A method and device for constructing an open circuit voltage-state of charge model of a battery

Abstract

The present invention discloses a method and device for constructing an open circuit voltage-state of charge model of a battery, and relates to the technical field of battery management systems. When constructing an OCV-SOC model of a battery based on an SVR model, the present invention continuously performs binary segmentation and binary search on the initial search space of the parameters to be optimized in each dimension of the SVR model, so that the optimal search subspace can be quickly located from the large-scale initial search space of the parameters to be optimized, and then further performs detailed parameter optimization based on the optimal search subspace with a particle optimization algorithm, which greatly reduces the amount of calculation in the construction of the OCV-SOC model. At the same time, the binary segmentation and binary search of the initial search space can comprehensively evaluate the parameter performance of each search subspace in the large-scale initial search space, and can accurately determine the target search space of the PSO algorithm, avoiding the possibility of losing the optimal parameters, and improving the accuracy of the construction of the OCV-SOC model.

Description

Technical Field

[0001] The present invention relates to the technical field of battery management systems, and in particular to a method and device for constructing an open circuit voltage-state of charge model of a battery. Background Art

[0002] At present, the state of charge (SOC) of the battery is an important parameter of the battery management system, which reflects the remaining available power of the battery. Accurately estimating the state of charge of the battery is an important basis for the stable operation of the battery management system. The open circuit voltage (OCV) is the terminal voltage of the battery in the open circuit state. At a certain temperature, there is a one-to-one correspondence between OCV and SOC. An OCV-SOC model can be established, which is used to estimate the battery SOC when the battery OCV is known or to estimate the battery OCV when the battery SOC is known. In the battery management system, both OCV and SOC are short-time scale parameters and need to be estimated online. Therefore, while achieving high-precision estimation, the amount of calculation must not be too high, otherwise it will be difficult to apply.

[0003] In the prior art, temperature parameters are usually introduced into the method model, and then the full-temperature OCV-SOC model is established by considering the battery temperature factor. At present, there are methods for establishing the battery OCV-SOC model based on a less complex model to reduce the amount of calculation, such as establishing the battery OCV-SOC model based on the SVR model, in which the parameters of the SVR model are optimized through a parameter optimization algorithm.

[0004] However, although the existing technical solutions take into account both the computational complexity and accuracy in the OCV-SOC model construction process, the parameter optimization process based on the model parameter search space is difficult to quickly and accurately obtain the optimal parameters, resulting in a large amount of computation and poor accuracy in constructing the battery's OCV-SOC model. Summary of the invention

[0005] Based on this, it is necessary to provide a method and device for constructing an open circuit voltage-state of charge model of a battery in response to the above technical problems.

[0006] The present invention adopts the following technical solutions:

[0007] The present invention provides a method for constructing an open circuit voltage-state of charge model of a battery. The present invention first obtains a mapping relationship data set of the state of charge and the open circuit voltage of the battery at different temperatures, and divides the mapping relationship data set into a training set and a verification set; then, the initial search space of the parameters to be optimized in each dimension of the SVR model is divided into two parts, and the search spaces after the two-part division of different dimensions are combined to obtain a plurality of search subspaces; thereby, a plurality of particles corresponding to the parameters to be optimized in each dimension are generated in each search subspace, and the SVR model is trained by a gradient descent algorithm based on the training set, and the current optimal search subspace with the smallest error after training is determined by the verification set; finally, it is judged whether the error corresponding to the current optimal search subspace is less than a first error threshold; if so, the current optimal search subspace is used as the target search space, and the SVR model is trained by a particle optimization algorithm to obtain an OCV-SOC model of the battery at all temperatures; if not, the current optimal search subspace is used as the initial search space for iterative two-part division search, and the current optimal search subspace is updated until the error corresponding to the current optimal search subspace is less than the first error threshold.

