Lithium battery soc intelligent estimation method, electronic device and medium

Abstract

The application provides a lithium battery SOC intelligent estimation method, an electronic device and a medium. The method comprises the following steps: establishing a first-order equivalent circuit model of a lithium battery; fitting an end voltage based on the first-order equivalent circuit model to obtain a first fitted end voltage; establishing an end voltage prediction model of the lithium battery based on a characteristic time multilayer perception machine, and fitting the end voltage to obtain a second fitted end voltage; fusing the two fitted end voltages according to the voltage difference between the first fitted end voltage and an actual end voltage and the voltage difference between the second fitted end voltage and the actual end voltage to obtain a fused voltage; constructing a target state space model based on the first-order equivalent circuit model and the fused voltage; inputting a real-time end voltage, a real-time current and a real-time temperature of the lithium battery into the target state space model, and performing state estimation on the target state space model by using an adaptive unscented Kalman filtering algorithm to obtain an SOC estimation value. The scheme is helpful to realize accurate estimation of the SOC.

Description

Technical Field

[0001] This invention relates to the field of lithium-ion battery technology, and more specifically to a lithium battery SOC intelligent estimation method, electronic device, and medium. Background Technology

[0002] Against the backdrop of the global green energy transition, electrochemical energy storage has become a key support for new power systems due to the synergistic needs of "source-grid-load-storage". Lithium-ion batteries, with their advantages of high energy density, low self-discharge rate, and long lifespan, have become the core carrier of electrochemical energy storage. As a core functional module of the battery management system, the accurate estimation of the battery's state of charge (SOC) is crucial to the system's safety and efficiency. SOC refers to the ratio of the battery's remaining capacity to its fully charged state. It is a key parameter characterizing the remaining capacity of a lithium-ion battery, exhibiting characteristics such as nonlinearity, time-varying nature, and inability to be directly measured. Accurate SOC estimation faces significant challenges under multiple disturbances, including temperature fluctuations, aging degradation, and dynamic operating conditions.

[0003] In view of this, the present invention is hereby proposed. Summary of the Invention

[0004] The present invention is proposed in view of the above-mentioned problems. According to one aspect of the present invention, a lithium battery SOC intelligent estimation method is provided, comprising:

[0005] Establish a first-order equivalent circuit model for a lithium battery;

[0006] The first fitted terminal voltage is obtained by fitting the terminal voltage based on the first-order equivalent circuit model.

[0007] A terminal voltage prediction model for the lithium battery is established based on a feature-time multilayer perceptron.

[0008] Using the aforementioned terminal voltage prediction model, the terminal voltage is fitted to obtain a second fitted terminal voltage;

[0009] Based on the voltage difference between the first fitted terminal voltage and the actual terminal voltage, and the voltage difference between the second fitted terminal voltage and the actual terminal voltage, the first fitted terminal voltage and the second fitted terminal voltage are fused together to obtain a fused voltage.

[0010] Based on the first-order equivalent circuit model and the fused voltage, a target state-space model is constructed.

[0011] The real-time terminal voltage, real-time current, and real-time temperature of the lithium battery are input into the target state space model, and the state of the target state space model is estimated using an adaptive unscented Kalman filter algorithm to obtain the SOC estimate.

[0012] The filtering parameters of the adaptive unscented Kalman filter algorithm are determined based on the fused voltage obtained in the previous SOC estimation process.

[0013] For example, constructing the target state-space model based on the first-order equivalent circuit model and the fused voltage includes:

[0014] Based on the first-order equivalent circuit model, an initial state space model is constructed, which includes the state space equation of the lithium battery and the initial observation equation established based on the state space equation.

[0015] Substitute the fused voltage into the initial observation equation to obtain the target observation equation;

[0016] The target state-space model includes the state-space equation and the target observation equation.

[0017] For example, the state-space equation in the target state-space model is:

[0018] ;

[0019] The target observation equation of the target state-space model is:

[0020] ;

[0021] in, ; express k The estimated SOC value at time t; express k Polarization voltage at time; R p Indicates polarization resistance; C p Indicates polarization capacitance; Indicates the sampling time interval; Indicates the rated capacity; express k The current value at time -1; Indicates process noise; Indicates measurement noise; express k Fusion voltage at any given moment.

[0022] For example, before fitting the terminal voltage based on the first-order equivalent circuit model, the method further includes:

[0023] A multi-strategy improved serpentine optimization algorithm is used to identify the parameters of the first-order equivalent circuit model, and the parameters of the first-order equivalent circuit model are parameterized based on the identified parameters.

[0024] For example, the parameterized characterization results include:

[0025] ;

[0026] ;

[0027] ;

[0028] In the formula, This represents the ohmic internal resistance of the fit; Represents the polarization resistance of the fit; Represents the polarization capacitance of the fit; , , Represents a quadratic function related to the temperature of lithium batteries; , , In a polynomial, the th q A function of the order term as a function of temperature.

[0029] For example, fusing the first fitted terminal voltage and the second fitted terminal voltage based on the voltage difference between the first fitted terminal voltage and the actual terminal voltage, and the voltage difference between the second fitted terminal voltage and the actual terminal voltage, includes:

[0030] The weight of the first fitted terminal voltage is determined based on the ratio of the first voltage difference to the sum of voltage differences, wherein the first voltage difference is the voltage difference between the first fitted terminal voltage and the actual terminal voltage, the sum of voltage differences is the sum of the first voltage difference and the second voltage difference, and the second voltage difference is the voltage difference between the second fitted terminal voltage and the actual terminal voltage.

[0031] The weight of the second fitted terminal voltage is determined based on the ratio of the second voltage difference to the sum of the voltage differences.

[0032] The weighted sum of the first fitted terminal voltage and the second fitted terminal voltage is calculated to obtain the fused voltage.

