A lithium battery residual capacity estimation method, device, equipment and medium

Abstract

The application discloses a lithium battery residual capacity estimation method, and belongs to the technical field of batteries. The method comprises the following steps: preprocessing open-source lithium battery discharge data to obtain source domain data samples with time sequence characteristics; constructing a global feature encoder based on a gating recurrent unit and a neural ordinary differential equation; processing the source domain data samples through the global feature encoder to obtain a global feature time sequence; training the neural ordinary differential equation through the global feature time sequence to obtain a source domain model; receiving constant-current discharge process data of a lithium battery and preprocessing the constant-current discharge process data to obtain target domain data samples with time sequence characteristics; training the source domain model through the target domain data samples to obtain a target domain model; and estimating the residual capacity of the lithium battery through the target domain model. The method has both high precision of N-ODE and high calculation efficiency of transfer learning.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of batteries, in particular to a lithium battery residual capacity estimation method, device, equipment and medium. BACKGROUND

[0002] In recent years, the state has begun to change the energy supply structure of our country, and vigorously develops new energy to alleviate the energy crisis and environmental pollution problems. The continuous development and utilization of new energy and its energy storage devices lay the foundation for the development of electric vehicles. On the one hand, clean electric energy from new energy power generation provides sufficient energy for the development of electric vehicles. On the other hand, the continuous introduction of various types of energy storage devices realizes the on-board mobile energy supply. The continuous maturity of the above two aspects provides the necessary conditions for the energy transformation of the automobile industry. Guided by energy saving and emission reduction, the automobile industry has begun to transform, and energy-saving and new energy vehicles represented by electric vehicles have been listed as one of China's strategic emerging industries in the future.

[0003] With the strong support of national policy, a part of excellent new energy automobile enterprises begin to emerge. China's new energy automobile enterprises have entered the accelerated development stage from the early development stage. Various types of new energy vehicles are continuously developed and quickly pushed to the market. Traditional fuel buses are gradually replaced by new energy vehicles, and part of passenger cars and commercial vehicles are also replaced by new energy vehicles. However, there are still many problems to be solved in the use process of electric vehicles, and the overcharge and overdischarge of the battery still exist. Overcharge or overdischarge will not only cause the battery to heat seriously, but also damage the battery life, and even may cause the battery to have thermal runaway, cause personal safety and property loss and other problems. Lithium ion battery as the power source of electric vehicles, in the actual charging and discharging process, the working environment is complex, and the battery will have irreversible degradation. When the performance of the battery is degraded to the point that it cannot meet the requirements of the power, reliability and safety of the automobile in use, it will cause the performance of the automobile to decline, and the above problems are directly related to the battery management system.

[0004] The battery management system is the "brain" of the power battery system, which provides the necessary conditions for the normal operation of the battery module. Because the accurate SOC (state of charge) estimation has important significance in preventing overcharge or overdischarge, improving battery energy utilization, ensuring the safety and stability of the electric vehicle battery system, and providing the necessary conditions for subsequent optimization of vehicle energy distribution; accurate capacity estimation is the premise of predicting the remaining driving range of the electric vehicle, which can avoid the phenomenon that the battery cannot meet the energy and power requirements of the electric vehicle driving, and has important significance for ensuring the safe driving of the electric vehicle and the safety of the life and property of the driver. At present, the problems of battery overcharge and discharge and range anxiety still exist, and the core reason is that the accuracy of the battery SOC and capacity is not high, especially in the late stage of battery discharge, the accuracy of the current estimation method will have a great fluctuation. Therefore, it is necessary to further explore the main factors affecting the accuracy of SOC and capacity estimation, and to put forward the corresponding improvement method, so as to improve the accuracy of battery SOC estimation. SUMMARY

[0005] In view of the problem of inaccurate lithium battery SOC estimation in the prior art, the purpose of the present application is to provide a lithium battery remaining capacity estimation method, device, equipment and medium, so as to at least partially solve the above problems.

