Power supply thermal runaway prediction model generation method and power supply thermal runaway prediction method

Abstract

The invention provides a power supply thermal runaway prediction model generation method and a power supply thermal runaway prediction method, and the method comprises the steps: collecting normal and thermal runaway sample data of a lithium battery in a special working condition of a coal mine, generating a thermal runaway data set of a lithium battery cell, and obtaining a training set and a sample probability value of power supply thermal runaway; constructing a deep learning model comprising a multi-dimensional attention module and a gating circulation unit module, and determining model parameters by adopting a firefly algorithm; using the training set to train the deep learning model; obtaining a loss value according to the sample probability value and a model prediction probability value output by the deep learning model, and optimizing parameters of a gating cycle unit module by adopting an Adam algorithm; and integrating the trained deep learning model into the battery management system for thermal runaway prediction, so that thermal runaway can be predicted more accurately, safely and reliably, and practical requirements on safety and reliability in special scenes such as a coal mine are met.

Description

Technical Field

[0001] The present invention relates to the field of new energy, and in particular to a power supply thermal runaway prediction model generation method and a power supply thermal runaway prediction method. Background Art

[0002] With the rapid development of new energy electric vehicles and energy storage power supplies, the Battery Management System (BMS), as the core brain of lithium battery power products, is responsible for managing and maintaining the normal operation of the battery throughout its life cycle and evaluating the current status of each battery cell in real time. In harsh environments such as coal mines, power products based on lithium batteries have also begun to be gradually popularized and applied, but due to the particularity of the scenarios, there are extremely high requirements for the safety of lithium battery power products. If a lithium battery has thermal runaway, it is very easy to cause a major production accident. Therefore, the early prediction of thermal runaway of lithium batteries is crucial, which is directly related to the safety and stability of production, especially in special production scenarios such as underground coal mines.

[0003] In the existing technology, most mining power supply BMS mainly detects the temperature and voltage of power supply products, but cannot predict and alarm the thermal runaway of lithium batteries in advance. It is difficult to respond in time and prevent accidents in the event of thermal runaway, which can easily lead to safety accidents and affect the safety of coal mine production. Although some BMS provide early warning algorithms based on temperature data, the electrochemical reaction of lithium batteries is an extremely complex and changeable nonlinear change, which is difficult to accurately predict based on temperature data alone. In addition, these algorithms usually react and warn only after the temperature exceeds the set threshold, which cannot meet the needs of early prevention and control of thermal runaway risks in current mining scenarios.

[0004] How to provide a thermal runaway prediction algorithm for mine power supplies that can predict thermal runaway with a more accurate, safer and more reliable algorithm to meet the actual needs of safety and reliability in special scenarios such as coal mines. Summary of the invention

[0005] The technical problem to be solved by the present invention is to provide a method for generating a power supply thermal runaway prediction model and a method for predicting power supply thermal runaway, which can predict thermal runaway with a more accurate, safer and more reliable algorithm, and meet the actual needs for safety and reliability in special scenarios such as coal mines.

[0006] In order to solve the above problems, the present invention provides a method for generating a power supply thermal runaway prediction model, including: collecting normal sample data and thermal runaway sample data of lithium battery use process in special working conditions of coal mines, generating a thermal runaway data set of lithium battery cells, generating a training set based on the thermal runaway data set and sample probability values ​​of power supply thermal runaway occurring in samples in the training set; constructing a deep learning model, the deep learning model including a multi-dimensional attention module and a gated recurrent unit module, the multi-dimensional attention module including a spatial attention module and a channel attention module, and the output of the deep learning model is a model prediction probability value; The firefly algorithm is used to determine the parameters of the deep learning model, including: the position of each firefly represents a set of parameters of the deep learning model, in each iteration, the position of the firefly is updated according to the position update formula, the pre-trained model is trained to obtain the pre-trained predicted probability value, the firefly brightness function value of the firefly at the new position is calculated using the pre-trained predicted probability value, the firefly position with the smallest firefly brightness value in each iteration is retained as the current optimal solution, the iteration is performed cyclically until the number of iterations or the accuracy requirements are met, and the parameters of the deep learning model are determined by the current optimal solution at the end of the iteration. The position of the firefly is: ,in, Represents the learning rate of the deep learning model, with a value range of [0.0001, 0.01], is the number of iterations for training the deep learning model, ranging from [1,200], is the number of convolutional layers in the spatial attention module, ranging from [1,20], is the number of convolutional layers of the channel attention module, ranging from [1,20], is the number of gated recurrent units in the gated recurrent unit module, ranging from [1,10]. The firefly brightness function is: ,in, The number of samples contained in the training set used to train the pre-trained model. For the The pre-trained prediction probability value of samples, For the The sample probability value of samples, the relative attraction between fireflies is: ,in, is the initial attraction of fireflies, is the light absorption coefficient, For the Firefly and The distance between fireflies, The calculation formula is: , For the The location of the fireflies. For the The location of the fireflies. For the Firefly in The position of the dimensional space, For the Firefly in The position of the dimensional space, is the vector dimension of the firefly's position, and the relative brightness between fireflies is: , the position update formula is: ,in and They are and Fireflies in The position at the moment; using the training set to train the deep learning model; according to the sample probability value and the model prediction probability value, a loss value is calculated by a loss function; with the goal of minimizing the loss value, the Adam algorithm is used to optimize the parameters of the gated cyclic unit module during the training of the deep learning model; after the training is completed, the deep learning model is output as the power supply thermal runaway prediction model.

