Method, device and equipment for predicting service life of battery and medium

Abstract

The invention provides a battery life prediction method and device, equipment and a medium, and the method comprises the steps: carrying out the feature extraction of the battery data of a target battery, obtaining a battery feature vector, inputting the battery feature vector into a trained probability prediction model, carrying out the prediction of the battery life, and obtaining a predicted life set of the target battery, finally, a probability density estimation method is adopted, the predicted life of the target battery is determined based on the predicted life set, and a probability prediction model is obtained through training based on the battery data sample set and the trained physical information neural network. The trained physical information neural network is used for carrying out life decay prediction on the battery life predicted by the probability prediction model in the training process to obtain a life decay rate, and the life decay rate is used for determining decay physical loss under a preset battery physical constraint condition so as to adjust parameters of the probability prediction model in a back propagation manner. The embodiment of the invention can improve the precision of battery life prediction.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of batteries, in particular to a battery life prediction method and device, equipment and storage medium. BACKGROUND

[0002] Battery life refers to the ratio of current battery capacity to rated capacity. Battery life prediction is one of the core links of battery health management. As the core energy source of key devices such as satellites, electric vehicles and energy storage power stations, the accurate prediction of battery life is related to the reliable operation, cost control and safety of the devices.

[0003] Battery life is usually dynamically changed by various factors, such as battery materials, use environment, charging and discharging strategy, manufacturing process, etc. Therefore, how to improve the accuracy of battery life prediction is also a problem to be solved. SUMMARY

[0004] Therefore, the present application provides a battery life prediction method, device, equipment and storage medium to at least solve the problems in the related art.

[0005] Specifically, the present application is realized by the following technical solutions: The present application provides a battery life prediction method, comprising: Obtaining battery data of a target battery, and extracting features of the battery data according to a preset dimension to obtain a battery feature vector; Inputting the battery feature vector into a trained probability prediction model to predict the battery life, and obtaining a prediction life set of the target battery; the probability prediction model is trained based on a battery data sample set and a trained physical information neural network, the battery data sample set includes a plurality of battery data sample pairs of batteries, each battery data sample pair includes battery data samples of each battery at different charging stages; the trained physical information neural network is used to predict the life decay of the battery life predicted by the probability prediction model during training, to obtain a life decay rate, and the life decay rate is used to determine the decay physical loss under a preset battery physical constraint condition, to adjust the parameters of the probability prediction model by back propagation; Using a probability density estimation method, determining the life probability density corresponding to each prediction life in the prediction life set, and determining the prediction life of the target battery according to each life probability density.

[0006] In some embodiments, the trained probability prediction model is trained by the following steps: selecting a subset from the set of battery data samples, and determining, for each pair of battery data samples in the subset, a probability distribution parameter of a latent variable corresponding to each battery data sample in the pair of battery data samples; determining a probability difference loss based on the true value of the life span of each battery data sample in the subset and the probability distribution parameter of the latent variable corresponding thereto respectively; performing, according to the battery data samples, decay rate prediction by using the trained physical information neural network to obtain a battery life span decay rate; determining a rate decay loss based on the battery life span decay rate corresponding to each battery data sample, and determining the decay physical loss based on the battery life span decay rate corresponding to each battery data sample and a preset battery physical constraint condition; determining a total loss based on the probability difference loss, the rate decay loss and the decay physical loss, and adjusting model parameters of the probability prediction model based on the total loss, and repeating the training steps until a preset training requirement is met to obtain the trained probability prediction model.

[0007] In some embodiments, the probability prediction model comprises an encoder and a decoder; and the determining, for each pair of battery data samples in the subset, a probability distribution parameter of a latent variable corresponding to each battery data sample in the pair of battery data samples comprises: inputting each battery data sample in the pair of battery data samples into the encoder to obtain the probability distribution parameter of the latent variable corresponding to the battery data sample; performing, according to the battery data samples, decay rate prediction by using the trained physical information neural network to obtain a battery life span decay rate, comprises: obtaining gradient information corresponding to the battery data sample by using an automatic differentiation method to perform partial derivation on the battery data sample; obtaining battery physical information of a battery to which the battery data sample belongs, and performing, based on the battery data sample, the gradient information and the battery physical information, decay rate prediction by using the trained physical information neural network to obtain the battery life span decay rate; adjusting model parameters of the probability prediction model based on the total loss comprises: adjusting parameters of the encoder and the decoder based on the total loss respectively.

[0008] In some embodiments, the determining a rate decay loss based on the battery life span decay rate corresponding to each battery data sample comprises: determine a predicted life span corresponding to each battery data sample based on a probability distribution parameter of a latent variable corresponding to the battery data sample, and automatically differentiate the predicted life span to obtain a predicted decay rate; For each battery data sample, determine a sub-rate decay loss based on the predicted decay rate corresponding to the battery data sample and the battery life span decay rate corresponding to the battery data sample. Determine the rate decay loss based on the sub-rate decay loss corresponding to each battery data sample.

[0009] In some embodiments, the preset life span decay constraint condition includes that the battery life span is inversely proportional to the number of charge and discharge times, and the decay rate of the predicted life span is within a preset decay rate range; and the determination of the decay physical loss based on the battery life span decay rate corresponding to each battery data sample and the preset battery physical constraint condition comprises: For each battery data sample, determine a first sub-physical loss based on the difference between the two predicted life spans corresponding to the battery data sample, and determine a first physical loss based on the first sub-physical loss corresponding to each battery data sample; and For any predicted life span pair, take the partial derivative of the two predicted life spans respectively to obtain the predicted decay probability corresponding to the two predicted life spans respectively, and determine a second sub-physical loss based on the predicted decay probability corresponding to the two predicted life spans respectively and the preset decay rate range using a linear rectifier function, and determine the sum of each second sub-physical loss as a second physical loss. Determine the decay physical loss based on the first physical loss and the second physical loss.

[0010] In some embodiments, the probability prediction model comprises an encoder and a decoder; and the inputting of the battery feature vector into the trained probability prediction model for battery life span prediction to obtain a predicted life span set of the target battery comprises: Encode the battery feature vector using the encoder to obtain the probability distribution parameter of the latent variable corresponding to the battery feature vector; Based on the probability distribution parameter, sample the latent variable to obtain a plurality of latent variable samples and use the decoder to respectively map the battery life span of the plurality of latent variable samples to obtain the predicted life span set.

[0011] In some embodiments, the obtaining of the battery data of the target battery comprises: Obtain initial battery data of the target battery in a timing state during a charging process; Perform anomaly detection on the initial battery data to determine abnormal data, and adjust the abnormal data to obtain the battery data.

[0012] In some embodiments, the preset dimensions include at least one of a mean value, a standard deviation, a skewness, a kurtosis, a charging duration, a total charging amount, and a rate of change of battery data during charging.

[0013] The application further provides a battery life prediction device, comprising: An acquisition module is configured to acquire battery data of a target battery and perform feature extraction on the battery data according to preset dimensions to obtain a battery feature vector. A prediction module is configured to input the battery feature vector into a trained probability prediction model to predict the battery life of the target battery and obtain a prediction life set of the target battery. The probability prediction model is trained based on a battery data sample set and a trained physical information neural network. The battery data sample set includes a plurality of battery data sample pairs of batteries, and each battery data sample pair includes battery data samples of each battery at different charging stages. The trained physical information neural network is configured to predict a life decay rate of the battery life predicted by the probability prediction model during training, determine a decay physical loss under a preset battery physical constraint condition, and adjust parameters of the probability prediction model through back propagation. A determination module is configured to determine a life probability density corresponding to each prediction life in the prediction life set by using a probability density estimation method and determine the prediction life of the target battery according to the life probability densities.