[0006] To achieve the above purpose, the technical scheme of the present application is:

[0007] In the first aspect, the present application provides a lithium battery remaining capacity estimation method, which comprises the following steps:

[0008] Pretreatment is performed on the open source lithium battery discharge data to obtain source domain data samples with time sequence characteristics;

[0009] A global feature encoder is constructed based on a gated recurrent unit and a neural ordinary differential equation;

[0010] The source domain data samples are processed by the global feature encoder to obtain a global feature time sequence;

[0011] A neural ordinary differential equation is trained by the global feature time sequence to obtain a source domain model;

[0012] The constant current discharge process data of the lithium battery is received and pretreated to obtain target domain data samples with time sequence characteristics;

[0013] The source domain model is trained by the target domain data samples, and the source domain model is parameter fine-tuned and network layer frozen, and the inference is performed by forward propagation to obtain a target domain model;

[0014] After inputting the discharge process data of the lithium battery to be estimated into the target domain model, the remaining capacity of the lithium battery to be estimated is obtained.

[0015] In a preferred embodiment, the open-source lithium battery discharge data and the constant current discharge process data of the lithium battery both include discharge time, discharge voltage, discharge current, full charge capacity of the battery, and current remaining charge of the battery; the discharge process data of the lithium battery to be estimated includes discharge time, discharge voltage, discharge current, and full charge capacity of the battery.

[0016] In a preferred embodiment, the step of preprocessing the open-source lithium battery discharge data includes screening and removing abnormal data; the step of preprocessing the constant current discharge process data of the lithium battery includes sparsification processing.

[0017] In a preferred embodiment, the calculation process of the global feature encoder is as follows:

[0018] Equation (1)

[0019] Equation (2)

[0020] in, This represents the hyperparameters in the network structure. Indicates input, This is the battery's discharge voltage. Let f be the discharge current of the battery, m≤t≤M, where m and M are the maximum and minimum values ​​of time, respectively, and f represents the expression of the constructed general neural differential equation.

[0021] right The first step is to perform the operation where the input sequence is encoded using GRU to obtain the initial embedding vector of the time series process, i.e., equation (3):

[0022] (Equation 3)

[0023] in, and These are the update gate and the reset gate at the time step. The output, and The outputs of the hidden layer and the temporary layer are respectively. For time step Input voltage, and It is an activation function. and These are weight and threshold, respectively.

[0024] In a preferred embodiment, the source domain model is obtained through the following steps:

[0025] The embedding vector is fed into a neural network ordinary differential equation to obtain a continuous embedding vector, then is determined as:

[0026] Equation (4)

[0027] Equation (5)

[0028] Here, and are given a new sequence , ; k represents a newly introduced time sequence, and the original time sequence t is changed to , n represents the maximum value of k, and N represents the maximum value; the new sequence refers to the time sequence obtained after the global feature encoder processes the source domain data sample, and k pairs of time sequences are further divided on this basis;

[0029] Three new parameters , and are introduced, and are used for secondary calculation (equation 3) of the output and , to obtain:

[0030] Equation (6)

[0031] wherein, is the calculation step, is the weight parameter of the secondary calculation, is the amount of a single sample for network propagation, and the amount of samples for forward and backward propagation is the same; y ^ represents the process variable of the secondary calculation;

[0032] Then the continuous embedding vector uses a variational autoencoder to restore the initial sequence, and the remaining capacity SOC N-ODE (t k ) of the lithium battery is output according to equation (1) and equation (4) as:

[0033] Equation (7)

[0034] Equation (8)

[0035] Equation (9)

[0036] Equation (10)

[0037] wherein, is the full charge capacity of the battery, is the coulomb efficiency and is approximately 1.

[0038] In a preferred embodiment, the parameter fine-tuning and network layer freezing of the source domain model specifically comprises: replacing the Adam optimizer of the source domain model with a Decay-Adam optimizer, increasing the initial learning rate by 10 times, and freezing the hidden layers of the source domain model.

[0039] In a preferred embodiment, the difference between the estimated value and the true value of the target domain model output is minimized by a root mean square error loss function, which is:

[0040] Formula (11);

[0041] wherein L represents the loss function of network training, SOC * (t k ) represents the experimental value of the battery capacity.

