Electric vehicle charging fire early warning method based on multi-sensor fusion

Abstract

The invention discloses an electric vehicle charging fire early warning method based on multi-sensor fusion, and belongs to the technical field of fire early warning, and the method comprises the steps: obtaining the real-time monitoring data of a charging pile based on a preset sensor; generating a multi-dimensional primary feature vector based on the real-time monitoring data; analyzing the multi-dimensional primary feature vector to generate a comprehensive correlation index; scanning the battery management system to generate internal state data; generating a collaborative verification result based on the comprehensive association degree index and the internal state data; and generating a hierarchical early warning decision based on the collaborative verification result. And a multi-dimensional feature collaborative analysis and internal and external state verification mechanism is adopted, so that early-stage hidden danger accurate identification and graded emergency disposal can be realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of fire warning, in particular to an electric vehicle charging fire warning method based on multi-sensor fusion. BACKGROUND

[0002] The electric vehicle charging fire warning technology belongs to the field of safety monitoring, and is mainly aimed at the fire risk caused by thermal runaway during the charging process of lithium ion batteries. The technology realizes early identification and intervention of fire hazards by monitoring charging pile operating parameters and battery status.

[0003] The prior art usually uses a single type of sensor for monitoring, such as relying on temperature threshold to trigger an alarm, or generating a fault signal through the voltage and current abnormality detection mechanism built in the battery management system. Some solutions realize composite detection by combining temperature and smoke sensors.

[0004] The prior art has three defects. First, a single sensor is prone to false positives due to environmental interference, and the response to early slow-changing thermal runaway is lagging. Second, external sensors and battery management system data are analyzed in isolation, and cannot identify hidden risks where the internal and external states are inconsistent. Third, the fixed threshold alarm mechanism is difficult to adapt to different evolution speeds of fire scenes, and is prone to miss reports or delayed responses. SUMMARY

[0005] To solve the above problems, the present application provides an electric vehicle charging fire warning method based on multi-sensor fusion, which uses multi-dimensional feature collaborative analysis and internal and external state verification mechanism to realize early hidden danger accurate identification and graded emergency disposal.

[0006] The above object can be achieved by the following scheme: An electric vehicle charging fire warning method based on multi-sensor fusion, comprising: acquiring real-time monitoring data of a charging pile based on a preset sensor; generating a multi-dimensional primary feature vector based on the real-time monitoring data; analyzing the multi-dimensional primary feature vector to generate a comprehensive correlation index; scanning a battery management system to generate internal state data; generating a collaborative verification result based on the comprehensive correlation index and the internal state data; and generating a graded warning decision based on the collaborative verification result.

[0007] Optionally, the generation of a multidimensional primary feature vector includes: identifying key point temperature values ​​and maximum temperature differences in preset temperature sensor array data, calculating the maximum temperature rise rate, and generating a temperature feature vector; calculating the concentration rise rate and concentration ratio of the electrolyte decomposition characteristic gas in the preset composite gas sensor data to generate a gas feature vector; calculating the current effective value fluctuation amplitude and harmonic content change rate and charging efficiency decrease rate in the current sensor data to generate an electrical feature vector; and fusing the temperature feature vector, gas feature vector, and electrical feature vector to generate a multidimensional primary feature vector.

[0008] Optionally, generating a comprehensive correlation index includes: generating a time correlation factor based on the time series correlation of each vector in the multi-dimensional primary feature vector; analyzing the spatial propagation characteristics of sensor data at different locations to generate a spatial correlation factor; calculating the statistical relationship between feature vectors of different dimensions to generate a feature cross-correlation factor; and fusing the time correlation factor, spatial correlation factor and feature cross-correlation factor to generate a comprehensive correlation index.

[0009] Optionally, generating the characteristic cross-correlation factor includes: establishing a dynamic ratio model based on the maximum temperature rise rate and the characteristic gas concentration rise rate; correlating the current fluctuation spectrum characteristics with the local temperature difference to generate a spatial distribution relationship; when the dynamic ratio model and the spatial distribution relationship conform to the preset battery thermal runaway evolution pattern, generating the characteristic cross-correlation factor.

[0010] Optionally, generating the collaborative verification result includes: comparing the comprehensive correlation index with the internal state data to generate an abnormality factor; and generating the collaborative verification result based on the abnormality factor through preset association rules.

[0011] Optionally, the graded warning decision includes: when the comprehensive correlation index exceeds a first preset threshold and the internal status data does not trigger a fault alarm, executing a primary warning operation; when the comprehensive correlation index exceeds a second preset threshold, or the comprehensive correlation index continues to rise after the primary warning, or the internal status data is associated with an abnormal mode of an external sensor, executing an advanced alarm operation.

