Lithium ion battery state-of-charge estimation method, apparatus and device, and storage medium

Abstract

The invention discloses a lithium ion battery charge state estimation method, device and equipment and a storage medium, and relates to the technical field of battery detection, and the method comprises the steps: replacing an ideal capacitor in a second-order resistance-capacitance equivalent circuit model with a constant phase element to construct a fractional-order second-order equivalent circuit model, carrying out discretization processing to obtain a state space equation and a state space expression; identifying model parameters by using a whale algorithm, determining a historical open-circuit voltage estimated value in combination with charging and discharging experimental data, a state-space equation and a state-space expression, and constructing a historical feature vector in combination with a historical terminal voltage measured value and a historical current measured value of the lithium ion battery; a time sequence feature is extracted by fusing an improved long-short-term memory structure, feature association is optimized through a time-enhanced attention mechanism, a time sequence prediction model is trained, and a state-of-charge estimation value is determined based on the model, a current terminal voltage measurement value, a current current measurement value and a current open-circuit voltage estimation value, so that the efficiency of estimating the state-of-charge of the battery is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of battery detection, and in particular to a method, device, equipment and storage medium for estimating the state of charge of a lithium-ion battery. Background Art

[0002] Lithium-ion batteries, with their long cycle life, environmental friendliness, and high energy density, are widely used in electric vehicles, energy storage, and electronic devices. However, lithium-ion batteries face numerous challenges during use, such as aging and thermal runaway. Therefore, a battery management system (BMS) is essential for continuously monitoring battery performance. State of Charge (SOC), a key indicator of battery performance, is a fundamental component of a battery management system (BMS). Therefore, accurately estimating SOC is crucial for battery health management.

[0003] SOC estimation methods primarily include ampere-hour counting, open-circuit voltage-based methods, impedance-based methods, equivalent circuit model methods, Kalman filtering, and data-driven methods. Different algorithms employ different SOC calculation processes and yield different results. Prior art studies have used various nonlinear Kalman filters (NLKFs) to conduct SOC estimation studies under both constant and varying temperatures. It has been found that a polynomial regression-based battery model (PRBM) combined with an adaptive unscented Kalman filter (AUKF) achieves superior accuracy. However, the effectiveness of these methods is limited by the significant correlation between battery SOC changes and open-circuit voltage, and more complex physical models require greater computational effort.

[0004] To address the above issues, some studies have attempted to combine neural networks with Kalman filters. This involves combining machine learning methods with equivalent circuits and Kalman filtering methods. Specifically, Kalman filtering or an improved extended Kalman filter is combined with artificial neural networks and ablation experiments are performed. However, deep learning models have strict requirements for the experimental data used, and the noise present in the data can mask the original characteristics of the data. Furthermore, while preprocessing experimental data can improve model accuracy, the input features of traditional deep learning models are generally average voltage, average current, voltage change rate, and current change rate, which still lack accuracy and robustness.

[0005] As can be seen from the above, how to improve the efficiency of estimating the state of charge of lithium-ion batteries during the state of charge estimation process of lithium-ion batteries is a problem that needs to be solved urgently. Summary of the Invention

[0006] In view of this, the present invention aims to provide a method, apparatus, device, and storage medium for estimating the state of charge of a lithium-ion battery, which can improve the efficiency of estimating the state of charge of the lithium-ion battery during the process of estimating the state of charge of the lithium-ion battery, thereby improving the safety of the production process. The specific scheme is as follows:

[0007] In a first aspect, the present application provides a method for estimating the state of charge of a lithium-ion battery, comprising:

[0008] The preset constant phase element is replaced by the ideal capacitor element in the second-order resistance and capacitance equivalent circuit model to obtain a fractional-order second-order equivalent circuit model corresponding to the lithium-ion battery, and the fractional-order second-order equivalent circuit model is discretized to obtain a state-space equation and a state-space expression. Then, the whale algorithm is used to perform offline identification of the parameters of the fractional-order second-order equivalent circuit model based on a charge and discharge experimental data set including different temperatures and operating conditions, and parameter identification results including ohmic internal resistance, polarization resistance, fractional capacitance value corresponding to polarization capacitance, and fractional order are obtained;

[0009] Determine a historical open circuit voltage estimate corresponding to the lithium-ion battery using the state-space equation and the state-space expression and based on a parameter identification result, and construct a historical feature vector based on the historical open circuit voltage estimate, historical terminal voltage measurements, and historical current measurements of the lithium-ion battery;

[0010] The historical feature vector is extracted using the fused improved long short-term memory structure in the initial time series prediction model to obtain target time series features. Then, the target time series features are subjected to global correlation between features and weighted enhancement of time series dimensions using the time-enhanced attention mechanism in the initial time series prediction model to obtain time series node features.

[0011] The initial timing prediction model is trained based on the timing node features and the corresponding true state of charge labels to obtain a target timing prediction model, so as to determine a state of charge estimate value based on the target timing prediction model and the current terminal voltage measurement value, the current current measurement value and the current open circuit voltage estimate value corresponding to the lithium-ion battery.

[0012] Optionally, the preset constant phase element is replaced with the ideal capacitor element in the second-order resistor-capacitor equivalent circuit model to obtain a fractional-order second-order equivalent circuit model corresponding to the lithium-ion battery, and the fractional-order second-order equivalent circuit model is discretized to obtain a state-space equation and a state-space expression, including:

[0013] Determining a position corresponding to an ideal capacitor element in a second-order resistor-capacitor equivalent circuit model, and then replacing the ideal capacitor element in the second-order resistor-capacitor equivalent circuit model with a preset constant phase element based on the position to obtain a fractional-order second-order equivalent circuit model corresponding to the lithium-ion battery;

[0014] Based on preset fractional-order differential definition rules, sampling time, preset memory length, system noise and preset Newton binomial coefficients, a differential equation for describing the polarization process corresponding to the fractional-order second-order equivalent circuit model is determined, and the differential equation is discretized to obtain a state-space equation and a state-space expression; wherein the state-space equation and the state-space expression are used to describe the changing dynamics corresponding to the polarization capacitor voltage and the diffusion capacitor voltage respectively.

[0015] Optionally, the whale algorithm is used to perform offline identification of the parameters of the fractional second-order equivalent circuit model based on a charge and discharge experimental data set including different temperatures and working conditions, and parameter identification results including ohmic internal resistance, polarization resistance, fractional capacitance value corresponding to polarization capacitance, and fractional order are obtained, including:

[0016] Designing a battery test experiment that includes intermittent charge and discharge and static operation to obtain a charge and discharge experimental data set including battery voltage data and current time series data under different temperatures and operating conditions. Then, using a polynomial fitting algorithm and based on the charge and discharge experimental data set, establish a functional relationship curve between open circuit voltage and state of charge.

[0017] Determining a position vector corresponding to the fractional-order second-order equivalent circuit model based on the functional relationship curve using the whale algorithm, and then initializing a population position corresponding to the fractional-order second-order equivalent circuit model based on the position vector using the whale optimization algorithm;

[0018] Determine a current control parameter corresponding to the population position, and determine whether an absolute value corresponding to the current control parameter is less than a preset threshold; if the absolute value corresponding to the current control parameter is less than the preset threshold, use the whale algorithm and perform a local optimal solution search operation on the population position based on the function relationship curve to obtain a local optimal solution determination result;

[0019] If the absolute value corresponding to the current control parameter is not less than the preset threshold, the whale algorithm is used to perform a global optimal solution search operation on the population position based on the functional relationship curve to obtain a global optimal solution determination result;

[0020] A parameter identification result is determined based on the local optimal solution determination result and the global optimal solution determination result; the parameter identification result includes ohmic internal resistance, polarization resistance, fractional capacitance value corresponding to polarization capacitance, and fractional order.

[0021] Optionally, the determining, using the state-space equation and the state-space expression and based on a parameter identification result, a historical open-circuit voltage estimate corresponding to the lithium-ion battery, and constructing a historical feature vector based on the historical open-circuit voltage estimate, historical terminal voltage measurements, and historical current measurements of the lithium-ion battery, includes:

[0022] The parameter identification result is processed using the fractional-order second-order equivalent circuit model and based on the state-space equation and the state-space expression to obtain a historical open-circuit voltage estimate corresponding to the lithium-ion battery; the historical open-circuit voltage estimate is an independent time series feature;

[0023] Obtain historical terminal voltage measurement values ​​and historical current measurement values ​​corresponding to the lithium-ion battery, and concatenate the historical open-circuit voltage estimate, the historical terminal voltage measurement values, and the historical current measurement values ​​in a feature dimension to obtain a historical feature vector.

