Power battery thermal runaway diagnosis method and system based on hybrid prediction framework

Abstract

The application provides a power battery thermal runaway diagnosis method and system based on a hybrid prediction framework, sample data at time t of the battery is collected, and the sample data is preprocessed; a first-order equivalent circuit model of the power battery is constructed, and the least square method is used to identify the voltage, ohmic resistance, polarization resistance and polarization capacitance in the initial state; a SOC estimation model is constructed by using a current integration method; a battery internal temperature estimation model is constructed based on the internal Joule heat of the power battery and the heat exchange heat of the external environment; the standardized data in step S2 are input into a recurrent neural network GRU for training to obtain estimated sample data at time t+1; state space models are respectively constructed based on the SOC estimation model and the battery internal temperature estimation model; the estimated sample data is input into an adaptive unscented Kalman filter AUKF model as observation at time t+1 based on the state space models, and prediction data at time t+1 is calculated; and the thermal runaway is predicted through the prediction data.

Description

Technical Field

[0001] This invention relates to the field of power battery technology, and specifically to a method and system for diagnosing thermal runaway in power batteries based on a hybrid prediction framework. Background Technology

[0002] With the development of the electric vehicle industry, consumers' demands for the performance of electric vehicles are gradually increasing, especially in terms of driving range and battery capacity. Lithium-ion batteries have been widely used due to their high energy density, high specific energy, high voltage, long lifespan, and low self-discharge rate. However, under abnormal conditions, while the energy density of lithium-ion batteries continues to increase, the risk of thermal runaway due to heat from irreversible exothermic reactions also increases. Thermal runaway refers to the overheating phenomenon where an exothermic chain reaction occurs inside the battery, causing a rapid change in the battery's temperature rise rate.

[0003] There are many causes of thermal runaway in power batteries, broadly categorized as mechanical abuse due to impact, compression, or puncture; electrical abuse primarily caused by overcharging, over-discharging, and internal short circuits; and thermal abuse such as localized overheating. When a lithium-ion battery experiences thermal runaway, it can directly lead to the collapse of the battery pack structure, potentially causing a fire or explosion. Therefore, risk monitoring of the thermal runaway state of lithium-ion batteries is necessary.

[0004] Currently, thermal runaway state assessment methods can be broadly categorized into two types: one is detection methods based on thermal runaway characteristics. These methods primarily utilize characteristic signals of lithium-ion battery thermal runaway for detection. These include overcharging, over-discharging, and internal short-circuit detection for battery abuse; detection of lithium-ion battery-specific gas components during thermal runaway; and detection of lithium-ion battery surface temperature and temperature rise rate. Such methods are cumbersome, costly, and time-consuming. The second type is thermal runaway detection methods driven by big data. These methods ignore complex battery models and thermal runaway propagation processes and require the acquisition of large amounts of training data. To overcome the shortcomings of the above methods, this invention discloses an online thermal runaway state prediction method based on a hybrid model-driven and data-driven approach. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention proposes a method and system for diagnosing thermal runaway of power batteries based on a hybrid prediction framework. The method uses a recurrent neural network (GRU) to predict the battery's operating data and uses the predicted values ​​as observations for a model-driven method, thereby obtaining parameters that take into account the battery mechanism model and enabling the prediction of the battery's internal and external temperatures.

[0006] To achieve the above objectives, this invention provides a method for diagnosing thermal runaway of power batteries based on a hybrid prediction framework, characterized by the following steps:

[0007] Step S1: Collect sample data of the power battery at time t, including voltage, current, internal temperature, external temperature and ambient temperature;

[0008] Step S2: Preprocess the sample data and standardize it using the second moment of the density function to obtain standardized data;

[0009] Step S3: Construct a first-order equivalent circuit model of the power battery, and use the least squares method to identify the voltage, ohmic internal resistance, polarization internal resistance and polarization capacitance in the initial state;

[0010] Step S4: Construct a SOC estimation model using the current integration method; construct a battery internal temperature estimation model based on the internal Joule heat of the power battery and the heat exchange heat of the external environment;

[0011] Step S5: Input the standardized data described in step S2 into the recurrent neural network GRU for training to obtain the predicted sample data at time t+1;

[0012] Step S6: Based on the SOC estimation model and the battery internal temperature estimation model, construct state-space models respectively;

[0013] Step S7: Based on the state space model, use the estimated sample data as the observation input at time t+1 to the adaptive unscented Kalman filter (AUKF) model, and calculate the predicted data at time t+1.