[0008] The present invention provides a device for constructing an open circuit voltage-state of charge model of a battery, comprising:

[0009] An acquisition module is used to acquire a mapping relationship data set between the state of charge and the open circuit voltage of the battery at different temperatures, and divide the mapping relationship data set into a training set and a validation set;

[0010] A segmentation module is used to divide the initial search space of the parameters to be optimized in each dimension of the SVR model into two parts, and combine the search spaces after the binary segmentation of different dimensions to obtain multiple search subspaces;

[0011] The optimal search subspace determination module is used to generate multiple particles corresponding to the parameters to be optimized in each dimension in each search subspace, and train the SVR model through the gradient descent algorithm based on the training set, and determine the current optimal search subspace with the minimum error after training through the verification set;

[0012] A construction module is used to determine whether the error corresponding to the current optimal search subspace is less than a first error threshold; if so, the current optimal search subspace is used as the target search space, and the SVR model is trained by the particle optimization algorithm to obtain the OCV-SOC model of the battery; if not, the current optimal search subspace is used as the initial search space for an iterative binary search, and the current optimal search subspace is updated until the error corresponding to the current optimal search subspace is less than the first error threshold.

[0013] The present invention provides a computer-readable storage medium, wherein the storage medium stores a computer program, and when the computer program is executed by a processor, the method for constructing an open circuit voltage-state of charge model of the battery is implemented.

[0014] The present invention provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned method for constructing an open circuit voltage-state of charge model of a battery when executing the program.

[0015] At least one of the above technical solutions adopted by the present invention can achieve the following beneficial effects:

[0016] When constructing the OCV-SOC model of the battery based on the SVR model, the present invention continuously performs binary division and binary search on the initial search space of the parameters to be optimized in each dimension of the SVR model, so that the optimal search subspace can be quickly located from the large-scale initial search space of the parameters to be optimized through a small amount of data, and then further performs detailed parameter optimization based on the optimal search subspace with a particle optimization algorithm, which greatly reduces the amount of calculation in the construction of the OCV-SOC model. At the same time, the binary division and binary search of the initial search space can comprehensively evaluate the parameter performance of each search subspace in the large-scale initial search space, effectively making up for the problem that the target search space range of the traditional PSO algorithm is difficult to accurately determine, avoiding the possibility of losing the optimal parameters, and improving the accuracy of the construction of the OCV-SOC model. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0018] Figure 1 A schematic diagram of a flow chart of a method for constructing an open circuit voltage-state of charge model of a battery provided by the present invention;

[0019] Figure 2 A schematic diagram of a process for establishing a full-temperature OCV-SOC model based on CPPS-PSO-SVR provided by the present invention;

[0020] Figure 3 A schematic diagram of the estimation error of a test set 1 provided by the present invention;

[0021] Figure 4 A schematic diagram of the estimation error of a test set 2 provided by the present invention;

[0022] Figure 5 A schematic diagram of the estimation error of a test set 3 provided by the present invention;

[0023] Figure 6 A schematic diagram of the estimation error of a test set 4 provided by the present invention;

[0024] Figure 7 A schematic diagram of a device for constructing an open circuit voltage-state of charge model of a battery provided by the present invention. DETAILED DESCRIPTION

[0025] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the specific embodiments of the present invention and the corresponding drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0026] At present, in battery management systems, both OCV and SOC are short-time scale parameters that need to be estimated online. Therefore, while achieving high-precision estimation, the amount of calculation must not be too high, otherwise it will be difficult to apply. Existing methods reduce the amount of calculation by selecting models with lower complexity, but the accuracy of the model will also decrease. The method of discretizing the model into a table can avoid calculations, but it requires a large amount of stored data, which reduces the corresponding table lookup speed and increases the storage cost.

[0027] The present invention proposes a full-temperature OCV-SOC model based on CPPS-PSO-SVR, which is a high-precision online OCV-SOC model adapting to the full operating temperature range. The model is divided into two parts: cosine partition particle synergy (CPPS) improved particle swarm optimization (PSO) parameter optimization and support vector machine regression (SVR) modeling. The CPPS-PSO-SVR algorithm can effectively balance the complexity and accuracy of the full-temperature OCV-SOC model and significantly reduce the training cost.

[0028] The technical solutions provided by various embodiments of the present invention are described in detail below in conjunction with the accompanying drawings.

