Method and device for detecting internal short circuit of lithium ion battery and electronic equipment thereof

Abstract

The internal short circuit detection method comprises the following steps: acquiring training data and test data; training a preset deep learning model by using the training data to obtain an internal short circuit prediction model; the input information of the deep learning model comprises a relaxation time distribution function, characteristics multiplied by the relaxation time distribution function and characteristics multiplied by the electrochemical alternating current impedance spectroscopy; calculating internal short circuit probability distribution corresponding to input information of test data by using the internal short circuit prediction model; and determining an internal short circuit detection result of the lithium ion battery corresponding to the test data according to the internal short circuit probability distribution.

Description

[Technical field]

[0001] The present application relates to the technical field of lithium-ion batteries, and in particular to a method, a detection device and an electronic device for detecting an internal short circuit of a lithium-ion battery. [Background technology]

[0002] Lithium-ion batteries play an important role in electric vehicles, drones and energy storage due to their high energy density and long cycle life. However, safety issues such as thermal runaway, fire and explosion of lithium-ion batteries are still a concern. In incidents such as the Boeing 787 and Samsung mobile phone fires, internal short circuit (ISC) is considered to be the main cause of thermal runaway of lithium-ion batteries. Therefore, effective detection of internal short circuit faults in lithium-ion batteries before they develop into thermal runaway is of great significance for the safe and reliable use of lithium-ion batteries. [Summary of the invention]

[0003] The inventors of this application discovered when developing a product for detecting internal short circuits in lithium-ion batteries that the electrochemical reaction mechanism inside lithium-ion batteries is very complex. Therefore, the manifestation of early internal short circuit failures is mainly concentrated in a small decrease in voltage. Limited by the limitations of voltage acquisition accuracy and sampling frequency, current lithium-ion battery management systems are difficult to capture such small voltage changes caused by early internal short circuits.

[0004] To this end, the inventor of the present application provides a method for detecting an internal short circuit of an ion battery in the first aspect. The internal short circuit detection method comprises: obtaining training data; the training data comprises: a first AC impedance spectrum of a normal lithium ion battery at different states of charge and a second AC impedance spectrum of an abnormal lithium ion battery with an internal short circuit at different states of charge; calculating the relaxation time distribution function of the normal lithium ion battery and the abnormal lithium ion battery respectively through the first AC impedance spectrum and the second AC impedance spectrum; using the training data, training a preset deep learning model to obtain an internal short circuit prediction model; the input information of the deep learning model comprises: a relaxation time distribution function, features derived from the relaxation time distribution function, and features derived from the electrochemical AC impedance spectrum; obtaining test data; the test data comprises: a third AC impedance spectrum of a lithium ion battery at different states of charge; determining the input information of the test data through the third AC impedance spectrum; using the internal short circuit prediction model, calculating the internal short circuit probability distribution corresponding to the input information of the test data; and determining the internal short circuit detection result of the lithium ion battery corresponding to the test data according to the internal short circuit probability distribution.

[0005] The above-mentioned AC impedance spectrum is used as training data, and the relaxation time distribution function or the characteristics derived from it are used as input information to train the deep learning model to obtain the corresponding internal short circuit prediction model. Such an internal short circuit detection model does not need to rely on long-term internal short circuit process data when performing internal short circuit detection. It only needs to perform electrochemical AC impedance testing at a specific frequency for a few minutes to output the internal short circuit detection results of lithium-ion batteries in different health states and any charge state, thereby improving the diagnosis speed of lithium-ion battery internal short circuits.

[0006] In combination with the first aspect, in a possible implementation manner, the internal short circuit probability distribution includes: a first probability that the lithium ion battery is a normal lithium ion battery and a second probability that the lithium ion battery is an abnormal lithium ion battery; determining the internal short circuit detection result of the lithium ion battery corresponding to the test data according to the internal short circuit probability distribution specifically includes: when the first probability is greater than the second probability, determining that the lithium ion battery corresponding to the test data is a normal lithium ion battery; when the second probability is less than or equal to the second probability, determining that the lithium ion battery corresponding to the test data is an abnormal lithium ion battery.

[0007] In combination with the first aspect, or any of the above-mentioned possible implementations of the first aspect, in another possible implementation, the step of obtaining training data specifically includes: performing an AC impedance spectrum test before and after the test resistor is connected in parallel to both ends of a normal lithium-ion battery, respectively, to obtain the first AC impedance spectrum and the second AC impedance spectrum; wherein the test resistor has a preset resistance value, and the normal lithium-ion battery has a preset state of charge; after adjusting the lithium-ion battery in the preset state of charge to another state of charge, performing an AC impedance spectrum test again before and after the test resistor is connected in parallel to both ends of the normal lithium-ion battery, to obtain the first AC impedance spectrum and the second AC impedance spectrum of the lithium-ion battery in another state of charge; or preparing a battery with an internal short-circuit fault defect of the same type of battery, performing an AC impedance test on the normal lithium-ion battery and the battery with the internal short-circuit fault, to obtain the first AC impedance spectrum and the second AC impedance spectrum;

[0008] In combination with the first aspect, or any one of the above possible implementations of the first aspect, in another possible implementation, the step of obtaining training data further includes: after replacing the test resistor with a preset resistance value with another test resistor with a different resistance value, performing an AC impedance spectrum test again before the test resistor is connected in parallel to both ends of a normal lithium-ion battery and after the test resistor is connected in parallel to both ends of the normal lithium-ion battery to obtain a first AC impedance spectrum and a second AC impedance spectrum corresponding to the lithium-ion battery; or preparing a battery with an internal short-circuit fault defect of the same type, performing an AC impedance test on the normal lithium-ion battery and the battery with the internal short-circuit fault defect, and obtaining the first AC impedance spectrum and the second AC impedance spectrum.

[0009] In combination with the first aspect, or any one of the above possible implementations of the first aspect, in another possible implementation, the relaxation time distribution functions of the normal lithium-ion battery and the abnormal lithium-ion battery are calculated respectively by the first AC impedance spectrum and the second AC impedance spectrum, specifically including: calculating the corresponding first AC impedance value according to the AC excitation and AC response data in the first AC impedance spectrum; solving the corresponding first relaxation time distribution function by the Debye relaxation time formula based on the first AC impedance value; calculating the corresponding second AC impedance value according to the AC excitation and AC response data in the second AC impedance spectrum; solving the corresponding second relaxation time distribution function by the Debye relaxation time formula based on the second AC impedance value.

[0010] In combination with the first aspect, or any one of the above-mentioned possible implementation methods of the first aspect, in another possible implementation method, the use of the training data to train a preset deep learning model to obtain an internal short circuit prediction model specifically includes: determining the input information of the training data and inputting it into the deep learning model; using a preset first activation function to extract the information contained in the input information of the training data to obtain a corresponding calculation result; using a preset second activation function to convert the calculation result into a corresponding probability distribution; determining the current model fitting error based on the probability distribution; using a back propagation algorithm to update the model weights of the deep learning model; and obtaining the internal short circuit prediction model after a preset number of model weight updates.

[0011] In combination with the first aspect, or any one of the above possible implementations of the first aspect, in another possible implementation, the second activation function includes: any activation function such as a normalized exponential function; the first activation function includes: any activation function such as a hyperbolic tangent function.