[0014] Step S8: Set a threshold and compare the predicted data with the measured data to determine the thermal runaway phenomenon of the power battery.

[0015] Preferably, the method for preprocessing the sample data in step S2 is as follows: the sample data is transformed into a sample matrix X using the following expression:

[0016]

[0017] In the formula, x tj This represents the data of sample j at time t.

[0018] Preferably, the formula for the standardization process in step S2 is:

[0019]

[0020]

[0021] In the formula, Var() represents the second moment of the probability density function, and x j This represents the data for sample j over all time periods.

[0022] Preferably, the expression for the SOC estimation model in step S3 is:

[0023]

[0024] In the formula, SOC t Let SOC represent the estimated SOC at time t, where SOC0 represents the initial SOC, C represents the nominal capacity of the battery, ξ represents the correction factor, η represents the charge / discharge efficiency of the power battery, and I represents the SOC at time t. t This represents the charging and discharging current at time t.

[0025] Preferably, the expression for the battery internal temperature estimation model in step S3 is:

[0026] Q j =I 2 (R0+R d )

[0027] Q e =H(T) surf -T amb )

[0028]

[0029] In the formula, Q j This represents the Joule heat inside the power battery, I represents the current, R0 represents the initial voltage, and R d Q represents the polarization resistance. e T represents the amount of heat exchanged between the inside of the power battery and the external environment, where H represents the convective heat exchange coefficient. surf Indicates surface temperature, T amb The ambient temperature is represented by C, and T(t) represents the internal temperature of the power battery at time t. p Let m represent specific heat capacity, m represent mass, and Δt represent unit time.

[0030] Preferably, in step S5, the expression for the state-space model of the SOC estimation model is:

[0031]

[0032] In the formula, SOC0 represents the initial SOC, ξ t C represents the correction coefficient at time t. t I represents the nominal capacity at time t, η represents the charge / discharge efficiency of the power battery, and I t Represents the current I at time t t x t U represents the state at time t. t I represents the external input at time t. t .

[0033] Preferably, in step S5, the expression for the state-space model of the battery internal temperature estimation model is:

[0034]

[0035] In the formula, T(t-1) represents the internal temperature at time t-1, and H... t C represents the heat exchange coefficient at time t. p T represents specific heat capacity, m represents mass, and T represents the specific heat capacity. surf Indicates surface temperature, T amb Indicates ambient temperature, I t Represents the current I at time t t R0 represents the initial ohmic resistance, R d Let x represent the initial polarization resistance. t U represents the state at time t. t I represents the external input at time t. t .

[0036] Preferably, in step S8, the method for determining the thermal runaway phenomenon of the power battery includes the following steps:

[0037] Step S81: If the difference between the measured internal temperature of the battery and the predicted internal temperature of the battery is less than a set threshold, no thermal runaway occurs; otherwise, proceed to step S82.

[0038] Step S82: If the difference between the measured surface temperature and the predicted surface temperature is less than a set threshold, thermal shock occurs, leading to thermal runaway; if the difference between the measured surface temperature and the predicted surface temperature is greater than a set threshold, overshoot occurs, leading to thermal runaway.

[0039] This invention also proposes a power battery thermal runaway diagnostic system based on a hybrid prediction framework. The system includes a data acquisition module, a data preprocessing module, a recurrent neural network prediction module, a state-space model construction module, a prediction module, and a thermal runaway monitoring module.

[0040] The data acquisition module is used to acquire sample data of the power battery at time t. The sample data includes voltage, current, internal temperature, external temperature and ambient temperature, and outputs it to the data preprocessing module.

[0041] The data preprocessing module is used to preprocess the sample data and standardize it using the second moment of the density function to obtain standardized data, which is then output to the recurrent neural network prediction module.