[0033] For example, calculating the weighted sum of the first fitted terminal voltage and the second fitted terminal voltage includes calculation using the following formula:

[0034] ;

[0035] ;

[0036] ;

[0037] ;

[0038] In the formula, express k The first fitted terminal voltage at time 1; express k The second fitted terminal voltage at time 1; express k The first voltage difference at time; express k The second voltage difference at time; express k The weight of the first fitted terminal voltage at time step; express k The actual terminal voltage at that moment.

[0039] For example, the step of fitting the terminal voltage based on the first-order equivalent circuit model to obtain the first fitted terminal voltage includes calculating the first fitted terminal voltage using the following formula:

[0040] ;

[0041] In the formula, express k Open-circuit voltage at any given moment; express k Polarization voltage at time; express k Load current at any given moment; This represents the ohmic internal resistance of the fit;

[0042] Among them, the open-circuit voltage is related to the SOC value and temperature, and the ohmic internal resistance is related to the SOC value and temperature.

[0043] According to another aspect of the present invention, an electronic device is provided, including a processor and a memory, wherein the memory stores a computer program, and the processor is used to execute the computer program to implement the method as described above.

[0044] According to another aspect of the present invention, a computer-readable storage medium is provided, which stores a computer program / instructions that, when executed by a processor, implement the method described above.

[0045] In the above technical solution, by fusing an equivalent circuit model with good physical interpretation with a neural network model possessing strong fitting ability and high efficiency to establish the target space state equation, the neural network model can compensate for the insufficient fitting accuracy of the first-order equivalent circuit model. By fusing the first and second fitted terminal voltages based on the voltage difference between the first and actual terminal voltages, and the voltage difference between the second and actual terminal voltages, the two terminal voltages can be dynamically fused. This allows the neural network model to effectively compensate the equivalent circuit model when the voltage fitting accuracy is insufficient, minimizing the terminal voltage error and thus minimizing the SOC estimation error. By combining the adaptive unscented Kalman filter algorithm for closed-loop estimation of the lithium battery's SOC, the accuracy and robustness of the estimation can be effectively improved, enabling dynamic tracking of the lithium battery's operating state. This solution helps provide effective guidance for the charging and discharging management of lithium batteries.

[0046] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and in order to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0047] The above and other objects, features, and advantages of the present invention will become more apparent from the more detailed description of the embodiments of the invention in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same parts or steps.

[0048] Figure 1 A schematic flowchart of a lithium battery SOC intelligent estimation method according to an embodiment of the present invention is shown.

[0049] Figure 2 A schematic diagram of a first-order equivalent circuit model according to an embodiment of the present invention is shown;

[0050] Figure 3 A schematic diagram of a feature time multilayer perceptron according to an embodiment of the present invention is shown;

[0051] Figure 4 This illustrates the relationship between battery terminal voltage and SOC changes according to an embodiment of the present invention;

[0052] Figure 5 The diagram shows a comparison curve between the terminal voltage obtained by fitting a first-order equivalent circuit model according to an embodiment of the present invention and the actual terminal voltage.

[0053] Figure 6A schematic block diagram of an electronic device according to an embodiment of the present invention is shown. Detailed Implementation

[0054] To make the objectives, technical solutions, and advantages of the present invention more apparent, exemplary embodiments according to the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are merely a part of the embodiments of the present invention, and not all of the embodiments of the present invention. It should be understood that the present invention is not limited to the exemplary embodiments described herein. Based on the embodiments of the present invention described herein, all other embodiments obtained by those skilled in the art without inventive effort should fall within the protection scope of the present invention.

[0055] As mentioned above, accurate SOC estimation faces significant challenges under multiple disturbances such as temperature fluctuations, aging degradation, and dynamic operating conditions. Specifically, lithium battery SOC estimation methods are mainly divided into physical model methods and data-driven methods. However, existing physical model methods lack sufficient nonlinear processing capabilities, and data-driven methods suffer from weak interpretability of the data-driven models. In recent years, more and more researchers have focused on improving equivalent circuit models, attempting to enhance SOC estimation accuracy through more refined modeling. However, while improving estimation accuracy, this method inevitably introduces complexity to the model structure. In conclusion, there is an urgent need for a method with better nonlinear processing capabilities and interpretability, while also achieving high SOC estimation accuracy. Therefore, this invention provides a smart SOC estimation method, electronic device, and medium for lithium batteries. This method balances model simplification and physical interpretability, achieving better SOC estimation accuracy.

[0056] According to one aspect of the present invention, a method for intelligent estimation of the state of charge (SOC) of a lithium battery is provided. Figure 1 A schematic flowchart illustrating a lithium battery SOC intelligent estimation method according to an embodiment of the present invention is shown. Figure 1 As shown, the method may include steps S110, S120, S130, S140, S150, S160 and S170.

[0057] In step S110, a first-order equivalent circuit model of the lithium battery is established.

[0058] In this embodiment, a first-order RC model is used as the equivalent circuit model of the lithium battery. A voltage source represents the open-circuit voltage associated with SOC and temperature, an ohmic resistor characterizes the voltage drop caused by internal current excitation, and a parallel RC network describes the transient dynamic characteristics inside the lithium battery. Figure 2As shown, the first-order equivalent circuit model consists of a polarization loop composed of a polarization capacitor and a polarization resistor, and an internal resistance of one ohm to simulate the dynamic characteristics of a lithium battery. This first-order equivalent circuit model can be established using the following mathematical equations based on Kirchhoff's laws:

[0059] ;

[0060] ;

[0061] in, Indicates the terminal voltage of the lithium battery; This indicates the open-circuit voltage of the lithium battery; Indicates the polarization voltage of the lithium battery; Indicates the internal resistance of the ohm; Indicates polarization resistance; Indicates capacitance; This indicates the load current.

[0062] In this example scheme, to facilitate parameter identification, the polarization voltage is approximately discretized using an indicator function, resulting in a discrete-time form:

[0063] ;

[0064] In the formula, This represents the sampling time interval used to identify model parameters.