[0024] Optionally, the fusion-improved long short-term memory structure in the initial time series prediction model is used to extract features from the historical feature vector to obtain target time series features, and then the time-enhanced attention mechanism in the initial time series prediction model is used to perform global correlation between features and time series dimension weighted enhancement operations on the target time series features to obtain time series node features, including:

[0025] Calling the fused improved long short-term memory structure in the initial time series prediction model to control the input gate and the forget gate using the exponential gating function, and using the input gate to control the inflow speed of the historical feature vector;

[0026] Extracting features from the historical feature vector using a preset state normalization technique and the fused improved long short-term memory structure to obtain initial time series features, and performing a noise interference suppression operation on the initial time series features to obtain time series features to be processed;

[0027] Determining the change frequencies corresponding to the sine function and the cosine function, respectively, and using the change frequencies to perform a time feature embedding operation on the time series feature to be processed to obtain a target time series feature; the target time series feature is a feature sequence with a time series dependency relationship;

[0028] Determine the time dimension corresponding to the target time series feature, and then use the time enhanced attention mechanism to compress the target time series feature based on the time dimension to obtain a feature mean corresponding to the target time series feature;

[0029] The feature mean is processed using the preset fully connected network, preset activation function and preset nonlinear change technology in the time-enhanced attention mechanism to obtain the corresponding attention weight, and the target time series feature is adaptively enhanced based on the attention weight to obtain the time step information to obtain the time series node feature; the network structure corresponding to the preset fully connected network is a network structure including a two-layer nonlinear transformation.

[0030] Optionally, the training of the initial time series prediction model based on the time series node features and the corresponding true state of charge labels to obtain a target time series prediction model includes:

[0031] Setting a data set of the lithium-ion battery under specific temperature and operating conditions as a source domain data set, and determining a true state of charge label corresponding to the source domain data set, so as to train an initial time series prediction model using the time series node features and the corresponding true state of charge labels to obtain a first time series prediction model to be adjusted;

[0032] Obtaining a current output result of the time series prediction model to be adjusted, and adjusting the fully connected layer parameters in the time series prediction model to be adjusted based on the current output result to obtain a current time series prediction model to be adjusted, and performing error verification on the current time series prediction model to be adjusted using a preset verification set to obtain an error verification result;

[0033] Determine whether the error verification result is greater than a preset threshold. If the error verification result is greater than the preset threshold, use the current output result and adjust the parameters corresponding to the fused improved long short-term memory structure based on the preset learning rate to obtain the target time series prediction model.

[0034] Optionally, after determining the state of charge estimate value by using the target time series prediction model and based on the current terminal voltage measurement value, the current current measurement value, and the current open circuit voltage estimate value corresponding to the lithium-ion battery, the method further includes:

[0035] Determine a coefficient of determination index, a root mean square error index, and a mean absolute error index; wherein the coefficient of determination index is used to evaluate the strength of the linear relationship between the model prediction value and the true value of the target time series prediction model; the root mean square error index is used to evaluate the average value of the square root of the difference between the prediction value and the true value of the target time series prediction model; the mean absolute error index is used to evaluate the average value of the absolute difference between the prediction value and the true value of the target time series prediction model;

[0036] The determination coefficient index, the root mean square error index and the mean absolute error index are used to evaluate the accuracy and stability of the target time series prediction model, and corresponding evaluation results are obtained, so as to use the evaluation results to adjust the model parameters of the target time series prediction model.

[0037] In a second aspect, the present application provides a device for estimating the state of charge of a lithium-ion battery, comprising:

[0038] A parameter identification result generation module is used to replace the ideal capacitor element in the second-order resistance and capacitance equivalent circuit model with a preset constant phase element to obtain a fractional-order second-order equivalent circuit model corresponding to the lithium-ion battery, and discretize the fractional-order second-order equivalent circuit model to obtain a state-space equation and a state-space expression. Then, the whale algorithm is used to perform offline identification of the parameters of the fractional-order second-order equivalent circuit model based on a charge and discharge experimental data set including different temperatures and operating conditions, and parameter identification results including ohmic internal resistance, polarization resistance, fractional capacitance value corresponding to polarization capacitance, and fractional order are obtained;

[0039] a historical characteristic vector determination module, configured to determine a historical open circuit voltage estimate corresponding to the lithium-ion battery using the state-space equation and the state-space expression and based on a parameter identification result, and construct a historical characteristic vector based on the historical open circuit voltage estimate, historical terminal voltage measurements, and historical current measurements of the lithium-ion battery;

[0040] A time series node feature determination module is used to extract features from the historical feature vector using the fused improved long short-term memory structure in the initial time series prediction model to obtain target time series features, and then use the time-enhanced attention mechanism in the initial time series prediction model to perform global correlation between features and time series dimension weighted enhancement operations on the target time series features to obtain time series node features;

[0041] The state of charge estimation module is used to train the initial time series prediction model based on the time series node features and the corresponding true state of charge labels to obtain a target time series prediction model, so as to use the target time series prediction model and determine the state of charge estimation value based on the current terminal voltage measurement value, current current measurement value and current open circuit voltage estimation value corresponding to the lithium-ion battery.

[0042] In a third aspect, the present application provides an electronic device, comprising:

[0043] Memory, used to store computer programs;

[0044] The processor is configured to execute the computer program to implement the aforementioned method for estimating the state of charge of a lithium-ion battery.

[0045] In a fourth aspect, the present application provides a computer-readable storage medium for storing a computer program, wherein the computer program implements the aforementioned method for estimating the state of charge of a lithium-ion battery when executed by a processor.

[0046] As can be seen from the above, before the state of charge estimation of the lithium-ion battery, the present application needs to replace the preset constant phase element with the ideal capacitor element in the second-order resistance and capacitance equivalent circuit model to obtain a fractional-order second-order equivalent circuit model corresponding to the lithium-ion battery, and discretize the fractional-order second-order equivalent circuit model to obtain the state-space equation and state-space expression. Then, the whale algorithm is used to perform offline identification of the parameters of the fractional-order second-order equivalent circuit model based on a charge and discharge experimental data set including different temperatures and working conditions to obtain the parameter identification results; the state-space equation and state-space expression are used to determine the historical open-circuit voltage estimate corresponding to the lithium-ion battery based on the parameter identification results, and the historical open-circuit voltage estimate is determined based on the historical open-circuit voltage estimate. , the historical terminal voltage measurement values ​​and historical current measurement values ​​of the lithium-ion battery are used to construct a historical feature vector; the fusion improved long short-term memory structure in the initial time series prediction model is used to extract features from the historical feature vector to obtain the target time series features, and then the time enhanced attention mechanism in the initial time series prediction model is used to perform global correlation between features and time dimension weighted enhancement operations on the target time series features to obtain time series node features; the initial time series prediction model is trained based on the time series node features and the corresponding true state of charge labels to obtain the target time series prediction model, so as to use the target time series prediction model to determine the state of charge estimate for the current terminal voltage measurement value, current current measurement value and current open circuit voltage estimate corresponding to the lithium-ion battery.

[0047] It can be seen that the present application first needs to replace the preset constant phase element with the ideal capacitor element in the second-order resistance and capacitance equivalent circuit model to obtain a fractional-order second-order equivalent circuit model corresponding to the lithium-ion battery, and discretize the fractional-order second-order equivalent circuit model to obtain the state-space equation and state-space expression, and then use the whale algorithm and based on the charge and discharge experimental data set including different temperatures and working conditions to perform offline identification on the parameters of the fractional-order second-order equivalent circuit model to obtain the parameter identification results; secondly, use the state-space equation and state-space expression and based on the parameter identification results to determine the historical open-circuit voltage estimate corresponding to the lithium-ion battery, and based on the historical open-circuit voltage estimate and the historical value of the lithium-ion battery The terminal voltage measurement value and the historical current measurement value are used to construct a historical feature vector. Furthermore, the fused improved long short-term memory structure in the initial time series prediction model is used to extract features from the historical feature vector to obtain the target time series features. The time-enhanced attention mechanism in the initial time series prediction model is then used to perform global correlation between features and weighted enhancement of the time series dimension on the target time series features to obtain time series node features. Finally, the initial time series prediction model is trained based on the time series node features and the corresponding true state of charge labels to obtain the target time series prediction model. The target time series prediction model is then used to determine the state of charge estimate for the current terminal voltage measurement value, current current measurement value, and current open circuit voltage estimate corresponding to the lithium-ion battery. This improves the efficiency of estimating the state of charge of lithium-ion batteries during the process, thereby enhancing the safety of the production process. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0049] Figure 1 This is a flow chart of a method for estimating the state of charge of a lithium-ion battery disclosed in this application;

[0050] Figure 2 A schematic diagram of a specific fractional-order second-order equivalent circuit model disclosed in this application;

[0051] Figure 3 This is a specific discharge current schematic diagram disclosed in this application; wherein, Figure 3 (a) is the current diagram of HPPC working condition, Figure 3 (b) Schematic diagram of battery capacity under HPPC working conditions;

[0052] Figure 4This is a schematic diagram comparing a specific integer-order model and a fractional-order model disclosed in this application; wherein, Figure 4 (a) Comparison diagram of terminal voltage predictions by integer-order model and fractional-order model; Figure 4 (b) Schematic diagram showing the comparison of the absolute error of predictions between the integer-order model and the fractional-order model;

[0053] Figure 5 A schematic diagram of a specific target time series prediction model disclosed in this application;