[0042] The recurrent neural network prediction module is used to train the standardized data into a recurrent neural network GRU to obtain the predicted sample data at time t+1, and output it to the prediction module.

[0043] The state-space model construction module is used to construct a SOC estimation model using the current integration method, construct a battery internal temperature estimation model based on the internal Joule heat of the power battery and the heat exchange heat of the external environment, and construct a state-space model.

[0044] The prediction module is used to employ an adaptive unscented Kalman filter (AUKF) model and, based on the state-space model and the estimated sample data, calculate the predicted data at time t+1.

[0045] The thermal runaway monitoring module is used to set a threshold and compare the predicted data with the measured data to determine whether thermal runaway has occurred.

[0046] Preferably, the thermal runaway monitoring module determines whether thermal runaway has occurred by comprising the following steps: if the difference between the measured internal battery temperature and the predicted internal battery temperature is less than a set threshold, no thermal runaway has occurred; otherwise, proceed to subsequent judgments; if the difference between the measured external surface temperature and the predicted external surface temperature is less than a set threshold, thermal shock has occurred, leading to thermal runaway; if the difference between the measured external surface temperature and the predicted external surface temperature is greater than a set threshold, overshoot has occurred, leading to thermal runaway.

[0047] The beneficial effects of this invention are as follows: It uses a recurrent neural network (GRU) to predict battery operating data, and then uses these predicted values ​​as observations for a model-driven method. This yields parameters that consider the battery mechanism model, enabling the prediction of both the internal and external temperatures of the battery. The iterative prediction method, which combines data-driven prediction using the GRU and model-driven prediction using an adaptive unscented Kalman filter (AUKF) model, can analyze the mechanism of battery thermal runaway and accurately predict non-mechanistic parameters, effectively improving the fault tolerance and accuracy of thermal runaway state monitoring. Attached Figure Description

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

[0049] Figure 2 This is a schematic diagram of part of the algorithm flow of the present invention;

[0050] Figure 3 This is a schematic diagram of the thermal runaway judgment process of the present invention. Detailed Implementation

[0051] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0052] like Figures 1-2 As shown, the present invention proposes a method for diagnosing thermal runaway of power batteries based on a hybrid prediction framework, comprising the following steps:

[0053] Step S1: Collect sample data of the power battery at time t, including voltage, current, internal temperature, external temperature and ambient temperature;

[0054] Step S2: Preprocess the sample data and standardize it using the second moment of the density function to obtain standardized data;

[0055] In order to enable the sample data obtained in step S1 to be input into the recurrent neural network GRU for training, the present invention adopts the following method to preprocess the sample data to obtain standardized data.

[0056] Specifically, the sample data is transformed into a sample matrix X:

[0057]

[0058] In the formula, x tj This represents sample data at time t. In this embodiment of the invention, the sample data includes voltage U, current I, and internal temperature T. in Surface temperature T surf Ambient temperature T amb .

[0059] The following formula was then used for standardization:

[0060]

[0061]

[0062] In the formula, Var() represents the second moment of the probability density function, and x j This represents the data for sample j over all time periods.

[0063] Step S3: Construct a first-order equivalent circuit model of the power battery, and use the least squares method to identify the voltage U0, ohmic internal resistance R0, and polarization internal resistance R in the initial state. d With polarization capacitor C d Constructing a first-order equivalent circuit model of a power battery and using the least squares method to identify parameters in the initial state are conventional techniques for those skilled in the art, and therefore will not be elaborated upon here.

[0064] Step S4: Construct a SOC estimation model using the current integration method; construct a battery internal temperature estimation model based on the internal Joule heat of the power battery and the heat exchange heat of the external environment;

[0065] Specifically, the expression for constructing the SOC estimation model using the current integral method is as follows:

[0066]

[0067] In the formula, SOC t Let SOC represent the estimated SOC at time t, where SOC0 represents the initial SOC, C represents the nominal capacity of the battery, ξ represents the correction factor, η represents the charge / discharge efficiency of the power battery, and I represents the SOC at time t. t This represents the charging and discharging current at time t.

