State of charge prediction method and apparatus, computer device, and storage medium

Abstract

This application discloses a method, apparatus, computer device, and storage medium for predicting the state of charge (SOC). The method includes: acquiring initial state parameters, including the initial SOC and battery electrical parameters; inputting the initial state parameters into a target model to obtain a predicted voltage, the target model including an equivalent circuit model and a neural network model for predicting the model parameters of the equivalent circuit model; and updating the initial SOC of the battery based on the predicted voltage to obtain the target SOC. This application effectively integrates the nonlinear fitting capability of deep learning with the physical consistency of the equivalent circuit model, improving prediction accuracy while enhancing the algorithm's generalization ability under different operating conditions and environments, thereby achieving higher accuracy and stronger generalization capability in SOC prediction.

Description

Technical Field

[0001] This application relates to the field of computer technology, and specifically to a method, apparatus, computer device, and storage medium for predicting the state of charge. Background Technology

[0002] The state of charge (SOC) of a battery is a core parameter that measures the ratio of the battery's remaining usable capacity to its fully charged capacity. It is also a key parameter in the battery management system (BMS) and directly affects the battery's performance, lifespan, and safety.

[0003] Currently, the estimation of state of charge (SOC) typically relies on the equivalent circuit model (ECM), which can effectively describe the physical characteristics of the battery and thus has strong generalization ability. However, the accuracy of this approach is limited by the complexity of the model and the accuracy of parameter identification, affecting the prediction accuracy of SOC. Summary of the Invention

[0004] To at least partially solve the above problems, this application provides a method, apparatus, computer device, and storage medium for predicting the state of charge, aiming to achieve higher accuracy and stronger generalization ability in predicting the state of charge.

[0005] In a first aspect, this application provides a method for predicting the state of charge, comprising:

[0006] Obtain the initial state parameters of the battery under test, including the initial state of charge and battery electrical parameters;

[0007] The initial state parameters are input into the trained target model for processing to obtain the first predicted voltage of the battery under test. The target model includes an equivalent circuit model and a neural network model for predicting the model parameters of the equivalent circuit model.

[0008] The initial state of charge of the battery under test is updated based on the first predicted voltage to obtain the target state of charge of the battery under test:

[0009] In some embodiments of this application, the step of inputting the initial state parameters into a trained target model for processing to obtain the first predicted voltage of the battery under test includes:

[0010] The initial state parameters are input into the trained neural network model to obtain the first identification parameters;

[0011] The model parameters of the equivalent circuit model are determined based on the first identification parameters to determine the target equivalent circuit model;

[0012] The initial state parameters are input into the target equivalent circuit model to obtain the first predicted voltage of the battery under test.

[0013] In some embodiments of this application, determining the model parameters of the equivalent circuit model based on the first identification parameter includes:

[0014] Based on multiple sets of state parameters of the battery under test, parameter identification is performed on the battery under test to obtain the second identification parameters;

[0015] The model parameters of the equivalent circuit model are determined based on the first identification parameter and the second identification parameter to determine the target equivalent circuit model.

[0016] In some embodiments of this application, determining the model parameters of the equivalent circuit model based on the first identification parameter and the second identification parameter includes:

[0017] The first identification parameter and the second identification parameter are weighted according to preset weights to obtain the model parameters of the equivalent circuit model;

[0018] The preset weights are determined by at least one of the confidence level of the first identification parameter output by the neural network model, the battery health status of the battery under test, and the operating condition status of the battery under test.

[0019] In some embodiments of this application, after determining the model parameters of the equivalent circuit model based on the first identification parameter and the second identification parameter, the method further includes:

[0020] The model parameters of the equivalent circuit model are transformed to a preset range of values ​​according to preset constraints; and

[0021] The target equivalent circuit model is determined based on the model parameters of the transformed equivalent circuit model.

[0022] In some embodiments of this application, the target model is trained through the following steps:

[0023] Determine the initial equivalent circuit model of the battery under test, and determine the initial neural network model based on the order of the initial equivalent circuit model;

[0024] Obtain training state parameters for model training, the training state parameters including a time series of charged states;

[0025] The training state parameters are input into the initial neural network model and the initial equivalent circuit model to obtain the second predicted voltage corresponding to the training state parameters;

[0026] The initial neural network model is updated based on the second predicted voltage and the target voltage associated with the training state parameters until a trained target model is obtained when preset training conditions are met.

[0027] In some embodiments of this application, updating the initial neural network model based on the target voltage associated with the second predicted voltage and the training state parameters includes:

[0028] The first model loss value of the initial neural network model is determined based on the difference between the second predicted voltage and the target voltage;

[0029] The second model loss value of the initial neural network model is determined based on the difference result of the second predicted voltage;

[0030] The initial neural network model is updated based on the first model loss value and the second model loss value to obtain the updated neural network model.

[0031] In some embodiments of this application, updating the initial state of charge of the battery under test based on the first predicted voltage to obtain the target state of charge of the battery under test includes:

[0032] Based on the difference between the first predicted voltage and the measured voltage of the battery under test, the initial state of charge of the battery under test is filtered and updated to obtain the target state of charge of the battery under test.

[0033] Secondly, this application also provides a state of charge prediction device, comprising:

[0034] The acquisition module is used to acquire the initial state parameters of the battery under test, including the initial state of charge and battery electrical parameters.

[0035] The prediction module is used to input the initial state parameters into the trained target model for processing to obtain the first predicted voltage of the battery under test. The target model includes an equivalent circuit model and a neural network model for predicting the model parameters of the equivalent circuit model.

[0036] An update module is used to update the initial state of charge of the battery under test based on the first predicted voltage to obtain the target state of charge of the battery under test.

[0037] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, it implements the state of charge prediction method as described in any of the preceding claims.

[0038] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, the computer program being loaded by a processor to execute the state of charge prediction method described in any of the preceding claims.

