Battery thermal runaway prediction method based on improved multi-modal dynamic weight fusion algorithm

Abstract

The invention relates to a battery thermal runaway prediction method based on an improved multi-modal dynamic weight fusion algorithm, belongs to the technical field of lithium ion battery safety monitoring, in particular to an electric tool battery, and comprises the following steps: S1, collecting gas, temperature, pressure, deformation and voltage data of a lithium ion battery; s2, performing preprocessing, feature extraction and data fusion on the acquired data based on a risk assessment model of a multi-modal dynamic weight fusion algorithm MM-DWFA, converting multi-source data into feature vectors, analyzing the feature vectors and outputting early warning signals; and S3, executing corresponding early warning operation according to the early warning signal.

Description

Technical Field

[0001] This invention belongs to the field of lithium-ion battery safety monitoring technology, particularly power tool batteries, and relates to a battery thermal runaway prediction method based on an improved multimodal dynamic weight fusion algorithm. Background Technology

[0002] The thermal runaway early warning technology for lithium-ion batteries suffers from significant shortcomings in model accuracy, becoming a key bottleneck restricting the safety of power tools. Existing models mostly employ fixed parameters and linear assumptions, failing to accurately characterize the nonlinear dynamic characteristics of the thermal runaway process—for example, the risk parameter weights differ between the initial and violent reaction stages of thermal runaway. Static models, using fixed coefficients, result in large risk assessment errors. Furthermore, these models do not consider individual battery differences and the degradation patterns throughout the battery's lifespan; for batteries aged after 500 cycles, the early warning accuracy is far lower than that of new batteries.

[0003] In terms of prediction accuracy, existing models lack the ability to model the spatiotemporal correlations of multiphysics coupling. In the temporal dimension, the lack of a dynamic transmission model for gas concentration-temperature-pressure leads to significant prediction errors in the rate of thermal runaway evolution. In the spatial dimension, neglecting the thermal diffusion effect between battery cells results in high delays in identifying local hotspots due to single-point temperature measurement. This dual deficiency in spatiotemporal accuracy leads to a high false negative rate for early risks in practical applications, severely impacting the reliability of early warning systems. Summary of the Invention

[0004] In view of this, the purpose of this invention is to provide a battery thermal runaway prediction method based on an improved multimodal dynamic weight fusion algorithm.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] A battery thermal runaway prediction method based on an improved multimodal dynamic weight fusion algorithm includes the following steps:

[0007] S1: Collect gas, temperature, pressure, strain, and voltage data of the lithium-ion battery;

[0008] S2: A risk assessment model based on the multimodal dynamic weight fusion algorithm MM-DWFA preprocesses, extracts features, and fuses the collected data, transforms multi-source data into feature vectors, analyzes the feature vectors, and outputs early warning signals.

[0009] S3: Execute the corresponding warning operation based on the warning signal.

[0010] Furthermore, in step S1, a gas sensor is used to detect hydrogen, carbon monoxide, and carbon dioxide data. The gas concentration acquisition formula is as follows:

[0011] C i (t)=k i ×V out,i (t)+b i

[0012] Where k i b i V is the sensor calibration coefficient. out,i (t) represents the output voltage, and i includes three types: H2, CO, and CO2.

[0013] Furthermore, it also includes measuring temperature using a temperature sensor.

[0014] further,

[0015] The pressure data is calculated using a dynamic pressure change model:

[0016]

[0017] in k represents the cumulative total concentration of the characteristic gas over time during thermal runaway. p k is the gas pressure coefficient. t P0 is the coefficient of thermal expansion; T(t) is the initial ambient pressure; T(t) is the temperature that changes with time; and T0 is the initial temperature.

[0018]

[0019] in These represent the cumulative concentrations of hydrogen, carbon monoxide, and carbon dioxide over time during thermal runaway.