[0029] Figure 1 The figure is a flow chart of a method for constructing an open circuit voltage-state of charge model of a battery in the present invention, which specifically includes the following steps:

[0030] S101: Obtain a mapping relationship data set between the state of charge and the open circuit voltage of a battery at different temperatures, and divide the mapping relationship data set into a training set and a validation set.

[0031] S102: The initial search space of the parameters to be optimized in each dimension of the SVR model is divided into two parts, and the search spaces after the two-part division in different dimensions are combined to obtain a plurality of search subspaces.

[0032] S103: Generate multiple particles corresponding to the parameters to be optimized in each dimension in each search subspace, and train the SVR model through the gradient descent algorithm based on the training set, and determine the current optimal search subspace with the minimum error after training through the verification set.

[0033] S104: Determine whether the error corresponding to the current optimal search subspace is less than a first error threshold; if so, execute step S105; if not, execute step S106.

[0034] S105: Taking the current optimal search subspace as the target search space, training the SVR model through a particle optimization algorithm to obtain an OCV-SOC model of the battery.

[0035] S106: Perform an iterative binary search using the current optimal search subspace as the initial search space, and update the current optimal search subspace until the error corresponding to the current optimal search subspace is less than a first error threshold.

[0036] For the convenience of explanation, the following description is only based on the server as the execution subject. The server mentioned in the present invention can be a server set up on the business platform, or a device such as a desktop computer, a notebook computer, etc. that can execute the solution of the present invention.

[0037] The present invention proposes a method for establishing a full-temperature OCV-SOC model based on CPPS-PSO-SVR. In the process of establishing the full-temperature OCV-SOC model based on the SVR model, the PSO algorithm is improved by the CPPS algorithm to obtain the CPPS-PSO algorithm, so that the parameters of the SVR model are optimized by the CPPS-PSO algorithm to effectively balance the complexity and accuracy of establishing the full-temperature OCV-SOC model.

[0038] In one or more embodiments of the present invention, the server may establish an OCV-SOC model based on the SVR model. D OCV-SOC mapping relationship dataset of samples , where It is the input feature vector, which can contain two features: SOC and battery temperature; is the corresponding true value of OCV. The SVR model regression prediction function is: .

[0039] Where x is the model input, is the model output, is the weight parameter,b is the bias parameter, is the model mapping. SVR determines the parameters by maximizing the margin and minimizing the loss and b Unlike polynomial regression, it can tolerate The deviation of the model output from the true value is only The loss is calculated only after the loss is calculated. Therefore, the parameter determination can be transformed into the following convex optimization problem: .

[0040] In the formula, It is an insensitive parameter.

[0041] In order to overcome the influence of noisy data, a relaxation factor can be introduced and Then we get: .

[0042] In the formula, C is the penalty factor.

[0043] To solve the nonlinear mapping in the solution process and The problem of high complexity of inner product operation, among which, , introduce the kernel function , so that it satisfies: .

[0044] In one or more embodiments of the invention, the selected kernel function may be a radial basis kernel function: .

[0045] In the formula, The bandwidth parameter of the radial basis kernel function. Finally, the final OCV-SOC model can be obtained through the Lagrange multiplier method and KKT conditions: .

[0046] In the formula, and is the Lagrange multiplier.

[0047] In the process of establishing the OCV-SOC model, CPPS-PSO can be used to optimize the key hyperparameters that affect the performance of the SVR algorithm to establish an OCV-SOC model with excellent performance. Figure 2 As shown, Figure 2 It is a schematic diagram of a process of establishing a full-temperature OCV-SOC model based on CPPS-PSO-SVR in the present invention.

[0048] Specifically, after obtaining the OCV-SOC mapping relationship dataset of the battery at different battery temperatures, the server can first pre-process the data in the mapping relationship dataset, for example, interpolate missing values, discard outliers, perform min-max normalization on the SOC and battery temperature data, and divide the training set , validation set and test set .

[0049] Then, an OCV-SOC model is established based on SVR, with the input features being battery temperature and SOC, and the output being OCV; a large range of parameter initial search space is set, other CPPS-PSO parameters are set, and the cost function Set to: .