[0014] The application further provides a computer device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps of the battery life prediction method in any of the preceding embodiments are implemented.

[0015] The application further provides a computer-readable storage medium, which stores a computer program. When the program is executed by a processor, the steps of the battery life prediction method in any of the preceding embodiments are implemented.

[0016] The application further provides a computer program product, which includes a computer program. When the computer program is executed by a processor, the steps of the battery life prediction method in any of the preceding embodiments are implemented.

[0017] The technical solutions provided by the embodiments of the application can have the following beneficial effects: In the embodiment of the present application, the life attenuation rate output by the physical information neural network is used to determine the attenuation physical loss under the preset battery physical constraint condition. It can be understood that the preset battery physical constraint condition is used to constrain the physical law, that is, the data-driven and physical law are fused in the training process. In this way, the model can learn the prediction results that conform to the physical law, thereby improving the accuracy of the probability prediction model. Furthermore, the trained probability prediction model is used to predict the life of the target battery, which can improve the accuracy of the predicted life.

[0018] It should be understood that the foregoing general description and the following detailed description are only exemplary and explanatory, and cannot limit the present specification. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1 is a flowchart of a battery life prediction method according to an example embodiment of the present application; Figure 2 is a structural diagram of a probability prediction model according to an example embodiment of the present application; Figure 3 is a comparison diagram of a probability prediction interval according to an example embodiment of the present application; Figure 4 is a comparison diagram of the average absolute error of the predicted life according to an example embodiment of the present application; Figure 5 is a flowchart of a training process of a probability prediction model according to an example embodiment of the present application; Figure 6 is a schematic diagram of a voltage data extraction interval according to an example embodiment of the present application; Figure 7 is a schematic diagram of a current data extraction interval according to an example embodiment of the present application; Figure 8 is a structural diagram of a physical information neural network according to an example embodiment of the present application; Figure 9 is a training diagram of a probability prediction model according to an example embodiment of the present application; Figure 10 is a structural diagram of a battery life prediction device according to an example embodiment of the present application; Figure 11 is a structural diagram of another battery life prediction device according to an example embodiment of the present application; Figure 12 is a hardware structure diagram of a computer device according to an example embodiment of the present application. DETAILED DESCRIPTION

[0020] The exemplary embodiments will be described in detail herein with reference to the attached drawings. In the following description, like reference numerals refer to like elements, unless the context clearly dictates otherwise. The following description of exemplary embodiments is not intended to represent all embodiments in accordance with the present application. Rather, they are merely examples in accordance with some aspects of the present application as detailed in the appended claims.

[0021] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the present application. As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms "comprises" and / or "comprising," when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0022] It is to be understood that the singular forms "a", "an", and "the" include plural referents unless the context clearly dictates otherwise. It is to be further understood that the term "or" as used herein encompasses both exclusive and inclusive or unless the context clearly dictates otherwise. The terms "comprise", "comprising", "comprises" and / or "comprising" when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and / or components but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0023] Battery life refers to the ratio of current battery capacity to rated capacity, and the prediction of battery life is one of the core links of battery health management. As the core energy source of key equipment such as satellites, electric vehicles, and energy storage power stations, the accurate prediction of the life of the battery is related to the reliable operation, cost control, and safety protection of the equipment.

[0024] Battery life is usually dynamically changed by various factors, such as battery materials, use environment, charging and discharging strategy, manufacturing process, etc. Therefore, how to improve the accuracy of battery life prediction is also a problem to be solved.

[0025] Traditional model-driven methods construct mathematical models based on physical or electrochemical principles, with good interpretability and extrapolation ability, but the model construction process is complex and involves many unknown or difficult to accurately measure parameters, limiting the improvement of prediction accuracy. Data-driven methods rely on a large amount of historical data for modeling and prediction, and can handle complex nonlinear relationships, but require high data quality and coverage. In the satellite scenario, it is difficult to obtain sufficient representative data due to high data acquisition costs and harsh environments, affecting the prediction performance and generalization ability.

[0026] The physical information neural network (PINN) combines the advantages of data-driven and model-driven by embedding physical principles into neural networks, and can maintain good prediction performance even with small amounts of data. However, it only gives a single deterministic prediction value and cannot measure prediction uncertainty, making it difficult to support risk assessment and decision making. Although traditional probabilistic prediction methods can provide prediction intervals and quantify prediction uncertainty, the model assumptions are too idealized, parameter estimation is complex and difficult, and the ability to handle complex nonlinear relationships is poor, making it difficult to meet the demand for high-precision prediction.

[0027] Based on the above research, the present disclosure provides a battery life prediction method, which first acquires battery data of a target battery, and extracts features from the battery data according to a preset feature dimension to obtain a battery feature vector; secondly, the battery feature vector is input into a trained probability prediction model for battery life prediction to obtain a predicted life set of the target battery; the probability prediction model is trained based on a battery data sample set and a trained physical information neural network, the battery data sample set includes a plurality of battery data sample pairs of batteries, and each battery data sample pair includes battery data samples of each battery at different charging stages; the trained physical information neural network is used to predict the life attenuation of the battery life predicted by the probability prediction model during the training process to obtain a life attenuation rate, the life attenuation rate is used to determine the attenuation physical loss under a preset battery physical constraint condition to adjust the parameters of the probability prediction model through back propagation; then a probability density estimation method is used to determine the life probability density corresponding to each predicted life in the predicted life set, and the predicted life of the target battery is determined according to each life probability density.

[0028] In the embodiments of the present application, the life attenuation rate output by the physical information neural network is used to determine the attenuation physical loss under the preset battery physical constraint condition. It can be understood that the preset battery physical constraint condition is used to constrain the physical law, that is, the data-driven and physical law are combined during the training process. In this way, the model can be guided to learn the prediction results that conform to the physical law, thereby improving the accuracy of the probability prediction model. Furthermore, the trained probability prediction model is used to predict the life of the target battery, which can improve the accuracy of the predicted life.

[0029] For the convenience of understanding the present embodiment, first, a battery life prediction method disclosed by the present embodiment is introduced in detail. The execution subject of the battery life prediction method provided by the present embodiment is generally a computer device. The computer device can be a server. The server can be a physical server, a server cluster composed of multiple physical servers, or a distributed system. The server can also be a cloud server providing basic cloud computing services such as cloud service, cloud database, cloud computing, cloud storage, big data, and artificial intelligence platform. In other embodiments, the computer device can also be a terminal device. The terminal device can be a mobile device, a terminal, a handheld device, a computing device, a vehicle-mounted device, etc.

[0030] In other embodiments, the method can also be applied to an implementation environment composed of a computer device and a server, or an implementation environment composed of a terminal device and a server. In addition, the battery life prediction method can also be realized by a processor calling computer readable instructions stored in a memory.

[0031] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application.

[0032] Please refer to the drawings of the present application Figure 1 A flowchart of a battery life prediction method according to an exemplary embodiment of the present application is shown. As shown in Figure 1 The battery life prediction method in the present embodiment can include the following steps S101-S103: S101: Obtain battery data of a target battery, and perform feature extraction on the battery data according to a preset feature dimension to obtain a battery feature vector.