[0020] further,

[0021] The strain data includes radial strain and axial strain, wherein:

[0022] The radial strain is calculated as follows:

[0023] ε radial (t)=0.35×ε R1 +0.15×ε R2 +0.35×ε R3 +0.15×ε R4

[0024] Where ε radial (t) represents the radial strain, reflecting the overall deformation / failure risk of the structure in the radial direction; ε R1 ε represents the radial strain at the edge of the battery electrode. R2 ε represents the radial strain at the top of the battery casing. R3 ε represents the radial strain in the central region of the diaphragm. R4 Radial strain at the edge of the electrode core;

[0025] The axial strain is calculated as follows:

[0026] ε axial (t)=0.5×ε A1 +0.5×ε A2

[0027] Where ε A1 , ε A2 The strain is at two locations along the axial direction of the battery;

[0028] The total strain is:

[0029] ε total (t)=0.6×ε radial (t)+0.4×ε axial (t).

[0030] further,

[0031] The voltage data includes voltage fluctuations, calculated as follows:

[0032]

[0033] Where σ V (t) represents the voltage fluctuation rate; N represents the number of battery cells; Vi represents the voltage of the i-th cell; Average voltage of the battery pack. Furthermore,

[0034] It also includes the correction index R1(t) for the concentration ratio of H2 to CO and the correction index R2(t) for the concentration ratio of CO to CO2.

[0035]

[0036] Thermal runaway is divided into the following four stages:

[0037]

[0038] Furthermore, the risk assessment model processing steps based on the multimodal dynamic weight fusion algorithm MM-DWFA include:

[0039] S21: Preprocess the collected and calculated data, including removing outliers and normalizing parameters, including:

[0040] Using the improved 3σ criterion, a sliding window is constructed for each parameter, and the mean μ and standard deviation σ are calculated. When a data point satisfies |x(t)-μ|>3σ, it is replaced with the second nearest neighbor value within the window.

[0041] x corr (t) = x(t-1)×0.7 + x(t+1)×0.3

[0042] Map the data to the [0,1] interval and use a stage-adaptive threshold:

[0043]

[0044] S22: Perform feature extraction on the preprocessed data, including temporal features and cross-parameter coupling, including:

[0045] Calculate the time-domain features of normalized data, including:

[0046] Mean:

[0047] Peak value: x peak =max(x i )

[0048] kurtosis:

[0049] Rate of change:

[0050] Construct a 5×5 coupling matrix and calculate the nonlinear interaction characteristics:

[0051] F i,j (t)=x i (t)×x j (t)+sin(x i (t)-x j (t))

[0052] Where i and j represent different parameters;

[0053] S23: Dynamically assign weights to each feature to achieve data fusion, including:

[0054] First, calculate the correction index R1(t) for the concentration ratio of H2 to CO and the correction index R2(t) for the concentration ratio of CO to CO2:

[0055]

[0056] Thermal runaway is divided into the following four stages:

[0057]

[0058] Then, a smooth switching mechanism is used to achieve the weight transition:

[0059] When the thermal runaway phase switches from S to S+1, the weights achieve a linear transition:

[0060] W(S,t)=W0×M(S)

[0061] Where W0 represents the initial weights, and M(S) is the stage adjustment matrix;

[0062] S24: Based on the characteristics of fusion, assess the risk of thermal runaway and output early warning signals, including:

[0063] Calculate the risk index:

[0064] RI(t)=∑(F i,j (t)×W(S,t))+0.2×Cov(ΠF i,j (t))

[0065] The warning levels are as follows:

[0066]

[0067] Furthermore, the stage adjustment matrix M(S) includes:

[0068] When S=0 during the thermal runaway phase, it indicates a normal state. At this time, the gas regulation coefficient is 0.57, the temperature regulation coefficient is 0.80, the pressure regulation coefficient is 1.00, the deformation regulation coefficient is 1.00, and the voltage regulation coefficient is 3.00.

[0069] When S=1 during the thermal runaway stage, it indicates the SEI decomposition period. At this time, the gas regulation coefficient is 1.71, the temperature regulation coefficient is 0.80, the pressure regulation coefficient is 0.33, the deformation regulation coefficient is 0.33, and the voltage regulation coefficient is 0.33.

[0070] When S=2 in the thermal runaway stage, it represents the electrolyte decomposition period. At this time, the gas regulation coefficient is 1.14, the temperature regulation coefficient is 1.20, the pressure regulation coefficient is 0.67, the deformation regulation coefficient is 0.67, and the voltage regulation coefficient is 0.33.