[0042] In a second aspect, the present application also provides a battery remaining capacity estimation device, comprising:

[0043] An acquisition module is configured to acquire open-source lithium battery discharge data, lithium battery constant-current discharge process data, and discharge process data of a lithium battery to be estimated.

[0044] A preprocessing module is configured to preprocess the open-source lithium battery discharge data and the lithium battery constant-current discharge process data.

[0045] A global feature encoder construction module is configured to construct a global feature encoder based on a gated recurrent unit and a neural ordinary differential equation.

[0046] A global feature encoder calling module is configured to process source domain data samples by calling the function of the global feature encoder, thereby obtaining a global feature time series.

[0047] A source domain model training module is configured to train a neural ordinary differential equation through the global feature time series, thereby obtaining a source domain model.

[0048] A target domain model training module is configured to train the source domain model through target domain data samples, fine-tune the parameters of the source domain model, freeze the network layers of the source domain model, and perform inference through forward propagation, thereby obtaining a target domain model.

[0049] An estimation module is configured to input the discharge process data of the lithium battery to be estimated into the target domain model, thereby obtaining the remaining capacity of the lithium battery to be estimated.

[0050] In a third aspect, the present application also provides an electronic device, comprising a memory storing executable program codes and a processor coupled with the memory; wherein the processor invokes the executable program codes stored in the memory to execute the method as described above.

[0051] In a fourth aspect, the present application also provides a computer readable storage medium storing a computer program, wherein the computer program is executed by a processor to perform the method as described above.

[0052] With the above technical solution, the present application has the following advantages:

[0053] 1. By using neural ordinary differential equation (N-ODE) to predict the lithium battery capacity, the fine learning of the entire discharge process is realized, the problem of increasing estimation error in the later stage of the discharge of the traditional lithium battery estimation model is solved, and the capacity estimation accuracy of the entire discharge process is improved;

[0054] 2. By using the transfer learning method, the problem of high memory occupation and high computing resource requirement caused by the complex calculation process of neural ordinary differential equation is solved. The neural ordinary differential equation has a wider application scenario;

[0055] 3. By combining transfer learning and neural ordinary differential equation, not only the problem of large error in the later stage of battery capacity estimation is solved, but also the problem of low calculation efficiency that needs to be overcome in the traditional deep learning-based estimation method is solved, which is more valuable for industrial application. BRIEF DESCRIPTION OF DRAWINGS

[0056] Figure 1 The method flowchart of the first embodiment of the present application is shown in the figure;

[0057] Figure 2 The structural schematic diagram of the second embodiment of the present application is shown in the figure;

[0058] Figure 3 The structural schematic diagram of the third embodiment of the present application is shown in the figure;

[0059] Figure 4 The discharge process estimation error diagram of the present application is shown in the figure;

[0060] Figure 5 The error value at different stages of the discharge of the present application is shown in the figure. DETAILED DESCRIPTION

[0061] The specific embodiments of the present application will be further described below in combination with the drawings. It should be noted that the description of these embodiments is used to help understand the present application, but does not constitute a limitation on the present application. In addition, the technical features involved in the various embodiments of the present application described below can be combined with each other as long as they do not conflict with each other.

[0062] Embodiment one

[0063] A lithium battery residual capacity estimation method is used to accurately estimate the residual capacity of a lithium battery, as shown in Figure 1 The method comprises the following steps:

[0064] S1, pre-processing the open source lithium battery discharge data to obtain source domain data samples with time sequence characteristics.

[0065] The open source lithium battery discharge data includes discharge time, discharge voltage, discharge current, full charge capacity of the battery, current residual capacity of the battery, etc. The pre-processing of the open source lithium battery discharge data specifically includes screening and removing abnormal data.

[0066] S2, constructing a global feature encoder based on a gated recurrent unit (GRU) and a neural ordinary differential equation (N-ODE).

