells into a multi-channel feature tensor according to their spatial arrangement within the battery compartment. Input the multi-channel feature tensor into the graph convolutional network model and output the anomaly score for each battery cell. There are strong electrical and spatial thermal coupling relationships between battery cells within the battery compartment. The features of a single cell cannot reflect the overall correlation anomaly. The graph convolutional network can effectively mine the spatial and electrical correlation features between cells. Step S3 specifically includes: Obtain the three-dimensional spatial arrangement coordinates of all battery cells within the battery compartment. The unit is centimeters (cm). Based on the series and parallel electrical connections and spatial adjacency of individual battery cells, an undirected graph structure is constructed, with each battery cell as a node and electrical connections or spatial distances less than a preset threshold as edges. The preset spatial distance threshold is conventionally set to 5cm, meaning that a significant thermal coupling relationship is considered to exist when the center-to-center distance between two cells is less than 5cm. adjacency matrix If node With nodes If there is an electrical connection or the spatial distance is less than a threshold, then ,otherwise Adjacency matrix diagonal elements (Add self-loops to preserve the node's own characteristics).

[0040] The five-dimensional recursive quantization feature vector of each battery cell is used as the initial feature of the corresponding node. The initial features of all nodes are stacked vertically according to the physical arrangement of the nodes in the battery compartment to generate... Two-dimensional feature matrix This matrix is ​​the multi-channel feature tensor of the input graph convolutional network. Each row corresponds to the feature vector of a single battery cell, and each column corresponds to a type of recursive quantization feature.

[0041] Construct a two-layer graph convolutional network (GCN). This network aggregates the features of a node itself and the features of its neighboring nodes layer by layer to mine abnormal features related to individual nodes; and converts the multi-channel feature tensor... and the adjacency matrix of the undirected graph structure Input the two-layer graph convolutional network.

[0042] The first layer of graph convolution aggregates the initial features of each node with the features of its neighboring nodes, outputting the hidden features. The formula for calculating the convolution of the first layer of the graph is: In the formula: To add a self-loop adjacency matrix, for The identity matrix; For degree matrix, That is, the degree of each node; This is the first layer trainable weight matrix, with dimension . ; The ReLU activation function is used to introduce a nonlinear transformation, and its expression is: ; The first layer of hidden features has a dimension of . .

[0043] The hidden features are aggregated again using a second layer of graph convolution, outputting a one-dimensional scalar value for each node. The formula for calculating the convolution of the second layer graph is: In the formula: This is the trainable weight matrix for the second layer, with dimension . ; The second layer output features have a dimension of Each element corresponds to the original abnormal feature value of a single battery cell.

[0044] The one-dimensional scalar value is mapped to the interval between 0 and 1 using the Sigmoid function to obtain the standardized anomaly score for each battery cell. The formula for calculating the Sigmoid function is: In the formula: For the first Abnormal scores for individual battery cells; for The One element; The closer it is to 1, the more it represents the first The higher the probability of anomalies in each individual battery cell; The closer it is to 0, the more normal the unit's operating status.

[0045] Step S4: Select the battery cell with the highest anomaly score as a candidate anomaly cell. Perform a sliding window analysis of the recursion rate on the voltage time series of the candidate anomaly cell to generate a recursion rate change curve. Calculate the first and second derivatives of the recursion rate change curve within the current window. When the first derivative is greater than a positive threshold and the second derivative is greater than an acceleration threshold, mark the candidate anomaly cell as a confirmed anomaly cell. This step eliminates random errors in single feature calculations through time series trend analysis, accurately distinguishing between instantaneous voltage fluctuations and actual fault deterioration trends. Step S4 specifically includes: Step S41: Obtain the original voltage timing sequence of the candidate abnormal unit, and set a sliding window of fixed length, wherein the width of the sliding window is... The standard engineering value is 50 sampling points (corresponding to a 5-second duration), which balances the sensitivity and anti-interference capability of trend recognition; sliding step size. The standard engineering sampling value is 5 sampling points (corresponding to a duration of 0.5 seconds), and the window slides forward from the start of the time sequence to cover voltage data for the entire time period.