[0071] When S=3 in the thermal runaway stage, it represents the positive electrode decomposition period. At this time, the gas regulation coefficient is 0.57, the temperature regulation coefficient is 0.80, the pressure regulation coefficient is 2.00, the deformation regulation coefficient is 1.33, and the voltage regulation coefficient is 0.33.

[0072] Furthermore, in step S3, a first-level warning triggers a flashing green LED, a second-level warning activates a buzzer and Bluetooth alarm, and a third-level warning triggers a relay to cut off the battery output and provide a continuous audible and visual alarm.

[0073] The beneficial effects of this invention are as follows: This invention achieves accurate characterization of the entire thermal runaway cycle through a three-gas synergistic monitoring and four-stage dynamic segmentation mechanism. Specifically, the specific monitoring and ratio analysis of hydrogen (H2), carbon monoxide (CO), and carbon dioxide (CO2) solves the problem of delayed early warning for single gases. Combined with weighted fusion of radial / axial deformation, it improves the accuracy of identifying casing rupture risk. The multimodal dynamic weighted fusion algorithm (MM-DWFA), through stage adjustment matrix and nonlinear interaction features, improves the early warning accuracy of batteries at different aging stages, identifying risks earlier than traditional technologies and effectively covering the entire stage from SEI film decomposition to cathode material decomposition.

[0074] The system boasts significant advantages in engineering practicality: eliminating redundant current parameters reduces hardware costs and allows it to adapt to the complex operating conditions of power tools, including high temperatures, high dust levels, and strong vibrations. The dynamic weighting mechanism ensures stable performance throughout its entire lifecycle, accurately capturing early, weak signals while filtering out transient load interference. This provides a highly reliable and economical safety monitoring solution for lithium batteries in power tools, making it particularly suitable for the harsh working environments of industrial-grade power tools.

[0075] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination, or may be learned from practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description

[0076] To make the objectives, technical solutions, and advantages of the present invention clearer, the preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings, wherein:

[0077] Figure 1 This is a flowchart of the overall process for predicting battery thermal runaway based on an improved multimodal dynamic weight fusion algorithm as described in this invention.

[0078] Figure 2 Here is a flowchart of the multimodal dynamic weight fusion algorithm;

[0079] Figure 3 A schematic diagram of sensor array deployment;

[0080] Figure 4 This is a comparison chart of thermal runaway risk prediction with existing technologies. Detailed Implementation

[0081] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0082] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Therefore, the drawings only show the components related to the present invention and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0083] In the following description, numerous details are explored to provide a more thorough explanation of embodiments of the invention. However, it will be apparent to those skilled in the art that embodiments of the invention may be practiced without these specific details. In other embodiments, well-known structures and devices are shown in block diagram form rather than in detail to avoid obscuring embodiments of the invention.

[0084] Example 1:

[0085] like Figure 1 As shown, this invention addresses the technical pain points of thermal runaway early warning in lithium-ion batteries for power tools by providing a dynamic early warning scheme based on multi-physics coupling modeling. Through a three-layer architecture design of "multidimensional perception layer, adaptive fusion layer, and predictive decision layer," combined with an innovative algorithm model, it achieves early and accurate identification and graded early warning of thermal runaway risks. The following details the system architecture, core technologies, algorithm model, and early warning mechanism.

[0086] This invention constructs a fully closed-loop architecture of "perception-fusion-decision-execution," with clearly defined functions at each level and coordinated linkage, as detailed below:

[0087] Multi-physical quantity sensing layer: including gas sensor array, miniature pressure sensor, strain gauge array, distributed temperature sensor, and voltage acquisition module, used to acquire gas, temperature, pressure, strain, and voltage data of lithium-ion batteries.

[0088] Data processing layer: Includes data cleaning, feature extraction, and data fusion modules, which transform multi-source data into feature vectors.

[0089] Intelligent decision-making layer: Based on the improved MM-DWFA algorithm and risk assessment model, it analyzes feature vectors and outputs early warning signals.

[0090] Early warning execution layer: includes audible and visual alarms, wireless push notifications, and forced power-off modules, which execute corresponding operations based on the early warning signal.

[0091] The gas data is detected by gas sensors, including hydrogen, carbon monoxide, and carbon dioxide data. The gas concentration acquisition formula is as follows:

[0092] C i (t)=k i ×V out,i (t)+b i

[0093] Where k i b i V is the sensor calibration coefficient. out,i (t) represents the output voltage, and i includes three types: H2, CO, and CO2.