[0054] Figure 6 This is the SOC prediction comparison diagram of the four working conditions at 0 degrees and the corresponding absolute error comparison diagram disclosed in this application; wherein, Figure 6 (a1), Figure 6 (a2) are the SOC prediction comparison diagram and absolute error comparison diagram under 0 degree EUDC working condition, Figure 6 (b1), Figure 6 (b2) are the SOC prediction comparison diagram and absolute error comparison diagram under 0 degree FUDS working condition, Figure 6 (c1), Figure 6 (c2) are the SOC prediction comparison chart and absolute error comparison chart under 0 degree NEDC condition; Figure 6 (d1), Figure 6 (d2) are the SOC prediction comparison diagram and absolute error comparison diagram under 0 degree UDDS working condition;

[0055] Figure 7 This is a schematic diagram of three error evaluation index values ​​corresponding to each characteristic number at 0 degrees disclosed in this application; wherein, Figure 7 (a) is the corresponding characteristic number at 0 degrees Schematic diagram of indicator values; Figure 7 (b) Schematic diagram of the RMSE index value corresponding to each feature number at 0 degrees; Figure 7 (c) is a schematic diagram of the MAE index values ​​corresponding to each characteristic number at 0 degrees;

[0056] Figure 8 This is the SOC prediction comparison diagram of the four working conditions at 10 degrees and the corresponding absolute error comparison diagram disclosed in this application; wherein, Figure 8 (e1), Figure 8 (e2) are the SOC prediction comparison diagram and absolute error comparison diagram under 10-degree EUDC working condition; Figure 8 (f1), Figure 8 (f2) are the SOC prediction comparison diagram and absolute error comparison diagram under 10-degree FUDS working condition; Figure 8 (h1), Figure 8 (h2) are the SOC prediction comparison chart and absolute error comparison chart under 10 degrees NEDC condition; Figure 8 (g1), Figure 8 (g2) are the SOC prediction comparison diagram and absolute error comparison diagram under 10-degree UDDS working condition;

[0057] Figure 9 This is a schematic diagram of three error evaluation index values ​​corresponding to each characteristic number at 10 degrees disclosed in this application; wherein, Figure 9 (a) is the corresponding characteristic number at 10 degrees Schematic diagram of indicator values; Figure 9 (b) is a schematic diagram of the RMSE index value corresponding to each characteristic number at 10 degrees; Figure 9 (c) is a schematic diagram of the MAE index values ​​corresponding to each characteristic number at 10 degrees;

[0058] Figure 10 This is the SOC prediction comparison diagram of the four working conditions at 25 degrees and the corresponding absolute error comparison diagram disclosed in this application; wherein, Figure 10 (k1), Figure 10 (k2) are the SOC prediction comparison diagram and absolute error comparison diagram under 25-degree EUDC working condition; Figure 10 (l1), Figure 10 (l2) are the SOC prediction comparison diagram and absolute error comparison diagram under 25 degree FUDS working condition; Figure 10 (m1), Figure 10 (m2) are the SOC prediction comparison chart and absolute error comparison chart under 25 degrees NEDC condition; Figure 10 (n1), Figure 10 (n2) are the SOC prediction comparison diagram and absolute error comparison diagram under 25-degree UDDS working condition;

[0059] Figure 11 This is a schematic diagram of three error evaluation index values ​​corresponding to each characteristic number at 25 degrees disclosed in this application; wherein, Figure 11 (a) is the corresponding characteristic number at 25 degrees Schematic diagram of indicator values; Figure 11 (b) is a schematic diagram of the RMSE index value corresponding to each feature number at 25 degrees; Figure 11 (c) is a schematic diagram of the MAE index values ​​corresponding to each characteristic number at 25 degrees;

[0060] Figure 12 A specific SOC prediction comparison diagram and error comparison diagram of different models disclosed in this application;

[0061] Figure 13 This is a schematic structural diagram of a lithium-ion battery state of charge estimation device disclosed in this application;

[0062] Figure 14This is a structural diagram of an electronic device disclosed in this application. DETAILED DESCRIPTION

[0063] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0064] At present, lithium-ion batteries have the advantages of long cycle life, green environmental protection and high energy density, and are widely used in electric vehicles, energy storage, and electronic devices. However, lithium-ion batteries face many challenges during use, such as aging, thermal runaway, etc. The SOC estimation methods mainly include ampere-hour counting method, open circuit voltage-based method, impedance-based method, equivalent circuit model method, Kalman filtering and data-driven methods. However, the effectiveness of the above methods is limited by the fact that there is a significant correlation between the battery SOC change and the open circuit voltage, and the more complex the physical model, the greater the amount of calculation required. To this end, the present application provides a method for estimating the state of charge of a lithium-ion battery, which can improve the efficiency of estimating the state of charge of a lithium-ion battery during the process of estimating the state of charge of the lithium-ion battery, thereby improving the safety of the production process.

[0065] See also Figure 1 As shown, an embodiment of the present invention discloses a method for estimating the state of charge of a lithium-ion battery, comprising:

[0066] Step S11, replacing the ideal capacitor element in the second-order resistor-capacitor equivalent circuit model with a preset constant phase element to obtain a fractional-order second-order equivalent circuit model corresponding to the lithium-ion battery, and discretizing the fractional-order second-order equivalent circuit model to obtain a state-space equation and a state-space expression, and then using the whale algorithm and based on a charge and discharge experimental data set including different temperatures and working conditions, offline identifying the parameters of the fractional-order second-order equivalent circuit model to obtain parameter identification results including ohmic internal resistance, polarization resistance, fractional capacitance value corresponding to polarization capacitance, and fractional order.

[0067] In this embodiment, the common equivalent circuit models are the integer-order model (IOM) and the fractional-order model (FOM). However, as the requirements for SOC prediction accuracy become increasingly higher, and the advantage of the fractional-order model in more accurately depicting the complex reactions inside the battery has gradually become prominent, the embodiment of the present application adopts the fractional-order model for state of charge estimation. It is worth mentioning that fractional-order integration means that the order of differentiation and integration can take any real number and is not limited to integers. The embodiment of the present application divides the fractional-order definition into three categories: Grunwald-Letnikov (Grünwald-Letnikov Fractional Calculus) (G-L) definition, Riemann-Liouville (Riemann-Liouville Fractional Calculus) (R-L) definition, and Caputo (Caputo Fractional Calculus) definition. Among them, the Grunwald-Letnikov (G-L) definition is more suitable for the discretization of the time domain and is more in line with the dynamic characteristics of the actual physical system. The expression of the G-L definition is as follows:

[0068] ;

[0069] in, Representation function Regarding the variable t Fractional derivatives of order, is the order of the fractional derivative, is an integer variable, is the time step, t is the function The independent variable represents the function The value at a specific time point t, Indicates the initial time, represents the sampling time, is the memory or window length, and the expression is as follows:

[0070] .

[0071] Furthermore, are the Newton binomial coefficients, and the expression is as follows:

[0072] ; ;

[0073] Then, the embodiment of the present application can be derived according to the above formula The expression:

[0074] ;

[0075] in, Represents the sampling interval.

[0076] It is worth mentioning that, according to the definition of G-L, the fractional order model has memory effect and hysteresis effect, which has significant advantages in establishing dynamic physical models. It should be noted that the memory length will increase exponentially with time, so it is necessary to set the memory length reasonably. , in order to achieve a balance between model accuracy and computational efficiency.

[0077] In addition, the fractional-order model uses CPE components instead of RC components based on the integer-order model. Its core idea is to use non-integer-order differentials and integrals to replace traditional integer-order calculus to describe the dynamic behavior of the system. The expression of the CPE formula is as follows:

[0078] ; ;

[0079] ; ;

[0080] in, Represents capacitance, m and n represent fractional orders. The specific structure of the model is to use CPE elements instead of ideal capacitors based on the second-order RC equivalent circuit, and the schematic diagram is as follows Figure 2 shown.

[0081] Specifically, replacing the ideal capacitor element in the second-order resistor-capacitor equivalent circuit model with a preset constant phase element to obtain a fractional-order second-order equivalent circuit model corresponding to the lithium-ion battery, and discretizing the fractional-order second-order equivalent circuit model to obtain a state-space equation and a state-space expression, may include: determining the position corresponding to the ideal capacitor element in the second-order resistor-capacitor equivalent circuit model, and then replacing the ideal capacitor element in the second-order resistor-capacitor equivalent circuit model with a preset constant phase element based on the position to obtain a fractional-order second-order equivalent circuit model corresponding to the lithium-ion battery; determining the differential equation for describing the polarization process corresponding to the fractional-order second-order equivalent circuit model based on a preset fractional-order differential definition rule, sampling time, preset memory length, system noise, and preset Newton binomial coefficient, and discretizing the differential equation to obtain a state-space equation and a state-space expression; wherein the state-space equation and the state-space expression are used to describe the changing dynamics corresponding to the polarization capacitor voltage and the diffusion capacitor voltage, respectively.

[0082] In a specific embodiment, the state space expression of the fractional-order second-order RC equivalent circuit model is as follows:

[0083] ;

[0084] ;

[0085] in, and are the polarization capacitance voltage and the diffusion capacitance voltage, is the battery current, and is the polarization resistance, is the equivalent ohmic internal resistance, is the open circuit voltage compensation term.