[0039] The state of charge prediction method provided in this application processes the initial state parameters of the battery by using a target model of a neural network model that integrates an equivalent circuit model and the model parameters of the equivalent circuit model to obtain the predicted voltage of the battery under test, thereby updating and predicting the state of charge of the battery. It can effectively integrate the nonlinear fitting ability of deep learning and the physical consistency of the equivalent circuit model, and can improve the prediction accuracy while enhancing the generalization ability of the algorithm under different operating conditions and environmental conditions, thereby achieving higher accuracy and stronger generalization ability in state of charge prediction. Attached Figure Description

[0040] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0041] Figure 1 A schematic flowchart illustrating the steps of a method for predicting the state of charge provided in an embodiment of this application;

[0042] Figure 2 This is a schematic flowchart illustrating the steps for obtaining predicted voltage by processing state parameters through a fusion model, as provided in an embodiment of this application.

[0043] Figure 3 This application provides a flowchart illustrating the steps for determining equivalent circuit model parameters by integrating multi-parameter identification results in an embodiment of the present application.

[0044] Figure 4 This application provides a schematic flowchart illustrating the steps involved in training a target model.

[0045] Figure 5 This application provides a flowchart illustrating the steps for determining model loss values ​​using fusion parameter smoothness loss.

[0046] Figure 6a This diagram illustrates the effect of data processing based on a fusion model to obtain predicted voltage.

[0047] Figure 6b This is a schematic diagram of the complete process of state of charge prediction provided in the embodiments of this application;

[0048] Figure 7A schematic diagram of the structure of a state of charge prediction device provided in an embodiment of this application;

[0049] Figure 8 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0050] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. 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.

[0051] The use of "applies to" or "configured to" in this application implies open and inclusive language, which does not exclude applicability to or configuration to devices performing additional tasks or steps. Additionally, the use of "based on" implies openness and inclusivity, because processes, steps, calculations, or other actions "based on" one or more of the stated conditions or values ​​may in practice be based on additional conditions or values ​​beyond those stated.

[0052] In this application, the term "exemplary" is used to mean "used as an example, illustration, or description." Any embodiment described as "exemplary" in this application is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use this application. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that this application can be made without using these specific details. In other instances, well-known structures and processes are not described in detail to avoid obscuring the description of this application with unnecessary detail. Therefore, this application is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed in this application.

[0053] To clearly understand the application scenarios of the state of charge (SOC) prediction method, apparatus, computer equipment, and storage medium provided in the embodiments of this application, the relevant background of the embodiments of this application will be specifically described below. Specifically, SOC is a core parameter used to measure the ratio of a battery's remaining usable capacity to its fully charged capacity, directly affecting the battery's performance, lifespan, and safety. However, conventional open-circuit voltage methods and ampere-hour integration methods have limited generalization ability under different operating conditions, especially insufficient accuracy under dynamic loads and complex environmental conditions. Based on this, related technologies improve the generalization effect of SOC prediction by constructing an equivalent circuit model (ECM) to describe the battery's physical characteristics. However, the prediction of the state of charge by the equivalent circuit model depends on the model complexity and the accuracy of parameter identification. For example, commonly used equivalent circuit models usually use at least one parallel RC network to represent the relaxation process, that is, the process of voltage slowly changing with time, so as to simulate the dynamic response of chemical reactions and diffusion processes inside the battery. The more parallel RC networks used, that is, the higher the order of the equivalent circuit model, the more detailed the dynamic process that the equivalent circuit model can describe and the higher the accuracy. However, the model complexity, computational load and parameter identification difficulty also increase significantly. The mutual constraint between the model complexity and the accuracy of parameter identification also affects the prediction accuracy of the battery state of charge to a certain extent.

[0054] To address the aforementioned technical problems, this application provides a method for predicting the state of charge (SOC). By integrating the physical consistency of the equivalent circuit model with the nonlinear fitting capability of deep learning, it achieves accurate prediction of the SOC. For clarity, specific embodiments will be described below. Please refer to the detailed description. Figure 1 , Figure 1 This application provides a flowchart illustrating the steps of a method for predicting the state of charge, specifically including steps S110 to S130:

[0055] S110, obtain the initial state parameters of the battery under test.

[0056] In some embodiments of this application, the battery under test refers to the battery for which state of charge prediction is required. It can typically be a single battery or a battery module composed of multiple batteries, existing in the form of a battery pack or battery group. This application does not limit the form of the battery under test. Similarly, the state of charge prediction provided in this application can also be applied to different types of chemical batteries, such as lithium batteries, lead-acid batteries, nickel-based batteries, etc. This application also does not limit the type of battery. For ease of understanding, subsequent embodiments of this application will use lithium batteries as an example for explanation.

[0057] Specifically, in the embodiments of this application, the state parameters of the battery under test can generally be classified as parameters closely related to the state of charge (SOC) and used to predict the SOC of the battery. Since the SOC of a lithium battery is usually dynamically changing, the state parameters here include, but are not limited to, a given initial SOC, and can also include other electrical parameters that affect the SOC, such as, but not limited to, voltage and other information. Of course, based on actual needs, these electrical parameters can also include information such as resistance, capacitance, and inductance. Furthermore, considering that temperature can affect chemical reactions to some extent, thereby affecting the performance of the chemical battery, in some embodiments of this application, the state parameters can also include the ambient temperature of the battery's operating environment, etc. Of course, in addition to the parameters provided above, based on the differences in actual batteries and operating environments, the state parameters can also include any other information that can affect the battery's SOC. The embodiments of this application do not limit this, but for ease of description, in some embodiments of this application, the initial state parameters including the initial SOC, initial polarization voltage, current, and temperature will be used as examples for illustration, but the above scheme should not be considered a limitation of the scheme of this application.

[0058] In particular, the state of charge prediction method provided in this application can also be regarded as a cyclic process of dynamically predicting the state of charge of a battery. That is, after the state of charge is predicted by the state of charge prediction method, the obtained state of charge will be used as a new initial state of charge and other electrical parameters that affect the state of charge in real time. These parameters will be input into the trained target model to predict the state of charge at the next moment. In other words, the initial state of charge here is not the state of charge at a specific moment, but refers to the real-time state of charge used to predict the target state of charge at the next moment in each round of the cyclic prediction process.