[0065] Furthermore, the state equations of the lithium battery equivalent circuit model are discretized using the Laplace transform, resulting in the following transfer function form:

[0066]

[0067] Furthermore, a bilinear transformation is used to transform the above... s Domain equations mapped to z The domain, where the bilinear transformation formula is:

[0068] ;

[0069] The transformed transfer function is:

[0070] ;

[0071] In the above formula, c 1. c 2. c 3. Intermediate variables introduced to facilitate the identification of model parameters.

[0072] After differential discretization of the above transfer function, the following time-domain expression is obtained:

[0073] ;

[0074] Meanwhile, the voltage residual is defined as the difference between the terminal voltage and the open-circuit voltage, that is:

[0075] .

[0076] In this embodiment, a low-current test method is used to obtain the relationship between open-circuit voltage and state of charge (SOC) at different temperatures, and the result is fitted as a function of the following form:

[0077] ;

[0078] In the formula, This represents the functional relationship between the battery open-circuit voltage and the state of charge (SOC). Represents the coefficient of the constant term. In a polynomial, the th p Order term coefficients.

[0079] Meanwhile, by analyzing the function results obtained from fitting at different temperatures, it was found that battery parameters not only change with SOC but also with operating temperature. Based on this, in this embodiment, each coefficient of the polynomial function is represented as a quadratic function of temperature T, making it temperature-dependent, thereby extending the univariate function of battery parameters with respect to SOC into a bivariate function with respect to both SOC and temperature T.

[0080] ;

[0081] In the formula: This represents the functional relationship between the battery open-circuit voltage and the state of charge (SOC) and temperature. Represents a quadratic function related to the temperature of lithium batteries; Let m represent the function of the p-th order term in the polynomial as a function of temperature, and m be the highest order of the polynomial to be fitted. During the function fitting process, it can be observed that when the polynomial order is too high, overfitting is likely to occur; while when the order is too low, underfitting may occur. When the highest order m of the polynomial is 7, a good fit can be achieved for the battery open-circuit voltage.

[0082] Based on this, the voltage residual value at each time point can be calculated, thus representing the first-order equivalent circuit model of the lithium battery in the following regression form:

[0083] ;

[0084] In the formula, the input vector parameter vector By input k The voltage residual at time -1 and k Time and k The load current at time -1 is estimated. c1. c 2. c 3. Three intermediate variables.

[0085] In this embodiment, a hybrid pulse power characteristic test is used to obtain test data to facilitate fitting the parameters in the model (including terminal voltage, current, and temperature data during the lithium battery charging and discharging process). The hybrid pulse power characteristic test is a standard test method for characterizing the pulse charge and discharge performance of lithium batteries and is widely used for identifying parameters of the equivalent circuit model of lithium batteries. A series of pulse currents are applied at different SOC levels, and by analyzing the response characteristics of the lithium battery under dynamic current excitation, its dynamic behavior is revealed, thereby establishing an equivalent circuit model that reflects the characteristics of the lithium battery. The specific steps of this test process are understood by those in the art and will not be elaborated upon.

[0086] In some embodiments, after obtaining the test data, outlier removal and missing value imputation can be performed on the data, which helps to obtain a complete, high-quality dataset.

[0087] In step S120, the terminal voltage is fitted based on a first-order equivalent circuit model to obtain the first fitted terminal voltage.

[0088] After establishing the first-order equivalent circuit model, the fitted terminal voltage (i.e., the first fitted terminal voltage) under the first-order equivalent circuit model can be obtained based on the state equation of the first-order equivalent circuit model and the discrete-time form of the polarization voltage.

[0089] ;

[0090] In the formula: , , and They respectively represent the first-order equivalent circuit model in k The terminal voltage, open-circuit voltage, polarization voltage, and load current at any given time; express k SOC value at time t; This represents the fitted ohmic internal resistance.

[0091] In step S130, a prediction model for the terminal voltage of the lithium battery is established based on a feature time multilayer perceptron.

[0092] In this embodiment, based on the voltage, current and temperature data collected during the actual operation of the lithium battery, a three-input one-output characteristic time multilayer perceptron model is constructed to establish a nonlinear mapping relationship between the battery terminal voltage and historical voltage, current and temperature, which helps to achieve high-precision prediction of the battery terminal voltage.

[0093] In step S140, the terminal voltage is fitted using the terminal voltage prediction model to obtain the second fitted terminal voltage.

[0094] In this embodiment, to effectively capture the temporal dependence and inter-feature interactions of multidimensional input data, a Feature-temporal multilayer perceptron (FTMLP) model is selected to construct a terminal voltage prediction neural network. The model structure is as follows: Figure 3 As shown, this model introduces feature blocks and time blocks to enhance the expressive power of input data in the feature dimension and time dimension, respectively, making it suitable for handling historical multivariate sequence prediction tasks.

[0095] The FTMLP model consists of two core modules: feature blocks and temporal blocks. The feature blocks are used to extract the static correlations of the input sequences along the feature dimensions, and the specific steps include steps 1, 2, and 3.

[0096] Step 1, Input Compression: Compress the input matrix Global average pooling is performed on each row (i.e., the time series of each variable) to obtain the feature statistics vector: In the formula: Combine the compressed representations of all variables into .

[0097] Step 2, Nonlinear Feature Modeling: The compressed vector is nonlinearly projected using an MLP structure, and the activation function GeLU is introduced. .

[0098] Step 3, Feature Interaction Enhancement: Expand the output of the MLP back to the original input shape. Multiply element-wise with the original input and perform residual concatenation: In the formula: This represents the Sigmoid function. This indicates element-wise multiplication.

[0099] Time blocks are used to mine the dependency features of the input sequence in the time dimension, and mainly include the following steps 1, 2, 3 and 4.

[0100] Step 1, Frequency Domain Transformation: Convert the input matrix... Perform a Fast Fourier Transform to convert to the frequency domain: In the formula: This indicates the Fast Fourier Transform operation.

[0101] Step 2, Frequency Domain Filtering: To modulate the spectrum, a learnable frequency filter is used. Perform element-wise multiplication: .

