A method and device for predicting the state of charge of a lithium battery and a medium

Abstract

The application discloses a kind of battery state of charge prediction method, device and medium of lithium battery, it is related to new energy technology field, the method includes: obtaining the operation data of lithium battery to be detected;The operation data obtained is input into the state of charge prediction model constructed, and the predicted state of charge of lithium battery to be detected predicted by state of charge prediction model is obtained;State of charge prediction model is based on extended Kalman filtering framework to predict state of charge, and in extended Kalman filtering framework, the error covariance matrix of each round of iteration process is updated using Levenberg-Marquardt method, and, through the iteration coefficient of numerical adjustment generated according to the influence of Levenberg-Marquardt method in iteration, each round of iteration is carried out.The application ensures the stability and reliability of state estimation, while effectively preventing the predicted state of charge finally predicted from deviating from actual result.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of new energy, in particular to a battery state of charge prediction method and device for lithium battery and medium. BACKGROUND

[0002] Lithium-ion batteries have shown significant advantages in the field of energy storage due to their excellent energy density, high power, and long life cycle. To ensure the safe and reliable use of the battery, the battery management system (BMS) needs to monitor the battery state in real time, and the state of charge (SOC) is a core monitoring parameter. Accurate estimation of the SOC is of great significance to improve system performance, enhance reliability, and prolong battery life.

[0003] Currently, SOC prediction techniques mainly include four categories: traditional methods, model-driven methods, filtering estimation methods, and data-driven methods. The traditional methods mainly include open-circuit voltage method and ampere-hour integral method, which are simple to operate but have relatively large errors due to the limitations of sensor accuracy. Model-driven methods include equivalent circuit models, electrochemical models, and electrochemical impedance models, which are directly affected by the complexity of the model in terms of calculation and prediction accuracy. The models established are often universal between different types of batteries. Filtering estimation methods, such as Kalman filtering and particle filtering, effectively combine models for SOC prediction, but their prediction accuracy is limited by the accuracy of the model. Data-driven methods, including support vector machines, Gaussian process regression, and neural network algorithms, rely on a large amount of historical data to build network models, which have relatively large calculation amounts, and the prediction accuracy is closely related to the selected training data set, and the generalization performance of the model is difficult to guarantee.

[0004] It can be seen that the existing SOC prediction techniques cannot reflect the true state of charge of lithium batteries, and therefore a new method for predicting the state of charge of lithium batteries is needed. SUMMARY

[0005] Therefore, the embodiments of the present application provide a battery state of charge prediction method and device for lithium battery and medium to solve the problem that the existing SOC prediction techniques cannot reflect the true state of charge of lithium batteries.

[0006] According to a first aspect, the embodiments of the present application provide a battery state of charge prediction method for lithium battery, which comprises:

[0007] obtaining the operating data of the lithium battery to be detected;

[0008] The obtained running data is input into the constructed state of charge prediction model to obtain a prediction state vector of the lithium battery to be detected predicted by the state of charge prediction model; the prediction state vector is a vector composed of the predicted state of charge and the polarization voltage;

[0009] The state of charge prediction model predicts the state of charge based on an extended Kalman filter framework, and uses a Levenberg-Marquardt method to update an error covariance matrix of each round of iteration in the extended Kalman filter framework, and performs iteration of each round by using an iteration coefficient; when the Levenberg-Marquardt method has a positive influence on iteration, the value of the iteration coefficient is reduced in the next round of iteration; and when the Levenberg-Marquardt method has a reverse influence on iteration, the value of the iteration coefficient is increased in the next round of iteration.

[0010] With reference to the first aspect, in a first implementation manner of the first aspect, the state of charge prediction model is constructed by the following steps:

[0011] An actual open-circuit voltage corresponding to each actual state of charge of the sample lithium battery in the discharging process is obtained, and a state of charge-open-circuit voltage curve is determined according to the actual state of charge and the corresponding actual open-circuit voltage;

[0012] The state of charge-open-circuit voltage curve is fitted by using a least square method;

[0013] The extended Kalman filter framework is constructed according to the fitted state of charge-open-circuit voltage curve, the error covariance matrix of each round of iteration in the extended Kalman filter framework is updated by using the Levenberg-Marquardt method, and each round of iteration is performed by using the iteration coefficient.

[0014] With reference to the first implementation manner of the first aspect, in a second implementation manner of the first aspect, the actual open-circuit voltage corresponding to each actual state of charge of the sample lithium battery in the discharging process is obtained, and the state of charge-open-circuit voltage curve is determined according to the actual state of charge and the corresponding actual open-circuit voltage, and specifically includes the following steps:

[0015] The sample lithium battery is left to stand for a preset time to ensure that the sample lithium battery reaches a stable state;

[0016] The sample lithium battery is charged by using a preset charging current, and when the sample lithium battery is charged to a preset state of charge, the sample lithium battery is charged by using a preset charging voltage until the sample lithium battery is charged to a preset current state; the preset charging current in the charging process is a constant current, and the preset charging voltage in the charging process is a constant voltage;

[0017] Discharge the sample lithium battery into a preset current state using a constant discharge current, and determine an actual open circuit voltage value of the sample lithium battery at an actual battery state of charge at each preset charge value of discharge until the sample lithium battery is discharged to the end;

[0018] Determine an open circuit voltage value-battery state of charge curve according to the actual open circuit voltage value and the actual battery state of charge corresponding to the actual open circuit voltage value.

[0019] In combination with the first aspect and the first implementation, in a third implementation of the first aspect, the extended Kalman filter framework is constructed, an error covariance matrix of each round of an iteration process of the extended Kalman filter framework is updated using a Levenberg-Marquardt method, and each round of iteration is performed through an iteration coefficient, and specifically includes:

[0020] The extended Kalman filter framework is constructed, and an initial state vector and an initial covariance matrix are solved according to the battery state of charge-open circuit voltage curve obtained by fitting;

[0021] The error covariance matrix of each round of the iteration process of the extended Kalman filter framework is updated using the Levenberg-Marquardt method, and each round of iteration is performed through the iteration coefficient;

[0022] It is judged whether the iteration stage of the extended Kalman filter satisfies an iteration convergence condition, in the case that the iteration convergence condition is not satisfied, iteration parameters at the next iteration are adjusted and updated until the iteration convergence condition is satisfied and the iteration is stopped;

[0023] In the case that the iteration stage of the extended Kalman filter satisfies the iteration convergence condition, a Jacobian matrix is updated, the error covariance matrix at the iteration stop is updated according to the updated Jacobian matrix, and the updated error covariance matrix is verified, and according to the verification result, the optimal estimation value at the iteration stop or the optimal estimation value before the update is taken as a predicted state vector.