[0012] Optionally, the advanced alarm operation includes: triggering a preset strong sound and light alarm device; sending an emergency alarm signal to a remote monitoring platform; and controlling the charging pile to start a fire extinguishing device and cut off the charging circuit.

[0013] Optionally, the method also includes: recording the multi-dimensional primary feature vectors, comprehensive correlation index and decision results in historical warning events to generate a historical warning matrix; based on the historical warning matrix, optimizing the parameter weights for analyzing the multi-dimensional primary feature vectors and the thresholds for graded warning decisions.

[0014] Optionally, the optimization operation comprises: extracting interference patterns of feature vectors in false alarm events, and reducing the correlation weight of similar interference patterns; extracting weak correlation patterns in missed alarm events, and enhancing the sensitivity of specific feature cross factors.

[0015] Based on the same inventive concept, the application also provides a multi-sensor fusion-based electric vehicle charging fire early warning system, comprising: a multi-source data acquisition module: obtaining real-time monitoring data of a charging pile based on a preset sensor; a feature engineering processing module: generating a multi-dimensional primary feature vector based on the real-time monitoring data; an association degree analysis module: analyzing the multi-dimensional primary feature vector to generate a comprehensive association degree index; a BMS state scanning module: scanning a battery management system to generate internal state data; a state cooperative verification module: generating a cooperative verification result based on the comprehensive association degree index and the internal state data; a hierarchical early warning decision module: generating a hierarchical early warning decision based on the cooperative verification result.

[0016] Compared with the prior art, the application has the following advantages: 1. By fusing temperature, gas, and current multi-source sensor data to construct a multi-dimensional feature vector, the collaborative perception of the abnormal changes of multiple physical quantities in the early stage of thermal runaway is realized, and the recognition sensitivity of early fire hazards is significantly improved.

[0017] 2. A dual verification mechanism combining external sensor data and internal state of the battery management system is used to effectively identify hidden faults that are not triggered by the BMS, thereby reducing the risk of false positives caused by failure of a single data source.

[0018] 3. Based on a dynamic analysis model of time, space, and feature cross correlation, the evolution law of battery thermal runaway is accurately captured, and the anti-interference ability in complex charging environments is enhanced.

[0019] 4. A hierarchical early warning decision and self-optimization mechanism are used to match differentiated emergency response measures according to risk levels, optimize resource allocation, and avoid over-response.

[0020] Other features and advantages of the application will be set forth in the following description of the application, and in part will become apparent to those skilled in the art from the description, or can be learned by practice of the application. The objects and other advantages of the application will be realized and attained by the structure particularly pointed out in the written description and claims thereof as well as the appended drawings. BRIEF DESCRIPTION OF DRAWINGS

[0021] In order to more clearly illustrate the technical solutions in the embodiments of the application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiment or prior art description. Obviously, the drawings in the following description are some embodiments of the application, and for those skilled in the art, other drawings can also be obtained without creative labor.

[0022] Figure 1 is a flowchart of a multi-sensor fusion-based electric vehicle charging fire early warning method of an embodiment of the present application.

[0023] Figure 2 is a battery thermal runaway evolution mode identification schematic diagram of an embodiment of the present application.

[0024] Figure 3 is a three-dimensional feature space risk probability cloud chart schematic diagram of an embodiment of the present application.

[0025] Figure 4 is a system structure schematic diagram of a multi-sensor fusion-based electric vehicle charging fire early warning method of an embodiment of the present application. DETAILED DESCRIPTION

[0026] To make the objectives, technical solutions, and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described below in detail with reference to the accompanying drawings of the embodiments of the present application. Obviously, the described embodiments are some, but not all, of the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of protection of the present application.

[0027] With reference to Figure 1 , one embodiment of the present application proposes a multi-sensor fusion-based electric vehicle charging fire early warning method, which adopts multi-dimensional feature collaborative analysis and internal and external state verification mechanism, and can realize early hidden danger accurate identification and hierarchical emergency disposal.

[0028] The method of the embodiment specifically includes: acquiring real-time monitoring data of a charging pile based on a preset sensor; generating a multi-dimensional primary feature vector based on the real-time monitoring data; analyzing the multi-dimensional primary feature vector to generate a comprehensive correlation degree index; scanning a battery management system to generate internal state data; generating a collaborative verification result based on the comprehensive correlation degree index and the internal state data; generating a hierarchical early warning decision based on the collaborative verification result.