[0046] Step S42: Move the sliding window sequentially to each position of the original voltage timing sequence, and truncate a length of [length missing] at each window position. For the voltage subsequence, the phase space reconstruction, recursive graph generation, and feature extraction processes of step S2 are repeated for this subsequence to obtain the recursive quantization feature vector corresponding to the window subsequence. The recursion rate value is then extracted from the recursive quantization feature vector. ,in For window number, , This represents the total number of windows. .

[0047] Step S43: Arrange the recursion rate values ​​calculated for all window positions in chronological order of the window center time to form the original recursion rate change curve. ,in The sampling time corresponding to the center of the window is expressed in seconds (s).

[0048] Step S44: Perform local polynomial smoothing on the recursion rate change curve, specifically using the Savitzky-Golay smoothing algorithm. The smoothing window length is set to 11 window points, and the polynomial order is set to 3. This eliminates curve spikes caused by sampling noise and calculation errors, resulting in a smoothed recursion rate curve. .

[0049] Step S45: On the smoothed recursion rate curve, for the current window position Take the recursion rate value one step before and after the current window position. and The first derivative is calculated using the central difference method. The first derivative characterizes the rate of change of the recursion rate, and the calculation formula is: In the formula: for The first derivative at time t, in units of 1 / second; a positive first derivative indicates a continuously increasing recursion rate and a continuously enhanced abnormal battery characteristic.

[0050] Step S46: Repeat the central difference operation of step S45 on the obtained first derivative sequence to obtain the second derivative sequence. The second derivative characterizes the acceleration of the recursion rate change, reflecting the rate of fault deterioration, and is expressed in units of 1 / second²; a positive threshold is set for the first derivative. (Conventional engineering value: 0.02 / s) and acceleration threshold of the second derivative (The standard value for engineering is 0.005 / second²). When the first derivative of the current window position is greater than the positive threshold and the second derivative of the current window position is greater than the acceleration threshold, it proves that the abnormal characteristics of the battery are not only continuously increasing, but also increasing at an accelerating rate, and the fault is in a state of accelerated deterioration. The candidate abnormal cell is then marked as a confirmed abnormal cell.

[0051] Step S5: Using the confirmed abnormal cell as the center, and combining the recursive quantized feature vectors of adjacent battery cells that have electrical connections and thermal coupling relationships with it, construct a local anomaly propagation network, and calculate the average path length and clustering coefficient of the local anomaly propagation network; battery cell failures will spread rapidly to the surrounding area through electrical conduction and thermal radiation, and the local network topology features can effectively quantify the range and speed of anomaly propagation. Step S5 specifically includes: Using the confirmed abnormal cell as the central node, all battery cells that are directly connected in series or in parallel with the central node are obtained from the electrical topology database of the battery management system and serve as the first-layer neighbor nodes. These nodes have strong electrical coupling with the central node, and abnormal signals will propagate rapidly through electrical loops.

[0052] The flow direction of the cooling medium (such as coolant or air) in the battery compartment thermal management system is obtained. The battery cells located downstream of the central node and separated from the central node by no more than two battery cells are regarded as the second-layer neighbor nodes. These nodes have strong thermal coupling with the central node. The heat generated by the central node will spread downstream with the cooling medium, causing the temperature of the downstream cells to rise and triggering a chain of abnormalities.

[0053] The set of nodes that together constitute the local anomaly propagation network includes the central node, the first-layer neighbor nodes, and the second-layer neighbor nodes. Suppose the set of nodes contains 1 node Typically, there are 5-15, depending on the topology of the battery compartment.

[0054] For each pair of nodes in the node set, the cosine similarity between the recursively quantized feature vectors of the two nodes is calculated. The cosine similarity is used to measure the similarity of the abnormal features of two entities. The higher the similarity, the closer the operating states of the two entities are, and the greater the possibility of anomaly propagation. The calculation formula is as follows: In the formula: For nodes With nodes The cosine similarity between them ranges from [-1, 1]. It is the dot product of two eigenvectors; and These are the magnitudes of the two feature vectors; a preset similarity threshold is set. (The standard value for engineering is 0.85). When the cosine similarity is greater than the preset similarity threshold, an undirected edge is established between the pair of nodes, and the cosine similarity is used as the weight of the edge. The larger the weight, the stronger the abnormal propagation association between the two nodes.

