Battery life prediction method and related equipment

Abstract

The present application provides a battery life prediction method and related equipment. The method includes: testing a battery under multiple preset test conditions to obtain multiple initial feature groups and battery capacity measurement intervals corresponding to each initial feature group under each test condition; inputting each initial feature group and corresponding battery capacity measurement interval into a preset point cloud semantic segmentation network to obtain a target feature group and corresponding battery capacity measurement interval corresponding to the initial feature group; constructing a prediction model using a gradient descent-based loss function, inputting all target feature groups and corresponding battery capacity measurement intervals into the prediction model, and training the prediction model through multiple rounds of iterations; completing the training of the prediction model in response to the gradient descent of the loss function until a preset convergence accuracy is reached, and using the trained prediction model to predict battery life. This method solves the problems of computational redundancy and overfitting, thereby improving prediction accuracy.

Description

Technical Field

[0001] The embodiments of the present application relate to the technical field of battery health status assessment, and in particular to a battery life prediction method and related devices. Background Art

[0002] In the relevant battery life or capacity prediction methods, when faced with models for multiple scenarios, it is often necessary to manually adjust the model parameters, and not all of the large number of battery-related features used are helpful in judging the battery capacity in a specific scenario. When the prediction model runs a large number of features, it will inevitably lead to increased operating costs and reduced efficiency. In addition, the overly complex structure not only increases the redundancy of calculations, but also makes overfitting very likely to occur.

[0003] Based on this, a solution is needed that can select features with strong discriminative ability and effectively improve prediction efficiency. Summary of the Invention

[0004] In view of this, the purpose of this application is to propose a battery life prediction method and related devices.

[0005] Based on the above objectives, this application provides a battery life prediction method, including:

[0006] Testing the battery under multiple preset test conditions to obtain multiple groups of initial feature sets under each test condition and battery capacity measurement intervals corresponding to each group of initial feature sets;

[0007] Inputting each of the initial feature groups and the corresponding battery capacity measurement interval into a preset point cloud semantic segmentation network to obtain a target feature group corresponding to the initial feature group and the corresponding battery capacity measurement interval;

[0008] Constructing a prediction model using a loss function based on gradient descent, inputting all the target feature groups and the corresponding battery capacity measurement intervals into the prediction model, and training the prediction model through multiple rounds of iterations;

[0009] In response to the gradient descent of the loss function reaching a preset convergence accuracy, the training of the prediction model is completed, and the battery life is predicted using the trained prediction model.

[0010] Furthermore, the battery is tested under multiple preset test conditions to obtain multiple groups of initial feature groups under each test condition and battery capacity measurement intervals corresponding to each group of initial feature groups, including:

[0011] Performing a vibration stress test by applying a preset vibration stress to the battery;

[0012] The battery was subjected to a cycle aging test at discharge rates of 0C, 1C, 2C and 0.5C respectively;

[0013] Performing a capacity test on the battery at a 0.5C rate charge and discharge;

[0014] The battery is Integral capacity-differential voltage curve test of charge and discharge;

[0015] Performing an impedance spectrum test by applying an AC signal of a preset frequency to the battery;

[0016] Under each test condition, determining a battery capacity percentage corresponding to each initial feature group, and dividing the battery capacity percentage into a plurality of intervals according to predetermined interval values;

[0017] Determining intervals corresponding to the respective battery capacity percentages and using them as the battery capacity measurement intervals;

[0018] Each of the initial feature groups includes multiple initial features.

[0019] Furthermore, each of the initial feature groups and the corresponding battery capacity measurement interval is input into a preset point cloud semantic segmentation network to obtain a target feature group corresponding to the initial feature group and the corresponding battery capacity measurement interval, including:

[0020] Construct a point cloud semantic segmentation network consisting of convolutional layers, pooling layers, and fully connected layers;

[0021] Inputting all the initial features and the corresponding battery capacity measurement intervals into the point cloud semantic segmentation network in the form of a matrix;

[0022] Selecting target features from the initial features through the convolution layer, the pooling layer, and the fully connected layer, and forming a plurality of target feature groups;

[0023] Each target feature group includes multiple target features:

[0024] Conductivity loss, active material loss, lithium ion loss, ohmic impedance, charge transfer impedance, solid electrolyte interface film impedance, diffusion impedance, temperature and constant pressure drop time

[0025] Furthermore, a prediction model is constructed using a loss function based on gradient descent, including:

[0026] Construct the loss function as shown below:

[0027]

[0028] Wherein, x represents the residual between the battery capacity prediction interval predicted by the prediction model and the battery capacity measurement interval, a represents the hyperparameter preset in the prediction model, and c represents the coordination parameter preset in the prediction model.

[0029] Furthermore, the training of the prediction model through multiple rounds of iterations includes:

[0030] In each round of iteration, the prediction model is enabled to perform prediction using each target feature group to determine the battery capacity prediction interval corresponding to the target feature group;

[0031] determining a residual between the battery capacity prediction interval and the battery capacity measurement interval;

[0032] Calculating the loss function according to the residual;

[0033] The descent gradient of the loss function is determined according to the calculated result.

[0034] Furthermore, after completing the training of the prediction model, the method further includes:

[0035] Determining the battery capacity prediction interval output by the prediction model in the last round of iteration;

[0036] Determining a root mean square error, an absolute error, an average percentage error, and a maximum error of the battery capacity prediction interval and the battery capacity measurement interval;

[0037] Determine the total number of iterations;

[0038] Determining the memory usage of the prediction model when it is running;

[0039] determining how quickly the model makes predictions;

[0040] Establishing an evaluation function using the root mean square error, absolute error, average percentage error, maximum error, memory usage, and speed according to preset weights;

[0041] The prediction model is evaluated using the evaluation function.

