A dual-model prediction method

Abstract

This application provides a dual-model prediction method, comprising: obtaining the input feature vectors of the lithium battery at k time points before the sampling time in the k-th discharge cycle, respectively inputting them into a first SOC sub-model and a first SOH sub-model, and outputting the predicted SOC mean, predicted SOC variance, predicted SOH mean, and predicted SOH variance at the k subsequent time points; further determining a first total prediction error at the k subsequent time points, including the prediction error of the first SOC sub-model and the prediction error of the first SOH sub-model; updating the parameters of the first SOC sub-model and the first SOH sub-model respectively by an optimizer based on the first total prediction error to obtain a second SOC sub-model and a second SOH sub-model; if the prediction error of the second SOC sub-model is greater than the prediction error of the first SOC sub-model, then the first SOC sub-model is used as the trained SOC sub-model; if the prediction error of the second SOH sub-model is greater than the prediction error of the first SOH sub-model, then the first SOH sub-model is used as the trained SOH sub-model.

Description

Technical Field

[0001] This application relates to the field of battery life prediction technology, and in particular to a dual-model prediction method. Background Technology

[0002] Lithium-ion batteries are widely used in electric vehicles, energy storage systems, portable electronic devices, and industrial backup power supplies due to their high energy density, long cycle life, and low self-discharge rate. A Battery Management System (BMS) continuously estimates the battery's State of Charge (SOC) and State of Health (SOH) to support range estimation, charge / discharge power limiting, safety protection, lifespan prediction, and maintenance decisions.

[0003] Existing methods often model SOC and SOH separately, or use a single neural network to output both SOC and SOH simultaneously. When modeling separately, it is easy to overlook the coulometric coupling between the dynamics of SOC and the capacity decay of SOH. When modeling multiple tasks with a single model, SOC is a rapidly changing variable at the sampling level, while SOH is a slowly changing variable at the cyclic level. The time scales of the two types of tasks are significantly different, which can easily lead to the problem that the fast variable dominates the shared features of the task, while the generalization ability of the slow variable SOH decreases. Summary of the Invention

[0004] To address the aforementioned technical problems, this application provides a dual-model prediction method, which... The inputs are respectively fed into the first SOC sub-model and the first SOH sub-model, and the outputs are... After The predicted mean SOC, predicted variance SOC, predicted mean SOH, and predicted variance SOH at each time point are used to determine the... After The first total prediction error, including the prediction errors of the first SOC sub-model and the first SOH sub-model, is used by the optimizer to update the parameters of the first SOC sub-model and the first SOH sub-model based on the first total prediction error, respectively, to obtain the second SOC sub-model and the second SOH sub-model. If the prediction error of the second SOC sub-model is greater than the prediction error of the first SOC sub-model, then the first SOC sub-model is used as the trained SOC sub-model; if the prediction error of the second SOH sub-model is greater than the prediction error of the first SOH sub-model, then the first SOH sub-model is used as the trained SOH sub-model. The prediction error of the first SOC sub-model is determined by... After any time in time The predicted average SOC and lithium battery in After any time in time The reconstructed SOC tag value was determined, and the lithium battery was in After any time in time The reconstructed SOC tag value is consistent with that of the lithium battery. After Time in time The reconstructed SOH label values ​​exhibit a coupling relationship; the first SOC sub-model and the first SOH sub-model are mutually coupled. By utilizing the mutual coupling relationship between the first SOC sub-model and the first SOH sub-model, the SOC sub-model trained in this application significantly improves the prediction accuracy of the SOC value of lithium batteries, and the SOH sub-model trained in this application significantly improves the prediction accuracy of the SOH value of lithium batteries.

[0005] Firstly, this application provides a dual-model prediction method, including:

[0006] Obtaining lithium batteries in the first Sampling time during the next discharge cycle Previous Input feature vector at time 1 , Indicates the sampling time The input feature vector, where, , , , , , and These represent the first cell in the [number]th [year]. Under the condition of the next discharge cycle, the first The sampling time includes terminal voltage, current, temperature, normalization time, normalization cycle number, and operating condition segment code. It is a positive number. It is a positive integer;

[0007] Will get Previous Input feature vector at time 1 The input is given to the first SOC sub-model, causing the first SOC sub-model to output... After The predicted SOC mean and predicted SOC variance at each time point; and the obtained Previous Input feature vector at time 1 The input is given to the first SOH sub-model, causing the first SOH sub-model to output... After The mean and variance of predicted SOH at each time point;

[0008] According to the continuous The mean and variance of the predicted SOC at each time point, and the continuous The predicted mean and variance of SOH at each time point are used to determine the... After The first total prediction error, including the prediction error of the first SOC sub-model and the prediction error of the first SOH sub-model, is backpropagated. The gradients of the first total prediction error with respect to the parameters of the first SOC sub-model and the first total prediction error with respect to the parameters of the first SOH sub-model are calculated. The optimizer updates the parameters of the first SOC sub-model based on the gradients of the first total prediction error with respect to the parameters of the first SOC sub-model, obtaining a second SOC sub-model with the parameters of the second SOC sub-model. Similarly, the optimizer updates the parameters of the first SOH sub-model based on the gradients of the first total prediction error with respect to the parameters of the first SOC sub-model, obtaining a second SOH sub-model with the parameters of the second SOH sub-model. The prediction error of the first SOC sub-model is obtained through... After any time in time The predicted average SOC and lithium battery in After any time in time The reconstructed SOC tag value was determined, and the lithium battery was in After any time in time The reconstructed SOC tag value is consistent with that of the lithium battery. After Time in time The reconstructed SOH label values ​​exhibit coupling; the first SOC sub-model and the first SOH sub-model are mutually coupled. , All are positive; the lithium battery at the sampling time Previous At that moment and the lithium battery After At any given moment, the lithium battery is in its first... During the next discharge cycle;

[0009] The prediction errors of the second SOC sub-model and the second SOH sub-model are determined. If the prediction error of the second SOC sub-model is greater than the prediction error of the first SOC sub-model, then the first SOC sub-model is used as the trained SOC sub-model. If the prediction error of the second SOH sub-model is greater than the prediction error of the first SOH sub-model, then the first SOH sub-model is used as the trained SOH sub-model. The second SOC sub-model and the second SOH sub-model are coupled together.