[0050] In the formula, is the number of samples in the validation set, The validation sample for the SVR model Output OCV prediction value, To verify the sample The corresponding OCV true value.

[0051] In the process of establishing the OCV-SOC model based on SVR, CPPS-PSO is used to adjust the penalty factor in the SVR model. C , insensitive parameters and the bandwidth parameter of the radial basis kernel function Optimize. So according to the optimal parameters and and The collection of SVRs is used to establish the final full-temperature OCV-SOC model.

[0052] When optimizing the parameters in the SVR model based on CPPS-PSO, in one or more embodiments of the present invention, it is assumed that n The initial search space for dimension parameters is: .

[0053] In the formula, S represents the initial search space, Indicates i The lower limit of the parameter to be optimized. Indicates i The upper limit of the parameter to be optimized.

[0054] The initial search space of the parameters to be optimized in each dimension of the SVR model can be divided into two parts by the following formula, and the search spaces after the two-division divisions in different dimensions can be combined to obtain multiple search subspaces: , .

[0055] In the formula, Indicates j search subspace, Indicates j Search subspace i The lower limit of the parameter to be optimized. Indicates j Search subspace i The upper limit of the parameter to be optimized. n Indicates the number of dimensions of the parameter to be optimized.

[0056] The server can generate the corresponding parameters to be optimized in each dimension in each search subspace. Particles, k The position of a particle is: By generating a small number of particles in each search subspace to optimize the search subspace, the computational complexity of establishing the OCV-SOC model can be further reduced.

[0057] In the formula, Indicates j Search subspace k The position of a particle, Indicates j Search subspace i Dimensional location.

[0058] If you will k -1 Use Bit binary representation, then: .

[0059] In the formula, for k -1 Use When the bit is represented in binary i Binary number.

[0060] Afterwards, the server can train the SVR model based on the training set through the gradient descent algorithm, and determine the current optimal search subspace with the smallest error after training through the validation set. Specifically, the server can normalize each particle in each search subspace, and set the normalization standard formula as: .

[0061] In the formula, is the normalized particle position.

[0062] The server can also calculate the increment based on the normalized particle position and gradient of each particle in each search subspace, and determine the partial derivative of each particle in each search subspace based on the cost function for training the SVR model, as shown in the following formula: .

[0063] In the formula, For the particle j Subspace i The partial derivative of , is the cost function, is the anti-normalization function, is the normalized particle position vector, is the increment of gradient calculation, For the i -dimensional unit vector.

[0064] The denormalization formula can be inferred from the normalization standard formula. Then, the server can normalize each search subspace according to the learning rate of gradient descent and the partial derivative of each particle. Particles perform gradient descent: ; .

[0065] In the formula, , T is the total number of steps in the current cycle, T The initial value is , after each cycle ends, , ; If after each cycle is greater than the remaining number of iterations of gradient descent, then let T Equal to the number of remaining iterations. is the initial learning rate, is the final learning rate. In each cycle, the cosine decreases. The early cycle is short and the decrease is faster, while the later cycle is long and the decrease is slower. This is conducive to the particles jumping out of the local optimal solution and converging to the global optimal solution.

[0066] The server can perform multiple gradient descents on each particle in each search subspace until the maximum number of gradient descent iterations is reached. , .

[0067] Finally, the server can denormalize each particle in each search subspace after gradient descent, determine the average error of each particle in each search subspace according to the cost function for training the SVR model through the validation set, and take the search subspace with the smallest average error as the current optimal search subspace.

[0068] After a round of subspace optimization, the server can determine whether the error of the optimal subspace is less than the first error threshold. Or equal to the maximum number of iterations of cosine partition particle collaborative optimization If yes, the current optimal search subspace is used as the target search space of the PSO algorithm. If no, the optimal subspace is used as the initial search space for the next round of binary search, and the current optimal search subspace is updated until the error corresponding to the current optimal search subspace is less than the first error threshold.

[0069] After obtaining the target search space that meets the error requirements, the server can use the current optimal search subspace as the target search space, and generate the parameters to be optimized in each dimension of the SVR model according to the center point of the target search space. M The initial position of the particle, m A particle is represented as a n dimensional vector: In the formula, Indicates m The particle in i Position in one dimension.