[0012] In combination with the first aspect, or any one of the above-mentioned possible implementations of the first aspect, in another possible implementation, the features derived from the relaxation time distribution function include: the peak height, peak area and relaxation time corresponding to the highest point of the peak corresponding to the diffusion impedance in the relaxation time distribution function, etc., which may be features extracted from the relaxation time distribution function.

[0013] In a second aspect, the present application provides an internal short circuit detection device for a lithium-ion battery. The internal short circuit detection device comprises: a training data acquisition module, used for acquiring training data; the training data comprises: a first AC impedance spectrum of a normal lithium ion battery at different states of charge and a second AC impedance spectrum of an abnormal lithium ion battery with an internal short circuit at different states of charge; and the relaxation time distribution functions of the normal lithium ion battery and the abnormal lithium ion battery are calculated respectively through the first AC impedance spectrum and the second AC impedance spectrum; a model training module, used for training a preset deep learning model using the training data to obtain an internal short circuit prediction model; the input information of the deep learning model comprises: a relaxation time distribution function, a feature derived from the relaxation time distribution function, and a feature derived from the electrochemical AC impedance spectrum; a test data acquisition module, used for acquiring test data; the test data comprises: a third AC impedance spectrum of the lithium ion battery at different states of charge; and the input information of the test data is determined through the third AC impedance spectrum; an internal short circuit determination module, used for calculating the internal short circuit probability distribution corresponding to the input information of the test data using the internal short circuit prediction model; and the internal short circuit detection result of the lithium ion battery corresponding to the test data is determined according to the internal short circuit probability distribution.

[0014] In a third aspect, the present application provides an electronic device, which includes a processor and a memory, wherein a non-volatile storage medium is stored in the memory, and when the non-volatile storage medium is called by the processor, the processor executes the internal short circuit detection method as described above.

[0015] The beneficial effects of the related devices provided in the second and third aspects of this application can refer to the beneficial effects of the technical solution of the first aspect, and will not be repeated here.

Brief Description of the Drawings

[0016] One or more embodiments are exemplarily described by pictures in the corresponding drawings, and these exemplified descriptions do not constitute limitations on the embodiments. Elements with the same reference numerals in the drawings represent similar elements, and unless otherwise stated, the figures in the drawings do not constitute proportional limitations.

[0017] Figure 1 A method flow chart of an internal short circuit detection method provided in an embodiment of the present application;

[0018] Figure 2 A flow chart of a method for determining an internal short circuit detection result based on an internal short circuit probability distribution provided in an embodiment of the present application;

[0019] Figure 3 A flow chart of a method for obtaining training data provided in an embodiment of the present application;

[0020] Figure 4 A flow chart of a method for calculating a relaxation time distribution function provided in an embodiment of the present application;

[0021] Figure 5 A flow chart of a method for training a deep learning model provided in an embodiment of the present application;

[0022] Figure 6 This is a method flow chart of the internal short circuit detection method of Example 1 of the present application, showing the complete process from training data acquisition, deep learning model training and test data verification;

[0023] Fig. 7A The electrochemical AC impedance spectrum of the lithium-ion battery of Example 1 of the present application shows a situation where the state of charge is 100% and the resistance of the test resistor connected in parallel at both ends is 10Ω;

[0024] Figure 7B The electrochemical AC impedance spectrum of the lithium-ion battery of Example 1 of the present application shows a situation where the state of charge is 100% and the resistance of the test resistor connected in parallel at both ends is 20Ω;

[0025] Figure 7C The electrochemical AC impedance spectrum of the lithium-ion battery of Example 1 of the present application shows that the state of charge is 100% and the resistance of the test resistor connected in parallel at both ends is 30Ω;

[0026] Fig.7D The electrochemical AC impedance spectrum of the lithium-ion battery of Example 1 of the present application shows a situation where the state of charge is 100% and the resistance of the test resistor connected in parallel at both ends is 50Ω;

[0027] Fig. 7E The electrochemical AC impedance spectrum of the lithium-ion battery of Example 1 of the present application shows a situation where the state of charge is 100% and the resistance of the test resistor connected in parallel at both ends is 100Ω;

[0028] Figure 7F The electrochemical AC impedance spectrum of the lithium-ion battery of Example 1 of the present application shows a situation where the state of charge is 100% and the resistance of the test resistor connected in parallel at both ends is 200Ω;

[0029] Fig. 8A for Fig. 7A The relaxation time distribution diagram corresponding to the electrochemical impedance spectroscopy;

[0030] Figure 8B for Figure 7B The relaxation time distribution diagram corresponding to the electrochemical impedance spectroscopy;

[0031] Figure 8C for Figure 7C The relaxation time distribution diagram corresponding to the electrochemical impedance spectroscopy;

[0032] Fig.8D for Fig.7D The relaxation time distribution diagram corresponding to the electrochemical impedance spectroscopy;

[0033] Fig. 8E for Fig. 7E The relaxation time distribution diagram corresponding to the electrochemical impedance spectroscopy;

[0034] Fig.8F for Figure 7F The relaxation time distribution diagram corresponding to the electrochemical impedance spectroscopy;

[0035] Fig. 9 A comparison diagram of the internal short circuit probability distribution obtained by the internal short circuit prediction model of Example 1 of the present application and the internal short circuit true value corresponding to the test data set;

[0036] Fig.10 This is a method flow chart of the internal short circuit detection method of Example 2 of the present application, showing the complete process from training data acquisition, deep learning model training and test data verification;

[0037] Fig.11A A schematic diagram of the peak area corresponding to the diffusion impedance in the relaxation time distribution function of the lithium-ion battery of Example 2 of the present application;

[0038] Fig. 11B A schematic diagram of the peak height corresponding to the diffusion impedance in the relaxation time distribution function of the lithium-ion battery of Example 2 of the present application;

[0039] Fig. 11C A schematic diagram of the relaxation time corresponding to the highest point of the peak in the relaxation time distribution function of the lithium-ion battery of Example 2 of the present application;

[0040] Fig.12 A comparison diagram of the internal short circuit probability distribution obtained by the internal short circuit prediction model of Example 2 of the present application and the internal short circuit true value corresponding to the test data set;

[0041] Fig.13 A functional block diagram of an internal short circuit detection device provided in an embodiment of the present application;

[0042] Fig.14 A schematic diagram of a control terminal provided in an embodiment of the present application. [Specific implementation method]

[0043] In order to facilitate the understanding of the present application, the present application is described in more detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that when an element is described as "fixed to" another element, it can be directly on another element or there can be one or more centered elements therebetween. When an element is described as "connected" to another element, it can be directly connected to another element or there can be one or more centered elements therebetween. The orientation or positional relationship indicated by the terms "upper", "lower", "inner", "outer", "bottom" and the like used in this specification is based on the orientation or positional relationship shown in the accompanying drawings, only for the convenience of describing the present application and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present application. In addition, the terms "first", "second", "third" and the like are used only for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0044] Unless otherwise defined, all technical and scientific terms used in this specification have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used in this specification and in the specification of this application are only for the purpose of describing specific embodiments and are not intended to limit this application. The term "and / or" used in this specification includes any and all combinations of one or more of the related listed items.

[0045] In addition, the technical features involved in the different embodiments of the present application described below can be combined with each other as long as they do not conflict with each other.