[0010] In conjunction with the first aspect, in one alternative implementation,

[0011] After determining the prediction errors of the second SOC sub-model and the second SOH sub-model, the following steps are also included:

[0012] After determining the prediction errors of the second SOC sub-model and the second SOH sub-model, the following steps are also included:

[0013] If the prediction error of the second SOC sub-model is less than the prediction error of the first SOC sub-model, and the absolute value of the difference between the prediction error of the second SOC sub-model and the prediction error of the first SOC sub-model is less than the first error threshold, then the second SOC sub-model is used as a trained SOC sub-model; if the prediction error of the second SOH sub-model is less than the prediction error of the first SOH sub-model, and the absolute value of the difference between the prediction error of the second SOH sub-model and the prediction error of the first SOH sub-model is less than the second error threshold, then the second SOH sub-model is used as a trained SOH sub-model.

[0014] In conjunction with the first aspect, in one alternative implementation,

[0015] Will get Previous Input feature vector at time 1 The input is given to the first SOC sub-model, causing the first SOC sub-model to output... After The predicted SOC mean and predicted SOC variance at each time point, specifically including:

[0016] Will get Previous Input feature vector at time 1 Input into the first SOC sub-model, and obtain according to equations (1) to (9) After any time in time Predicted SOC mean Predicting SOC variance , ;

[0017] (1)

[0018] (2)

[0019] (3)

[0020] (4)

[0021] (5)

[0022] (6)

[0023] (7)

[0024] (8)

[0025] (9)

[0026] in, It is a positive number. This represents the prediction log-variance of the first SOC sub-model. Used to guarantee SOC variance It is a positive number. This represents the nonlinear transformation in the prediction head. and Both represent the variance prediction head parameters of the first SOC sub-model. and Both represent the mean prediction head parameters of the first SOC sub-model. express After any time in time The predicted SOC mean, This represents the global context information obtained by averaging all historical hidden states in the first SOC sub-model. express After any time in time Contextual information, , , These represent the query matrix, key matrix, and value mapping matrix of the cross-attention module, respectively. This indicates the first SOC sub-model set in the first SOC sub-model. A learnable future query vector This represents the timing encoder in the first SOC sub-model. Encoded output data, This represents the initial hidden state sequence of the first SOC sub-model. The model parameters are: The first SOC sub-model parameters, include: , , and , The dimension is The input linear mapping matrix of the first SOC sub-model, The dimension is Learnable positional encoding of the first SOC sub-model Indicates length is A column vector of all 1s. express The transpose of .

[0027] In conjunction with the first aspect, in one alternative implementation,

[0028] Will get Previous Input feature vector at time 1 The input is given to the first SOH sub-model, causing the first SOH sub-model to output... After The predicted mean and variance of SOH at each time point specifically include:

[0029] Will get Previous Input feature vector at time 1 Input into the first SOH sub-model, and obtain according to equations (10) to (18) After any time in time Predicted mean SOH Predicting SOC variance , ;

[0030] (10)

[0031] (11)

[0032] (12)

[0033] (13)

[0034] (14)

[0035] (15)

[0036] (16)

[0037] (17)

[0038] (18)

[0039] in, It is a positive number. Indicates the first SOH sub-model in The predicted log-variance at time t. Used to ensure SOH variance It is a positive number. This represents the nonlinear transformation in the prediction head. and These are all variance prediction head parameters of the first SOH sub-model. express The predicted mean SOH at time 10:00. and These are all mean prediction head parameters of the first SOH sub-model. This represents the global context information obtained by averaging all historical hidden states in the first SOH sub-model. This represents the timing encoder in the first SOH sub-model. Encoded output data, This represents the initial hidden state sequence of the first SOH sub-model. The model parameters are: The first SOH sub-model parameters, include: , , and , The dimension is The input linear mapping matrix of the first SOH sub-model, The dimension is Learnable position encoding of the first SOH sub-model Indicates length is A column vector of all 1s. express The transpose of .

[0040] In conjunction with the first aspect, in one alternative implementation,

[0041] According to the continuous The mean and variance of the predicted SOC at each time point, and the continuous The predicted mean and variance of SOH at each time point are used to determine the... After The first total prediction error at each time step includes the prediction error of the first SOC sub-model and the prediction error of the first SOH sub-model, specifically including:

[0042] According to the continuous The mean and variance of the predicted SOC at each time point, and the continuous The mean and variance of the predicted SOH at each time point are used to determine the value of the predicted SOH using equations (19)-(33). After The first total prediction error at each time step includes the prediction errors of the first SOC sub-model and the first SOH sub-model. ,

[0043] (19)

[0044] (20)

[0045] (twenty one)

[0046] (twenty two)

[0047] (twenty three)

[0048] (twenty four)

[0049] (25)

[0050] (26)

[0051] (27)

[0052] (28)

[0053] (29)

[0054] (30)

[0055] (31)

[0056] (32)

[0057] (33)