[0055] Calculate the average path length of the local anomaly propagation network. The average path length represents the average distance an anomaly travels in the network; a shorter path length indicates a faster propagation speed. The average path length is equal to the sum of the shortest path lengths between all node pairs divided by the total number of node pairs. The shortest path length between node pairs is defined as the minimum sum of the reciprocals of the weights of each edge on the path. The Dijkstra algorithm is used to calculate the shortest path between any two nodes, and the formula for calculating the average path length is: In the formula: This represents the average path length. For nodes To the node All possible paths; This indicates taking the minimum path length; a decrease in average path length indicates that abnormal connections between nodes in the local network are strengthened, and the speed of abnormal propagation is accelerated.

[0056] Calculate the clustering coefficient of the local anomaly propagation network. The clustering coefficient characterizes the degree of clustering in a local network. A higher clustering coefficient indicates a stronger connection between nodes within the local area, making it easier to form a large-scale fault cluster. For each node... Calculate the actual number of edges between all neighboring nodes of the given node. Divide by the maximum number of edges that may exist between the neighboring nodes. The local clustering coefficients of the nodes are obtained. ,in For nodes The total number of neighbors; then, the arithmetic mean of the local clustering coefficients of all nodes is taken to obtain the clustering coefficient of the local anomaly propagation network. The calculation formula is: In the formula: The global clustering coefficient; For nodes The local clustering coefficient; For nodes The actual number of edges between neighbors; For nodes The number of neighbors; an increase in the clustering coefficient indicates that multiple individuals in a local area are showing similar abnormal characteristics at the same time, and the fault has begun to spread to the surrounding area and form a clustered potential hazard.

[0057] Step S6: When the rate of decrease of the average path length exceeds the path decrease threshold and the rate of increase of the clustering coefficient exceeds the cluster increase threshold, an early warning signal is generated; the changes in the topology parameters of the local anomaly propagation network are monitored in real time within three consecutive sliding windows, and the rate of decrease of the average path length is calculated. and the rate of increase of clustering coefficients The calculation formula is: In the formula: The descent rate is the average path length, expressed in units of 1 / second. The rate of increase of the clustering coefficient is expressed in units of 1 / second. and They are respectively Average path length and clustering coefficient at time step; preset path descent threshold. (The standard engineering value is 0.1 / second) and the clustering rise threshold (The standard engineering value is 0.05 / second); when and At that time, it was confirmed that the abnormality in the battery compartment was spreading rapidly, and a large-scale fault cluster had formed in a local area. The risk of thermal runaway increased sharply. Immediately, a level one fire warning signal for the battery compartment was generated, and at the same time, the emergency power-off of the battery management system, the activation of fire sprinklers, and the audible and visual alarms were triggered. The location of the abnormal individual and the fault data were uploaded to the cloud monitoring platform.

[0058] Example 2 This embodiment proposes an automatic early warning system for battery compartment fires based on Embodiment 1, including: The voltage timing acquisition module continuously acquires the terminal voltage value of each battery cell at a preset sampling period during the charging or discharging operation of the battery management system, forming a voltage timing sequence for each battery cell. The phase space reconstruction and recursive graph generation module performs phase space reconstruction on the voltage time sequence of each battery cell to generate an embedded trajectory matrix, calculates the Euclidean distance between any two phase points in the embedded trajectory matrix, and obtains the recursive graph matrix by binarization according to the recursive threshold. The recursive quantization feature extraction module extracts a recursive quantization feature vector from the recursive graph matrix, which includes recursion rate, determination rate, laminar flow rate, recursion entropy, and capture time. The graph convolutional anomaly detection module is used to organize the recursively quantized feature vectors of all battery cells into a multi-channel feature tensor according to the spatial arrangement position within the battery compartment. The multi-channel feature tensor is then input into the graph convolutional network model to output the anomaly score for each battery cell. The recursion rate evolution analysis module selects the battery cell with the highest anomaly score as a candidate anomaly cell, performs a sliding window analysis of the recursion rate on the voltage time series of the candidate anomaly cell to generate a recursion rate change curve, calculates the first and second derivatives of the recursion rate change curve within the current window, and marks the candidate anomaly cell as a confirmed anomaly cell when the first derivative is greater than a positive threshold and the second derivative is greater than an acceleration threshold. The local anomaly propagation network module, centered on the confirmed abnormal cell, combines the recursive quantized feature vectors of adjacent battery cells that have electrical connections and thermal coupling relationships with it to construct a local anomaly propagation network, and calculates the average path length and clustering coefficient of the local anomaly propagation network. The early warning generation module generates an early warning signal when the rate of decrease of the average path length exceeds the path decrease threshold and the rate of increase of the clustering coefficient exceeds the cluster increase threshold.