[0029] Specifically, the application is based on multi-sensor fusion technology, generates multi-dimensional primary feature vectors by real-time collection of charging pile multi-source monitoring data, constructs a comprehensive correlation index through comprehensive time, space and feature cross correlation analysis, realizes internal and external collaborative verification by synchronous scanning of internal state data of the battery management system, and finally dynamically triggers a hierarchical early warning decision according to the verification result. This method constructs a closed-loop early warning system from feature extraction, correlation analysis to collaborative decision-making through the dual verification mechanism of external multi-sensor data and internal BMS state. It can significantly improve the accuracy and timeliness of electric vehicle charging fire warning, effectively overcome the false alarm and missed alarm defects of a single sensor. Through internal and external data collaborative verification, the recognition ability of early thermal runaway signals is enhanced, the hierarchical early warning mechanism optimizes the allocation of emergency disposal resources, and the risk of misoperation is reduced. At the same time, multi-dimensional feature fusion analysis enhances the anti-interference ability in complex environments, and provides comprehensive and reliable technical support for charging safety.

[0030] Optionally, the generating multi-dimensional primary feature vectors comprises: identifying key point temperature values in preset temperature sensor array data, maximum temperature difference, calculating the highest temperature rise rate, and generating a temperature feature vector; calculating the concentration rise rate and concentration ratio of electrolyte decomposition characteristic gas in the preset composite gas sensor data, and generating a gas feature vector; calculating the current effective value fluctuation amplitude and harmonic content change rate and charging efficiency decline rate in the current sensor data, and generating an electrical feature vector; and fusing the temperature feature vector, the gas feature vector and the electrical feature vector to generate a multi-dimensional primary feature vector.

[0031] Specifically, the key point temperature values in the preset temperature sensor array data are the highest value in the real-time monitoring temperature and the maximum temperature difference between the highest temperature and the lowest temperature at the same time point in different position sensor readings. The highest temperature rise rate is calculated as: , Wherein, is the maximum value of the temperature difference between consecutive sampling points, obtained by comparing the temperature difference values of adjacent time points; and Δt is the corresponding sampling time interval determined by the sensor sampling frequency. The temperature feature vector is generated by fusing the key point temperature values, the maximum temperature difference and the highest temperature rise rate: , Wherein is the key point temperature value, is the maximum temperature difference. The preset composite gas sensor data is processed to identify the electrolyte decomposition characteristic gas. The concentration rise rate of the characteristic gas is calculated as: , Wherein, is the current sampling concentration value, is the concentration value of the previous sampling, Δt is the sampling interval, and the concentration ratio is directly calculated by directly dividing the real-time concentration values of the two characteristic gases. The concentration rise rate of the key gas and the concentration ratio are fused to generate a gas feature vector: , wherein, , The concentration rise rate of the characteristic gases hydrogen and carbon monoxide. is the concentration ratio of the characteristic gases hydrogen and carbon monoxide. The current sensor data is analyzed, the standard deviation of the effective value of the current in the set time window is calculated to obtain the fluctuation amplitude of the effective value of the current, the fundamental wave and the main harmonic component are extracted by fast Fourier transform, and the change rate of the current harmonic distortion rate and the previous sampling distortion rate is calculated: , wherein, is the current harmonic distortion rate, is the previous total harmonic distortion rate. The charging efficiency decline rate is calculated: , wherein, is the input power of the charging pile, is the received power of the battery. The fluctuation amplitude of the effective value of the current, the harmonic content change rate and the charging efficiency decline rate are fused to form an electrical feature vector: , wherein, is the fluctuation amplitude of the effective value of the current. The temperature feature vector, the gas feature vector and the electrical feature vector are spliced in the order of dimensions to generate a multi-dimensional primary feature vector.

[0032] Exemplarily, during the charging process of a certain charging pile, the temperature sensor array detects that the maximum temperature is 52.3℃, the maximum temperature difference on the upper surface of the battery pack at the same time is 8.7℃, and the maximum temperature rise rate within 10 seconds is 1.5℃ / s. The composite gas sensor measures that the H2 concentration rises from 120ppm to 185ppm within 5 seconds, so the hydrogen concentration rise rate is 13ppm / s, the CO concentration rises from 45ppm to 60ppm, so the carbon monoxide concentration rise rate is 3ppm / s, and the ratio of hydrogen concentration to carbon monoxide concentration is 3.08. The current sensor monitors that the effective value fluctuation amplitude is 5% of the rated current, the harmonic distortion rate rises from the previous 2.1% to the current 3.0%, and the charging efficiency decreases from 95% to 89%. The system generates a temperature feature vector [52.3, 8.7, 1.5], a gas feature vector [13, 3, 3.08], and an electrical feature vector [5, 42.9, 6.3] according to this, and finally fuses into a multi-dimensional primary feature vector [52.3, 8.7, 1.5, 13, 3, 3.08, 5, 42.9, 6.3]. The beneficial effect of this verification example is that by quantifying the abnormal dynamic changes of the three dimensions of heat, gas and electricity (such as rapid temperature rise, sudden increase of characteristic gas, and sudden drop of efficiency) at the same time, the recognition sensitivity of early fire hazards is significantly improved, the shortcomings of single sensor being easily disturbed by the environment or having a lag in response are overcome, high information density basic data are provided for subsequent comprehensive correlation analysis, and the risk of early warning failure caused by sensor false alarm or missed alarm is effectively reduced.