[0058] in, Indicates including and All trainable parameters, , , , , , Each loss item , , , , , The weight, They are positive numbers, Indicates that the lithium battery in the first The actual usable capacity measured during 1C discharge under one discharge cycle condition. This indicates the rated capacity of the lithium battery. Indicates that the lithium battery in the first Timing under the next discharge cycle condition After any time in time The reconstructed SOC label values, Indicates that the lithium battery in the first Under the condition of the next discharge cycle, the first The SOC tag value reconstructed at time step, Indicates that the lithium battery in the first Timing under the next discharge cycle condition The next The SOC tag value reconstructed at time step, Indicates that the lithium battery in the first Under the condition of the next discharge cycle, the first The reconstructed SOH label value at time 10:00. Represents coulomb measurement, such that When smaller, Decrease The rate of decline increased; Indicates that the lithium battery in the first Under the condition of the next discharge cycle, from the discharge start time At the time The cumulative discharge capacity, Indicates that the lithium battery in the first Under the condition of the next discharge cycle, from the discharge start time At the time The cumulative discharge capacity, Indicates that the lithium battery in the first Under the condition of the next discharge cycle, from the discharge start time At the time The cumulative discharge capacity, Indicates time The current, measured in amperes. This represents the time interval between adjacent sampling moments, in seconds. The unit is ampere-hour. It is a positive number.

[0059] In conjunction with the first aspect, in one alternative implementation,

[0060] After determining the prediction errors of the second SOC sub-model and the second SOH sub-model, the following steps are also included:

[0061] If the prediction error of the second SOC sub-model is less than the prediction error of the first SOC sub-model, and the absolute value of the difference between the prediction errors of the second SOC sub-model and the first SOC sub-model is greater than the first error threshold, a second total prediction error is calculated based on the prediction errors of the second SOC sub-model and the second SOH sub-model. This second total prediction error is used to update the parameters of the second SOC sub-model; or...

[0062] If the prediction error of the second SOH sub-model is less than the prediction error of the first SOH sub-model, and the absolute value of the difference between the prediction error of the second SOH sub-model and the prediction error of the first SOH sub-model is greater than the second error threshold, the second total prediction error is calculated based on the prediction error of the second SOC sub-model and the prediction error of the second SOH sub-model. The second total prediction error is used to update the parameters of the second SOH sub-model.

[0063] In conjunction with the first aspect, in one alternative implementation,

[0064] Based on equation (20), the prediction error of the first SOC sub-model pass After any time in time Predicted SOC mean and lithium batteries After any time in time The reconstructed SOC tag value The determination is made, and the lithium battery is explained by equations (34)-(35). After any time in time The reconstructed SOC tag value With lithium batteries After Time in time The reconstructed SOH tag values ​​exhibit a coupling relationship.

[0065] (34)

[0066] (35)

[0067] in, Indicates that lithium batteries are After any time in time The reconstructed SOC tag value, Lithium batteries in After any time in time The reconstructed SOH tag value, Represents coulomb measurement, Indicates in After Time in time The current, measured in amperes. express Subsequent adjacent sampling times and The time interval between them is in seconds. The unit is ampere-hour. Indicates process noise. It is a positive number.

[0068] In conjunction with the first aspect, in one alternative implementation,

[0069] Prediction error of the first SOH sub-model pass Predicted mean SOH at time and lithium batteries SOH label value reconstructed at time step The determination is made, and the lithium battery is explained by equation (36). SOH label value reconstructed at time step With lithium batteries SOC tag value reconstructed at time step There is a coupling relationship.

[0070] (36)

[0071] in, Represents coulomb measurement, Indicates time The current, measured in amperes. Indicates time adjacent sampling times and The time interval between them is in seconds. The unit is ampere-hour. Indicates process noise. It is a positive number.

[0072] This application provides a dual-model prediction method, including: obtaining the lithium battery's prediction in the 1st... Sampling time during the next discharge cycle Previous Input feature vector at time 1 , Indicates the sampling time The input feature vector, where, , , , , , and These represent the first cell in the [number]th [year]. Under the condition of the next discharge cycle, the first The sampling time includes terminal voltage, current, temperature, normalization time, normalization cycle number, and operating condition segment code. It is a positive number. It is a positive integer;

[0073] Will get Previous Input feature vector at time 1 The input is given to the first SOC sub-model, causing the first SOC sub-model to output... After The predicted SOC mean and predicted SOC variance at each time point; and the obtained Previous Input feature vector at time 1 The input is given to the first SOH sub-model, causing the first SOH sub-model to output... After The mean and variance of predicted SOH at each time point;

[0074] According to the continuous The mean and variance of the predicted SOC at each time point, and the continuous The predicted mean and variance of SOH at each time point are used to determine the... After The first total prediction error, including the prediction error of the first SOC sub-model and the prediction error of the first SOH sub-model, is backpropagated. The gradients of the first total prediction error with respect to the parameters of the first SOC sub-model and the first total prediction error with respect to the parameters of the first SOH sub-model are calculated. The optimizer updates the parameters of the first SOC sub-model based on the gradients of the first total prediction error with respect to the parameters of the first SOC sub-model, obtaining a second SOC sub-model with the parameters of the second SOC sub-model. Similarly, the optimizer updates the parameters of the first SOH sub-model based on the gradients of the first total prediction error with respect to the parameters of the first SOC sub-model, obtaining a second SOH sub-model with the parameters of the second SOH sub-model. The prediction error of the first SOC sub-model is obtained through... After any time in time The predicted average SOC and lithium battery in After any time in time The reconstructed SOC tag value was determined, and the lithium battery was in After any time in time The reconstructed SOC tag value is consistent with that of the lithium battery. After Time in time The reconstructed SOH label values ​​exhibit coupling; the first SOC sub-model and the first SOH sub-model are mutually coupled. , All are positive; the lithium battery at the sampling time Previous At that moment and the lithium battery After At any given moment, the lithium battery is in its first... During the next discharge cycle;

[0075] The prediction errors of the second SOC sub-model and the second SOH sub-model are determined. If the prediction error of the second SOC sub-model is greater than the prediction error of the first SOC sub-model, then the first SOC sub-model is used as the trained SOC sub-model. If the prediction error of the second SOH sub-model is greater than the prediction error of the first SOH sub-model, then the first SOH sub-model is used as the trained SOH sub-model. The second SOC sub-model and the second SOH sub-model are coupled together.

