A lithium ion battery soc and soh combined estimation method

Abstract

This invention discloses a joint estimation method for SOC and SOH of lithium-ion batteries, comprising the following steps: establishing a circuit model based on a first-order fractional equivalent circuit model of a lithium-ion power battery; designing battery aging experiments and HPPC test experiments for the circuit model; extracting the SOC-OCV curve; establishing the state-space equation of the lithium-ion power battery; constructing a parameter identifier based on a genetic algorithm to identify parameters of the first-order fractional equivalent circuit model; constructing an unscented Kalman filter based on the state equation and observation equation; constructing an unscented Kalman filter based on the nearest symmetric matrix algorithm of the F-norm to construct an SOC state estimator for SOC state estimation; inputting SOC data, terminal voltage, and terminal current into a BP neural network for training; constructing an SOH state estimator to achieve online SOH estimation; and coupling the parameter identifier and the state estimator to achieve joint estimation of SOC and SOH. This invention fully considers the coupling relationship between SOC and SOH, improving the estimation accuracy of SOC and SOH.

Description

Technical Field

[0001] This invention relates to the field of lithium-ion battery technology, and specifically to a method for jointly estimating the SOC and SOH of a lithium-ion battery. Background Technology

[0002] The state of charge (SOC) and state of health (SOH) of electric vehicle power batteries are important parameters of battery operation and key state variables for battery management systems to control the batteries. SOC reflects the current state of charge storage, typically expressed as the ratio of discharged capacity to current usable capacity, describing short-term state changes. SOH reflects the aging state over the battery's entire life cycle, typically expressed as the ratio of current usable capacity to initial capacity, describing the degree of battery degradation under different cycles. Accurate estimation of SOH is helpful for battery health diagnosis and timely replacement of degraded batteries. However, SOH and SOC are strongly coupled and influence each other; a single state estimation often cannot achieve sufficient accuracy.

[0003] Among existing State of Charge (SOC) estimation methods, those based on equivalent circuit models, such as the Kalman filter (EKF), are widely used in lithium battery state estimation due to their tracking characteristics and real-time performance. However, due to the nonlinear nature of the state and observation equations established based on the EKF, the traditional Kalman filter is no longer applicable. Its first-order approximation, the Extended Kalman Filter (EKF), is widely used in SOC estimation applications. However, the EKF only performs a first-order Taylor expansion on the nonlinear observation equations, which significantly limits its accuracy. The Unscented Kalman Filter (UKF) utilizes the UT transform to sample point sets of the iterative state variables. This approach achieves at least the accuracy of a second-order EKF without significantly increasing computation time, and even under Gaussian noise conditions, it can achieve the accuracy of a third-order EKF without a substantial increase in computation time. This addresses the insufficient SOC estimation accuracy problem of the EKF. However, the traditional UKF algorithm requires Cholesky decomposition of the covariance matrix during iteration, which necessitates that the covariance matrix be positive definite. In practice, initial value errors, noise disturbances, and floating-point errors in the computation module can easily lead to non-positive definiteness of the covariance matrix, causing iteration to halt. To address this issue, the Square Root Unscented Kalman Filter (SRUKF) algorithm was proposed, which uses the Cholesky factor (i.e., the square root of the covariance matrix) instead of the covariance matrix for iteration, improving numerical stability and ensuring the positive definiteness of the iteration matrix. However, the SRUKF algorithm includes a first-order update step for the Cholesky factor, which still requires Cholesky decomposition. Therefore, the problem of iteration halting due to matrix non-positive definiteness persists, and the accuracy of the integer-order circuit equivalent model is limited, resulting in insufficient SOC estimation accuracy.

[0004] Among existing SOH estimation methods, data model-based estimation methods can achieve higher accuracy. Commonly used data-driven methods include Gaussian process regression, support vector machine and correlation vector machine. Conventional data-driven SOH estimation is mostly based on parameters such as battery voltage and current, while ignoring the characterization of SOH by SOC. Summary of the Invention

[0005] To address the aforementioned shortcomings in existing technologies, this invention provides a joint estimation method for SOC and SOH of lithium-ion batteries. First, based on a first-order fractional equivalent circuit model of a lithium-ion power battery, a circuit model is established. Aging experiments and HPPC testing experiments are designed to extract the SOC-OCV curve. Then, based on the SOC-OCV curve and the circuit model, a parameter identifier and an SOH state estimator are constructed using a genetic algorithm and an unscented Kalman filter improved based on the F-norm nearest symmetric matrix algorithm, respectively. Simultaneously, a trained BP neural network is used as the SOH state estimator to estimate the SOH state. Finally, the established parameter identifier is coupled with the SOC and SOH state estimators to achieve joint online estimation of SOC and SOH.