[0046] Due to the limitations of actual voltage detection accuracy and sampling frequency, the battery management system cannot directly rely on the collected voltage changes to determine whether the lithium-ion battery has an internal short circuit. Usually, when identifying and detecting whether a lithium-ion battery has an internal short circuit, the battery management system uses the battery-related parameters (for example, voltage, current and temperature characteristics) collected by the battery management system as input, and calculates through a specific model and algorithm to output the corresponding internal short circuit prediction result.

[0047] In the process of implementing this application, the applicant found that the method of using battery-related parameters as input and outputting the corresponding internal short circuit prediction results after calculation through a specific model and algorithm requires process data of battery-related parameters for a long period of time. Moreover, the diagnostic accuracy and maximum diagnostic internal short circuit resistance of this detection method are difficult to meet the current needs of lithium-ion batteries in actual use.

[0048] The internal short circuit detection method for lithium-ion batteries provided in the embodiment of the present application combines the use of a deep learning model and a relaxation time distribution function, using the relaxation time distribution function calculated by the AC impedance spectrum, the features derived from the relaxation time distribution function, and the features derived from the electrochemical AC impedance spectrum as input information of the deep learning model to train the deep learning model, thereby constructing the required internal short circuit prediction model.

[0049] Therefore, by using the internal short circuit prediction model, the corresponding internal short circuit detection result can be determined by calculating the AC impedance spectrum of the lithium ion battery. Since the relaxation time distribution function contains the internal electrochemical reaction information of the lithium ion battery, the early diagnosis accuracy of the internal short circuit of the lithium ion battery and the maximum diagnosis internal short circuit resistance are improved.

[0050] Figure 1 A method flow chart of the internal short circuit detection method of a lithium ion battery provided in an embodiment of the present application. The internal short circuit detection method can be roughly divided into a training phase and a test use phase. The steps involved in the training phase are no longer repeated after the model training is completed and the required internal short circuit prediction model is obtained. When performing actual lithium ion internal short circuit detection, it is only necessary to use the internal short circuit prediction model that has been obtained.

[0051] like Figure 1 As shown, the training phase of the internal short circuit detection method includes:

[0052] S11. Obtain training data.

[0053] The training data is data information required for deep learning model training, which specifically includes multiple different AC impedance spectra, and whether the lithium-ion battery corresponding to these AC impedance spectra has an internal short circuit and the charge state are known.

[0054] "AC impedance spectroscopy" is an electrochemical characterization method that applies a small AC signal to a lithium-ion battery, measures its impedance response, and obtains impedance characteristic curves at different frequencies. It reflects the characteristics of multiple electrochemical processes in lithium-ion batteries, such as electrolyte resistance, charge transfer impedance, and solid-state diffusion processes.

[0055] For ease of description and distinction, a lithium-ion battery without internal short circuit is referred to as a "normal lithium-ion battery" and a lithium-ion battery with internal short circuit is referred to as an "abnormal lithium-ion battery". Accordingly, the AC impedance spectrum of a normal lithium-ion battery is referred to as a "first AC impedance spectrum" and the AC impedance spectrum of an abnormal lithium-ion battery is referred to as a "second AC impedance spectrum".

[0056] In other words, the acquired training data includes: a plurality of first AC impedance spectra of normal lithium-ion batteries at different states of charge and a plurality of second AC impedance spectra of abnormal lithium-ion batteries with internal short circuit at different states of charge.

[0057] S12, respectively calculating the relaxation time distribution functions of the normal lithium-ion battery and the abnormal lithium-ion battery through the first AC impedance spectrum and the second AC impedance spectrum.

[0058] Among them, the "relaxation time distribution function" is a function used to describe the time scale distribution of various relaxation processes. It is obtained by performing corresponding mathematical operations on the AC impedance spectroscopy data, and the superimposed electrochemical processes are separated according to their characteristic time scales.

[0059] S13. Use the training data to train the preset deep learning model to obtain an internal short circuit prediction model.

[0060] Among them, "training the deep learning model" refers to the process of inputting part of the training data information into the deep learning model as input information, and then updating the model parameters through iterative optimization, so that the deep learning model gradually learns the correspondence between the input information and the probability of internal short circuit, and obtains model parameters with predictive ability.

[0061] In some embodiments, the input information used by the preset deep learning model is a relaxation time distribution function. In other embodiments, one or more features derived from the relaxation time distribution function are used as input information of the preset deep learning model to effectively characterize the kinetic characteristics of the internal diffusion impedance of the battery. In some other embodiments, features derived from the electrochemical AC impedance spectrum are used as input information of the preset deep learning model. Alternatively, two or all of the above-mentioned relaxation time distribution function, features derived from the relaxation time distribution function, and features derived from the electrochemical AC impedance spectrum are used as input information of the preset deep learning model.

[0062] Specifically, the characteristics of the relaxation time distribution function include but are not limited to: peak height, peak area and relaxation time corresponding to the highest peak point corresponding to the diffusion impedance in the relaxation time distribution function, and other characteristics that can be extracted from the relaxation time distribution function.

[0063] Among them, "peak height" represents the maximum value of the distribution function at a specific relaxation time, which reflects the dominance of the relaxation process. "Peak area" is a feature obtained by integrating the relaxation time distribution function, which reflects the total amount of charge involved in the relaxation process. "Relaxation time corresponding to the highest point of the peak" represents the characteristic time scale of the most important relaxation process.

[0064] The internal short circuit prediction model obtained through training in step S13 is stored in any suitable type of memory for use in subsequent internal short circuit detection of lithium-ion batteries.

[0065] Please continue reading Figure 1 Based on the obtained internal short circuit prediction model, the test use phase of the internal short circuit detection method includes:

[0066] S21. Obtain test data.

[0067] Among them, the test data is an AC impedance spectrum that is different from the training data. For the convenience of description and distinction, the AC impedance spectrum of the test data is referred to as the "third AC impedance spectrum" below. In the test scenario, it is known whether the lithium-ion battery of the third AC impedance spectrum has an internal short circuit, which is used to verify the performance of the internal short circuit prediction model. In the usage scenario, it is unknown whether the lithium-ion battery of the third AC impedance spectrum has an internal short circuit.

[0068] S22. Determine input information of test data through the third AC impedance spectrum.

[0069] The input information is specifically determined by step S13 and is consistent with the input information used in training the internal short circuit prediction model to ensure that the input information meets the use requirements of the internal short circuit prediction model.

[0070] S23. Using the internal short circuit prediction model, calculate the internal short circuit probability distribution corresponding to the input information of the test data.

[0071] Among them, "internal short circuit probability distribution" is the prediction output of the internal short circuit prediction model, which is used to quantify the possibility of internal short circuit in lithium-ion batteries under the current input information. The internal short circuit probability distribution is presented in many different forms, such as a two-dimensional probability vector, a normalized risk score, or a multi-level probability distribution.

[0072] S24. Determine the internal short circuit detection result of the lithium ion battery corresponding to the test data according to the internal short circuit probability distribution.

[0073] Among them, based on the specific presentation form of the internal short circuit probability distribution, a corresponding method is used to determine the specific internal short circuit detection result (ie, whether the lithium-ion battery has an internal short circuit).

