Lithium battery thermal runaway feature recognition algorithm based on BP neural network

Abstract

The invention relates to the technical field of battery management, and discloses a lithium battery thermal runaway feature recognition algorithm based on a BP neural network, and the algorithm comprises the following steps: data collection and preprocessing: employing a temperature sensor and a gas sensor to monitor the temperature and gas information of a lithium battery system in real time; the temperature change of the temperature sensor before and after the thermal runaway and the temperature change slope at the moment of the thermal runaway event are used as main characteristic parameters; for a gas sensor, gas concentration change before and after thermal runaway and gas concentration change slope are considered as main characteristic parameters; and taking the collected temperature, gas data and characteristic parameters as a data set. According to the invention, the temperature and gas changes of the lithium battery are monitored in real time through the temperature sensor and the gas sensor, the characteristics of thermal runaway can be quickly identified, and early warning can be timely given out, so that the real-time monitoring capability of the system is improved, and the accuracy and timeliness of thermal runaway prediction are enhanced.

Description

Technical Field

[0001] The present invention relates to the technical field of battery management, and in particular to a lithium battery thermal runaway feature recognition algorithm based on a BP neural network. Background Art

[0002] As a core component of modern electronic devices, transportation tools (such as electric vehicles) and energy storage systems, lithium-ion batteries are widely used for their high energy density, long life and low self-discharge rate. However, lithium batteries generate heat during operation, especially in the case of overcharge, over-discharge, short circuit or external impact, the temperature inside the battery will rise sharply, leading to "thermal runaway". Thermal runaway is a dangerous state that may cause the battery to spontaneously combust or explode, posing a serious threat to personal safety and equipment stability.

[0003] In order to deal with this problem, lithium battery thermal runaway detection and early warning technology has become a research focus. Related technologies monitor the internal temperature, gas composition and other parameters of the battery, and combine data analysis and machine learning algorithms to identify and predict the occurrence of thermal runaway in real time, thereby improving the safety of the battery system. Traditional battery management systems (BMS) mainly rely on rule threshold control and simple monitoring technology, which have certain limitations and make it difficult to achieve accurate thermal runaway prediction and early warning.

[0004] Existing battery management systems (BMS) usually use the following methods to monitor and manage lithium batteries:

[0005] Threshold detection method: Based on the monitoring of basic parameters such as battery temperature, pressure difference, and current, when these parameters exceed the preset safety threshold, the system will issue an alarm signal. For example, when the temperature rises to the set safety threshold, the system will shut down the charging or discharging process to avoid further temperature rise. Although this method is simple, it has hysteresis and inaccuracy, and cannot predict the occurrence of thermal runaway in advance.

[0006] Monitoring method based on equivalent circuit model (ECM): This method treats the battery as equivalent to circuit elements such as resistors and capacitors and analyzes changes in its electrical characteristics. Although this method can capture changes in the internal state of the battery, due to the significant nonlinear characteristics of the battery, the model is difficult to accurately reflect the actual thermodynamic state, especially in predicting thermal runaway.

[0007] Limited multi-sensor monitoring: Some systems introduce multiple sensors, such as temperature sensors and gas sensors, to monitor the temperature and gas concentration of the battery. However, most existing systems only trigger alarms when the battery exceeds a certain fixed parameter value, lacking in-depth analysis of the data, resulting in inaccurate warnings.

[0008] However, the inventors have discovered that the prior art still has the following defects:

[0009] Traditional threshold-based monitoring technology mainly relies on preset single physical parameters such as temperature and voltage to monitor battery status. This method can only trigger an alarm when the battery reaches its limit conditions. It lacks the ability to provide early warning of potential problems, resulting in the inability to effectively prevent the occurrence of thermal runaway.

[0010] To this end, we propose a lithium battery thermal runaway feature recognition algorithm based on BP neural network. Summary of the invention

[0011] The present invention mainly solves the technical problems existing in the above-mentioned prior art and provides a lithium battery thermal runaway feature recognition algorithm based on BP neural network.