[0102] Step 3, Inverse Transform: Use the inverse Fourier transform to restore the frequency domain information back to the time domain: .

[0103] Step 4, Temporal Feature Extraction: Input the feature into the MLP for nonlinear temporal modeling, and maintain information consistency through the residual structure. .

[0104] Finally, the predicted trend component and seasonal component are weighted and summed to obtain the final output: In the formula: and The two components are obtained by inverse normalization of the linear layer RevIN after prediction by the aforementioned feature block and time block models, respectively.

[0105] In this embodiment, the input to the FTMLP model is historical time. arrive terminal voltage V t Current I and temperature T The sequence is output as the predicted time. k Terminal voltage: In the formula: This indicates that the neural network model at time 10 is... k The terminal voltage. , and These represent the voltage, current, and temperature data of the lithium battery, respectively. H This indicates the size of the sliding window.

[0106] Compared to traditional MLP, FTMLP models are more suitable for modeling the diverse operating conditions and strong coupling of input dimensions in lithium batteries, enabling higher-precision voltage prediction and providing a stable reference for subsequent steps.

[0107] In step S150, the first fitted terminal voltage and the second fitted terminal voltage are fused together based on the voltage difference between the first fitted terminal voltage and the actual terminal voltage and the voltage difference between the second fitted terminal voltage and the actual terminal voltage to obtain a fused voltage.

[0108] State of charge (SOC) cannot be measured directly, but its changes can significantly affect the battery's terminal voltage. Figure 4 This illustrates the relationship between battery terminal voltage and SOC changes according to one embodiment of the present invention. Figure 4 As shown in the figure, under the mixed pulse power characteristic test conditions, the battery terminal voltage changes significantly as the SOC gradually decreases from 1 to 0, indicating that the change in SOC has a direct impact on the voltage. Figure 5 The diagram shows a comparison curve between the terminal voltage obtained by fitting a first-order equivalent circuit model according to an embodiment of the present invention and the actual terminal voltage. For example... Figure 5 As shown, the identification results of the first-order equivalent circuit model basically coincide with the actual voltage curve measured in the experiment, with a small overall error. Under different SOC conditions, the model voltage can track the dynamic change trend of the actual battery voltage well, demonstrating the effectiveness of the equivalent circuit model. However, in the voltage recovery stage after the pulse current ends, there is still a significant error between the model-fitted voltage and the measured voltage, reflecting the inadequacy of the first-order equivalent circuit model in describing the hysteresis effect and dynamic recovery characteristics of the battery voltage. Therefore, this embodiment considers using a terminal voltage prediction model based on FTMLP to compensate for the insufficient accuracy of the terminal voltage fitting of the first-order equivalent circuit model. The terminal voltage fitted by the first-order equivalent circuit model and the terminal voltage predicted by the terminal voltage prediction model are fused to obtain a fused voltage, which is used as the basis for correcting the SOC estimation accuracy in subsequent processes.

[0109] In step S160, a target state-space model is constructed based on the first-order equivalent circuit model and the fused voltage.

[0110] It is understood that a state-space model typically includes state-space equations and observation equations based on the state-space equations. In this embodiment, the fused voltage is incorporated into the state-space model. This allows the final target state-space model to integrate an equivalent circuit model with good physical interpretation with a neural network model with strong fitting ability and high efficiency, thereby helping to improve the accuracy and reliability of the model.

[0111] In step S170, the real-time terminal voltage, real-time current, and real-time temperature of the lithium battery are input into the target state space model, and the state of the target state space model is estimated using an adaptive unscented Kalman filter algorithm to obtain the SOC estimate; wherein, the filtering parameters of the adaptive unscented Kalman filter algorithm are determined based on the fused voltage obtained in the previous SOC estimation process.

[0112] It is understandable that when performing state estimation using the adaptive unscented Kalman filter algorithm, parameters such as the Kalman gain matrix need to be considered in conjunction with the previous state estimation (i.e., ...). k The observation matrix at time -1 is determined. In this embodiment, the fused voltage is incorporated into the state-space model. Thus, in the next SOC estimation, the fused voltage calculated previously is used, and the fused voltage obtained after each SOC estimation can correct the filtering parameters for the next SOC estimation, which helps improve the accuracy of SOC estimation.

[0113] In this invention, the state of charge (SOC) of a lithium battery is estimated by fusing an equivalent circuit model and a neural network model. A first-order equivalent circuit model is chosen for the equivalent circuit modeling part, while the FTMLP network is used for error compensation and performance enhancement in the neural network part. This combination is not arbitrary but based on specific modeling objectives and engineering requirements. The first-order equivalent circuit model, due to its simple structure, low computational complexity, and stable parameter identification, is suitable for the real-time and reliability requirements of embedded systems; simultaneously, it already possesses the ability to describe the main dynamic behaviors of the battery and can effectively provide a stable baseline estimate. The FTMLP neural network, on the other hand, has excellent time-frequency feature fusion capabilities. By combining the frequency domain representation of time-series data with the original time-domain input, it significantly enhances the model's ability to model the nonlinear characteristics and long-term dependent behaviors of the battery, making it particularly suitable for capturing voltage dynamic changes under complex operating conditions. While maintaining the model's expressive power, FTMLP can also reduce sensitivity to high-frequency noise, improving the convergence speed and generalization performance of network training. In summary, this invention effectively combines a first-order equivalent circuit model with stability and physical interpretability with an FTMLP neural network with strong time-series modeling capabilities, achieving an optimal balance between accuracy, robustness, and real-time performance. It proves that the selection and fusion of the two types of models have clear theoretical basis and practical value, providing a lightweight and high-precision modeling scheme for SOC estimation.

