A method for early warning and detection of lithium battery fires

Abstract

This invention discloses an early warning detection method for lithium battery fires, comprising the following steps: collecting characteristic parameter data of lithium battery fires using multiple detectors; synchronizing and normalizing the characteristic parameter data; constructing a basic probability allocation (BPA) based on D-S evidence theory, and synthesizing D-S evidence on the normalized data; performing conflict detection on the synthesized evidence, and refining the BPA based on the conflict detection results; and obtaining a fire early warning detection decision output based on the refined BPA. This invention employs a composite detection method to simultaneously collect typical early characteristic parameters of lithium battery fires. Through deep fusion and intelligent inference of multi-source parameters using D-S evidence theory, it can quantify the degree of support of various parameters for different fire states (normal, warning, fire, interference / fault), thereby obtaining a more reliable and accurate comprehensive judgment than any single sensor.

Description

Technical Field

[0001] This invention relates to the field of lithium battery fire early warning and detection technology, and in particular to an early warning and detection method for lithium battery fires. Background Technology

[0002] In the field of lithium battery fire detection technology, photoelectric smoke detectors generally employ a single-wavelength light source, sensing a fire by detecting the intensity of scattering of incident light by smoke particles. While this method boasts high sensitivity, it is susceptible to interference sources in complex environments (such as dust and water mist), leading to a persistently high false alarm rate. The fundamental reason is that the extinction and scattering characteristics of fire smoke and interference sources are influenced by multiple factors, including the wavelength of the light source, the scattering angle, and the type and size distribution of particles. A low-dimensional signal with a single wavelength and angle cannot distinguish between fire smoke particles and interference source particles.

[0003] Current lithium battery fire detection technology uses a single light source to measure the extinction or scattering of fire smoke. However, the signal acquired by a single light source has dimensional limitations and cannot distinguish interference sources such as dust and water mist. Currently, the operating temperature of fire smoke detectors is generally between -10 and 55°C, which is insufficient to meet the detection requirements of lithium battery fire smoke in the extreme environment where lithium batteries operate at temperatures of -40 to 85°C. Furthermore, there is a lack of effective means to suppress the cross-interference of combustible gases and high temperatures present in lithium battery fires.

[0004] For example, invention application number 202511086454.X discloses a fire monitoring system and method for electrochemical energy storage facilities based on big data. This solution can scientifically diagnose lithium battery fires according to their early, middle, and late stages; and perform in-depth analysis of the fire extinguishing procedures and fire extinguishing linkage control for lithium battery fires. However, this solution also has the following problems: it uses a single light source wavelength, resulting in poor discrimination of interference sources, limited operational stability, and a single detection method.

[0005] Therefore, in reality, there is a need for an early warning and detection method for lithium battery fires that can effectively distinguish between fire smoke and interference sources, adapt to extreme temperature environments, and suppress cross-interference from multiple factors. Summary of the Invention

[0006] To address the aforementioned problems, this invention proposes an early warning detection method for lithium battery fires. It employs a composite detection approach to simultaneously collect typical early-stage lithium battery fire characteristic parameters such as smoke, temperature, carbon monoxide (CO), and hydrogen (H2). Through DS evidence theory, it performs deep fusion and intelligent inference of multi-source parameters, quantifying the degree to which various parameters support different fire states (normal, warning, fire, interference / fault), thereby arriving at a more reliable and accurate comprehensive judgment.

[0007] The objective of this invention can be achieved through the following technical solution: a method for early warning and detection of lithium battery fires, comprising the following steps:

[0008] S1. Use multiple detectors to collect data on various characteristic parameters of lithium battery fires;

[0009] S2. Perform data synchronization and normalization on the data of each feature parameter;

[0010] S3. Based on the DS evidence theory, construct the Basic Probability Allocation (BPA) and synthesize DS evidence on the normalized data.

[0011] S4. Perform conflict detection on the synthesized evidence and improve the Basic Probability Allocation (BPA) based on the conflict detection results.