[0033] Optionally, the generating a comprehensive correlation degree index comprises: generating a time correlation factor based on the time sequence correlation of each vector in the multi-dimensional primary feature vector; analyzing the spatial propagation characteristics of sensor data at different positions to generate a spatial correlation factor; calculating the statistical relationship of different dimension feature vectors to generate a feature cross-correlation factor; and fusing the time correlation factor, the spatial correlation factor and the feature cross-correlation factor to generate a comprehensive correlation degree index.

[0034] Specifically, based on the time sequence characteristics of each dimension data in the multi-dimensional primary feature vector, the fluctuation synchronization of each sensor data on the time axis is calculated, the dynamic coincidence degree of the temperature change trend and the gas concentration change trend, and the time matching degree of the current harmonic pre-response characteristics are calculated, the covariance is calculated by sliding time window, and the time correlation factor is obtained; by grid positioning of sensor data in different physical positions, the consistency of the spatial propagation path of temperature abnormal points and the diffusion direction of gas concentration gradient at the same time node is analyzed, the diffusion rate convergence characteristics of local abnormal phenomena are calculated by vector field divergence, and the spatial correlation factor is obtained; the statistical relationship matrix of the feature vector is constructed, the dynamic fitting curve of the temperature rise rate and the gas concentration ratio is established, the spatial mapping model of the main frequency band shift of the current harmonic spectrum and the local temperature difference region is formed, when the curvature change of the dynamic fitting curve is consistent with the battery electrolyte decomposition experimental data, and the heat flow distribution coefficient of the spatial mapping model exceeds the critical value, the feature cross correlation factor is generated; the time correlation factor, the spatial correlation factor and the feature cross correlation factor are linearly superimposed according to the set weight coefficient and input into the standardization function to generate the comprehensive correlation degree index I: , Where FT is the time correlation factor, FS is the spatial correlation factor, FC is the feature cross correlation factor, α represents the time dimension weight, β represents the spatial dimension weight, and γ represents the cross feature weight. Each weight coefficient is determined by multi-objective optimization of historical accident data.

[0035] For example, in an electric vehicle fast charging station equipment, when a temperature sensor detects that the temperature of a certain charging gun connection point rises from 25°C to 48°C in 3 minutes, a composite gas sensor at the adjacent position measures that the electrolyte decomposition characteristic gas concentration ratio CO / H2 reaches 1:0.8, and a current sensor detects that the three harmonic component ratio exceeds 15%. At this time, the system first analyzes the covariance of the temperature rise curve and the H2 concentration curve, which reaches 0.92, and determines that the time correlation factor is 82; secondly, through the position distribution of abnormal points, it is found that the diffusion path of the high temperature region along the cable sheath to the charging cabinet direction is consistent with the gas concentration gradient direction, and the spatial correlation factor is 75; finally, through the comparison of the temperature-gas feature cross model, it is found that the similarity of this mode to the laboratory sample of battery separator melting reaches 87%, and the feature cross correlation factor is 89. With the weight setting of α=0.4, β=0.3, γ=0.3, the comprehensive correlation degree index is calculated as 81.4.

[0036] Optionally, the generating the feature cross correlation factor comprises: establishing a dynamic ratio model based on the highest temperature rise rate and the characteristic gas concentration rise rate; correlating the current fluctuation frequency spectrum characteristics and the local temperature difference to generate a spatial distribution relationship; when the dynamic ratio model and the spatial distribution relationship conform to the preset battery thermal runaway evolution mode, the feature cross correlation factor is generated.

[0037] Specifically, the highest temperature rise rate in the temperature feature vector and the concentration rise rate of the key feature gas in the gas feature vector are extracted, and a dynamic ratio model is established: , wherein the highest temperature rise rate is calculated by temperature sensor data, the concentration rise rate is calculated by gas sensor data. The current harmonic content change rate is extracted from the electrical feature vector, and the energy proportion of the main harmonic number is obtained by fast Fourier transform; the local temperature difference of different regions of the battery pack is extracted from the temperature feature vector; and the normalized correlation coefficient S of the harmonic energy proportion and the local temperature difference is calculated: , wherein k is the number of spatial partitions, is the harmonic energy proportion of the i-th partition, is the temperature difference of the corresponding partition, and is the harmonic energy proportion and the average temperature difference. When the dynamic ratio falls within the preset threshold range or the spatial distribution correlation coefficient is greater than the preset threshold, a feature cross correlation factor is generated, as shown in Fig. 1 is a schematic diagram of battery thermal runaway evolution mode recognition. Figure 2