[0067] The calculation process of the global feature encoder is as follows:

[0068] Formula (1)

[0069] Formula (2)

[0070] Wherein, represents the hyperparameters in the network structure, represents the input, is the discharge voltage of the battery, is the discharge current of the battery, m≤t≤M, m, M are the maximum and minimum values of time respectively, and f represents the general neural differential equation expression constructed;

[0071] The first step operation is performed on , wherein the input sequence is encoded by GRU to obtain the initial embedding vector of the time sequence process, i.e. formula (3):

[0072] Formula 3

[0073] Wherein, and are the outputs of the update gate and the reset gate at time step , and are the outputs of the hidden layer and the temporary layer, is the input at time step , and are activation functions, and are weights and thresholds, respectively.

[0074] S3, processing the source domain data sample through a global feature encoder to obtain a global feature time sequence.

[0075] S4, training a neural ordinary differential equation through the global feature time sequence to obtain a source domain model, and saving in an interpreter.

[0076] In this embodiment, the source domain model is obtained through the following steps:

[0077] The embedding vector is fed into the neural network ordinary differential equation to obtain a continuous embedding vector, and the in formula (1) is determined as:

[0078] Formula (4)

[0079] Formula (5)

[0080] Here, and are given new sequences , , k represents a newly introduced time sequence, and the original time sequence t is changed to , n represents the maximum value of k, and N represents the maximum value of ; the new sequence refers to the time sequence obtained after the global feature encoder processes the source domain data sample, and k is further added to the time sequence on this basis.

[0081] Three new parameters , and are introduced again, and are used for secondary calculation (formula 3) of the outputs and , to obtain:

[0082] Formula (6)

[0083] wherein, is the calculation step, is the weight parameter of the secondary calculation, is the amount of a single sample for network propagation, and the amount of samples for forward and backward propagation is the same, and y ^ represents the process variable of the secondary calculation.

[0084] Then the continuous embedding vector uses the variational autoencoder to restore the initial sequence, and the remaining capacity SOC N-ODE of the lithium battery k is output according to formula (1) and formula (4) as:

[0085] ​ Formula (7)

[0086] Formula (8)

[0087] Formula (9)

[0088] Formula (10)

[0089] wherein, is the full charge capacity of the battery, is the coulomb efficiency and is approximately 1.

[0090] S5, receiving the constant current discharge process data of the lithium battery and preprocessing to obtain the target domain data sample with time sequence characteristics.

[0091] In this embodiment, the constant current discharge process data of different lithium batteries is received, which includes discharge time, discharge voltage, discharge current, full charge capacity of the battery, and current remaining capacity of the battery. In addition to the above-mentioned screening and rejection of abnormal data, the preprocessing of the constant current discharge process data of the lithium battery also includes sparse processing.

[0092] S6, training the source domain model obtained in S4 with the target domain data sample, parameter fine-tuning and network layer freezing of the source domain model, and inferring through forward propagation to obtain the target domain model.

[0093] In this embodiment, the parameter fine-tuning and network layer freezing of the source domain model specifically include: replacing the Adam optimizer of the source domain model with a Decay-Adam optimizer, increasing the initial learning rate by 10 times, and freezing the hidden layer of the source domain model.

[0094] In a preferred embodiment, the difference between the estimated value output by the target domain model and the true value is minimized by continuing to use the root mean square error loss function, and the root mean square error loss function is:

[0095] Formula (11)

[0096] wherein, L represents the loss function of network training, SOC * (t k ) represents the experimental value of the battery capacity.

[0097] Correspondingly, the target domain data sample is usually divided into training samples and test samples, wherein the training samples are used to train the target domain model, and the test samples are input into the target domain model, and the estimated value output by the target domain model is input into the root mean square error loss function for calculation.

[0098] S7, inputting the discharge process data of the lithium battery to be estimated into the target domain model, obtaining the residual capacity of the lithium battery to be estimated.

[0099] In this embodiment, the discharge process data of the lithium battery to be estimated includes discharge time, discharge voltage, discharge current and full charge capacity of the battery.

[0100] In summary, the embodiment of the present application utilizes the similarity of different battery discharge processes, reduces the calculation amount of the target battery using the N-ODE model for capacity estimation through the method of transfer learning, and obtains a lithium battery capacity estimation method with high precision of N-ODE and high calculation efficiency of transfer learning through the combination of N-ODE and transfer learning. In particular, after the lithium battery is discharged to 40%, the estimation accuracy of the present method is greatly improved compared with existing methods. As shown in Figure 4 and Figure 5 , respectively, are the discharge process estimation error diagram and the error value diagram at different stages of discharge.