[0012] S5. Obtain fire early warning detection decision output based on the improved Basic Probability Allocation (BPA).

[0013] As a further embodiment of the present invention, the multi-detector includes a dual-band smoke detector, a temperature sensor, and a gas sensor, wherein:

[0014] The dual-band smoke detector uses 460nm and 940nm wavelengths. After temperature compensation and baseline drift correction, it collects the forward and backward scattered light power and performs particle size analysis.

[0015] The temperature sensor uses a thermistor to monitor the ambient temperature and the surface temperature of the lithium battery.

[0016] Gas sensors monitor CO and H2 concentrations.

[0017] As a further embodiment of the present invention, the dual-band smoke detector performs temperature compensation and baseline drift correction, expressed by the following formula:

[0018] B(T) = B0 + k*(T−T) ref )

[0019] S c = S r − B(T)

[0020] Where B(T) is the temperature-compensated calibration baseline, and B0 is the reference temperature T. ref The baseline value is given below, where k is the temperature coefficient, T is the ambient temperature, and S is the base value. r S represents the smoke detector output value at the current ambient temperature. c This is used to calibrate the output value of the smoke detector.

[0021] As a further embodiment of the present invention, the particle size analysis includes smoke particle size, dust particle size, and water mist particle size analysis, expressed by the following formula:

[0022]

[0023] in, For asymmetric ratios, P F P represents the forward-scattered light power. B This represents the power of the backscattered light.

[0024] As a further embodiment of the present invention, the data of each feature parameter is normalized, and the formula is expressed as follows:

[0025]

[0026] Where V is the original feature parameter data, V min and V max This represents the sensor's measurement range boundary.

[0027] As a further embodiment of the present invention, the DS evidence theory state proposition identification framework includes:

[0028] A1: Normal state; A2: Warning state; A3: Fire state; A 4： Interference / fault status, where:

[0029] Θ={A1, A2, A3, A4}

[0030] Θ is the complete set of mutually exclusive state propositions.

[0031] The basic probability assignment (BPA) assigns a probability to a subset of each Θ, satisfying:

[0032] ,

[0033] in, Let be an empty set of mutually exclusive state propositions. It is a subset.

[0034] As a further aspect of the present invention, in the DS evidence theory:

[0035] Smoke particle size: R < 3 supports A1; R ≥ 3 supports A2; R ≥ 3 and photodiode voltage rise rate < 2.4 V / s supports A3; photodiode voltage rises and R ≈ 1 supports A4; R ≥ 3 and photodiode voltage rise rate ≥ 2.4 V / s supports A4.

[0036] Temperature range: Temperature < 50℃ supports A1, 50℃ ≤ Temperature < 65℃ supports A2, temperature ≥ 65℃ supports A3, temperature without reading supports A4.

[0037] CO concentration: Threshold classification <50 ppm supports A1; 50–100 ppm supports A2; >100 ppm supports A3; no value supports A4.

[0038] H2 concentration: Threshold classification: <100 ppm supports A1; 100-200 ppm supports A2; >200 ppm supports A3; no value supports A4.

[0039] As a further aspect of the present invention, the conflict detection of the synthetic evidence is expressed by the following formula:

[0040]

[0041]

[0042] Where, m T and m S For two independent synthetic pieces of evidence, B and C ⊆ Θ, and K is the conflict coefficient.

[0043] The beneficial effects of this invention are:

[0044] 1. The method of this invention constructs a multi-sensor collaborative detection system. It simultaneously collects typical early characteristic parameters of lithium battery thermal runaway, such as smoke (using a unique dual-band), temperature, carbon monoxide (CO), and hydrogen (H2). By employing this specific parameter combination and its fire judgment logic of simultaneous acquisition and collaborative analysis, it ensures cross-verification of the fire situation from multiple physical dimensions.