[0038] For example, when a battery pack is charging, the highest temperature rise rate is detected to be 1.8℃ / s, the H2 concentration rise rate is 15ppm / s, and the dynamic ratio R=0.12 is calculated. At the same time, the current spectrum analysis shows that the 3rd harmonic energy proportion in three partitions is 0.22, 0.18, and 0.25, respectively, and the corresponding regional temperature difference is 7.2℃, 5.8℃, and 8.5℃, respectively, and the normalized correlation coefficient S=0.85 is calculated. The system determines that R=0.12 is within the preset threshold range of 0.08-0.15, and S=0.85>0.7, which meets the battery thermal runaway evolution mode, and therefore generates a feature cross correlation factor . The beneficial effect of this verification example is that by strictly matching the dynamic proportional relationship and spatial coupling law between physical quantities, the key features of the cooperative evolution of battery internal electrolyte decomposition and local overheating can be accurately identified, external environmental interference such as charger harmonic pollution or sensor drift can be significantly excluded to cause false triggering, and it is ensured that the feature cross correlation factor is only activated when the real thermal runaway initial feature appears, which greatly improves the reliability of subsequent comprehensive correlation analysis.

[0039] Optionally, the generating a cooperative verification result comprises: comparing the comprehensive correlation degree index with the internal state data to generate an abnormal factor; and generating a cooperative verification result based on the abnormal factor through a preset correlation rule.

[0040] ​Specifically, the comprehensive correlation index and the battery management system internal state data are obtained, the latter including the single cell voltage difference, the insulation resistance value and the maximum temperature value. The abnormal factor D of the external index and the internal data is calculated: , wherein the fusion function of the internal state data has: , , , is a preset weight sum of 1, is the maximum allowable voltage difference, is the minimum allowable insulation resistance, is the ambient temperature, is the upper limit of the battery temperature, all from the battery technical specification, index is the comprehensive correlation index, is the single cell voltage difference, is the insulation resistance value, is the maximum temperature value. Based on the abnormal factor, a cooperative verification result is generated through a preset correlation rule, wherein if the abnormal factor and the comprehensive correlation index are greater than a preset threshold, an external abnormality and an internal undetected cooperative verification result is generated; if the abnormal factor is less than the preset threshold and the maximum temperature value is greater than the preset threshold, an internal abnormality and an external undetected cooperative verification result is generated; if the comprehensive correlation index and the single cell voltage difference are greater than the preset threshold and the insulation resistance value is less than the preset threshold, an internal and external cooperative abnormality cooperative verification result is generated.

[0041] Exemplarily, in a certain charging, the comprehensive correlation index Index=0.82, the internal state data is the single cell voltage difference 0.15V, the maximum allowable voltage difference 0.2V, the insulation resistance value 1800Ω, the minimum allowable insulation resistance 1500Ω, the maximum temperature value 48℃, the ambient temperature 25℃, and the upper limit of the battery temperature 60℃. Assuming that the weights are 0.4, 0.3 and 0.3 respectively, the difference D is calculated. The difference D=0.088. The preset thresholds α=0.15 and β=0.7, because D=0.088<α but Index=0.82>β, an external abnormality and an internal undetected cooperative verification result is generated. The beneficial effect of this verification example is that by quantifying the difference degree of the external multi-sensor fusion index and the internal BMS data, and based on the battery failure mechanism, a multi-level judgment rule is preset, the early thermal runaway risk that may be missed by the BMS system is effectively identified, and false judgment caused by single data source failure is avoided, thereby providing a double protection mechanism for hierarchical early warning.

[0042] Optionally, the hierarchical early warning decision comprises: when the comprehensive correlation index exceeds a first preset threshold and the internal state data does not trigger a fault alarm, performing a primary early warning operation; when the comprehensive correlation index exceeds a second preset threshold, or the comprehensive correlation index continues to rise after the primary early warning, or the internal state data is associated with an external sensor abnormal mode, performing a high-level alarm operation.

[0043] Specifically, preset first and second thresholds, wherein the second threshold is greater than the first threshold and both are in the range of 0 to 1 and are calibrated by historical fault data. Real-time comprehensive correlation index and battery management system internal state data are obtained, the latter including a fault alarm flag and specific parameters. The condition for performing a primary early warning operation is that the comprehensive correlation index is greater than the first threshold and the fault alarm flag is not triggered. If the comprehensive correlation index is greater than the second threshold or the comprehensive correlation index continues to rise in the continuous sampling period after the primary early warning, or the internal state data is associated with an external sensor abnormal mode, a high-level alarm operation is performed. The rising determination formula is , Where t is the current time; the correlation condition specifically includes that when the single cell voltage difference exceeds the threshold, the concentration ratio of the gas feature vector is lower than the preset threshold, or when the insulation resistance is lower than the threshold, the charging efficiency of the electric feature vector decreases by more than the preset threshold.