[0101] Embodiment Two

[0102] A battery residual capacity estimation device, as shown in Figure 2 , comprises:

[0103] The acquisition module is configured to acquire open source lithium battery discharge data, lithium battery constant current discharge process data and discharge process data of a lithium battery to be estimated.

[0104] The preprocessing module is configured to preprocess the open source lithium battery discharge data and the lithium battery constant current discharge process data.

[0105] The global feature encoder construction module is configured to construct a global feature encoder based on a gated recurrent unit and a neural ordinary differential equation.

[0106] The global feature encoder calling module is configured to process the source domain data samples by calling the function of the global feature encoder, thereby obtaining a global feature time sequence.

[0107] The source domain model training module is configured to train the neural ordinary differential equation through the global feature time sequence, thereby obtaining a source domain model.

[0108] The target domain model training module is configured to train the source domain model through the target domain data samples, and to perform parameter fine-tuning and network layer freezing on the source domain model, and to perform inference through forward propagation, thereby obtaining a target domain model.

[0109] The estimation module is configured to input the discharge process data of the lithium battery to be estimated into the target domain model, thereby obtaining the residual capacity of the lithium battery to be estimated.

[0110] Embodiment Three

[0111] An electronic device, such as Figure 3 As shown, the electronic device includes a memory storing executable program code and a processor coupled with the memory; wherein the processor invokes the executable program code stored in the memory to execute the method steps disclosed in the above embodiments.

[0112] Embodiment Four

[0113] A computer storage medium storing a computer program, the computer program being run by a processor to execute the method steps disclosed in the above embodiments.

[0114] The present application is described with reference to the flowchart and / or block diagram illustrations of the methods, apparatus (systems) and computer program products according to embodiments of the present application. It will be understood that each block of the flowchart and / or block diagrams, and combinations of blocks in the flowchart and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart and / or block diagram block or blocks. Figure 1 The flowchart and / or block diagrams can include one or more flowcharts and / or one or more block diagrams. Figure 1 The flowchart and / or block diagrams can include one or more flowcharts and / or one or more block diagrams.

[0115] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instructions which implement the function specified in the flowchart and / or block diagram block or blocks. Figure 1 The flowchart and / or block diagrams can include one or more flowcharts and / or one or more block diagrams. Figure 1 The flowchart and / or block diagrams can include one or more flowcharts and / or one or more block diagrams.

[0116] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the function specified in the flowchart and / or block diagram block or blocks. Figure 1 The flowchart and / or block diagrams can include one or more flowcharts and / or one or more block diagrams. Figure 1 The flowchart and / or block diagrams can include one or more flowcharts and / or one or more block diagrams.

[0117] The above is only an operation example of the present application, and the parameter adjustment process of transfer learning and the N-ODE source domain model training process through open source data do not limit the present application. Any person skilled in the art can make any form of equivalent replacement, modification or change to the technical solutions and technical contents disclosed by the present application without departing from the scope of the technical solutions of the present application, and still belongs to the protection scope of the present application.

Claims

1. A method of estimating the remaining capacity of a lithium battery, characterized by: The method comprises the following steps: Preprocessing open source lithium battery discharge data to obtain source domain data samples with time sequence characteristics; Building a global feature encoder based on a gated recurrent unit and a neural ordinary differential equation; Processing the source domain data samples through the global feature encoder to obtain a global feature time sequence; Training a neural ordinary differential equation through the global feature time sequence to obtain a source domain model; Receiving constant current discharge process data of a lithium battery and preprocessing it to obtain target domain data samples with time sequence characteristics; Training the source domain model with the target domain data samples, fine-tuning the parameters of the source domain model, freezing the network layers, and performing inference through forward propagation to obtain a target domain model; Inputting discharge process data of a lithium battery to be estimated into the target domain model to obtain the remaining capacity of the lithium battery to be estimated.