[0094] Pressure data is calculated using a dynamic pressure change model:

[0095]

[0096] This pressure calculation model is a core quantitative tool for measuring pressure evolution during thermal runaway, used to accurately characterize pressure changes during battery thermal runaway. Here, P0 represents the initial ambient pressure (usually atmospheric pressure), signifying the pressure evolution baseline; k p It is the gas production pressure coefficient (experimentally calibrated, such as 0.5 kPa·s / mL, reflecting the pressure change corresponding to 1 mL of gas production). The time integral of the total thermal runaway gas production (C) total It is the total gas production rate, which describes the process of gas accumulation leading to pressure rise; k t Let T(t) be the temperature-pressure coefficient (determined by the thermal expansion coefficient of the gas, 0.2 kPa / ℃), T(t) be the temperature changing with time, and T0 be the initial temperature. The difference between the two is expressed as (kPa / ℃). t The product reflects the pressure thermal expansion effect caused by an increase in temperature. When using it, first define the physical parameters (P0, k...). p k t Based on the time series (T0), prepare the total gas production rate C. total (Simulated or measured gas production at different stages of thermal runaway) and temperature T(t) data, the gas production accumulation term is calculated by numerical integration, and then the pressure P(t) is calculated time by time. τ has no special meaning and is the same as X for calculating definite integrals.

[0097]

[0098] in These represent the cumulative concentrations of hydrogen, carbon monoxide, and carbon dioxide over time during thermal runaway.

[0099] Strain data includes radial strain and axial strain, wherein:

[0100] The radial strain is calculated as follows:

[0101] ε radial (t)=0.35×ε R1 +0.15×ε R2 +0.35×ε R3 +0.15×ε R4

[0102] Where ε radial (t) represents the radial strain, reflecting the overall deformation / failure risk of the structure in the radial direction; ε R1 ε represents the radial strain at the edge of the battery electrode. R2 ε represents the radial strain at the top of the battery casing. R3 ε represents the radial strain in the central region of the diaphragm. R4 Radial strain at the edge of the electrode core;

[0103] The axial strain is calculated as follows:

[0104] ε axial (t)=0.5×ε A1 +0.5×ε A2

[0105] Where ε A1 , ε A2 The strain is at two locations along the axial direction of the battery;

[0106] The total strain is:

[0107] ε total (t)=0.6×ε radial (t)+0.4×ε axial (t)

[0108] Voltage data includes voltage fluctuations calculated using the following formula:

[0109]

[0110] Where σ V (t) represents the voltage fluctuation rate (reflecting the voltage consistency of the battery pack; the larger the value, the worse the consistency); N represents the number of individual battery cells (e.g., the number of cells in the battery pack); and Vi represents the voltage of the i-th individual cell (real-time collected value). Average voltage of the battery pack.

[0111] The Multimodal Dynamic Weight Fusion Algorithm (MM-DWFA) is the core algorithm framework of this invention, achieving adaptive fusion of five-dimensional parameters. Its specific structure includes four parts: a data preprocessing layer, a feature extraction layer, a dynamic weighting layer, and a decision output layer. Figure 2 As shown, the algorithm steps are as follows:

[0112] 1. Data Preprocessing Layer

[0113] Step 1: Outlier Removal

[0114] Using the improved 3σ criterion, a sliding window (window size N = 50) is constructed for each parameter. The mean μ and standard deviation σ are calculated. When a data point satisfies |x(t) - μ| > 3σ, it is replaced with the second nearest neighbor value within the window.

[0115] x corr (t) = x(t-1)×0.7 + x(t+1)×0.3

[0116] Step 2: Parameter Normalization

[0117] The five-dimensional parameters are mapped to the [0,1] interval, and a stage-adaptive threshold is used:

[0118]

[0119] Where x min (S) and x max (S) represents the minimum and critical values ​​of the parameters under the current stage S.