[0070] You can also initialize the initial speed of each particle. The initial speed of each particle can be a random number in [-0.25, 0.25]. m The "flying" speed of a particle is also n dimensional vector, which can be written as: In the formula, Indicates m The particle in i Speed ​​in one dimension.

[0071] Afterwards, the fitness of each particle can be determined based on the cost function of the SVR model training and the initial position of each particle. Then the position of each particle is normalized, and the normalization standard formula is set as: .

[0072] In the formula, is the normalized particle position, and They represent the PSO target search space i The lower and upper limits of the dimension. The anti-normalization formula can be obtained by inverting the normalization standard formula.

[0073] Then, the initial position and velocity of each particle can be updated according to the particle velocity and position update process. The first m Particles to During the evolution of generations, the speed and position update formulas are as follows: , .

[0074] In the formula, is the inertia coefficient, which indicates the degree to which the particle maintains its original velocity. is the weight coefficient of the particle tracking its own historical optimal value, and They are random numbers in [0,1] to increase the randomness of particle search. Indicates Daidi m The local optimal position of a particle, is the weight coefficient of the particle tracking group's historical optimal value, Indicates The global optimal position of each particle.

[0075] Next, the server can determine the updated fitness of each particle according to the cost function of training the SVR model based on the sample set and the updated position of each particle; and determine the local optimal position of each particle and the global optimal position of each particle according to the fitness of each particle before and after the update.

[0076] The local optimal position of each particle and the global optimal position of each particle are updated through multiple rounds of iterations until the fitness corresponding to the global optimal position of each particle is less than the second error threshold Or the maximum number of PSO iterations is reached , the parameters to be optimized in each dimension of the SVR model are determined according to the global optimal position of each particle; thus, the SVR model after parameter optimization is trained through the training set to obtain the OCV-SOC model of the battery at all temperatures.

[0077] based on Figure 1 The open circuit voltage-state of charge model construction method of the battery shown in the figure, when the OCV-SOC model of the battery is constructed based on the SVR model, the initial search space of the parameters to be optimized in each dimension of the SVR model is continuously binary divided and binary searched, so that the optimal search subspace can be quickly located from the large-scale initial search space of the parameters to be optimized through a small amount of data, and then the particle optimization algorithm is further used based on the optimal search subspace to perform detailed parameter optimization, which greatly reduces the amount of calculation in the construction of the OCV-SOC model. At the same time, the binary division and binary search of the initial search space can comprehensively evaluate the parameter performance of each search subspace in the large-scale initial search space, effectively making up for the problem that the target search space range of the traditional PSO algorithm is difficult to accurately determine, avoiding the possibility of losing the optimal parameters, and improving the accuracy of the OCV-SOC model construction.

[0078] The CPPS algorithm of the present invention has a high parameter adjustment margin and is easy to implement parameter configuration. By partitioning the large-scale initial search space of arbitrary scale and combining specific normalization and particle gradient descent, the CPPS algorithm can comprehensively evaluate the parameter performance of each subspace and quickly narrow the target search space range. This method effectively makes up for the problem that the target search space range of the traditional PSO algorithm is difficult to accurately determine, and avoids the possibility of losing the optimal parameters. In addition, compared with the traditional method of repeatedly adjusting the target search space and re-optimizing the PSO parameters, the CPPS-PSO algorithm only needs to provide a large-scale target search space of arbitrary scale, which can run autonomously and efficiently complete parameter optimization, thereby significantly reducing the training cost.

[0079] The present invention optimizes the parameters of the SVR algorithm based on the CPPS-PSO algorithm, which can effectively balance the complexity and accuracy of the full-temperature OCV-SOC model, and then establish an OCV-SOC model that meets the online estimation requirements of the battery management system.

[0080] When applying the method for constructing the open circuit voltage-state of charge model of the battery provided by the present invention, it is not necessary to Figure 1 The steps are executed in the order shown. The specific execution order of the steps can be determined according to needs, and the present invention does not limit this.