[0074] The internal short circuit detection performed by the above-mentioned internal short circuit detection model does not need to rely on long-term internal short circuit process data. It only needs to perform an electrochemical AC impedance test for a few minutes at a specific frequency to obtain the corresponding test results as input information to the internal short circuit detection model, which can output the internal short circuit detection results of lithium-ion batteries in any health state and any charge state, thereby improving the diagnosis speed of internal short circuits in lithium-ion batteries.

[0075] In some embodiments, the internal short circuit probability distribution is presented as a two-dimensional probability vector. For ease of description below, the probability that the lithium-ion battery has no internal short circuit is called the "first probability", and the probability that the lithium-ion battery has an internal short circuit is called the "second probability".

[0076] Accordingly, if Figure 2 As shown, the step of determining the internal short circuit detection result of the lithium ion battery corresponding to the test data according to the internal short circuit probability distribution specifically includes:

[0077] S241. Determine whether the first probability is greater than the second probability; if so, execute step S242; if not, execute step S243.

[0078] S242: Determine that the lithium-ion battery corresponding to the test data is a normal lithium-ion battery.

[0079] Among them, "normal lithium-ion battery" means that there is a high degree of confidence that the current lithium-ion battery has not suffered an internal short circuit and is in a safe state.

[0080] S243. Determine that the lithium-ion battery corresponding to the test data is an abnormal lithium-ion battery.

[0081] Among them, "normal lithium-ion battery" means that there is a high degree of certainty that the current lithium-ion battery has experienced an internal short circuit and needs to be dealt with in a timely manner.

[0082] Figure 3 is a flow chart of a method for obtaining training data provided in an embodiment of the present application. Figure 3 As shown, the method of obtaining training data includes:

[0083] S111 . Performing an AC impedance spectrum test on a normal lithium-ion battery in a preset state of charge to obtain a first AC impedance spectrum.

[0084] S112, connecting a test resistor with a preset resistance value in parallel to both ends of the normal lithium-ion battery, and then performing an AC impedance spectrum test to obtain a second AC impedance spectrum.

[0085] S113, after adjusting the lithium-ion battery in the preset state of charge to a different state of charge, repeat the above steps S111 and S112 to obtain a first AC impedance spectrum and a second AC impedance spectrum of the lithium-ion battery in another state of charge.

[0086] In some other embodiments, N test resistors with different resistance values ​​are further selected to simulate and obtain training data with different internal short circuit degrees.

[0087] Specifically, after completing the first AC impedance spectrum and the second AC impedance spectrum of the lithium-ion battery with one test resistor under multiple charge states, another test resistor with a different resistance value is connected in parallel at both ends of the normal lithium-ion battery, and the above-mentioned AC impedance spectrum test is performed before and after the test resistors are connected in parallel to obtain N first AC impedance spectra and second AC impedance spectra corresponding to the test resistors with different resistance values.

[0088] In other embodiments, in addition to obtaining training data by connecting a test resistor in parallel with a normal lithium-ion battery, training data is obtained by directly preparing an internal short-circuit fault defective battery. First, an internal short-circuit fault defective battery of the same type is prepared. Then, an AC impedance test is performed on the normal lithium-ion battery and the internal short-circuit fault defective battery, respectively, to obtain a first AC impedance spectrum and a second AC impedance spectrum as training data.

[0089] Figure 4 is a flow chart of a method for solving the relaxation time distribution function provided in an embodiment of the present application. In some embodiments, Figure 4 As shown, the steps of calculating the relaxation time distribution function of a lithium-ion battery include:

[0090] S121. Calculate the corresponding AC impedance value according to the AC excitation and AC response data in the AC impedance spectrum.

[0091] Specifically, the AC impedance value is calculated by the following formula (1):

[0092] Z(ω)=X(ω) / J(ω)(1)

[0093] Where ω is the angular frequency, J(ω) is the AC excitation, X(ω) is the AC response, and Z(ω) is the AC impedance of the lithium-ion battery.

[0094] S122. Based on the AC impedance value, the corresponding relaxation time distribution function is solved by using the Debye relaxation time formula.

[0095] Specifically, the Debye relaxation time formula is expressed by the following equations (2) and (3):

[0096]

[0097] Among them, R ohm is the ohmic impedance of the lithium-ion battery, R pol is the total polarization impedance of the lithium-ion battery, τ is the relaxation time, j is the complex imaginary part, and γ(τ) is the relaxation time distribution function of the lithium-ion battery.

[0098] It can be applied to different first AC impedance spectra, second AC impedance spectra and third AC impedance spectra Figure 4The method steps shown are used to calculate the corresponding relaxation time distribution function, and then obtain the input information required by the model. For simplicity of presentation, the specific calculation process of the first AC impedance spectrum, the second AC impedance spectrum and the third AC impedance spectrum will not be repeated below.

[0099] Figure 5 A flowchart of a method for training a deep learning model provided in an embodiment of the present application. In some embodiments, Figure 5 As shown, the method for training a deep learning model includes the following steps:

[0100] S131. Determine the input information of the training data and input it into the deep learning model.

[0101] Among them, the "deep learning model" is a neural network structure composed of an input layer that receives and preliminarily processes input information, a hidden layer between the input layer and the output layer, and an output layer responsible for generating the final prediction results. The hidden layer consists of multiple neuron nodes that transmit and transform information through weighted connections.

[0102] S132: Use a preset first activation function to extract information contained in the input information of the training data to obtain a corresponding calculation result.

[0103] Among them, the "first activation function" is a nonlinear mathematical function. It is used to introduce nonlinear transformations into deep learning models and provide feature extraction capabilities for the model, including any activation functions such as tangent functions.

[0104] Specifically, step S132 is expressed by the following equations (4) and (5):

[0105] γ(τ) hidden =tanh(γ(τ))(4)

[0106]

[0107] Among them, tanh is the hyperbolic tangent function, which is the first activation function used in the input layer and hidden layer, γ(τ) hidden is the result calculated by the input layer and the hidden layer, and is the input of the output layer.

[0108] S133. Use a preset second activation function to convert the calculation result into a corresponding probability distribution.

[0109] The "second activation function" is a function used to convert the calculation result into an output form that meets specific requirements, including any activation function such as a normalized exponential function.

[0110] Specifically, step S133 is expressed by the following equations (6) and (7):

[0111]

[0112] γ(τ) output =softmax(γ(τ) hidden ) (7)

[0113] Among them, softmax is a normalized exponential function and is the second activation function used in the above step S133. γ(τ) output is the two-dimensional probability distribution calculated by the softmax function.

[0114] S134. Determine the current model fitting error according to the probability distribution.

[0115] Among them, "model fitting error" refers to the difference between the current model prediction value and the true value. It is a parameter used to measure the prediction accuracy of the model. Specifically, the model fitting error is expressed by means of mean square error or cross entropy.

[0116] Specifically, step S134 is expressed by the following formula (8):

[0117]

[0118] Among them, Loss is the cross entropy loss function, N is the number of samples of training data, and p i Predict results and ISC for training data i real Same probability, ISC real , is an indicator of whether the lithium-ion battery in the training data is an abnormal lithium-ion battery. For example, ISC real =1, it means that the lithium-ion battery in the training data is an abnormal lithium-ion battery, ISC real =0, it indicates that the lithium-ion battery in the training data is a normal lithium-ion battery.

[0119] S135. Use the back propagation algorithm to update the model weights of the deep learning model.