[0033] In the present embodiment, the type of the target battery can be a lithium ion battery. In other embodiments, the target battery can also be a sodium ion battery or a nickel-hydrogen battery, etc., which is not limited here.

[0034] The battery data of the target battery can include a plurality of data with time sequence states in any charging and discharging stage. The battery data includes voltage data and current data.

[0035] Here, the battery data can be feature extracted according to a preset feature dimension based on a preset feature extraction network. The preset feature dimension includes at least one of the following: mean, standard deviation, skewness, kurtosis, charging duration, total charging amount, and change rate of battery data in the charging process.

[0036] As mentioned above, battery data includes voltage data and current data. Therefore, during feature extraction, features are extracted from the voltage data and current data according to the preset feature dimensions to obtain the battery feature vector.

[0037] The following section provides a detailed introduction to each preset feature dimension based on formulas (1) to (10).

[0038] (1) Mean : Used to reflect the average level of voltage and current data, its expression is shown in formula (1): (1) Where n is the total amount of data, Data for any time-series state during any charge / discharge phase.

[0039] (2) Standard deviation : Used to reflect the dispersion of current and voltage signals, its expression is shown in formula (2): (2) (3) Skewness : Used to reflect the symmetry of current and voltage signals, its expression is shown in formula (3): (3) (4) Kurtosis : Used to reflect the sharpness of the current and voltage signal distribution, its expression is shown in formula (4): (4) (5) Charging time: used to reflect the time from the start of the charging process. End time The duration of is expressed as shown in formula (5): (5) (6) Cumulative electricity consumption Used to reflect the current during the charging process. The total amount is expressed as shown in formula (6): (6) (7) Curve slope Used to reflect the start time Battery data End time Battery data The rate of change of is expressed as shown in formula (7): (7) In some embodiments, since the battery data is collected by a sensor, the sensor may be affected by environmental interference and equipment failure during data collection, resulting in abnormal values in the battery data. Therefore, after obtaining the battery data of the target battery, the battery data can be preprocessed, specifically, the initial battery data of the target battery in a charging process can be obtained, and then the initial battery data is detected for abnormal data, and the abnormal data is adjusted to obtain the battery data.

[0040] In this embodiment, the Hampel outlier detection algorithm is used to detect the abnormal data in the initial battery data. Specifically, the initial battery data of the time sequence state is taken by a preset sliding window, for each sub-time sequence state data obtained by sliding, the median of the sub-time sequence state data is determined, and the median absolute deviation of the sub-time sequence state data is determined based on the median. Then, the standard deviation is determined according to the median absolute deviation, the absolute deviation threshold is determined based on the standard deviation and the preset threshold, if the absolute deviation of any data of the sub-time sequence state data and the median is greater than the absolute deviation threshold, the data is determined as abnormal data, and the abnormal data is modified to the median. In this way, the abnormal data in each sub-time sequence state data can be modified, and the battery data can be obtained based on each modified sub-time sequence state data.

[0041] For example, the specific steps of the Hampel algorithm are as follows: define the radius of the sliding window K and the threshold (typically 3), and the total window size is . From the beginning to the end of the time sequence, take the sub-sequence in the sliding window centered on each data point . For the part at both ends of the sequence that is insufficient for the size of the sliding window, symmetric padding (mirror boundary value) or only taking the side with data is used. Sort the data in the sliding window, and take the median of the data . Calculate the absolute deviation of each data point in the window from , and then take the median of the deviations to get the median absolute deviation MAD.

[0042] (8) Under the assumption of normal distribution, MAD can be converted to an unbiased estimate of the standard deviation by a coefficient of 1.4826 (approximately equal to 1 / 0.6745) (as formula (9)): (9) If the current data point satisfies , mark as abnormal data, and replace the abnormal data with the median of the current sliding window ; for non-outlier data, the original value is retained.

[0043] The Hampel algorithm uses the median and MAD which are insensitive to outliers to robustly detect outliers, and then by locally and selectively correcting outliers, the structure and features of the original data can be maximally retained.

[0044] S102: input the battery feature vector into the trained probability prediction model for battery life prediction to obtain a predicted life set of the target battery; the probability prediction model is trained based on a battery data sample set and a trained physical information neural network, the battery data sample set includes a plurality of battery data sample pairs of batteries, each battery data sample pair includes battery data samples of each battery at different charging stages; the trained physical information neural network is used to predict life decay of the battery life predicted by the probability prediction model during training to obtain a life decay rate, and the life decay rate is used to determine a decay physical loss under a preset battery physical constraint condition to adjust parameters of the probability prediction model in a back propagation manner.

[0045] Wherein, the probability prediction model is constructed based on a variational inference framework, please refer to Figure 2 , which is a structural diagram of a probability prediction model provided by an exemplary embodiment of the present application. As Figure 2 shown, the probability prediction model includes an encoder and a decoder , wherein, and represent the parameters of the encoder and the decoder respectively.

[0046] For detailed training process of the probability prediction model, please refer to the following.

[0047] Optionally, when the battery feature vector is input into the trained probability prediction model for battery life prediction to obtain the predicted life set of the target battery, the following steps can be included: (A) using the encoder to encode the battery feature vector to obtain the probability distribution parameters of the latent variable corresponding to the battery feature vector.

[0048] Here, the encoder maps the input battery feature vector to the latent variable space, and outputs the probability distribution parameters of the latent variable , as shown in formula (10): (10) Wherein, is the mean of the latent variable, is the standard deviation of the latent variable, and The calculation is performed by an encoder network (such as formula (11)): (11) The present application is based on a multi-layer perceptron (MLP) to construct an encoder, which maps high-dimensional input data to a low-dimensional representation space. The encoder structure includes an alternating stack of multiple linear transformation layers and activation function layers. Specifically, the input layer first projects the original high-dimensional data to a hidden feature space through linear transformation, and then applies a sine function activation; the intermediate hidden layer continues to transform the features through linear transformation combined with the sine activation function, and introduces a random dropout regularization mechanism after each activation to prevent overfitting; the output layer finally maps the hidden layer features to the target dimension through linear transformation.

[0049] In terms of parameter initialization, the present application uses the Xavier normal distribution initialization method to set a suitable initial range for the weights of each layer of the network. This initialization strategy can maintain the stability of the signal amplitude during forward propagation and backpropagation, thereby promoting the convergence of the training process and improving the training efficiency.

[0050] (B) Based on the probability distribution parameters, sample the latent variables to obtain a plurality of latent variable samples, and use the decoder to respectively perform battery life mapping on the plurality of latent variable samples to obtain the set of predicted lives.

[0051] In the present application, the reparameterization technique is used to randomly sample the latent variables based on the probability distribution parameters, as shown in formula (12): wherein, represents an element-wise product, is a noise vector sampled from a standard normal distribution.

[0052] After obtaining a plurality of latent variables, the plurality of latent variables are input into the decoder respectively, so that the decoder maps the plurality of latent variables to the battery life space to obtain the set of predicted lives .

[0053] S103: Adopting a probability density estimation method, determining the life probability density based on each predicted life in the set of predicted lives, and determining the predicted life of the target battery according to the life probability density.

[0054] In the present embodiment, the probability density estimation method includes a kernel density estimation method, the expression of which is shown in formula (13).

[0055] (13) wherein, is a probability density function, For predicting the lifetime, c is a battery lifetime variable, is a kernel function (e.g., Gaussian kernel), is a bandwidth parameter.