[0042] Based on the same inventive concept, the present application also provides a battery life prediction device, comprising: a test data acquisition module, a feature selection module, a training module and a prediction module;

[0043] The test data acquisition module is configured to test the battery under multiple preset test conditions to obtain multiple groups of initial feature sets under each test condition and the battery capacity measurement intervals corresponding to each group of initial feature sets;

[0044] The feature selection module is configured to input each of the initial feature groups and the corresponding battery capacity measurement interval into a preset point cloud semantic segmentation network to obtain a target feature group corresponding to the initial feature group and the corresponding battery capacity measurement interval;

[0045] a model training module configured to construct a prediction model using a loss function based on gradient descent, input all the target feature groups and the corresponding battery capacity measurement intervals into the prediction model, and train the prediction model through multiple rounds of iterations;

[0046] The prediction module is configured to complete the training of the prediction model in response to the gradient descent of the loss function reaching a preset convergence accuracy, and predict the battery life using the trained prediction model.

[0047] Furthermore, the battery life prediction device further includes: an evaluation module,

[0048] The evaluation module is configured to determine the battery capacity prediction interval output by the prediction model in the last round of iteration;

[0049] Determining a root mean square error, an absolute error, an average percentage error, and a maximum error of the battery capacity prediction interval and the battery capacity measurement interval;

[0050] Determine the total number of iterations;

[0051] Determining the memory usage of the prediction model when it is running;

[0052] determining how quickly the model makes predictions;

[0053] Establishing an evaluation function using the root mean square error, absolute error, average percentage error, maximum error, memory usage, and speed according to preset weights;

[0054] The prediction model is evaluated using the evaluation function.

[0055] Based on the same inventive concept, the present application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and runnable on the processor, wherein when the processor executes the program, the battery life prediction method as described in any one of the above items is implemented.

[0056] Based on the same inventive concept, the present application also provides a non-transitory computer-readable storage medium, wherein the non-transitory computer-readable storage medium stores computer instructions, and the computer instructions are used to enable the computer to execute the battery life prediction method as described above.

[0057] From the above, it can be seen that the battery life prediction method provided by this application obtains a large number of initial features for various working conditions based on tests under different conditions. It comprehensively considers the computational redundancy and overfitting of the prediction model and the network model during operation to select the initial features, so that the prediction model can directly use features with strong discriminative ability to predict the battery life, that is, the battery capacity, thereby reducing the computational burden of the prediction model and improving the accuracy of the prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] In order to more clearly illustrate the technical solutions in this application or related technologies, the following briefly introduces the drawings required for use in the embodiments or related technical descriptions. Obviously, the drawings described below are merely embodiments of this application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0059] Figure 1 This is a flowchart of a battery life prediction method according to an embodiment of the present application;

[0060] Figure 2 This is a schematic diagram of the structure of a battery life prediction method device according to an embodiment of the present application;

[0061] Figure 3 The figure shows the structure of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0062] In order to make the objectives, technical solutions and advantages of this application more clear, this application is further described in detail below in combination with specific embodiments and with reference to the accompanying drawings.

[0063] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in the embodiments of this application should have the usual meanings understood by people with ordinary skills in the field to which this application belongs. "Include" or "comprising" and similar words mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect.

[0064] As described in the background technology section, related battery life prediction methods are still unable to meet the needs of actual prediction work.

[0065] In the battery market, power batteries mainly include lead-acid batteries, nickel-metal hydride batteries and lithium-ion batteries. Among them, lithium-ion batteries are widely used in electric vehicles or hybrid electric vehicles due to their excellent performance in energy density, cycle life, self-discharge rate, charge and discharge performance, operating temperature range and other aspects.

[0066] However, as the number of charge and discharge cycles of lithium-ion batteries increases, the electrochemical reactions within the battery intensify. In addition to the redox reactions caused by lithium ion insertion and removal, there are also many side reactions, such as electrolyte decomposition, active material dissolution, and metallic lithium deposition. These side reactions lead to battery capacity degradation. Under actual electric vehicle driving conditions, the operating environment and performance degradation of batteries are complex. Accurately predicting the remaining battery life, and therefore the battery capacity, requires a large amount of battery-related characteristic data.

[0067] In the process of implementing this application, the applicant discovered that the main problems with the relevant battery life prediction methods are:

[0068] In the relevant battery life or capacity prediction methods, when faced with models for multiple scenarios, it is often necessary to manually adjust the model parameters, and not all of the large number of battery-related features used are helpful in judging the battery capacity in a specific scenario. When the prediction model runs a large number of features, it will inevitably lead to increased operating costs and reduced efficiency. In addition, the overly complex structure not only increases the redundancy of calculations, but also makes overfitting very likely to occur.

[0069] The embodiments of the present application are described in detail below with reference to the accompanying drawings.

[0070] refer to Figure 1 A battery life prediction method according to an embodiment of the present application includes the following steps:

[0071] Step S101 : Testing a battery under multiple preset test conditions to obtain multiple initial feature groups under each test condition and battery capacity measurement intervals corresponding to each initial feature group.

[0072] In the embodiments of the present application, in order to ensure the accuracy of the battery life prediction results, it is necessary to obtain as many battery attenuation conditions as possible under various operating conditions and use them as features of the input prediction model.

[0073] In this embodiment, a vibration stress test is set up to simulate the vibration that the battery may be subjected to in actual use, so as to simulate the working environment of the battery. Specifically, an experimental prototype can be used to apply preset different vibration stresses to the battery to simulate the working environment under various vibration conditions.

[0074] Furthermore, a cyclic aging test is also set up. In this cyclic aging test, the battery to be tested is charged and discharged in cycles at different preset discharge rates until the battery is damaged. Specifically, for example, for the test of the battery of an electric vehicle, the discharge rate can be set to 0C, 1C, 2C and 0.5C respectively to simulate the battery capacity degradation under the conditions of idling, constant speed, acceleration and deceleration of the electric vehicle.

[0075] Furthermore, a capacity test is provided, in which the battery can be charged and discharged at a fixed rate. Specifically, for example, the battery can be fully charged and discharged at a rate of 0.5C to test the degradation of the battery capacity under such operating conditions.