[0076] In this application, the prediction error of the first SOC sub-model is determined by... After any time in time The predicted average SOC and lithium battery in After any time in time The reconstructed SOC tag value was determined, and the lithium battery was in After any time in time The reconstructed SOC tag value is consistent with that of the lithium battery. After Time in time The reconstructed SOH label values ​​exhibit a coupling relationship; the first SOC sub-model and the first SOH sub-model are mutually coupled. By utilizing the mutual coupling relationship between the first SOC sub-model and the first SOH sub-model, the SOC sub-model trained in this application significantly improves the prediction accuracy of the SOC value of lithium batteries, and the SOH sub-model trained in this application significantly improves the prediction accuracy of the SOH value of lithium batteries. Attached Figure Description

[0077] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0078] Figure 1 This is a flowchart illustrating a dual-model prediction method provided in this application. Detailed Implementation

[0079] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0080] It should be noted that the terms "first" and "second" in this application are only used to distinguish different SOC sub-models, SOH sub-models, total prediction error, error threshold, etc., and have no other special meaning, and should not limit the scope of protection of this application.

[0081] Figure 1 This is a flowchart illustrating a dual-model prediction method.

[0082] This can be combined with the appendix in this application. Figure 1 The dual-model prediction method provided in this application may include, but is not limited to, the following steps:

[0083] S101, Obtaining the lithium battery in the first... Sampling time during the next discharge cycle Previous Input feature vector at time 1 .

[0084] In this application, Indicates the sampling time The input feature vector, where, , , , , , and These represent the first cell in the [number]th [year]. Under the condition of the next discharge cycle, the first The sampling time includes terminal voltage, current, temperature, normalization time, normalization cycle number, and operating condition segment code. It is a positive number. It is a positive integer; express The dimension is , The value of is a real number. Optionally, It can be 120.

[0085] S102, the obtained Previous Input feature vector at time 1 The input is given to the first SOC sub-model, causing the first SOC sub-model to output... After The predicted SOC mean and predicted SOC variance at each time point; and the obtained Previous Input feature vector at time 1 The input is given to the first SOH sub-model, causing the first SOH sub-model to output... After The mean and variance of the predicted SOH at each time point.

[0086] In this application, the obtained Previous Input feature vector at time 1 The input is given to the first SOC (State of Charge) sub-model, causing the first SOC sub-model to output... After The predicted SOC mean and predicted SOC variance at each time point, specifically including but not limited to:

[0087] Will get Previous Input feature vector at time 1 Input into the first SOC sub-model, and obtain according to equations (1) to (9) After any time in time Predicted SOC mean Predicting SOC variance , ;

[0088] (1)

[0089] (2)

[0090] (3)

[0091] (4)

[0092] (5)

[0093] (6)

[0094] (7)

[0095] (8)

[0096] (9)

[0097] in, It is a positive number. This represents the prediction log-variance of the first SOC sub-model. Used to guarantee SOC variance It is a positive number. This represents the nonlinear transformation in the prediction head. and Both represent the variance prediction head parameters of the first SOC sub-model. and Both represent the mean prediction head parameters of the first SOC sub-model. express After any time in time The predicted SOC mean, This represents the global context information obtained by averaging all historical hidden states in the first SOC sub-model. express After any time in time Contextual information, , , These represent the query matrix, key matrix, and value mapping matrix of the cross-attention module, respectively. This indicates the first SOC sub-model set in the first SOC sub-model. A learnable future query vector This represents the timing encoder in the first SOC sub-model. Encoded output data, This represents the initial hidden state sequence of the first SOC sub-model. The model parameters are: The first SOC sub-model parameters, include: , , and , The dimension is The input linear mapping matrix of the first SOC sub-model, The dimension is Learnable positional encoding of the first SOC sub-model Indicates length is A column vector of all 1s. express The transpose of, where, It is a positive integer. , , , Both can be positive numbers.

[0098] Optionally, It can be 30.

[0099] It should be noted that the same input Inputting different first SOC sub-models and first SOH (State of Health) sub-models allows for the acquisition of different information, achieved through a parameter-unshared time encoder, a parameter-unshared mean prediction head, a variance prediction head, and different model prediction losses. The first SOC sub-model focuses on learning the current integral, short-time changes in terminal voltage, the SOC decreasing trend, and local window dynamics (e.g., ...). Input feature vector at time 1 The first SOH sub-model focuses on learning cycle numbering (e.g., discharge cycle). (Voltage plateau changes, capacity decay, and long-term aging trends)

[0100] It should be noted that the obtained Previous Input feature vector at time 1 The input is given to the first SOH sub-model, causing the first SOH sub-model to output... After The predicted mean and variance of SOH at each time point, which may include, but are not limited to:

[0101] Will get Previous Input feature vector at time 1 Input into the first SOH sub-model, and obtain according to equations (10) to (18) After any time in time Predicted mean SOH Predicting SOC variance , ;

[0102] (10)

[0103] (11)

[0104] (12)

[0105] (13)

[0106] (14)

[0107] (15)

[0108] (16)

[0109] (17)

[0110] (18)

[0111] in, It is a positive number. Indicates the first SOH sub-model in The predicted log-variance at time t. Used to ensure SOH variance It is a positive number. This represents the nonlinear transformation in the prediction head. and These are all variance prediction head parameters of the first SOH sub-model. express The predicted mean SOH at time 10:00. and These are all mean prediction head parameters of the first SOH sub-model. This represents the global context information obtained by averaging all historical hidden states in the first SOH sub-model. This represents the timing encoder in the first SOH sub-model. Encoded output data, This represents the initial hidden state sequence of the first SOH sub-model. The model parameters are: The first SOH sub-model parameters, include: , , and , The dimension is The input linear mapping matrix of the first SOH sub-model, The dimension is Learnable position encoding of the first SOH sub-model Indicates length is A column vector of all 1s. express The transpose of .