[0081] In addition, the present invention also provides an embodiment of the method for constructing an open circuit voltage-state of charge model of a battery provided by the present invention. In this embodiment, the open source A123 lithium-ion battery OCV-SOC mapping relationship data set of the University of Maryland in the United States at -10°C, 0°C, 10°C, 20°C, 25°C, 30°C, 40°C, and 50°C is first obtained. N .

[0082] Preprocess the collected data, interpolate missing values, discard outliers, normalize SOC and battery temperature data to min-max, and divide the training set , validation set and test set .in, and Contains 7000 sample data in total. : =4:1, consisting of data at -10℃, 0℃, 10℃, 20℃, 30℃, 40℃, and 50℃; It contains 4000 sample data in total, consisting of data at -10℃, 0℃, 25℃, and 50℃.

[0083] An OCV-SOC model is established based on SVR, with the input features being battery temperature and SOC, and the output being OCV. A large-scale parameter three-dimensional initial search space is set [(1.0, 1000.0), (0.00001, 0.01), (1.0, 1000.0)], where the three dimensions represent penalty factors. C , insensitive parameters and the bandwidth parameter of the radial basis kernel function , set other parameters of CPPS-PSO, =2, =10, =0.001, =0.001, =30, =0.0005, =5, =20, =0.8, =2, =2, =0.0002, =50, cost function Set to: .

[0084] Using CPPS-PSO to find insensitive parameters in OCV-SOC model , penalty factor C and the bandwidth parameter of the radial basis kernel function Optimize and obtain the final full-temperature OCV-SOC model. The temperature is divided into four test sets at -10℃, 0℃, 25℃, and 50℃, and the above-mentioned full-temperature OCV-SOC model is tested on the four test sets respectively. The root mean square error (RMSE) and mean absolute error (MAE) of the four tests are shown in Table 1:

[0085] Table 1 Estimation error of OCV-SOC model

[0086]

[0087] Figure 3 This is a schematic diagram of the estimation error of a test set 1 in the present invention, Figure 4 This is a schematic diagram of the estimation error of a test set 2 in the present invention, Figure 5 This is a schematic diagram of the estimation error of a test set 3 in the present invention, Figure 6 This is a schematic diagram of the estimated error of a test set 4 in the present invention. Figure 3 , Figure 4 , Figure 5 , Figure 6It can be seen from Table 1 that in a large range of parameter initial search space [(1.0, 1000.0), (0.00001, 0.01), (1.0, 1000.0)], the estimation error of the full-temperature OCV-SOC model based on CPPS-PSO-SVR is small. These values ​​can further illustrate the effectiveness of the OCV-SOC model construction method.

[0088] The above is a method for constructing a battery open circuit voltage-state of charge model provided by one or more embodiments of the present invention. Based on the same idea, the present invention also provides a corresponding battery open circuit voltage-state of charge model construction device, such as Figure 7 shown.

[0089] Figure 7 A schematic diagram of a device for constructing an open circuit voltage-state of charge model of a battery provided by the present invention, comprising:

[0090] An acquisition module 201 is used to acquire a mapping relationship data set between the state of charge and the open circuit voltage of the battery at different temperatures, and divide the mapping relationship data set into a training set and a validation set;

[0091] A segmentation module 202 is used to divide the initial search space of the parameters to be optimized in each dimension of the SVR model into two parts, and combine the search spaces after the binary segmentation in different dimensions to obtain multiple search subspaces;

[0092] The optimal search subspace determination module 203 is used to generate multiple particles corresponding to the parameters to be optimized in each dimension in each search subspace, and train the SVR model through the gradient descent algorithm based on the training set, and determine the current optimal search subspace with the minimum error after training through the verification set;

[0093] The construction module 204 is used to determine whether the error corresponding to the current optimal search subspace is less than the first error threshold; if so, the current optimal search subspace is used as the target search space, and the SVR model is trained by the particle optimization algorithm to obtain the OCV-SOC model of the battery; if not, the current optimal search subspace is used as the initial search space for iterative binary search, and the current optimal search subspace is updated until the error corresponding to the current optimal search subspace is less than the first error threshold.