[0059] S120, the initial state parameters are input into the trained target model for processing to obtain the first predicted voltage of the battery under test.

[0060] Specifically, based on the foregoing, in order to accurately predict the state of charge of the battery by utilizing the initial state parameters of the battery, in this embodiment of the application, the acquired state parameters are input into the trained target model for processing, thereby obtaining the first predicted voltage of the battery under test.

[0061] Specifically, in the embodiments of this application, the target model is usually a fusion model, that is, it is composed of multiple models. It can typically include the constructed equivalent circuit model and the neural network model used to predict the model parameters of the equivalent circuit model. The following will be described in conjunction with specific embodiments.

[0062] Specifically, equivalent circuit models typically use at least one parallel RC network to represent the relaxation process. For example, in a first-order equivalent circuit model, an open-circuit voltage (OCV), internal resistance (R0), and an RC parallel network (R1, C1) are commonly used. The open-circuit voltage OCV represents the battery's steady-state voltage, a function of SOC, and the internal resistance R0 represents the battery's ohmic internal resistance (instantaneous impedance from the electrodes, electrolyte, and separator). The RC network (R1, C1) simulates the battery's polarization effects (electrochemical polarization and concentration polarization), representing the voltage relaxation phenomenon during charging and discharging. A second-order equivalent circuit adds another parallel RC network, using two parallel RC networks to simulate electrochemical polarization (charge transfer process) and concentration polarization (lithium-ion diffusion process), respectively. A third-order equivalent circuit builds upon the second-order equivalent circuit by adding another parallel RC network to attempt to capture the battery's more complex internal processes. By identifying the parameters in the circuit model through experimental fitting, the state of charge can be predicted by describing the physical characteristics of the battery.

[0063] However, identifying the model parameters of equivalent circuit models often relies on extensive data computation with poor accuracy. This is especially true when a higher-order equivalent circuit model is desired to better fit battery performance, as excessive parameters can lead to coupling issues. To address these problems, this application proposes a neural network-based approach to predict the model parameters of equivalent circuit models by incorporating the nonlinear fitting capabilities of deep learning. For example, consider a possible third-order equivalent circuit:

[0064]

[0065]

[0066] in, The polarization voltage is approximately zero at the end of the stationary section, while the model parameters to be identified typically include ohmic internal resistance. Polarization internal resistance time constant And charge / discharge data OCV, etc. Of course, the above is illustrated using a third-order equivalent circuit as an example. In fact, depending on the selected equivalent circuit, the parameters of the equivalent circuit model that need to be identified usually vary to some extent. This application does not limit this.

[0067] Building upon the foregoing, in this embodiment, by constructing a neural network model for predicting the model parameters of the equivalent circuit model—that is, a neural network model for predicting the mapping relationship between battery parameter influence factors and equivalent circuit model parameters—the model parameters that need to be identified in the equivalent circuit model can be determined using the initial state parameters of the battery under test. The obtained model parameters will then be used in the equivalent circuit model to further predict the battery's state of charge. For details, please refer to [link to relevant documentation]. Figure 2 , Figure 2 The flowchart illustrating the steps for obtaining predicted voltage by processing state parameters through a fusion model, as provided in this application embodiment, specifically includes steps S210 to S230:

[0068] S210, the initial state parameters are input into the trained neural network model to obtain the first identification parameters.

[0069] In one embodiment of this application, the initial state parameters are input into the trained neural network model. The neural network model processes the initial state parameters by learning the mapping relationship between the battery parameter influence factors and the equivalent circuit model parameters during the training process, thereby predicting the first identification parameters.

[0070] S220, determine the model parameters of the equivalent circuit model based on the first identification parameters, so as to determine the target equivalent circuit model.

[0071] Specifically, in one embodiment of this application, considering that the neural network model can accurately predict the equivalent circuit model parameters through the battery parameter influence factors, the first identified parameters can be directly used as the model parameters of the equivalent circuit model. That is, the identified ohmic internal resistance, polarization internal resistance, time constant, and charge / discharge data OCV, etc., are substituted into the equivalent circuit model to obtain the target equivalent circuit model with given parameters.

[0072] Of course, the aforementioned method is illustrated by assuming that the neural network model has fully learned the mapping relationship between battery parameter influencing factors and equivalent circuit model parameters. In fact, in some embodiments of this application, more accurate parameter identification results can be achieved by introducing traditional parameter identification. For details, please refer to [link to relevant documentation]. Figure 3 , Figure 3 This application provides a flowchart illustrating the steps for determining equivalent circuit model parameters by integrating multi-parameter identification results, specifically including steps S310-S320:

[0073] S310, Based on multiple sets of state parameters of the battery under test, perform parameter identification on the battery under test to obtain the second identification parameter.

[0074] In some embodiments of this application, traditional parameter identification methods can be relied upon, such as least squares fitting, to identify parameters by using multiple sets of state parameters measured in advance experiments, thereby achieving parameter identification of the equivalent circuit model of the battery under test. Specifically, for ease of understanding, taking the Hybrid Pulse Power Characterization (HPPC) method for offline identification of a first-order equivalent circuit model as an example, the battery is fully charged in a constant-temperature environment, i.e., 100% SOC. Then, a short (e.g., 10 seconds / 30 seconds / 60 seconds) constant current discharge pulse is applied. After the pulse ends, the battery is left to rest for a relatively long time (e.g., 40 seconds / several minutes) to allow the voltage to fully relax. Then, a charging pulse of the same duration and rate is applied again, and the battery is left to rest again. Finally, a smaller constant current is used to discharge the battery for a period of time, causing the SOC to decrease at a fixed interval (e.g., 5% or 10%). The aforementioned steps are repeated at new SOC points (e.g., 95%, 90%,...) until the battery is discharged to the cutoff voltage. Furthermore, the entire process described above is repeated at different temperature points (e.g., 0°C, 10°C, 25°C, 40°C). Through this processing, parameters can be identified based on experimentally measured data. For example, the ohmic internal resistance can be determined by the instantaneous voltage jump at the start or end of the pulse and the magnitude of the pulse current; the polarization resistance can be determined by the total polarization voltage change during the entire pulse; and the polarization capacitance can be obtained by fitting the voltage recovery curve during the relaxation phase. Specifically, voltage recovery is an exponential process, and the time constant τ can be calculated through curve fitting, thus yielding the polarization capacitance.