[0076] In this embodiment, the amount of electricity discharged by the battery per unit time is taken as the battery capacity. The unit may be ampere-hour, ie, A·H, or the symbol C to express the battery capacity, and 1A·H=3600C.

[0077] Furthermore, an IC-DV curve (integral capacity-differential voltage curve) test is also set up. In this IC-DV curve test, the battery to be tested can be charged and discharged at a fixed small rate. Specifically, for example, the battery can be fully charged and discharged at a rate of 0.04, and the IC-DV curve of the battery under such conditions can be determined.

[0078] Furthermore, an impedance spectrum test is provided, in which an excitation may be applied to the battery. Specifically, for example, an excitation of 0.01 kHz to 100 kHz may be applied to the battery. In the above-mentioned example of the battery of the electric vehicle, the excitation may be an AC signal.

[0079] In this embodiment, it is necessary to obtain multiple initial features related to battery capacity degradation in each type of test, and obtain the value of each initial feature.

[0080] Specifically, a plurality of different initial characteristics can be determined according to specific circumstances. In this embodiment, voltage, conductivity loss, active material loss, lithium ion loss, ohmic impedance, charge transfer impedance, EIS membrane impedance (solid electrolyte interface membrane impedance), diffusion impedance, temperature and constant pressure drop time are included.

[0081] In this embodiment, in each type of test, the value of each initial feature needs to be obtained multiple times, and the values ​​of multiple initial features obtained simultaneously each time are regarded as a group of initial feature groups.

[0082] Specifically, in each type of test, the initial value of the battery capacity and the initial value of the voltage may be measured before the test, and the measured value of the battery capacity and the measured value of the voltage at the current moment may be measured during multiple charge and discharge cycles.

[0083] Furthermore, the above-determined initial value and measured value of the voltage are used to calculate CL (conductivity loss) using the following formula:

[0084]

[0085] Wherein, V0 represents the initial value of the battery voltage, Vi represents the measured value of the battery voltage at the i-th charge or discharge, or charge and discharge cycle, and LCi represents the conductivity loss at the i-th above cycle.

[0086] Furthermore, the initial value and the measured value of the battery capacity determined above are used to calculate LLI (lithium ion loss) using the following formula:

[0087]

[0088] Wherein, Q0 represents the initial value of the battery capacity, Qi represents the measured value of the battery capacity at the i-th charge or discharge, or charge and discharge cycle, and LLIi represents the lithium ion loss at the i-th above cycle.

[0089] It can be seen that in each of the above types of tests, the voltage and battery capacity can be measured multiple times in each cycle, and the maximum value thereof can be used for calculation in the above formula.

[0090] Furthermore, in combination with the voltage and battery capacity determined above, the following formula can be used to calculate the LAM (active material loss) in the cycle by derivative:

[0091]

[0092] Wherein LAMi represents the loss of active substance at the i-th cycle.

[0093] Furthermore, the ohmic impedance, charge transfer impedance, EIS film impedance, and diffusion impedance can be obtained by converting the test circuit into a second-order equivalent circuit model in various types of tests, and using the second-order equivalent circuit model to determine the above-mentioned ohmic impedance, charge transfer impedance, EIS film impedance, and diffusion impedance.

[0094] Furthermore, in various types of tests, the temperature of the battery can be acquired by using a thermocouple collector to acquire the temperature of the battery surface and use it as the battery temperature.

[0095] Furthermore, in each type of test, the constant voltage drop time can be determined by drawing an SOC curve (charge and discharge curve) regarding voltage and battery capacity and obtaining the time corresponding to the same voltage drop starting from the same moment.

[0096] In some embodiments, the SOC curve may also be drawn based on voltage and capacitance.

[0097] Furthermore, in the IC-DV curve test, by charging and discharging the battery at a rate of 0.04C, an IC-DV curve consisting of an IC curve and a DV curve can be constructed in the following manner:

[0098] First, based on the above-obtained measured values ​​of battery capacity and voltage, the following relationship is determined:

[0099] V i =f(Q i )=f(I i t)

[0100] Where Ii represents the current in the i-th cycle, and f() is used to describe the relationship between Vi and Qi. The relationship between Vi and Qi may be different in different types of tests.

[0101] Furthermore, the derivative of battery capacity with respect to voltage is calculated as follows:

[0102]

[0103] Furthermore, since the IC curve represents the relationship between the battery capacity and the voltage after the voltage is derived, the voltage can be used as the abscissa of the IC curve, and the derivative Δc of the battery capacity with respect to the voltage can be used as the ordinate of the IC curve.

[0104] Furthermore, the derivative of the voltage with respect to the battery capacity is calculated as follows:

[0105]

[0106] Since the DV curve represents the relationship between the voltage and the battery capacity, the battery capacity can be used as the abscissa of the DV curve, and the derivative Δu of the voltage with respect to the battery capacity can be used as the ordinate of the DV curve.

[0107] The above calculations can determine the values ​​of each initial feature under each type of test.

[0108] Furthermore, in this embodiment, in order to use the following point cloud semantic segmentation network to select the most discriminative target feature from the above-mentioned multiple initial features, that is, to select the target feature that is more effective in determining the battery capacity, it is necessary to abstract the relationship between each initial feature and the battery capacity into a classification problem.

[0109] Therefore, in this embodiment, the battery capacity needs to be converted into a plurality of fixed selected battery capacity intervals. Specifically, the initial value of the battery capacity is taken as 100%, and the SOH (battery state of health) percentage corresponding to each measured value of the battery capacity is estimated based on this.

[0110] Furthermore, the SOH percentage is divided into multiple intervals with a fixed interval value. For example, it can be divided into 5 intervals with an interval value of 20% between 0% and 100%, that is, the classification type corresponding to the 5 initial features, and the SOH percentage corresponding to each battery capacity measurement value is determined according to the above-defined intervals, and the determined interval is called the battery capacity measurement interval of the battery capacity.

[0111] In some embodiments, in order to achieve better prediction results, the granularity of the classification type can be increased, that is, more intervals can be divided with smaller interval values.