[0024] In combination with the third implementation of the first aspect, in a fourth implementation of the first aspect, the extended Kalman filter framework is constructed, and an initial state vector and an initial covariance matrix are solved according to the battery state of charge-open circuit voltage curve obtained by fitting, and specifically includes:

[0025] The total state expression and the total measurement expression are constructed based on an ampere-hour integral method calculation formula and a discrete state space expression of a first-order resistance-capacitance equivalent circuit model, and the total state expression and the total measurement expression are solved according to the battery state of charge-open circuit voltage curve obtained by fitting;

[0026] The parameters of the extended Kalman filter framework are initialized;

[0027] The initial state vector and the initial covariance matrix are determined;

[0028] The discrete state space expression of the first-order resistance-capacitance equivalent circuit model is obtained by the following steps:

[0029] A first-order resistance-capacitance equivalent circuit is constructed, and a first-order resistance-capacitance equivalent circuit model is constructed based on the first-order resistance-capacitance equivalent circuit; the first-order resistance-capacitance equivalent circuit model comprises a voltage source of an open circuit voltage, an ohmic internal resistance, a transfer internal resistance, and a polarization capacitance, and the voltage source, the ohmic internal resistance, the transfer internal resistance, and the polarization capacitance are connected in parallel and jointly describe the transient response characteristics of the lithium battery;

[0030] The first-order resistance-capacitance equivalent circuit model is discretized to obtain a discrete state expression, and the discrete state space expression is determined based on the discrete state expression.

[0031] In combination with the fourth implementation manner of the first aspect, in a fifth implementation manner of the first aspect, the error covariance matrix of each round of the iterative process of the extended Kalman filter framework is updated using the Levenberg-Marquardt method, and each round of iteration is performed through an iteration coefficient, and specifically includes:

[0032] The iteration update of the extended Kalman filter framework is performed according to the initial covariance matrix and the iteration coefficient, and the error covariance matrix of each round of the iteration stage of the extended Kalman filter is corrected using the Levenberg-Marquardt method;

[0033] The Kalman gain of each round of the iteration stage of the extended Kalman filter is determined according to the error covariance matrix of each round;

[0034] The optimal estimation value of the state of charge and the error covariance matrix of the corresponding iteration round are updated according to the Kalman gain of each round.

[0035] In combination with the third implementation manner of the first aspect, in a sixth implementation manner of the first aspect, it is judged whether the iteration stage of the extended Kalman filter satisfies the iteration convergence condition, and in the case that the iteration convergence condition is not satisfied, the iteration parameter of the next iteration is adjusted and updated until the iteration convergence condition is satisfied and the iteration is stopped, and specifically includes:

[0036] It is judged whether the iteration stage of the extended Kalman filter satisfies the iteration convergence condition;

[0037] In the case that it is determined that the iteration convergence condition is not satisfied, the loss function values of the current iteration round and the last iteration round are obtained;

[0038] In the case that it is determined that the loss function value of the current iteration round is less than the loss function value of the last iteration round, the iteration parameter is reduced and the next round of iteration update is performed until the iteration convergence condition is satisfied and the iteration is stopped;

[0039] When it is determined that the loss function value of the current iteration round is not less than the loss function value of the last iteration round, the iteration parameter is expanded and the next round of iteration update is performed until the iteration convergence condition is met and the iteration is stopped.

[0040] In combination with the third implementation manner of the first aspect, in a seventh implementation manner of the first aspect, when the iteration stage of the extended Kalman filter satisfies the iteration convergence condition, the Jacobian matrix is updated, the error covariance matrix at the iteration stop is updated according to the updated Jacobian matrix, and the updated error covariance matrix is verified, and according to the verification result, the optimal estimation value at the iteration stop or the optimal estimation value before the update is taken as the predicted state vector, and specifically comprising:

[0041] When the iteration stage of the extended Kalman filter satisfies the iteration convergence condition, the Jacobian matrix is updated through a re-approximation method;

[0042] The error covariance matrix at the iteration stop is updated according to the updated Jacobian matrix;

[0043] The traces of the error covariance matrices before and after the update are obtained, and when it is determined that the trace of the updated error covariance matrix is less than the trace of the error covariance matrix before the update, the updated optimal estimation value is taken as the predicted state vector of the sample lithium battery;

[0044] When it is determined that the trace of the updated error covariance matrix is not less than the trace of the error covariance matrix before the update, the optimal estimation value at the iteration stop is taken as the predicted state vector of the sample lithium battery.

[0045] According to the second aspect, an embodiment of the present application further provides a battery state of charge prediction device of a lithium battery, and the device comprises:

[0046] A data acquisition module is configured to acquire operation data of a lithium battery to be detected.

[0047] A state prediction module is configured to input the acquired operation data into a constructed state of charge prediction model to obtain a predicted state vector of the lithium battery to be detected predicted by the state of charge prediction model, wherein the predicted state vector is a vector composed of a predicted state of charge and a polarization voltage.

[0048] The state of charge prediction model predicts the state of charge based on an extended Kalman filter framework, and uses a Levenberg-Marquardt method to update an error covariance matrix of each round of iteration in the extended Kalman filter framework, and each round of iteration is performed by using an iteration coefficient; when the Levenberg-Marquardt method has a positive impact on iteration, the value of the iteration coefficient is reduced in the next round of iteration; and when the Levenberg-Marquardt method has a negative impact on iteration, the value of the iteration coefficient is increased in the next round of iteration.

[0049] According to a third aspect, the embodiments of the present application further provide an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the method for predicting the state of charge of a lithium battery according to any one of the above aspects when executing the program.

[0050] According to a fourth aspect, the embodiments of the present application further provide a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the method for predicting the state of charge of a lithium battery according to any one of the above aspects.

[0051] The method, device and medium for predicting the state of charge of a lithium battery according to the present application predict the predicted state of charge of a lithium battery to be detected by using a completed state of charge prediction model, the state of charge prediction model predicts the state of charge based on an extended Kalman filter framework, and uses a Levenberg-Marquardt method to update an error covariance matrix of each round of iteration in the extended Kalman filter framework, and each round of iteration is performed by using an iteration coefficient adjusted in value according to the impact of the Levenberg-Marquardt method in iteration, more specifically, when the Levenberg-Marquardt method has a positive impact on iteration, i.e., the iteration direction is consistent with the convergence direction, the value of the iteration coefficient is reduced in the next round of iteration; and when the Levenberg-Marquardt method has a negative impact on iteration, i.e., the iteration direction is opposite to the convergence direction, the value of the iteration coefficient is increased in the next round of iteration. The introduction of the Levenberg-Marquardt algorithm into the extended Kalman filter ensures global convergence, and the dynamic adjustment of the value of the iteration coefficient avoids the local optimal solution trap that may be encountered by the traditional method. Compared with the traditional extended Kalman filter, the improved extended Kalman filter framework used by the state of charge prediction model solves the convergence problem, pursues accurate estimation of the covariance matrix, and pursues a better Kalman gain, thereby ensuring the stability and reliability of state estimation while effectively preventing the predicted state of charge from deviating from the actual result. BRIEF DESCRIPTION OF DRAWINGS

[0052] The features and advantages of the present application will be more clearly understood through reference to the following drawings, which are presented as exemplary and should not be construed as limiting the application, in which:

[0053] Figure 1 A flowchart of a method for predicting a state of charge of a lithium battery is shown.

[0054] Figure 2 A structure diagram of a first-order RC equivalent circuit model constructed in the method for predicting a state of charge of a lithium battery is shown.

[0055] Figure 3 A diagram of an OCV-SOC curve fitted in the method for predicting a state of charge of a lithium battery is shown.

[0056] Figure 4 A diagram of current data under a dynamic stability test (DST) working condition simulated in the method for predicting a state of charge of a lithium battery is shown.

[0057] Figure 5 A diagram of voltage data under a dynamic stability test (DST) working condition simulated in the method for predicting a state of charge of a lithium battery is shown.