[0075] Of course, the above-mentioned solution is based on the offline identification method of the first-order equivalent circuit model. For the second-order or higher-order equivalent circuit models, multiple time constants can be separated by extending the resting time, and curve fitting can be performed by nonlinear least squares method, etc.

[0076] Of course, besides the solutions mentioned above, parameters can also be identified through the open-circuit voltage (OCV) - state of charge (SOC) curve, or through other online parameter identification methods, such as updating model parameters in real time during battery operation to address the effects of battery aging and temperature changes. For example, as a possible implementation, the transfer function of the battery model can be discretized in the Z-domain using recursive least squares, transforming it into a linear equation. By minimizing the sum of squared prediction errors, the parameter estimates are recursively updated using new sampled data, thereby achieving parameter identification.

[0077] Of course, regardless of the method used to identify the parameters of the battery under test, the obtained second identification parameters can be fused with the first identification parameters provided in this application to determine the model parameters of the final equivalent circuit model. This application does not impose any limitations on the embodiments herein.

[0078] S320, determine the model parameters of the equivalent circuit model based on the first identification parameter and the second identification parameter, so as to determine the target equivalent circuit model.

[0079] In some embodiments of this application, after the parameters of the equivalent circuit model are identified by traditional identification methods, such as online or offline identification, and the second identification parameters are obtained, the two can be fused based on the first identification parameters output by the aforementioned neural network model to provide more accurate model parameters.

[0080] For example, in one embodiment of this application, the sum of the first identification parameter and the second identification parameter can be used as the final model parameter. For instance, taking the aforementioned third-order equivalent circuit as an example, in one embodiment, the second identification parameter obtained using the aforementioned parameter identification method can be denoted as... Based on this, the first identification parameter predicted by the neural network model is: At this point, the model parameters of the final equivalent circuit model can be determined, that is:

[0081]

[0082]

[0083] Of course, in addition to determining the model parameters of the final equivalent circuit model based on the sum of the first and second identification parameters, in some embodiments of this application, the first and second identification parameters can be weighted according to preset weights to obtain the model parameters of the equivalent circuit model. Specifically, the weights here can be determined based on the actual scenario. For example, in one possible implementation, the preset weights are determined by at least one of the confidence level of the first identification parameter output by the neural network model, the battery health status of the battery under test, and the operating condition status of the battery under test.

[0084] For example, taking the confidence level of the first identification parameter output by the neural network model as an example, the neural network model can determine the uncertainty estimate of its predicted value, i.e., the confidence level, based on the operating conditions of the battery under test. This confidence level can be positively correlated with the weight of the first identification parameter. That is, the higher the confidence level, the higher the weight can be set for the first identification parameter.

[0085] Alternatively, the preset weights can be based on the battery health status of the battery under test. Specifically, after the parameters of the equivalent circuit model are identified through offline identification or other identification methods, the parameters will change as the battery ages. Therefore, a decay function related to the battery health status (SOH) or the number of cycles can be established. The lower the SOH, or the more severe the aging, the lower the weight can be assigned to the second identification parameter identified by offline identification methods, while a higher weight can be assigned to the first identification parameter predicted by the neural network.

[0086] Alternatively, the preset weights can be based on the operating conditions of the battery under test, specifically the degree of matching between the operating conditions and those used during offline identification. For example, the difference between the current real-time operating conditions (current, temperature) and the HPPC experimental operating conditions can be determined. The greater the difference, the lower the weight of the second identification parameter can be.

[0087] The aforementioned solution determines the weights of the identification parameters obtained through different methods based on the identification scenario of the battery under test, thereby determining the model parameters of the equivalent circuit model used to determine the state of charge, which can be used more accurately for subsequent prediction of the battery's state of charge.

[0088] Furthermore, in order to make the parameters obtained from model training more consistent with real-world scenarios, some embodiments of this application may introduce constraints to limit the range of parameter values. For example, the obtained parameters should conform to physical laws. That is, after determining the model parameters of the equivalent circuit model based on the first identification parameters and the second identification parameters, the method further includes:

[0089] After determining the model parameters of the equivalent circuit model based on the first identification parameter and the second identification parameter, the method further includes:

[0090] The model parameters of the equivalent circuit model are transformed to a preset range of values ​​according to preset constraints; and

[0091] The target equivalent circuit model is determined based on the model parameters of the transformed equivalent circuit model.

[0092] For example, in one embodiment of this application, the model parameters of the equivalent circuit model should be within a preset value range determined based on the actual application scenario; that is, the preset constraints may include the upper and lower limits of the model parameters. Specifically, in embodiments of this application, the constraints here typically refer to physical constraints that ensure the model parameters conform to physical properties, such as the lower limit of the model parameters, like the polarization resistance in the model parameters. The time constant should be greater than the lower limit, that is:

[0093]

[0094]

[0095] in, This is the lower limit for the polarization internal resistance value. It should be greater than 0. This is the lower bound of the time constant value. It should be greater than 1. In particular, when the model parameters of the equivalent circuit model do not meet the preset constraints, the model parameters of the equivalent circuit model can be transformed to a preset value range according to the preset constraints, thereby determining the target equivalent circuit model based on the transformed model parameters of the equivalent circuit model.