[0112] Based on the above division, the classification corresponding to each initial feature, that is, the corresponding battery capacity measurement interval, can be determined.

[0113] Step S102: Input each of the initial feature groups and the corresponding battery capacity measurement interval into a preset point cloud semantic segmentation network to obtain a target feature group corresponding to the initial feature group and the corresponding battery capacity measurement interval.

[0114] In an embodiment of the present application, the values ​​of the initial features in all the initial feature groups obtained above need to be input into a pre-built point cloud semantic segmentation network to select target features.

[0115] In this embodiment, it can be determined that the more degradation features there are, the more accurate the prediction model is in predicting battery life. However, when there are too many degradation features, unnecessary computational redundancy in the prediction process will be greatly increased. Moreover, when there are more features, a more complex and sophisticated structure will need to be set up when constructing the prediction model, and an overly sophisticated structure can easily lead to overfitting.

[0116] Therefore, in order to solve the above problem, the multiple initial features obtained above can be screened to select the most discriminative initial features as the target features used in the prediction model. Among them, the most discriminative initial features are the multiple features that have a greater impact on the predicted battery capacity.

[0117] In this embodiment, a pre-trained point cloud semantic segmentation network can be used to screen the above-mentioned multiple initial features. In this embodiment, the point cloud semantic segmentation network is also called a network model, and the network model can include a convolutional layer, a pooling layer, and a fully connected layer.

[0118] Specifically, the initial features obtained above are first normalized to unify the scales of the initial features in the calculation. Furthermore, the normalized initial features and the corresponding battery capacity measurement intervals are used to form multiple input matrices, and all the input matrices are input into the convolutional layer of the point cloud semantic segmentation network. It can be seen that the input matrix is ​​used as the feature space in the initial stage in this embodiment.

[0119] In this embodiment, the convolution layer is used to abstract higher-level features. However, since the convolution layer does not actually reduce the dimension of the input matrix, overfitting is likely to occur. Therefore, a pooling layer is required after the convolution layer to reduce the dimension of the input matrix.

[0120] In the network model of this embodiment, the pre-trained point cloud semantic segmentation network is improved. In order to achieve maximum lightweightness and avoid unnecessary computational redundancy, no activation function is set in the network model of this embodiment. That is, after abstracting the battery life prediction problem into a grouping problem, the initial features and battery capacity measurement intervals input therein can be used as linearly separable data sets, and the battery life prediction problem in this application can be regarded as a linearly separable problem. Therefore, for the linearly separable data, activation functions such as sigmoid function or tanh function can be not set, thereby further lightweighting the structure of the traditional point cloud semantic segmentation network.

[0121] Furthermore, a fully connected layer is provided in the network model to integrate the local information in the pooling layer.

[0122] In some embodiments, an attention mechanism can be used instead of the fully connected layer, and by calculating the attention value of each input matrix, each initial feature is weighed by setting weights to achieve the entire information aggregation process, thereby solving the problem of fixed weights in the fully connected layer.

[0123] In some embodiments, the network model may not be pre-trained, and a preset training data set is used to train the constructed network model, and the training process is completed by setting a cross entropy loss function in the network model.

[0124] Furthermore, by screening the initial features through the network model, multiple features with better discriminative grouping capabilities can be selected from the initial features as target features for the input prediction model.

[0125] In some other embodiments, the target feature may also be selected using the Pearson correlation coefficient.

[0126] Specifically, the Pearson correlation coefficient between each initial feature and the battery capacity can be calculated respectively, and the Pearson correlation coefficient between the two can be used to describe the degree of correlation between the initial feature and the battery capacity. Then, multiple initial features with higher correlation are determined as target features to obtain the feature space in the following steps.

[0127] In some other embodiments, the method of using the above-mentioned Pearson correlation coefficient to describe the degree of correlation between the initial features and the battery capacity can not only be used to screen out target features from the initial features, but can also be combined with the above-mentioned point cloud semantic segmentation network. By calculating the Pearson correlation coefficient, the degree of correlation between each initial feature and the battery capacity is determined to select multiple candidate features from the initial features, and then the selected candidate features are input into the point cloud semantic segmentation network for further feature selection.

[0128] It can be seen that adding a screening step for initial features before the point cloud semantic segmentation network performs feature selection can effectively reduce the resource consumption of the point cloud semantic segmentation network and make the acquisition of target features faster.

[0129] Specifically, the Pearson correlation coefficient between the above initial characteristics and the battery capacity can be calculated using the following formula:

[0130]

[0131] Among them, RUL j represents the output result of the battery capacity under the influence of the jth initial feature, The output result representing the battery capacity under the combined influence of all initial characteristics; Indicates RUL j and Var() represents the variance calculated for the contents in the brackets.

[0132] It can be seen that the Pearson correlation coefficient is obtained by statistically analyzing two variables, namely the above RUL j and The linear correlation between the jth initial feature and the battery capacity is used to evaluate the correlation between the jth initial feature and the battery capacity, so as to determine the influence of the initial feature on the battery capacity.

[0133] Furthermore, the value of the Pearson correlation coefficient can be between 0 and 1. The larger the value, the stronger the correlation between the two variables.

[0134] Among them, in the above formula of Pearson correlation coefficient, if the data within the absolute value is greater than 0, it means RUL j and There is a positive correlation between them; if it is less than 0, it means RULj and There is a negative correlation between them; if it is equal to 0, it means RUL j and There is no correlation between them.

[0135] In this embodiment, the selected target characteristics specifically include: conductivity loss, active material loss, lithium ion loss, ohmic impedance, charge transfer impedance, solid electrolyte interface film impedance, diffusion impedance, temperature and constant pressure drop time.

[0136] In some other embodiments, for the target features determined above, the degree of correlation between any two target features may still be determined by calculating the Pearson correlation coefficient between the two target features.

[0137] Specifically, the Pearson correlation coefficient between any two target features can be calculated using the following formula:

[0138]

[0139] Among them, RUL j Represents the output result of battery capacity under the influence of the jth initial feature, RUL k Represents the output result of the battery capacity under the influence of the kth initial feature; Cov(RUL j ,RUL k ) indicates RUL j With RUL k Var() represents the variance calculated for the contents in the brackets.