[0045] 2. To address the false alarm problem caused by interference sources such as dust and water mist, this invention introduces dual-band smoke detection technology and constructs a particle size distribution database of fire smoke, dust, and water mist based on Mie scattering theory and the "three regions" law. By combining the differences in scattering characteristics of different substances in this database, it can effectively distinguish fire smoke from non-fire interference materials such as dust and water mist. The scattering ratio R under dual-band conditions exhibits a significantly different characteristic range compared to regularly spherical and relatively uniformly sized dust particles and larger water mist particles, thereby greatly reducing the false alarm rate caused by environmental interference.

[0046] 3. The method of this invention has adaptive correction capability for extreme ambient temperatures. Specifically, this invention proposes a temperature compensation method to correct the baseline drift of the smoke sensor in the extreme temperature range (-40℃ to 85℃) of the lithium battery operating environment. Through a specific temperature drift model and dynamic calibration algorithm, the accuracy of the detector is ensured across the entire temperature range.

[0047] 4. The method of this invention is based on an intelligent decision-making model using uncertainty reasoning. The innovation of this invention lies in its decision-making mechanism, which adopts DS evidence theory as the core algorithm for multi-source information fusion. It can handle the uncertainty and conflict of sensor information, transforming each sensor data into a basic probability assignment (BPA) for propositions such as "normal, warning, fire, interference / fault", and then performing intelligent inference through evidence combination rules, thereby achieving accurate assessment of the risk status of lithium battery fires. Attached Figure Description

[0048] Figure 1 This is a schematic flowchart of the method of the present invention;

[0049] Figure 2 This is a flowchart illustrating the principle of the method of the present invention. Detailed Implementation

[0050] Embodiments of the present invention are described in detail below. Examples of these embodiments are illustrated in the accompanying drawings, wherein the same or similar symbols denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0051] like Figures 1-2 As shown, this invention discloses an early warning detection method for lithium battery fires, comprising the following steps:

[0052] S1. Multiple detectors are used to collect characteristic parameter data of lithium battery fires.

[0053] The multi-detector system includes dual-band smoke detectors, temperature sensors, and gas sensors, among others:

[0054] The dual-band smoke detector uses 460nm and 940nm wavelengths (emitting 460nm blue light and 940nm infrared light).

[0055] First, temperature compensation and baseline drift correction were performed on the dual-band smoke detector. Within the range of -40℃ to 85℃, the relationship between the baseline output of the dual-band smoke detector and the ambient temperature T was experimentally calibrated, and a drift model was established.

[0056] B(T) = B0 + k*(T−T) ref )

[0057] Where B(T) is the temperature-compensated calibration baseline, and B0 is the reference temperature T. ref The baseline value is T, where T is the ambient temperature and k is the temperature coefficient.

[0058] The ambient temperature is collected in real time, and a piecewise linear compensation formula is used to correct the smoke signal to ensure stable output of the detector under extreme conditions.

[0059] S c = S r − B(T)

[0060] Where S r The current ambient temperature is the smoke detector output value for the dual-band smoke detector, B(T) is the temperature-compensated calibration baseline, and S is the output value for the smoke detector. c Calibrate the output value for the dual-band smoke detector.

[0061] After temperature compensation and baseline drift correction, the forward (85°) and backward (120°) scattered light power was collected for particle size analysis.

[0062] Particle size analysis, based on Mie scattering theory, calculates the ratio (asymmetric ratio) of forward and backward scattered light power in dual-band (460nm / 940nm), using the following formula:

[0063]

[0064] in, For asymmetric ratios, P F P represents the forward-scattered light power. B This represents the power of the backscattered light.

[0065] By comparing the measured asymmetry ratio with the pre-defined particle size distribution database of fire smoke, dust, and water mist (particle size range 0.1–10 μm) under the "three regions" law, fire smoke, dust, and water mist can be distinguished.

[0066] Fire smoke particles are mostly concentrated in the range of 0.1–1 μm, with an asymmetry ratio R usually > 3; dust particles are mostly larger than 1 μm, with an asymmetry ratio close to 1; water mist particles have a wide particle size distribution but their scattering intensity rises very quickly (within 0.5 s), and the asymmetry ratio of water mist particles with a particle size mostly larger than 1 μm is close to 1. Water mist particles with a size of 0.1–1 μm need to be judged in conjunction with the rise rate of the photodiode voltage.