[0114] The above technical solution establishes the target space state equation by fusing an equivalent circuit model with good physical interpretation with a neural network model possessing strong fitting ability and high efficiency. The neural network model can compensate for the insufficient fitting accuracy of the first-order equivalent circuit model. By fusing the first and second fitted terminal voltages based on the voltage difference between the first and actual terminal voltages, and the voltage difference between the second and actual terminal voltages, the two terminal voltages can be dynamically fused. This allows the neural network model to effectively compensate for insufficient voltage fitting accuracy in the equivalent circuit model, minimizing terminal voltage errors and thus minimizing SOC estimation errors. By combining an adaptive unscented Kalman filter algorithm for closed-loop estimation of the lithium battery's SOC, the accuracy and robustness of the estimation can be effectively improved, enabling dynamic tracking of the lithium battery's operating state. This solution helps provide effective guidance for the charging and discharging management of lithium batteries.

[0115] For example, before fitting the terminal voltage based on the first-order equivalent circuit model in step S120, the method may further include the following step S180.

[0116] In step S180, the parameters of the first-order equivalent circuit model are identified using a multi-strategy improved snake optimization algorithm, and the parameters of the first-order equivalent circuit model are parameterized based on the identified parameters.

[0117] In this method, the optimization objective is to optimize the parameter vector. The objective function for minimizing the sum of squared residuals between the model output and the actual measured values ​​is defined as follows: In the formula: This represents the measured voltage residual.

[0118] In some embodiments, the specific steps of improving the snake optimization algorithm are as follows:

[0119] Step 1, Initialization Phase: Use a multi-strategy chaotic system to generate an initial population with high randomness and uniform distribution, thereby improving the quality of the initial solution and enhancing the robustness of the algorithm.

[0120] Step 2, Fitness Calculation: For each individual According to the objective function Calculate the fitness score; the smaller the score, the better the fit.

[0121] Step 3, Population Evolution Strategy: In the exploration phase, an anti-predation strategy is introduced to mathematically simulate the natural behavior of snakes escaping predators, encouraging the population to break out of local optima; in the development phase, a two-way population evolution mechanism is introduced to perform fine mutations on the top 20% of individuals in terms of fitness, while the rest are eliminated or migrated to enhance the global optimization ability.

[0122] Step 4, Termination Condition: When the maximum number of iterations or the error threshold is reached, output the optimal parameters. .

[0123] In this embodiment, a multi-strategy improved snake optimization algorithm is used to identify the parameters of the first-order equivalent circuit model. Considering the characteristics of strong nonlinearity, high dynamics, and frequent changes in operating conditions of battery parameters, it exhibits significant performance advantages over traditional methods, making it the optimal and targeted choice in this scheme. Compared to the traditional Forgotten Factor Least Squares (FFRLS) method, this method not only has advantages in identification accuracy but also excels in adaptability and stability. Although FFRLS has high computational efficiency and is suitable for online recursive updates, its core problem lies in its strong dependence on the forgetting factor. Improper selection of the forgetting factor can easily lead to estimation divergence or system response lag. Especially when facing dynamic changes in battery parameters with temperature, state of charge, or aging, its identification results often fail to remain stable and are significantly affected by model structure errors and noise interference. In contrast, the multi-strategy improved snake optimization algorithm, by simulating the cooperative behavior of snake groups and integrating multi-strategy chaotic systems, anti-predator strategies, and bidirectional population evolution mechanisms, possesses strong global search capabilities and local convergence accuracy, enabling efficient search in complex, non-convex, and multi-peak parameter spaces. Furthermore, compared to common intelligent optimization algorithms such as Particle Swarm Optimization (PSO), this algorithm better addresses the problems of premature convergence and insufficient population diversity in PSO, exhibiting more stable convergence and higher accuracy, especially in high-dimensional spaces. Its optimization process does not rely on gradient information, enabling it to handle non-differentiable, noisy, or highly uncertain measurement problems, thus demonstrating stronger robustness and practical adaptability. In addition, this algorithm is insensitive to initial values ​​and offers flexible search strategies, allowing for adjustments based on different modeling needs. In summary, this embodiment improves the snake-like optimization algorithm by specifically selecting multiple strategies to identify parameters of the first-order equivalent circuit model, significantly improving parameter identification accuracy. This facilitates accurate fitting of the terminal voltage, thereby contributing to improved SOC estimation accuracy.

[0124] After achieving optimal parameters After identification, the ohmic internal resistance, polarization capacitance, and polarization internal resistance parameters in the first-order equivalent circuit model of a lithium battery can be calculated using the following mathematical expressions:

[0125] .

[0126] In this embodiment, for the equivalent circuit model parameters identified under different temperature and SOC conditions, a parameterized characterization function is constructed using the same polynomial fitting method as for the open-circuit voltage, thereby establishing a mapping relationship between the model parameters and SOC and temperature. The parameterized characterization results include:

[0127] ;

[0128] ;

[0129] ;

[0130] In the formula, This represents the ohmic internal resistance of the fit; Represents the polarization resistance of the fit; Represents the polarization capacitance of the fit; , , Represents a quadratic function related to the temperature of lithium batteries; , , In a polynomial, the th q The term is a function of temperature. This is understandable. , , They represent the ohmic internal resistance respectively. Polarization resistance and polarization capacitor The function representation of . The parameters are similar and will not be elaborated upon.

[0131] In this embodiment, n =6. During the fitting process, it was found that when the highest order... n When the order is 6, a good fit to the data can be achieved. Of course, the highest order can be chosen reasonably according to the actual situation to avoid overfitting and underfitting.

[0132] The above scheme establishes a mapping relationship between ohmic internal resistance, polarization resistance, polarization capacitance, SOC, and temperature, which facilitates accurate estimation of SOC in subsequent steps.

[0133] For example, step S120, fitting the terminal voltage based on a first-order equivalent circuit model to obtain a first fitted terminal voltage, includes calculating the first fitted terminal voltage using the following formula:

[0134] ;

[0135] In the formula, express k Open-circuit voltage at any given moment; express k Polarization voltage at time; express k Load current at any given moment; The ohmic internal resistance is the fitted value; where the open-circuit voltage is related to the SOC value and temperature, and the ohmic internal resistance is related to the SOC value and temperature.