[0007] In some embodiments, the label of the thermal runaway data set includes at least one or more of total voltage, total current, battery state of charge, battery health status, voltage change rate, current change rate, temperature change rate, maximum voltage, minimum voltage, maximum temperature, minimum temperature, and charge and discharge rate.

[0008] In some embodiments, the multi-dimensional attention module further includes a baseline model that extracts thermal runaway deep features from the thermal runaway dataset.

[0009] In some embodiments, the deep learning model also includes an integration module, which includes a fully connected layer and a mapping function.

[0010] In some embodiments, the loss function is: ,in, is the sample probability value, is the model predicted probability value, is the actual sample weight value.

[0011] In some embodiments, the sample probability value includes a normal sample probability value and a thermal runaway sample probability value, the true sample weight value for the normal sample probability value is 0.2, and the true sample weight value for the thermal runaway sample probability value is 3.

[0012] In order to solve the above problems, the present invention also provides a power supply thermal runaway prediction method, comprising: integrating a power supply thermal runaway prediction model generated according to the power supply thermal runaway prediction model generation method into a battery management system; the battery management system inputs the collected power supply data into the power supply thermal runaway prediction model to obtain a power supply thermal runaway prediction probability value.

[0013] In some embodiments, when the power supply thermal runaway prediction probability value is greater than 0.6, a power supply thermal runaway warning is triggered, and when the power supply thermal runaway prediction probability value is greater than 0.8, a power supply thermal runaway alarm is triggered.

[0014] The above technical scheme generates a thermal runaway data set for lithium battery cells by collecting normal sample data and thermal runaway sample data during the use of lithium batteries in special working conditions in coal mines, generates a training set and sample probability values ​​of power supply thermal runaway for samples in the training set based on the thermal runaway data set; constructs a deep learning model including a multi-dimensional attention module and a gated cyclic unit module, and the output of the deep learning model is the model prediction probability value; uses the firefly algorithm to determine the parameters of the deep learning model; uses the training set to train the deep learning model; calculates the loss value through the loss function according to the sample probability value and the model prediction probability value, and uses the Adam algorithm to optimize the parameters of the gated cyclic unit module; integrates the trained deep learning model into the battery management system for thermal runaway prediction, which can predict thermal runaway more accurately, safely and reliably, and meet the actual needs of safety and reliability in these special scenarios of coal mines.

[0015] It should be understood that the above general description and the detailed description below are exemplary and explanatory only and cannot limit the present invention. The techniques, methods and devices known to ordinary technicians in the relevant field may not be discussed in detail, but where appropriate, the techniques, methods and devices should be considered as part of the authorization specification. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments of the present invention are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0017] Figure 1 is a flow chart of a method for generating a power supply thermal runaway prediction model provided by an embodiment of the present invention; Figure 2 is a schematic diagram of a multi-dimensional attention module provided by an embodiment of the present invention; Figure 3is a schematic diagram of a method for generating a power supply thermal runaway prediction model provided by an embodiment of the present invention; Figure 4 It is a flow chart of a source thermal runaway prediction method provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0018] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings. 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.