[0067] If the asymmetry ratio or the rate of rise of the photodiode voltage falls within the dust or water mist range, it is marked as an interference signal and will not trigger a fire alarm.

[0068] A temperature sensor (using a thermistor) monitors the ambient temperature and the surface temperature of the lithium battery; a gas sensor monitors the concentrations of CO and H2.

[0069] S2. Perform data synchronization and normalization on the data of each feature parameter.

[0070] Data synchronization and normalization are performed. The original feature parameter data output by multiple sensors are time-synchronized and calibrated, and then normalized to the [0,1] interval using the following formula to eliminate the influence of dimensions.

[0071]

[0072] Where V is the original feature parameter data, V min and V max This represents the sensor's measurement range boundary.

[0073] S3. Based on the DS evidence theory, construct the Basic Probability Allocation (BPA) and synthesize DS evidence on the normalized data.

[0074] Using the DS evidence theory, normalized multi-sensor data is mapped to four states of confidence: normal, warning, fire, and interference / fault. This enables multi-sensor fusion, anti-interference identification, and environmental adaptive correction, thereby improving the accuracy of early detection of lithium battery fires.

[0075] The DS evidence theory state proposition identification framework includes: A1: Normal state; A2: Warning state; A3: Fire state; A 4： Interference / fault status, where:

[0076] Θ={A1, A2, A3, A4}

[0077] Θ is the complete set of mutually exclusive state propositions.

[0078] The basic probability assignment (BPA) assigns a probability to a subset of each Θ, satisfying:

[0079] ,

[0080] in, Let be an empty set of mutually exclusive state propositions. It is a subset.

[0081] Taking a temperature sensor as an example, BPA mapping is constructed to include: defining the identification framework and the temperature proposition.

[0082] A1: Normal state, indicating the temperature is within a safe range. A2: Warning state, indicating an abnormal temperature increase, potentially posing a risk. A3: Fire state, indicating the temperature has reached or exceeded the fire threshold. A4: Interference / Fault state, indicating temperature data may be abnormal due to sensor malfunction or other non-fire-related factors.

[0083] Determine the temperature threshold and membership function, and set a temperature critical value for each state. Normal range (A1): temperature < 50℃; Warning range (A2): 50℃ ≤ temperature < 65℃; Fire range (A3): temperature ≥ 65℃; Interference / fault range (A4): no temperature reading.

[0084] Next, a trapezoidal membership function is selected for each interval, as shown in Table 1:

[0085] Table 1. Parameter Table for State Propositions

[0086]

[0087] Combining multiple detectors such as dual-band smoke detectors, temperature sensors, and gas sensors, similar methods are applied to dual-band smoke detectors and CO and H2 gas sensors, where:

[0088] Smoke particle size: R < 3 supports A1; R ≥ 3 supports A2; R ≥ 3 and photodiode voltage rise rate < 2.4 V / s supports A3; photodiode voltage rises and R ≈ 1 supports A4; R ≥ 3 and photodiode voltage rise rate ≥ 2.4 V / s supports A4.

[0089] CO concentration: Threshold classification <50 ppm supports A1; 50–100 ppm supports A2; >100 ppm supports A3; no value supports A4.

[0090] H2 concentration: Threshold classification: <100 ppm supports A1; 100-200 ppm supports A2; >200 ppm supports A3; no value supports A4.

[0091] S4. Perform conflict detection on the synthetic evidence and improve the Basic Probability Allocation (BPA) based on the conflict detection results.

[0092] Conflict detection is performed on the synthesized evidence. The similarity between different pieces of evidence is calculated using the Jousselme distance function, and then the support level of each piece of evidence is calculated. After normalizing the support levels, the credibility of the corresponding evidence is obtained. The credibility is used as a weight to perform a weighted average of the conflicting evidence. Finally, the mass function of each piece of evidence is processed using the Dempster fusion rule. The steps include:

[0093] First, obtain the distances between each piece of evidence:

[0094]

[0095] in For i evidence, For j evidence, It is the dot product of vectors.