[0136] As described above, battery parameters not only change with SOC but also with operating temperature. In this embodiment, by fitting the first fitting terminal voltage using open-circuit voltage and ohmic resistance, which are dependent on SOC and temperature, the accuracy of the terminal voltage fitting can be improved.

[0137] For example, step S150, which fuses the first fitted terminal voltage and the second fitted terminal voltage based on the voltage difference between the first fitted terminal voltage and the actual terminal voltage and the voltage difference between the second fitted terminal voltage and the actual terminal voltage, includes the following steps S151, S152 and S153.

[0138] In step S151, the weight of the first fitted terminal voltage is determined according to the ratio of the first voltage difference to the sum of voltage differences, wherein the first voltage difference is the voltage difference between the first fitted terminal voltage and the actual terminal voltage, the sum of voltage differences is the sum of the first voltage difference and the second voltage difference, and the second voltage difference is the voltage difference between the second fitted terminal voltage and the actual terminal voltage.

[0139] In step S152, the weight of the second fitting terminal voltage is determined based on the ratio of the second voltage difference to the sum of voltage differences.

[0140] In step S153, the weighted sum of the first fitting terminal voltage and the second fitting terminal voltage is calculated to obtain the fused voltage.

[0141] In this embodiment, the voltage difference between the first fitted terminal voltage and the actual terminal voltage is the absolute value of the difference between the first fitted terminal voltage and the actual terminal voltage, and the voltage difference between the second fitted terminal voltage and the actual terminal voltage is the absolute value of the difference between the second fitted terminal voltage and the actual terminal voltage.

[0142] In this embodiment, the ratio of the voltage difference to the sum of voltage differences is used as the weight of the corresponding fitted terminal voltage to perform weighted fusion of the two fitted terminal voltages. This weighted fusion method based on the ratio of the voltage difference to the sum of voltage differences achieves high-precision terminal voltage fitting through adaptive weight allocation: when the error of a fitted terminal voltage is small (i.e., the voltage difference is small), its corresponding weight will automatically increase; conversely, when the error is large, the weight will decrease, thereby effectively suppressing the influence of large-error fitted values ​​on the overall result. At the same time, the normalization of the sum of voltage differences ensures that the total weight is always 1, avoiding extreme weight allocation and enhancing the stability of the fusion result. In addition, this method only requires simple calculation of the voltage difference ratio without complex parameter adjustments, possessing real-time performance and computational efficiency. It is particularly suitable for scenarios with high requirements for accuracy and robustness, and helps to output fusion results close to the true value, essentially realizing intelligent collaboration of "the smaller the error, the larger the weight".

[0143] For example, the weighted sum of the first fitted terminal voltage and the second fitted terminal voltage is calculated using the following formula:

[0144] ;

[0145] ;

[0146] ;

[0147] ;

[0148] In the formula, express k The first fitted terminal voltage at time 1; express k The second fitted terminal voltage at time 1; express k The first voltage difference at time; express k The second voltage difference at time; express k The weight of the first fitted terminal voltage at time step; express k The actual terminal voltage at that moment.

[0149] The above scheme integrates the two fitted terminal voltages through adaptive dynamic weights, which can improve the compensation effect of the neural network model on the equivalent circuit model, and thus help improve the estimation accuracy of SOC.

[0150] For example, step S160, based on the first-order equivalent circuit model and the fused voltage, constructs a target state space model, including: constructing an initial state space model based on the first-order equivalent circuit model, the initial state space model including the state space equation of the lithium battery and the initial observation equation established based on the state space equation; substituting the fused voltage into the initial observation equation to obtain the target observation equation; wherein, the target state space model includes the state space equation and the target observation equation.

[0151] In this example, the scheme considers incorporating the fused voltage into the initial observation equations. It can be understood that the observation equations are used to improve the accuracy of state estimation. By incorporating the fused voltage into the initial observation equations, the terminal voltage prediction model can be used to compensate for the first-order equivalent circuit model, minimizing the terminal voltage error, thereby reducing the SOC estimation error and improving the SOC estimation accuracy.

[0152] For example, the state-space equations in the target state-space model are:

[0153] ;

[0154] The target observation equation of the target state-space model is:

[0155] ;

[0156] Among them, state variables ; express k The estimated SOC value at time t; express k Polarization voltage at time; R p Indicates polarization resistance; C p Indicates polarization capacitance; Indicates the sampling time interval; Indicates the rated capacity; express k The current value at time -1; Indicates process noise; Indicates measurement noise; express k Fusion voltage at any given moment.

[0157] It is understandable that SOC exhibits significant nonlinearity and time-varying characteristics. SOC can be defined as:

[0158] ;

[0159] In the formula: I For battery charging and discharging current, Indicates the battery at time k SOC, Indicates the rated capacity. Indicates the sampling time interval. The Coulomb efficiency is represented by a constant 1 in this invention.

[0160] In traditional equivalent circuit modeling, combining the lithium battery SOC definition and RC network response, the state equation of the lithium battery equivalent circuit model can be obtained as follows:

[0161] .

[0162] The observation equations that combine open-circuit voltage and polarization voltage can be expressed as follows:

[0163] ;

[0164] Where: state variables Input vector u For battery charging and discharging current, u k-1 It means k The current value at time -1. and These represent process noise and measurement noise, respectively. It is a nonlinear state transition function. It can be called the observation function. This represents the model output, i.e., the estimated lithium battery terminal voltage.

[0165] Traditional voltage fitting methods are typically based on equivalent circuit models, representing the output voltage as a combination of open-circuit voltage, polarization voltage, and current resistance voltage drop. While this method has a clear physical meaning and is easy to understand and implement in engineering, its accuracy is highly dependent on the quality of RC parameter modeling and the fitting accuracy of open-circuit voltage and SOC. It is difficult to cope with nonlinear errors under complex operating conditions, such as model deviations caused by low-temperature conditions, battery aging, or sudden load changes.

[0166] Therefore, this embodiment introduces a dual-model fusion structure to replace the traditional voltage fitting method, and expresses the output voltage as:

[0167] ;

[0168] In the formula: The fused voltage is the result of the fusion of the physical model output and the data model output.