[0068] The internal heat of the battery is mainly caused by internal Joule heating Q. j Heat exchange with the external environment Q e Therefore, a battery internal temperature estimation model can be constructed based on its composition, and its expression is:

[0069] Q j =I 2 (R0+R d )

[0070] Q e =H(T) surf -T amb )

[0071]

[0072] In the formula, Q j This represents the Joule heat inside the power battery, I represents the current, R0 represents the initial voltage, and R d Q represents the polarization resistance. e T represents the amount of heat exchanged between the inside of the power battery and the external environment, where H represents the convective heat exchange coefficient. surf Indicates surface temperature, T amb The ambient temperature is represented by C, and T(t) represents the internal temperature of the power battery at time t. p Let m represent specific heat capacity, m represent mass, and Δt represent unit time.

[0073] Step S5: Input the standardized data described in step S2 into the recurrent neural network GRU for training to obtain the predicted sample data at time t+1;

[0074] The following explains the principle of constructing a recurrent neural network (GRU) for iterative training:

[0075] Step A), Input: x input =concat[h t-1 ,x t ];

[0076] Where concat[] means combining text from multiple regions or strings, h t-1 x represents the state transmitted from the previous file. t The input for the current node is [x] in the sample matrix. t1 ,x t2 ,…,x tj ].

[0077] Step B), reset the gate neuron: r t =σ(x input W τ +b τ );

[0078] Where σ() is the reset gate activation function, which transforms the data into values ​​within the range of 0-1, thus acting as a gating signal. W τ b τ To reset the parameters of the gate neuron;

[0079] Step C), memory gate neurons:

[0080] Where tanh() is the memory gate activation function, W h b h For the parameters of the memory gate neuron, ⊙ represents the element-wise product;

[0081] Step D), Input gate neuron: z t =σ(x input W z +b z );

[0082] Where z t This represents the gate that controls the update at time t, where σ() is the input gate activation function, and W z b z These are the parameters of the input gate neuron;

[0083] Step E), input and memorize:

[0084] This indicates that the current node information is included. To selectively "memorize";

[0085] Step F), forgetting gate neuron: f t =1-z t ;

[0086] f t This represents the forget gate, where h is forgotten. t-1 Some unimportant information in the dimension;

[0087] Step G), remember after forgetting: h′t-1 =f t ⊙h t-1 ;

[0088] Step H), memorization at time t:

[0089] The sample data is input into the recurrent neural network GRU for iterative training using the above method to obtain the predicted sample data at time t+1.

[0090] Step S6: Based on the SOC estimation model and the battery internal temperature estimation model, construct state-space models respectively;

[0091] In order to accurately estimate the state parameters, this embodiment of the invention describes the entire battery state system as a nonlinear system, which is expressed by the following formula:

[0092]

[0093] In the formula, x t Let y represent the system state at time t, C be the state transition function, and y be the system state at time t. t u t These are the observations at time t and the external input, respectively; g() is the observation equation; v t-1 w t These represent the process noise and observation noise at time t, respectively. The process and measurement should follow a Gaussian distribution. t-1 w t There are two covariance matrices, Q and R, with zero mean and zero variance, respectively.

[0094] Based on the nonlinear system constructed above, and using the SOC estimation model and the battery internal temperature estimation model constructed in step S4, the state-space equations are constructed as follows:

[0095] 1) State-space model of SOC estimation model:

[0096] x t =[ξ t C t ] T

[0097]

[0098] In the formula, SOC0 represents the initial SOC, ξ t C represents the correction coefficient at time t. t I represents the nominal capacity at time t, η represents the charge / discharge efficiency of the power battery, and I t Represents the current I at time t t x t U represents the state at time t.t I represents the external input at time t. t .

[0099] 2) State-space model of battery internal temperature estimation model:

[0100] x t =H t

[0101]

[0102] In the formula, T(t-1) represents the internal temperature at time t-1, and H... t C represents the heat exchange coefficient at time t. p T represents specific heat capacity, m represents mass, and T represents the specific heat capacity. surf Indicates surface temperature, T amb Indicates ambient temperature, I t Represents the current I at time t t R0 represents the initial ohmic resistance, R d Let x represent the initial polarization resistance. t U represents the state at time t. t I represents the external input at time t. t .