[0019] Figure 1 FIG. 1 is a flow chart of a method for generating a power supply thermal runaway prediction model provided by an embodiment of the present invention. Figure 1 As shown, the power supply thermal runaway prediction model generation method includes: step S11, collecting normal sample data and thermal runaway sample data of the lithium battery use process in the special working conditions of the coal mine, generating a thermal runaway data set of the lithium battery cell, generating a training set based on the thermal runaway data set, and calculating the sample probability value of the power supply thermal runaway of the samples in the training set; step S12, constructing a deep learning model, the deep learning model includes a multi-dimensional attention module and a gated cyclic unit module, and the output of the deep learning model is the model prediction probability value; step S13, using the firefly algorithm to determine the parameters of the deep learning model; step S14, using the training set to train the deep learning model; step S15, according to the sample probability value and the model prediction probability value, using the loss function to calculate the loss value; step S16, with the goal of minimizing the loss value, using the Adam algorithm to optimize and update the parameters of the gated cyclic unit module during the training of the deep learning model; step S17, after the training is completed, output the deep learning model as the power supply thermal runaway prediction model.

[0020] Referring to step S11, normal sample data and thermal runaway sample data of the lithium battery use process in the special working conditions of coal mines are collected to generate a thermal runaway data set for the lithium battery cell, and a training set and sample probability values ​​of power supply thermal runaway occurring in the samples in the training set are generated based on the thermal runaway data set. The thermal runaway data set for lithium battery cells is crucial to generating the power supply thermal runaway prediction model, but currently there is no good thermal runaway data set in the lithium battery field, and it does not meet the special working conditions of coal mines. Therefore, in this step, normal sample data and thermal runaway sample data of the lithium battery use process in the special working conditions of coal mines are collected, and finally a thermal runaway data set for lithium battery cells is constructed. Furthermore, in the step of generating the thermal runaway data set for lithium battery cells, some abnormal data are eliminated and the multidimensional data is normalized.

[0021] The sample probability values ​​include a normal sample probability value and a thermal runaway sample probability value.

[0022] The label of the thermal runaway data set includes at least one or more of the total voltage, total current, battery state of charge, battery health status, voltage change rate, current change rate, temperature change rate, maximum voltage, minimum voltage, maximum temperature, minimum temperature, and charge and discharge rate. In this embodiment, the label of the thermal runaway data set includes all of the above. Among them, the total current includes charging current and discharging current.

[0023] Referring to step S12, a deep learning model is constructed, wherein the deep learning model includes a multi-dimensional attention module and a gated recurrent unit module, and the firefly algorithm is used to determine the parameters of the deep learning model. The output of the deep learning model is a model prediction probability value.

[0024] Figure 2 is a schematic diagram of a multi-dimensional attention module provided by an embodiment of the present invention. represents the weighted sum, The multidimensional attention module includes a spatial attention module and a channel attention module, which are used to capture short-term patterns and local dependencies in the thermal runaway data of lithium batteries.

[0025] like Figure 2 As shown, the multi-dimensional attention module also includes a benchmark model, and the benchmark model extracts the thermal runaway deep features from the thermal runaway data set. The benchmark model is a MobileNet model or an Inception-ResNet model.

[0026] like Figure 2 As shown, the benchmark model is used to extract thermal runaway deep features from the thermal runaway data set, the thermal runaway deep features are respectively input into the channel attention module and the spatial attention module to obtain the channel attention weight and the spatial attention weight, the channel attention weight and the spatial attention weight are weighted and summed to obtain the multidimensional attention weight, and the thermal runaway deep features are multiplied by the multidimensional attention weight to obtain the thermal runaway key features.

[0027] The gated recurrent unit (GRU) includes a reset gate and an update gate. The reset gate is used to combine the new input information with the previous memory, thereby determining which information in the hidden state of the previous moment should be ignored; the update gate is used to control the extent to which the state information of the previous moment is brought into the current state, thereby determining how much information in the hidden state of the previous moment is retained and passed to the current moment. The gated recurrent unit module adopts the algorithm of the gated recurrent unit, which can discover the time variation law hidden in the key features of thermal runaway.

[0028] The specific calculation formula in the gated recurrent unit module is as follows: , , , , in, is the input feature sequence, is the new hidden state, is the old hidden state, is a candidate hidden state, To reset the gate output, Output for hidden door. , , are the weight matrices, That is the Sigmoid function, which is defined as: , The function is the hyperbolic tangent function, which is defined as: .

[0029] The deep learning model also includes an integration module, which includes a fully connected layer and a mapping function. The fully connected layer (FC) acts as a classifier in the entire convolutional neural network. The mapping function is a sigmoid function. The output of the integration module is the model prediction probability value.