[0056] Further, after obtaining the probability density function, the maximum point of the probability density function is determined directly by formula (14) for optimization, and the predicted lifetime corresponding to the maximum point is determined as the predicted lifetime of the target battery.

[0057] (14) wherein, is the predicted lifetime of the target battery.

[0058] In the embodiments of the present application, in order to further evaluate the distribution characteristics of the predicted lifetime, quantile can be calculated by numerical method, so as to quantify the distribution characteristics of the data.

[0059] In the present application, the lifetime cumulative distribution function is constructed based on the lifetime probability density, the lifetime limit of the cumulative distribution function is found by using the quantile finding function, and the probability prediction interval of the predicted lifetime is determined.

[0060] Specifically, first, the cumulative distribution function (CDF) is constructed by discrete numerical integration: (15) wherein, k is the number of discrete probability densities, and the discrete is the grid spacing.

[0061] After obtaining the discrete cumulative distribution, the distribution result is normalized to ensure that the cumulative probability sum is 1.

[0062] As known, the definition of quantile is that for a given quantile point p (0 p <1, such as p =0.1 corresponds to the 10% quantile), find the value , such that It can be understood that since the cumulative distribution function is a discrete sequence, the probability of directly finding is low, therefore the quantile finding function needs to be realized by "locating the index closest to p " and "linear interpolation".

[0063] Here, the core of the probability prediction interval is a "numerical range containing a specified probability", which is achieved by calculating the two-sided quantile. Specifically, by calculating the specific quantile points (such as 10% and 90%), an 80% probability prediction interval of the data can be constructed. Thus, the uncertainty range of the predicted life can be quantified according to the determined 80% probability prediction interval, wherein the smaller the interval width, the smaller the prediction uncertainty, and the wider the interval width, the greater the uncertainty.

[0064] See Figure 3 A comparative diagram of a probability prediction interval is provided for an exemplary embodiment of the present application. The present application compares the probability prediction interval obtained using DeepAR with the probability prediction interval obtained by the method described in the present application on the four public data sets of MIT, HUST, TJU and XJTU, with 80% confidence interval as the evaluation standard, as shown in Figure 3 The probability prediction interval obtained by the present application reduces the probability prediction interval width by 64% compared with the traditional probability prediction method, effectively reducing the probability prediction interval range, without significantly reducing the coverage probability.

[0065] See Figure 4 A comparative diagram of the mean absolute error of the predicted life is provided for an exemplary embodiment of the present application. The present application uses kernel density estimation method to fit the probability density function of the predicted value on the four public data sets of MIT, HUST, TJU and XJTU, and selects the predicted life corresponding to the maximum probability density as the final prediction result.

[0066] On this basis, the prediction performance is evaluated by mean absolute error (MAE), and compared with other prediction methods, as shown in Figure 4 The DeepAR model uses the same set processing method; the CNN and MLP are implemented based on the classical network structure, and the parameter size is consistent with the probability prediction model based on physical information neural network (ProbPINN), that is, the network structure used in the inference stage. From Figure 4 It can be seen that the prediction method proposed in the present application performs better than the traditional point prediction model on the MIT, HUST and TJU data sets, and the effect on the XJTU data set is generally flat with the traditional model. Overall, the life prediction method described in the present application improves the prediction accuracy on the premise of not increasing the network complexity.

[0067] Figure 5 A flowchart of the training process of a probability prediction model is provided for an exemplary embodiment of the present application. The training process of the probability prediction model in step S102 is described in detail below. Figure 5

[0068] As shown in Figure 5 ​As shown, the training step of the probability prediction model includes the following steps S501-S506: S501: Select a subset from the battery data sample set, and for each battery data sample pair in the subset, determine the probability distribution parameters of the hidden variable corresponding to each battery data sample in the battery data sample pair.

[0069] The battery data sample set includes a plurality of battery data sample pairs of a plurality of batteries, and each battery data sample pair includes battery data samples of each battery at different charging stages. It should be understood that the battery will be repeatedly charged and discharged, so that the battery data samples of the battery at different charging stages can be obtained. It should be noted that in the discharge stage, the battery data will be affected by the load, and therefore, the present application only selects the battery data in the charging stage when constructing the battery data sample set.

[0071] In the present application, different ranges of battery data are selected according to different data types when constructing the battery data sample. Specifically, for voltage data, data in the constant current charging segment cutoff voltage range is selected; for current data, data in the constant voltage charging segment range is selected for feature extraction to obtain the battery data sample.

[0072] Please refer to Figures 6-7 , Figure 6 a schematic diagram of a voltage data extraction interval provided by an exemplary embodiment of the present application, Figure 7 a schematic diagram of a current data extraction interval provided by an exemplary embodiment of the present application. As Figure 6 shown. The horizontal axis is the charging time, and the vertical axis is the voltage amplitude. As can be seen from the figure, the cutoff voltage is 4.2V, and the extraction range of the voltage data is 4V-4.2V. The blue curve represents the voltage change curve in the 10th charging stage, the yellow curve represents the voltage change curve in the 40th charging stage, and the purple curve represents the voltage change curve in the 70th charging stage.

[0073] Similarly, as Figure 7 shown. The horizontal axis is the charging time, and the vertical axis is the current amplitude. As can be seen from the figure, the extraction range of the current data is 0.1A-0.5A. The green curve represents the current change curve in the 10th charging stage, the orange curve represents the current change curve in the 40th charging stage, and the gray curve represents the current change curve in the 70th charging stage.

[0074] In addition, in constructing the battery data sample, the battery data of different batteries in different charge and discharge stages is also extracted according to the preset feature dimension, and the process is similar to the content described in the foregoing step S101, and thus, details are not described herein.

[0075] An exemplary battery data sample set Each battery data sample in the battery data sample set represents the data features of the battery in a certain charge and discharge cycle, and specifically, the battery data sample includes a feature vector and the current charge and discharge cycle That is, , d represents the preset feature dimension, is a real number set.

[0076] For each battery data sample pair in the subset, the probability distribution parameters of the hidden variables corresponding to each battery data sample in the battery data sample pair are determined, and for example, any two battery data samples of the same battery and are respectively input into the encoder of the probability prediction model to obtain the corresponding hidden variable distribution parameters and .

[0077] S502: Based on the life true value of each battery data sample in the subset and the probability distribution parameters of the corresponding hidden variables, a total probability difference loss is determined.

[0078] It can be understood that each battery data sample has a corresponding life true value, and in the embodiment, a Gaussian negative log-likelihood loss function (as shown in formula (16)) is used for measurement to obtain the probability difference loss: (16) Wherein, is the probability difference loss, represents the number of samples, and M represents the dimension of the hidden variable z, is the remaining life of the charge and discharge cycle.

[0079] The Gaussian negative log-likelihood loss function can evaluate the matching degree between the hidden variable distribution parameters output by the model and the life true value, and the probability difference losses of the battery data samples are summed to obtain the total probability difference loss.

[0080] It can be understood that the smaller the total probability difference loss value is, the more matched the hidden variable distribution parameters output by the model and the true value are, that is, the stronger the model's representation ability for the battery data is. The loss function not only considers the accuracy of the mean prediction, but also considers the estimation of the prediction uncertainty of the model.

[0081] S503: For each battery data sample, according to the battery data sample, the physical information neural network is trained to predict the decay rate, and the battery life decay rate is obtained.

[0082] In this application, the physical neural information network is constructed according to the battery type, that is, according to the foregoing, the battery type can include lithium ion battery, sodium ion battery or nickel hydrogen battery, etc. Because the physical reaction of different types of batteries is different, the establishment of the physical information neural network is also different.