[0006] To achieve the above-mentioned objectives, the technical solution adopted by this invention is as follows:

[0007] A joint estimation method for SOC and SOH of lithium-ion batteries includes the following steps:

[0008] S1. Based on the first-order fractional equivalent circuit model of lithium-ion power battery, establish the circuit model, and design the battery aging experiment and HPPC test experiment of the circuit model to extract the SOC-OCV curve.

[0009] S2. Based on the SOC-OCV curve and circuit model in step S1, establish the state-space equation of the lithium-ion power battery, and construct a parameter identifier based on the genetic algorithm to identify the parameters of the first-order fractional equivalent circuit model.

[0010] S3. Based on the parameters of the first-order fractional equivalent circuit model identified in step S2, construct an unscented Kalman filter based on the state equation and the observation equation, and build an unscented Kalman filter improved by the nearest symmetric matrix algorithm based on the F-norm to construct a SOC state estimator for SOC state estimation.

[0011] S4. Input the SOC data, lithium-ion power battery terminal voltage, and lithium-ion power battery terminal current into the BP neural network for training, and construct an SOH state estimator to estimate the SOH state.

[0012] S5. Couple the parameter identifier constructed in step S2, the SOC state estimator constructed in step S3, and the SOH state estimator constructed in step S4 to achieve joint online estimation of SOC and SOH.

[0013] The present invention has the following beneficial effects:

[0014] This invention couples the established parameter identifier with the SOC state estimator and the SOH state estimator to achieve joint estimation of SOC and SOH, making full use of the SOC's characterization of SOH and improving the estimation accuracy of SOC and SOH. Attached Figure Description

[0015] Figure 1 This is a schematic flowchart of a joint estimation method for SOC and SOH of lithium-ion batteries proposed in this invention.

[0016] Figure 2 This is a schematic diagram of the circuit topology of the first-order fractional-order equivalent circuit model proposed in this invention.

[0017] Figure 3 This is a schematic diagram of the BP neural network used in this invention;

[0018] Figure 4 This is a schematic diagram of the SOC-SOH joint algorithm proposed in this invention. Detailed Implementation

[0019] The specific embodiments of the present invention are described below to enable those skilled in the art to understand the present invention. However, it should be understood that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the present invention as defined and determined by the appended claims. All inventions utilizing the concept of the present invention are protected.

[0020] like Figure 1 As shown, a joint estimation method for SOC and SOH of a lithium-ion battery includes the following steps S1-S5:

[0021] S1. Based on the first-order fractional equivalent circuit model of lithium-ion power battery, establish the circuit model, and design the battery aging experiment and HPPC test experiment of the circuit model to extract the SOC-OCV curve.

[0022] In an optional embodiment of the present invention, the present invention is based on the first-order fractional equivalent circuit model of lithium-ion power battery, and designs battery aging experiment and HPPC test experiment, and obtains capacity aging curve through battery aging experiment, and then combines the entire HPPC voltage and current curve to obtain SOC-OCV curve.

[0023] Specifically, step S1 includes the following steps S11-S13:

[0024] S11. Based on the cell type and model of the lithium-ion power battery, obtain the battery model and operating parameters of the lithium-ion power battery.

[0025] S12. Based on the first-order fractional equivalent circuit model of lithium-ion power battery, establish the circuit model and design the battery aging experiment and HPPC test experiment of the circuit model.

[0026] S13. Based on the battery aging experiment in step S12, obtain the capacity aging curve, and combine it with the entire HPPC voltage and current curve to obtain the SOC-OCV curve.

[0027] In this embodiment, the NCR18650PF ternary lithium battery cell produced by Zhejiang Aoyou Power System Co., Ltd. is used as an example to obtain its nominal capacity C (2.8Ah, 1C) and charging cut-off voltage V. c (4.2V), discharge cutoff voltage V d (2.5V) Three basic operating parameters. In the specific operation process, a first-order fractional equivalent circuit model, a first-order fractional Weber model, a multi-order fractional model, etc. can be selected for experiments, and the final selection of the fractional equivalent circuit model needs to be determined by comprehensively considering the model complexity, model accuracy, and robustness. Therefore, in this embodiment, a first-order fractional equivalent circuit model is selected to design the cell aging experiment and HPPC test experiment. In the cell aging experiment, the test method is to fully charge the lithium-ion power battery under the standard current using a constant current and constant voltage method, let it stand for one hour, and then discharge it in a constant current manner to the discharge cutoff voltage, and perform a cyclic test, recording the cell capacity C. n The process continues until 80% of the remaining rated capacity is reached, thus obtaining the capacity aging curve. Typically, the charge / discharge rate in the constant current stage is 1.0C. In the HPPC test of the battery cell, the cell under test is fully charged using a standard current in a constant current and constant voltage manner, and then left to stand for 1 hour to allow it to approach equilibrium. A mixed pulse current excitation sequence is then applied to the cell to induce a 10% C charge. n After discharging the cell at 10% SOC and allowing it to rest until it is close to equilibrium, the excitation sequence is applied again. This process is repeated until the cell's equilibrium open-circuit voltage and excitation response curves at 90%, 80%, ..., 10% SOC are obtained, thus yielding the SOC-OCV curve.