[0058] Figure 6 A diagram of experimental results under a dynamic stability test (DST) working condition simulated in the method for predicting a state of charge of a lithium battery is shown.

[0059] Figure 7 A diagram of experimental results under a US06 working condition simulated in the method for predicting a state of charge of a lithium battery is shown.

[0060] Figure 8 A structure diagram of a device for predicting a state of charge of a lithium battery is shown.

[0061] Figure 9 A hardware structure diagram of an electronic device is shown. DETAILED DESCRIPTION

[0062] To make the objectives, technical solutions, and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some but not all of the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0063] With the development of related technologies, new energy vehicles represented by pure electric vehicles, hybrid electric vehicles and fuel cell vehicles have developed rapidly. Lithium-ion batteries have shown significant advantages in energy storage due to their excellent energy density, high power and long life cycle. To ensure the safe and reliable use of batteries, BMS needs to monitor the battery status in real time. In actual use, the SOC of lithium batteries will change with the battery capacity, internal resistance and capacitance parameters, temperature, discharge rate and aging degree. As a core monitoring parameter, the accurate estimation of SOC is of great significance to improve system performance, enhance reliability and prolong battery life.

[0064] Currently, SOC prediction techniques mainly include four categories: traditional methods, model-driven methods, filtering estimation methods and data-driven methods.

[0065] Among them, the traditional method mainly includes open circuit voltage method and ampere-hour integral method, these two methods are simple to operate, but are limited by the accuracy of the sensor, and the error is relatively large; Model-driven method covers equivalent circuit model, electrochemical model and electrochemical impedance model, equivalent circuit model is used to describe and simulate the dynamic characteristics of the battery, which regards the battery as a two-port network, and uses voltage source, resistance, capacitance and other devices to form a circuit to simulate the internal characteristics of the battery, electrochemical model calculates the terminal voltage and SOC of the battery according to the electrochemical reaction process, which is a battery model based on porous electrode and solution concentration theory, the electrochemical model mainly reflects the internal chemical reaction mechanism of the battery, and has high model accuracy, but it is difficult to determine all the parameters, and has great calculation complexity and time-consuming, the electrochemical impedance model can accurately describe the battery characteristics, but the matching process is difficult, complex and not intuitive in practical application, and the impedance model is only useful at a specific SOC and temperature, and cannot predict the direct current reaction and battery running time, the calculation amount and prediction accuracy of these models are directly affected by the complexity of the model, and the model established is often universal between different types of batteries; Filter estimation method, such as Kalman filter and particle filter, effectively combines model for SOC prediction, but its prediction accuracy is also limited by the accuracy of the model; Data-driven method, including support vector machine, Gaussian process regression, neural network algorithm, etc., the SOC prediction method based on data-driven does not need to consider the complex chemical reaction mechanism inside the battery, but establishes and trains the mapping relationship model between external characteristics parameters such as current, voltage and temperature and SOC based on a large amount of battery experimental test data, then relies on a large amount of historical data to construct the network model, the calculation amount is relatively large, the neural network model is a typical representative, this method has high fitting ability while ignoring the details of the internal chemical reaction of the battery, and is suitable for SOC estimation of various power batteries, and has high estimation accuracy, but a large amount of data is required for training, and the calculation amount is large, in practical application, a high-performance chip must be equipped, which increases the cost of BMS, the prediction accuracy is closely related to the selected training data set, and the generalization performance of the model is difficult to guarantee.

[0066] In order to realize dynamic SOC estimation, the filter and observer are often combined with the battery model to form a model-based SOC prediction method for SOC prediction. The commonly used filters and observers include Kalman filter, particle filter, H∞ filter and other state observers (such as sliding mode observer). Such estimation method uses a closed-loop structure, which continuously corrects the SOC estimation value, so that the SOC estimation value continuously approaches the true value, and the algorithm has certain robustness.

[0067] However, the prediction accuracy of the SOC prediction method combining filter and observer with battery model is often limited by covariance matrix estimation, which is difficult to reflect the true battery state of charge of lithium battery.

[0068] It can be seen that the existing SOC prediction technology is difficult to reflect the true state of charge of the lithium battery. In view of the above problems, it is necessary to provide a new method for predicting the state of charge of a lithium battery.

[0069] In order to solve the above problems, a method for predicting the state of charge of a lithium battery is provided in the embodiment, which aims to obtain a better Kalman gain while reflecting the true state of the lithium battery, thereby improving the accuracy of the state of charge prediction. The method for predicting the state of charge of a lithium battery according to the embodiment of the application can be used in electronic devices, including but not limited to computers, mobile terminals, etc. Figure 1 is a flowchart of the method for predicting the state of charge of a lithium battery according to the embodiment of the application, as shown in Figure 1 The method can include the following steps:

[0070] S10, obtaining the operating data of the lithium battery to be detected, in the embodiment, the operating data includes but is not limited to the open-circuit voltage, rated capacity, terminal voltage, current flowing through, sampling time interval, etc. of the lithium battery to be detected.

[0071] S20, inputting the obtained operating data into the constructed state of charge prediction model to obtain a predicted state vector of the lithium battery to be detected predicted by the state of charge prediction model, the predicted state vector being a vector composed of the predicted state of charge and the polarization voltage, specifically, the predicted state vector being a two-row-one-column matrix composed of the predicted state of charge and the polarization voltage.

[0072] The state of charge prediction model is based on an extended Kalman filter framework to predict the state of charge, and uses a Levenberg-Marquardt (L-M) method to update the error covariance matrix of each round of iteration in the extended Kalman filter framework, and each round of iteration is performed by adjusting the iteration coefficient according to the influence generated by the L-M method in the iteration, more specifically, when the L-M method produces a positive influence in the iteration, i.e. the iteration direction is consistent with the convergence direction, the value of the iteration coefficient is reduced in the next round of iteration, and when the L-M method produces a negative influence in the iteration, i.e. the iteration direction is opposite to the convergence direction, the value of the iteration coefficient is increased in the next round of iteration.

[0073] Compared with the traditional iterative extended Kalman filter, the improved extended Kalman filter framework used by the state of charge prediction model solves the convergence problem. The introduction of the L-M algorithm iterative extended Kalman filter can guarantee global convergence to pursue accurate estimation of the covariance matrix, and through the dynamic change of the iteration coefficient, a better Kalman gain is pursued. The dynamic adjustment of the value of the parameter iteration coefficient avoids the local optimal solution trap that the traditional method may encounter, ensures the stability and reliability of state estimation, and effectively prevents the predicted state of charge from deviating from the actual result.

[0074] In the embodiment, the state of charge prediction model is constructed by the following steps:

[0075] S30, the actual open circuit voltage corresponding to each actual battery state of charge during the discharging process of the sample lithium battery is obtained, and the battery state of charge-open circuit voltage curve (SOC-OCV curve) is determined according to the actual battery state of charge and the corresponding actual open circuit voltage.

[0076] In the embodiment, step S30 specifically comprises:

[0077] S31, the sample lithium battery is left for a preset time to ensure that the internal state of the sample lithium battery reaches a stable state. This step is to leave the sample lithium battery.

[0078] In the embodiment, the preset time can be configured by the user, for example, the preset time is set to 1h.