[0083] Taking lithium ion battery as an example, the main side reaction of lithium ion battery is the formation and thickening of SEI film, which consumes active lithium and is the main factor leading to capacity attenuation and life shortening of lithium battery. Therefore, the establishment of the physical information neural network for lithium battery needs to be constructed according to the above-mentioned side reaction. As shown in formula (17) to formula (19), the SEI film thickness growth rate is affected by reaction rate constant, activation energy, gas constant, temperature, battery state of charge and electrolyte concentration, etc. (17) Among them, is the thickness of SEI film, represents the reaction rate constant, represents the activation energy, represents the gas constant, represents the temperature, represents the battery state of charge, represents the electrolyte concentration.

[0084] The formation of SEI film will permanently consume lithium ions, leading to capacity attenuation: (18) Among them, is the decay capacity of lithium battery, that is, and are positively correlated, which means that the thicker the SEI film grows, the more the battery life decreases.

[0085] Therefore, the battery life decay rate can be expressed as: (19) Among them, is the decay rate, represents the initial battery capacity, is a function.

[0086] Since the battery life decay rate is difficult to characterize in an explicit analytical form, the model framework with physical information neural network as the core is constructed in this application to approximate the implicit relationship, which is expressed as wherein, are network parameters of the physical information neural network.

[0087] The architecture of the physical information neural network adopts the same form as the encoder, and by embedding the residual constraint of the battery life decay rate physical equation, the physical information neural network is forced to satisfy the physical consistency during the training process, thereby realizing the interpretable prediction of the life decay rate.

[0088] Specifically, for step S503, when the battery life decay rate is obtained by training the physical information neural network according to the battery data sample, the following steps (A)~(B) can be included: (A) The partial derivative of the battery data sample is obtained by using the automatic differentiation method, and the gradient information corresponding to the battery data sample is obtained.

[0089] As shown in formula (20), the expression of the gradient information is: (20) wherein, is the battery data sample, is the probability prediction model, is the gradient information.

[0090] (B) The battery physical information of the battery to which the battery data sample belongs is obtained, and the decay rate prediction is performed by the trained physical information neural network based on the battery data sample, the gradient information and the battery physical information, to obtain the battery life decay rate.

[0091] Here, the battery physical information can include temperature, internal resistance and the like, which is not limited herein.

[0092] Exemplarily, please refer to Figure 8 is a structural schematic diagram of a physical information neural network provided by an exemplary embodiment of the present application. As shown in Figure 8 The physical information neural network receives the predicted life output by the probability prediction model, t represents the current charge and discharge cycle, represents the feature vector in the battery data sample, represents the predicted life, represents the differential of the predicted life of the probability prediction model with respect to the battery data sample, represents the differential of the predicted life of the probability prediction model with respect to the current charge and discharge cycle.

[0093] Limited by the inconsistency of the actual battery measurable sensor data, the physical information neural network model is not limited to fixed physical information (temperature internal resistance In this application, the input vector of the physical information neural network can be represented as formula (21): (21) In this way, based on the battery data samples, gradient information and battery physical information, the battery life (SOH) decay rate can be obtained by the trained physical information neural network for decay rate prediction. .

[0094] S504: Based on the battery life decay rate corresponding to each battery data sample, determine the rate decay loss, and based on the battery life decay rate corresponding to each battery data sample and the preset battery physical constraint condition, determine the decay physical loss.

[0095] Optionally, when determining the rate decay loss based on the battery life decay rate corresponding to each battery data sample, the sub-rate decay loss can be determined based on the change rate of the predicted life corresponding to each battery data sample and the battery life decay rate, and the rate decay loss can be determined based on each sub-rate decay loss.

[0096] That is, according to the change rate of the predicted life (the differential of the predicted life of the probability prediction model with respect to the current charge and discharge cycle ) and the battery life decay rate output by the physical information neural network , the sub-rate decay loss is calculated, and in this embodiment, the sub-rate decay loss is determined by the mean square error calculation method, and the rate decay loss is determined based on the sum of each sub-rate decay loss, as shown in formula (22): (22) wherein, is the rate decay loss, N is the total number of battery data samples.

[0097] In this application, by measuring and constraining the deviation between the battery life decay rate output by the physical information neural network and the change rate of the predicted life obtained by automatic differentiation, the consistency of the data can be improved, and thus the accuracy of the probability prediction model can be improved.

[0098] Regarding step S504, when determining the physical loss of degradation based on the battery life degradation rate, as mentioned above, the physical loss of degradation is determined based on the life degradation rate under preset battery physical constraints. The preset battery physical constraints include that the battery life is inversely proportional to the number of charge and discharge cycles, and the degradation rate of the predicted life is within the preset degradation rate range. That is, according to the battery physical characteristics, the battery life should decrease monotonically with the increase of the number of charge and discharge cycles, and its degradation rate is between the degradation rates corresponding to the two battery data samples in the battery data sample pair.

[0099] Therefore, when determining the physical loss due to degradation, it is necessary to calculate the loss for the predicted lifetime that violates the above-mentioned preset battery physical constraints, so as to guide the model to learn the physical law.

[0100] Specifically, when determining the physical loss due to battery degradation based on the battery life degradation rate, the following steps may be included: (I) For each battery data sample pair, based on the difference between the two predicted lifetimes, determine the first sub-physical loss, and based on the first sub-physical loss corresponding to each battery data sample pair, determine the first physical loss.

[0101] As mentioned above, when performing lifetime prediction on two battery data samples that include the same battery, two corresponding predicted lifetimes can be obtained. Thus, the first sub-physical loss can be determined based on the difference between the two predicted lifetimes.

[0102] Please refer to formulas (23)-(24) for the expression to determine the first physical loss: (twenty three) in, and For the two battery data samples in the battery data sample pair, as mentioned above, battery life should monotonically decrease with the increase of charge-discharge cycles. Therefore, the above conditions need to be met. This condition.

[0103] Furthermore, the portion violating the constraints shown in Equation (23) can be converted into the first physical loss using a preset activation function, such as the Rectified LinearUnit (ReLU) (Equation (24)). (twenty four) in, The first physical loss, M Indicates the number of batteries. Indicates the first j The total number of charge and discharge cycles for each battery.

[0104] (II) For any predicted life pair, the partial derivative of the two predicted lives is calculated respectively to obtain the predicted attenuation probability corresponding to the two predicted lives respectively, and a linear rectifier function is used to determine the second sub-physical loss based on the predicted attenuation probability corresponding to the two predicted lives respectively and the preset attenuation rate range, and the sum of each second sub-physical loss is determined as the second physical loss.

[0105] See formula (25) for the expression of the second physical loss: (25) Wherein, and constitute a preset attenuation rate range, is the difference between the predicted lives, is the difference between the charge and discharge cycles.

[0106] (III) The attenuation physical loss is determined based on the first physical loss and the second physical loss.

[0107] (26) Wherein, is the physical loss.

[0108] S505: Determine the total loss based on the probability difference loss, the rate attenuation loss and the attenuation physical loss, and adjust the model parameters of the probability prediction model based on the total loss. Repeat the above training steps until the preset training requirements are met to obtain the trained probability prediction model.

[0109] It can be understood that after the probability difference loss, the rate attenuation loss and the attenuation physical loss are calculated, the sum of the probability difference loss, the rate attenuation loss and the attenuation physical loss is determined as the total loss (27) Wherein, is the total loss.