[0112] S103, according to continuous The mean and variance of the predicted SOC at each time point, and the continuous The predicted mean and variance of SOH at each time point are used to determine the... After The first total prediction error, including the prediction errors of the first SOC sub-model and the first SOH sub-model, is backpropagated. The gradients of the first total prediction error with respect to the parameters of the first SOC sub-model and the first total prediction error with respect to the parameters of the first SOH sub-model are calculated. The first SOC sub-model parameters are updated by the optimizer based on the gradients of the first total prediction error with respect to the parameters of the first SOC sub-model, resulting in a second SOC sub-model with the parameters of the second SOC sub-model. The first SOH sub-model parameters are updated by the optimizer based on the gradients of the first total prediction error with respect to the parameters of the first SOC sub-model, resulting in a second SOH sub-model with the parameters of the second SOH sub-model.

[0113] In this application, the gradient of the first total prediction error with respect to the parameters of the first SOC sub-model of the first SOC sub-model and the gradient of the first total prediction error with respect to the parameters of the first SOH sub-model of the first SOH sub-model are calculated. That is, the derivative of the first total prediction error with respect to the parameters of the first SOC sub-model of the first SOC sub-model and the derivative of the first total prediction error with respect to the parameters of the first SOH sub-model of the first SOH sub-model are calculated.

[0114] The optimizer in this application may include, but is not limited to, the AdamW optimizer. It should be noted that updating the model parameters based on the gradient of the model parameters is a conventional technique, which will not be described in detail in this application.

[0115] In this application, based on continuity The mean and variance of the predicted SOC at each time point, and the continuous The predicted mean and variance of SOH at each time point are used to determine the... After The first total prediction error at each time step includes the prediction error of the first SOC sub-model and the prediction error of the first SOH sub-model, and may specifically include, but is not limited to:

[0116] According to the continuous The mean and variance of the predicted SOC at each time point, and the continuous The mean and variance of the predicted SOH at each time point are used to determine the value of the predicted SOH using equations (19)-(33). After The first total prediction error at each time step includes the prediction errors of the first SOC sub-model and the first SOH sub-model. ,

[0117] (19)

[0118] (20)

[0119] (twenty one)

[0120] (twenty two)

[0121] (twenty three)

[0122] (twenty four)

[0123] (25)

[0124] (26)

[0125] (27)

[0126] (28)

[0127] (29)

[0128] (30)

[0129] (31)

[0130] (32)

[0131] (33)

[0132] in, Indicates including and All trainable parameters, , , , , , Each loss item , , , , , The weight, They are positive numbers, Indicates that the lithium battery in the first The actual usable capacity measured during 1C discharge under one discharge cycle condition. This indicates the rated capacity of the lithium battery. Indicates that the lithium battery in the first Timing under the next discharge cycle condition After any time in time The reconstructed SOC label values, Indicates that the lithium battery in the first Under the condition of the next discharge cycle, the first The SOC tag value reconstructed at time step, Indicates that the lithium battery in the first Timing under the next discharge cycle condition The next The SOC tag value reconstructed at time step, Indicates that the lithium battery in the first Under the condition of the next discharge cycle, the first The reconstructed SOH label value at time 10:00. Represents coulomb measurement, such that When smaller, Decrease The rate of decline increased; Indicates that the lithium battery in the first Under the condition of the next discharge cycle, from the discharge start time At the time The cumulative discharge capacity, Indicates that the lithium battery in the first Under the condition of the next discharge cycle, from the discharge start time At the time The cumulative discharge capacity, Indicates that the lithium battery in the first Under the condition of the next discharge cycle, from the discharge start time At the time The cumulative discharge capacity, Indicates time The current, measured in amperes. This represents the time interval between adjacent sampling moments, in seconds. The unit is ampere-hour. It is a positive number.

[0133] It should be noted that, This can be expressed as SOC heteroscedasticity loss. This can be expressed as SOH heteroscedasticity loss. This can be expressed as the loss of physical consistency (e.g., the mean square error between the reconstructed SOC label values ​​and the predicted SOC mean). It can represent the error at the end of the SOC.

[0134] Among them, the first SOC sub-model is mainly driven by SOC heteroscedasticity loss, SOC terminal error and coulomb consistency constraints; the first SOH sub-model is mainly driven by SOH heteroscedasticity loss, smoothing constraints and uncertainty constraints.

[0135] It should be noted that the first SOC sub-model and the first SOH sub-model can be coupled together through physical consistency loss.

[0136] From equation (26), it can be deduced that lithium batteries in After any time in time The reconstructed SOC tag value With lithium batteries The predicted SOC mean at each time point is coupled, which also reflects the mutual coupling between the first SOC sub-model and the first SOH sub-model. When the parameters of the first SOC sub-model and the first SOH sub-model are updated, the second total prediction error is calculated based on (26) in the same way. Again, the mutual coupling between the second SOC sub-model and the second SOH sub-model can be reflected by equation (26). By utilizing the mutual coupling between the first SOC sub-model and the first SOH sub-model, the SOC sub-model trained in this application greatly improves the prediction accuracy of the SOC value of lithium batteries, and the SOH sub-model trained in this application greatly improves the prediction accuracy of the SOH value of lithium batteries.