[0028] S2. Based on the SOC-OCV curve and circuit model in step S1, establish the state-space equation of the lithium-ion power battery, and construct a parameter identifier based on the genetic algorithm to identify the parameters of the first-order fractional equivalent circuit model.

[0029] like Figure 2 As shown, in an optional embodiment of the present invention, this embodiment uses a genetic algorithm to construct a parameter identifier based on the SOC-OCV curve and the circuit model to identify complex fractional-order model parameters, while improving the algorithm execution speed. The formula for SOC in this embodiment can be expressed as: Where SOC(t) represents the SOC value at time t, and Q SOH This indicates the battery capacity of a lithium-ion power battery under aging conditions, with an initial rated value of Q0. This indicates that the lithium-ion battery terminal current I(t) is integrated from time 0 to time t, and Q0 represents the initial rated capacity of the lithium-ion battery. The State of Charge (SOC) is directly affected by the State of Health (SOH), but the SOC timescale is smaller than that of SOH. Therefore, SOH is used as a fixed value in the parameter identifier constructed in this embodiment and the SOC state estimator constructed in step S3, with the initial value set to the rated value.

[0030] Specifically, step S2 includes the following steps S21-S27:

[0031] S21. Based on the first-order fractional equivalent circuit model, establish the circuit equations.

[0032] The circuit equation in step S21 is:

[0033]

[0034] Where U represents the terminal voltage of the lithium-ion power battery, U OCV Represents the open-circuit voltage of the lithium-ion power battery, I represents the terminal current of the lithium-ion power battery, R0 represents the ohmic internal resistance of the lithium-ion power battery, U1 represents the terminal voltage of the constant phase element, and D... a This represents the fractional derivative of the constant-phase element with order a, R1 represents the parallel resistance of the constant-phase element, and C... CPE This indicates a constant-phase element; SOC represents the state of charge of the lithium-ion battery; SOC0 represents the initial value of the state of charge of the lithium-ion battery; Q SOH This indicates the battery capacity of a lithium-ion power battery under aging conditions, with its initial value being the rated capacity. This indicates that the integral operation is performed on the lithium-ion power battery terminal current I(t) from time 0 to time t.

[0035] S22. Introduce noise parameters into the circuit equations in step S21, perform fractional derivatives based on the GL definition and discretize them to obtain the state-space equations of the lithium-ion power battery.

[0036] The state-space equation of the lithium-ion power battery in step S22 is:

[0037]

[0038] Where u(k+1) represents the terminal voltage of the lithium-ion power battery at the (k+1)th step, U OCV The open-circuit voltage of the lithium-ion battery is represented by SOC(k+1), the state of charge of the lithium-ion battery at step k+1, the terminal current of the lithium-ion battery at step k+1, R0, the ohmic internal resistance of the lithium-ion battery, u1(k+1), the terminal voltage of the constant-phase element at step k+1, ω(k), the system white noise at step k, u1(k), the terminal voltage of the constant-phase element at step k, R1, and the parallel resistance of the constant-phase element are all represented by C. CPE Let i(k) represent the constant phase element, i(k) represent the lithium-ion battery terminal current at step k, L represent the fractional step size, a represent the order of the constant phase element, u1(k+j) represent the constant phase element terminal voltage at step k+j, ν(k) represent the measurement white noise at step k, SOC(k) represent the state of charge of the lithium-ion battery at step k, and Q represent the lithium-ion battery capacity. The system white noise and the measurement white noise are independent of each other, and their mean is 0.

[0039] S23. The parameters of the state-space equation of the lithium-ion power battery in step S22 are randomly assigned and encoded using population chromosomes.

[0040] S24. Use roulette wheel selection to select the population.

[0041] S25, Cross over the chromosomes.

[0042] S26. Perform chaotic random mutations on the chromosomes.

[0043] S27. Return to the new population and repeat steps S24-S26 until the maximum number of generations is reached or the root mean square error of the chromosome is lower than the expected value. Then the parameters of the identified first-order fractional equivalent circuit model are obtained.

[0044] S3. Based on the parameters of the first-order fractional equivalent circuit model identified in step S2, construct an unscented Kalman filter based on the state equation and the observation equation, and construct an unscented Kalman filter based on the nearest symmetric matrix algorithm of the F-norm to construct a SOC state estimator for SOC state estimation.