[0140] It can be seen that the Pearson correlation coefficient is obtained by statistically analyzing two variables, namely the above RUL j With RUL k The linear correlation between the jth initial feature and the jth initial feature is used to evaluate the correlation between the jth initial feature and the jth initial feature in affecting the battery usage.

[0141] Furthermore, the value of the Pearson correlation coefficient can be between 0 and 1. The larger the value, the stronger the correlation between the two variables.

[0142] Among them, in the above formula of Pearson correlation coefficient, if the data within the absolute value is greater than 0, it means RUL j With RUL k There is a positive correlation between them; if it is less than 0, it means RUL j With RUL k There is a negative correlation between them; if it is equal to 0, it means RUL j With RUL k There is no correlation between them.

[0143] Step S103: construct a prediction model using a loss function based on gradient descent, input all the target feature groups and the corresponding battery capacity measurement intervals into the prediction model, and train the prediction model through multiple rounds of iterations.

[0144] In an embodiment of the present application, based on the target features determined above, a prediction model may be constructed and the target features may be used in the prediction model to predict the battery capacity, that is, to determine a battery capacity prediction interval.

[0145] In some other embodiments, when the number of target features input into the prediction model is insufficient, other available target features may be found based on the correlation coefficient between the two target features calculated above.

[0146] For example, if the Pearson correlation coefficient between the target feature: lithium ion loss and another target feature: active material loss is calculated, it is determined that the correlation between the two target features is high. When one of the target features is missing, the other target feature can be used to replace the missing target feature; or when there are too few target features, another target feature with a higher correlation can be added to the existing target features to ensure that the target features in the input prediction model are sufficient.

[0147] In this embodiment, a prediction model can be constructed based on the principle of decision tree. Specifically, LightGBM (Lightweight Gradient Boosting Machine) is used as the base learner in the prediction model.

[0148] Furthermore, the target features determined above can be input as feature space and input into the base learner of the prediction model in matrix form.

[0149] Furthermore, multiple base learners can be set along the gradient descent direction, and the ensemble formed by connecting multiple base learners can be used as a strong learner, and the loss function shown below can be set in it to train the prediction model:

[0150]

[0151] Among them, x represents the residual between the battery capacity prediction interval predicted by the prediction model and the battery capacity measurement interval, a represents the hyperparameter in the prediction model, and c represents the coordination parameter in the prediction model, where a is used to control the robustness of the above loss function, and c is used to adjust the scale of the curve bending of the loss function at x = 0.

[0152] Furthermore, in this embodiment, the above loss function can be converted into the following form according to different values ​​of the hyperparameters:

[0153]

[0154] in, represents the Cauchy loss function; represents the L2 norm loss function; Represents the Welsh loss function. It can be seen that although a is not defined at a = 0, a = 2, and a = -∞, an approximate result can still be obtained by taking its limit.

[0155] Specifically, when a=2, it is the L2 norm loss function, which is used to represent the minimum square error (the sum of all square differences between the true value and the predicted value). When there are outliers in the data, the L2 norm loss function will lead to poor prediction results due to considering the square difference.

[0156] When a=0, it is the Cauchy loss function, which performs better when training samples containing data noise.

[0157] When a=-∞, it is the Welsch loss function, which has better robustness when training samples containing outliers.

[0158] Furthermore, it can be determined as follows:

[0159]

[0160] Furthermore, this loss function can not only represent the relatively traditional loss function as above, but in some embodiments, depending on the value of a, it can also represent the following L1 norm loss function L1(x) and Geman-McClure loss function G(x):

[0161]

[0162] Furthermore, based on the loss function determined above, a prediction model can be used for training.

[0163] Specifically, the target feature groups output by the above-mentioned point cloud semantic segmentation network and the battery capacity measurement intervals corresponding to each target feature group can be input into the prediction model. By using multiple base classifiers to fit the existing model separately in multiple iterations, the loss can be minimized, that is, gradient descent.

[0164] In some other embodiments, the aforementioned multiple base learners may be combined into a strong learner by calculating the Pearson correlation coefficient.

[0165] Specifically, the Pearson correlation coefficient can be calculated using the following formula:

[0166]

[0167] Among them, α i represents the output of the i-th base learner, β represents the output of the strong learner as a whole; Cov(α i ,β) represents the covariance between the output of the i-th base learner and the overall output of the strong learner, and Var() represents the variance calculated for the content in the brackets.

[0168] It can be seen that the Pearson correlation coefficient evaluates the correlation between the base learner and the strong learner as a whole by statistically analyzing the linear correlation between the two variables, namely α and β, to determine whether the base learner is suitable for selection.

[0169] Furthermore, the value of the Pearson correlation coefficient can be between 0 and 1. The larger the value, the stronger the correlation between the two variables.

[0170] Among them, in the above formula of Pearson correlation coefficient, if the data within the absolute value is greater than 0, it means that there is a positive correlation between α and β; if it is less than 0, it means that there is a negative correlation between α and β; if it is equal to 0, it means that there is no correlation between α and β.

[0171] In some other embodiments of the present application, a strong learner may also be constructed according to different scenario requirements.

[0172] Specifically, as described in the aforementioned step S102, an attention mechanism is used to replace the fully connected layer. That is, the attention mechanism can be used to integrate the base learners into strong learners by paying attention to different scenario requirements, allocating more computing resources to scenario requirements that need to be focused on, and selecting evaluation indicators and cost functions suitable for different scenario requirements.

[0173] Among them, for different scenario requirements, the evaluation indicators can be selected from the various evaluation indicators in the following steps.

[0174] It can be seen that the above-mentioned method of integrating strong learners based on different scenario requirements through the attention mechanism can effectively improve the adaptability and generalization ability of this solution, enabling it to effectively respond to different scenario requirements.

[0175] Step S104: In response to the gradient descent of the loss function reaching a preset convergence accuracy, the prediction model is trained, and the battery life is predicted using the trained prediction model.