[0120] Among them, the "back propagation algorithm" is a gradient descent optimization method that adjusts the weight parameters layer by layer from the output layer to the input layer by calculating the gradient of the error to the weight of each layer. The "model weight" is the parameter connecting the neurons in each layer, which determines the importance and conversion method of each input feature.

[0121] S136. After a preset number of model weight updates, an internal short circuit prediction model is obtained.

[0122] The “number of model weight updates” refers to the number of iterations of model training, which is set by technical personnel based on actual needs.

[0123] The following two specific examples are provided to fully illustrate and describe in detail the internal short circuit detection methods of lithium-ion batteries in the above-mentioned multiple embodiments. These specific examples show the complete detection process from obtaining training data, training a deep learning model to testing the trained deep learning model.

[0124] Embodiment 1:

[0125] 1) Acquisition of AC impedance spectroscopy data:

[0126] The selected lithium-ion battery is: 2.75Ah 18650 type, graphite-based anode / LiNi 0.8 Co 0.1 Mn 0.1 O2 cathode battery. The different states of charge that need to be obtained include: 10%, 20%, 30%, 40%, 50%, 60%, 70%, 80%, 90% and 100% states of charge. The test resistors of different resistance values ​​that need to be used include: 200Ω, 100Ω, 50Ω, 30Ω, 20Ω, 10Ω test resistors.

[0127] like Figure 6 As shown, perform the following steps to obtain AC impedance spectroscopy data:

[0128] S101. Perform an AC impedance spectrum test on the lithium-ion battery at 10% charge state and record the ISC at this time. real is 0.

[0129] S102, connecting the test resistors in parallel to both ends of the lithium-ion battery in order from large resistance to small resistance, and performing an AC impedance spectrum test.

[0130] In which, between each two parallel test resistor AC impedance tests, an AC impedance test without parallel test resistors is performed. When the AC impedance test is performed without parallel test resistors, the ISC at this time is recorded. real When the resistor is connected in parallel for AC impedance test, record the ISC at this time. real is 1.

[0131] S103, charging the lithium-ion battery from 10% SOC to 20%, 30%, 40%, 50%, 60%, 70%, 80%, 90% and 100% SOC in sequence with a current of 2.75A, and repeating steps S101 and S102 at each different SOC.

[0132] Thus, we finally obtained 10 different states of charge, 6 different internal short-circuit resistances, a total of 120 AC impedance spectra and their corresponding ISCs. realThe AC impedance spectrum data set is divided into a training data set and a test data set in a ratio of 7:3, which are used for training and testing of the deep learning model respectively (i.e., the training data set contains 84 AC impedance spectrum test data, and the test data set contains 36 AC impedance spectrum test data).

[0133] For example, Fig. 7A The electrochemical impedance spectroscopy of a lithium-ion battery is shown in FIG. 1 when the state of charge is 100% and the test resistance connected in parallel at both ends is 10Ω. Figure 7B The electrochemical impedance spectroscopy of a lithium-ion battery is shown in FIG. 1 when the state of charge is 100% and the test resistance connected in parallel at both ends is 20Ω. Figure 7C The electrochemical impedance spectroscopy of a lithium-ion battery is shown in FIG. 1 when the state of charge is 100% and the test resistance connected in parallel at both ends is 30Ω. Fig.7D The electrochemical impedance spectroscopy of a lithium-ion battery is shown in FIG. 1 when the state of charge is 100% and the test resistance connected in parallel at both ends is 50Ω. Fig. 7E The electrochemical impedance spectroscopy of a lithium-ion battery is shown in FIG. 1 when the state of charge is 100% and the test resistance connected in parallel at both ends is 100Ω.

[0134] Figure 7F The electrochemical impedance spectroscopy of a lithium-ion battery is shown in FIG. 1 when the state of charge is 100% and the test resistance connected in parallel at both ends is 200Ω.

[0135] 2) Calculation of relaxation time distribution function:

[0136] Please continue reading Figure 6 , for each AC impedance spectrum data, perform the following steps to calculate the corresponding relaxation time distribution function:

[0137] S201, obtaining AC excitation and AC response data in AC impedance spectrum data, and calculating the AC impedance value by formula (1).

[0138] S202, numerically analyzing the AC impedance value calculated in step S201, solving equations (2) and (3), thereby obtaining a corresponding lithium-ion battery relaxation time distribution function.

[0139] For example, Fig. 8A It is shown in Fig. 7A The corresponding relaxation time distribution function diagram obtained by calculating the electrochemical impedance spectroscopy shown. Figure 8B It is shown in Figure 7B The corresponding relaxation time distribution function diagram obtained by calculating the electrochemical impedance spectroscopy shown. Figure 8C It is shown in Figure 7C The corresponding relaxation time distribution function diagram obtained by calculating the electrochemical impedance spectroscopy shown. Fig.8D It is shown in Fig.7D The corresponding relaxation time distribution function diagram obtained by calculating the electrochemical impedance spectroscopy shown. Fig. 8E It is shown in Fig. 7E The corresponding relaxation time distribution function diagram obtained by calculating the electrochemical impedance spectroscopy shown. Fig.8F It is shown in Figure 7F The corresponding relaxation time distribution function diagram obtained by calculating the electrochemical impedance spectroscopy shown.

[0140] 3) Training of deep learning models:

[0141] Please continue reading Figure 6 , based on the relaxation time distribution function corresponding to the AC impedance spectroscopy data in the training data set, perform the following steps to complete the training of the deep learning model:

[0142] S301a, inputting the relaxation time distribution function into the input layer and the hidden layer in the deep learning model, and using the tanh activation function to extract the internal short circuit information contained in the relaxation time distribution function.

[0143] S302, input the results calculated by the input layer and the hidden layer into the output layer, and generate a two-dimensional probability distribution through the softmax function, which respectively represent the probability of no internal short circuit occurring and the probability of internal short circuit occurring.

[0144] S303. Calculate the model fitting error according to the cross entropy loss function, and apply the back propagation algorithm to update the model weights.

[0145] S304, repeat the above steps S301 to S303, update the weight parameters of the model, and complete the training after the number of iterations reaches 200. The deep learning model after training is referred to as the "internal short circuit prediction model" below.

[0146] 4) Test of internal short circuit prediction model:

[0147] Please continue reading Figure 6 , using the AC impedance spectroscopy data and its corresponding relaxation time distribution function in the remaining unused test data set, perform the following steps to test the model:

[0148] S401a, inputting the relaxation time distribution function in the test data set into the internal short circuit prediction model to obtain the corresponding internal short circuit prediction two-dimensional probability distribution (ie, a first probability of no internal short circuit and a second probability of internal short circuit).

[0149] S402: Use the larger one of the first probability and the second probability as the prediction result of the lithium-ion battery.

[0150] For example, Fig. 9 The actual values ​​of 36 AC impedance spectroscopy test data in the test data set (i.e. ISC real The comparison between the internal short circuit prediction probability value calculated and output by the internal short circuit prediction model of Example 1 is shown.

[0151] like Fig. 9 As shown in the figure, in the 36 test data, the prediction results of the internal short circuit prediction model are basically consistent with the true value, indicating that the model has good prediction performance. It can effectively distinguish between the two different states of internal short circuit and no internal short circuit, showing a strong ability to identify the internal short circuit state, and the reliability of the prediction results is high.