[0169] This embodiment employs a dual-model fusion mechanism. The physical model characterizes the main trends in battery voltage changes, ensuring the physical rationality and stability of the modeling structure. The data model compensates for the fitting bias of the physical model under complex operating conditions, effectively enhancing its responsiveness to nonlinear behavior. This fusion strategy combines accuracy and interpretability, significantly improving the accuracy of battery state estimation and the model's adaptability.

[0170] In this embodiment, the specific steps for state estimation of the target state space model using the adaptive unscented Kalman filter algorithm are as follows: Steps 1-10.

[0171] Step 1: Initialize state variables x The mean and covariance matrix P :

[0172] ;

[0173] ;

[0174] In the formula: E This represents the expected value, which is the average value. Represents the initial state variable. This represents the average value of the initial state variables. Let represent the initial covariance matrix.

[0175] Step 2: Calculate the mean weight and covariance weight:

[0176] ;

[0177] ;

[0178] ;

[0179] In the formula: and These are the mean weight and the covariance weight, respectively; for a Gaussian random variable, ; n The dimension of the state variable; The scale correction factor can be calculated using the following expression: In the formula: and These are all scale parameters. It describes the distance between the sample point and the mean point.

[0180] Step 3: Obtain the result through singular value decomposition. k The 2n+1 Sigma sampling points at time -1:

[0181] ;

[0182] ;

[0183] In the formula: In the constructed Sigma point set, the first... i One point. for k The mean of the state estimate at time -1 is used as the center for constructing the Sigma point. Indicates to be decomposed k Covariance matrix at time -1. and They are two orthogonal matrices, derived from the pair The singular value decomposition of . It is the diagonal matrix obtained by the singular value decomposition of the covariance matrix.

[0184] Step 4: Further predict the state variables and covariance matrix:

[0185] ;

[0186] In the formula: It is the nonlinear state transition function of the system. Indicates the previously generated first i Sigma points. Representation system kThe input at time -1 refers to the battery current. The result of propagating the Sigma point through the system function is used for state prediction at the next time step. For the current moment k The mean of the state prediction. for k The state covariance prediction matrix at time 1. This represents the system process noise covariance matrix.

[0187] Step 5, in k At time t, by analyzing the state covariance matrix Perform singular value decomposition to construct a 2-valued vector for observing propagation. n +1 Sigma sampling point:

[0188] ;

[0189] .

[0190] In the formula: express k The mean of the predicted state at time t. It is the state covariance matrix The diagonal matrix obtained by performing singular value decomposition. and It is the corresponding left-right orthogonal matrix.

[0191] Step 6: Substitute the Sigma points constructed in Step 5 into the observation function to complete the nonlinear observation propagation, and calculate the weighted average of the observation predictions:

[0192] ;

[0193] In the formula: It is the observation function. For the first i The predicted values ​​of the observed points after propagation through the observation function. This is the weighted predicted mean of the observed values.

[0194] Step 7: Calculate the covariance matrix of the predicted observations and the cross-covariance matrix between the state and the observations:

[0195] ;

[0196] .

[0197] In the formula: The covariance matrix for predicting observations includes the covariance of process uncertainty and observation noise. . This represents the covariance between the state and the observed variables, which is subsequently used to calculate the Kalman gain.

[0198] Step 8: Calculate the Kalman gain matrix :

[0199] .

[0200] Step 9: Update the state matrix and error covariance matrix:

[0201] .

[0202] In the formula: for k State estimate at time 1. This is the mean of the state prediction. These are actual observed values. In this invention, the observed value is a fused voltage value, which is used to predict the observed value. This is the updated covariance matrix.

[0203] Step 10: Adaptively adjust the process noise covariance based on the error. and measurement noise covariance :

[0204] Error residual is defined as: ;

[0205] Estimating observation covariance error using the sliding window method: ;

[0206] Further updates to process noise covariance and measurement noise covariance: ;

[0207] .

[0208] In the formula: Is the battery model in the current k Voltage residual at time t. Let be the residual covariance matrix. M This is the window size for covariance matching, which is set to 5 in this invention.

[0209] Therefore, based on the established battery model and data samples, the estimation accuracy of the model can be effectively verified, thereby providing support for improving the accuracy of lithium battery SOC estimation.

[0210] According to another aspect of the present invention, an electronic device is also provided. Figure 6 A schematic block diagram of an electronic device according to an embodiment of the present invention is shown. Figure 6As shown, the electronic device 600 includes a processor 610 and a memory 620. The memory 620 stores a computer program, which the processor 610 executes to implement the method described above.

[0211] According to another aspect of the present invention, a computer-readable storage medium is also provided. The storage medium stores a computer program / instructions that, when executed by a processor, implement the method described above. The storage medium may, for example, include a read-only memory (ROM), an erasable programmable read-only memory (EPROM), a portable compact disc read-only memory (CD-ROM), a USB memory, or any combination of the above storage media. The computer-readable storage medium may be any combination of one or more computer-readable storage media.

[0212] Those skilled in the art will readily understand the implementation structure, working principle, and beneficial effects of electronic devices and computer-readable storage media by reading the above methods. For the sake of brevity, further details will not be elaborated upon here.

[0213] Although exemplary embodiments have been described herein with reference to the accompanying drawings, it should be understood that the above exemplary embodiments are merely illustrative and are not intended to limit the scope of the invention. Various changes and modifications can be made therein by those skilled in the art without departing from the scope and spirit of the invention.

[0214] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0215] In the several embodiments provided by this invention, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed.

[0216] Numerous specific details are set forth in the specification provided herein. However, it will be understood that embodiments of the invention may be practiced without these specific details. In some instances, well-known methods, structures, and techniques have not been shown in detail so as not to obscure the understanding of this specification.

[0217] Similarly, it should be understood that, in order to streamline the invention and aid in understanding one or more of the various aspects of the invention, in the description of exemplary embodiments of the invention, various features of the invention are sometimes grouped together into a single embodiment, figure, or description thereof.