[0103] Step S7: Based on the state space model, use the estimated sample data as the observation input at time t+1 to the adaptive unscented Kalman filter (AUKF) model, and calculate the predicted data at time t+1.

[0104] The following explains the calculation method for the Adaptive Unscented Kalman Filter (AUKF) model:

[0105] Step A), initialization of the state and its covariance:

[0106]

[0107] Where x0 is the initial state setting, E() represents the expected value, and Cov() represents the covariance. Let P be the mathematical expectation of the initial state variables. 0∣0 This indicates the initialization of covariance.

[0108] Step B) The unscented transformation is used to transform the n-dimensional state vector into an n+2-dimensional state vector:

[0109]

[0110] in Let be the expected value at time t-1, n be the dimension of the state variables, and λ be the scaling parameter. This represents the error covariance at time t-1.

[0111] Step C) Calculate the corresponding weights:

[0112]

[0113] Where ω m ω is the mean of the sampled points. c Let α be the covariance of the sampling points, β be the parameter to be selected, the selection of α controls the distribution of the sampling points, and λ be the scaling parameter.

[0114] Step D) Predict the state and prior error covariance First update:

[0115]

[0116]

[0117]

[0118] Where A is the state transition matrix. The average of the sampled points. For the sampling point covariance, Let Q be the mathematical expectation at time t. x Let be the covariance between state variables.

[0119] Step E) Determine the updated state variable covariance At time k|k-1, a new Sigma point is generated:

[0120]

[0121] in Let be the expected value at time t, n be the dimension of the state variables, and λ be the scaling parameter. This represents the error covariance at time t.

[0122] Step F) Measurement update based on output equation and measurement noise:

[0123]

[0124]

[0125]

[0126]

[0127] in These are the estimated values ​​of the observations at time t and the average value of the estimated observations, respectively. The system equations at time t are used to estimate the covariance, R. xThe covariance of the observation equation.

[0128] Step G) The covariance matching method is used to adaptively adjust the covariance R:

[0129]

[0130]

[0131]

[0132] Where y t , Let z be the actual and estimated voltage values ​​at time t, respectively. t F represents the residual between the estimated voltage value and the actual voltage value. t Represents the voltage residual value z t The approximate value of the covariance is given by L, where L is the window size of the covariance matching method.

[0133] Step H) Calculate the Kalman gain K, and update the system state and error covariance based on system measurements:

[0134]

[0135] Q x =KF t K T

[0136]

[0137]

[0138] Where K is the Kalman gain. These represent the updated system state and error covariance, respectively.

[0139] Step I), using the system state variables obtained in step H) at time t+1, respectively, are substituted into the SOC mechanism model and the internal temperature T. in In the mechanistic model, the predicted data at time t+1 is calculated. In this embodiment of the invention, the predicted data includes the internal temperature T′ at time t+1. in The surface temperature T′ at time (t+1) surf The voltage U′(t+1) at time t+1 and the charge value SOC′(t+1) at time t+1.

[0140] Step S8: Set a threshold and compare the predicted data with the measured data to determine the thermal runaway phenomenon of the power battery.

[0141] Specifically, such as Figure 3 As shown, the judgment conditions set in the embodiments of the present invention are as follows:

[0142] 1) Set the battery voltage monitoring threshold γ3. When the predicted voltage U′(t+1) is greater than γ3, the battery overcharge alarm will sound. At the same time, analyze whether the predicted SOC′(t+1) is greater than 99. If it is, charging needs to be stopped to prevent thermal runaway caused by overcharging.

[0143] 2) Set monitoring thresholds γ1 and γ2 for the internal and external temperatures of the battery respectively. If the difference between the measured internal temperature of the battery and the predicted internal temperature is less than the set threshold γ1, no thermal runaway occurs. If the difference between the measured external temperature and the predicted external temperature is less than the set threshold γ2, thermal shock occurs, leading to thermal runaway. If the difference between the measured external temperature and the predicted external temperature is greater than the set threshold, overshoot occurs, leading to thermal runaway.