[0030] Referring to step S13, the firefly algorithm is used to determine the parameters of the deep learning model. The firefly algorithm (FA) is an optimization algorithm based on swarm intelligence, which simulates the flashing behavior of fireflies to find the optimal solution to the problem. In this step, the firefly module is used to determine the parameters of the deep learning model, and the parameters include the learning rate of the deep learning model, the number of iterations for training the deep learning model, the number of convolutional layers of the spatial attention module, the number of convolutional layers of the channel attention module, and the number of gated recurrent units of the gated recurrent unit module.

[0031] In the step of using the firefly algorithm to determine the parameters of the deep learning model, the position of each firefly represents a set of parameters of the deep learning model; in each iteration, the position of the firefly is updated according to the position update formula, the pre-trained model is trained to obtain the pre-trained predicted probability value, the firefly brightness function value of the firefly at the new position is calculated using the pre-trained predicted probability value, and the firefly position with the smallest firefly brightness value in each iteration is retained as the current optimal solution; the iteration is executed in a loop until the number of iterations or the accuracy requirements are met, and the parameters of the deep learning model are determined with the current optimal solution at the end of the iteration.

[0032] Specifically, this step includes: (1) initializing firefly swarm parameters; (2) determining the firefly brightness function; (3) updating the firefly position; (4) adjusting the step size coefficient; (5) evaluating the firefly brightness function; (6) recording the optimal solution; (7) iterative optimization; and (8) outputting the optimal solution.

[0033] In step (1), a set of firefly positions are randomly initialized in the high-dimensional solution space. The position of each firefly represents a set of parameters of the deep learning model. The position of the firefly is: ,in, Represents the learning rate of the deep learning model, with a value range of [0.0001, 0.01], is the number of iterations for training the deep learning model, ranging from [1,200], is the number of convolutional layers in the spatial attention module, ranging from [1,20], is the number of convolutional layers of the channel attention module, ranging from [1,20], is the number of gated recurrent units in the gated recurrent unit module, and its value range is [1,10].

[0034] In step (2), the firefly brightness function is: ,in, The number of samples contained in the training set used to train the pre-trained model. For the The pre-trained prediction probability value of samples, For the The sample probability value of the samples.

[0035] In this embodiment, the pre-trained model is a residual neural network (Residual Neural Network, referred to as Resnet) model.

[0036] The brightness of fireflies can represent the predictive performance of the model. The higher the brightness, the higher the predictive performance index. The firefly population is balanced towards the initial parameters with high brightness and good predictive performance.

[0037] The relative attraction between fireflies is: ,in, is the initial attraction of fireflies, is the light absorption coefficient, For the Firefly and The distance between fireflies, The calculation formula is: , For the The location of the fireflies. For the The location of the fireflies. For the Firefly in The position of the dimensional space, For the Firefly in The position of the dimensional space, is the vector dimension of the firefly's position, and the relative brightness between fireflies is: .

[0038] In step (3), the position update formula is: ,in, and They are and Fireflies in The location at the moment, is the relative attraction between the fireflies, i.e., the step size coefficient.

[0039] In step (4), the step size coefficient gradually decreases as the number of iterations increases, ensuring that the algorithm can quickly explore the solution space in the early stage and gradually refine the search in the later stage.

[0040] In step (5), the firefly brightness function value of each firefly at the new position is calculated.

[0041] In step (6), find the firefly position with the smallest firefly brightness value in each iteration, and use this firefly position as the current optimal solution.

[0042] In step (7), steps (3) to (6) are repeated until the number of iterations or accuracy requirements are met.

[0043] In step (8), the current optimal solution retained at the end of the iteration is the global optimal solution, and the parameters of the deep learning model are determined according to the parameter values ​​in the global optimal solution.

[0044] For example, the firefly algorithm is used to determine that the number of convolutional layers of the spatial attention module is 2, the number of convolutional layers of the channel attention module is 4, and the number of gated recurrent units of the gated recurrent unit module is 2.

[0045] Referring to step S14, the deep learning model is trained using the training set. In step S13, the parameters of the deep learning model have been determined. The training set is generated based on the thermal runaway data set. During the training process, the deep learning model outputs the model prediction probability value, and the deep learning model can be optimized by processing the model prediction probability value.

[0046] Referring to step S15, the loss value is calculated by the loss function according to the sample probability value and the model prediction probability value. Considering that the sample data category distribution is uneven and the data difference between normal samples and thermal runaway samples is large, this step adopts the weighted binary cross entropy loss function to improve the training effect by assigning different weights to the two categories of samples.