[0152] Embodiment 2:

[0153] 1) Acquisition of AC impedance spectroscopy data:

[0154] The selected lithium-ion battery is: 2.75Ah 18650 type, graphite-based anode / LiNi 0.8 Co 0.1 Mn 0.1 O2 cathode battery. The different states of charge that need to be obtained include: 10%, 20%, 30%, 40%, 50%, 60%, 70%, 80%, 90% and 100% states of charge. The test resistors of different resistance values ​​that need to be used include: 200Ω, 100Ω, 50Ω, 30Ω, 20Ω, 10Ω test resistors.

[0155] like Fig.10 As shown, perform the following steps to obtain AC impedance spectroscopy data:

[0156] S101. Perform an AC impedance spectrum test on the lithium-ion battery at 10% charge state and record the ISC at this time. real is 0.

[0157] S102, connecting the test resistors in parallel to both ends of the lithium-ion battery in descending order of resistance value, and performing an AC impedance spectrum test.

[0158] In between every two parallel resistor AC impedance tests, an AC impedance test without parallel resistors is performed. When the AC impedance test is performed without parallel resistors, the ISC at this time is recorded. real When the resistor is connected in parallel for AC impedance test, record the ISC at this time. real is 1.

[0159] S103, charging the lithium-ion battery from 10% SOC to 20%, 30%, 40%, 50%, 60%, 70%, 80%, 90% and 100% SOC in sequence with a current of 2.75A, and repeating steps S101 and S102 at each different SOC.

[0160] Thus, we finally obtained 10 different charge states, 6 test resistors with different resistance values, and a total of 120 AC impedance spectra and their corresponding ISCs. real The AC impedance spectrum data set is divided into a training data set and a test data set in a ratio of 7:3, which are used for training and testing of the deep learning model respectively (i.e., the training data set contains 84 AC impedance spectrum test data, and the test data set contains 36 AC impedance spectrum test data).

[0161] 2) Calculation of relaxation time distribution function:

[0162] Please continue reading Fig.10 , for each AC impedance spectrum data, perform the following steps to calculate the corresponding relaxation time distribution function:

[0163] S201, obtaining AC excitation and AC response data in AC impedance spectrum data, and calculating the AC impedance value by formula (1).

[0164] S202, numerically analyzing the AC impedance value calculated in step S201, solving equations (2) and (3), thereby obtaining a corresponding lithium-ion battery relaxation time distribution function.

[0165] For example, Fig.11A It shows the peak area corresponding to the diffusion impedance in the corresponding relaxation time distribution function calculated from the electrochemical impedance spectrum when the lithium-ion battery is at a 100% state of charge and the resistance values ​​of the test resistors connected in parallel at both ends are 200Ω, 100Ω, 50Ω, 30Ω, 20Ω and 10Ω respectively. Fig. 11B The figure shows the peak height corresponding to the diffusion impedance in the corresponding relaxation time distribution function calculated from the electrochemical impedance spectrum when the lithium-ion battery is at a 100% state of charge and the resistance values ​​of the test resistors connected in parallel at both ends are 200Ω, 100Ω, 50Ω, 30Ω, 20Ω and 10Ω respectively. Fig. 11C It shows the relaxation time corresponding to the highest peak of the diffusion impedance in the corresponding relaxation time distribution function calculated from the electrochemical impedance spectrum when the state of charge of the lithium-ion battery is 100% and the resistance values ​​of the test resistors connected in parallel at both ends are 200Ω, 100Ω, 50Ω, 30Ω, 20Ω and 10Ω respectively.

[0166] 3) Training of deep learning models:

[0167] Please continue reading Fig.10 , based on the peak area, peak height and relaxation time corresponding to the highest point of the diffusion impedance in the relaxation time distribution function corresponding to the AC impedance spectrum data in the training data set, perform the following steps to complete the training of the deep learning model:

[0168] S301b. The peak area, peak height and relaxation time corresponding to the highest point of the peak corresponding to the diffusion impedance in the relaxation time distribution function are used as input features and input into the input layer and hidden layer of the deep learning model, and the internal short circuit information contained in the relaxation time distribution function is extracted using the tanh activation function.

[0169] S302, input the results calculated by the input layer and the hidden layer into the output layer, and generate a two-dimensional probability distribution through the softmax function, which respectively represent the probability of no internal short circuit occurring and the probability of internal short circuit occurring.

[0170] S303. Calculate the model fitting error according to the cross entropy loss function, and apply the back propagation algorithm to update the model weights.

[0171] S304, repeat the above steps S301 to S303, update the weight parameters of the model, and complete the training after the number of iterations reaches 200. The deep learning model after training is referred to as the "internal short circuit prediction model" below.

[0172] 4) Test of internal short circuit prediction model:

[0173] Please continue reading Fig.10 , using the AC impedance spectrum data in the remaining unused test data set and the peak area, peak height and relaxation time corresponding to the peak highest point of the diffusion impedance in the corresponding relaxation time distribution function, perform the following steps to test the model:

[0174] S401 b. The peak area, peak height and relaxation time corresponding to the highest point of the diffusion impedance in the relaxation time distribution function in the test data set are used as input features and input into the internal short circuit prediction model to obtain the corresponding two-dimensional probability distribution of internal short circuit prediction (i.e., the first probability of no internal short circuit occurring and the second probability of an internal short circuit occurring).

[0175] S402: Use the larger one of the first probability and the second probability as the prediction result of the lithium-ion battery.

[0176] For example, Fig.12 The actual values ​​of 36 AC impedance spectroscopy test data in the test data set (i.e. ISC realThe comparison between the internal short circuit prediction probability value calculated and output by the internal short circuit prediction model of Example 2 is shown.

[0177] like Fig.12 As shown in the figure, in the 36 test data, the prediction results of the internal short circuit prediction model are basically consistent with the true value, indicating that the model has good prediction performance. It can effectively distinguish between the two different states of internal short circuit and no internal short circuit, showing a strong ability to identify the internal short circuit state, and the reliability of the prediction results is high.

[0178] Based on the internal short circuit detection method of the lithium ion battery provided in the above embodiment, the embodiment of the present application further provides an internal short circuit detection device of the lithium ion battery. Fig.13 This is a functional block diagram of the internal short circuit detection device provided in an embodiment of the present application.

[0179] In some embodiments, Fig.13 As shown, the internal short circuit detection device 10 includes: a training data acquisition module 11, a model training module 12, a test data acquisition module 13 and an internal short circuit determination module 14.

[0180] Among them, the training data acquisition module 11 is used to acquire training data and calculate the relaxation time distribution function of normal lithium-ion batteries and abnormal lithium-ion batteries respectively through the first AC impedance spectrum and the second AC impedance spectrum. The training data includes: the first AC impedance spectrum of normal lithium-ion batteries at different states of charge and the second AC impedance spectrum of abnormal lithium-ion batteries with internal short circuit at different states of charge. The model training module 12 is used to use the training data to train the preset deep learning model to obtain an internal short circuit prediction model. The input information of the deep learning model includes: the relaxation time distribution function, the features derived from the relaxation time distribution function, and the features derived from the electrochemical AC impedance spectrum. The test data acquisition module 13 is used to acquire test data and determine the input information of the test data through the third AC impedance spectrum. The test data includes: the third AC impedance spectrum of the lithium-ion battery at different states of charge. The internal short circuit determination module 14 is used to use the internal short circuit prediction model to calculate the internal short circuit probability distribution corresponding to the input information of the test data; and according to the internal short circuit probability distribution, determine the internal short circuit detection result of the lithium-ion battery corresponding to the test data.