[0176] In an embodiment of the present application, for the above-mentioned prediction model, the training process can be controlled by a preset convergence accuracy.

[0177] Specifically, the convergence accuracy can be set according to the specific situation. When the gradient of the loss function drops to the preset convergence accuracy and remains stable, the training of the prediction model can be completed.

[0178] In some embodiments, the training process of the prediction function may also be controlled in other ways.

[0179] For example, a predetermined step size may be set along the direction of gradient descent, and when the training of the prediction model reaches the preset step size, the training may be completed.

[0180] Furthermore, a fixed number of iterations may be set, and when the training of the prediction model reaches the preset number of iterations, the training may be completed.

[0181] In this embodiment, after the prediction model is trained, it can be applied to predict battery life.

[0182] In some embodiments, after the prediction model is trained, it may be evaluated to determine the prediction effect of the prediction model.

[0183] Specifically, the evaluation of the prediction model can be performed using multiple evaluation criteria. For example, multiple evaluation indicators can be set to: root mean square error, mean absolute error, percentage error, maximum error, memory usage, number of iterations, and prediction speed.

[0184] The root mean square error can be calculated using the following formula:

[0185]

[0186] Wherein, RUL represents the battery capacity, the subscript pred.i represents the predicted value for the i-th cycle in the test of the aforementioned embodiment, and the subscript true.i represents the measured value for the i-th cycle.

[0187] Furthermore, the mean absolute error can be calculated using the following formula:

[0188]

[0189] Furthermore, the average percentage error can be calculated using the following formula:

[0190]

[0191] Furthermore, the maximum error can be calculated using the following formula:

[0192]

[0193] In this embodiment, based on the various evaluation indicators determined above, different weights may be set for the various evaluation indicators, and all the evaluation indicators may be integrated through the preset weights.

[0194] In some other embodiments, the set weights may be adjusted according to the specific working conditions of the battery to suit various working conditions.

[0195] It can be seen that the battery life prediction method of the embodiment of the present application obtains a large number of initial features for various working conditions based on tests under different conditions, and comprehensively considers the computational redundancy and overfitting of the prediction model and the network model during operation to select the initial features, so that the prediction model can directly use features with strong discriminative ability to predict the battery life, that is, the battery capacity, thereby reducing the computational burden of the prediction model and improving the accuracy of the prediction.

[0196] It should be noted that the method of the embodiment of the present application can be performed by a single device, such as a computer or server. The method of this embodiment can also be applied in a distributed scenario and completed by multiple devices working together. In the case of such a distributed scenario, one of the multiple devices may only perform one or more steps in the method of the embodiment of the present application, and the multiple devices will interact with each other to complete the method described.

[0197] It should be noted that the above description is limited to some embodiments of the present application. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in an order different from that described in the above embodiments and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require the specific order or sequential order shown to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0198] Based on the same inventive concept, corresponding to any of the above-mentioned embodiments and methods, an embodiment of the present application further provides a battery life prediction device.

[0199] refer to Figure 2 , the battery life prediction device includes: a test data acquisition module 201, a feature selection module 202, a training module 203 and a prediction module 204;

[0200] The test data acquisition module 201 is configured to test the battery under multiple preset test conditions to obtain multiple sets of initial feature groups under each test condition and the battery capacity measurement intervals corresponding to each set of initial feature groups;

[0201] The feature selection module 202 is configured to input each of the initial feature groups and the corresponding battery capacity measurement interval into a preset point cloud semantic segmentation network to obtain a target feature group corresponding to the initial feature group and the corresponding battery capacity measurement interval;

[0202] The model training module 203 is configured to construct a prediction model using a loss function based on gradient descent, input all the target feature groups and the corresponding battery capacity measurement intervals into the prediction model, and train the prediction model through multiple rounds of iterations;

[0203] The prediction module 204 is configured to complete the training of the prediction model in response to the gradient descent of the loss function reaching a preset convergence accuracy, and use the trained prediction model to predict the battery life.

[0204] As an optional embodiment, the test data acquisition module 201 is specifically configured to:

[0205] Performing a vibration stress test by applying a preset vibration stress to the battery;

[0206] The battery was subjected to a cycle aging test at discharge rates of 0C, 1C, 2C and 0.5C respectively;

[0207] Performing a capacity test on the battery at a 0.5C rate charge and discharge;

[0208] The battery is Integral capacity-differential voltage curve test of charge and discharge;

[0209] Performing an impedance spectrum test by applying an AC signal of a preset frequency to the battery;

[0210] Under each test condition, determining a battery capacity percentage corresponding to each initial feature group, and dividing the battery capacity percentage into a plurality of intervals according to predetermined interval values;

[0211] Determining intervals corresponding to the respective battery capacity percentages and using them as the battery capacity measurement intervals;

[0212] Each of the initial feature groups includes multiple initial features.

[0213] As an optional embodiment, the feature selection module 202 is specifically configured to:

[0214] Construct a point cloud semantic segmentation network consisting of convolutional layers, pooling layers, and fully connected layers;

[0215] Inputting all the initial features and the corresponding battery capacity measurement intervals into the point cloud semantic segmentation network in the form of a matrix;

[0216] Selecting target features from the initial features through the convolution layer, the pooling layer, and the fully connected layer, and forming a plurality of target feature groups;

[0217] Each target feature group includes multiple target features:

[0218] Conductivity loss, active material loss, lithium ion loss, ohmic impedance, charge transfer impedance, solid electrolyte interface film impedance, diffusion impedance, temperature and constant pressure drop time.

[0219] As an optional embodiment, the model training module 203 is specifically configured to:

[0220] Construct the loss function as shown below:

[0221]

[0222] Wherein, x represents the residual between the battery capacity prediction interval predicted by the prediction model and the battery capacity measurement interval, a represents the hyperparameter preset in the prediction model, and c represents the coordination parameter preset in the prediction model.