[0045] In an optional embodiment of the present invention, this embodiment establishes an unscented Kalman filter based on the nearest symmetric matrix algorithm of F-norm to construct the SOC state estimator for SOC state estimation. Similarly, the nearest symmetric matrix algorithm based on F-norm performs positive definite processing on the covariance matrix before each UT sampling, which ensures the effectiveness of Cholesky decomposition and overcomes the problem of iteration stopping due to non-positive definite covariance matrix caused by initial value error, noise disturbance, floating point error of calculation module, etc., thereby enhancing the numerical stability of the filtering process and the robustness of the algorithm.

[0046] Specifically, step S3 includes the following steps S31-S34:

[0047] S31. Transform the state-space equation of the lithium-ion power battery in step S2 to obtain the state equation and observation equation, namely:

[0048]

[0049] Where, x k Let f(·) represent the 2D system state vector at step k, specifically (U, U1), where f(·) represents the state function, and x k-1 pm represents the 2D system state vector at step k-1. k-1 This represents the 4-dimensional model parameter vector at step k-1, specifically {R0,1,PE,}, where i L,k-1 The input vector of a 1-dimensional system with a fractional-order step size of L, representing the (k-1)th step, specifically the terminal current, ω. k-1 u represents the system white noise at step k-1. k The output vector of the system at step k is represented by the 1-dimensional vector, specifically the terminal voltage, h(·) represents the observation function, and pm k Represents the 4-dimensional model parameter vector at step k, i L,k Let ν(k) represent the 1-dimensional system input vector of the k-th step with fractional step size L, and let ν(k) represent the measurement white noise of the k-th step.

[0050] S32. Establish the unscented Kalman filter based on the state equation and observation equation in step S31, specifically including the following steps S321-S328:

[0051] S321. Initialize the system state vector x0, model parameter vector pm0, covariance matrix P0, system white noise covariance matrix Q, and measurement white noise covariance matrix R.

[0052] S322. Calculate the 5 sampling points of the system state vector using the UT transformation, and calculate the weights of the sampling points, i.e.:

[0053]

[0054]

[0055] in, This represents the system state vector at the (k-1)th step of the first sampling point. Let L represent the system state vector at the (k-1)th step of the i-th sampling point. ch(·) represents the Cholesky decomposition of the positive definite matrix, outputting an upper triangular matrix. ′ L represents the dimension of the system state variables. ′ =2, λ represents the scaling parameter and λ = a 2 (L ′ +)- ′ This is used to reduce the total prediction error, P. k-1 Let T denote the covariance matrix of the (k-1)th step, and let T denote the transpose. This represents the weight of the first sampling point with a mean of n. This represents the weight of the first sampling point with covariance c. This indicates the distribution state of the selected control sampling points, where β represents a non-negative weighting coefficient and β≥0. This represents the weight of the i-th sampling point with a mean of m. Let c represent the weight of the i-th sampling point with covariance c.

[0056] S323. Calculate the one-step prediction for the set of 5 sampling points, i.e.:

[0057]

[0058] in, This represents a one-step prediction based on the i-th sampling point at the k-th step.

[0059] S324. Calculate the one-step prediction and covariance matrix of the system state vector, i.e.:

[0060]

[0061] Among them, x k|k-1 Let P represent the 2D system state vector at the k-th step based on the (k-1)-th step number. k|k-1 Let Q represent the covariance matrix of the k-th step based on the (k-1)-th step number. ′ This represents the system's white noise covariance matrix.

[0062] S325. Based on the predicted values ​​from the first step, resampling is performed using the UT transform to generate a new set of sampling points, i.e.:

[0063]

[0064] in, This represents a one-step prediction based on the first sampling point of the k-th step, which is k-1 steps in total.

[0065] S326. Substitute the new set of sampling points generated in step S325 into the observation equation of the Kalman filter iteration in step S31 to obtain the predicted set of observation points, i.e.:

[0066]

[0067] in, This represents the observation predicted based on the i-th sampling point at the k-th step, which is based on k-1 steps.

[0068] S327. Calculate the mean and covariance of the system predictions using the weighted summation method, i.e.:

[0069]

[0070] Among them, u k|k-1 Let P represent the 1D system output vector at the k-th step based on the (k-1)-th step, R represent the measurement white noise covariance matrix, and P represent the system output vector at the k-th step. uu P xu Let represent the mean matrix and the covariance matrix, respectively.

[0071] S328. Calculate the Kalman gain matrix and update the system state vector and covariance matrix. Establish an unscented Kalman filter based on the state equation and observation equation, i.e.:

[0072]

[0073] in, Let K represent the Kalman gain matrix at step k.