2. The method of claim 1, wherein: The open source lithium battery discharge data and the constant current discharge process data of the lithium battery each comprise discharge time, discharge voltage, discharge current, full charge capacity of the battery, and current remaining capacity of the battery; and the discharge process data of the lithium battery to be estimated comprises discharge time, discharge voltage, discharge current, and full charge capacity of the battery.

3. The method of claim 1, wherein: The preprocessing of the open source lithium battery discharge data comprises screening and removing abnormal data; and the preprocessing of the constant current discharge process data of the lithium battery comprises sparse processing.

4. The method of claim 1, wherein: The calculation process of the global feature encoder is as follows: Formula (1) Formula (2) wherein, denotes a hyperparameter in the network structure, denotes an input, is the discharge voltage of the battery, is the discharge current of the battery, m≤t≤M, m, M are the maximum and minimum values of time, respectively, and f denotes a constructed universal neural differential equation expression; For The first step operation is performed where the input sequence is encoded with a GRU to obtain the initial embedding vectors for the time series process, i.e., Equation (3): (Formula 3) wherein, and are the outputs of the update gate and the reset gate at time step , and are the outputs of the hidden layer and the temporary layer, is the input voltage at time step , and are activation functions, and are weights and thresholds, respectively.

5. The method of claim 4, wherein: The source domain model is obtained through the following steps: feeding the embedding vectors into a neural network ordinary differential equation to obtain continuous embedding vectors, then is determined to be: Formula (4) Formula (5) Here will and New sequence , ; k represents a newly introduced time sequence, which changes the original time sequence t to , n represents the maximum value of k, and N represents The maximum value; the new sequence refers to the time sequence obtained by processing the source domain data sample by the global feature encoder, and k pairs of time sequences are further divided on this basis; Three new parameters are introduced , and and used in the output of the quadratic calculation (eq. 3) and resulting in: Formula (6) wherein, is the step size for the computation, is the weight parameter for the quadratic computation, is the single sample size for the network propagation and the sample size for the forward and backward propagation are the same; y ^ denotes the process variable for the quadratic computation; The initial sequence is then recovered from the successive embeddings of vectors using a variational autoencoder. The state of charge (SOC) of the lithium battery N-ODE (t k ) is output according to equation (1) and equation (4). Formula (7) Formula (8) Formula (9) Formula (10) wherein, Cfull is the full charge capacity of the battery, is the coulombic efficiency and is approximately 1.

6. The method of claim 5, wherein: Fine-tuning the parameters of the source domain model and freezing the network layers specifically comprises replacing the Adam optimizer of the source domain model with a Decay-Adam optimizer, increasing the initial learning rate by 10 times, and freezing the hidden layers of the source domain model.

7. The method of claim 5, wherein: The difference between the estimated value output by the target domain model and the true value is minimized through a root mean square error loss function, which is as follows: Formula (11) where L represents a loss function of network training, SOC * (t k ) represents an experimental value of the battery capacity.

8. A battery remaining capacity estimation device characterized by comprising: It comprises: An acquisition module is configured to acquire open source lithium battery discharge data, constant current discharge process data of a lithium battery, and discharge process data of a lithium battery to be estimated; A preprocessing module is configured to preprocess the open source lithium battery discharge data and the constant current discharge process data of the lithium battery; A global feature encoder construction module is configured to build a global feature encoder based on a gated recurrent unit and a neural ordinary differential equation; A global feature encoder calling module is configured to process source domain data samples through the function of the global feature encoder to obtain a global feature time sequence; A source domain model training module is configured to train a neural ordinary differential equation through the global feature time sequence to obtain a source domain model; A target domain model training module is configured to train the source domain model with target domain data samples, fine-tune the parameters of the source domain model, freeze the network layers, and perform inference through forward propagation to obtain a target domain model; An estimation module is configured to input discharge process data of a lithium battery to be estimated into the target domain model to obtain the remaining capacity of the lithium battery to be estimated.

9. An electronic device, comprising: A computer program product comprising a memory storing executable program code and a processor coupled with the memory; wherein the processor invokes the executable program code stored in the memory to perform the method of any of claims 1-7.

10. A computer readable storage medium storing a computer program, characterized in that: The computer program, when run by a processor, performs the method of any of claims 1-7.

Images