[0059] Example 3 This embodiment proposes an electronic device based on Embodiment 1, including a processor, a memory, and a device stored in the memory and capable of being used in various applications. The computer program running on the processor executes an automatic early warning method for battery compartment fire.

[0060] Example 4 This embodiment is based on embodiment 1 and provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it implements an automatic early warning method for battery compartment fire.

[0061] The circuits, electronic components, and modules involved are all existing technologies, which can be fully implemented by those skilled in the art, and need not be elaborated upon. The scope of protection of this invention does not involve any improvement to the software and methods.

[0062] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0063] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An automatic early warning method for battery compartment fire, characterized in that, include: Step S1: During the charging or discharging operation of the battery management system, the terminal voltage value of each battery cell is continuously collected at a preset sampling period to form a voltage time sequence of each battery cell. Step S2: Perform phase space reconstruction on the voltage time sequence of each battery cell to generate an embedded trajectory matrix, calculate the Euclidean distance between any two phase points in the embedded trajectory matrix and obtain a recursive graph matrix by binarization according to the recursive threshold, and extract a recursive quantization feature vector containing recursion rate, determination rate, laminar flow rate, recursion entropy and capture time from the recursive graph matrix. Step S3: Organize the recursively quantized feature vectors of all battery cells into a multi-channel feature tensor according to the spatial arrangement position within the battery compartment. Input the multi-channel feature tensor into the graph convolutional network model and output the anomaly score of each battery cell. Step S4: Select the battery cell with the highest anomaly score as the candidate anomaly cell, perform a sliding window analysis of the recursion rate on the voltage time series of the candidate anomaly cell to generate a recursion rate change curve, calculate the first and second derivatives of the recursion rate change curve in the current window, and mark the candidate anomaly cell as a confirmed anomaly cell when the first derivative is greater than the positive threshold and the second derivative is greater than the acceleration threshold. Step S5: Using the confirmed abnormal cell as the center, and combining the recursive quantized feature vectors of adjacent battery cells that have electrical connection and thermal coupling relationship with it, construct a local anomaly propagation network, and calculate the average path length and clustering coefficient of the local anomaly propagation network. Step S6: When the rate of decrease of the average path length exceeds the path decrease threshold and the rate of increase of the clustering coefficient exceeds the cluster increase threshold, an early warning signal is generated.

2. The automatic early warning method for battery compartment fire according to claim 1, characterized in that, In step S2, the step of performing phase space reconstruction to generate the embedded trajectory matrix specifically includes: For the voltage timing sequence of each battery cell, the mutual information function is calculated using the mutual information method, and the delay value when the mutual information function first appears at a local minimum is taken as the delay time for phase space reconstruction. Starting with a delay time equal to one sampling period, the embedding dimension is gradually increased. The proportion of false nearest neighbors under each embedding dimension is calculated. When the proportion first drops below a preset threshold, the embedding dimension is determined as the final embedding dimension. Based on the determined delay time and embedding dimension, each data point in the voltage time sequence is combined with several subsequent data points with delay intervals to form a high-dimensional phase point. All high-dimensional phase points are arranged in chronological order to form the embedding trajectory matrix.