[0181] In some embodiments, the training data acquisition module 11 is specifically used to: perform an AC impedance spectrum test on a normal lithium-ion battery at a preset state of charge to obtain the first AC impedance spectrum; connect a resistor with a preset resistance value in parallel across the two ends of the normal lithium-ion battery, and then perform the AC impedance spectrum test to obtain the second AC impedance spectrum; after adjusting the lithium-ion battery at the preset state of charge to a different state of charge, repeat the above-mentioned AC impedance spectrum test steps, or perform an AC impedance test on a normal lithium-ion battery and a pre-prepared internal short-circuit fault defective battery to obtain a first AC impedance spectrum and a second AC impedance spectrum.

[0182] Specifically, when acquiring the second AC impedance spectrum data, the training data acquisition module 11 is also used to: sequentially connect N test resistors with different resistance values ​​in parallel at both ends of a normal lithium-ion battery, perform AC impedance spectrum testing, and acquire N second AC impedance spectra corresponding to each test resistor.

[0183] In some embodiments, when calculating the AC impedance value, the training data acquisition module 11 is specifically used to: calculate the corresponding first AC impedance value according to the AC excitation and AC response data in the first AC impedance spectrum; based on the first AC impedance value, solve the corresponding first relaxation time distribution function through the Debye relaxation time formula; calculate the corresponding second AC impedance value according to the AC excitation and AC response data in the second AC impedance spectrum; based on the second AC impedance value, solve the corresponding second relaxation time distribution function through the Debye relaxation time formula.

[0184] In some embodiments, when performing deep learning model training, the model training module 12 is specifically used to: determine the input information of the training data and input it into the deep learning model; use a preset first activation function to extract the information contained in the input information of the training data to obtain the corresponding calculation result; use a preset second activation function to convert the calculation result into a corresponding probability distribution; determine the current model fitting error based on the probability distribution; use a back propagation algorithm to update the model weights of the deep learning model; and obtain an internal short circuit prediction model after a preset number of model weight updates.

[0185] Specifically, the second activation function includes any activation function such as a normalized exponential function. The first activation function includes any activation function such as a hyperbolic tangent function. The above-mentioned features derived from the relaxation time distribution function include: the peak height corresponding to the diffusion impedance in the relaxation time distribution function, the peak area and the relaxation time corresponding to the highest peak point, etc., which can be extracted from the relaxation time distribution function.

[0186] In some embodiments, the internal short circuit probability distribution includes: a first probability that the lithium ion battery is a normal lithium ion battery and a second probability that the lithium ion battery is an abnormal lithium ion battery. The model training module 12 is specifically used to: when the first probability is greater than the second probability, determine that the lithium ion battery corresponding to the test data is a normal lithium ion battery; when the second probability is less than or equal to the second probability, determine that the lithium ion battery corresponding to the test data is an abnormal lithium ion battery.

[0187] It should be noted that the internal short circuit detection device provided by the embodiment of the present application can be performed by an electronic computing platform of any suitable type. The electronic computing platform can realize or perform the above-mentioned one or more functions with electronic hardware, computer software program or a combination of the two. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to realize the described functions for each specific application, but this realization should not be considered to exceed the scope of the present application.

[0188] Alternatively, when the internal short circuit prediction model has been stored, the internal short circuit detection device only includes a test data acquisition module 13 and an internal short circuit determination module 14, and there is no need to additionally set up a training data acquisition module 11 and a model training module 12.

[0189] Fig.14 A schematic diagram of a control terminal provided in an embodiment of the present application. The control terminal can be used to perform one or more steps in the internal short circuit detection method described in one or more embodiments above. The control terminal is implemented by various forms of digital computers, such as workstations, servers, blade servers, mainframe computers and other appropriate electronic computing platforms.

[0190] It should be noted that Fig.14 The components, their connections and relationships, and their functions shown are for illustration only and are not intended to limit the specific implementation of the control terminal.

[0191] like Fig.14 As shown, the control terminal 20 includes a processor 21, a memory 22, a storage device 23, a high-speed interface 25 connected to the memory 22 and multiple high-speed expansion ports 24, and a low-speed interface 27 connected to the low-speed expansion port 26 and the storage device 23.

[0192] Each of the processor 21, memory 22, storage device 23, high-speed interface 25, high-speed expansion port 24, and low-speed interface 27 are interconnected using various buses and may be mounted on a common motherboard or in other suitable ways.

[0193] The processor 21 processes computer program instructions stored in the memory 22 or on the storage device 23 for executing computer program instructions for displaying graphic information on an external input / output device (eg, a display device 28 coupled to the high-speed interface 25).

[0194] In some embodiments, multiple processors and / or multiple buses are also used together with multiple memories and multiple types of memories. Also, multiple electronic devices are connected, each of which provides part of the necessary operations (e.g., as a server group, blade server group, or multi-processor system).

[0195] The memory 22 stores information within the control terminal and is one or more volatile memory units, non-volatile memory units, or another form of computer-readable media, such as a magnetic disk or optical disk.

[0196] The storage device 23 provides mass storage for the control terminal. It contains computer readable media, such as a floppy disk device, a hard disk device, an optical disk device or a tape device, a flash memory or other similar solid-state memory device, or a device array, including devices in a storage area network or other configurations.

[0197] The computer program instructions are stored in the information carrier. When the computer program instructions are executed by one or more processing devices (such as the processor 21), the internal short circuit detection method of the lithium ion battery described in one or more embodiments above is implemented.

[0198] The high-speed interface 25 manages bandwidth-intensive operations for controlling the terminal, while the low-speed interface 27 manages less bandwidth-intensive operations. In some embodiments, the high-speed interface 25 is coupled to the memory 22, the display 28 (e.g., via a graphics processor or accelerator), and the high-speed expansion port 24 that can accept various expansion cards. The low-speed interface 27 is coupled to the storage device 23 and the low-speed expansion port 26.

[0199] The low-speed expansion port 26 includes a communication port (eg, USB, Ethernet, wireless Ethernet) and is coupled to one or more input / output devices, such as a keyboard, pointing device, scanner, or networking device such as a switch (eg, through a network adapter).

[0200] For example, Fig.14 As shown, in order to provide interaction with the user, the control terminal 20 also has a display device 28 (e.g., a cathode ray tube or a liquid crystal display monitor) for displaying information to the user and a pointing device 29 (e.g., a mouse) for the user to provide input to the computer. Other types of interactive devices are also used to provide interaction with the user;

[0201] Of course, the feedback provided to the user is any form of sensory feedback (eg, visual feedback, auditory feedback, or tactile feedback); and input from the user is received in any form, including acoustic, voice, or tactile input.

[0202] The cargo handling method described in one or more embodiments of the present application can be implemented in digital electronic circuits, integrated circuits, specially designed ASICs (application-specific integrated circuits), computer hardware, firmware, software and / or a combination thereof. It includes one or more embodiments in a computer program that is executed and / or interpreted on a programmable system that includes at least one programmable processor, which is dedicated or general purpose, coupled to receive data and instructions from a storage system, at least one input device, and at least one output device, and to send data and instructions to the storage system, at least one input device, and at least one output device.