[0120] 2. Feature Extraction Layer

[0121] Step 1: Temporal Feature Extraction

[0122] Calculate the time-domain features of normalized data, including:

[0123] mean

[0124] Peak x peak =max(x i )

[0125] cliff

[0126] rate of change

[0127] Step 2: Cross-parameter coupling features

[0128] Construct a 5×5 coupling matrix and calculate the nonlinear interaction characteristics:

[0129] F i,j (t)=x i (t)×x j(t)+sin(x i (t)-x j (t))

[0130] Where i and j represent different parameters, such as the coupling characteristics of gas and temperature. (Structure and process of a five-dimensional parameter nonlinear coupling model)

[0131] 3. Dynamic weight layer

[0132] First, calculate the correction index R1(t) for the concentration ratio of H2 to CO and the correction index R2(t) for the concentration ratio of CO to CO2:

[0133]

[0134] Thermal runaway is divided into the following four stages:

[0135]

[0136] Then, a smooth switching mechanism is used to achieve the weight transition:

[0137] When the stage switches from S to S+1, the weights must achieve a linear transition:

[0138] W(S,t)=W0×M(S)

[0139] Where W0 represents the initial weight, which is set to [0.35, 0.25, 0.15, 0.15, 0.10] in this embodiment, and M(S) is the stage adjustment matrix as shown in Table 1.

[0140] Table 1

[0141]

[0142] Physical meaning:

[0143] Normal state: The voltage regulation coefficient is the highest, because voltage stability is the core indicator during normal charging and discharging;

[0144] SEI decomposition period: The gas adjustment coefficient is the highest, highlighting the monitoring weight of early characteristic gases such as H2;

[0145] During the positive electrode decomposition period, the pressure regulation coefficient and deformation regulation coefficient are the highest, because the risk of shell rupture is mainly dominated by pressure and deformation during this stage.

[0146] This matrix allows the system to automatically amplify the weights of key parameters according to the thermal runaway process, solving the problem of poor adaptability of traditional fixed weights to dynamic processes.

[0147] 4. Decision Output Layer

[0148] Risk index: RI(t)=∑(Fi,j (t)×W(S,t))+0.2×Cov(ΠF i,j (t)).

[0149] Warning level:

[0150] Example 2:

[0151] like Figure 3 As shown, the hardware of this system adopts a modular integrated design. The core sensing components include a three-channel gas sensor array, a 16-channel distributed temperature monitoring network, a miniature pressure sensor, a 6-channel strain gauge array, and a high-precision voltage acquisition module. The gas sensors are ME2 series electrochemical sensors (CO, CO2, H2), integrated into a 30×20×10mm sealed cavity. A 1mm diameter vent hole is opened on the front of the cavity and covered with a polytetrafluoroethylene film. The sensor is mounted 5mm above the center of the top surface of the battery pack, operating at 5V, with a sampling frequency of 200Hz. The output signal is conditioned by an INA3221 amplifier before being connected to the main control unit. Temperature monitoring uses 16 K-type thermocouples (0.2mm wire diameter), distributed in a 4×4 grid. Eight thermocouples are embedded in the cell gaps, four are attached to the shell surface, and four are close to the gas sensors. The thermocouple signals are converted by a MAX31856 chip (resolution 0.0625℃) at a sampling frequency of 500Hz. The pressure sensor is an MPX5700 (range 0-700kPa), embedded in a pre-drilled hole inside the battery top cover and sealed with silicone. Of the six strain gauges (BF120-3AA), four axial strain gauges are attached to the long side of the square battery (20mm from the top and bottom edges, 30mm spacing), and two axial strain gauges are attached to the centers of the positive and negative end faces respectively. The strain signals are processed by an HX711 amplifier (gain 128), with a sampling frequency of 500Hz. The voltage acquisition module uses an ADS1256 analog-to-digital converter (24-bit resolution), connected to the positive and negative terminals of the battery via gold-plated probes. The sampling line length is 4cm, the sampling frequency is 1kHz, and only terminal voltage data is acquired. The main control unit uses an STM32H743 microprocessor (480MHz, 1MB RAM), integrating a BLE5.0 Bluetooth module and a power management unit, adapting to 12-24V power tool battery input, with overall power consumption controlled below 150mW.