[0086] In addition, the expression of the state space equation of the fractional second-order RC circuit model is as follows:

[0087] ; ;

[0088] in, , , , 、 and are parameter matrices, namely the state transfer matrix, input matrix and observation matrix, for Current flowing through the ohmic internal resistance The current value, Indicates the rated capacity of the battery. for Estimated value of the moment, polarization capacitance The order is , polarized capacitance The order is , open circuit voltage , and are the system process noise and observation noise, is the state vector at time kj, for Output vector at time instant.

[0089] From this we can further obtain that the expression of the state vector is as follows:

[0090] ,in, for The voltage across the RC network at time for The battery's state of charge at all times.

[0091] Subsequently, the embodiment of the present application needs to use intermittent charge and discharge experiments to identify the model parameters corresponding to the above model, and then fit the OCV-SOC curve according to the parameter identification results obtained from the battery charge and discharge experiments to obtain the corresponding 8th-order polynomial fitting function.

[0092] It is worth mentioning that the corresponding battery parameters during the charge and discharge experiments are shown in Table 1:

[0093] Table 1 Battery parameters

[0094]

[0095] Subsequently, the embodiment of the present application needs to charge the battery in constant current and constant voltage charging mode, and the constant current value and constant voltage value are 2A and 4.2V respectively, and maintain a constant current of 6.5A for 3 minutes, then let the battery stand for 3 hours to reach electrochemical equilibrium, and finally repeat the above steps until the cut-off voltage is reached. Furthermore, the embodiment of the present application needs to perform polynomial fitting on the experimental data obtained to obtain the OCV-SOC fitting curve diagram, and the discharge current schematic diagram is as follows Figure 3 As shown in the figure, the comparison between the integer order model and the fractional order model is shown in Figure 4 shown.

[0096] It is worth mentioning that, after analyzing the fitting results of different orders, the embodiment of the present application believes that the fitting result corresponding to the 8th order fitting is more accurate and the fitting curve is closer to the real result. Therefore, in a specific embodiment, the function expression of the 8th order fitting is as follows:

[0097] .

[0098] It is worth mentioning that the embodiment of the present application uses the Whale Optimization Algorithm (WOA) for offline parameter identification, and the obtained parameter identification results include R0, R1, R2, C1, C2, m, and n. Among them, the core of the Whale Optimization Algorithm is to simulate the hunting behavior of humpback whales in nature. In the Whale Algorithm, the coefficient vectors of the shrinkage, encirclement, and exploration processes are adjusted. and ,The algorithm can switch between local optimal solution search and global optimal search to adapt to different search stages.

[0099] In one embodiment, when | When | < 1, the algorithm performs a local optimal solution search operation:

[0100] ;

[0101] When | When | ≥ 1, the algorithm performs a global optimal search operation:

[0102] ;

[0103] ;

[0104] in, to are the corresponding fitting equation coefficients, is the battery's state of charge, represents the position of a random individual.

[0105] Subsequently, in a specific embodiment, the embodiment of the present application uses multiple data sets to verify the effectiveness of the model, including self-test data sets and University of Maryland data. The relevant information is as follows: Among them, the experimental platform for the self-test data (Ours) is a battery tester (NewareCT-4008Tn-5V12A-S1), the accuracy error of the tester is 0.05%, and the thermostat (NewareMGDW-225-20) is controlled at 0°C, 10°C, 25°C and 45°C. The experiment is run under four working conditions, with a total of 24 sets of experimental data. The lithium battery data set collected by the CALCE battery research team was used for preprocessing, and the INR 18650-20R battery data set with a rated capacity of 2,000mAh was selected, which includes four working conditions, three temperatures, and two initial and cutoff capacity data sets. A total of 24 sets of experimental data are used, and the data sets are shown in Table 2:

[0106] Table 2 Dataset

[0107]

[0108] In this embodiment, parameter identification is first performed on the self-test data set. The identification results are shown in Table 3. As can be seen from Table 3, the chemical reaction inside the battery can be better described by incorporating the fractional-order model, which effectively improves the accuracy of parameter identification.

[0109] Table 3 Schematic diagram of parameter identification results

[0110]

[0111] Specifically, the parameters of the fractional-order second-order equivalent circuit model are offline identified based on a charge and discharge experimental data set including different temperatures and working conditions using the whale algorithm, and parameter identification results including ohmic internal resistance, polarization resistance, fractional capacitance value corresponding to polarization capacitance, and fractional order are obtained. The method may include: designing a battery test experiment including intermittent charge and discharge and static operation to obtain a charge and discharge experimental data set including battery voltage data and current timing data under different temperatures and working conditions, and then using a polynomial fitting algorithm to establish a functional relationship curve between open circuit voltage and state of charge based on the charge and discharge experimental data set; using the whale algorithm to determine the position vector corresponding to the fractional-order second-order equivalent circuit model based on the functional relationship curve, and then using the whale optimization algorithm to initialize the position vector based on the position vector. The population position corresponding to the fractional-order second-order equivalent circuit model; determining the current control parameter corresponding to the population position, and judging whether the absolute value corresponding to the current control parameter is less than the preset threshold value; if the absolute value corresponding to the current control parameter is less than the preset threshold value, then using the whale algorithm and based on the function relationship curve to perform a local optimal solution search operation on the population position to obtain a local optimal solution determination result; if the absolute value corresponding to the current control parameter is not less than the preset threshold value, then using the whale algorithm and based on the function relationship curve to perform a global optimal solution search operation on the population position to obtain a global optimal solution determination result; determining the parameter identification result based on the local optimal solution determination result and the global optimal solution determination result; the parameter identification result includes the ohmic internal resistance, polarization resistance, the fractional-order capacitance value corresponding to the polarization capacitance, and the fractional-order order.

[0112] Step S12: Determine a historical open circuit voltage estimate corresponding to the lithium-ion battery using the state-space equation and the state-space expression based on a parameter identification result, and construct a historical feature vector based on the historical open circuit voltage estimate, historical terminal voltage measurements, and historical current measurements of the lithium-ion battery.

[0113] In this embodiment, after obtaining the state-space equation, state-space expression, and parameter identification results, the embodiment of the present application needs to determine the historical open-circuit voltage estimate corresponding to the lithium-ion battery, so as to construct a historical feature vector based on the historical open-circuit voltage estimate, the historical terminal voltage measurement values, and the historical current measurement values ​​of the lithium-ion battery. Specifically, using the state-space equation and state-space expression and based on the parameter identification results to determine the historical open-circuit voltage estimate corresponding to the lithium-ion battery, and constructing the historical feature vector based on the historical open-circuit voltage estimate, the historical terminal voltage measurement values, and the historical current measurement values ​​of the lithium-ion battery, can include: using a fractional-order second-order equivalent circuit model and processing the parameter identification results based on the state-space equation and state-space expression to obtain the historical open-circuit voltage estimate corresponding to the lithium-ion battery; the historical open-circuit voltage estimate is an independent time series feature; obtaining the historical terminal voltage measurement values ​​and the historical current measurement values ​​corresponding to the lithium-ion battery, and splicing the historical open-circuit voltage estimate, the historical terminal voltage measurement values, and the historical current measurement values ​​on the feature dimension to obtain the historical feature vector.

[0114] Step S13: Use the fused improved long short-term memory structure in the initial time series prediction model to extract features from the historical feature vector to obtain target time series features, and then use the time-enhanced attention mechanism in the initial time series prediction model to perform global correlation between features and time series dimension weighted enhancement operations on the target time series features to obtain time series node features.

[0115] In this embodiment, the xLSTM model is used to estimate state of charge. xLSTM is an improved LSTM (Long Short-Term Memory) network. Unlike the sigmoid gating function used in traditional LSTMs, xLSTM's exponential gating allows the input and forget gates to control memory updates exponentially, thereby improving information processing efficiency. Furthermore, the xLSTM design is suitable for processing large-scale and high-dimensional input datasets, addressing the scalability and performance limitations of traditional RNN models.

[0116] It's worth noting that positional encoding is a key component of the Transformer (a model based on the self-attention mechanism). It uses the changing frequency of sine and cosine functions as temporal features of the data embedding, thereby enhancing the model's ability to discern sequence order. This encoding method allows the model to parse the temporal properties of sequence data and is the first step after the self-attention mechanism processes the vector. The resulting vector is then passed through a feedforward network, where it is subjected to nonlinear transformations to extract higher-level features.

[0117] In addition, to enhance the Transformer's ability to discriminate between time channels, this embodiment of the present application introduces a temporal attention mechanism that compresses the time dimension to obtain the feature mean of the entire sequence. A two-layer fully connected network and a sigmoid activation function are then used to generate attention weights for the feature dimension. Subsequently, the attention weights are multiplied by the original sequence features before the final output to achieve adaptive enhancement of important time step features. This improves the model's focus on key time steps in time series data and makes it more suitable for SOC estimation than the standard Transformer.