[0079] S32, the sample lithium battery is charged using a preset charging current, and when the sample lithium battery is charged to a preset state of charge, the sample lithium battery is charged using a preset charging voltage until the sample lithium battery is charged to a preset current state. This step is to charge the sample lithium battery.

[0080] In the embodiment, the preset state of charge is that the lithium battery is close to the full state, for example, 90% SOC; and the preset current state is that the current of the lithium battery decreases to a certain threshold to ensure that the battery is fully charged.

[0081] In order to ensure the controllability and stability during the initial charging process, the preset charging current of the charging process is a constant current, and the preset charging voltage of the charging process is a constant voltage, that is, in this step, the sample lithium battery is first charged using a preset constant current, and then charged using a preset constant voltage.

[0082] S33, the sample lithium battery in the preset current state is discharged using a constant discharge current, and the actual open circuit voltage value of the sample lithium battery at the actual battery state of charge is determined when each preset state of charge is discharged, that is , until the end of the discharge of the sample lithium battery. This step is to discharge the sample lithium battery.

[0083] In this embodiment, the discharge is carried out at a constant current, and the corresponding actual open circuit voltage value is recorded once every time the sample lithium battery is discharged by a preset state of charge value, for example, 2% SOC. In this way, the corresponding actual open circuit voltage value can be recorded once every time the SOC of the sample lithium battery reaches an actual SOC value during the discharge process. Considering the calculation accuracy, the SOC reduction amount during the discharge of the sample lithium battery is accurately calculated by the ampere-hour integration method.

[0084] S34, determining the open circuit voltage value-battery state of charge curve according to the actual open circuit voltage value and the actual battery state of charge corresponding to the actual open circuit voltage value. This step will plot the recorded OCV value and the corresponding SOC value into a battery state of charge-open circuit voltage curve (SOC-OCV curve).

[0085] In this embodiment, in order to ensure that the plotted OCV-SOC curve can be applied to the subsequent prediction of the SOC of the battery, the standing, charging, discharging and recording processes are repeated multiple times, and after each discharge, the lithium battery to be tested is allowed to stand for a long enough time, and then the open circuit voltage value is measured, that is, steps S31 to S33 are repeated.

[0086] S40, fitting the battery state of charge-open circuit voltage curve (SOC-OCV curve) using the least squares method.

[0087] Among them, the x-axis of the recorded SOC-OCV curve is the SOC value, and the y-axis is the OCV value. The functional relationship between SOC and OCV can be obtained by data fitting method.

[0088] Since the fitting effect of the SOC-OCV curve is affected by the order of the polynomial, after completing the fitting, the root mean squared error (Root Mean Squared Error, RMSE), the coefficient of determination (Coefficient of Determination, R-square), and the sum of squares due to error (Sum of Squares due to Error, SSE) are used to evaluate the pros and cons of the curve fitting effect; if the fitting effect is not good, the order of the polynomial is increased to improve the fitting effect of the curve, until the evaluation index result has no obvious improvement or exceeds the set threshold, and the evaluation of the fitting effect is shown in Table 1.

[0089] Table 1 Polynomial fitting results

[0090]

[0091] The power of the function has a positive correlation with the fitting accuracy, but the larger the power is, the larger the calculation amount is; when the order is low, the curve cannot better reflect the relationship between SOC and OCV, but when the order is high, the curve will oscillate, thereby affecting the accuracy. Please refer to Figure 2 In this embodiment, a 9th order polynomial is used for fitting the SOC-OCV curve, and accordingly, the function relationship between SOC and OCV obtained is:

[0092] OCV(SOC)=2952.557SOC 9 -13998.9416SOC 8 +28345.1537SSOC 7 -32005.7919SOC 6 +22068.5577SOC 5 -9558.2579SOC 4 +2573.9902SOC 3 -411.0355SOC 2 +35.3359SOC+1.9273

[0093] S50, construct an extended Kalman filter framework, use the L-M method to update the error covariance matrix of each round of iteration process of the extended Kalman filter framework, and perform iteration of each round through an iteration coefficient.

[0094] In this embodiment, step S50 specifically includes:

[0095] S51, construct an extended Kalman filter framework, and solve an initial state vector and an initial covariance matrix according to the fitted battery state of charge-open circuit voltage curve. Lay a foundation for solving the optimal estimation value of the state of charge later.

[0096] In this embodiment, step S51 specifically includes:

[0097] S511, based on the ampere-hour integral method calculation formula and the discrete state space expression of the first order resistor-capacitance (Resistor-Capacitance) equivalent circuit model, construct a total state expression That is, the total state expression equation (calculation formula) and the total measurement expression That is, the total measurement equation (calculation formula), and solve the total state expression according to the fitted SOC-OCV curve And the total measurement expression Specifically:

[0098]

[0099] Wherein, Indicates Battery state of charge of the sample lithium battery at time t; represents Open-circuit voltage corresponding to the SOC value of the sample lithium battery at time t; represents Polarization voltage of the sample lithium battery at time t; represents represents represents represents represents represents 1) Coulombic efficiency, which can be used to represent the ratio of the actual available electric quantity to the theoretical electric quantity in the charging and discharging process of the sample lithium battery represents , that is, an exponential function with as the base; represents Terminal voltage of the sample lithium battery at time t; represents Dynamic current of the sample lithium battery at time t (that is, at a corresponding discrete time), a positive value representing that the lithium battery is in a charging state, and a negative value representing that the lithium battery is discharging; represents represents

[0100] The total state expression and the total measurement expression are constructed, and each variable parameter is also constructed, specifically:

[0101]

[0102]

[0103]

[0104]

[0105] This process selects the state variable as ; represents a preset state transition function (matrix); represents a preset control input matrix; represents the slope of the OCV-SOC curve at the predicted SOC; represents the partial derivative of the terminal voltage with respect to the polarization voltage, taking the sensitivity of the terminal voltage with respect to the polarization voltage as a fixed value ; represents the polarization voltage at time ; the Jacobian matrix predicted at time t, which is expressed by the total measurement matrix predicted at time t; is the state vector predicted at time t; the state vector predicted at time t, which is expressed by the total state matrix predicted at time t.

[0106] In this embodiment, the discrete state space expression of the first-order RC equivalent circuit model is obtained by the following steps:

[0107] A10, first construct a first-order resistor-capacitance (Resistor-Capacitance) equivalent circuit, based on the first-order RC equivalent circuit, which is composed of an RC network structure and a resistor in series, the core components of the first-order RC equivalent circuit model, including an open-circuit voltage source , an ohmic internal resistance , a transfer internal resistance and a polarization capacitance , these components are connected in parallel to each other, which collectively describes the transient response characteristics of the lithium battery.

[0108] As Figure 3 shown in the first-order RC equivalent circuit model, denotes the terminal voltage at time t (this parameter reflects the voltage state of the battery at any time t), denotes the ohmic internal resistance, denotes the transfer internal resistance, denotes the polarization capacitance. According to Kirchhoff Voltage Law (KVL) and Kirchhoff's Current Law (KCL).

[0109] At this time, for the sample lithium battery, the terminal voltage at time t and the dynamic current can be expressed as:

[0110]

[0111]

[0112] where, denotes the open-circuit voltage at time t; denotes the voltage value of the battery when no current flows through the battery, which is usually a complex nonlinear function.