[0110] Of course, in other embodiments, the weighted sum of the probability difference loss, the rate attenuation loss and the attenuation physical loss can also be determined as the total loss, which is not limited here.

[0111] After determining the total loss, the model parameters of the probability prediction model can be adjusted based on the total loss. Repeat the above training steps, that is, return to execute the step of determining the subset from the battery data sample set. The model parameters of the probability prediction model are adjusted using the back propagation algorithm and the gradient update strategy until the preset training requirements are met to obtain the trained probability prediction model.

[0112] The preset training requirement can include that the total number of training reaches a preset number, or the variation of the total loss in continuous multiple rounds of training is less than a preset threshold (i.e., no improvement in continuous multiple rounds).

[0113] According to the foregoing, the probability prediction model includes an encoder and a decoder, and therefore, adjusting the model parameters of the probability prediction model is adjusting the parameters of the encoder and the decoder respectively.

[0114] Please refer to Figure 9 A training schematic diagram of a probability prediction model provided for an exemplary embodiment of the present application is as shown in Figure 9 The training schematic diagram includes a front-stage network probability prediction model and a rear-stage network physical information neural network First, the battery data samples in each battery data sample pair in the subset are respectively input into the encoder of the probability prediction model, to obtain the probability distribution parameters of the latent variables. Here, since the battery data samples in the subset are multiple, multiple probability distribution parameters of the latent variables can be obtained , … , are the dimensions of the latent variables. Then, the latent variables are sampled according to the probability distribution parameters of the latent variables, to obtain multiple latent variable samples. Then, the latent variable samples are mapped in the life space by the decoder, to obtain the predicted life Then, based on the predicted life , the input vector of the physical information neural network is generated as The physical information neural network encodes and decodes the input vector, to output the predicted life decay rate. In this way, the probability difference loss, the rate decay loss and the decay physical loss can be determined, and the total loss can be determined for back propagation to adjust the parameters of the encoder and the decoder of the probability prediction model.

[0115] In the embodiments of the present application, the decay physical loss is determined based on the battery life decay rate predicted by the physical information neural network, and the model parameters are adjusted based on the probability density loss, the rate decay loss and the decay physical loss. In this way, the probability prediction model can take into account the physical law of the battery on the basis of data-driven training, so as to improve the accuracy and robustness of the probability prediction model.

[0116] ​Further, based on the application of the probability prediction model in the inference process, the accuracy of the battery life prediction can be improved, thereby improving the accuracy of the battery health state evaluation. In addition, the physical information neural network in the present application is trained specifically for different types of batteries, that is, for different types of batteries, a corresponding physical information neural network can be constructed according to the principle of battery attenuation, influencing factors, etc. of the battery, so that the specificity and accuracy of different types of batteries can be improved.

[0117] Corresponding to the embodiments of the battery life prediction method, the present application also provides embodiments of a battery life prediction device.

[0118] Please refer to Figure 10 A structural schematic diagram of a battery life prediction device according to an example embodiment of the present application is shown. As shown in Figure 10 The battery life prediction device 1000 comprises: The acquisition module 1010 is configured to acquire battery data of a target battery, and perform feature extraction on the battery data according to a preset dimension to obtain a battery feature vector. The prediction module 1020 is configured to input the battery feature vector into a trained probability prediction model to predict the battery life, and obtain a prediction life set of the target battery. The probability prediction model is trained based on a battery data sample set and a trained physical information neural network. The battery data sample set comprises a plurality of battery data sample pairs of batteries. Each battery data sample pair comprises battery data samples of each battery at different charging stages. The trained physical information neural network is used to predict the life attenuation of the battery life predicted by the probability prediction model during the training process, to obtain a life attenuation rate. The life attenuation rate is used to determine the physical loss of attenuation under a preset battery physical constraint condition, to adjust the parameters of the probability prediction model through back propagation. The determination module 1030 is configured to determine the life probability density corresponding to each prediction life in the prediction life set by using a probability density estimation method, and determine the prediction life of the target battery according to each life probability density.

[0119] Please refer to Figure 11 A structural schematic diagram of another battery life prediction device according to an example embodiment of the present application is shown. As shown in Figure 11 The battery life prediction device 1000 further comprises a training module 1040. The training module 1040 is configured to: Select a subset from the battery data sample set, and determine the probability distribution parameters of the latent variables corresponding to each battery data sample in each battery data sample pair for each battery data sample pair in the subset. determine a probability difference loss based on the true value of the life span of each battery data sample in the subset and the probability distribution parameter of the corresponding latent variable respectively; for each battery data sample, perform life span decay rate prediction on the battery data sample by using the trained physical information neural network to obtain a battery life span decay rate; determine a rate decay loss based on the battery life span decay rate corresponding to each battery data sample respectively, and determine the decay physical loss based on the battery life span decay rate corresponding to each battery data sample respectively and a preset battery physical constraint condition; determine a total loss based on the probability difference loss, the rate decay loss and the decay physical loss, and adjust the model parameters of the probability prediction model based on the total loss, and repeat the above training steps until a preset training requirement is met to obtain the trained probability prediction model.

[0120] In some embodiments, the probability prediction model includes an encoder and a decoder; and the training module 1040 is specifically configured to, when determining the probability distribution parameter of the latent variable corresponding to each battery data sample in each battery data sample pair in the subset: input each battery data sample in the battery data sample pair into the encoder to obtain the probability distribution parameter of the latent variable corresponding to the battery data sample pair; The training module 1040 is specifically configured to, when performing life span decay rate prediction on the battery data sample by using the trained physical information neural network to obtain a battery life span decay rate: obtain the gradient information corresponding to the battery data sample by using an automatic differentiation method to perform partial derivation on the battery data sample; obtain the battery physical information of the battery to which the battery data sample belongs, and perform decay rate prediction on the battery data sample, the gradient information and the battery physical information by using the trained physical information neural network to obtain the battery life span decay rate; The training module 1040 is specifically configured to, when adjusting the model parameters of the probability prediction model based on the total loss: adjust the parameters of the encoder and the decoder based on the total loss respectively.

[0121] In some embodiments, the training module 1040 is specifically configured to: determine the predicted life span corresponding to each battery data sample based on the probability distribution parameter of the latent variable corresponding to each battery data sample, and perform automatic differentiation on the predicted life span to obtain a predicted decay rate; For each battery data sample, determine a sub-rate attenuation loss based on a corresponding predicted attenuation rate and a corresponding battery life attenuation rate of the battery data sample; Determine the rate attenuation loss based on the sub-rate attenuation loss corresponding to each battery data sample.

[0122] In some embodiments, the preset life attenuation constraint condition includes that the battery life is inversely proportional to the number of charging and discharging times, and the attenuation rate of the predicted life is within a preset attenuation rate range; and the training module 1040 is specifically configured to: For each battery data sample, determine a first sub-physical loss based on the difference between the two predicted lives corresponding to the battery data sample, and determine a first physical loss based on the first sub-physical loss corresponding to each battery data sample; and For any predicted life pair, take the partial derivative of the two predicted lives respectively to obtain a predicted attenuation probability corresponding to the two predicted lives respectively, and determine a second sub-physical loss based on the predicted attenuation probability corresponding to the two predicted lives respectively and the preset attenuation rate range by using a linear rectifier function, and determine the sum of each second sub-physical loss as a second physical loss; Determine the attenuation physical loss based on the first physical loss and the second physical loss.