[0118] Specifically, the fused improved long short-term memory structure in the initial time series prediction model is used to extract features from the historical feature vector to obtain the target time series features, and then the time-enhanced attention mechanism in the initial time series prediction model is used to perform global correlation between features and time-series dimension weighted enhancement operations on the target time series features to obtain time series node features, which may include: calling the fused improved long short-term memory structure in the initial time series prediction model to control the input gate and the forget gate using the exponential gating function, and using the input gate to control the inflow speed of the historical feature vector; using the preset state normalization technology and the fused improved long short-term memory structure to extract features from the historical feature vector to obtain the initial time series features, and performing noise interference suppression operations on the initial time series features to obtain the time series features to be processed; determining the sinusoidal The change frequencies corresponding to the function and the cosine function are used to perform time feature embedding operations on the time series features to be processed to obtain target time series features; the target time series features are feature sequences with time series dependencies; the time dimension corresponding to the target time series features is determined, and then the time-enhanced attention mechanism is used to compress the target time series features based on the time dimension to obtain the feature mean corresponding to the target time series features; the preset fully connected network, preset activation function and preset nonlinear change technology in the time-enhanced attention mechanism are used to process the feature mean to obtain the corresponding attention weight, and based on the attention weight, the target time series features are adaptively enhanced in time step information to obtain time series node features; the network structure corresponding to the preset fully connected network is a network structure containing a two-layer nonlinear transformation.

[0119] That is, in order to improve the modeling capability of complex time series data, the embodiment of the present application proposes a hybrid time series analysis framework that integrates extended long short-term memory (XLSTM) and Transformer-timeattention. The above framework adopts a hierarchical feature processing strategy and uses XLSTM to perform front-end feature extraction on the input sequence. Its unique state normalization mechanism can effectively capture long-term time series dependencies and suppress noise interference, providing a stable basic time series feature representation for subsequent analysis; the feature sequence output by XLSTM is then input into the Transformer-timeattention module, and the global correlation between features is modeled through a multi-head self-attention mechanism. At the same time, the time attention mechanism is embedded to perform weighted enhancement on the time series dimension. That is, the time dimension is first compressed to obtain the overall feature distribution of the sequence, and then a dynamic weight coefficient of the feature dimension is generated through a nonlinear transformation. Finally, the adaptive enhancement of key time step information is achieved by fusing the weights with the original features, thereby highlighting the time series node features that are critical to the task while mining the nonlinear dependencies between features of different time steps.

[0120] It is understandable that the above architecture, through the synergy of XLSTM's time series feature purification capabilities and Transformer-timeattention's dual mechanism of "global correlation modeling and key time step focusing", not only retains the ability to accurately capture long-term time series trends, but also enhances sensitivity to local key time series nodes. It shows better adaptability in tasks that rely on time series dynamic characteristics, such as battery state of charge (SOC) estimation, and provides an efficient feature fusion solution for time series data analysis under complex working conditions.

[0121] Step S14: training the initial timing prediction model based on the timing node features and the corresponding true state of charge labels to obtain a target timing prediction model, and determining a state of charge estimate based on the target timing prediction model and the current terminal voltage measurement value, current current measurement value, and current open circuit voltage estimate corresponding to the lithium-ion battery.

[0122] In this embodiment, after obtaining the time series node features, the embodiment of the present application needs to train the initial time series prediction model to obtain the target time series prediction model, and the basic process is as follows: first train a basic model, then migrate the new data set to continue training the model, and choose to only adjust the parameters of the XLSTM and fully connected layers so that it can adapt to the new data. And when fine-tuning, a slight learning rate adjustment is also made. It is worth mentioning that the embodiment of the present application proposes a two-stage migration method: first fine-tune the FC layer and judge the task error, so that when the above task error is greater than a threshold, fine-tune some parameters of the XLSTM to achieve high-precision migration, and the overall model diagram is as shown. Figure 5 shown.

[0123] Specifically, training an initial time series prediction model based on time series node features and corresponding real state of charge labels to obtain a target time series prediction model can include: setting a data set of lithium-ion batteries under specific temperature and working conditions as a source domain data set, and determining the real state of charge label corresponding to the source domain data set, so as to train the initial time series prediction model using the time series node features and the corresponding real state of charge label to obtain a first time series prediction model to be adjusted; obtaining the current output result of the time series prediction model to be adjusted, and adjusting the fully connected layer parameters in the time series prediction model to be adjusted based on the current output result to obtain the current time series prediction model to be adjusted, and using a preset verification set to perform error verification on the current time series prediction model to be adjusted to obtain an error verification result; judging whether the error verification result is greater than a preset threshold. If the error verification result is greater than the preset threshold, adjusting the parameters corresponding to the fused improved long short-term memory structure using the current output result and based on a preset learning rate to obtain the target time series prediction model.

[0124] It is worth mentioning that in the process of training the model, the embodiment of the present application needs to evaluate the model based on the evaluation index, so as to adjust the model parameters accordingly based on the evaluation results. It is set as the degree to which the model explains the variation in the data. The value range is 0 to 1. The closer it is to 1, the better the model fit is. Specifically, It reflects the ability of the independent variable to explain the variation in the dependent variable. RMSE is the average of the square roots of the differences between the model's predicted values ​​and the true values, measuring the magnitude of the model's prediction error. A smaller value indicates a smaller model's prediction error. MAE reflects the average of the absolute errors between the predicted values ​​and the true values, reflecting the average degree of deviation between the model's prediction results and the true values. The calculation formulas for the above three evaluation indicators are as follows:

[0125] ;

[0126] ;

[0127] ;

[0128] in, For the The actual value of the observations, For the The predicted value of the observation, is the average of the actual values ​​of all observations, is the actual value of the observation, is the predicted value of the observed value, For the The actual value of the observation at time t, For the The predicted value of the observation at time instant.

[0129] Specifically, after using the target time series prediction model and determining the state of charge estimation value based on the current terminal voltage measurement value, current current measurement value and current open circuit voltage estimation value corresponding to the lithium-ion battery, it can also include: determining the determination coefficient index, the root mean square error index and the mean absolute error index; wherein the determination coefficient index is used to evaluate the strength of the linear relationship between the model prediction value and the true value of the target time series prediction model; the root mean square error index is used to evaluate the average value of the square root of the difference between the prediction value and the true value of the target time series prediction model; the mean absolute error index is used to evaluate the average value of the absolute difference between the prediction value and the true value of the target time series prediction model; the determination coefficient index, the root mean square error index and the mean absolute error index are used to evaluate the accuracy and stability of the target time series prediction model, and obtain corresponding evaluation results, so as to use the evaluation results to adjust the model parameters of the target time series prediction model.

[0130] It is worth mentioning that in order to further verify the generalization ability of the model, the present embodiment conducted migration experiments on different batteries. In a specific embodiment, the present embodiment used the EUDC, FUDS, NEDC, and UDDS operating conditions of the University of Maryland's 80SOC as the basic training set to train the model. Figure 6 (a1)(a2), (b1)(b2), (c1)(c2) and (d1)(d2) are the SOC prediction comparison diagrams and the corresponding absolute error comparison diagrams of the above four working conditions at 0 degrees, respectively. F-Number is the feature number, and Figure 7 The histogram in the figure is from left to right the corresponding characteristic numbers at 0 degrees. , RMSE and MAE three error evaluation index values; Figure 8(e1)(e2), (f1)(f2), (h1)(h2) and (g1)(g2) are the SOC prediction comparison diagrams and the corresponding absolute error comparison diagrams of the above four working conditions at 10 degrees, respectively. Figure 9 The bar graph in the figure is from left to right the corresponding characteristic numbers under 10 degrees. , RMSE and MAE three error evaluation index values; Figure 10 The (k1)(k2), (l1)(l2), (m1)(m2) and (n1)(n2) are the SOC prediction comparison diagrams and the corresponding absolute error comparison diagrams of the above four working conditions at 25 degrees, respectively. Figure 9 The bar graph in the figure is from left to right the corresponding characteristic numbers at 25 degrees. , RMSE and MAE three error evaluation index values.

[0131] In this embodiment, a comparative diagram of the same battery studied in the same equipment under different working conditions and different temperatures is shown in Table 4:

[0132] Table 4 Schematic diagram of battery error evaluation under different temperatures and working conditions

[0133]

[0134] It is worth mentioning that the input features in the traditional deep learning model are The number of feature inputs is 2. A common approach uses average voltage, average current, voltage change rate, and current change rate as augmented feature inputs for the input data, bringing the number of feature inputs to 6. In this embodiment, the FUDS and EUDC operating conditions at 25 degrees Celsius were used as the basic training set. Experiments were conducted at different temperatures and operating conditions to verify the generalization capability of the model in this embodiment.

[0135] From the data in Table 4, it can be seen that the accuracy performance at low temperatures (0℃ and 10℃) when the number of feature inputs is 6 is actually lower than that when the number of feature inputs is 2. The model with the number of features 3 performs best under all working conditions, corresponding to the highest And the lowest RMSE and MAE values. Among them, The value increased by more than 0.5%, and the RMSE and MAE values ​​decreased by more than 60%. The ESM-XLTAT model is stable above 0.999, with both RMSE and MAE below 0.7%. These results demonstrate that the ESM-XLTAT model can effectively capture the inherent patterns of the data and exhibit high robustness. Furthermore, the present embodiment can visualize the model prediction results and absolute error, and when the feature number is 3, the error value is smaller and the error fluctuation is smaller under different temperatures and working conditions.