[0223] In each round of iteration, the prediction model is enabled to perform prediction using each target feature group to determine the battery capacity prediction interval corresponding to the target feature group;

[0224] determining a residual between the battery capacity prediction interval and the battery capacity measurement interval;

[0225] Calculating the loss function according to the residual;

[0226] The descent gradient of the loss function is determined according to the calculated result.

[0227] In some embodiments, the battery life prediction device further includes: an evaluation module;

[0228] The evaluation module is configured to determine the battery capacity prediction interval output by the prediction model in the last round of iteration;

[0229] Determining a root mean square error, an absolute error, an average percentage error, and a maximum error of the battery capacity prediction interval and the battery capacity measurement interval;

[0230] Determine the total number of iterations;

[0231] Determining the memory usage of the prediction model when it is running;

[0232] determining how quickly the model makes predictions;

[0233] Establishing an evaluation function using the root mean square error, absolute error, average percentage error, maximum error, memory usage, and speed according to preset weights;

[0234] The prediction model is evaluated using the evaluation function.

[0235] For the convenience of description, the above devices are described as being divided into various modules according to their functions. Of course, when implementing the embodiments of the present application, the functions of each module can be implemented in the same or multiple software and / or hardware.

[0236] The apparatus of the above embodiment is used to implement the corresponding battery life prediction method in any of the above embodiments, and has the beneficial effects of the corresponding method embodiment, which will not be described in detail here.

[0237] Based on the same inventive concept, corresponding to any of the above-mentioned embodiments and methods, an embodiment of the present application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and runnable on the processor, wherein when the processor executes the program, the battery life prediction method as described in any of the above embodiments is implemented.

[0238] Figure 3 10 is a schematic diagram showing a more specific hardware structure of an electronic device provided in this embodiment. The device may include: a processor 1010, a memory 1020, an input / output interface 1030, a communication interface 1040, and a bus 1050. The processor 1010, the memory 1020, the input / output interface 1030, and the communication interface 1040 are communicatively connected to each other within the device via the bus 1050.

[0239] The processor 1010 can be implemented using a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of the present application.

[0240] The memory 1020 can be implemented in the form of ROM (Read Only Memory), RAM (Random Access Memory), static storage devices, dynamic storage devices, etc. The memory 1020 can store an operating system and other application programs. When the technical solutions provided in the embodiments of the present application are implemented through software or firmware, the relevant program codes are stored in the memory 1020 and called and executed by the processor 1010.

[0241] The input / output interface 1030 is used to connect an input / output module to implement information input and output. The input / output module can be configured as a component in the device (not shown in the figure) or can be externally connected to the device to provide corresponding functions. Input devices may include a keyboard, mouse, touch screen, microphone, various sensors, etc., and output devices may include a display, speaker, vibrator, indicator light, etc.

[0242] The communication interface 1040 is used to connect to a communication module (not shown) to enable communication between the device and other devices. The communication module can communicate via a wired method (such as USB, network cable, etc.) or a wireless method (such as mobile network, WiFi, Bluetooth, etc.).

[0243] The bus 1050 comprises a path for transmitting information between the various components of the device (eg, the processor 1010 , the memory 1020 , the input / output interface 1030 , and the communication interface 1040 ).

[0244] It should be noted that although the above device only shows the processor 1010, the memory 1020, the input / output interface 1030, the communication interface 1040, and the bus 1050, in a specific implementation, the device may also include other components necessary for normal operation. In addition, it will be understood by those skilled in the art that the above device may only include the components necessary to implement the embodiments of the present application, and does not necessarily include all the components shown in the figure.

[0245] The apparatus of the above embodiment is used to implement the corresponding battery life prediction method in any of the above embodiments, and has the beneficial effects of the corresponding method embodiment, which will not be described in detail here.

[0246] Based on the same inventive concept, corresponding to any of the above-mentioned embodiment methods, the present application also provides a non-transitory computer-readable storage medium, wherein the non-transitory computer-readable storage medium stores computer instructions, and the computer instructions are used to enable the computer to execute the battery life prediction method described in any of the above embodiments.

[0247] The computer-readable media of this embodiment include permanent and non-permanent, removable and non-removable media that can be used to store information by any method or technology. The information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, read-only compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, tape disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device.

[0248] The computer instructions stored in the storage medium of the above embodiment are used to enable the computer to execute the battery life prediction method described in any of the above embodiments, and have the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0249] Those skilled in the art should understand that the discussion of any of the above embodiments is merely illustrative and is not intended to imply that the scope of the present application (including the claims) is limited to these examples. In line with the present application, the technical features in the above embodiments or different embodiments may be combined, the steps may be implemented in any order, and there are many other variations of the different aspects of the embodiments of the present application as described above, which are not provided in detail for the sake of simplicity.

[0250] In addition, to simplify the description and discussion, and in order not to make the embodiments of the present application difficult to understand, the known power supply / ground connection with the integrated circuit (IC) chip and other components may or may not be shown in the drawings provided. In addition, the device can be shown in the form of a block diagram to avoid making the embodiments of the present application difficult to understand, and this also takes into account the following fact, that is, the details of the implementation of these block diagram devices are highly dependent on the platform of the embodiment to be implemented in the present application (that is, these details should be fully within the scope of understanding of those skilled in the art). When specific details (e.g., circuits) are set forth to describe exemplary embodiments of the present application, it will be apparent to those skilled in the art that the embodiments of the present application can be implemented without these specific details or when these specific details are changed. Therefore, these descriptions should be considered to be illustrative rather than restrictive.

[0251] Although the present invention has been described in conjunction with specific embodiments thereof, many alternatives, modifications, and variations of these embodiments will be apparent to those skilled in the art based on the foregoing description. For example, other memory architectures (e.g., dynamic RAM (DRAM)) may utilize the embodiments discussed.

[0252] The embodiments of the present application are intended to encompass all such substitutions, modifications, and variations that fall within the broad scope of the appended claims. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the embodiments of the present application should be included in the scope of protection of the present application.