[0074] S33. Based on the unscented Kalman filter established in step S32 using the state equation and observation equation, a nearest-nearest-symmetric matrix algorithm based on the F-norm is introduced. Before each UT sampling, the covariance matrix P is positive definite and then fed into the UT sampling for Cholesky decomposition. Specifically, this includes the following steps S331-S336:

[0075] S331. According to the F-norm definition, the distance between the covariance matrix P and its nearest symmetric positive definite matrix X is:

[0076]

[0077] Where, δ F (·) represents the distance between the covariance matrix P and its nearest symmetric positive definite matrix X under the F-norm. Let T denote the minimum value of a symmetric positive definite matrix X based on the F-norm, where X is equal to and greater than its transpose.

[0078] S332. Calculate the symmetric part B and the antisymmetric part C of the covariance matrix P, i.e.:

[0079]

[0080] Where B represents the symmetric part of the covariance matrix P, and C represents the antisymmetric part of the covariance matrix P.

[0081] S333. Perform polar decomposition on the symmetric part B of the covariance matrix P, i.e.:

[0082] B = UH

[0083] Where U represents an orthogonal matrix and U T U = I, H denotes a positive definite symmetric matrix and H = H T ≥0.

[0084] S334. Calculate the real symmetric positive definite matrix X′ that is uniquely closest to the distance covariance matrix P under the F-norm, i.e.:

[0085] X′=(B+H) / 2

[0086] Where X′ represents the nearest real symmetric positive definite matrix.

[0087] S335. Calculate the distance from the covariance matrix P to the real symmetric matrix X′, i.e.:

[0088]

[0089] Where, δ F (p)′ represents the distance from the covariance matrix P to the real symmetric matrix X′ under the F-norm, λ i Represents the eigenvalue.

[0090] S336. Substitute the real symmetric matrix X′ into the UT calculation in the next stage.

[0091] S34. Construct an unscented Kalman filter based on the nearest symmetric matrix algorithm of F-norm to perform SOC state estimation.

[0092] In this embodiment, the formula for SOC is... It can be seen that the battery state of charge at any two times t1 and t2 has the following relationship: Where SOC(t1) represents the SOC value at time t1, and SOC(t2) represents the SOC value at time t2. This indicates that the integral operation is performed on the lithium-ion power battery terminal current I(t) from time 0 to t2. This indicates that the lithium-ion battery terminal current I(t) is integrated over time from 0 to t1. However, since the time scale of SOC is smaller than that of SOH, SOH will not change significantly in this embodiment, thus yielding the following formula: It can be found that there is a strong coupling relationship between SOC and SOH. Therefore, SOC data can be used as a feature parameter of SOH data, that is, SOC value can be used as input value of SOH estimator.

[0093] S4. Input the SOC data, lithium-ion power battery terminal voltage, and lithium-ion power battery terminal current into the BP neural network for training, and construct the SOH state estimator to estimate the SOH state.

[0094] like Figure 3 As shown, in an optional embodiment of the present invention, this embodiment uses a BP neural network for SOH state estimation, by inputting a set of training samples (x t The true value (h) is used to train the BP neural network. During the training process, the steepest descent method is used to adjust the connection weights so that the error E quickly reaches the minimum value, resulting in a trained BP neural network used to estimate the SOH value.

[0095] Specifically, step S4 includes the following steps S41-S47:

[0096] S41. Network initialization: Determine the number of input layer nodes and output layer nodes of the BP neural network based on the input learning samples, calculate the number of hidden layer nodes, initialize the connection weights between the input layer and hidden layer, the connection weights between the hidden layer and output layer, the hidden layer threshold, the output layer threshold, and set the desired output, i.e.:

[0097]

[0098] Where p represents the number of hidden layer nodes, m represents the number of output layer nodes, n represents the number of input layer nodes, and τ represents an integer between 1 and 10.

[0099] In this embodiment, the number of learning iterations is set to 0, and the weights and thresholds are set to random numbers between [-1, 1].

[0100] S42. Using the bipolar S-function, calculate the output of the hidden layer, i.e.:

[0101]

[0102] Among them, Y o The output of the hidden layer is represented by f1(·), and the transfer function of the hidden layer is represented by w. tox represents the weights of the hidden layer and the input layer. t θ represents the value output from the input layer. o This represents the threshold of the hidden layer.

[0103] S43. Using a linear transfer function, calculate the output of the output layer, i.e.:

[0104]

[0105] Among them, Y h f2(·) represents the output of the output layer, and w represents the transfer function of the output layer. oh b represents the weights of the hidden layer and the output layer. o θ represents the value output by the hidden layer. h This represents the threshold of the output layer.

[0106] S44. Based on the output of the output layer and the expected output in step S43, calculate the error of the output layer nodes, i.e.:

[0107] δ h =-(T) h -Y h )Y h (1-Y h )

[0108] Where, δ h T represents the error of the output layer nodes. h This indicates the expected output.