[0113] A20、For lithium battery, SOC of the battery is calculated as the ratio of the remaining battery capacity to the rated battery capacity, based on which, the first-order RC equivalent circuit model based on the first-order RC equivalent circuit is discretized to obtain a discrete state expression, and a discrete state space expression is determined based on the discrete state expression, wherein the discrete state expression is also a discrete state equation (formula) as follows:

[0114]

[0115] The discrete state space expression is also a discrete state equation (formula) as follows:

[0116]

[0117] The voltage jump caused by the current transient in the lithium battery is caused by the ohmic internal resistance of the lithium battery , specifically:

[0118]

[0119] wherein, represents the voltage value at a moment before the lithium battery starts discharging; represents the voltage value when the lithium battery starts discharging; represents the voltage value when the lithium battery finishes discharging; represents the voltage value at a moment after the lithium battery finishes discharging; represents the discharging current during the discharging process of the lithium battery.

[0120] During the discharging process, the voltage jump of the lithium battery is caused by the ohmic internal resistance , which is manifested as a transient voltage drop. The slow decrease of the voltage is dominated by the polarization effect, and the dynamic characteristics are related to the parallel connection of the transfer internal resistance and the polarization capacitance in the first-order RC equivalent circuit (i.e. the response under step current excitation); after the discharging is finished, the polarization effect of the battery disappears, and the voltage of the lithium battery gradually rises, which is related to the parallel connection of the transfer internal resistance and the polarization capacitance in the first-order RC equivalent circuit, at this time, the lithium battery is in a zero-input response state.

[0121] According to the exponential fitting of the voltage rising process at the end of discharging, the specific values of the transfer internal resistance and the polarization capacitance of the first-order RC equivalent circuit can be obtained, and the fitting form of the two parameters is as follows:

[0122]

[0123] wherein, represents a time constant, which is used to reflect the decay rate of the polarization voltage, ; assuming the initial value of the polarization voltage , the transfer resistance is calculated by the initial value of the polarization voltage and the discharge current , specifically:

[0124]

[0125] The polarization capacitance is calculated by the time constant obtained by fitting, , specifically:

[0126]

[0127] S512, initialize the parameters of the extended Kalman filter framework, that is, set the initial values of the Kalman filter framework parameters , , , and , wherein the parameters , , and The parameters of the three parameters are described above, represent the initial covariance matrix, these parameters are involved in the extended Kalman filter process, and an effective initial value is set for the state of charge prediction model to avoid filter divergence.

[0128] S513, determine the initial state vector and the initial covariance matrix, in this embodiment, the state vector is updated by the following calculation formula, so as to predict the initial state vector of the iteration process, specifically:

[0129]

[0130] wherein, represents the initial state vector, which is also the predicted state vector at time to time, it should be noted that the initial time of the iteration process is usually set to 0 time, so the initial state vector is ; represents the predicted state vector at time.

[0131] The specific values of the parameters and can be obtained through the previous steps, and will not be described in more detail here.

[0132] Then, the initial covariance matrix corresponding to the initial state vector is updated using the following calculation formula. Specifically:

[0133]

[0134] in, Denotes the initial covariance matrix, which is also Always The error covariance matrix at time 1 is given. It should be noted that the initial time step in the iteration process is usually set to time 0; therefore, the initial covariance matrix is... ; express Time (the starting point of the recursion) The error covariance matrix at time ( ); This represents the transpose of a predefined state transition function (matrix); The covariance represents the process noise.

[0135] S52. Use the LM method to update the error covariance moment of each round of the extended Kalman filter framework iteration process and perform each round of iteration through the iteration coefficients.

[0136] In this embodiment, step S52 specifically includes:

[0137] S521. Based on the initial covariance matrix and iteration coefficients, the extended Kalman filter framework is iteratively updated, and the LM method is used to correct the error covariance matrix of each round of the extended Kalman filter iteration stage, i.e., the prediction stage. Specifically, the predicted state vector and error covariance matrix in each iteration are updated using the following formula:

[0138]

[0139] in, Indicates the first During the round of iteration Always The predicted state vector at time step; Indicates the first During the round of iteration Always The error covariance matrix of the predicted state vector at time t.

[0140] The correction process for the error covariance matrix is ​​as follows:

[0141]

[0142] in, Indicates the first The error covariance matrix after the round of iterations, that is, the result of the error covariance matrix after correction; This represents the iteration coefficient. The iteration coefficient is introduced during the correction of the error covariance matrix. The introduction of this ensures the global convergence of the extended Kalman filter.

[0143] When the value of is large, it means that the iterative results of the extended Kalman filter with the LM method are closer to the results of gradient descent; conversely, When the value is small, it means that the iterative results of the extended Kalman filter with the LM method are closer to the calculation results obtained by the Gauss-Newton method.

[0144] In this embodiment, the actual calculation process will... When the value is set small, the iteration result will be consistent with the convergence direction. For example, the parameter can be set to a smaller value. The initial value is set to 0.15.

[0145] S522. Determine the Kalman gain for each round of the extended Kalman filter iteration stage based on the error covariance matrix of each round. For example, determine the Kalman gain for the first round of the extended Kalman filter iteration stage based on the first round, i.e., the initial covariance matrix. Specifically:

[0146]

[0147]

[0148]

[0149] in, Indicates the first During the round of iteration Always The covariance between the predicted state vector at time t and the observed predicted value; This represents the predicted state vector used in the calculation process; This represents the observed and predicted values ​​used in the calculation process. Indicates the first During the round of iteration Always Observe the covariance of the predicted value at any time; express Always Jacobian matrix at time; Representing the Jacobian matrix Transpose of; Indicates the first i Kalman gain during round iteration; The noise variance of the Jacobian matrix, that is, the total measurement expression The noise variance of the measurement value obtained, that is, the observation data.

[0150] In the embodiment, the observation function The total measurement expression Remove the noise part, that is, The predicted state vector is substituted into the observation function The observation prediction value corresponding to the time point of the predicted state vector is obtained.

[0151] It can be understood that, similar to the predicted state vector, the real state vector is a two-row-one-column matrix composed of a real state of charge and a polarization voltage.

[0152] S523, update the optimal estimation value of the state of charge and the error covariance matrix of each iteration according to the Kalman gain of each round, specifically:

[0153]

[0154]

[0155] Wherein, The optimal estimation value in the th iteration process is represented by The actual observation value is represented by The total measurement expression is represented by The Jacobian matrix at The time point to The time point, that is, the current time.

[0156] In fact, The actual observation value is obtained by substituting the real state vector into the total measurement expression .

[0157] S53, determine whether the extended Kalman filter iteration stage meets the iteration convergence condition, and adjust the iteration parameters and update in the next iteration under the condition that the iteration convergence condition is not met, until the iteration convergence condition is met and the iteration is stopped.

[0158] In the embodiment, the iteration convergence condition is that the iteration error between adjacent iterations is less than a preset error or reaches the maximum iteration round. The preset error represents the error between adjacent iterations. If the value is less than a certain value, it can be considered that the iteration result has converged to a local minimum point, at which time the iteration can be terminated, for example, the parameter is set to 0.0001; or the current iteration round has reached the maximum number of iterations, which can be set to 20.

[0159] The calculation method of the iteration error is:

[0160]

[0161] Secondly, by comparing the loss function values after the adjacent two iteration rounds, it can be judged whether the L-M method has a positive impact on the iteration. The calculation formula of the loss function is:

[0162]

[0163] wherein, The loss function value of the first iteration round is .