[0094] For the specific definition of the battery open circuit voltage-state of charge model construction device, please refer to the definition of the battery open circuit voltage-state of charge model construction method mentioned above, which will not be repeated here. Each module in the above-mentioned battery open circuit voltage-state of charge model construction device can be implemented in whole or in part by software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0095] The present invention also provides a computer-readable storage medium, which stores a computer program, which can be used to execute the above Figure 1 A method for constructing an open circuit voltage-state of charge model of a battery is provided.

[0096] The present invention also provides a computer device. At the hardware level, the computer device includes a processor, an internal bus, a network interface, a memory, and a non-volatile memory. Of course, it may also include hardware required for other services. The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it to achieve the above Figure 1 A method for constructing an open circuit voltage-state of charge model of a battery is provided.

[0097] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided by the present invention can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory or optical memory, etc. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0098] The technical features of the above embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of the present invention.

Claims

1. A method for constructing an open circuit voltage-state of charge model of a battery, characterized in that: include: Obtain a mapping relationship data set between the state of charge and the open circuit voltage of the battery at different temperatures, and divide the mapping relationship data set into a training set and a validation set; The initial search space of the parameters to be optimized in each dimension of the SVR model is divided into two parts, and the search spaces after the two-part divisions in different dimensions are combined to obtain multiple search subspaces; Generate multiple particles corresponding to the parameters to be optimized in each dimension in each search subspace, and train the SVR model through the gradient descent algorithm based on the training set, and determine the current optimal search subspace with the minimum error after training through the validation set; Determine whether the error corresponding to the current optimal search subspace is less than a first error threshold; If so, the current optimal search subspace is used as the target search space, and the SVR model is trained by the particle optimization algorithm to obtain the OCV-SOC model of the battery at all temperatures; If not, the current optimal search subspace is used as the initial search space for iterative binary split search, and the current optimal search subspace is updated until the error corresponding to the current optimal search subspace is less than the first error threshold; The initial search space of the parameters to be optimized in each dimension of the SVR model is divided into two parts, and the search spaces after the two-part division in different dimensions are combined to obtain multiple search subspaces, specifically including: The initial search space of the parameters to be optimized in each dimension of the SVR model is: ; The initial search space of the parameters to be optimized in each dimension of the SVR model is divided into two parts by the following formula, and the search spaces after the two-division divisions in different dimensions are combined to obtain multiple search subspaces: , ; in, Indicates i The lower limit of the parameter to be optimized. Indicates i The upper limit of the parameter to be optimized. S represents the initial search space, Indicates j search subspace, Indicates j Search subspace i The lower limit of the parameter to be optimized. Indicates j Search subspace i The upper limit of the parameter to be optimized. n Indicates the number of dimensions of the parameters to be optimized; The step of generating a plurality of particles corresponding to the parameters to be optimized in each dimension in each search subspace specifically includes: Generate the corresponding parameters to be optimized in each dimension in each search subspace particles, the position of each particle is: ， ； in, n Indicates the number of dimensions of the parameters to be optimized, Indicates j Search subspace k The position of a particle, Indicates j Search subspace i The location of the dimension, Indicates j Search subspace i The lower limit of the parameter to be optimized. Indicates j Search subspace i The upper limit of the parameter to be optimized. for k- 1 Use When the bit is represented in binary i Bit binary number; The SVR model is trained by a gradient descent algorithm based on the training set, and the current optimal search subspace with the minimum error after training is determined by a validation set, specifically including: Normalize each particle in each search subspace; According to the normalized particle position and gradient calculation increment of each particle in each search subspace, the partial derivative of each particle in each search subspace is determined based on the cost function for training the SVR model; The learning rate for gradient descent is determined by ; According to the learning rate of gradient descent and the partial derivative of each particle, the normalized particles in each search subspace are gradient-descented; After denormalizing each particle in each search subspace after gradient descent, the average error of each particle in each search subspace is determined according to the cost function for training the SVR model through the validation set, and the search subspace with the smallest average error is taken as the current optimal search subspace; in, is the learning rate of gradient descent, is the initial learning rate, is the final learning rate, is the total number of steps in the current cycle, .