[0137] In this application, the prediction error of the first SOC sub-model is determined by... After any time in time The predicted average SOC and lithium battery in After any time in time The reconstructed SOC tag value was determined, and the lithium battery was in After any time in time The reconstructed SOC tag value is consistent with that of the lithium battery. After Time in time The reconstructed SOH label values ​​exhibit coupling; the first SOC sub-model and the first SOH sub-model are mutually coupled. , All are positive; the lithium battery at the sampling time Previous At that moment and the lithium battery After At any given moment, the lithium battery is in its first... During the next discharge cycle;

[0138] Specifically, based on equation (20), the prediction error of the first SOC sub-model pass After any time in time Predicted SOC mean and lithium batteries After any time in time The reconstructed SOC tag value The determination is made, and the lithium battery is explained by equations (34)-(35). After any time in time The reconstructed SOC tag value With lithium batteries After Time in time The reconstructed SOH label values ​​are coupled to achieve mutual coupling between the first SOC sub-model and the first SOH sub-model;

[0139] (34)

[0140] (35)

[0141] in, Indicates that lithium batteries are After any time in time The reconstructed SOC tag value, Lithium batteries in After any time in time The reconstructed SOH tag value, Represents coulomb measurement, Indicates in After Time in time The current, measured in amperes. express Subsequent adjacent sampling times and The time interval between them is in seconds. The unit is ampere-hour. This represents process noise and can be used to characterize temperature, polarization, measurement errors, and model simplification errors. It is a positive number. Specifically, the coupling between the first SOC sub-model and the first SOH sub-model can be achieved through equation (34), which can be specifically reflected in equation (34). , There is a related coupling relationship. By utilizing the mutual coupling relationship between the first SOC sub-model and the first SOH sub-model, the SOC sub-model trained in this application greatly improves the prediction accuracy of the SOC value of lithium batteries, and the SOH sub-model trained in this application greatly improves the prediction accuracy of the SOH value of lithium batteries.

[0142] Specifically, the prediction error of the first SOH sub-model pass Predicted mean SOH at time and lithium batteries SOH label value reconstructed at time step The determination is made, and the lithium battery is explained by equation (36). SOH label value reconstructed at time step With lithium batteries SOC tag value reconstructed at time step There is a coupling relationship to achieve mutual coupling between the first SOC sub-model and the first SOH sub-model;

[0143] (36)

[0144] in, Represents coulomb measurement, Indicates time The current, measured in amperes. Indicates time adjacent sampling times and The time interval between them is in seconds. The unit is ampere-hour. This represents process noise and can be used to characterize temperature, polarization, measurement errors, and model simplification errors. It is a positive number.

[0145] Specifically, the first SOC sub-model and the first SOH sub-model are coupled through equation (36), which can be specifically reflected in equation (36). and The coupling relationship between them.

[0146] S104. Determine the prediction error of the second SOC sub-model and the prediction error of the second SOH sub-model. If the prediction error of the second SOC sub-model is greater than the prediction error of the first SOC sub-model, then the first SOC sub-model is used as the trained SOC sub-model. If the prediction error of the second SOH sub-model is greater than the prediction error of the first SOH sub-model, then the first SOH sub-model is used as the trained SOH sub-model.

[0147] In this application, before determining the prediction error of the second SOC sub-model and the prediction error of the second SOH sub-model, the specific steps may include, but are not limited to:

[0148] The aforementioned obtained Previous Input feature vector at time 1 The input is given to the second SOC sub-model, causing the second SOC sub-model to output... After The mean and variance of the target predicted SOC at each time point; and the aforementioned values ​​can be used to obtain Previous Input feature vector at time 1 The input is given to the second SOH sub-model, causing the second SOH sub-model to output... After Mean and variance of the target predicted SOH at each time point;

[0149] According to the continuous The mean and variance of the target predicted SOC at each time point, and the continuous The mean and variance of the target predicted SOH at each time point are determined. After The prediction errors at each time point include the prediction errors of the second SOC sub-model and the second SOH sub-model.

[0150] In this application, after determining the prediction error of the second SOC sub-model and the prediction error of the second SOH sub-model, the following may also be included, but are not limited to:

[0151] If the prediction error of the second SOC sub-model is less than the prediction error of the first SOC sub-model, and the absolute value of the difference between the prediction errors of the second SOC sub-model and the first SOC sub-model is less than the first error threshold, it indicates that the prediction error of the second SOC sub-model has converged. In this case, the second SOC sub-model is used as a trained SOC sub-model, achieving early stopping of SOC sub-model training and avoiding waste of resources. If the prediction error of the second SOH sub-model is less than the prediction error of the first SOH sub-model, and the absolute value of the difference between the prediction errors of the second SOH sub-model and the first SOH sub-model is less than the second error threshold, it indicates that the prediction error of the second SOH sub-model has converged. In this case, the second SOH sub-model is used as a trained SOH sub-model, achieving early stopping of SOH sub-model training and avoiding waste of resources.

[0152] The first error threshold and the second error threshold may be equal or unequal.

[0153] It should be noted that after determining the prediction error of the second SOC sub-model and the prediction error of the second SOH sub-model, the following may also be included, but are not limited to:

[0154] If the prediction error of the second SOC sub-model is less than the prediction error of the first SOC sub-model, and the absolute value of the difference between the prediction errors of the second SOC sub-model and the first SOC sub-model is greater than the first error threshold, a second total prediction error is calculated based on the prediction errors of the second SOC sub-model and the second SOH sub-model. This second total prediction error is used to update the parameters of the second SOC sub-model; or...