[0109] S45. Based on the error of the output layer node in step S44 and the output of the hidden layer in step S42, calculate the error of the hidden layer node, that is:

[0110]

[0111] Where, δ o This represents the error of the hidden layer nodes.

[0112] S46. Calculate the error function. If the error function is lower than the set value or the maximum number of iterations, the BP neural network training is complete; otherwise, update the weights and repeat steps S42-S46, i.e.:

[0113]

[0114]

[0115] Where E represents the error function, M represents the number of samples, w represents the updated weights, and η represents the learning rate.

[0116] S47. Based on the trained BP neural network, the SOC data, lithium-ion power battery terminal voltage, and lithium-ion power battery terminal current are used as inputs, and the BP neural network outputs a state estimate of SOH.

[0117] S5. Couple the parameter identifier constructed in step S2 with the SOC state estimator constructed in step S3 and the SOH state estimator constructed in step S4 to achieve joint online estimation of SOC and SOH.

[0118] like Figure 4 As shown, in this embodiment, the established parameter identifier is coupled with each state estimator. Specifically, the parameter identifier is connected to the SOC state estimator, the SOC state estimator is connected to the SOH state estimator, and the SOH state estimator is connected to both the parameter identifier and the SOC state estimator. This coupling structure ensures both the accuracy and speed of the computation, while also enabling accurate estimation of SOC and SOH.

[0119] Specifically, step S5 includes the following steps S51-S52:

[0120] S51. Input the SOC value estimated by the SOC state estimator in step S3 into the SOH state estimator, and output the SOH estimated value.

[0121] S52. Input the SOH value estimated by the SOH state estimator in step S4 into the SOC state estimator and parameter identifier, and simultaneously input the parameters identified by the parameter identifier in step S2 into the SOC estimator, and output the SOC estimated value.

[0122] Specific embodiments have been used to illustrate the principles and implementation methods of this invention. The descriptions of the embodiments above are only for the purpose of helping to understand the method and core ideas of this invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this invention. Therefore, the content of this specification should not be construed as a limitation of this invention.

[0123] Those skilled in the art will recognize that the embodiments described herein are intended to help the reader understand the principles of the invention, and should be understood that the scope of protection of the invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific modifications and combinations based on the technical teachings disclosed in this invention without departing from the spirit of the invention, and these modifications and combinations are still within the scope of protection of this invention.

Claims

1. A joint estimation method for SOC and SOH of lithium-ion batteries, characterized in that, Includes the following steps: S1. Based on the first-order fractional equivalent circuit model of lithium-ion power battery, establish the circuit model, and design the battery aging experiment and HPPC test experiment of the circuit model to extract the SOC-OCV curve. S2. Based on the SOC-OCV curve and circuit model in step S1, establish the state-space equation of the lithium-ion power battery, and construct a parameter identifier based on a genetic algorithm to identify the parameters of the first-order fractional-order equivalent circuit model. Specifically: S21. Based on the first-order fractional equivalent circuit model, establish the circuit equations; S22. Introduce noise parameters into the circuit equations in step S21, perform fractional derivatives and discretize them based on the definition of GL, and obtain the state-space equations of the lithium-ion power battery. S23. Randomly assign values ​​to the parameters of the state-space equation of the lithium-ion power battery in step S22. S24. Use the roulette wheel selection method to select the population; S25. Cross the chromosomes; S26. Perform chaotic random mutations on the chromosomes; S27. Return to the new population and repeat steps S24-S26 until the maximum number of generations is reached or the root mean square error of the chromosome is lower than the expected value. Then the parameters of the identified first-order fractional equivalent circuit model are obtained. S3. Based on the parameters of the first-order fractional equivalent circuit model identified in step S2, construct an unscented Kalman filter based on the state equation and observation equation, and establish an unscented Kalman filter improved by the nearest symmetric matrix algorithm based on the F-norm to construct a SOC state estimator for SOC state estimation, specifically as follows: S31. Transform the state-space equation of the lithium-ion power battery in step S2 to obtain the state equation and observation equation, namely: in, Indicates the first The 2D system state vector of the number of steps. Represents the state function. Indicates the first The 2D system state vector of the number of steps. Indicates the first The 4D model parameter vector of the number of steps. Indicates the fractional step size as The The 1D system input vector of the number of steps, Indicates the first System white noise for steps, Indicates the first The 1D system output vector of the number of steps. Represents the observation function, Indicates the first The 4D model parameter vector of the number of steps. Indicates the fractional step size as The The 1D system input vector of the number of steps, Indicates the first White noise for step counting; S32. Establish the unscented Kalman filter based on the state equation and observation equation in step S31; S33. Based on the unscented Kalman filter established in step S32 based on the state equation and observation equation, the nearest symmetric matrix algorithm based on the F-norm is introduced. Before each UT sampling, the covariance matrix P is positive definite and then passed into the UT sampling for Cholesky decomposition. S34. Construct an unscented Kalman filter based on the F-norm nearest symmetric matrix algorithm to perform SOC state estimation. S4. Input the SOC data, lithium-ion power battery terminal voltage, and lithium-ion power battery terminal current into the BP neural network for training, and construct an SOH state estimator to estimate the SOH state. S5. Couple the parameter identifier constructed in step S2, the SOC state estimator constructed in step S3, and the SOH state estimator constructed in step S4 to achieve joint online estimation of SOC and SOH.