[0047] The loss function is specifically: ,in, is the sample probability value, is the model predicted probability value, is the actual sample weight value.

[0048] The sample probability value includes a normal sample probability value and a thermal runaway sample probability value. The true sample weight value for the normal sample probability value is 0.2, and the true sample weight value for the thermal runaway sample probability value is 3.

[0049] Referring to step S16, with the goal of minimizing the loss value, the Adam algorithm is used to optimize and update the parameters of the gated recurrent unit module during the training of the deep learning model. The Adam algorithm is Adaptive Moment Estimation, which is a deep learning optimization algorithm. The parameters of the gated recurrent unit module are the weight matrix , , .

[0050] When the deep learning model is trained, each time the model is trained, a predicted probability value of the model is obtained, and the Adam algorithm iteratively optimizes the loss value output by the loss function to obtain a new weight matrix , , .

[0051] Figure 3 It is a schematic diagram of a method for generating a power supply thermal runaway prediction model provided by an embodiment of the present invention.

[0052] like Figure 3 As shown, the deep learning model is Figure 3 As shown in the dashed box, it includes: a multi-dimensional attention module, a gated recurrent unit module and an integration module, wherein the gated recurrent unit module includes a weight matrix , , When the deep learning model is trained, each training step further includes: the multi-dimensional attention module extracts the thermal runaway key features from the thermal runaway data set, and the thermal runaway key features output the model prediction probability value after passing through the gated cycle unit module and the integration module.

[0053] According to the sample probability value obtained from the thermal runaway data set and the model predicted probability value, the loss value is obtained through the loss function, and the Adam algorithm is iteratively optimized with the goal of minimizing the loss value to obtain a new weight matrix , , The updated weight matrix is ​​used for the next training of the deep learning model, so that the weight matrix gradually reaches the optimum.

[0054] In this embodiment, the maximum number of iterations of the Adam algorithm is set to 50, the initial learning rate is 0.1, the size of each batch is 128, and the optimal weight matrix of the gated recurrent unit module is found after 50 iterations.

[0055] Referring to step S17, after the training is completed, the deep learning model is output as the power supply thermal runaway prediction model. After multiple training iterations and Adam algorithm optimization, the loss value obtained by the deep learning model has met the preset value, and the deep learning model training is completed.

[0056] Based on the same inventive concept, an embodiment of the present invention further provides a method for predicting thermal runaway of a power supply.

[0057] Figure 4 FIG. 1 is a flow chart of a source thermal runaway prediction method provided by an embodiment of the present invention. Figure 4As shown, the power supply thermal runaway prediction method includes: step S41, integrating the power supply thermal runaway prediction model generated according to the power supply thermal runaway prediction model generation method into the battery management system; step S42, the battery management system inputs the collected power supply data into the power supply thermal runaway prediction model to obtain a power supply thermal runaway prediction probability value. The power supply thermal runaway prediction probability value has a value range of [0,1].

[0058] Furthermore, the battery management system determines whether it is necessary to trigger a power supply thermal runaway alarm based on the power supply thermal runaway prediction probability value. In this embodiment, when the power supply thermal runaway prediction probability value is greater than 0.6, a power supply thermal runaway warning is triggered to remind the technician to analyze the performance indicators of the power supply; when the power supply thermal runaway prediction probability value is greater than 0.8, a power supply thermal runaway alarm is triggered to notify the technician to adjust the power supply parameters according to the current state of the power supply to eliminate the alarm.

[0059] The above technical scheme generates a thermal runaway data set for lithium battery cells by collecting normal sample data and thermal runaway sample data during the use of lithium batteries in special working conditions in coal mines, generates a training set and sample probability values ​​of power supply thermal runaway for samples in the training set based on the thermal runaway data set; constructs a deep learning model including a multi-dimensional attention module and a gated cyclic unit module, and the output of the deep learning model is the model prediction probability value; uses the firefly algorithm to determine the parameters of the deep learning model; uses the training set to train the deep learning model; calculates the loss value through the loss function according to the sample probability value and the model prediction probability value, and uses the Adam algorithm to optimize the parameters of the gated cyclic unit module; integrates the trained deep learning model into the battery management system for thermal runaway prediction, which can predict thermal runaway more accurately, safely and reliably, and meet the actual needs of safety and reliability in these special scenarios of coal mines.