[0012] In order to achieve the above object, the present invention adopts the following technical solution, a lithium battery thermal runaway feature recognition algorithm based on BP neural network, comprising the following steps:

[0013] The first step is data collection and preprocessing: use temperature and gas sensors to monitor the temperature and gas information of the lithium battery system in real time; use the temperature change of the temperature sensor before and after thermal runaway, and the temperature change slope at the time of thermal runaway as the main characteristic parameters; for the gas sensor, consider the change of gas concentration before and after thermal runaway, and the gas concentration change slope as the main characteristic parameters; use the collected temperature, gas data and characteristic parameters as a data set;

[0014] The second step is feature extraction: PCA is used as the feature extraction algorithm. The PCA algorithm can reduce the dimension of feature data to two or three dimensions of the original data, and normalize the feature data to satisfy the standard normal distribution.

[0015] The third step is model construction: use the main feature components to build a BP neural network model, use the error back propagation algorithm to optimize the model parameters, and improve the model prediction accuracy and stability;

[0016] Step 4: Prediction and decision-making: The data obtained from real-time monitoring is input into the neural network model. If an abnormality is detected, an alarm will be issued; if there is no abnormality, the model will continue to monitor in real time to ensure the safety of the system.

[0017] Step 5: Result interpretation: The results are interpreted through the lithium battery thermal runaway model and verified using the corresponding fault interpretation algorithm.

[0018] Preferably, the first step of obtaining the temperature change before and after the thermal runaway occurs uses a temperature sensor to collect data, and the collected data needs to be feature extracted before being input into the model as a data set. Since multiple temperature sensors collect data at the same time, the data needs to be filtered, and the data collected by the same ambient temperature sensor is selected as the input of the prediction model. After the data preprocessing, it needs to be normalized to make the model converge faster.

[0019] Preferably, in the second step, the PCA algorithm is used as the feature extraction algorithm. The PCA algorithm is mainly based on the collinearity and covariance characteristics between the feature data, and can reduce the dimension of the feature data to two or three dimensions of the original data, and normalize the feature data to satisfy the standard normal distribution.

[0020] Preferably, the BP neural network model in the third step includes an input layer, a hidden layer and an output layer, wherein the input layer is used to input data features, the hidden layer acts as an intermediate neuron, and the output layer is the final result to be predicted. When the data is input into the trained BP neural network model, the output layer outputs the final prediction result.

[0021] Preferably, the prediction model for thermal runaway in the fourth step is predicted by a trained prediction model to predict whether the lithium battery will experience thermal runaway under actual working conditions, and when it is predicted that the lithium battery is about to experience thermal runaway, the model issues a warning in a timely manner.

[0022] Preferably, the specific method for establishing the BP neural network model in the third step is as follows: first, determine the input variables and output variables of the neural network, select temperature and gas concentration data as input features to reduce model complexity and improve prediction accuracy, and select thermal runaway flag as output variable; determine the number of hidden layer nodes according to the number of input and output variables, and generally use the number of hidden layer nodes formula to calculate the number of hidden layer nodes, the formula is as follows: Where H represents the number of hidden layer nodes, I represents the number of input layer nodes, O represents the number of output layer nodes, and e represents a value between 10 -4 A constant between 10 and 10; the excitation function of the model uses the tanh or sigmoid function, where the tanh excitation function uses the S-shaped function, which is as follows: the sigmoid excitation function uses a piecewise linear function, as follows: Where e is the base of the natural logarithm, x is the function independent variable, and S(x) represents the function result; the activation function of the hidden layer is selected as follows: h1=tanh(S1), h0=sigmoid(S0), where S1 and S0 represent the output values ​​of the two hidden layer nodes respectively, and the activation function of the output layer is: o=a·s0+b·s1, where a and b represent the weights between the output layer and the hidden layer.

[0023] As a preference, the back propagation algorithm is used in the BP neural network model to optimize the model parameters and improve the prediction accuracy and stability. In the BP neural network, the nonlinearity of the activation function of the neuron will cause the gradient descent search to fall into a local extreme value when the surface of the error function converges, thereby destroying the prediction accuracy. The adaptive moment estimation (Adam) optimization algorithm is used to optimize the model parameters. The algorithm can calculate the network learning speed and adjust the adaptive weights by analyzing the historical gradient changes. The specific method is as follows: First, the expected value and square difference of the gradient are calculated respectively: where t is the current training round, E is the expectation of the historical gradient value, D is the square difference of the historical gradient value, is the global learning rate, θt represents the network parameter value at time t, θ t-1 represents the network parameter value at time t-1, represents the speed of global learning, ρ1 and ρ2 represent the exponential decay rate of the global learning rate; then calculate the first-order moment estimate and second-order moment estimate of the global offset: where m represents the first-order moment estimate of the offset, n represents the second-order moment estimate of the offset, is the global learning rate, and represents the deviation of the global learning speed: where represents the first-order moment estimate deviation, represents the second-order moment estimate deviation, δ and ε are two small positive numbers used to prevent the denominator from being 0; calculate the deviation correction: where ms represents the deviation of the historical first-order moment estimate, and ns represents the deviation of the historical second-order moment estimate; finally calculate the iterative update of the parameters: where α represents the final update step size, m represents the first-order moment estimate, and n represents the second-order moment estimate.