[0164] The loss function values after the adjacent two iterations are and respectively. Next, the values of the two parameters and are compared. When , it indicates that the iteration is convergent, that is, the iteration direction is correct. At this time, the value of is reduced and the reduced is taken as the new iteration parameter. Then the next iteration round is updated with the new iteration parameter. When , it indicates that the iteration direction is opposite to the convergent trend. At this time, the value of is enlarged and the enlarged is taken as the new iteration parameter. Then the next iteration round is updated with the new iteration parameter. That is, when the positive impact is generated, the iteration parameter is reduced and the next iteration round is updated. When the reverse impact is generated, the iteration parameter is enlarged and the next iteration round is updated.

[0165] Preferably, when the value of is reduced, the value of is reduced to 0.5 . When the value of is enlarged, the value of is enlarged to 4 . By dynamically adjusting the value of the parameter , the local optimal solution trap in the traditional method can be avoided.

[0166] In the embodiment, the step S53 specifically comprises:

[0167] S531, judging whether the extended Kalman filter iteration stage satisfies the iteration convergence condition.

[0168] S532, in the case of determining that the iteration convergence condition is not met, obtaining the loss function value of the current iteration round and the last iteration round, that is, obtaining and .

[0169] S533, in the case of determining that the loss function value of the current iteration round is less than the loss function value of the last iteration round, reducing the iteration parameter and performing the next round of iteration update until the iteration convergence condition is met and the iteration is stopped.

[0170] S534, in the case of determining that the loss function value of the current iteration round is not less than the loss function value of the last iteration round, expanding the iteration parameter and performing the next round of iteration update until the iteration convergence condition is met and the iteration is stopped.

[0171] S54, in the case of determining that the extended Kalman filter iteration stage meets the iteration convergence condition, updating the Jacobian matrix, updating the error covariance matrix at the iteration stop according to the updated Jacobian matrix, and verifying the updated error covariance matrix, and according to the verification result, the optimal estimation value at the iteration stop or the optimal estimation value before updating will be used as the predicted state vector.

[0172] Through the re-calibration of the error covariance matrix and the verification of the re-calibration result, the approximation process can be re-processed after updating the state vector. In this process, by comparing the observation data and the model prediction result, the existing deviation is accurately identified and corrected, and then the covariance matrix estimation is updated and optimized. The verification of the re-calibration result of the error covariance matrix further ensures the convergence of the state of charge prediction model, avoids divergence, and thus improves the accuracy of the predicted state of charge.

[0173] S541, in the case of determining that the extended Kalman filter iteration stage meets the iteration convergence condition, the Jacobian matrix is updated by re-approximation method, specifically:

[0174]

[0175] wherein, represents the total measurement expression at the moment the predicted Jacobian matrix for the total state expression at the moment the partial derivative of the predicted state vector.

[0176] S542, updating the error covariance matrix at the iteration stop according to the updated Jacobian matrix, specifically:

[0177]

[0178]

[0179]

[0180] in, This indicates when the iteration stops. The covariance between the predicted state vector at time t and the observed predicted value; This indicates when the iteration stops. The transpose of the Jacobian matrix at time t; This indicates when the iteration stops. Observe the covariance of the predicted value at any given time; express Transpose of; ; Indicates the Kalman gain at the point where the iteration stops; express Transpose of; This represents the updated error covariance matrix, i.e. The error covariance matrix of the predicted state vector at time step [time] can be seen from... From the previous version get, This represents the error covariance matrix at the point where the iteration stops, i.e., the matrix at which the iteration stops. Always The error covariance matrix of the predicted state vector at time t.

[0181] S543. Obtain the trace of the error covariance matrix before and after the update. If it is determined that the trace of the updated error covariance matrix is ​​less than the trace of the original error covariance matrix, use the updated optimal estimate as the predicted state vector of the sample lithium battery.

[0182] In this embodiment, the trace of the error covariance matrix before and after the update is first calculated. Specifically, if... Then The result after the update Corresponding ( The updated optimal estimate is used as the predicted state vector of the sample lithium battery.

[0183] S544. If the trace of the updated error covariance matrix is ​​not less than the trace of the original error covariance matrix, the optimal estimate at the time of iteration termination is used as the predicted state vector of the sample lithium battery.

[0184] Specific if Then Corresponding ( The optimal estimate at the time of iteration termination is used as the predicted state of charge (SOC) of the sample lithium battery. denotes the solution of the trace-free part operation, i.e., finding the trace.

[0185] In the embodiment, after updating the prediction state vector, whether the optimal estimation value before or after updating is taken as the prediction state vector of the sample lithium battery is determined according to the trace of the error covariance matrix, so that the re-calibration of the prediction state vector is performed, the accurate estimation of the battery state of charge is realized, the convergence of the model is further ensured, divergence is avoided, and the accuracy of the prediction state vector obtained by prediction is improved

[0186] The battery state of charge prediction method of the lithium battery provided by the application obtains the state of charge of the lithium battery to be detected through the completed state of charge prediction model, the state of charge prediction model is based on the extended Kalman filtering framework to predict the state of charge, and the L-M method is used to update the error covariance matrix of each round of iteration in the extended Kalman filtering framework, and each round of iteration is performed through the iteration coefficient adjusted in value according to the influence generated in iteration of the L-M method, and more specifically, when the L-M method generates a positive influence in iteration, that is, the iteration direction is consistent with the convergence direction, the value of the iteration coefficient is reduced in the next round of iteration, and when the L-M method generates a reverse influence in iteration, that is, the iteration direction is opposite to the convergence direction, the value of the iteration coefficient is increased in the next round of iteration, and the introduction of the L-M algorithm iteration extended Kalman filter can guarantee the global convergence, and the dynamic adjustment of the value of the iteration coefficient avoids the local optimal solution trap that may be encountered in the traditional method. Compared with the traditional iteration extended Kalman filter, the improved extended Kalman filtering framework used by the state of charge prediction model solves the convergence problem, pursues accurate estimation of the covariance matrix, and pursues better Kalman gain, so that the stability and reliability of state estimation are ensured, and the predicted state of charge deviating from the actual result is effectively prevented.

[0187] Based on the completed state of charge prediction model, the battery model A in the laboratory environment is verified by charging and discharging. Specifically, please refer to Figure 4 and Figure 5 The units of the horizontal coordinates in the two graphs are seconds (S), and the units of the vertical coordinates are amperes (A) and volts (V), respectively. The data set used in the verification process is the battery voltage and current data obtained by simulating the DST and US06 working conditions of the battery model A in the laboratory; please refer to Figure 6 and Figure 7 , wherein Figure 6 shows the experimental result graph in the simulation of the dynamic stability test (DST) working condition in the battery state of charge prediction method of the lithium battery provided by the application, Figure 7The present application provides a lithium battery state of charge prediction method, and the experimental results of simulating the US06 working condition are shown in the following two figures, the horizontal coordinate unit is second (S), and the vertical coordinate unit is the percentage of the specific SOC, and it can be seen that there is a certain deviation between the SOC calculated by the ampere-hour integration method at the second level and the SOC calculated by the ampere-hour integration method at the millisecond level. In order to more accurately evaluate, it is assumed that the ampere-hour integration calculation result at the millisecond level is closer to the true value, and under this premise, the SOC value estimated by the improved extended Kalman filtering algorithm shows high consistency with the assumed true value, and the error range is relatively small. The results show that the constructed state of charge prediction model has a significant advantage in improving the SOC estimation accuracy.