[0123] In some embodiments, the probability prediction model includes an encoder and a decoder; and the prediction module 1020 is specifically configured to: Encode the battery feature vector by using the encoder to obtain a probability distribution parameter of a latent variable corresponding to the battery feature vector; Sample the latent variable based on the probability distribution parameter to obtain a plurality of latent variable samples, and perform battery life mapping on the plurality of latent variable samples respectively by using the decoder to obtain the set of predicted lives.

[0124] In some embodiments, the acquisition module 1010 is specifically configured to: Acquire initial battery data of a time sequence state of a target battery in a charging process; Perform anomaly detection on the initial battery data to determine abnormal data, and adjust the abnormal data to obtain the battery data.

[0125] In some embodiments, the preset dimension includes at least one of the following: mean, standard deviation, skewness, kurtosis, charging duration, total charging amount, and change rate of battery data in the charging process.

[0126] The implementation process of the functions and roles of each unit in the above device is specifically described in the implementation process of the corresponding steps in the above method, which will not be repeated here.

[0127] For the device embodiment, since it basically corresponds to the method embodiment, the relevant part can be referred to the part of the method embodiment. The device embodiment described above is only illustrative, and the units described as separate components can or can not be physically separated, and the components shown as units can or can not be physical units, that is, they can be located in one place or distributed on multiple network units. Some or all of the modules can be selected to achieve the purpose of the present application according to actual needs. Those skilled in the art can understand and implement without creative labor.

[0128] Corresponding to the above-mentioned battery life prediction method, the embodiment of the disclosure also provides a computer device, as shown in Figure 12 The structure schematic diagram of the computer device provided by the embodiment of the disclosure is shown in Figure 12 The computer device 1200 includes a processor 1210, an internal bus 1220, a memory 1230, a network interface 1240, and a non-volatile memory 1250, and of course, it can also include other hardware required by functions. One or more embodiments of the present specification can be implemented in a software manner, such as reading the corresponding computer program from the non-volatile memory 1250 into the memory 1230 by the processor 1210 and then running. Of course, in addition to the software implementation, one or more embodiments of the present specification do not exclude other implementation manners, such as logic devices or a combination of software and hardware, and the like, that is, the execution subject of the following processing flow is not limited to each logic unit, but can also be hardware or logic devices.

[0129] The memory 1230 is also called an internal memory, which is used to temporarily store operation data in the processor 1210 and data exchanged with the non-volatile memory 1250 such as a hard disk. The processor 1210 exchanges data with the non-volatile memory 1250 through the memory 1230.

[0130] In the embodiment of the present application, the memory 1230 is specifically used to store the application program code for executing the present application scheme, and is controlled to execute by the processor 1210. That is, when the computer device is running, the processor 1210 communicates with the network interface 1240, the memory 1230, and the non-volatile memory 1250 through the internal bus 1220 respectively, so that the processor 1210 executes the application program code stored in the memory 1230 and the non-volatile memory 1250, and further executes the battery life prediction method described in the above method embodiment.

[0131] The processor 1210 can be an integrated circuit chip having a processing capability for signals. The processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; or can be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic device, a discrete gate or transistor logic device, a discrete hardware microservice. The disclosed methods, steps and logic block diagrams in the embodiments of the present application can be implemented or executed. The general-purpose processor can be a microprocessor or the processor can be any conventional processor.

[0132] It can be understood that the structure illustrated in the embodiments of the present application does not constitute a specific limitation on the computer device 1200. In other embodiments of the present application, the computer device 1200 can include more or fewer components than the illustrated components, or combine certain components, or split certain components, or different component arrangements. The illustrated components can be implemented in hardware, software or a combination of software and hardware.

[0133] The embodiments of the present disclosure also provide a computer readable storage medium, which stores a computer program. When the computer program is run by a processor, the steps of the battery life prediction method in the above method embodiments are executed. The storage medium can be a volatile or non-volatile computer readable storage medium.

[0134] The embodiments of the present disclosure also provide a computer program product carrying a program code. The instructions included in the program code can be used to execute the steps of the battery life prediction method in the above method embodiments. For details, please refer to the above method embodiments, which will not be repeated here.

[0135] The computer program product can be specifically implemented by hardware, software or a combination thereof. In one optional embodiment, the computer program product is specifically embodied as a computer storage medium. In another optional embodiment, the computer program product is specifically embodied as a software product, such as a software development kit (SDK) and the like.

[0136] Embodiments of the subject matter and the functional operations described in this specification can be implemented in digital electronic circuitry, in tangibly-embodied computer software or firmware, in computer hardware, including the structures disclosed in this specification and their structural equivalents, or in combinations of one or more of them. Embodiments of the subject matter described in this specification can be implemented as one or more computer programs, i.e., one or more modules of computer program instructions encoded on a tangible non-transitory program carrier for execution by, or to control the operation of, data processing apparatus. Alternatively or additionally, the program instructions can be encoded on an artificially generated propagated signal, e.g., a machine-generated electrical, optical, or electromagnetic signal that is generated to encode information for transmission to suitable receiver apparatus for execution by a data processing apparatus. The computer storage medium can be a machine-readable storage device, a machine-readable storage substrate, a random or serial access memory device, or a combination of one or more of them.

[0137] The processes and logic flows described in this specification can be performed by one or more programmable computers executing one or more computer programs to perform functions by operating on input data and generating output. The processes and logic flows can also be performed by special purpose logic circuitry, e.g., an FPGA (field programmable gate array) or an ASIC (application specific integrated circuit), and the apparatus can be implemented as special purpose logic circuitry.

[0138] Computers suitable for the execution of a computer program include, by way of example, general and / or special purpose microprocessors, or any other kind of central processing unit. Generally, a central processing unit will receive instructions and data from a read-only memory and / or a random access memory. The essential elements of a computer are a central processing unit for performing instructions and one or more memory devices for storing instructions and data. Generally, a computer will also include, or be operatively coupled to receive data from or transfer data to, or both, one or more mass storage devices for storing data, e.g., magnetic, magneto-optical disks, or optical disks. However, a computer need not have such devices. Moreover, a computer can be embedded in another device, e.g., a mobile telephone, a personal digital assistant (PDA), a mobile audio or video player, a game console, a Global Positioning System (GPS) receiver, or a portable storage device (e.g., a universal serial bus (USB) flash drive), to name just a few.

[0139] Computer readable media suitable for storing computer program instructions and data include all forms of non-volatile memory, media and memory devices, including by way of example semiconductor memory devices, e.g., EPROM, EEPROM, and flash memory devices; magnetic disks, e.g., internal hard disks or removable disks; magneto-optical disks; and CD-ROM and DVD-ROM disks. The processor and the memory can be supplemented by, or incorporated in, special purpose logic circuitry.

[0140] While the specification contains many specifics, these should not be construed as limiting the scope of any invention or of what can be claimed, but as merely providing illustrations of some of the embodiments of the inventions. Certain features that are described in this specification in the context of separate embodiments can also be implemented in combination in a single embodiment. Conversely, various features that are described in the context of a single embodiment can also be implemented in multiple embodiments separately or in any suitable subcombination. Moreover, although features can be described above as acting in certain combinations and even initially claimed as such, one or more features from a claimed combination can in some cases be excised from the combination and the claimed combination can be directed to a subcombination or variation of a subcombination.