[0136] In addition, to demonstrate the effectiveness of the components and improvements, the present invention conducted a performance comparison experiment of different models in the lithium-ion battery SOC estimation task, and the experimental results are shown in Table 5:

[0137] Table 5. Performance comparison of different models in lithium-ion battery SOC estimation tasks.

[0138]

[0139] Among them, Table 5 shows the performance comparison of different models in the lithium-ion battery SOC estimation task. The embodiment of the present application lists traditional machine learning models, such as CNN (Convolutional Neural Network), GRU (Gated Recurrent Unit), LSTM, and deep learning models, such as BiLSTM (Bidirectional Long Short-Term Memory, i.e. bidirectional long short-term memory network), BiGRU (Bidirectional Gated Recurrent Unit, i.e. bidirectional gated recurrent unit), Transformer-BiLSTM (Transformer-Bidirectional LSTM, i.e. Transformer and bidirectional LSTM combined model), Transformer-BiGRU (Bidirectional Long Short-Term Memory, i.e. bidirectional long short-term memory network) and the XLSTM-TAT model proposed in the embodiment of the present application, and gives three evaluation indicators for each model: , RMSE (Root Mean Square Error), MAE (Mean Absolute Error). As can be seen from the table, the XLSTM-TAT model proposed in the embodiment of this application shows excellent performance in the SOC estimation task. The value reached 0.9995, while the RMSE and MAPE were as low as 0.36% and 0.49%, respectively. This result significantly outperformed other mainstream models, including CNN, LSTM, BiGRU, BiLSTM, and their variants. Compared to the next-best performing BiGRU model, XLSTM-TAT achieved a MAE reduction of approximately 73.3% and a RMSE reduction of approximately 71.0%, demonstrating significant advantages in prediction accuracy and stability. This performance improvement is primarily attributed to the unique architectural design of XLSTM-TAT.

[0140] It is worth mentioning that the performance comparison of different model structures in the lithium-ion battery SOC estimation task is shown in Table 6:

[0141] Table 6 Performance comparison of different model structures in lithium-ion battery SOC estimation tasks

[0142]

[0143] Then, convert the above table data into a line graph, and the SOC prediction comparison chart and error comparison chart of different models are as follows: Figure 12 As shown in the figure, traditional convolutional neural networks (CNNs) and recurrent neural networks, such as LSTMs and GRUs, exhibit certain performance differences in the SOC estimation task. CNNs, due to their lack of modeling of temporal dependencies when processing time series data, exhibit low prediction accuracy (MAE = 1.52%, RMSE = 2.11%). While LSTMs and GRUs also employ gating mechanisms to better capture the dynamic characteristics of time series data, GRUs offer a simpler structure than LSTMs, avoiding errors caused by excessive ineffective parameters and achieving significantly higher accuracy. However, these unidirectional recurrent structures still suffer from information loss or vanishing gradients when processing long sequences of data. Bidirectional recurrent neural networks, such as BiLSTMs and BiGRUs, mitigate these issues to some extent by simultaneously considering both forward and backward information, but their performance improvements are limited. Hybrid architectures, such as Transformer-BiLSTMs and Transformer-BiGRUs, attempt to combine the local feature extraction capabilities of convolutional networks with the temporal modeling capabilities of the Transformer architecture, but their performance degrades. This is because the Transformer's self-attention mechanism inherently captures long-sequence dependencies, while bidirectional recurrent neural networks can capture dependencies within time series. This functional overlap leads to information redundancy when the two are combined, resulting in a decrease in prediction accuracy. In contrast, XLSTM-TAT, by combining improved XLSTM units with the time-series parallel processing capabilities of Transformer-TimeAttention, effectively enhances the ability to model global state dependencies while capturing local dynamic characteristics, resulting in significant improvements in both prediction accuracy and robustness.

[0144] In this way, the embodiment of the present application first constructs a fractional-order second-order RC model and uses the whale algorithm to identify the model's parameters, so that the OCV obtained from the discretized circuit equation is input as an extended feature into the model for learning, thereby enriching the model input. In addition, the embodiment of the present application proposes an ESM-XLSTM-TAT model, and experimentally verifies the model on a single battery and compares the model's combination with deep learning methods with general deep learning methods. The experimental results show that changing the traditional data-driven input and inputting OCV as an additional feature into the model can greatly improve the accuracy of the model prediction and the model's generalization ability, and the ability to predict state of charge under different temperatures and operating conditions has been significantly improved. In addition, when the XLSTM-TAT model is compared with common deep learning methods, the XLSTM-TAT model has obvious advantages in predicting SOC. Specifically, the developed model performs well under different temperatures, operating conditions, and transfer learning methods. Therefore, the ESM-XLSTM-TAT proposed in the embodiment of the present application has strong practicality.

[0145] As can be seen from the above, the embodiment of the present application first needs to replace the preset constant phase element with the ideal capacitor element in the second-order resistor-capacitor equivalent circuit model to obtain a fractional-order second-order equivalent circuit model corresponding to the lithium-ion battery, and discretize the fractional-order second-order equivalent circuit model to obtain a state-space equation and a state-space expression. Then, the whale algorithm is used to perform offline identification on the parameters of the fractional-order second-order equivalent circuit model based on a charge and discharge experimental data set including different temperatures and working conditions to obtain a parameter identification result; secondly, the state-space equation and the state-space expression are used to determine the historical open-circuit voltage estimate corresponding to the lithium-ion battery based on the parameter identification result, and based on the historical open-circuit voltage estimate and the historical state of the lithium-ion battery, the historical open-circuit voltage estimate is obtained. The historical terminal voltage measurement values ​​and historical current measurement values ​​are used to construct a historical feature vector. Furthermore, the fused improved long short-term memory structure in the initial time series prediction model is used to extract features from the historical feature vector to obtain the target time series features. The time-enhanced attention mechanism in the initial time series prediction model is then used to perform global correlation between features and weighted enhancement of the time series dimension on the target time series features to obtain time series node features. Finally, the initial time series prediction model is trained based on the time series node features and the corresponding true state of charge labels to obtain the target time series prediction model. The target time series prediction model is then used to determine the state of charge estimate for the current terminal voltage measurement value, current current measurement value, and current open circuit voltage estimate corresponding to the lithium-ion battery. This improves the efficiency of estimating the state of charge of lithium-ion batteries during the process, thereby enhancing the safety of the production process.

[0146] Accordingly, see Figure 13 As shown, the present application also provides a device for estimating the state of charge of a lithium-ion battery, comprising:

[0147] A parameter identification result generation module 11 is used to replace the ideal capacitor element in the second-order resistance and capacitance equivalent circuit model with a preset constant phase element to obtain a fractional-order second-order equivalent circuit model corresponding to the lithium-ion battery, and discretize the fractional-order second-order equivalent circuit model to obtain a state-space equation and a state-space expression. Then, the parameters of the fractional-order second-order equivalent circuit model are offline identified using a whale algorithm based on a charge and discharge experimental data set including different temperatures and working conditions to obtain parameter identification results including ohmic internal resistance, polarization resistance, fractional capacitance value corresponding to polarization capacitance, and fractional order.

[0148] a historical feature vector determination module 12, configured to determine a historical open circuit voltage estimate corresponding to the lithium-ion battery using the state-space equation and the state-space expression based on a parameter identification result, and construct a historical feature vector based on the historical open circuit voltage estimate, historical terminal voltage measurements, and historical current measurements of the lithium-ion battery;

[0149] A time series node feature determination module 13 is configured to extract features from the historical feature vector using the fused improved long short-term memory structure in the initial time series prediction model to obtain target time series features, and then perform inter-feature global correlation and time series dimension weighted enhancement operations on the target time series features using the time enhanced attention mechanism in the initial time series prediction model to obtain time series node features;

[0150] The state of charge estimation module 14 is used to train the initial time series prediction model based on the time series node characteristics and the corresponding true state of charge labels to obtain a target time series prediction model, so as to use the target time series prediction model and determine the state of charge estimation value based on the current terminal voltage measurement value, current current measurement value and current open circuit voltage estimation value corresponding to the lithium-ion battery.

[0151] In some specific implementations, the parameter identification result generating module 11 may specifically include:

[0152] an equivalent circuit model determining unit, configured to determine a position corresponding to an ideal capacitor element in a second-order resistor-capacitor equivalent circuit model, and then replace the ideal capacitor element in the second-order resistor-capacitor equivalent circuit model with a preset constant phase element based on the position, to obtain a fractional-order second-order equivalent circuit model corresponding to the lithium-ion battery;

[0153] A state-space equation determination unit is used to determine a differential equation for describing the polarization process corresponding to the fractional-order second-order equivalent circuit model based on a preset fractional-order differential definition rule, a sampling time, a preset memory length, system noise, and a preset Newton binomial coefficient, and discretize the differential equation to obtain a state-space equation and a state-space expression; wherein the state-space equation and the state-space expression are used to describe the changing dynamics corresponding to the polarization capacitor voltage and the diffusion capacitor voltage, respectively.