[0096] Of course, it should be noted that in some embodiments of this application, in order for the neural network model in the fusion model, i.e., the target model, to accurately predict the model parameters in the equivalent circuit model, it is often necessary to collect battery test data so that the neural network model can learn the mapping relationship between battery parameter influencing factors and equivalent circuit model parameters. Specifically, the training process of the neural network model usually involves inputting training data into the initial model to obtain prediction results, and determining the model's loss value based on the difference between the prediction results and the labels, thereby updating the parameters in the initial model until a well-trained target model is obtained. Furthermore, based on the aforementioned solution, in one embodiment of this application, it is also considered to determine a neural network model structure that is more suitable for the initial equivalent circuit model based on the order difference of the constructed battery under test. For a clearer understanding of the above, please refer to [link to relevant documentation]. Figure 4 , Figure 4 This application provides a flowchart illustrating the steps for training a target model, specifically including steps S410 to S440:

[0097] S410, determine the initial equivalent circuit model of the battery under test, and determine the initial neural network model based on the order of the initial equivalent circuit model.

[0098] In one embodiment of this application, the initial equivalent circuit model of the battery under test can be determined based on the predicted state of charge (SOC) scenario of the battery under test. Specifically, as can be seen from the foregoing description, using equivalent circuit models of different orders can meet the user's needs for prediction results of different accuracies in SOC. For example, as the user's requirements for the accuracy of the SOC prediction results increase, a higher-order equivalent circuit model can be selected, but the computational load for parameter identification of the equivalent circuit model is usually also higher.

[0099] Having determined the initial equivalent circuit model of the battery under test, such as a first-order, second-order, or third-order equivalent circuit model, in one embodiment of this application, a neural network model is further established based on the order of the initial equivalent circuit model. For example, as a possible implementation of this application, considering that the dynamics of the first-order equivalent circuit model are relatively weak, its model parameters are often strongly correlated with the current SOC, temperature, and current rate. Therefore, a multilayer perceptron (MLP), which is good at learning such complex static nonlinear mappings, can often be selected. Of course, if it is desired to capture some short-term dependencies (such as the small influence of instantaneous current magnitude on internal resistance), a simple recurrent neural network (SNN) or a small gated recurrent unit (Small GRU) can also be selected to further improve the prediction accuracy of the first-order equivalent circuit model. For second-order equivalent circuit models, since more model parameters need to be predicted, and these parameters depend not only on the current state but also on state parameters over a past period, in one embodiment of this application, a gated recurrent unit (GRU) or a long short-term memory network (LSTM) that can effectively capture long-term dependencies in the time series can be selected. Similarly, for third-order equivalent circuit models, a deeper mapping relationship between battery parameter influencing factors and equivalent circuit model parameters is often required. Therefore, a deep LSTM can be used to extract more abstract temporal features layer by layer, or an encoder with a fused attention mechanism can be used. Of course, the solutions provided above are only some possible implementation schemes. In addition to the network models provided above, the network models used in this application can also include feedforward neural network models and convolutional neural network models, and can also include a fusion of any of the aforementioned model structures. This application embodiment does not limit this.

[0100] In particular, it should be noted that the initial equivalent circuit model and the initial neural network model refer to a type of model in which the weights are randomly initialized in the initial model construction. They often need to be trained to obtain more accurate weight values.

[0101] S420, Obtain training state parameters for model training, the training state parameters including the time series of the state of charge.

[0102] Based on the aforementioned solution, by collecting battery test data, i.e. training state parameters, which typically include voltage, temperature, current, and SOC time series data calculated by device ampere-hour integration, for example, by collecting HPPC charge and discharge test data of the battery cell at different temperatures, such as charging / discharging and letting it stand for two hours every 5% SOC, test data can be obtained.

[0103] S430, the training state parameters are input into the initial neural network model and the initial equivalent circuit model to obtain the second predicted voltage corresponding to the training state parameters.

[0104] Building upon the aforementioned scheme, by inputting test data, i.e., training state parameters, into the initially constructed neural network model, the model parameters for the initial equivalent circuit model can be predicted. These parameters are then further input into the initial equivalent circuit model to process the training state parameters and obtain the corresponding predicted voltage. It should be noted that if it is necessary to fuse the second identification parameters obtained offline, during the aforementioned training process, the output of the initial neural network model and the identified second identification parameters also need to be fused in a given manner to ultimately determine the model parameters of the initial equivalent circuit model for subsequent processing to obtain the second predicted voltage.

[0105] Of course, it should be noted that in some embodiments of this application, a range of physical constraint parameters may be further introduced, such as:

[0106]

[0107]

[0108]

[0109] in, Use the Tahn activation function to constrain the model output. In scope Alternatively, you can use other activation functions. This is the lower limit for the polarization internal resistance value. It should be greater than 0. This is the lower bound of the time constant value. It should be greater than 1.

[0110] S440, the initial neural network model is updated according to the second predicted voltage and the target voltage associated with the training state parameters until a trained target model is obtained when the preset training conditions are met.

[0111] After determining the second predicted voltage using the aforementioned method, it is compared with the label associated with the training state parameters, i.e., the target voltage. The loss value of the neural network model can be determined based on the difference between the second predicted voltage and the target voltage. This loss value can be understood as the neural network model's predictive ability for the model parameters of the equivalent circuit model. Based on this loss value, the backpropagation algorithm is used to update and optimize the parameters in the neural network model, enabling it to better capture the mapping relationship between battery parameter influencing factors and equivalent circuit model parameters. The neural network model can then be iteratively updated until the final training of the model is completed. For example, the model can be considered trained when the difference between the second predicted voltage obtained by inputting the training state parameters used for model training into the trained target model and its associated target voltage is less than a certain difference threshold. Of course, using other conditions, such as the number of training rounds, as preset training conditions is also feasible; this embodiment does not impose any limitations on this.

[0112] Based on the aforementioned solutions, in some embodiments of this application, besides determining the loss value of the neural network model based on the difference between the second predicted voltage and the target voltage, this application also provides a flowchart illustrating the steps for determining the model loss value by fusing parameter smoothness loss. Specifically, Figure 5 A flowchart illustrating the steps for determining model loss values ​​using fusion parameter smoothness loss, as provided in this application embodiment, includes steps S510 to S530:

[0113] S510, determine the first model loss value of the initial neural network model based on the difference between the second predicted voltage and the target voltage.