[0155] If the prediction error of the second SOH sub-model is less than the prediction error of the first SOH sub-model, and the absolute value of the difference between the prediction error of the second SOH sub-model and the prediction error of the first SOH sub-model is greater than the second error threshold, the second total prediction error is calculated based on the prediction error of the second SOC sub-model and the prediction error of the second SOH sub-model. The second total prediction error is used to update the parameters of the second SOH sub-model.

[0156] Figure 1 This is only used to illustrate the embodiments of this application and should not be construed as limiting the scope of protection of this application.

[0157] Those skilled in the art will recognize that the method steps of the various examples described in conjunction with the embodiments disclosed in this application can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of each example have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this application.

[0158] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process of the devices and equipment described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0159] In the several embodiments provided in this application, it should be understood that the methods can be implemented in other ways. For example, the composition and steps of each example have been described. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0160] The embodiments described above are merely illustrative. The mutual coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection through some interface or module, or it may be an electrical, mechanical or other form of connection.

[0161] Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0162] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A dual-model prediction method, characterized in that, include: Obtaining lithium batteries in the first Sampling time during the next discharge cycle Previous Input feature vector at time 1 , Indicates the sampling time The input feature vector, where, , , , , , and These represent the first cell in the [number]th [year]. Under the condition of the next discharge cycle, the first The sampling time includes terminal voltage, current, temperature, normalization time, normalization cycle number, and operating condition segment code. It is a positive number. It is a positive integer; Will get Previous Input feature vector at time 1 The input is given to the first SOC sub-model, causing the first SOC sub-model to output... After The predicted SOC mean and predicted SOC variance at each time point; and the obtained Previous Input feature vector at time 1 The input is given to the first SOH sub-model, causing the first SOH sub-model to output... After The mean and variance of predicted SOH at each time point; According to the continuous The mean and variance of the predicted SOC at each time point, and the continuous The predicted mean and predicted variance of SOH at each time point are used to determine the... After The first total prediction error, including the prediction error of the first SOC sub-model and the prediction error of the first SOH sub-model, is backpropagated. The gradients of the first total prediction error with respect to the parameters of the first SOC sub-model and the first total prediction error with respect to the parameters of the first SOH sub-model are calculated. The optimizer updates the parameters of the first SOC sub-model based on the gradients of the first total prediction error with respect to the parameters of the first SOC sub-model, obtaining a second SOC sub-model with the parameters of the second SOC sub-model. Similarly, the optimizer updates the parameters of the first SOH sub-model based on the gradients of the first total prediction error with respect to the parameters of the first SOC sub-model, obtaining a second SOH sub-model with the parameters of the second SOH sub-model. The prediction error of the first SOC sub-model is obtained through... After any time in time The predicted average SOC and lithium battery in After any time in time The reconstructed SOC tag value was determined, and the lithium battery was in After any time in time The reconstructed SOC tag value is consistent with that of the lithium battery. After Time in time The reconstructed SOH label values ​​are coupled to achieve mutual coupling between the first SOC sub-model and the first SOH sub-model. , All are positive; the lithium battery at the sampling time Previous At that moment and the lithium battery After At any given moment, the lithium battery is in its first... During the next discharge cycle; The prediction errors of the second SOC sub-model and the second SOH sub-model are determined. If the prediction error of the second SOC sub-model is greater than the prediction error of the first SOC sub-model, then the first SOC sub-model is used as the trained SOC sub-model. If the prediction error of the second SOH sub-model is greater than the prediction error of the first SOH sub-model, then the first SOH sub-model is used as the trained SOH sub-model. The second SOC sub-model and the second SOH sub-model are coupled together.

2. The dual-model prediction method as described in claim 1, characterized in that, After determining the prediction errors of the second SOC sub-model and the second SOH sub-model, the following steps are also included: If the prediction error of the second SOC sub-model is less than the prediction error of the first SOC sub-model, and the absolute value of the difference between the prediction error of the second SOC sub-model and the prediction error of the first SOC sub-model is less than the first error threshold, then the second SOC sub-model is used as a trained SOC sub-model; if the prediction error of the second SOH sub-model is less than the prediction error of the first SOH sub-model, and the absolute value of the difference between the prediction error of the second SOH sub-model and the prediction error of the first SOH sub-model is less than the second error threshold, then the second SOH sub-model is used as a trained SOH sub-model.

3. The dual-model prediction method as described in claim 1, characterized in that, Will get Previous Input feature vector at time 1 The input is given to the first SOC sub-model, causing the first SOC sub-model to output... After The predicted SOC mean and predicted SOC variance at each time point, specifically including: Will get Previous Input feature vector at time 1 Input into the first SOC sub-model, and obtain according to equations (1) to (9) After any time in time Predicted SOC mean Predicting SOC variance , ; (1) (2) (3) (4) (5) (6) (7) (8) (9) in, It is a positive number. This represents the prediction log-variance of the first SOC sub-model. Used to guarantee SOC variance It is a positive number. This represents the nonlinear transformation in the prediction head. and Both represent the variance prediction head parameters of the first SOC sub-model. and Both represent the mean prediction head parameters of the first SOC sub-model. express After any time in time The predicted SOC mean, This represents the global context information obtained by averaging all historical hidden states in the first SOC sub-model. express After any time in time Contextual information, , , These represent the query matrix, key matrix, and value mapping matrix of the cross-attention module, respectively. This indicates the first SOC sub-model set in the first SOC sub-model. A learnable future query vector This represents the timing encoder in the first SOC sub-model. Encoded output data, This represents the initial hidden state sequence of the first SOC sub-model. The model parameters are: The first SOC sub-model parameters, include: , , and , The dimension is The input linear mapping matrix of the first SOC sub-model, The dimension is Learnable positional encoding of the first SOC sub-model Indicates length is A column vector of all 1s. express The transpose of .