[0218] Those skilled in the art will understand that, apart from the mutual exclusion of features, all features disclosed in this specification and all processes or units of any method or apparatus so disclosed can be combined in any combination. Unless otherwise expressly stated, each feature disclosed in this specification may be replaced by an alternative feature that serves the same, equivalent, or similar purpose.

[0219] Furthermore, those skilled in the art will understand that although some embodiments described herein include certain features included in other embodiments but not others, combinations of features from different embodiments are meant to be within the scope of the invention and form different embodiments.

[0220] The various component embodiments of the present invention can be implemented in hardware, or as software modules running on one or more processors, or a combination thereof. Those skilled in the art will understand that microprocessors or digital signal processors (DSPs) can be used in practice to implement some or all of the functions of some modules in the electronic device according to embodiments of the present invention. The present invention can also be implemented as an apparatus program (e.g., a computer program and computer program product) for performing some or all of the methods described herein. Such programs implementing the present invention can be stored on a computer-readable medium or can be in the form of one or more signals. Such signals can be downloaded from an Internet website, provided on a carrier signal, or provided in any other form.

[0221] The above description is merely a specific embodiment of the present invention or an explanation of the specific embodiment. The scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for intelligent estimation of SOC of lithium batteries, characterized in that, include: Establish a first-order equivalent circuit model for a lithium battery; The first fitted terminal voltage is obtained by fitting the terminal voltage based on the first-order equivalent circuit model. A terminal voltage prediction model for the lithium battery is established based on a feature-time multilayer perceptron. Using the aforementioned terminal voltage prediction model, the terminal voltage is fitted to obtain a second fitted terminal voltage; Based on the voltage difference between the first fitted terminal voltage and the actual terminal voltage, and the voltage difference between the second fitted terminal voltage and the actual terminal voltage, the first fitted terminal voltage and the second fitted terminal voltage are fused together to obtain a fused voltage. Based on the first-order equivalent circuit model and the fused voltage, a target state-space model is constructed. The real-time terminal voltage, real-time current, and real-time temperature of the lithium battery are input into the target state space model, and the state of the target state space model is estimated using an adaptive unscented Kalman filter algorithm to obtain the SOC estimate. The filtering parameters of the adaptive unscented Kalman filter algorithm are determined based on the fused voltage obtained in the previous SOC estimation process. The construction of the target state-space model based on the first-order equivalent circuit model and the fused voltage includes: Based on the first-order equivalent circuit model, an initial state space model is constructed, which includes the state space equation of the lithium battery and the initial observation equation established based on the state space equation. Substitute the fused voltage into the initial observation equation to obtain the target observation equation; The target state-space model includes the state-space equation and the target observation equation.

2. The method according to claim 1, characterized in that, The state-space equations in the target state-space model are: ； The target observation equation of the target state-space model is: ； in, ; express k The estimated SOC value at time t; express k Polarization voltage at time; R p Indicates polarization resistance; C p Indicates polarization capacitance; Indicates the sampling time interval; Indicates the rated capacity; express k The current value at time -1; Indicates process noise; Indicates measurement noise; express k The fusion voltage at any given moment.

3. The method according to claim 1, characterized in that, Before fitting the terminal voltage based on the first-order equivalent circuit model, the method further includes: A multi-strategy improved serpentine optimization algorithm is used to identify the parameters of the first-order equivalent circuit model, and the parameters of the first-order equivalent circuit model are parameterized based on the identified parameters.

4. The method according to claim 3, characterized in that, The parameterized characterization results include: ； ； ； In the formula, This represents the ohmic internal resistance of the fit; Represents the polarization resistance of the fit; Represents the polarization capacitance of the fit; , , A quadratic function relating to lithium battery temperature; , , Represents the first polynomial in the polynomial. q A function of the order term as a function of temperature.

5. The method according to claim 1, characterized in that, The step of fusing the first fitted terminal voltage and the second fitted terminal voltage based on the voltage difference between the first fitted terminal voltage and the actual terminal voltage, and the voltage difference between the second fitted terminal voltage and the actual terminal voltage, includes: The weight of the first fitted terminal voltage is determined based on the ratio of the first voltage difference to the sum of voltage differences, wherein the first voltage difference is the voltage difference between the first fitted terminal voltage and the actual terminal voltage, the sum of voltage differences is the sum of the first voltage difference and the second voltage difference, and the second voltage difference is the voltage difference between the second fitted terminal voltage and the actual terminal voltage. The weight of the second fitted terminal voltage is determined based on the ratio of the second voltage difference to the sum of the voltage differences. The weighted sum of the first fitted terminal voltage and the second fitted terminal voltage is calculated to obtain the fused voltage.

6. The method according to claim 5, characterized in that, The calculation of the weighted sum of the first fitted terminal voltage and the second fitted terminal voltage includes calculation using the following formula: ； ； ； ； In the formula, express k The first fitted terminal voltage at time 1; express k The second fitted terminal voltage at time 1; express k The first voltage difference at time; express k The second voltage difference at time; express k The weight of the first fitted terminal voltage at time step; express k The actual terminal voltage at that moment.

7. The method according to claim 1, characterized in that, The step of fitting the terminal voltage based on the first-order equivalent circuit model to obtain the first fitted terminal voltage includes calculating the first fitted terminal voltage using the following formula: ； In the formula, express k Open-circuit voltage at any given moment; express k Polarization voltage at time; express k Load current at any given moment; This represents the ohmic internal resistance of the fit; Among them, the open-circuit voltage is related to the SOC value and temperature, and the ohmic internal resistance is related to the SOC value and temperature.

8. An electronic device, characterized in that, It includes a processor and a memory, wherein the memory stores a computer program, and the processor is used to execute the computer program to implement the method as claimed in any one of claims 1-7.

9. A computer-readable storage medium, characterized in that, The system stores a computer program / instructions that, when executed by a processor, implement the method as described in any one of claims 1-7.

Images