Claims

1. A battery life prediction method, characterized in that: include: Testing the battery under multiple preset test conditions to obtain multiple initial feature groups under each test condition and battery capacity measurement intervals corresponding to each initial feature group, wherein each initial feature group includes multiple initial features; Inputting each of the initial feature groups and the corresponding battery capacity measurement interval into a preset point cloud semantic segmentation network to obtain a target feature group corresponding to the initial feature group and the corresponding battery capacity measurement interval; Constructing a prediction model using a loss function based on gradient descent, inputting all the target feature groups and the corresponding battery capacity measurement intervals into the prediction model, and training the prediction model through multiple rounds of iterations; In response to the gradient descent of the loss function reaching a preset convergence accuracy, completing the training of the prediction model, and predicting the battery life using the trained prediction model; Inputting each of the initial feature groups and the corresponding battery capacity measurement interval into a preset point cloud semantic segmentation network to obtain a target feature group corresponding to the initial feature group and the corresponding battery capacity measurement interval includes: Construct a point cloud semantic segmentation network consisting of convolutional layers, pooling layers, and fully connected layers; Inputting all the initial features and the corresponding battery capacity measurement intervals into the point cloud semantic segmentation network in the form of a matrix; Selecting target features from the initial features through the convolution layer, the pooling layer, and the fully connected layer, and forming a plurality of target feature groups; Each target feature group includes multiple target features: Conductivity loss, active material loss, lithium ion loss, ohmic impedance, charge transfer impedance, solid electrolyte interface film impedance, diffusion impedance, temperature and constant pressure drop time.

2. The method according to claim 1, characterized in that The battery is tested under multiple preset test conditions to obtain multiple groups of initial feature sets under each test condition and battery capacity measurement intervals corresponding to each group of initial feature sets, including: Performing a vibration stress test by applying a preset vibration stress to the battery; The battery was subjected to a cycle aging test at discharge rates of 0C, 1C, 2C and 0.5C respectively; Performing a capacity test on the battery at a 0.5C rate charge and discharge; The battery is Integral capacity-differential voltage curve test of charge and discharge; Performing an impedance spectrum test by applying an AC signal of a preset frequency to the battery; Under each test condition, determining a battery capacity percentage corresponding to each initial feature group, and dividing the battery capacity percentage into a plurality of intervals according to predetermined interval values; The intervals corresponding to the respective battery capacity percentages are determined and used as the battery capacity measurement intervals.

3. The method according to claim 1, characterized in that The prediction model is constructed using a loss function based on gradient descent, including: Construct the loss function as shown below: Wherein, x represents the residual between the battery capacity prediction interval predicted by the prediction model and the battery capacity measurement interval, a represents the hyperparameter preset in the prediction model, and c represents the coordination parameter preset in the prediction model.

4. The method according to claim 3, characterized in that The prediction model is trained through multiple rounds of iterations, including: In each round of iteration, the prediction model is enabled to perform prediction using each target feature group to determine the battery capacity prediction interval corresponding to the target feature group; determining a residual between the battery capacity prediction interval and the battery capacity measurement interval; Calculating the loss function according to the residual; The descent gradient of the loss function is determined according to the calculated result.

5. The method according to claim 4, characterized in that After completing the training of the prediction model, the method further includes: Determining the battery capacity prediction interval output by the prediction model in the last round of iteration; Determining a root mean square error, an absolute error, an average percentage error, and a maximum error of the battery capacity prediction interval and the battery capacity measurement interval; Determine the total number of iterations; Determining the memory usage of the prediction model when it is running; determining how quickly the model makes predictions; Establishing an evaluation function using the root mean square error, absolute error, average percentage error, maximum error, memory usage, and speed according to preset weights; The prediction model is evaluated using the evaluation function.

6. A battery life prediction device, comprising: Test data acquisition module, feature selection module, training module and prediction module; The test data acquisition module is configured to test the battery under multiple preset test conditions to obtain multiple groups of initial feature groups under each test condition and battery capacity measurement intervals corresponding to each group of initial feature groups, wherein each initial feature group includes multiple initial features; The feature selection module is configured to input each of the initial feature groups and the corresponding battery capacity measurement interval into a preset point cloud semantic segmentation network to obtain a target feature group corresponding to the initial feature group and the corresponding battery capacity measurement interval; a model training module configured to construct a prediction model using a loss function based on gradient descent, input all the target feature groups and the corresponding battery capacity measurement intervals into the prediction model, and train the prediction model through multiple rounds of iterations; The prediction module is configured to complete the training of the prediction model in response to the gradient descent of the loss function reaching a preset convergence accuracy, and predict the battery life using the trained prediction model; The feature selection module is configured to: Construct a point cloud semantic segmentation network consisting of convolutional layers, pooling layers, and fully connected layers; Inputting all the initial features and the corresponding battery capacity measurement intervals into the point cloud semantic segmentation network in the form of a matrix; Selecting target features from the initial features through the convolution layer, the pooling layer, and the fully connected layer, and forming a plurality of target feature groups; Each target feature group includes multiple target features: Conductivity loss, active material loss, lithium ion loss, ohmic impedance, charge transfer impedance, solid electrolyte interface film impedance, diffusion impedance, temperature and constant pressure drop time.

7. The device according to claim 6, characterized in that Also includes: Evaluation module, The evaluation module is configured to determine the battery capacity prediction interval output by the prediction model in the last round of iteration; Determining a root mean square error, an absolute error, an average percentage error, and a maximum error of the battery capacity prediction interval and the battery capacity measurement interval; Determine the total number of iterations; Determining the memory usage of the prediction model when it is running; determining how quickly the model makes predictions; Establishing an evaluation function using the root mean square error, absolute error, average percentage error, maximum error, memory usage, and speed according to preset weights; The prediction model is evaluated using the evaluation function.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable by the processor, wherein: When the processor executes the computer program, the method according to any one of claims 1 to 5 is implemented.

9. A non-transitory computer-readable storage medium, characterized in that The non-transitory computer-readable storage medium stores computer instructions for causing a computer to execute the method according to any one of claims 1 to 5.

Images