[0044] Exemplarily, in a certain charging process, β1=0.65 and β2=0.85 are preset. The initial Index=0.70, the BMS has no fault alarm, and the primary early warning is triggered. In the subsequent 3 continuous sampling periods, the Index values are 0.73, 0.78 and 0.82 respectively, which meet the continuous rising condition, and are automatically upgraded to high-level alarm. In another scenario, Index=0.87 is greater than β2, which directly triggers high-level alarm. In the third scenario, Index=0.75 does not reach β2, but the BMS detects that the single cell voltage difference is 0.18V, the threshold is 0.15V, and the real-time concentration ratio is 2.1, the threshold is 2.5, which meets the internal abnormality and external gas feature correlation condition, and also triggers high-level alarm. The beneficial effect of this verification example is that through the multi-path early warning upgrade mechanism, both gradual development risks and sudden composite abnormalities can be responded to gradually and immediately, respectively, to ensure that appropriate emergency measures can be matched in time in different fire risk scenarios, and the risk of early warning lag or over-response can be effectively avoided.

[0045] Optionally, the high-level alarm operation comprises: triggering a preset strong sound and light alarm device; sending an emergency alarm signal to a remote monitoring platform; controlling the charging pile to perform a fire extinguishing device start and charging circuit cut-off operation.

[0046] Specifically, when the preset strong sound and light alarm device is triggered, a pulse width modulation signal is sent to the alarm controller, and the pulse duty cycle is obtained by linear mapping of the comprehensive correlation index: , wherein is the minimum duty cycle 20% corresponding to the minimum brightness of the alarm device, is the maximum duty cycle 90% corresponding to the maximum brightness, is the second preset threshold, and Index is the comprehensive correlation index. The sound intensity level is calculated according to the exponential growth model: , wherein is the reference sound intensity, and c is the adjustment coefficient calibrated by acoustic characteristics. When sending an emergency alarm signal to the remote monitoring platform, a protocol data unit containing a timestamp, a charging pile number, a comprehensive correlation index, and an internal state data digest is packaged, and the data digest compression algorithm is: , wherein Hash adopts SHA-256 algorithm to generate a 256-bit digest. The transmission protocol sets a retransmission mechanism, and when the confirmation signal is not returned within the preset time, it is retransmitted according to the exponential backoff algorithm. When controlling the charging pile to execute the fire extinguishing device start and charging circuit cut-off operation, the strict time sequence of the fire extinguishing agent release instruction and the circuit breaker opening instruction is generated: first cut off the charging circuit and then start the fire extinguishing device, to ensure that the electric arc is completely extinguished. Greater than the extinguishing device start duration T, there is: , wherein, is the basic injection time, and k is the expansion coefficient. Exemplarily, when the system triggers an advanced alarm, Index = 0.88, and β2 = 0.85. The sound and light alarm duty cycle K = 34%, and the sound intensity level L = 81.2dB. The remote alarm signal encapsulates the current time, the pile number P123, the Index value, and the internal state digest H, wherein H is generated by hashing after splicing the single cell voltage range, the insulation resistance value, and the maximum temperature value. The fire extinguishing control first sends the charging circuit cut-off instruction, and then starts the fire extinguishing device after 500 milliseconds for a duration T = 3.15 seconds. The beneficial effect of this verification example is to match the risk level by dynamically adjusting the alarm intensity, to guarantee transmission efficiency by using data digest technology, and to avoid secondary arc accidents caused by the fire extinguishing process by strict time sequence control, forming a complete safety closed loop from on-site warning to remote response and then to physical intervention.

[0047] Optionally, the method further comprises: recording the multi-dimensional primary feature vector, the comprehensive correlation index and the decision result in a historical early warning event, to generate a historical early warning matrix; and optimizing the parameter weight for analyzing the multi-dimensional primary feature vector and the threshold value for the graded early warning decision based on the historical early warning matrix.