[0114] In one embodiment of this application, the first model loss value can typically be an MSE loss function. Specifically, for predicting voltage... Measured voltage For example, the loss value of the first model is represented as follows:

[0115]

[0116] Of course, in addition to the MSE loss function, other loss functions, such as the log loss function, are also feasible.

[0117] S520, determine the second model loss value of the initial neural network model based on the difference result of the second predicted voltage.

[0118] Furthermore, considering that the battery parameters should change smoothly and continuously with SOC and temperature, without sudden and drastic changes, the parameter sequence predicted by the neural network model should also be smooth. Based on this, in one embodiment of this application, it is also possible to consider differentially processing multiple sequences of second predicted voltages obtained by processing the time series and penalizing excessive or drastic changes.

[0119] S530, the initial neural network model is updated based on the first model loss value and the second model loss value to obtain the updated neural network model.

[0120] After determining the first model loss value and the second model loss value using the aforementioned scheme, the final model loss value can be obtained by weighted fusion of the first model loss value and the second model loss value. This final model loss value can then be used to update and optimize the parameters of the neural network model, thereby improving the prediction performance of the neural network model.

[0121] S230, the initial state parameters are input into the target equivalent circuit model to obtain the first predicted voltage of the battery under test.

[0122] After training the neural network model using the solution provided in any of the aforementioned embodiments, the model parameters of the equivalent circuit model can be predicted using the trained neural network model. Finally, by inputting the initial state parameters into the target equivalent circuit model, the predicted voltage of the voltage to be measured can be predicted.

[0123] For details, please refer to [link / reference]. Figure 6a , Figure 6a A schematic diagram illustrating the effect of data processing based on a fusion model to obtain predicted voltage is shown.

[0124] S130, update the initial state of charge of the battery under test based on the first predicted voltage to obtain the target state of charge of the battery under test.

[0125] Based on the embodiments of this application, after the predicted voltage of the battery under test is obtained by the fusion model, the initial state of charge of the battery under test can be updated by filtering based on the difference between the predicted voltage and the measured voltage, thereby obtaining an accurate target state of charge.

[0126] Specifically, for ease of understanding, in one embodiment, updating the initial state of charge of the battery under test based on the first predicted voltage to obtain the target state of charge of the battery under test can be determined through the following steps:

[0127] Based on the difference between the first predicted voltage and the measured voltage of the battery under test, the initial state of charge of the battery under test is filtered and updated to obtain the target state of charge of the battery under test.

[0128] Specifically, for ease of understanding, in one embodiment of this application, the state of charge (SOC) parameter at the next moment can be updated by calculating the difference between the first predicted voltage and the measured voltage of the battery under test, and by using the Extended Kalman Filter (EKF) algorithm to calculate the EKF gain.

[0129] To clearly understand the complete implementation scheme of state of charge prediction provided in the embodiments of this application, please refer to... Figure 6b , Figure 6b The complete flowchart of state of charge prediction provided in the embodiments of this application is illustrated below.

[0130] In this embodiment, the battery's state parameters, such as temperature, current, and initial charge parameters, are input into the trained fusion model. The model parameters of the equivalent circuit model are then predicted sequentially through the neural network model. The predicted voltage is then input through the equivalent circuit model. By comparing the difference between the predicted voltage and the measured voltage, the EKF gain is calculated using the Kalman filter algorithm. This allows for the dynamic correction of the state of charge (SOC) parameter and serves as the data input for the next moment, enabling subsequent processing and thus completing the full prediction process for the state of charge.

[0131] Furthermore, in one embodiment of this application, the specific training process of the aforementioned fusion model may include:

[0132] (1) Collect cell test data: including voltage, temperature, current and SOC time series data calculated by the device ampere-hour integration.

[0133] (2) Construct a fusion model, which mainly consists of two parts: a deep learning model and an equivalent circuit model. The deep learning model is used to predict the parameters of the equivalent circuit model. The equivalent circuit model parameters and current are input into the equivalent circuit model to calculate the predicted voltage. The predicted equivalent circuit model parameters must match the equivalent circuit model.

[0134] (3) First, establish an equivalent circuit model, including but not limited to first-order, second-order, and third-order equivalent circuit models. Based on the established equivalent circuit model, determine the model parameters, such as the parameters of the third-order equivalent circuit model, including ohmic internal resistance, polarization internal resistance, time constant, and.

[0135] (4) The established deep learning model is mainly used to predict the mapping relationship between lithium battery parameter influencing factors and equivalent circuit model parameters; the lithium battery parameter influencing factors are the model inputs, including but not limited to SOC, temperature and current; the output is the equivalent circuit model parameters, such as the third-order equivalent circuit model parameters mentioned above. The deep learning model includes but is not limited to feedforward neural network models and convolutional neural network models. Physical constraints are added to the model to ensure that the prediction results conform to physical laws.

[0136] (5) Input the equivalent circuit model parameters and current into the equivalent circuit model. The equivalent circuit model parameters can be entirely derived from the prediction results of the deep learning model, or the sum of the prediction results of the deep learning model and the parameters obtained by traditional parameter identification. The latter introduces the parameters calculated by the traditional parameter identification method, which can accelerate the model convergence. Deep learning further captures the nonlinear changes of parameters based on the traditional parameter identification.

[0137] (6) Calculate the loss function based on the predicted voltage and the measured voltage, use the Adam optimizer to make the fusion model converge quickly, and finally train the weights of the deep learning model.

[0138] (7) Given the initial SOC and polarization voltage, input the real-time temperature and current, use the above-trained fusion model to predict the voltage in real time, and use the EKF algorithm to update the SOC prediction value.

[0139] Specifically, in one embodiment, the specific implementation scheme of the above steps includes:

[0140] Collect HPPC charge and discharge test data of the battery cells at different temperatures, such as charging / discharging and letting them stand for two hours after each 5% SOC.

[0141] A third-order equivalent circuit model is established, and the model formula is as follows:

[0142]

[0143]

[0144] Charge / discharge data (OCV) can be measured at the end of each resting period. The polarization voltage is approximately zero at the end of the stationary section. Preliminary parameters are obtained using traditional parameter identification methods, denoted as... .