[0203] These computer programs (also referred to as programs, software, software applications or codes) include machine instructions for programmable processors and are implemented in high-level procedural and / or object-oriented programming languages ​​and / or in assembly / machine languages. In the present embodiment, the above-mentioned machine-readable medium and computer-readable medium refer to any computer program product, device and / or device (e.g., disk, optical disk, memory, programmable logic device (PLD)) for providing machine instructions and / or data to a programmable processor, including machine-readable media that receive machine instructions as machine-readable signals. The above-mentioned machine-readable signals refer to any signals for providing machine instructions and / or data to a programmable processor.

[0204] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Under the concept of the present application, the technical features in the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other changes in different aspects of the present application as described above, which are not provided in detail for the sake of simplicity. Although the present application has been described in detail with reference to the aforementioned embodiments, a person of ordinary skill in the art should understand that the technical solutions described in the aforementioned embodiments can still be modified, or some of the technical features can be replaced by equivalents. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for detecting an internal short circuit of a lithium-ion battery, characterized in that: include: Get training data; The training data includes: a first AC impedance spectrum of a normal lithium-ion battery at different states of charge and a second AC impedance spectrum of an abnormal lithium-ion battery with an internal short circuit at different states of charge; Calculating relaxation time distribution functions of a normal lithium-ion battery and an abnormal lithium-ion battery respectively through the first AC impedance spectrum and the second AC impedance spectrum; Using the training data, a preset deep learning model is trained to obtain an internal short circuit prediction model; the input information of the deep learning model includes: a relaxation time distribution function, a feature derived from the relaxation time distribution function, and a feature derived from an electrochemical impedance spectrum; Acquire test data; the test data includes: a third AC impedance spectrum of the lithium-ion battery at different states of charge; Determining input information of the test data through the third AC impedance spectrum; Using the internal short circuit prediction model, calculating the internal short circuit probability distribution corresponding to the input information of the test data; According to the internal short circuit probability distribution, an internal short circuit detection result of the lithium ion battery corresponding to the test data is determined.

2. The internal short circuit detection method according to claim 1, characterized in that: The internal short circuit probability distribution includes: a first probability that the lithium ion battery is a normal lithium ion battery and a second probability that the lithium ion battery is an abnormal lithium ion battery; Determining the internal short circuit detection result of the lithium ion battery corresponding to the test data according to the internal short circuit probability distribution specifically includes: When the first probability is greater than the second probability, determining that the lithium ion battery corresponding to the test data is a normal lithium ion battery; When the second probability is less than or equal to the second probability, it is determined that the lithium ion battery corresponding to the test data is an abnormal lithium ion battery.

3. The internal short circuit detection method according to claim 1, characterized in that: The step of obtaining training data specifically includes: Performing an AC impedance spectrum test before and after the test resistor is connected in parallel to both ends of the normal lithium-ion battery to obtain the first AC impedance spectrum and the second AC impedance spectrum; Wherein, the test resistor has a preset resistance value, and the normal lithium-ion battery has a preset state of charge; After adjusting the lithium-ion battery in the preset state of charge to another state of charge, performing an AC impedance spectrum test again before the test resistor is connected in parallel to both ends of the normal lithium-ion battery and after the test resistor is connected in parallel to both ends of the normal lithium-ion battery to obtain a first AC impedance spectrum and a second AC impedance spectrum of the lithium-ion battery in another state of charge; Or prepare internal short-circuit fault defective batteries of the same type of batteries, perform AC impedance tests on normal lithium-ion batteries and the internal short-circuit fault defective batteries, and obtain the first AC impedance spectrum and the second AC impedance spectrum.

4. The internal short circuit detection method according to claim 3, characterized in that: The step of obtaining training data also includes: After replacing the test resistor with the preset resistance with another test resistor with a different resistance, performing an AC impedance spectrum test again before the test resistor is connected in parallel to both ends of the normal lithium-ion battery and after the test resistor is connected in parallel to both ends of the normal lithium-ion battery to obtain a first AC impedance spectrum and a second AC impedance spectrum corresponding to the lithium-ion battery; Or prepare internal short-circuit fault defective batteries of the same type of batteries, perform AC impedance tests on normal lithium-ion batteries and the internal short-circuit fault defective batteries, and obtain the first AC impedance spectrum and the second AC impedance spectrum.

5. The internal short circuit detection method according to claim 1, characterized in that: The step of calculating the relaxation time distribution functions of the normal lithium-ion battery and the abnormal lithium-ion battery respectively through the first AC impedance spectrum and the second AC impedance spectrum specifically includes: Calculating a corresponding first AC impedance value according to the AC excitation and AC response data in the first AC impedance spectrum; Based on the first AC impedance value, solving the corresponding first relaxation time distribution function by using the Debye relaxation time formula; Calculating a corresponding second AC impedance value according to the AC excitation and AC response data in the second AC impedance spectrum; Based on the second AC impedance value, the corresponding second relaxation time distribution function is solved by using the Debye relaxation time formula.

6. The internal short circuit detection method according to claim 1, characterized in that: The method of using the training data to train a preset deep learning model to obtain an internal short circuit prediction model specifically includes: Determine input information of the training data and input it into the deep learning model; Using a preset first activation function, extracting information contained in the input information of the training data to obtain a corresponding calculation result; Using a preset second activation function, converting the calculation result into a corresponding probability distribution; Determining a current model fitting error according to the probability distribution; Using a back propagation algorithm, updating the model weights of the deep learning model; After a preset number of model weight updates, the internal short circuit prediction model is obtained.

7. The internal short circuit detection method according to claim 6, characterized in that: The second activation function includes: a normalized exponential function; the first activation function includes: a hyperbolic tangent function.

8. The internal short circuit detection method according to claim 1, characterized in that: The characteristics derived from the relaxation time distribution function include: the peak height and peak area corresponding to the diffusion impedance in the relaxation time distribution function, and the relaxation time corresponding to the highest point of the peak.

9. A lithium-ion battery internal short circuit detection device, characterized in that: include: A training data acquisition module, used to acquire training data; The training data includes: a first AC impedance spectrum of a normal lithium-ion battery at different states of charge and a second AC impedance spectrum of an abnormal lithium-ion battery with an internal short circuit at different states of charge; and through the first AC impedance spectrum and the second AC impedance spectrum, respectively calculating the relaxation time distribution function of the normal lithium-ion battery and the abnormal lithium-ion battery; A model training module, used to train a preset deep learning model using the training data to obtain an internal short circuit prediction model; the input information of the deep learning model includes: a relaxation time distribution function, a feature derived from the relaxation time distribution function, and a feature derived from an electrochemical impedance spectrum; A test data acquisition module, used to acquire test data; the test data includes: a third AC impedance spectrum of the lithium-ion battery at different states of charge; and determining input information of the test data through the third AC impedance spectrum; The internal short circuit determination module is used to calculate the internal short circuit probability distribution corresponding to the input information of the test data using the internal short circuit prediction model; and determine the internal short circuit detection result of the lithium ion battery corresponding to the test data according to the internal short circuit probability distribution.

10. An electronic device, characterized in that: include: A processor and a memory, wherein a non-volatile storage medium is stored in the memory, and when the non-volatile storage medium is called by the processor, the processor executes the internal short circuit detection method according to any one of claims 1 to 8.

Images