3. The automatic early warning method for battery compartment fire according to claim 1, characterized in that, In step S2, calculating the Euclidean distance between any two phase points in the embedded trajectory matrix and obtaining the recursive graph matrix by binarization based on the recursive threshold specifically includes: Each row of the embedded trajectory matrix is ​​taken as a phase point, and the Euclidean distance between each phase point and all other phase points is calculated sequentially to obtain a distance matrix; Calculate the average value of all non-zero distances in the distance matrix, and use a preset ratio of the average value as a recursive threshold; The elements in the distance matrix that are less than or equal to the recursion threshold are set to the value 1, and the elements that are greater than the recursion threshold are set to the value 0, thus obtaining a recursion graph matrix composed of values ​​1 and 0.

4. The automatic early warning method for battery compartment fire according to claim 1, characterized in that, In step S2, extracting the recursive quantization feature vector containing recursion rate, determination rate, laminar flow rate, recursion entropy, and capture time from the recursion graph matrix specifically includes: Calculate the recursion rate, which is equal to the number of elements with a value of 1 in the recursion graph matrix divided by the total number of elements in the recursion graph matrix; To calculate the certainty rate, find all diagonal segments consisting of consecutive 1s in the recursive graph matrix, count the total number of 1s on diagonal segments whose length is greater than or equal to a preset minimum length, and divide the total number by the total number of 1s in the recursive graph matrix to obtain the certainty rate. To calculate the laminar flow rate, find all vertical line segments consisting of consecutive values ​​1 in the recursive graph matrix, count the total number of values ​​1 on vertical line segments whose length is greater than or equal to a preset minimum length, and divide the total number by the total number of values ​​1 in the recursive graph matrix to obtain the laminar flow rate. To calculate the recursive entropy, the probability of occurrence of diagonal segments of different lengths in the recursive graph matrix is ​​counted. For each probability of occurrence, the logarithm to the base of the natural constant is taken and multiplied by the corresponding probability of occurrence. Then, all the products are added together and the negative value is taken to obtain the recursive entropy. To calculate the capture time, multiply the length of each diagonal segment in the recursive graph matrix by its probability of occurrence, and then sum all the products to obtain the capture time. The recursion rate, determination rate, laminar flow rate, recursion entropy, and capture time are combined in sequence into a five-dimensional vector, which serves as the recursive quantization feature vector for each battery cell.

5. The automatic early warning method for battery compartment fire according to claim 1, characterized in that, Step S3 specifically includes: Obtain the spatial arrangement coordinates of all battery cells in the battery compartment. Based on the series and parallel electrical connection relationship and spatial adjacency relationship of the battery cells, construct an undirected graph structure with each battery cell as a node and electrical connection or spatial distance less than a preset threshold as edges. The recursive quantization feature vector of each battery cell is used as the initial feature of the corresponding node. The initial features of all nodes are stacked into a two-dimensional matrix according to the arrangement order of the nodes in the battery compartment. The two-dimensional matrix is ​​used as a multi-channel feature tensor. Construct a two-layer graph convolutional network and input the multi-channel feature tensor and the adjacency matrix of the undirected graph structure into the two-layer graph convolutional network; The initial features of each node are aggregated with the features of its neighboring nodes through the first layer of graph convolution, and the hidden features are output. The hidden features are aggregated again by a second layer of graph convolution, and a one-dimensional scalar value for each node is output. The one-dimensional scalar value is mapped to the interval between 0 and 1 using the Sigmoid function to obtain the anomaly score for each battery cell.

6. The automatic early warning method for battery compartment fire according to claim 1, characterized in that, Step S4 specifically includes: Step S41: Obtain the original voltage timing sequence of the candidate abnormal unit, set a sliding window of fixed length, the width of the sliding window is a preset window width range, and the sliding step size is a preset step size; Step S42: Move the sliding window sequentially to each position of the original voltage time series, repeat step S2 for the voltage subsequence within the window at each window position, obtain the recursive quantization feature vector corresponding to the window subsequence, and extract the recursion rate from the recursive quantization feature vector; Step S43: Arrange the recursion rate values ​​calculated for all window positions in chronological order of the window center time to form a recursion rate change curve; Step S44: Perform local polynomial smoothing on the recursion rate change curve to obtain a smoothed recursion rate curve; Step S45: On the smoothed recursion rate curve, for the current window position, take the recursion rate value one step before and after the current window position, calculate the difference between the previous recursion rate value and the next recursion rate value, divide it by twice the step size, and use the calculation result as the first derivative of the current window position. Step S46: Repeat the operation of step S45 on the obtained first derivative sequence to obtain the second derivative sequence. Set the positive threshold of the first derivative and the acceleration threshold of the second derivative. When the first derivative at the current window position is greater than the positive threshold and the second derivative at the current window position is greater than the acceleration threshold, mark the candidate abnormal entity as a confirmed abnormal entity.