[0096] The distance between pieces of evidence is used as a measure of similarity; the smaller the distance, the greater the similarity. Similarity is expressed as:

[0097] Sim(m i , mj )=1-d(m i , m j )

[0098] Based on the similarity score, the support level of each piece of evidence is further calculated, and the support level is expressed as follows:

[0099]

[0100] Then normalize the evidence m i Support level, received m i The credibility of Crd is expressed as:

[0101]

[0102] Using credibility CRD as evidence m i The weights are then applied to conflicting evidence m. i After preprocessing, the corrected mass function is obtained, which is expressed as:

[0103]

[0104] The weighted average mass function is used as the mass function provided by each classifier. Then, the new mass function is combined three times using the Dempster fusion rule to obtain the final classification output.

[0105] Dempster fusion rules are based on m T and m S For example, m T and m S For two independent sources of evidence, such as the mass functions of a temperature sensor and a dual-band smoke detector, respectively:

[0106]

[0107] Where B and C ⊆ Θ, and K is the conflict coefficient, calculated using the following formula:

[0108]

[0109] That is, list all propositional combinations, pairwise combine the focal elements of two pieces of evidence (propositions with BPA greater than 0); calculate the conflict coefficient K, find all propositional combinations with no intersection, and calculate the sum of their BPA products; calculate the synthetic BPA, find all combinations of the target proposition whose intersection is exactly equal to the target proposition, calculate the sum of their BPA products, and then divide by the normalization factor (1−K).

[0110] After fusion, we obtain m(A1), m(A2), m(A3), m(A4), and m(Θ). The decision must satisfy:

[0111] 1) Maximum value criterion: Final state A ∗ =argmaxm(A i ).

[0112] 2) Difference threshold: m(A) ∗ )−m(A 2nd )>ϵ (ϵ=0.1), to avoid fuzzy decision-making.

[0113] 3) Uncertainty threshold: m(Θ) < δ (δ = 0.2) to ensure reliable results.

[0114] This invention employs a composite detection method to simultaneously collect multi-parameter features such as smoke, temperature, and gases (e.g., CO, H2). Combining dual-band smoke detection technology with Mie scattering theory, a particle size distribution database of smoke particles and interfering sources (dust, water mist) is constructed. Simultaneously, to address the detector baseline drift problem caused by extreme ambient temperatures (-40℃ to 85℃), a dynamic temperature compensation algorithm is introduced to improve detection stability. Finally, based on DS evidence theory or fuzzy comprehensive evaluation method, multi-source data fusion is achieved to realize early and accurate identification and graded warning of fire conditions.

[0115] S5. Obtain fire early warning detection decision output based on the improved Basic Probability Allocation (BPA).

[0116] Fire status is determined using the maximum value criterion, difference threshold, and uncertainty threshold. From the improved BPA (Basic Probability Assessment), basic probability assignments m(Ai) are selected for each possible state (e.g., no fire, early-stage fire, typical fire). The target state A* with the highest probability value is determined using the maximum value criterion. Subsequently, the basic probability assignment of this target state A* and the second-highest basic probability assignment m(Ai) are calculated. 2nd The difference between the probabilities of different states (Θ) and the threshold ϵ (ϵ=0.1 in this embodiment) must be greater than the set threshold ϵ to exclude fuzzy decision-making situations caused by similar probabilities of different states. At the same time, it is also necessary to verify whether the uncertainty m(Θ) is less than the threshold δ (δ=0.2 in this embodiment) to ensure that the entire decision-making process is based on relatively certain information and to avoid making unreliable judgments under high uncertainty.

[0117] When all three conditions are met, the target state A* can be used as the final fire warning detection decision output, realizing early and accurate identification and graded warning of lithium battery fire status; if any condition is not met, the system will return to the pending decision state or carry out further data collection and fusion processing to improve the accuracy and reliability of the decision.