[0152] The system software is divided into a data acquisition layer, a feature extraction layer, a stage judgment layer, and a risk warning layer according to the data flow. Each layer operates collaboratively through real-time operating system scheduling. The data acquisition layer synchronously reads raw data from each sensor every 10ms. For voltage data, the average of 100 sampling points is taken every 10ms to reduce the sampling frequency to 500Hz. Temperature compensation and outlier correction are performed simultaneously, and temperature and humidity cross-calibration is performed on gas concentration. The feature extraction layer calculates 12-dimensional time-domain features (including mean, peak value, rate of change, etc.) and 10-dimensional nonlinear interaction features (such as the gas-temperature coupling term F) every 100ms. i,j (t)=x i (t)×x j (t)+sin(x i (t)-x j (t) is used to construct a 32-dimensional feature vector. The stage decision layer calculates the gas ratios R1(t) and R2(t) every 500ms, based on...

[0153]

[0154] The thermal runaway stage is determined, and the final weight is calculated based on the stage adjustment matrix M(S). The risk warning layer calculates the risk index RI(t) every 100ms and outputs the warning level: Level 1 warning triggers a flashing green LED (cycle 1s), Level 2 warning activates a buzzer (1kHz, interval 1s) and Bluetooth alarm, and Level 3 warning triggers a relay to cut off the battery output (response time <50ms) and provides continuous audible and visual alarm.

[0155] like Figure 4 As shown, to verify the advantages of the multi-physics coupled dynamic early warning model of this invention, three mainstream lithium batteries for power tools were selected and divided into an experimental group (this model) and a control group (using the traditional single threshold-temperature gradient model, with a set temperature threshold T). th Temperature gradient threshold dT th The duration of continuous triggering is determined by monitoring the battery surface temperature and the rate of temperature change (thermal runaway is determined). After battery pretreatment, normal (1C charge-discharge 50 cycles) and abnormal operating conditions (overcharge, short circuit, high temperature storage) simulations are performed. Multi-physics parameters are collected and the warning / thermal runaway time is recorded. From the warning lead time, false alarm / missed alarm rate, and robustness under different battery / environment conditions, it can be seen that the model of this invention has greater advantages over the traditional model in terms of timeliness and accuracy.

[0156] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A battery thermal runaway prediction method based on an improved multimodal dynamic weight fusion algorithm, characterized in that: Includes the following steps: S1: Collect gas, temperature, pressure, strain, and voltage data of the lithium-ion battery; S2: A risk assessment model based on the multimodal dynamic weight fusion algorithm MM-DWFA preprocesses, extracts features, and fuses the collected data, transforms multi-source data into feature vectors, analyzes the feature vectors, and outputs early warning signals. S3: Execute the corresponding warning operation based on the warning signal.

2. The battery thermal runaway prediction method based on the improved multimodal dynamic weight fusion algorithm according to claim 1, characterized in that: In step S1, gas sensors are used to detect hydrogen, carbon monoxide, and carbon dioxide data. The gas concentration acquisition formula is as follows: C i (t)=k i ×V out,i (t)+b i Where k i b i V is the sensor calibration coefficient. out,i (t) represents the output voltage, and i includes three types: H2, CO, and CO2.

3. The battery thermal runaway prediction method based on the improved multimodal dynamic weight fusion algorithm according to claim 1, characterized in that: It also includes measuring temperature using a temperature sensor.

4. The battery thermal runaway prediction method based on the improved multimodal dynamic weight fusion algorithm according to claim 1, characterized in that: The pressure data is calculated using a dynamic pressure change model: in k represents the cumulative total concentration of the characteristic gas over time during thermal runaway. p k is the gas pressure coefficient. t P0 is the coefficient of thermal expansion; T(t) is the initial ambient pressure; T(t) is the temperature that changes with time; and T0 is the initial temperature. in These represent the cumulative concentrations of hydrogen, carbon monoxide, and carbon dioxide over time during thermal runaway.