[0024] Preferably, the weights in the BP algorithm are calculated based on the loss function, and the loss function L is used to guide the optimization process. In the loss function, if the difference between the predicted value and the known value is small, the loss function is small; on the contrary, if the difference between the predicted value and the known value is large, the loss function will also be high.

[0025] Preferably, the fifth step of model result verification is as follows: first, the data within a period of time before and after the occurrence of thermal runaway is used as a training set, and the model is trained using the training set data; then, a test data set is used to test the accuracy and reliability of the model; based on the prediction effect and model accuracy, if the model needs to be improved, the training set data is used for training again; the performance comparison of the entire training and test data sets is used as the performance parameter for the final evaluation; the prediction results of the lithium battery thermal runaway prediction model are compared with the expert system to verify the accuracy of the prediction results, wherein the expert system is a decision-making model based on expert experience and knowledge and has a higher weight, and the test data set is used to verify whether the thermal runaway predicted by the model is consistent with the actual situation to ensure the accuracy of the model; in the thermal runaway scenario, historical data is used for training, and the test data may include data other than historical data; in conventional scenarios, the model is trained using training data, and then the data set used for testing includes data outside the range of the training data.

[0026] Preferably, in the actual application of the lithium battery thermal runaway prediction model in the fourth step, the thermal runaway of the lithium battery has a certain probability, and the model outputs the possibility of thermal runaway, that is, the thermal runaway with a prediction value greater than 0.5 is considered to have a high probability of thermal runaway, and the thermal runaway with a prediction value less than 0.5 is considered to have a low probability of thermal runaway. When the output probability value is between 0.5-0.7, the uncertainty of the model prediction result is strong, and thermal runaway may or may not occur. It is necessary to further monitor the prediction of the model and environmental factors to make decisions; the lithium battery thermal runaway prediction model can predict multiple thermal runaway events. When multiple prediction values ​​are greater than 0.5, the risk of thermal runaway is high, and corresponding measures (early warning) should be taken in time to reduce the risk or replace the battery; if multiple thermal runaway prediction values ​​are less than 0.5, the risk of thermal runaway is low, but monitoring still needs to be continued; when multiple consecutive thermal runaway prediction values ​​are less than 0.5, the risk is further reduced and the system can continue to run, and the monitoring frequency needs to be reduced.

[0027] Beneficial Effects

[0028] The present invention provides a lithium battery thermal runaway feature recognition algorithm based on BP neural network. It has the following beneficial effects:

[0029] (1) This lithium battery thermal runaway feature recognition algorithm based on BP neural network monitors the temperature and gas changes of lithium batteries in real time through temperature sensors and gas sensors, can quickly identify the characteristics of thermal runaway and issue early warnings in a timely manner, improves the real-time monitoring capability of the system, and enhances the accuracy and timeliness of thermal runaway prediction. The principal component analysis (PCA) algorithm is used for feature extraction and dimensionality reduction, which can effectively reduce the dimension of the data while retaining the main features of the data. This not only simplifies the complexity of the model, but also improves the operating efficiency and prediction performance of the model.

[0030] (2) This lithium battery thermal runaway feature recognition algorithm based on BP neural network can improve the prediction accuracy and stability of the model, reduce the problems of overfitting and local optimal traps, and thus ensure the stability of the prediction by using the error back propagation algorithm (BP algorithm) and adaptive moment estimation (Adam) to optimize the model parameters.

[0031] (3) This lithium battery thermal runaway feature recognition algorithm based on BP neural network can not only predict a single thermal runaway event, but also handle the prediction of multiple thermal runaway events, and can perform risk classification management according to the probability of thermal runaway. When the predicted value is greater than a certain threshold, the system will issue an early warning to reduce the risk of thermal runaway.

[0032] (4) This lithium battery thermal runaway feature recognition algorithm based on BP neural network can improve the convergence speed and stability of the model by filtering, feature extraction and normalization of data from multiple sensors, ensuring that the prediction model can obtain accurate results in a shorter time.