[0154] In some specific implementations, the parameter identification result generating module 11 may specifically include:

[0155] a functional relationship curve establishing unit, for designing a battery test experiment including intermittent charge and discharge and static operation to obtain a charge and discharge experimental data set including battery voltage data and current time series data under different temperatures and working conditions, and then using a polynomial fitting algorithm to establish a functional relationship curve between open circuit voltage and state of charge based on the charge and discharge experimental data set;

[0156] a population position determining unit, configured to determine a position vector corresponding to the fractional-order second-order equivalent circuit model using a whale algorithm and based on the functional relationship curve, and then initialize a population position corresponding to the fractional-order second-order equivalent circuit model based on the position vector using the whale optimization algorithm;

[0157] a local optimal solution determination result determination unit, configured to determine a current control parameter corresponding to the population position, and determine whether an absolute value corresponding to the current control parameter is less than a preset threshold; if the absolute value corresponding to the current control parameter is less than the preset threshold, performing a local optimal solution search operation on the population position using the whale algorithm and based on the functional relationship curve to obtain a local optimal solution determination result;

[0158] a global optimal solution determination result determination unit, configured to, if the absolute value corresponding to the current control parameter is not less than the preset threshold, perform a global optimal solution search operation on the population position using the whale algorithm and based on the functional relationship curve to obtain a global optimal solution determination result;

[0159] A parameter identification result generating subunit is used to determine a parameter identification result based on the local optimal solution determination result and the global optimal solution determination result; the parameter identification result includes ohmic internal resistance, polarization resistance, fractional capacitance value corresponding to polarization capacitance, and fractional order.

[0160] In some specific implementations, the historical feature vector determination module 12 may specifically include:

[0161] a parameter identification result processing unit, configured to process the parameter identification result using the fractional-order second-order equivalent circuit model and based on the state-space equation and the state-space expression to obtain a historical open-circuit voltage estimate corresponding to the lithium-ion battery; the historical open-circuit voltage estimate is an independent time series feature;

[0162] The historical feature vector determination subunit is used to obtain the historical terminal voltage measurement value and the historical current measurement value corresponding to the lithium-ion battery, and to splice the historical open circuit voltage estimation value, the historical terminal voltage measurement value and the historical current measurement value in the feature dimension to obtain a historical feature vector.

[0163] In some specific implementations, the timing node feature determination module 13 may specifically include:

[0164] A feature vector inflow speed control unit is used to call the fusion improved long short-term memory structure in the initial time series prediction model to control the input gate and the forget gate using the exponential gating function, and to control the inflow speed of the historical feature vector using the input gate;

[0165] a feature extraction unit, configured to extract features from the historical feature vector using a preset state normalization technique and the fused improved long short-term memory structure to obtain initial time series features, and perform a noise interference suppression operation on the initial time series features to obtain time series features to be processed;

[0166] A time series feature determination unit is used to determine the change frequencies corresponding to the sine function and the cosine function respectively, and use the change frequencies to perform a time feature embedding operation on the time series feature to be processed to obtain a target time series feature; the target time series feature is a feature sequence with a time series dependency relationship;

[0167] a feature mean determining unit, configured to determine a time dimension corresponding to the target time series feature, and then use the time enhanced attention mechanism to perform a compression operation on the target time series feature based on the time dimension to obtain a feature mean corresponding to the target time series feature;

[0168] A timing node feature determination unit is used to process the feature mean using the preset fully connected network, preset activation function and preset nonlinear change technology in the time enhanced attention mechanism to obtain the corresponding attention weight, and adaptively enhance the time step information of the target timing feature based on the attention weight to obtain the timing node feature; the network structure corresponding to the preset fully connected network is a network structure including a two-layer nonlinear transformation.

[0169] In some specific implementations, the state of charge estimation module 14 may specifically include:

[0170] a source domain data set determining unit, configured to set a data set of the lithium-ion battery under specific temperature and operating conditions as a source domain data set, and determine a true state of charge label corresponding to the source domain data set, so as to train an initial time series prediction model using the time series node features and the corresponding true state of charge labels to obtain a first time series prediction model to be adjusted;

[0171] an error verification result determining unit, configured to obtain a current output result of the time series prediction model to be adjusted, and adjust the fully connected layer parameters in the time series prediction model to be adjusted based on the current output result to obtain a current time series prediction model to be adjusted, and perform error verification on the current time series prediction model to be adjusted using a preset verification set to obtain an error verification result;

[0172] A parameter adjustment unit is used to determine whether the error verification result is greater than a preset threshold. If the error verification result is greater than the preset threshold, the parameters corresponding to the fused improved long short-term memory structure are adjusted using the current output result and based on a preset learning rate to obtain a target time series prediction model.

[0173] In some specific embodiments, the lithium-ion battery state of charge estimation device may further include:

[0174] An indicator determination unit is used to determine a coefficient of determination indicator, a root mean square error indicator, and a mean absolute error indicator; wherein the coefficient of determination indicator is used to evaluate the strength of the linear relationship between the model prediction value and the true value of the target time series prediction model; the root mean square error indicator is used to evaluate the average value of the square root of the difference between the prediction value and the true value of the target time series prediction model; the mean absolute error indicator is used to evaluate the average value of the absolute difference between the prediction value and the true value of the target time series prediction model;

[0175] An evaluation result determination unit is used to evaluate the accuracy and stability of the target time series prediction model using the determination coefficient index, the root mean square error index and the mean absolute error index, and obtain corresponding evaluation results, so as to adjust the model parameters of the target time series prediction model using the evaluation results.

[0176] Furthermore, the embodiment of the present application also discloses an electronic device, Figure 14This is a structural diagram of an electronic device 20 according to an exemplary embodiment. The content in the diagram should not be considered as any limitation on the scope of use of this application. The electronic device 20 may specifically include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 is used to store a computer program, which is loaded and executed by the processor 21 to implement the relevant steps in the method for estimating the state of charge of a lithium-ion battery disclosed in any of the aforementioned embodiments. In addition, the electronic device 20 in this embodiment may specifically be an electronic computer.

[0177] In this embodiment, the power supply 23 is used to provide operating voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and the external device. The communication protocol it follows is any communication protocol that can be applied to the technical solution of this application and is not specifically limited here; the input and output interface 25 is used to obtain external input data or output data to the outside world. Its specific interface type can be selected according to specific application needs and is not specifically limited here.

[0178] In addition, the memory 22, as a carrier for resource storage, can be a read-only memory, random access memory, disk or CD, etc. The resources stored thereon can include an operating system 221, a computer program 222, etc., and the storage method can be temporary storage or permanent storage.

[0179] The operating system 221 is used to manage and control the hardware devices and computer program 222 on the electronic device 20, and can be Windows Server, Netware, Unix, Linux, etc. In addition to including a computer program capable of implementing the method for estimating the state of charge of a lithium-ion battery executed by the electronic device 20 disclosed in any of the aforementioned embodiments, the computer program 222 may further include a computer program capable of implementing other specific tasks.

[0180] Furthermore, the present application discloses a computer-readable storage medium for storing a computer program; wherein, when executed by a processor, the computer program implements the aforementioned method for estimating the state of charge of a lithium-ion battery. The specific steps of this method can be referred to the corresponding contents disclosed in the aforementioned embodiments and will not be repeated here.

[0181] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from the other embodiments. Reference can be made to the descriptions of the identical or similar parts between the various embodiments. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple, and the relevant parts can be referred to the descriptions of the methods.

[0182] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the above description has generally described the components and steps of each example according to their functions. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0183] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein may be implemented directly using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0184] Finally, it should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article, or device comprising the element.

[0185] The above is a detailed introduction to the technical solution provided by the present application. Specific examples are used herein to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method of the present application and its core idea. At the same time, for those skilled in the art, according to the ideas of the present application, there may be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.

Claims

1. A method for estimating the state of charge of a lithium-ion battery, characterized in that: include: The preset constant phase element is replaced by the ideal capacitor element in the second-order resistance and capacitance equivalent circuit model to obtain a fractional-order second-order equivalent circuit model corresponding to the lithium-ion battery, and the fractional-order second-order equivalent circuit model is discretized to obtain a state-space equation and a state-space expression. Then, the whale algorithm is used to perform offline identification of the parameters of the fractional-order second-order equivalent circuit model based on a charge and discharge experimental data set including different temperatures and operating conditions, and parameter identification results including ohmic internal resistance, polarization resistance, fractional capacitance value corresponding to polarization capacitance, and fractional order are obtained; Determine a historical open circuit voltage estimate corresponding to the lithium-ion battery using the state-space equation and the state-space expression and based on a parameter identification result, and construct a historical feature vector based on the historical open circuit voltage estimate, historical terminal voltage measurements, and historical current measurements of the lithium-ion battery; The historical feature vector is extracted using the fused improved long short-term memory structure in the initial time series prediction model to obtain target time series features. Then, the target time series features are subjected to global correlation between features and weighted enhancement of time series dimensions using the time-enhanced attention mechanism in the initial time series prediction model to obtain time series node features. The initial timing prediction model is trained based on the timing node features and the corresponding true state of charge labels to obtain a target timing prediction model, so as to determine a state of charge estimate value based on the target timing prediction model and the current terminal voltage measurement value, the current current measurement value and the current open circuit voltage estimate value corresponding to the lithium-ion battery.