2. The method for jointly estimating the SOC and SOH of a lithium-ion battery according to claim 1, characterized in that, Step S1 specifically includes the following steps: S11. Based on the cell type and model of the lithium-ion power battery, obtain the battery model and operating parameters of the lithium-ion power battery. S12. Based on the first-order fractional equivalent circuit model of lithium-ion power battery, establish the circuit model and design the battery aging experiment and HPPC test experiment of the circuit model. S13. Based on the battery aging experiment in step S12, obtain the capacity aging curve, and combine it with the entire HPPC voltage and current curve to obtain the SOC-OCV curve.

3. The method for jointly estimating the SOC and SOH of a lithium-ion battery according to claim 1, characterized in that, The circuit equation in step S21 is: in, This indicates the terminal voltage of a lithium-ion power battery. This indicates the open-circuit voltage of a lithium-ion power battery. This indicates the terminal current of the lithium-ion power battery. This indicates the ohmic internal resistance of a lithium-ion power battery. This indicates the voltage at the terminals of the constant-phase element. The order of the constant phase element is... The fractional derivative operator, This indicates the parallel resistance of a constant-phase element. Indicates a constant phase element. This indicates the state of charge (SOC) of a lithium-ion battery. This represents the initial value of the state of charge of a lithium-ion power battery. This indicates the battery capacity of a lithium-ion power battery under aging conditions. Indicates from 0 to Constantly monitor the terminal current of the lithium-ion power battery Perform integration.

4. The method for jointly estimating the SOC and SOH of a lithium-ion battery according to claim 1, characterized in that, The state-space equation of the lithium-ion power battery in step S22 is: in, Indicates the first The terminal voltage of the lithium-ion power battery based on the number of steps. This indicates the open-circuit voltage of a lithium-ion power battery. Indicates the first The state of charge of lithium-ion power batteries based on the number of steps. Indicates the first The lithium-ion battery terminal current per step. This indicates the ohmic internal resistance of a lithium-ion power battery. Indicates the first The constant-phase element terminal voltage of the step count, Indicates the first System white noise for steps, Indicates the first The constant-phase element terminal voltage of the step count, This indicates the parallel resistance of a constant-phase element. Indicates a constant phase element. Indicates the first The lithium-ion battery terminal current per step. Indicates the fractional step size. Indicates the order of the constant phase element. Indicates the first The constant-phase element terminal voltage of the step count, Indicates the first White noise for step counting measurement Indicates the first The state of charge of lithium-ion power batteries based on the number of steps. This indicates the capacity of the lithium-ion power battery.

5. The method for jointly estimating SOC and SOH of a lithium-ion battery according to claim 1, characterized in that, Step S32 specifically includes the following steps: S321. Initialize the system state vector Model parameter vector Covariance matrix System white noise covariance matrix Measuring the covariance matrix of white noise ; S322. Calculate the 5 sampling points of the system state vector using the UT transformation, and calculate the weights of the sampling points, i.e.: in, Indicates the first sampling point is at the 1st dt. The system state vector of the number of steps, Indicates the first The sampling point at the th sampling point The system state vector of the number of steps, This indicates that the Cholesky decomposition is performed on a positive definite matrix, and the output is an upper triangular matrix. Represents the dimension of system state variables. This represents the scaling parameter. Indicates the first The covariance matrix of the number of steps, Indicates transpose. This indicates that the mean of the first sampling point is The weights, Indicates that the covariance of the first sampling point is The weights, This indicates the selection of the distribution state of the control sampling points. This represents a non-negative weight coefficient. Indicates the first The mean of each sampling point is The weights, Indicates the first The covariance of each sampling point is The weights; S323. Calculate the one-step prediction for the set of 5 sampling points, i.e.: in, Indicates based on The first step The first step One-step prediction for each sampling point; S324. Calculate the one-step prediction and covariance matrix of the system state vector, i.e.: in, Indicates based on The first step The 2D system state vector of the number of steps. Indicates based on The first step The covariance matrix of the number of steps, Represents the system's white noise covariance matrix; S325. Based on the predicted values ​​from the first step, resampling is performed using the UT transform to generate a new set of sampling points, i.e.: in, Indicates based on The first step One-step prediction for the first sampling point of the step; S326. Substitute the new set of sampling points generated in step S325 into the observation equation of the Kalman filter iteration in step S31 to obtain the predicted set of observation points, i.e.: in, Indicates based on The first step The first step Observations predicted at each sampling point; S327. Calculate the mean and covariance of the system predictions using the weighted summation method, i.e.: in, Indicates based on The first step The 1D system output vector of the number of steps. This represents the measurement white noise covariance matrix. , Let these represent the mean matrix and the covariance matrix, respectively. S328. Calculate the Kalman gain matrix and update the system state vector and covariance matrix. Establish an unscented Kalman filter based on the state equation and observation equation, i.e.: in, Indicates the first Kalman gain matrix for the number of steps.