[0033] (5) This lithium battery thermal runaway feature recognition algorithm based on BP neural network further enhances the credibility of the prediction by verifying the accuracy of the model prediction results through fault interpretation algorithm and expert system. This verification mechanism ensures that the model not only performs well on experimental data, but also can accurately predict the thermal runaway of lithium batteries in practical applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the implementation of the present invention or the technical solution in the prior art, the following briefly introduces the drawings required for the implementation or the description of the prior art. Obviously, the drawings in the following description are only exemplary, and for ordinary technicians in this field, other implementation drawings can be derived from the provided drawings without creative work.

[0035] The structures, proportions, sizes, etc. illustrated in this specification are only used to match the contents disclosed in the specification so as to facilitate understanding and reading by persons familiar with the technology. They are not used to limit the conditions under which the present invention can be implemented, and therefore have no substantial technical significance. Any structural modification, change in proportion or adjustment of size shall still fall within the scope of the technical contents disclosed in the present invention without affecting the effects and purposes that can be achieved by the present invention.

[0036] Figure 1 is a system architecture diagram of the present invention;

[0037] Figure 2 It is a flow chart of data collection and feature extraction in the present invention;

[0038] Figure 3 It is the structure diagram of BP neural network in the present invention;

[0039] Figure 4 It is a diagram of the BP neural network training process in the present invention;

[0040] Figure 5 This is a diagram showing the thermal runaway prediction results in the present invention;

[0041] Figure 6 It is a thermal runaway risk assessment view in the present invention;

[0042] Figure 7 This is a flowchart of fault explanation and diagnosis in the present invention. DETAILED DESCRIPTION

[0043] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0044] Embodiment: A lithium battery thermal runaway feature recognition algorithm based on BP neural network, such as Figure 1-Figure 7 As shown, the following steps are included:

[0045] The first step is data collection and preprocessing: use temperature and gas sensors to monitor the temperature and gas information of the lithium battery system in real time; use the temperature change of the temperature sensor before and after thermal runaway, and the temperature change slope at the time of thermal runaway as the main characteristic parameters; for the gas sensor, consider the change of gas concentration before and after thermal runaway, and the gas concentration change slope as the main characteristic parameters; use the collected temperature, gas data and characteristic parameters as a data set;

[0046] The second step is feature extraction: PCA is used as the feature extraction algorithm. The PCA algorithm can reduce the dimension of feature data to two or three dimensions of the original data, and normalize the feature data to satisfy the standard normal distribution.

[0047] The third step is model construction: use the main feature components to build a BP neural network model, use the error back propagation algorithm to optimize the model parameters, and improve the model prediction accuracy and stability;

[0048] Step 4: Prediction and decision-making: The data obtained from real-time monitoring is input into the neural network model. If an abnormality is detected, an alarm will be issued; if there is no abnormality, the model will continue to monitor in real time to ensure the safety of the system.

[0049] Step 5: Result interpretation: The results are interpreted through the lithium battery thermal runaway model and verified using the corresponding fault interpretation algorithm.

[0050] Furthermore, the first step of obtaining the temperature change before and after thermal runaway occurs uses a temperature sensor to collect data, and before the collected data is input into the model as a data set, it is necessary to extract features from the data. Since multiple temperature sensors collect data at the same time, the data needs to be filtered, and the data collected by the same ambient temperature sensor is selected as the input of the prediction model. After data preprocessing, normalization processing is also required to make the model converge faster.

[0051] Furthermore, in the second step, the PCA algorithm is used as the feature extraction algorithm. The PCA algorithm is mainly based on the collinearity and covariance characteristics between the feature data, and can reduce the dimension of the feature data to two or three dimensions of the original data, and normalize the feature data to satisfy the standard normal distribution.

[0052] Furthermore, the BP neural network model in the third step includes an input layer, a hidden layer and an output layer, wherein the input layer is used to input data features, the hidden layer acts as an intermediate neuron, and the output layer is the final result to be predicted. When the data is input into the trained BP neural network model, the output layer outputs the final prediction result.

[0053] Furthermore, the prediction model for thermal runaway in the fourth step is predicted by a trained prediction model to predict whether the lithium battery will experience thermal runaway under actual working conditions, and when it is predicted that the lithium battery is about to experience thermal runaway, the model issues a warning in a timely manner.