7. The automatic early warning method for battery compartment fire according to claim 1, characterized in that, Step S5 specifically includes: Using the confirmed abnormal cell as the central node, all battery cells that are directly connected in series or in parallel with the central node are obtained from the electrical topology of the battery management system and used as the first-layer neighbor nodes. Obtain the flow direction of the cooling medium in the battery compartment thermal management system, and designate the battery cells located downstream of the central node and separated from the central node by no more than two battery cells as the second-layer neighbor nodes. The central node, the first-layer neighbor nodes, and the second-layer neighbor nodes together constitute the node set of the local anomaly propagation network; For each pair of nodes in the node set, calculate the cosine similarity between the recursively quantized feature vectors of the two nodes. When the cosine similarity is greater than a preset similarity threshold, establish an edge between the pair of nodes and use the cosine similarity as the weight of the edge. Calculate the average path length of the local anomaly propagation network. The average path length is equal to the sum of the shortest path lengths between all node pairs divided by the total number of node pairs. The shortest path length between node pairs is defined as the minimum of the sum of the reciprocals of the weights of each edge on the path. To calculate the clustering coefficient of the local anomaly propagation network, for each node, the actual number of edges between all neighboring nodes is divided by the maximum possible number of edges between the neighboring nodes to obtain the local clustering coefficient of the node. Then, the arithmetic mean of the local clustering coefficients of all nodes is taken to obtain the clustering coefficient of the local anomaly propagation network.

8. An automatic early warning system for battery compartment fire, used to implement the automatic early warning method for battery compartment fire as described in claims 1-7, characterized in that, include: The voltage timing acquisition module continuously acquires the terminal voltage value of each battery cell at a preset sampling period during the charging or discharging operation of the battery management system, forming a voltage timing sequence for each battery cell. The phase space reconstruction and recursive graph generation module performs phase space reconstruction on the voltage time sequence of each battery cell to generate an embedded trajectory matrix, calculates the Euclidean distance between any two phase points in the embedded trajectory matrix, and obtains the recursive graph matrix by binarization according to the recursive threshold. The recursive quantization feature extraction module extracts a recursive quantization feature vector from the recursive graph matrix, which includes recursion rate, determination rate, laminar flow rate, recursion entropy, and capture time. The graph convolutional anomaly detection module is used to organize the recursively quantized feature vectors of all battery cells into a multi-channel feature tensor according to the spatial arrangement position within the battery compartment. The multi-channel feature tensor is then input into the graph convolutional network model to output the anomaly score for each battery cell. The recursion rate evolution analysis module selects the battery cell with the highest anomaly score as a candidate anomaly cell, performs a sliding window analysis of the recursion rate on the voltage time series of the candidate anomaly cell to generate a recursion rate change curve, calculates the first and second derivatives of the recursion rate change curve within the current window, and marks the candidate anomaly cell as a confirmed anomaly cell when the first derivative is greater than a positive threshold and the second derivative is greater than an acceleration threshold. The local anomaly propagation network module, centered on the confirmed abnormal cell, combines the recursive quantized feature vectors of adjacent battery cells that have electrical connections and thermal coupling relationships with it to construct a local anomaly propagation network, and calculates the average path length and clustering coefficient of the local anomaly propagation network. The early warning generation module generates an early warning signal when the rate of decrease of the average path length exceeds the path decrease threshold and the rate of increase of the clustering coefficient exceeds the cluster increase threshold.

9. An electronic device, characterized in that, Includes a processor, a memory, and components stored in the memory that can be used in the system. The processor runs a computer program, which, when executed, implements an automatic early warning method for battery compartment fire as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements an automatic early warning method for battery compartment fire as described in any one of claims 1-7.

Images