[0141] Similarly, while operations are depicted in the drawings in a particular order, this should not be understood as requiring or implying that such operations be performed in the particular order shown or in sequential order, or that all illustrated operations be performed, to achieve desirable results. In certain circumstances, multitasking and parallel processing can be advantageous. Moreover, the separation of various system modules and services in the embodiments described above should not be understood as requiring such separation in all embodiments, and it should be understood that the described program services and systems can generally be integrated in a single software product or packaged into multiple software products.

[0142] Accordingly, particular embodiments of the subject matter have been described. Other embodiments are within the scope of the following claims. In some cases, actions recited in the claims can be performed in a different order and still achieve desirable results. In addition, the processes depicted in the accompanying figures do not necessarily require the particular order shown, or sequential order, to achieve the desired results. In certain implementations, multitasking and parallel processing can be advantageous.

[0143] The above description is merely illustrative of the application and not restrictive. Since certain changes can be made in the application as described without departing from the spirit and scope of the application, it is intended that all of the possible modifications be included within the scope of the application.

Claims

1. A method of predicting battery life, characterized by, The method comprises the following steps: obtaining battery data of a target battery, and extracting features of the battery data according to a preset dimension to obtain a battery feature vector; inputting the battery feature vector into a trained probability prediction model to predict the battery life, and obtaining a set of predicted life of the target battery; the probability prediction model is trained based on a set of battery data samples and a trained physical information neural network, the set of battery data samples comprises a plurality of pairs of battery data samples of batteries, each pair of battery data samples comprises battery data samples of each battery at different charging stages; the trained physical information neural network is used to predict the life attenuation of the battery life predicted by the probability prediction model during training, to obtain a life attenuation rate, and the life attenuation rate is used to determine the attenuation physical loss under a preset battery physical constraint condition, to adjust the parameters of the probability prediction model through back propagation; using a probability density estimation method to determine the life probability density corresponding to each predicted life in the set of predicted life, and determining the predicted life of the target battery according to each life probability density.

2. The method of claim 1, wherein, The trained probability prediction model is trained by the following steps: selecting a subset from the set of battery data samples, and determining the probability distribution parameters of the hidden variables corresponding to each battery data sample in the subset for each pair of battery data samples in the subset; determining the probability difference loss based on the life true value of each battery data sample in the subset and the probability distribution parameters of the corresponding hidden variables; for each battery data sample, predicting the attenuation rate through the trained physical information neural network based on the battery data sample to obtain the battery life attenuation rate; based on the battery life attenuation rate corresponding to each battery data sample, determining the rate attenuation loss, and based on the battery life attenuation rate corresponding to each battery data sample and the preset battery physical constraint condition, determining the attenuation physical loss; determining the total loss based on the probability difference loss, the rate attenuation loss and the attenuation physical loss, adjusting the model parameters of the probability prediction model based on the total loss, repeating the above training steps until the preset training requirements are met, and obtaining the trained probability prediction model.

3. The method of claim 2, wherein, The probability prediction model comprises an encoder and a decoder; the step of determining the probability distribution parameters of the hidden variables corresponding to each battery data sample in the subset for each pair of battery data samples in the subset comprises: for each pair of battery data samples in the subset, inputting each battery data sample in the pair of battery data samples into the encoder to obtain the probability distribution parameters of the hidden variables corresponding to the pair of battery data samples; predicting the attenuation rate through the trained physical information neural network based on the battery data sample to obtain the battery life attenuation rate, comprising: using an automatic differentiation method to calculate the partial derivative of the battery data sample to obtain the gradient information corresponding to the battery data sample; obtain battery physical information of a battery to which the battery data sample belongs, and perform, based on the battery data sample, the gradient information, and the battery physical information, battery life decay rate prediction through the trained physical information neural network to obtain the battery life decay rate; adjust model parameters of the probability prediction model based on the total loss, including: adjust parameters of the encoder and the decoder respectively based on the total loss.

4. The method of claim 2, wherein, determine the rate decay loss based on the battery life decay rate corresponding to each battery data sample, including: determine the predicted life corresponding to each battery data sample based on the probability distribution parameters of the latent variable corresponding to each battery data sample, and automatically differentiate the predicted life to obtain the predicted decay rate; for each battery data sample, determine a sub-rate decay loss based on the predicted decay rate corresponding to the battery data sample and the battery life decay rate corresponding to the battery data sample; determine the rate decay loss based on the sub-rate decay loss corresponding to each battery data sample respectively.

5. The method of claim 2, wherein, The preset life decay constraint condition includes that the battery life is inversely proportional to the number of charge and discharge times, and the decay rate of the predicted life is within a preset decay rate range; and the decay physical loss is determined based on the battery life decay rate corresponding to each battery data sample and the preset battery physical constraint condition, including: for each battery data sample, determine a first sub-physical loss based on the difference between the two predicted lives corresponding to the battery data sample, and determine a first physical loss based on the first sub-physical loss corresponding to each battery data sample; and for any predicted life pair, take the partial derivative of the two predicted lives respectively to obtain the predicted decay probability corresponding to the two predicted lives respectively, and determine a second sub-physical loss based on the predicted decay probability corresponding to the two predicted lives respectively and the preset decay rate range by using a linear rectifier function, and determine the second physical loss as the sum of each second sub-physical loss; determine the decay physical loss based on the first physical loss and the second physical loss.

6. The method of claim 1, wherein, The probability prediction model includes an encoder and a decoder; and the battery life prediction of the battery feature vector input into the trained probability prediction model includes: use the encoder to encode the battery feature vector to obtain the probability distribution parameters of the latent variable corresponding to the battery feature vector; based on the probability distribution parameters, sample the latent variable to obtain a plurality of latent variable samples, and use the decoder to respectively perform battery life mapping on the plurality of latent variable samples to obtain the predicted life set.

7. The method of claim 1, wherein, The battery data of the target battery includes: obtain initial battery data of the target battery in a timing state during a charging process; perform anomaly detection on the initial battery data to determine abnormal data, and adjust the abnormal data to obtain the battery data.

8. The method of claim 1, wherein, The preset dimensions include at least one of the following: mean value, standard deviation, skewness, kurtosis, charging duration, total charging amount, and rate of change of battery data during charging.

9. A battery life prediction device, characterized by, Comprise: An acquisition module, configured to acquire battery data of a target battery, and perform feature extraction on the battery data according to preset dimensions to obtain a battery feature vector; A prediction module, configured to input the battery feature vector into a trained probability prediction model to perform battery life prediction and obtain a predicted life set of the target battery; the probability prediction model is trained based on a battery data sample set and a trained physical information neural network; the battery data sample set includes a plurality of battery data sample pairs of batteries; each battery data sample pair includes battery data samples of each battery at different charging stages; the trained physical information neural network is used to perform life attenuation prediction on the battery life predicted by the probability prediction model during training to obtain a life attenuation rate; the life attenuation rate is used to determine an attenuation physical loss under a preset battery physical constraint condition to adjust parameters of the probability prediction model through back propagation; A determination module, configured to determine a life probability density corresponding to each predicted life in the predicted life set by using a probability density estimation method, and determine a predicted life of the target battery according to the life probability densities.

10. A computer device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor executes the program to implement the steps of the battery life prediction method of any one of claims 1-8.

11. A computer readable storage medium having stored thereon a computer program, characterized in that The program is executed by the processor to implement the steps of the battery life prediction method of any one of claims 1-8. The program is executed by the processor to implement the steps of the battery life prediction method of any one of claims 1-8.

Images