[0048] Specifically, the historical early warning event is recorded, and each event generates a record containing a multi-dimensional primary feature vector, a comprehensive correlation index and a decision result label, to generate a historical early warning matrix: wherein is the multi-dimensional primary feature vector, is the comprehensive correlation index, is the decision result label, and m is the number of events. When the parameter weight is optimized, the time, space and cross-correlation factor weight coefficients of the correlation analysis module are adjusted adaptively by using a self-adaptive adjustment algorithm: wherein T is a set of missed events, F is a set of false events, are the i and j factor values of the k event, and η is a preset learning rate. The updated weight is normalized. When the graded threshold value is optimized, the distribution quantile point of the comprehensive correlation index of the specified label event is calculated by using a quantile function, and the primary early warning threshold value and the senior alarm threshold value are adjusted, as shown in FIG. 5. Figure 3

[0049] Exemplarily, the system accumulates 100 early warning events to form a historical matrix. It is found by analysis that 12 false alarms all have the characteristics of sudden surge of current harmonics but gentle temperature rise, which are stored in an interference mode dictionary. Five missed alarms all show a temperature rise rate of 1.2-1.5 ℃ / s and H2 / CO = 2.3-2.6. After optimization, the weight is adjusted from [0.4, 0.3, 0.3] to [0.35, 0.32, 0.33], the primary early warning threshold value is increased from 0.65 to 0.68, and the senior alarm threshold value is decreased from 0.85 to 0.82. When the similarity of the characteristics of a new event to the interference mode dictionary is 0.85, the system automatically reduces the weight of the event by 30%. The beneficial effect of this verification example lies in that the parameter sensitivity and the decision boundary are adjusted in a targeted manner through dynamic learning of the historical false alarm and missed alarm characteristic modes, the immunity of the system to new interference and the capture ability of the system to hidden risks are significantly improved, and a continuously evolving safety protection mechanism is formed.

[0050] Optionally, the optimization operation comprises: extracting an interference mode of a feature vector in a false alarm event, and reducing the correlation weight of a similar interference mode; and extracting a weak correlation mode in a missed alarm event, and enhancing the sensitivity of a specific feature cross factor.

[0051] ​​​​Specifically, for false alarm events, the multi-dimensional primary feature vector thereof is extracted, and a disturbance mode dictionary is generated through a mean clustering algorithm. The clustering process calculates the standardized Euclidean distance between the feature vectors: , wherein and are the i-th dimensional feature values of two vectors, is the standard deviation of the historical data in this dimension. After clustering, each cluster center vector is stored in the disturbance mode dictionary as a typical disturbance mode. When the cosine similarity between the feature vector of a new event and a mode in the disturbance mode dictionary exceeds a preset threshold, a preset weight attenuation coefficient is applied to the event when calculating the comprehensive correlation index. For missed alarm events, the feature values are first discretized to calculate the support of the item set, and when the support exceeds a minimum threshold, a weak correlation mode rule is generated. When a new event triggers the weak correlation mode rule, the feature cross correlation factor is multiplied by a gain coefficient.

[0052] By way of example, analysis of historical missed alarm events finds that in 5 events, 4 times simultaneously appear temperature rise rate 1.3±0.1℃ / s and H2 / CO=2.4±0.1, the support is supp=0.8, and the rule is generated: if the temperature rise rate=L2 and H2 / CO=P2, then the alarm is enhanced. When a new charging event detects a temperature rise of 1.32℃ / s and H2 / CO=2.38, the system automatically increases the feature cross correlation factor value from 0.68 to 0.884, causing the comprehensive correlation index to break through the threshold and trigger the warning. The beneficial effect of this verification example is that the machine learning technology automatically captures the hidden correlation rules in the historical missed alarms, significantly enhances the system's ability to recognize atypical risk patterns, and actively suppresses false alarms caused by environmental noise using the interference feature library, achieving two-way adaptive optimization.

[0053] Based on the same inventive concept, as shown in Figure 4 , the present application also provides a multi-sensor fusion-based electric vehicle charging fire warning system, which comprises: a multi-source data acquisition module for acquiring real-time monitoring data of a charging pile based on a preset sensor; a feature engineering processing module for generating a multi-dimensional primary feature vector based on the real-time monitoring data; a correlation analysis module for analyzing the multi-dimensional primary feature vector to generate a comprehensive correlation index; a BMS state scanning module for scanning a battery management system to generate internal state data; a state cooperative verification module for generating a cooperative verification result based on the comprehensive correlation index and the internal state data; a hierarchical warning decision module for generating a hierarchical warning decision based on the cooperative verification result.

[0054] It should be noted that the electrical connections between the various units described above are not necessarily direct connections, and that indirect connections are also possible in the embodiments of the present application. The above-described embodiments are merely exemplary embodiments of the present application, and should not be used in a limiting sense.

[0055] That is, any equivalent variations and modifications of the present application that follow the spirit of the present application are to be within the scope of the present application. Other embodiments of the present application will be readily apparent to those skilled in the art in view of the disclosure herein. The applicant intends to emcompass all alternatives, modifications and equivalents of the present application that fall within the scope of the present application as defined by the following claims.