2. The method for constructing an open circuit voltage-state of charge model of a battery according to claim 1, wherein: The current optimal search subspace is used as the target search space, and the SVR model is trained by the particle optimization algorithm to obtain the OCV-SOC model of the battery at all temperatures, which specifically includes: The current optimal search subspace is used as the target search space, and the initial positions of multiple particles corresponding to the parameters to be optimized in each dimension of the SVR model are generated according to the center point of the target search space, and the initial speed of each particle is initialized; Determine the fitness of each particle based on the cost function of training the SVR model and the initial position of each particle; The initial position and speed of each particle are updated according to the particle speed and position update process, and the fitness of each particle after the update is determined according to the cost function of training the SVR model based on the sample set and the updated position of each particle; According to the fitness of each particle before and after the update, the local optimal position of each particle and the global optimal position of each particle are determined; The local optimal position of each particle and the global optimal position of each particle are updated through multiple rounds of iterations until the fitness corresponding to the global optimal position of each particle is less than the second error threshold, and the parameters to be optimized in each dimension of the SVR model are determined according to the global optimal position of each particle; The SVR model after parameter optimization is trained through the training set to obtain the OCV-SOC model of the battery at all temperatures.

3. The method for constructing an open circuit voltage-state of charge model of a battery according to claim 1 or 2, characterized in that: The cost function for training the SVR model is: ; in, is the number of samples in the validation set, The validation sample for the SVR model Output OCV prediction value, To verify the sample The corresponding OCV true value.

4. The method for constructing an open circuit voltage-state of charge model of a battery according to claim 1, wherein: The parameters to be optimized in each dimension include: a penalty factor, an insensitive parameter and a bandwidth parameter of a radial basis kernel function.

5. A device for constructing an open circuit voltage-state of charge model of a battery based on the method according to any one of claims 1 to 4, characterized in that: include: An acquisition module is used to acquire a mapping relationship data set between the state of charge and the open circuit voltage of the battery at different temperatures, and divide the mapping relationship data set into a training set and a validation set; The initial search space of the parameters to be optimized in each dimension of the segmentation module for the SVR model is: The initial search space of the parameters to be optimized in each dimension of the SVR model is divided into two parts by the following formula, and the search spaces after the two-division divisions in different dimensions are combined to obtain multiple search subspaces: , ; The optimal search subspace determination module is used to generate the corresponding parameters to be optimized in each dimension in each search subspace. particles, and the position of each particle is: , ; and normalize each particle in each search subspace; calculate the increment based on the normalized particle position and gradient of each particle in each search subspace, and determine the partial derivative of each particle in each search subspace based on the cost function of training the SVR model; determine the learning rate of gradient descent by the following formula ; Perform gradient descent on each particle in each search subspace after normalization according to the learning rate of gradient descent and the partial derivative of each particle; After denormalizing each particle in each search subspace after gradient descent, determine the average error of each particle in each search subspace according to the cost function for training the SVR model through the validation set, and take the search subspace with the smallest average error as the current optimal search subspace; A construction module is used to determine whether the error corresponding to the current optimal search subspace is less than a first error threshold; If yes, the current optimal search subspace is used as the target search space, and the SVR model is trained by the particle optimization algorithm to obtain the OCV-SOC model of the battery; if no, the current optimal search subspace is used as the initial search space for iterative binary search, and the current optimal search subspace is updated until the error corresponding to the current optimal search subspace is less than the first error threshold; in, Indicates i The lower limit of the parameter to be optimized. Indicates i The upper limit of the parameter to be optimized. S represents the initial search space, Indicates j search subspace, Indicates j Search subspace i The lower limit of the parameter to be optimized. Indicates j Search subspace i The upper limit of the parameter to be optimized. n Indicates the number of dimensions of the parameters to be optimized; Indicates j Search subspace k The position of a particle, Indicates j Search subspace i The location of the dimension, for k- 1 Use When the bit is represented in binary i Bit binary number; is the learning rate of gradient descent, is the initial learning rate, is the final learning rate, is the total number of steps in the current cycle, .

6. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 4 is implemented.

7. A computer device, characterized in that: The method comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method according to any one of claims 1 to 4 is implemented.

Images