[0060] It should be noted that, in this document, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion. The various embodiments in this specification are described in a related manner, and the same and similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

[0061] The above is only a preferred embodiment of the present invention and is not intended to limit the protection scope of the present invention. It should be noted that a person skilled in the art can make several improvements and modifications without departing from the principle of the present invention, and these improvements and modifications should also be regarded as the protection scope of the present invention.

Claims

1. A method for generating a power supply thermal runaway prediction model, characterized in that: include: Collect normal sample data and thermal runaway sample data of lithium battery use process in special working conditions of coal mines, generate thermal runaway data set of lithium battery cells, generate training set and sample probability values ​​of power supply thermal runaway of samples in the training set based on the thermal runaway data set; construct a deep learning model, the deep learning model includes a multi-dimensional attention module and a gated recurrent unit module, the multi-dimensional attention module includes a spatial attention module and a channel attention module, and the output of the deep learning model is the model prediction probability value; The firefly algorithm is used to determine the parameters of the deep learning model, including: the position of each firefly represents a set of parameters of the deep learning model, in each iteration, the position of the firefly is updated according to the position update formula, the pre-trained model is trained to obtain the pre-trained predicted probability value, the firefly brightness function value of the firefly at the new position is calculated using the pre-trained predicted probability value, the firefly position with the smallest firefly brightness value in each iteration is retained as the current optimal solution, the iteration is performed cyclically until the number of iterations or the accuracy requirements are met, and the parameters of the deep learning model are determined by the current optimal solution at the end of the iteration. The position of the firefly is: ,in, Represents the learning rate of the deep learning model, with a value range of [0.0001, 0.01], is the number of iterations for training the deep learning model, ranging from [1,200], is the number of convolutional layers in the spatial attention module, ranging from [1,20], is the number of convolutional layers of the channel attention module, ranging from [1,20], is the number of gated recurrent units in the gated recurrent unit module, ranging from [1,10]. The firefly brightness function is: ,in, The number of samples contained in the training set used to train the pre-trained model. For the The pre-trained prediction probability value of samples, For the The sample probability value of samples, the relative attraction between fireflies is: ,in, is the initial attraction of fireflies, is the light absorption coefficient, For the Firefly and The distance between fireflies, The calculation formula is: , For the The location of the fireflies. For the The location of the fireflies. For the Firefly in The position of the dimensional space, For the Firefly in The position of the dimensional space, is the vector dimension of the firefly's position, and the relative brightness between fireflies is: , the position update formula is: ,in and They are and Fireflies in The position at the moment; using the training set to train the deep learning model; according to the sample probability value and the model prediction probability value, a loss value is calculated by a loss function; with the goal of minimizing the loss value, the Adam algorithm is used to optimize the parameters of the gated cyclic unit module during the training of the deep learning model; after the training is completed, the deep learning model is output as the power supply thermal runaway prediction model.

2. The method for generating a power supply thermal runaway prediction model according to claim 1, characterized in that: The labels of the thermal runaway data set include at least one or more of total voltage, total current, battery state of charge, battery health status, voltage change rate, current change rate, temperature change rate, maximum voltage, minimum voltage, maximum temperature, minimum temperature, and charge and discharge rate.

3. The method for generating a power supply thermal runaway prediction model according to claim 1, characterized in that: The multi-dimensional attention module also includes a benchmark model, which extracts thermal runaway deep features from the thermal runaway data set. The deep learning model also includes an integration module, which includes a fully connected layer and a mapping function.

4. The method for generating a power supply thermal runaway prediction model according to claim 1, characterized in that: The loss function is: ,in, is the sample probability value, is the model predicted probability value, is the actual sample weight value.

5. The method for generating a power supply thermal runaway prediction model according to claim 4, characterized in that: The sample probability value includes a normal sample probability value and a thermal runaway sample probability value. The true sample weight value for the normal sample probability value is 0.2, and the true sample weight value for the thermal runaway sample probability value is 3.

6. A method for predicting thermal runaway of a power supply, characterized in that: include: The power supply thermal runaway prediction model formed by the power supply thermal runaway prediction model generation method according to any one of claims 1 to 5 is integrated into a battery management system; the battery management system inputs the collected power supply data into the power supply thermal runaway prediction model to obtain a power supply thermal runaway prediction probability value.

7. The method for predicting thermal runaway of a power supply according to claim 6, characterized in that: When the power supply thermal runaway prediction probability value is greater than 0.6, a power supply thermal runaway warning is triggered, and when the power supply thermal runaway prediction probability value is greater than 0.8, a power supply thermal runaway alarm is triggered.

Images