CN-117129890-B - Combined estimation method for SOC and SOH of lithium ion battery

CN117129890BCN 117129890 BCN117129890 BCN 117129890BCN-117129890-B

Abstract

The invention discloses a combined estimation method of a lithium ion battery SOC and an SOH, which comprises the following steps of establishing a circuit model based on a first-order fractional order equivalent circuit model of the lithium ion power battery, designing a battery aging experiment and an HPPC test experiment of the circuit model, extracting an SOC-OCV curve, establishing a state space equation of the lithium ion power battery, establishing a parameter identifier based on a genetic algorithm to perform parameter identification on the first-order fractional order equivalent circuit model, establishing an unscented Kalman filter based on the state equation and an observation equation, establishing an unscented Kalman filter based on a nearest symmetric matrix algorithm of F-norm to perform SOC state estimation, inputting SOC data, terminal voltage and terminal current into a BP neural network to perform training, establishing an SOH state estimator to realize on-line estimation of the SOH, mutually coupling the parameter identifier with the state estimator to realize combined estimation of the SOC and the SOH.

Inventors

YU CHUANXIANG
ZHANG YINGJIAN
XIAO DONGPING
Pan Aoran
GUO HAOJIE
YAN CHENG
SHI ZHUXIN

Assignees

重庆大学

Dates

Publication Date: 20260505
Application Date: 20230816

Claims (8)