Claims

1. A method for early warning of electric vehicle charging fire based on multi-sensor fusion, characterized in that: include: Obtain real-time monitoring data of charging piles based on preset sensors; generating a multi-dimensional primary feature vector based on the real-time monitoring data; Analyzing the multi-dimensional primary feature vectors to generate a comprehensive correlation index; Scan the battery management system to generate internal status data; generating a collaborative verification result based on the comprehensive correlation index and the internal state data; Based on the collaborative verification results, a graded warning decision is generated.

2. The electric vehicle charging fire warning method based on multi-sensor fusion according to claim 1 is characterized in that: Generating a multi-dimensional primary feature vector includes: Identify key point temperature values ​​and maximum temperature differences in the preset temperature sensor array data, calculate the maximum temperature rise rate, and generate a temperature feature vector; Calculate the concentration increase rate and concentration ratio of the electrolyte decomposition characteristic gas in the preset composite gas sensor data to generate a gas characteristic vector; Calculate the current effective value fluctuation amplitude, harmonic content change rate and charging efficiency drop rate in the current sensor data to generate an electrical characteristic vector; The temperature feature vector, the gas feature vector and the electrical feature vector are fused to generate a multi-dimensional primary feature vector.

3. The electric vehicle charging fire warning method based on multi-sensor fusion according to claim 1 is characterized in that: Generating a comprehensive relevance index includes: generating a time correlation factor based on the time series correlation of each vector in the multi-dimensional primary feature vector; Analyze the spatial propagation characteristics of sensor data at different locations and generate spatial correlation factors; Calculate the statistical relationship between feature vectors of different dimensions and generate feature cross-correlation factors; The time correlation factor, space correlation factor and feature cross-correlation factor are integrated to generate a comprehensive correlation index.

4. The electric vehicle charging fire warning method based on multi-sensor fusion according to claim 3 is characterized in that: The generating feature cross-correlation factor comprises: A dynamic ratio model is established based on the maximum temperature rise rate and the characteristic gas concentration rise rate; Correlation between the spectral characteristics of current fluctuations and the spatial distribution of local temperature differences; When the dynamic ratio model and the spatial distribution relationship conform to a preset battery thermal runaway evolution pattern, a characteristic cross-correlation factor is generated.

5. The electric vehicle charging fire warning method based on multi-sensor fusion according to claim 1 is characterized in that: Generating the collaborative verification result includes: comparing the comprehensive correlation index with the internal state data to generate an anomaly factor; Based on the abnormal factors, collaborative verification results are generated through preset association rules.

6. The electric vehicle charging fire warning method based on multi-sensor fusion according to claim 1 is characterized in that: The graded warning decision-making includes: When the comprehensive correlation index exceeds a first preset threshold and the internal state data does not trigger a fault alarm, performing a primary warning operation; When the comprehensive correlation index exceeds a second preset threshold, or the comprehensive correlation index continues to rise after the primary warning, or the internal state data is associated with an abnormal pattern of an external sensor, a high-level alarm operation is executed.

7. The electric vehicle charging fire warning method based on multi-sensor fusion according to claim 6 is characterized in that: The advanced alarm operations include: Trigger the preset strong sound and light alarm device; Send emergency alarm signals to the remote monitoring platform; Control the charging pile to start the fire extinguishing device and cut off the charging circuit.

8. The electric vehicle charging fire warning method based on multi-sensor fusion according to claim 1 is characterized in that: The method further comprises: Record the multi-dimensional primary feature vectors, comprehensive correlation index and decision results in historical warning events to generate a historical warning matrix; Based on the historical warning matrix, the parameter weights for analyzing the multi-dimensional primary feature vector and the thresholds for graded warning decisions are optimized.

9. The method according to claim 8, characterized in that The optimization of the parameter weights for analyzing the multi-dimensional primary feature vector and the thresholds for graded warning decisions based on the historical warning matrix includes: Extract the interference pattern of the feature vector in the false alarm event and reduce the association weight of similar interference patterns; Extract weak correlation patterns from missed events and enhance the sensitivity of specific feature cross-correlation factors.

10. The electric vehicle charging fire warning system based on multi-sensor fusion is applied to the electric vehicle charging fire warning method based on multi-sensor fusion as described in any one of claims 1 to 9, characterized in that: The system comprises: Multi-source data acquisition module, used to obtain real-time monitoring data of charging piles based on preset sensors; A feature engineering processing module, configured to generate a multi-dimensional primary feature vector based on the real-time monitoring data; A correlation analysis module, configured to analyze the multi-dimensional primary feature vectors and generate a comprehensive correlation index; BMS status scanning module, used to scan the battery management system and generate internal status data; A state collaborative verification module, configured to generate a collaborative verification result based on the comprehensive correlation index and the internal state data; The hierarchical warning decision module is used to generate a hierarchical warning decision based on the collaborative verification result.

Images