[0054] Furthermore, the specific method for establishing the BP neural network model in the third step is as follows: first, determine the input variables and output variables of the neural network, select temperature and gas concentration data as input features to reduce model complexity and improve prediction accuracy, and select thermal runaway flag as output variable; determine the number of hidden layer nodes according to the number of input and output variables, and generally use the number of hidden layer nodes formula to calculate the number of hidden layer nodes, the formula is as follows: Where H represents the number of hidden layer nodes, I represents the number of input layer nodes, O represents the number of output layer nodes, and e represents a value between 10 -4 A constant between 10 and 10; the excitation function of the model uses the tanh or sigmoid function, where the tanh excitation function uses the S-shaped function, which is as follows: the sigmoid excitation function uses a piecewise linear function, as follows: Where e is the base of the natural logarithm, x is the function independent variable, and S(x) represents the function result; the activation function of the hidden layer is selected as follows: h1=tanh(S1), h0=sigmoid(S0), where S1 and S0 represent the output values ​​of the two hidden layer nodes respectively, and the activation function of the output layer is: o=a·s0+b·s1, where a and b represent the weights between the output layer and the hidden layer.

[0055] Furthermore, the back propagation algorithm is used in the BP neural network model to optimize the model parameters and improve the prediction accuracy and stability. In the BP neural network, the nonlinearity of the neuron activation function will cause the gradient descent search to fall into the local extreme value when the surface of the error function converges, thereby destroying the prediction accuracy. The adaptive moment estimation (Adam) optimization algorithm is used to optimize the model parameters. The algorithm can calculate the network learning speed and adjust the adaptive weights by analyzing the historical gradient changes. The specific method is as follows: First, the expected value and square difference of the gradient are calculated respectively: where t is the current training round, E is the expectation of the historical gradient value, D is the square difference of the historical gradient value, is the global learning rate, θt represents the network parameter value at time t, and θ t-1 represents the network parameter value at time t-1, represents the speed of global learning, ρ1 and ρ2 represent the exponential decay rate of the global learning rate; then calculate the first-order moment estimate and second-order moment estimate of the global offset: where m represents the first-order moment estimate of the offset, n represents the second-order moment estimate of the offset, is the global learning rate, and represents the deviation of the global learning speed: where represents the first-order moment estimate deviation, represents the second-order moment estimate deviation, δ and ε are two small positive numbers used to prevent the denominator from being 0; calculate the deviation correction: where ms represents the deviation of the historical first-order moment estimate, and ns represents the deviation of the historical second-order moment estimate; finally calculate the iterative update of the parameters: where α represents the final update step size, m represents the first-order moment estimate, and n represents the second-order moment estimate.

[0056] Furthermore, the calculation of weights in the BP algorithm is based on the loss function, and the loss function L is used to guide the optimization process. In the loss function, if the difference between the predicted value and the known value is small, the loss function is small; on the contrary, if the difference between the predicted value and the known value is large, the loss function will also be high.

[0057] Furthermore, the fifth step of model result verification is as follows: first, the data within a period of time before and after the occurrence of thermal runaway is used as a training set, and the model is trained using the training set data; then, a test data set is used to test the accuracy and reliability of the model; based on the prediction effect and model accuracy, if the model needs to be improved, the training set data is used again for training; the performance comparison of the entire training and test data sets is used as the performance parameter for the final evaluation; the prediction results of the lithium battery thermal runaway prediction model are compared with the expert system to verify the accuracy of the prediction results, wherein the expert system is a decision-making model based on expert experience and knowledge and has a higher weight, and the test data set is used to verify whether the thermal runaway predicted by the model is consistent with the actual situation to ensure the accuracy of the model; in the thermal runaway scenario, historical data is used for training, and the test data may include data other than historical data; in conventional scenarios, the model is trained using training data, and then the data set used for testing includes data outside the range of the training data.

[0058] Furthermore, in the actual application of the lithium battery thermal runaway prediction model in the fourth step, the thermal runaway of the lithium battery has a certain probability. The model outputs the possibility of thermal runaway, that is, the thermal runaway with a prediction value greater than 0.5 is considered to have a high probability of thermal runaway, and the thermal runaway with a prediction value less than 0.5 is considered to have a low probability of thermal runaway. When the output probability value is between 0.5-0.7, the uncertainty of the model prediction result is strong, and thermal runaway may or may not occur. It is necessary to further monitor the prediction of the model and environmental factors to make decisions; the lithium battery thermal runaway prediction model can predict multiple thermal runaway events. When multiple prediction values ​​are greater than 0.5, the risk of thermal runaway is high, and corresponding measures (early warning) should be taken in time to reduce the risk or replace the battery; if multiple thermal runaway prediction values ​​are less than 0.5, the risk of thermal runaway is low, but monitoring still needs to be continued; when multiple consecutive thermal runaway prediction values ​​are less than 0.5, the risk is further reduced and the system can continue to operate, and the monitoring frequency needs to be reduced.