[0188] The lithium battery state of charge prediction device provided by the embodiment of the present application is described below, and the lithium battery state of charge prediction device described below can be correspondingly referred to the lithium battery state of charge prediction method described above.

[0189] In order to solve the above problems, a lithium battery state of charge prediction device is provided in the embodiment, which aims to obtain a better Kalman gain and reflect the true state of the lithium battery, thereby improving the accuracy of the battery state of charge prediction. Figure 8 The structure diagram of the lithium battery state of charge prediction method according to the embodiment of the present application is shown in FIG. 1, and the device can include: Figure 8

[0190] The data acquisition module 10 is used to acquire the running data of the lithium battery to be detected, and in the embodiment, the running data includes but is not limited to the polarization voltage, rated capacity, terminal voltage, current, sampling time interval and the like of the lithium battery to be detected.

[0191] The state prediction module 20 is used to input the acquired running data into the constructed state of charge prediction model, so as to obtain the prediction state vector of the lithium battery to be detected predicted by the state of charge prediction model, and the prediction state vector is a vector composed of the predicted state of charge and the polarization voltage. Specifically, the prediction state vector is a two-row-one-column matrix composed of the predicted state of charge and the polarization voltage.

[0192] ​The battery state of charge prediction device of the lithium battery of the application predicts the predicted state of charge of the lithium battery to be detected through the completed state of charge prediction model, the state of charge prediction model predicts the state of charge based on the extended Kalman filtering framework, and the error covariance matrix of each round of iteration is updated in the extended Kalman filtering framework using the L-M method, and each round of iteration is performed through the iteration coefficient adjusted in value according to the influence generated in the iteration of the L-M method, and more specifically, when the L-M method generates a positive influence in the iteration, that is, the iteration direction is consistent with the convergence direction, the value of the iteration coefficient is reduced in the next round of iteration, and when the L-M method generates a reverse influence in the iteration, that is, the iteration direction is opposite to the convergence direction, the value of the iteration coefficient is enlarged in the next round of iteration, and the introduction of the L-M algorithm iteration extended Kalman filter can guarantee the global convergence, and the dynamic adjustment of the value of the iteration coefficient avoids the local optimal solution trap that the traditional method may encounter. Compared with the traditional iteration extended Kalman filter, the improved extended Kalman filtering framework used by the state of charge prediction model solves the convergence problem, pursues the accurate estimation of the covariance matrix, and pursues the better Kalman gain, ensures the stability and reliability of the state estimation, and effectively prevents the predicted state of charge finally predicted from deviating from the actual result.

[0193] Figure 9 An example of a schematic diagram of a physical structure of an electronic device is shown as Figure 9 The electronic device can include a processor 910, a communications interface 920, a memory 930, and a communications bus 940, wherein the processor 910, the communications interface 920, and the memory 930 communicate with each other through the communications bus 940. The processor 910 can invoke the logic command in the memory 930 to execute the battery state of charge prediction method of the lithium battery, which includes:

[0194] Obtaining the operating data of the lithium battery to be detected;

[0195] Inputting the obtained operating data into the constructed state of charge prediction model to obtain the predicted state vector of the lithium battery to be detected predicted by the state of charge prediction model; the predicted state vector is a vector composed of the predicted state of charge and the polarization voltage;

[0196] The state of charge prediction model predicts the state of charge based on an extended Kalman filter framework, and uses a Levenberg-Marquardt method to update an error covariance matrix of each round of iteration in the extended Kalman filter framework, and each round of iteration is performed through an iteration coefficient; when the Levenberg-Marquardt method has a positive impact on iteration, the value of the iteration coefficient is reduced in the next round of iteration; when the Levenberg-Marquardt method has a negative impact on iteration, the value of the iteration coefficient is increased in the next round of iteration.

[0197] In addition, the logic instructions in the memory 930 described above can be implemented in the form of a software function unit and sold or used as an independent product, and can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or the part that contributes to the prior art or part of the technical solutions can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a plurality of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the method described in various embodiments of the present application. The foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various program code storage media.

[0198] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement to part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for predicting the state of charge (SOC) of a lithium battery, characterized in that, The method includes: Obtain the operating data of the lithium battery under test; The acquired operational data is input into the constructed state of charge prediction model to obtain the predicted state vector of the lithium battery under test, which is predicted by the state of charge prediction model. The predicted state vector is a vector composed of the predicted state of charge and polarization voltage. The state of charge prediction model is based on an extended Kalman filter framework for predicting the state of charge. Within this framework, the Levenberg-Marquardt method is used to update the error covariance matrix for each iteration. Furthermore, each iteration is performed using iteration coefficients. When the Levenberg-Marquardt method has a positive impact on the iteration, the iteration coefficients are reduced in the next iteration; conversely, when the Levenberg-Marquardt method has a negative impact on the iteration, the iteration coefficients are increased in the next iteration. If the loss function value of the current iteration is less than that of the previous iteration, the Levenberg-Marquardt method is deemed to have a positive impact on the iteration, and the iteration coefficient is reduced in the next iteration. If the loss function value of the current iteration is not less than that of the previous iteration, the Levenberg-Marquardt method is deemed to have a negative impact on the iteration, and the iteration coefficient is increased in the next iteration. The formula for calculating the loss function value is: The formula for calculating the error covariance matrix in each iteration of the Levenberg-Marquardt method update process is as follows: in, Show the first The loss function value for each iteration round; Indicates the actual observed value; Represents the observation function; Represents the iteration coefficients; express Dynamic current at any given moment; Indicates the first During the round of iteration Always The predicted state vector at time step; Indicates the first The error covariance matrix after round iteration; Indicates the first During the round of iteration Always The error covariance matrix of the predicted state vector at time t.

2. The method for predicting the state of charge of a lithium battery according to claim 1, characterized in that, The state of charge prediction model is constructed through the following steps: Obtain the actual open-circuit voltage corresponding to each actual state of charge of the sample lithium battery during the discharge process, and determine the battery state of charge-open-circuit voltage curve based on the actual battery state of charge and the corresponding actual open-circuit voltage. The battery state-of-charge-open-circuit voltage curve was fitted using the least squares method. An extended Kalman filter framework is constructed based on the fitted battery state-of-charge-open-circuit voltage curve. The error covariance matrix of the extended Kalman filter framework is updated in each round of the iteration process using the Levenberg-Marquardt method, and each round of iteration is performed using the iteration coefficients.

3. The method for predicting the state of charge of a lithium battery according to claim 2, characterized in that, The process of obtaining the actual open-circuit voltage corresponding to each actual state of charge of the sample lithium battery during discharge, and determining the battery state of charge-open-circuit voltage curve based on the actual battery state of charge and the corresponding actual open-circuit voltage, specifically includes: The sample lithium battery was left to stand for a preset time to ensure that the internal state of the sample lithium battery reached a stable state. The sample lithium battery is charged using a preset charging current. When the sample lithium battery is charged to a preset state of charge, the sample lithium battery is charged using a preset charging voltage until the sample lithium battery is charged to a preset current state. The preset charging current and the preset charging voltage during the charging process are constant. A constant discharge current is used to discharge the sample lithium battery that has entered the preset current state, and the actual open circuit voltage value of the sample lithium battery under the actual battery charge state is determined at each preset charge value discharged until the sample lithium battery discharge ends. The open-circuit voltage-battery state of charge curve is determined based on the actual open-circuit voltage value and the corresponding actual battery state of charge.