[0145] Build a deep learning model, such as a feedforward neural network (FNN), taking real-time current, temperature, and device state of charge (SOC) as inputs, and outputting real-time predictions. The parameters predicted by deep learning are summed with the results of traditional parameter identification:

[0146]

[0147]

[0148] Introduce physical constraints to limit the range of parameter values, such as:

[0149]

[0150]

[0151]

[0152] in, Use the Tahn activation function to constrain the model output. Scope Alternatively, you can use other activation functions. This is the lower limit for the polarization internal resistance value. It should be greater than 0. This is the lower bound of the time constant value. It should be greater than 1.

[0153] The summation of the above parameters and the current data are input into the equivalent circuit model to calculate the predicted voltage. and compared with the measured voltage Calculate the loss function, taking the MSE loss function as an example:

[0154]

[0155] The above fusion model is trained using the Adam optimizer until the loss function converges, and the weights of the deep learning model are finally obtained.

[0156] The state of charge prediction method provided in this application processes the initial state parameters of the battery by using a target model of a neural network model that integrates an equivalent circuit model and the model parameters of the equivalent circuit model to obtain the predicted voltage of the battery under test, thereby updating and predicting the state of charge of the battery. It can effectively integrate the nonlinear fitting ability of deep learning and the physical consistency of the equivalent circuit model, and can improve the prediction accuracy while enhancing the generalization ability of the algorithm under different operating conditions and environmental conditions, thereby achieving higher accuracy and stronger generalization ability in state of charge prediction.

[0157] In one embodiment, this application also provides a state of charge prediction device; see [link to relevant documentation]. Figure 7 , Figure 7 A schematic diagram of a state of charge prediction device provided in this application embodiment, specifically including:

[0158] The acquisition module 710 is used to acquire the initial state parameters of the battery under test, including the initial state of charge and battery electrical parameters.

[0159] The prediction module 720 is used to input the initial state parameters into the trained target model for processing to obtain the first predicted voltage of the battery under test, wherein the target model includes an equivalent circuit model and a neural network model for predicting the model parameters of the equivalent circuit model.

[0160] The update module 730 is used to update the initial state of charge of the battery under test based on the first predicted voltage to obtain the target state of charge of the battery under test.

[0161] In some embodiments of this application, the prediction module 720 is further configured to input the initial state parameters into the trained neural network model to obtain the first identification parameters;

[0162] The model parameters of the equivalent circuit model are determined based on the first identification parameters to determine the target equivalent circuit model;

[0163] The initial state parameters are input into the target equivalent circuit model to obtain the first predicted voltage of the battery under test.

[0164] In some embodiments of this application, the prediction module 720 is further configured to perform parameter identification on the battery under test based on multiple sets of state parameters of the battery under test, and obtain a second identification parameter;

[0165] The model parameters of the equivalent circuit model are determined based on the first identification parameter and the second identification parameter to determine the target equivalent circuit model.

[0166] In some embodiments of this application, the prediction module 720 is further configured to weight the first identification parameter and the second identification parameter according to a preset weight to obtain the model parameters of the equivalent circuit model;

[0167] The preset weights are determined by at least one of the confidence level of the first identification parameter output by the neural network model, the battery health status of the battery under test, and the operating condition status of the battery under test.

[0168] In some embodiments of this application, the prediction module 720 is further configured to determine the initial equivalent circuit model of the battery under test, and to determine the initial neural network model based on the order of the initial equivalent circuit model.

[0169] Obtain training state parameters for model training, the training state parameters including a time series of charged states;

[0170] The training state parameters are input into the initial neural network model and the initial equivalent circuit model to obtain the second predicted voltage corresponding to the training state parameters;

[0171] The initial neural network model is updated based on the second predicted voltage and the target voltage associated with the training state parameters until a trained target model is obtained when preset training conditions are met.

[0172] In some embodiments of this application, the prediction module 720 is further configured to determine a first model loss value of the initial neural network model based on the difference between the second predicted voltage and the target voltage;

[0173] The second model loss value of the initial neural network model is determined based on the difference result of the second predicted voltage;

[0174] The initial neural network model is updated based on the first model loss value and the second model loss value to obtain the updated neural network model.

[0175] In some embodiments of this application, the update module 730 is further configured to filter and update the initial state of charge of the battery under test based on the difference between the first predicted voltage and the measured voltage of the battery under test, so as to obtain the target state of charge of the battery under test.

[0176] The state-of-charge (POC) prediction device provided in this application processes the initial state parameters of the battery by using a target model of a neural network model that integrates an equivalent circuit model and the model parameters of the equivalent circuit model to obtain the predicted voltage of the battery under test, thereby updating and predicting the POC of the battery. It can effectively integrate the nonlinear fitting ability of deep learning and the physical consistency of the equivalent circuit model, and can improve the prediction accuracy while enhancing the generalization ability of the algorithm under different operating conditions and environmental conditions, thereby achieving higher accuracy and stronger generalization ability in POC prediction.

[0177] In some embodiments of this application, the method for predicting the state of charge can be implemented as a computer program, which can be implemented in, for example... Figure 8 The computer device shown operates on this device. The computer device's memory can store the various program modules that constitute the method of this state of charge, for example, Figure 7 The acquisition module 710, prediction module 720, and update module 730 are shown. The computer program comprised of these modules causes the processor to execute the steps in the state of charge prediction methods of the various embodiments of this application described in this specification.

[0178] For example, Figure 8 The computer equipment shown can be used as follows Figure 7The acquisition module 710 in the illustrated state of charge prediction method apparatus executes step S110. A computer device can execute step S120 via the prediction module 720. The computer device can execute step S130 via the update module 730. The computer device includes a processor, memory, and a network interface connected via a system bus. The processor of the computer device provides computational and control capabilities. The memory of the computer device includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The network interface of the computer device is used for communication with external computer devices via a network connection. When the computer program is executed by the processor, it implements a state of charge prediction method.