2. The method for estimating the state of charge of a lithium-ion battery according to claim 1, wherein: The preset constant phase element is replaced with the ideal capacitor element in the second-order resistor-capacitor equivalent circuit model to obtain a fractional-order second-order equivalent circuit model corresponding to the lithium-ion battery, and the fractional-order second-order equivalent circuit model is discretized to obtain a state-space equation and a state-space expression, including: Determining a position corresponding to an ideal capacitor element in a second-order resistor-capacitor equivalent circuit model, and then replacing the ideal capacitor element in the second-order resistor-capacitor equivalent circuit model with a preset constant phase element based on the position to obtain a fractional-order second-order equivalent circuit model corresponding to the lithium-ion battery; Based on preset fractional-order differential definition rules, sampling time, preset memory length, system noise and preset Newton binomial coefficients, a differential equation for describing the polarization process corresponding to the fractional-order second-order equivalent circuit model is determined, and the differential equation is discretized to obtain a state-space equation and a state-space expression; wherein the state-space equation and the state-space expression are used to describe the changing dynamics corresponding to the polarization capacitor voltage and the diffusion capacitor voltage respectively.

3. The method for estimating the state of charge of a lithium-ion battery according to claim 1, wherein: The parameters of the fractional-order second-order equivalent circuit model are offline identified using the whale algorithm based on a charge-discharge experimental data set including different temperatures and working conditions, and parameter identification results including ohmic internal resistance, polarization resistance, fractional capacitance value corresponding to polarization capacitance, and fractional order are obtained, including: Designing a battery test experiment that includes intermittent charge and discharge and static operation to obtain a charge and discharge experimental data set including battery voltage data and current time series data under different temperatures and operating conditions. Then, using a polynomial fitting algorithm and based on the charge and discharge experimental data set, establish a functional relationship curve between open circuit voltage and state of charge. Determining a position vector corresponding to the fractional-order second-order equivalent circuit model based on the functional relationship curve using the whale algorithm, and then initializing a population position corresponding to the fractional-order second-order equivalent circuit model based on the position vector using the whale optimization algorithm; Determine a current control parameter corresponding to the population position, and determine whether an absolute value corresponding to the current control parameter is less than a preset threshold; if the absolute value corresponding to the current control parameter is less than the preset threshold, use the whale algorithm and perform a local optimal solution search operation on the population position based on the function relationship curve to obtain a local optimal solution determination result; If the absolute value corresponding to the current control parameter is not less than the preset threshold, the whale algorithm is used to perform a global optimal solution search operation on the population position based on the functional relationship curve to obtain a global optimal solution determination result; A parameter identification result is determined based on the local optimal solution determination result and the global optimal solution determination result; the parameter identification result includes ohmic internal resistance, polarization resistance, fractional capacitance value corresponding to polarization capacitance, and fractional order.

4. The method for estimating the state of charge of a lithium-ion battery according to claim 1, wherein: The method of determining a historical open circuit voltage estimate corresponding to the lithium-ion battery using the state-space equation and the state-space expression based on a parameter identification result, and constructing a historical feature vector based on the historical open circuit voltage estimate, historical terminal voltage measurement values, and historical current measurement values ​​of the lithium-ion battery, includes: The parameter identification result is processed using the fractional-order second-order equivalent circuit model and based on the state-space equation and the state-space expression to obtain a historical open-circuit voltage estimate corresponding to the lithium-ion battery; the historical open-circuit voltage estimate is an independent time series feature; Obtain historical terminal voltage measurement values ​​and historical current measurement values ​​corresponding to the lithium-ion battery, and concatenate the historical open-circuit voltage estimate, the historical terminal voltage measurement values, and the historical current measurement values ​​in a feature dimension to obtain a historical feature vector.

5. The method for estimating the state of charge of a lithium-ion battery according to claim 1, wherein: The fusion-improved long short-term memory structure in the initial time series prediction model is used to extract features from the historical feature vector to obtain target time series features, and then the time-enhanced attention mechanism in the initial time series prediction model is used to perform global correlation between features and time series dimension weighted enhancement operations on the target time series features to obtain time series node features, including: Calling the fused improved long short-term memory structure in the initial time series prediction model to control the input gate and the forget gate using the exponential gating function, and using the input gate to control the inflow speed of the historical feature vector; Extracting features from the historical feature vector using a preset state normalization technique and the fused improved long short-term memory structure to obtain initial time series features, and performing a noise interference suppression operation on the initial time series features to obtain time series features to be processed; Determining the change frequencies corresponding to the sine function and the cosine function, respectively, and using the change frequencies to perform a time feature embedding operation on the time series feature to be processed to obtain a target time series feature; the target time series feature is a feature sequence with a time series dependency relationship; Determine the time dimension corresponding to the target time series feature, and then use the time enhanced attention mechanism to compress the target time series feature based on the time dimension to obtain a feature mean corresponding to the target time series feature; The feature mean is processed using the preset fully connected network, preset activation function and preset nonlinear change technology in the time-enhanced attention mechanism to obtain the corresponding attention weight, and the target time series feature is adaptively enhanced based on the attention weight to obtain the time step information to obtain the time series node feature; the network structure corresponding to the preset fully connected network is a network structure including a two-layer nonlinear transformation.

6. The method for estimating the state of charge of a lithium-ion battery according to claim 1, wherein: The training of the initial time series prediction model based on the time series node features and the corresponding true state of charge labels to obtain a target time series prediction model includes: Setting a data set of the lithium-ion battery under specific temperature and operating conditions as a source domain data set, and determining a true state of charge label corresponding to the source domain data set, so as to train an initial time series prediction model using the time series node features and the corresponding true state of charge labels to obtain a first time series prediction model to be adjusted; Obtaining a current output result of the time series prediction model to be adjusted, and adjusting the fully connected layer parameters in the time series prediction model to be adjusted based on the current output result to obtain a current time series prediction model to be adjusted, and performing error verification on the current time series prediction model to be adjusted using a preset verification set to obtain an error verification result; Determine whether the error verification result is greater than a preset threshold. If the error verification result is greater than the preset threshold, use the current output result and adjust the parameters corresponding to the fused improved long short-term memory structure based on the preset learning rate to obtain the target time series prediction model.

7. The method for estimating the state of charge of a lithium-ion battery according to any one of claims 1 to 6, wherein: After determining the state of charge estimation value based on the current terminal voltage measurement value, the current current measurement value, and the current open circuit voltage estimation value corresponding to the lithium-ion battery using the target time series prediction model, the method further includes: Determine a coefficient of determination index, a root mean square error index, and a mean absolute error index; wherein the coefficient of determination index is used to evaluate the strength of the linear relationship between the model prediction value and the true value of the target time series prediction model; the root mean square error index is used to evaluate the average value of the square root of the difference between the prediction value and the true value of the target time series prediction model; the mean absolute error index is used to evaluate the average value of the absolute difference between the prediction value and the true value of the target time series prediction model; The determination coefficient index, the root mean square error index and the mean absolute error index are used to evaluate the accuracy and stability of the target time series prediction model, and corresponding evaluation results are obtained, so as to use the evaluation results to adjust the model parameters of the target time series prediction model.

8. A device for estimating the state of charge of a lithium-ion battery, characterized in that: include: A parameter identification result generation module is used to replace the ideal capacitor element in the second-order resistance and capacitance equivalent circuit model with a preset constant phase element to obtain a fractional-order second-order equivalent circuit model corresponding to the lithium-ion battery, and discretize the fractional-order second-order equivalent circuit model to obtain a state-space equation and a state-space expression. Then, the whale algorithm is used to perform offline identification of the parameters of the fractional-order second-order equivalent circuit model based on a charge and discharge experimental data set including different temperatures and operating conditions, and parameter identification results including ohmic internal resistance, polarization resistance, fractional capacitance value corresponding to polarization capacitance, and fractional order are obtained; a historical characteristic vector determination module, configured to determine a historical open circuit voltage estimate corresponding to the lithium-ion battery using the state-space equation and the state-space expression and based on a parameter identification result, and construct a historical characteristic vector based on the historical open circuit voltage estimate, historical terminal voltage measurements, and historical current measurements of the lithium-ion battery; A time series node feature determination module is used to extract features from the historical feature vector using the fused improved long short-term memory structure in the initial time series prediction model to obtain target time series features, and then use the time-enhanced attention mechanism in the initial time series prediction model to perform global correlation between features and time series dimension weighted enhancement operations on the target time series features to obtain time series node features; The state of charge estimation module is used to train the initial time series prediction model based on the time series node features and the corresponding true state of charge labels to obtain a target time series prediction model, so as to use the target time series prediction model and determine the state of charge estimation value based on the current terminal voltage measurement value, current current measurement value and current open circuit voltage estimation value corresponding to the lithium-ion battery.

9. An electronic device, characterized in that: include: Memory, used to store computer programs; A processor, configured to execute the computer program to implement the method for estimating the state of charge of a lithium-ion battery according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that Used to store a computer program, wherein when the computer program is executed by a processor, the method for estimating the state of charge of a lithium-ion battery according to any one of claims 1 to 7 is implemented.

Images