4. The method for predicting the state of charge of a lithium battery according to claim 2, characterized in that, The construction of the extended Kalman filter framework, which involves updating the error covariance matrix of each iteration using the Levenberg-Marquardt method and performing iterations using iteration coefficients, specifically includes: An extended Kalman filter framework is constructed, and the initial state vector and initial covariance matrix are solved based on the fitted battery state-of-charge-open-circuit voltage curve. The error covariance moment of each round of the extended Kalman filter framework is updated using the Levenberg-Marquardt method, and each round of iteration is performed using the iteration coefficients. Determine whether the iteration convergence condition is met during the extended Kalman filter iteration phase. If the iteration convergence condition is not met, adjust the iteration parameters for the next iteration and update them until the iteration convergence condition is met and the iteration stops. When the extended Kalman filter iteration stage meets the iteration convergence condition, the Jacobian matrix is ​​updated. The error covariance matrix at the time of iteration stop is updated based on the updated Jacobian matrix. The updated error covariance matrix is ​​then verified. Based on the verification results, the optimal estimate at the time of iteration stop or the optimal estimate before the update is used as the predicted state vector.

5. The method for predicting the state of charge of a lithium battery according to claim 4, characterized in that, The construction of the extended Kalman filter framework, and the solution of the initial state vector and initial covariance matrix based on the fitted battery state-of-charge-open-circuit voltage curve, specifically includes: Based on the discrete state-space representation of the ampere-hour integral method and the first-order resistor-capacitor equivalent circuit model, the total state expression and the total measurement expression are constructed, and the total state expression and the total measurement expression are solved according to the fitted battery state-of-charge-open-circuit voltage curve. Initialize the parameters of the extended Kalman filter framework; Determine the initial state vector and the initial covariance matrix; The discrete state-space representation of the first-order resistor-capacitor equivalent circuit model is obtained through the following steps: A first-order resistor-capacitor equivalent circuit is constructed, and a first-order resistor-capacitor equivalent circuit model is formed based on the first-order resistor-capacitor equivalent circuit. The first-order resistor-capacitor equivalent circuit model includes a voltage source with an open-circuit voltage, an ohmic internal resistance, a transmission internal resistance, and a polarized capacitor. The voltage source, the ohmic internal resistance, the transmission internal resistance, and the polarized capacitor are connected to each other in parallel and jointly describe the transient response characteristics of the lithium battery. The first-order resistor-capacitor equivalent circuit model is discretized to obtain a discrete state expression, and a discrete state space expression is determined based on the discrete state expression.

6. The method for predicting the state of charge of a lithium battery according to claim 5, characterized in that, The process of updating the error covariance moment of each round of the extended Kalman filter framework using the Levenberg-Marquardt method and performing each round of iteration through iteration coefficients specifically includes: The extended Kalman filter framework is iteratively updated based on the initial covariance matrix and iteration coefficients, and the error covariance matrix of each round of the extended Kalman filter iteration stage is corrected using the Levenberg-Marquardt method. The Kalman gain for each round of the extended Kalman filter iteration stage is determined based on the error covariance matrix of each round. The optimal estimate of the state of charge and the error covariance matrix for each iteration are updated based on the Kalman gain of each round.

7. The method for predicting the state of charge of a lithium battery according to claim 4, characterized in that, The process of determining whether the extended Kalman filter iteration stage meets the iteration convergence condition, and adjusting and updating the iteration parameters for the next iteration if the convergence condition is not met, continues until the iteration convergence condition is met and the iteration stops. Specifically, this includes: Determine whether the iterative convergence condition is met during the iterative phase of the extended Kalman filter; If it is determined that the iterative convergence condition is not met, obtain the loss function values ​​of the current iteration and the previous iteration. When the loss function value of the current iteration is determined to be less than the loss function value of the previous iteration, the iteration parameters are reduced and the next iteration is performed until the iteration convergence condition is met and the iteration stops. When it is determined that the loss function value of the current iteration is not less than the loss function value of the previous iteration, the iteration parameters are increased and the next iteration is performed until the iteration convergence condition is met and the iteration stops.

8. The method for predicting the state of charge of a lithium battery according to claim 4, characterized in that, When the extended Kalman filter iteration stage meets the iteration convergence condition, the Jacobian matrix is ​​updated. Based on the updated Jacobian matrix, the error covariance matrix at the time of iteration termination is updated, and the updated error covariance matrix is ​​verified. Based on the verification result, the optimal estimate at the time of iteration termination or the optimal estimate before the update is used as the predicted state of charge. Specifically, this includes: When the extended Kalman filter iteration stage meets the iterative convergence condition, the Jacobian matrix is ​​updated by a re-approximation method. Update the error covariance matrix at the stopping point of the iteration based on the updated Jacobian matrix; Obtain the trace of the error covariance matrix before and after the update. If the trace of the updated error covariance matrix is ​​less than the trace of the original error covariance matrix, use the updated optimal estimate as the predicted state vector of the sample lithium battery. If the trace of the updated error covariance matrix is ​​determined to be no less than the trace of the original error covariance matrix, the optimal estimate at the time of iteration termination is used as the predicted state vector of the sample lithium battery.

9. A battery state-of-charge prediction device for lithium batteries, characterized in that, The device includes: The data acquisition module is used to acquire the operating data of the lithium battery under test. The state prediction module is used to input the acquired operating data into the constructed state of charge prediction model to obtain the predicted state vector of the lithium battery under test, which is predicted by the state of charge prediction model. The predicted state vector is a vector composed of the predicted state of charge and polarization voltage. The state of charge prediction model is based on an extended Kalman filter framework for predicting the state of charge. Within this framework, the Levenberg-Marquardt method is used to update the error covariance matrix for each iteration. Furthermore, each iteration is performed using iteration coefficients. When the Levenberg-Marquardt method has a positive impact on the iteration, the iteration coefficients are reduced in the next iteration; conversely, when the Levenberg-Marquardt method has a negative impact on the iteration, the iteration coefficients are increased in the next iteration. If the loss function value of the current iteration is less than that of the previous iteration, the Levenberg-Marquardt method is deemed to have a positive impact on the iteration, and the iteration coefficient is reduced in the next iteration. If the loss function value of the current iteration is not less than that of the previous iteration, the Levenberg-Marquardt method is deemed to have a negative impact on the iteration, and the iteration coefficient is increased in the next iteration. The formula for calculating the loss function value is: The formula for calculating the error covariance matrix in each iteration of the Levenberg-Marquardt method update process is as follows: in, Show the first The loss function value for each iteration round; Indicates the actual observed value; Represents the observation function; Represents the iteration coefficients; express Dynamic current at any given moment; Indicates the first During the round of iteration Always The predicted state vector at time step; Indicates the first The error covariance matrix after round iteration; Indicates the first During the round of iteration Always The error covariance matrix of the predicted state vector at time t.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the battery state-of-charge prediction method for lithium batteries as described in any one of claims 1 to 8.

Images