[0059] The above shows and describes the basic principles and main features of the present invention and the advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A lithium battery thermal runaway feature recognition algorithm based on BP neural network, characterized by: The steps include: The first step is data collection and preprocessing: use temperature and gas sensors to monitor the temperature and gas information of the lithium battery system in real time; use the temperature change of the temperature sensor before and after thermal runaway, and the temperature change slope at the time of thermal runaway as the main characteristic parameters; for the gas sensor, consider the change of gas concentration before and after thermal runaway, and the gas concentration change slope as the main characteristic parameters; use the collected temperature, gas data and characteristic parameters as a data set; The second step is feature extraction: PCA is used as the feature extraction algorithm. The PCA algorithm can reduce the dimension of feature data to two or three dimensions of the original data, and normalize the feature data to satisfy the standard normal distribution. The third step is model construction: use the main feature components to build a BP neural network model, use the error back propagation algorithm to optimize the model parameters, and improve the model prediction accuracy and stability; Step 4: Prediction and decision-making: The data obtained from real-time monitoring is input into the neural network model. If an abnormality is detected, an alarm will be issued; if there is no abnormality, the model will continue to monitor in real time to ensure the safety of the system. Step 5: Result interpretation: The results are interpreted through the lithium battery thermal runaway model and verified using the corresponding fault interpretation algorithm.

2. According to claim 1, a lithium battery thermal runaway feature recognition algorithm based on BP neural network is characterized in that: The first step of obtaining the temperature change before and after thermal runaway occurs uses a temperature sensor to collect data, and before the collected data is input into the model as a data set, it is necessary to extract features from the data. Since multiple temperature sensors collect data at the same time, the data needs to be filtered, and the data collected by the same ambient temperature sensor is selected as the input of the prediction model. After data preprocessing, normalization processing is also required to make the model converge faster.

3. The lithium battery thermal runaway feature recognition algorithm based on BP neural network according to claim 1, characterized in that: In the second step, the PCA algorithm is used as the feature extraction algorithm. The PCA algorithm is mainly based on the collinearity and covariance characteristics between the feature data. It can reduce the dimension of the feature data to two or three dimensions of the original data and normalize the feature data to satisfy the standard normal distribution.

4. The lithium battery thermal runaway feature recognition algorithm based on BP neural network according to claim 1, characterized in that: In the third step, the BP neural network model includes an input layer, a hidden layer and an output layer, wherein the input layer is used to input data features, the hidden layer acts as an intermediate neuron, and the output layer is the final result to be predicted. When data is input into the trained BP neural network model, the output layer outputs the final prediction result.

5. The lithium battery thermal runaway feature recognition algorithm based on BP neural network according to claim 1, characterized in that: The prediction model for thermal runaway in the fourth step is predicted by a trained prediction model to predict whether the lithium battery will experience thermal runaway under actual working conditions, and when it is predicted that the lithium battery is about to experience thermal runaway, the model issues a warning in a timely manner.

6. The lithium battery thermal runaway feature recognition algorithm based on BP neural network according to claim 1, characterized in that: The specific method for establishing the BP neural network model in the third step is as follows: first, determine the input variables and output variables of the neural network, select temperature and gas concentration data as input features to reduce model complexity and improve prediction accuracy, and select thermal runaway flag as output variable; determine the number of hidden layer nodes according to the number of input and output variables, and generally use the number of hidden layer nodes formula to calculate the number of hidden layer nodes, the formula is as follows: Where H represents the number of hidden layer nodes, I represents the number of input layer nodes, O represents the number of output layer nodes, and e represents a value between 10 -4 A constant between 10 and 10; the excitation function of the model uses the tanh or sigmoid function, where the tanh excitation function uses the S-shaped function, which is as follows: the sigmoid excitation function uses a piecewise linear function, as follows: Where e is the base of the natural logarithm, x is the function independent variable, and S(x) represents the function result; the activation function of the hidden layer is selected as follows: h1=tanh(S1), h0=sigmoid(S0), where S1 and S0 represent the output values ​​of the two hidden layer nodes respectively, and the activation function of the output layer is: o=a·s0+b·s1, where a and b represent the weights between the output layer and the hidden layer.