6. The method for jointly estimating SOC and SOH of a lithium-ion battery according to claim 1, characterized in that, Step S33 specifically includes the following steps: S331. According to the F-norm definition, the distance between the covariance matrix P and its nearest symmetric positive definite matrix X is: in, This represents the distance between the covariance matrix P under the F-norm and its nearest symmetric positive definite matrix X. Let X denote the minimum value of a symmetric positive definite matrix X based on the F-norm, where X is equal to its transpose and greater than 0. Indicates transpose; S332. Calculate the symmetric part B and the antisymmetric part C of the covariance matrix P, i.e.: in, Represents the covariance matrix The symmetrical part, Represents the covariance matrix The antisymmetric part; S333. Perform polar decomposition on the symmetric part B of the covariance matrix P, i.e.: in, Represents an orthogonal matrix. Represents a positive definite symmetric matrix; S334. Calculate the unique nearest real symmetric positive definite matrix P of the distance covariance matrix under the F-norm. ,Right now: in, Denotes the nearest real symmetric positive definite matrix; S335. Calculate the covariance matrix P to a real symmetric matrix. The distance, that is: in, Describes the transformation from the covariance matrix P under the F-norm to the real symmetric matrix. distance, Represents eigenvalues; S336, Real symmetric matrix Substitute this into the UT sampling of the next stage for calculation.

7. The method for jointly estimating the SOC and SOH of a lithium-ion battery according to claim 1, characterized in that, Step S4 specifically includes the following steps: S41. Network initialization: Determine the number of input layer nodes and output layer nodes of the BP neural network based on the input learning samples, calculate the number of hidden layer nodes, initialize the connection weights between the input layer and hidden layer, the connection weights between the hidden layer and output layer, the hidden layer threshold, the output layer threshold, and set the desired output, i.e.: in, Indicates the number of hidden layer nodes. Indicates the number of nodes in the output layer. Indicates the number of nodes in the input layer. It represents an integer between 1 and 10; S42. Using the bipolar S-function, calculate the output of the hidden layer, i.e.: in, This represents the output of the hidden layer. This represents the transfer function of the hidden layer. This represents the weights of the hidden layer and the input layer. This represents the value output from the input layer. This represents the threshold of the hidden layer; S43. Using a linear transfer function, calculate the output of the output layer, i.e.: in, This indicates the output of the output layer. This represents the transfer function of the output layer. This represents the weights of the hidden layer and the output layer. This represents the value output by the hidden layer. Indicates the threshold of the output layer; S44. Based on the output of the output layer and the expected output in step S43, calculate the error of the output layer nodes, i.e.: in, This represents the error of the output layer nodes. Indicates the expected output; S45. Based on the error of the output layer node in step S44 and the output of the hidden layer in step S42, calculate the error of the hidden layer node, that is: in, This represents the error of the hidden layer nodes; S46. Calculate the error function. If the error function is lower than the set value or the maximum number of iterations, the BP neural network training is complete; otherwise, update the weights and repeat steps S42-S46, i.e.: in, Represents the error function. Indicates the number of samples. Indicates the updated weights. Indicates the learning rate; S47. Based on the trained BP neural network, the SOC data, lithium-ion power battery terminal voltage, and lithium-ion power battery terminal current are used as inputs, and the BP neural network outputs a state estimate of SOH.

8. The method for jointly estimating the SOC and SOH of a lithium-ion battery according to claim 1, characterized in that, Step S5 specifically includes the following steps: S51. Input the SOC value estimated by the SOC state estimator in step S3 into the SOH state estimator, and output the SOH estimated value. S52. Input the SOH value estimated by the SOH state estimator in step S4 into the SOC state estimator and parameter identifier, and simultaneously input the parameters identified by the parameter identifier in step S2 into the SOC estimator, and output the SOC estimated value.

Images