5. The battery thermal runaway prediction method based on the improved multimodal dynamic weight fusion algorithm according to claim 1, characterized in that: The strain data includes radial strain and axial strain, wherein: The radial strain is calculated as follows: e radial (t)=0.35×ε R1 +0.15×e R2 +0.35×e R3 +0.15×e R4 Where ε radial (t) represents the radial strain, reflecting the overall deformation / failure risk of the structure in the radial direction; ε R1 ε represents the radial strain at the edge of the battery electrode. R2 ε represents the radial strain at the top of the battery casing. R3 ε represents the radial strain in the central region of the diaphragm. R4 Radial strain at the edge of the electrode core; The axial strain is calculated as follows: e axial (t)=0.5×ε A1 +0.5×e A2 Where ε A1 , ε A2 The strain is at two locations along the axial direction of the battery; The total strain is: ε total (t)=0.6×ε radial (t)+0.4×ε axial (t)。 6. The battery thermal runaway prediction method based on the improved multimodal dynamic weight fusion algorithm according to claim 1, characterized in that: The voltage data includes voltage fluctuations, calculated as follows: Where σ V (t) represents the voltage fluctuation rate; N represents the number of battery cells; The voltage of the i-th cell; Average voltage of the battery pack.

7. The battery thermal runaway prediction method based on the improved multimodal dynamic weight fusion algorithm according to claim 1, characterized in that: The risk assessment model processing steps based on the multimodal dynamic weight fusion algorithm MM-DWFA include: S21: Preprocess the collected and calculated data, including removing outliers and normalizing parameters, including: Using the improved 3σ criterion, a sliding window is constructed for each parameter, and the mean μ and standard deviation σ are calculated. When a data point satisfies |x(t)-μ|>3σ, it is replaced with the second nearest neighbor value within the window. x corr (t)=x(t-1)×0.7+x(t+1)×0.3 Map the data to the [0,1] interval and use a stage-adaptive threshold: S22: Perform feature extraction on the preprocessed data, including temporal features and cross-parameter coupling, including: Calculate the time-domain features of normalized data, including: Mean: Peak value: x peak =max(x i ) kurtosis: Rate of change: Construct a 5×5 coupling matrix and calculate the nonlinear interaction characteristics: F i,j (t)=x i (t)×x j (t)+sin(x i (t)-x j (t)) Where i and j represent different parameters; S23: Dynamically assign weights to each feature to achieve data fusion, including: First, calculate the correction index R1(t) for the concentration ratio of H2 to CO and the correction index R2(t) for the concentration ratio of CO to CO2: Thermal runaway is divided into the following four stages: Then, a smooth switching mechanism is used to achieve the weight transition: When the thermal runaway phase switches from S to S+1, the weights achieve a linear transition: W(S,t)=W0×M(S) Where W0 represents the initial weights, and M(S) is the stage adjustment matrix; S24: Based on the characteristics of fusion, assess the risk of thermal runaway and output early warning signals, including: Calculate the risk index: RI(t)=∑(F i,j (t)×W(S,t))+0.2×(ΠF i,j (t)) The warning levels are as follows:

8. The battery thermal runaway prediction method based on the improved multimodal dynamic weight fusion algorithm according to claim 1, characterized in that: The stage adjustment matrix M(S) includes: When S=0 during the thermal runaway phase, it indicates a normal state. At this time, the gas regulation coefficient is 0.57, the temperature regulation coefficient is 0.80, the pressure regulation coefficient is 1.00, the deformation regulation coefficient is 1.00, and the voltage regulation coefficient is 3.

00. When S=1 during the thermal runaway stage, it indicates the SEI decomposition period. At this time, the gas regulation coefficient is 1.71, the temperature regulation coefficient is 0.80, the pressure regulation coefficient is 0.33, the deformation regulation coefficient is 0.33, and the voltage regulation coefficient is 0.

33. When S=2 in the thermal runaway stage, it represents the electrolyte decomposition period. At this time, the gas regulation coefficient is 1.14, the temperature regulation coefficient is 1.20, the pressure regulation coefficient is 0.67, the deformation regulation coefficient is 0.67, and the voltage regulation coefficient is 0.

33. When S=3 in the thermal runaway stage, it represents the positive electrode decomposition period. At this time, the gas regulation coefficient is 0.57, the temperature regulation coefficient is 0.80, the pressure regulation coefficient is 2.00, the deformation regulation coefficient is 1.33, and the voltage regulation coefficient is 0.

33.

9. The battery thermal runaway prediction method based on the improved multimodal dynamic weight fusion algorithm according to claim 1, characterized in that: In step S3, a first-level warning triggers a flashing green LED, a second-level warning activates a buzzer and Bluetooth alarm, and a third-level warning triggers a relay to cut off battery output and provide continuous audible and visual alarms.

Images