7. The lithium battery thermal runaway feature recognition algorithm based on BP neural network according to claim 6 is characterized in that: The back propagation algorithm is used in the BP neural network model to optimize the model parameters and improve the prediction accuracy and stability. In the BP neural network, the nonlinearity of the neuron activation function will cause the gradient descent search to fall into the local extreme value when the surface of the error function converges, thereby destroying the prediction accuracy. The adaptive moment estimation (Adam) optimization algorithm is used to optimize the model parameters. The algorithm can calculate the network learning speed and adjust the adaptive weights by analyzing the historical gradient changes. The specific method is as follows: First, the expected value and square difference of the gradient are calculated respectively: where t is the current training round, E is the expectation of the historical gradient value, D is the square difference of the historical gradient value, is the global learning rate, θt represents the network parameter value at time t, θt-1 represents the network parameter value at time t-1, represents the speed of global learning, ρ1 and ρ2 represent the exponential decay rate of the global learning rate; Then calculate the first-order moment estimate and second-order moment estimate of the global offset: where m represents the first-order moment estimate of the offset, n represents the second-order moment estimate of the offset, is the global learning rate, and represents the deviation of the global learning speed: where represents the first-order moment estimate deviation, represents the second-order moment estimate deviation, δ and ε are two small positive numbers used to prevent the denominator from being 0; calculate the deviation correction: where ms represents the deviation of the historical first-order moment estimate, and ns represents the deviation of the historical second-order moment estimate; finally calculate the iterative update of the parameters: where α represents the final update step size, m represents the first-order moment estimate, and n represents the second-order moment estimate.

8. The lithium battery thermal runaway feature recognition algorithm based on BP neural network according to claim 7, characterized in that: The calculation of weights in the BP algorithm is based on the loss function, and the loss function L is used to guide the optimization process. In the loss function, if the difference between the predicted value and the known value is small, the loss function is small; on the contrary, if the difference between the predicted value and the known value is large, the loss function will also be high.

9. The lithium battery thermal runaway feature recognition algorithm based on BP neural network according to claim 1, characterized in that: The fifth step of model result verification is as follows: first, the data within a period of time before and after the occurrence of thermal runaway is used as a training set, and the model is trained using the training set data; then, a test data set is used to test the accuracy and reliability of the model; based on the prediction effect and model accuracy, if the model needs to be improved, the training set data is used for training again; the performance comparison of the entire training and test data sets is used as the performance parameter for the final evaluation; the prediction results of the lithium battery thermal runaway prediction model are compared with the expert system to verify the accuracy of the prediction results, wherein the expert system is a decision-making model based on expert experience and knowledge and has a higher weight, and the test data set is used to verify whether the thermal runaway predicted by the model is consistent with the actual situation to ensure the accuracy of the model; in the thermal runaway scenario, historical data is used for training, and the test data may include data other than historical data; in conventional scenarios, the model is trained using training data, and then the data set used for testing includes data outside the range of the training data.

10. The lithium battery thermal runaway feature recognition algorithm based on BP neural network according to claim 1, characterized in that: In practical applications of the lithium battery thermal runaway prediction model of the fourth step, the thermal runaway of the lithium battery has a certain probability, and the model outputs the possibility of thermal runaway, that is, a thermal runaway with a prediction value greater than 0.5 is considered to have a high probability of thermal runaway, and a thermal runaway with a prediction value less than 0.5 is considered to have a low probability of thermal runaway. When the output probability value is between 0.5 and 0.7, the uncertainty of the model prediction result is strong, and thermal runaway may or may not occur. It is necessary to further monitor the model's prediction situation and environmental factors to make decisions; the lithium battery thermal runaway prediction model can predict multiple thermal runaway events. When multiple prediction values ​​are greater than 0.5, the risk of thermal runaway is high, and corresponding measures (early warning) should be taken in time to reduce the risk or replace the battery; If multiple thermal runaway prediction values ​​are less than 0.5, the risk of thermal runaway is low, but monitoring still needs to be continued; when multiple consecutive thermal runaway prediction values ​​are less than 0.5, the risk is further reduced and operation can continue, and the monitoring frequency needs to be reduced.

Images