[0144] This invention also proposes a power battery thermal runaway diagnostic system based on a hybrid prediction framework, characterized in that: the system includes a data acquisition module, a data preprocessing module, a recurrent neural network prediction module, a state space model construction module, a prediction module, and a thermal runaway monitoring module;

[0145] The data acquisition module is used to acquire sample data of the power battery at time t. The sample data includes voltage, current, internal temperature, external temperature and ambient temperature, and outputs it to the data preprocessing module.

[0146] The data preprocessing module is used to preprocess the sample data and standardize it using the second moment of the density function to obtain standardized data, which is then output to the recurrent neural network prediction module.

[0147] The recurrent neural network prediction module is used to train the standardized data into a recurrent neural network GRU to obtain the predicted sample data at time t+1, and output it to the prediction module.

[0148] The state-space model construction module is used to construct a SOC estimation model using the current integration method, construct a battery internal temperature estimation model based on the internal Joule heat of the power battery and the heat exchange heat of the external environment, and construct a state-space model.

[0149] The prediction module is used to employ an adaptive unscented Kalman filter (AUKF) model and, based on the state-space model and the estimated sample data, calculate the predicted data at time t+1.

[0150] The thermal runaway monitoring module is used to set a threshold and compare the predicted data with the measured data to determine whether thermal runaway has occurred. Specifically, the method for determining whether thermal runaway has occurred includes the following steps: if the difference between the measured internal battery temperature and the predicted internal battery temperature is less than the set threshold, no thermal runaway has occurred; otherwise, proceed to the next judgment. If the difference between the measured external surface temperature and the predicted external surface temperature is less than the set threshold, a thermal shock has occurred, leading to thermal runaway. If the difference between the measured external surface temperature and the predicted external surface temperature is greater than the set threshold, an overshoot has occurred, leading to thermal runaway.

[0151] Those skilled in the art will readily understand that the above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for diagnosing thermal runaway of power batteries based on a hybrid predictive framework, characterized in that: Includes the following steps: Step S1: Collect sample data of the power battery at time t, including voltage, current, internal temperature, external temperature and ambient temperature; Step S2: Preprocess the sample data and standardize it using the second moment of the density function to obtain standardized data; Step S3: Construct a first-order equivalent circuit model of the power battery, and use the least squares method to identify the voltage, ohmic internal resistance, polarization internal resistance and polarization capacitance in the initial state; Step S4: Construct a SOC estimation model using the current integration method; construct a battery internal temperature estimation model based on the internal Joule heat of the power battery and the heat exchange heat of the external environment; Step S5: Input the standardized data described in step S2 into the recurrent neural network GRU for training to obtain the predicted sample data at time t+1; Step S6: Based on the SOC estimation model and the battery internal temperature estimation model, construct state-space models respectively; Step S7: Based on the state space model, use the estimated sample data as the observation input at time t+1 to the adaptive unscented Kalman filter (AUKF) model, and calculate the predicted data at time t+1. Step S8: Set a threshold and compare the predicted data with the measured data to determine the thermal runaway phenomenon of the power battery.

2. The method for diagnosing thermal runaway of a power battery based on a hybrid prediction framework according to claim 1, characterized in that: The method for preprocessing the sample data in step S2 is as follows: the sample data is transformed into a sample matrix X using the following expression: ； In the formula, x tj This represents the data of sample j at time t.

3. The method for diagnosing thermal runaway of a power battery based on a hybrid prediction framework according to claim 2, characterized in that: The formula for the standardization process in step S2 is: ； In the formula, Let n represent the variance of sample j, and n be the total number of sampling times. This represents the data of sample j at time t after standardization.

4. The method for diagnosing thermal runaway of a power battery based on a hybrid prediction framework according to claim 1, characterized in that: The expression for the SOC estimation model in step S4 is: ； In the formula, SOC t Let SOC represent the estimated SOC at time t, where SOC0 represents the initial SOC, C represents the nominal capacity of the battery, ξ represents the correction factor, η represents the charge / discharge efficiency of the power battery, and I represents the SOC at time t. t This represents the charging and discharging current at time t.