[0118] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for early warning and detection of lithium battery fires, characterized in that, include: S1. Use multiple detectors to collect data on various characteristic parameters of lithium battery fires; S2. Perform data synchronization and normalization on the data of each feature parameter; S3. Based on the DS evidence theory, construct the Basic Probability Allocation (BPA) and synthesize DS evidence on the normalized data. S4. Perform conflict detection on the synthesized evidence and improve the Basic Probability Allocation (BPA) based on the conflict detection results. S5. Obtain fire early warning detection decision output based on the improved Basic Probability Allocation (BPA).

2. The method for early warning and detection of lithium battery fires according to claim 1, characterized in that, The multi-detector system includes a dual-band smoke detector, a temperature sensor, and a gas sensor, wherein: The dual-band smoke detector uses 460nm and 940nm wavelengths. After temperature compensation and baseline drift correction, it collects the forward and backward scattered light power and performs particle size analysis. The temperature sensor uses a thermistor to monitor the ambient temperature and the surface temperature of the lithium battery. Gas sensors monitor CO and H2 concentrations.

3. The method for early warning and detection of lithium battery fires according to claim 2, characterized in that, The dual-band smoke detector performs temperature compensation and baseline drift correction, expressed by the following formula: B(T)= B0+ k*(T−T ref ) S c = S r − B(T) Where B(T) is the temperature-compensated calibration baseline, and B0 is the reference temperature T. ref The baseline value is given below, where k is the temperature coefficient, T is the ambient temperature, and S is the base value. r S represents the smoke detector output value at the current ambient temperature. c This is used to calibrate the output value of the smoke detector.

4. The method for early warning and detection of lithium battery fires according to claim 2, characterized in that, The particle size analysis includes smoke particle size, dust particle size, and water mist particle size analysis, expressed by the following formula: in, For asymmetric ratios, P F P represents the forward-scattered light power. B This represents the power of the backscattered light.

5. A method for early warning detection of lithium battery fires according to any one of claims 1 to 4, characterized in that, The normalization process for each feature parameter data in S2 is expressed by the following formula: Where V is the original feature parameter data, V min and V max This represents the sensor's measurement range boundary.

6. The method for early warning detection of lithium battery fires according to claim 1, characterized in that, The S3 framework for identifying state propositions in DS evidence theory includes: A1: Normal state; A2: Warning state; A3: Fire state; A 4： Interference / fault status, where: Θ={A1, A2, A3, A4} Θ is the complete set of mutually exclusive state propositions; The basic probability assignment (BPA) assigns a probability to a subset of each Θ, satisfying: ， in, Let be an empty set of mutually exclusive state propositions. It is a subset.

7. The method for early warning detection of lithium battery fires according to claim 6, characterized in that, In the aforementioned DS evidence theory: Smoke particle size: R < 3 supports A1; R ≥ 3 supports A2; R ≥ 3 and photodiode voltage rise rate < 2.4 V / s supports A3; photodiode voltage rises and R ≈ 1 supports A4; R ≥ 3 and photodiode voltage rise rate ≥ 2.4 V / s supports A4. Temperature range: Temperature < 50℃ supports A1, 50℃ ≤ Temperature < 65℃ supports A2, temperature ≥ 65℃ supports A3, no temperature reading supports A4; CO concentration: Threshold classification: <50 ppm supports A1; 50–100 ppm supports A2; >100 ppm supports A3; no value supports A4; H2 concentration: Threshold classification: <100 ppm supports A1; 100-200 ppm supports A2; >200 ppm supports A3; no value supports A4.

8. The method for early warning detection of lithium battery fires according to claim 7, characterized in that, In step S4, conflict detection is performed on the synthesized evidence, expressed by the following formula: Where, m T and m S For two independent synthetic pieces of evidence, B and C ⊆ Θ, and K is the conflict coefficient.

Images