4. The dual-model prediction method as described in claim 3, characterized in that, Will get Previous Input feature vector at time 1 The input is given to the first SOH sub-model, causing the first SOH sub-model to output... After The predicted mean and variance of SOH at each time point specifically include: Will get Previous Input feature vector at time 1 Input into the first SOH sub-model, and obtain according to equations (10) to (18) After any time in time Predicted mean SOH Predicting SOC variance , ; (10) (11) (12) (13) (14) (15) (16) (17) (18) in, It is a positive number. Indicates the first SOH sub-model in The predicted log-variance at time t. Used to ensure SOH variance It is a positive number. This represents the nonlinear transformation in the prediction head. and These are all variance prediction head parameters of the first SOH sub-model. express The predicted mean SOH at time 10:

00. and These are all mean prediction head parameters of the first SOH sub-model. This represents the global context information obtained by averaging all historical hidden states in the first SOH sub-model. This represents the timing encoder in the first SOH sub-model. Encoded output data, This represents the initial hidden state sequence of the first SOH sub-model. The model parameters are: The first SOH sub-model parameters, include: , , and , The dimension is The input linear mapping matrix of the first SOH sub-model, The dimension is Learnable position encoding of the first SOH sub-model Indicates length is A column vector of all 1s. express The transpose of .

5. The dual-model prediction method as described in claim 4, characterized in that, According to the continuous The mean and variance of the predicted SOC at each time point, and the continuous The predicted mean and predicted variance of SOH at each time point are used to determine the... After The first total prediction error at each time step includes the prediction error of the first SOC sub-model and the prediction error of the first SOH sub-model, specifically including: According to the continuous The mean and variance of the predicted SOC at each time point, and the continuous The mean and variance of the predicted SOH at each time point are used to determine the value of the predicted SOH using equations (19)-(33). After The first total prediction error at each time step includes the prediction errors of the first SOC sub-model and the first SOH sub-model. , （19） （20） （21） （22） （23） （24） （25） （26） （27） （28） （29） (30) (31) (32) (33) in, Indicates including and All trainable parameters, , , , , , Each loss item , , , , , The weight, They are positive numbers, Indicates that the lithium battery in the first The actual usable capacity measured during 1C discharge under one discharge cycle condition. This indicates the rated capacity of the lithium battery. Indicates that the lithium battery in the first Timing under the next discharge cycle condition After any time in time The reconstructed SOC label values, Indicates that the lithium battery in the first Under the condition of the next discharge cycle, the first The SOC tag value reconstructed at time step, Indicates that the lithium battery in the first Timing under the next discharge cycle condition The next The SOC tag value reconstructed at time step, Indicates that the lithium battery in the first Under the condition of the next discharge cycle, the first The reconstructed SOH label value at time 10:

00. Represents coulomb measurement, such that When smaller, Decrease The rate of decline increased; Indicates that the lithium battery in the first Under the condition of the next discharge cycle, from the discharge start time At the time The cumulative discharge capacity, Indicates that the lithium battery in the first Under the condition of the next discharge cycle, from the discharge start time At the time The cumulative discharge capacity, Indicates that the lithium battery in the first Under the condition of the next discharge cycle, from the discharge start time At the time The cumulative discharge capacity, Indicates time The current, measured in amperes. This represents the time interval between adjacent sampling moments, in seconds. The unit is ampere-hour. It is a positive number. This is expressed as SOC heteroscedasticity loss. This is expressed as SOH heteroscedasticity loss. Represented as physical consistency loss, This indicates the error at the end of the SOC (State of Charge).

6. The dual-model prediction method as described in claim 2, characterized in that, After determining the prediction errors of the second SOC sub-model and the second SOH sub-model, the following steps are also included: If the prediction error of the second SOC sub-model is less than the prediction error of the first SOC sub-model, and the absolute value of the difference between the prediction errors of the second SOC sub-model and the first SOC sub-model is greater than the first error threshold, a second total prediction error is calculated based on the prediction errors of the second SOC sub-model and the second SOH sub-model. This second total prediction error is used to update the parameters of the second SOC sub-model; or... If the prediction error of the second SOH sub-model is less than the prediction error of the first SOH sub-model, and the absolute value of the difference between the prediction error of the second SOH sub-model and the prediction error of the first SOH sub-model is greater than the second error threshold, the second total prediction error is calculated based on the prediction error of the second SOC sub-model and the prediction error of the second SOH sub-model. The second total prediction error is used to update the parameters of the second SOH sub-model.

7. The dual-model prediction method as described in claim 5, characterized in that, Based on equation (20), the prediction error of the first SOC sub-model pass After any time in time Predicted SOC mean and lithium batteries After any time in time The reconstructed SOC tag value The determination is made, and the lithium battery is explained by equations (34)-(35). After any time in time The reconstructed SOC tag value With lithium batteries After Time in time The reconstructed SOH tag values ​​exhibit a coupling relationship. （34） （35） in, Indicates that lithium batteries are After any time in time The reconstructed SOC tag value, Lithium batteries in After any time in time The reconstructed SOH tag value, Represents coulomb measurement, Indicates in After Time in time The current, measured in amperes. express Subsequent adjacent sampling times and The time interval between them is in seconds. The unit is ampere-hour. Indicates process noise. It is a positive number.

8. The dual-model prediction method as described in claim 5, characterized in that, Prediction error of the first SOH sub-model pass Predicted mean SOH at time and lithium batteries SOH label value reconstructed at time step The determination is made, and the lithium battery is explained by equation (36). SOH label value reconstructed at time step With lithium batteries SOC tag value reconstructed at time step There is a coupling relationship. （36） in, Represents coulomb measurement, Indicates time The current, measured in amperes. Indicates time adjacent sampling times and The time interval between them is in seconds. The unit is ampere-hour. Indicates process noise. It is a positive number.

Images