[0179] Those skilled in the art will understand that Figure 8 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0180] In some embodiments of this application, a computer device is provided, including one or more processors; memory; and one or more application programs, wherein the one or more application programs are stored in the memory and configured to be executed by the processor to perform the following steps:

[0181] Obtain the initial state parameters of the battery under test, including the initial state of charge and battery electrical parameters;

[0182] The initial state parameters are input into the trained target model for processing to obtain the first predicted voltage of the battery under test. The target model includes an equivalent circuit model and a neural network model for predicting the model parameters of the equivalent circuit model.

[0183] The initial state of charge of the battery under test is updated based on the first predicted voltage to obtain the target state of charge of the battery under test.

[0184] In some embodiments of this application, a computer-readable storage medium is provided, storing a computer program that is loaded by a processor, causing the processor to perform the following steps:

[0185] Obtain the initial state parameters of the battery under test, including the initial state of charge and battery electrical parameters;

[0186] The initial state parameters are input into the trained target model for processing to obtain the first predicted voltage of the battery under test. The target model includes an equivalent circuit model and a neural network model for predicting the model parameters of the equivalent circuit model.

[0187] The initial state of charge of the battery under test is updated based on the first predicted voltage to obtain the target state of charge of the battery under test.

[0188] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. Any references to memory, storage, databases, or other media used in the embodiments provided in this application 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 storage, etc. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0189] The technical features of the above embodiments can be combined in any way. For the sake of brevity, 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 this specification.

[0190] The above provides a detailed description of a region identification method, apparatus, computer device, and storage medium provided in the embodiments of this application. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for predicting the state of charge, characterized in that, include: The initial state parameters of the battery under test are obtained. The initial state parameters include the initial state of charge and the battery electrical parameters. The initial state of charge is a real-time state of charge used to predict the target state of charge at the next moment. The battery electrical parameters include current and temperature. The initial state parameters are input into the trained target model for processing to obtain the first predicted voltage of the battery under test. The target model includes an equivalent circuit model and a neural network model for predicting the model parameters of the equivalent circuit model. The structure of the neural network model is determined based on the order of the equivalent circuit model. The initial state of charge of the battery under test is updated based on the first predicted voltage to obtain the target state of charge of the battery under test.

2. The method according to claim 1, characterized in that, The step of inputting the initial state parameters into the trained target model for processing to obtain the first predicted voltage of the battery under test includes: The initial state parameters are input into the trained neural network model to obtain the first identification parameters; The model parameters of the equivalent circuit model are determined based on the first identification parameters to determine the target equivalent circuit model; The initial state parameters are input into the target equivalent circuit model to obtain the first predicted voltage of the battery under test.

3. The method according to claim 2, characterized in that, The step of determining the model parameters of the equivalent circuit model based on the first identification parameters includes: Based on multiple sets of state parameters of the battery under test, parameter identification is performed on the battery under test to obtain the second identification parameters; The model parameters of the equivalent circuit model are determined based on the first identification parameter and the second identification parameter to determine the target equivalent circuit model.

4. The method according to claim 3, characterized in that, The step of determining the model parameters of the equivalent circuit model based on the first identification parameter and the second identification parameter includes: The first identification parameter and the second identification parameter are weighted according to preset weights to obtain the model parameters of the equivalent circuit model; The preset weights are determined by at least one of the confidence level of the first identification parameter output by the neural network model, the battery health status of the battery under test, and the operating condition status of the battery under test.

5. The method according to claim 3, characterized in that, After determining the model parameters of the equivalent circuit model based on the first identification parameter and the second identification parameter, the method further includes: The model parameters of the equivalent circuit model are transformed to a preset range of values ​​according to preset constraints; and The target equivalent circuit model is determined based on the model parameters of the transformed equivalent circuit model.

6. The method according to claim 1, characterized in that, The target model is trained through the following steps: Determine the initial equivalent circuit model of the battery under test, and determine the initial neural network model based on the order of the initial equivalent circuit model; Obtain training state parameters for model training, the training state parameters including a time series of charged states; The training state parameters are input into the initial neural network model and the initial equivalent circuit model to obtain the second predicted voltage corresponding to the training state parameters; The initial neural network model is updated based on the second predicted voltage and the target voltage associated with the training state parameters until a trained target model is obtained when preset training conditions are met.

7. The method according to claim 6, characterized in that, The step of updating the initial neural network model based on the target voltage associated with the second predicted voltage and the training state parameters includes: The first model loss value of the initial neural network model is determined based on the difference between the second predicted voltage and the target voltage; The second model loss value of the initial neural network model is determined based on the difference result of the second predicted voltage; The initial neural network model is updated based on the first model loss value and the second model loss value to obtain the updated neural network model.

8. The method according to any one of claims 1 to 7, characterized in that, The step of updating the initial state of charge of the battery under test based on the first predicted voltage to obtain the target state of charge of the battery under test includes: Based on the difference between the first predicted voltage and the measured voltage of the battery under test, the initial state of charge of the battery under test is filtered and updated to obtain the target state of charge of the battery under test.

9. A device for predicting the state of charge, characterized in that, include: The acquisition module is used to acquire the initial state parameters of the battery under test. The initial state parameters include the initial state of charge and battery electrical parameters. The initial state of charge is a real-time state of charge used to predict the target state of charge at the next moment. The battery electrical parameters include current and temperature. The prediction module is used to input the initial state parameters into the trained target model for processing to obtain the first predicted voltage of the battery under test. The target model includes an equivalent circuit model and a neural network model for predicting the model parameters of the equivalent circuit model. The structure of the neural network model is determined based on the order of the equivalent circuit model. An update module is used to update the initial state of charge of the battery under test based on the first predicted voltage to obtain the target state of charge of the battery under test.

10. A computer device, characterized in that, It includes a memory and a processor, wherein the memory stores a computer program that, when executed by the processor, implements the method for predicting the state of charge as described in any one of claims 1 to 8.

11. A computer-readable storage medium, characterized in that, It stores a computer program, which is loaded by a processor to execute the method for predicting the state of charge as described in any one of claims 1 to 8.

Images