5. The method for diagnosing thermal runaway of a power battery based on a hybrid prediction framework according to claim 4, characterized in that: The expression for the battery internal temperature estimation model in step S4 is: ； ； ； In the formula, Q j This represents the Joule heat inside the power battery, I represents the current, R0 represents the initial voltage, and R d Q represents the polarization resistance. e T represents the amount of heat exchanged between the inside of the power battery and the external environment, where H represents the convective heat exchange coefficient. surf Indicates surface temperature, T amb The ambient temperature is represented by C, and T(t) represents the internal temperature of the power battery at time t. p Let m represent specific heat capacity, m represent mass, and Δt represent unit time.

6. The method for diagnosing thermal runaway of a power battery based on a hybrid prediction framework according to claim 1, characterized in that: In step S6, the expression for the state-space model of the SOC estimation model is: ； ； In the formula, SOC0 represents the initial SOC, ξ t C represents the correction coefficient at time t. t I represents the nominal capacity at time t, η represents the charge / discharge efficiency of the power battery, and I t Represents the current I at time t t x t U represents the state at time t. t I represents the external input at time t. t .

7. The method for diagnosing thermal runaway of a power battery based on a hybrid prediction framework according to claim 6, characterized in that: In step S6, the expression for the state-space model of the battery internal temperature estimation model is: ； ； In the formula, T(t-1) represents the internal temperature at time t-1, and H... t C represents the heat exchange coefficient at time t. p T represents specific heat capacity, m represents mass, and T represents the specific heat capacity. surf Indicates surface temperature, T amb Indicates ambient temperature, I t Represents the current I at time t t R0 represents the initial ohmic resistance, R d Let x represent the initial polarization resistance. t U represents the state at time t. t I represents the external input at time t. t .

8. The method for diagnosing thermal runaway of a power battery based on a hybrid prediction framework according to claim 1, characterized in that: In step S8, the method for determining the thermal runaway phenomenon of the power battery includes the following steps: Step S81: If the difference between the measured internal temperature of the battery and the predicted internal temperature of the battery is less than a set threshold, no thermal runaway occurs; otherwise, proceed to step S82. Step S82: If the difference between the measured surface temperature and the predicted surface temperature is less than a set threshold, thermal shock occurs, leading to thermal runaway; if the difference between the measured surface temperature and the predicted surface temperature is greater than a set threshold, overshoot occurs, leading to thermal runaway.

9. A power battery thermal runaway diagnostic system based on a hybrid predictive framework, characterized in that: The system includes a data acquisition module, a data preprocessing module, a recurrent neural network prediction module, a state-space model construction module, a prediction module, and a thermal runaway monitoring module. The data acquisition module is used to acquire sample data of the power battery at time t. The sample data includes voltage, current, internal temperature, external temperature and ambient temperature, and outputs it to the data preprocessing module. The data preprocessing module is used to preprocess the sample data and standardize it using the second moment of the density function to obtain standardized data, which is then output to the recurrent neural network prediction module. The recurrent neural network prediction module is used to train the standardized data into a recurrent neural network GRU to obtain the predicted sample data at time t+1, and output it to the prediction module. The state-space model construction module is used to construct a SOC estimation model using the current integration method, construct a battery internal temperature estimation model based on the internal Joule heat of the power battery and the heat exchange heat of the external environment, and construct a state-space model. The prediction module is used to employ an adaptive unscented Kalman filter (AUKF) model and, based on the state-space model and the estimated sample data, calculate the predicted data at time t+1. The thermal runaway monitoring module is used to set a threshold and compare the predicted data with the measured data to determine whether thermal runaway has occurred.

10. A power battery thermal runaway diagnostic system based on a hybrid predictive framework according to claim 9, characterized in that: The thermal runaway monitoring module determines whether thermal runaway has occurred by comprising the following steps: if the difference between the measured internal battery temperature and the predicted internal battery temperature is less than a set threshold, no thermal runaway has occurred; otherwise, proceed to the next step. If the difference between the measured external surface temperature and the predicted external surface temperature is less than a set threshold, thermal shock has occurred, leading to thermal runaway. If the difference between the measured external surface temperature and the predicted external surface temperature is greater than a set threshold, overshoot has occurred, leading to thermal runaway.

Images