1. The combined estimation method of the SOC and the SOH of the lithium ion battery is characterized by comprising the following steps of: S1, establishing a circuit model based on a first-order fractional equivalent circuit model of a lithium ion power battery, designing a battery aging experiment and an HPPC test experiment of the circuit model, and extracting an SOC-OCV curve; s2, establishing a state space equation of the lithium ion power battery according to the SOC-OCV curve and the circuit model in the step S1, and constructing a parameter identifier based on a genetic algorithm to identify parameters of the first-order fractional equivalent circuit model, wherein the method specifically comprises the following steps: s21, establishing a circuit equation based on a first-order fractional order equivalent circuit model; s22, introducing the circuit equation in the step S21 into noise parameters, and carrying out fractional order derivation and discretization based on G-L definition to obtain a state space equation of the lithium ion power battery; S23, carrying out population chromosome random assignment coding on parameters of a state space equation of the lithium ion power battery in the step S22; s24, selecting a population by adopting a roulette selection method; S25, crossing chromosomes; S26, performing chaotic random variation on the chromosome; s27, returning to a new population, and repeating the steps S24-S26 until the maximum algebra or the root mean square error of the chromosome is lower than the expected value, so as to obtain the parameters of the identified first-order fractional order equivalent circuit model; S3, constructing an unscented Kalman filter based on a state equation and an observation equation according to the parameters of the first-order fractional order equivalent circuit model identified in the step S2, and constructing an SOC state estimator based on the unscented Kalman filter improved by the nearest symmetric matrix algorithm based on the F-norm to estimate the SOC state, wherein the method specifically comprises the following steps: S31, deforming the state space equation of the lithium ion power battery in the step S2 to obtain a state equation and an observation equation, namely: Wherein, the Represent the first The 2-dimensional system state vector of the number of steps, The state function is represented as a function of the state, Represent the first The 2-dimensional system state vector of the number of steps, Represent the first A 4-dimensional model parameter vector of the number of steps, Representing fractional step size as Is the first of (2) The 1-dimensional system of steps inputs a vector, Represent the first The number of steps is a systematic white noise, Represent the first The 1-dimensional system of steps outputs a vector, Representing the function of the observation(s), Represent the first A 4-dimensional model parameter vector of the number of steps, Representing fractional step size as Is the first of (2) The 1-dimensional system of steps inputs a vector, Represent the first Measuring white noise of the step number; s32, establishing an unscented Kalman filter based on a state equation and an observation equation in the step S31; S33, introducing a nearest symmetric matrix algorithm based on F-norm according to the unscented Kalman filter based on the state equation and the observation equation established in the step S32, performing positive customization treatment on the covariance matrix P before each UT sampling, and transmitting the covariance matrix P into the UT sampling for cholesky decomposition; s34, constructing an unscented Kalman filter improved based on a nearest symmetric matrix algorithm of F-norm to construct an SOC state estimator for SOC state estimation; s4, inputting the SOC data, the lithium ion power battery terminal voltage and the lithium ion power battery terminal current into a BP neural network for training, and constructing an SOH state estimator for estimating the SOH state; And S5, mutually coupling the parameter identifier constructed in the step S2, the SOC state estimator constructed in the step S3 and the SOH state estimator constructed in the step S4 to realize joint online estimation of the SOC and the SOH.
2. The method for jointly estimating SOC and SOH of a lithium ion battery according to claim 1, wherein the step S1 specifically includes the steps of: S11, acquiring the battery model and the operation parameters of the lithium ion power battery according to the type and the model of the battery core of the lithium ion power battery; s12, establishing a circuit model based on a first-order fractional equivalent circuit model of the lithium ion power battery, and designing a battery aging experiment and an HPPC test experiment of the circuit model; And S13, obtaining a capacity aging curve according to the battery aging experiment in the step S12, and combining the whole section of HPPC voltage and current curve to obtain an SOC-OCV curve.
3. The method for joint estimation of SOC and SOH of a lithium ion battery according to claim 1, wherein in step S21, the circuit equation is: Wherein, the Represents the terminal voltage of the lithium ion power battery, Represents the open circuit voltage of the lithium ion power battery, Represents the end current of the lithium ion power battery, Represents the ohmic internal resistance of the lithium ion power battery, Representing the terminal voltage of the constant phase element, Representing the constant phase element order as Is a derivative of the fractional order of (c), Representing the parallel resistance of the constant phase element, A constant phase element is shown as such, Representing the state of charge of the lithium-ion power battery, An initial value representing the state of charge of the lithium-ion power battery, Representing the battery capacity under aging conditions of a lithium-ion power battery, Represented by 0 to 0 The end current of the lithium ion power battery is measured at any time And performing integral operation.
4. The method for joint estimation of SOC and SOH of a lithium ion battery according to claim 1, wherein the state space equation of the lithium ion power battery in step S22 is: Wherein, the Represent the first The terminal voltage of the lithium ion power battery with the step number, Represents the open circuit voltage of the lithium ion power battery, Represent the first The state of charge of the lithium-ion power battery in steps, Represent the first The current of the lithium ion power battery end with the step number, Represents the ohmic internal resistance of the lithium ion power battery, Represent the first The constant phase element terminal voltage of the number of steps, Represent the first The number of steps is a systematic white noise, Represent the first The constant phase element terminal voltage of the number of steps, Representing the parallel resistance of the constant phase element, A constant phase element is shown as such, Represent the first The current of the lithium ion power battery end with the step number, Representing the step size of the fractional order, Representing the order of the constant phase element, Represent the first The constant phase element terminal voltage of the number of steps, Represent the first The measurement of the number of steps is white noise, Represent the first The state of charge of the lithium-ion power battery in steps, Representing 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, wherein step S32 specifically comprises the steps of: s321, initializing system state vector Model parameter vector Covariance matrix System white noise covariance matrix Measurement of white noise covariance matrix ; S322, calculating 5 sampling points of a system state vector by utilizing UT conversion, and calculating weights of the sampling points, namely: Wherein, the Indicating that the first sampling point is at the first A system state vector of the number of steps, Represent the first The sampling point is at the first A system state vector of the number of steps, Representing the positive definite matrix being subjected to cholesky decomposition, outputting an upper triangular matrix, Representing the dimension of the state quantity of the system, The scaling parameters are represented by a number of parameters, Represent the first A covariance matrix of the number of steps, The transpose is represented by the number, Representing the first sampling point as the mean value Is used for the weight of the (c), Representing the first sample point covariance as Is used for the weight of the (c), Indicating the distribution state of the selected control sampling points, Representing a non-negative weight coefficient, Represent the first The average value of the sampling points is Is used for the weight of the (c), Represent the first Covariance of each sampling point is Weight of (2); s323, calculating one-step prediction of 5 sampling point sets, namely: Wherein, the The representation is based on Number of steps Number of steps One-step prediction of the sampling points; S324, calculating a one-step prediction and covariance matrix of the system state vector, namely: Wherein, the The representation is based on Number of steps The 2-dimensional system state vector of the number of steps, The representation is based on Number of steps A covariance matrix of the number of steps, Representing a system white noise covariance matrix; S325, resampling is carried out by using UT conversion according to the one-step predicted value, and a new sampling point set is generated, namely: Wherein, the The representation is based on Number of steps One-step prediction of a first sample point of the step number; S326, the new sampling point set generated in the step S325 is brought into the iterative observation equation of the Kalman filter in the step S31 to obtain a predicted observed point set, namely: Wherein, the The representation is based on Number of steps Number of steps Observed quantity predicted by the sampling points; s327, calculating the mean and covariance of the system prediction by using a weighted summation method, namely: Wherein, the The representation is based on Number of steps The 1-dimensional system of steps outputs a vector, Representing the measurement white noise covariance matrix, 、 Respectively representing a mean matrix and a covariance matrix; S328, calculating a Kalman gain matrix and updating a system state vector and a covariance matrix, and establishing an unscented Kalman filter based on a state equation and an observation equation, namely: Wherein, the Represent the first A Kalman gain matrix of steps.
6. The method for jointly estimating SOC and SOH of a lithium ion battery according to claim 1, wherein step S33 specifically comprises the steps of: s331, defining the distance between the covariance matrix P and the nearest symmetric positive definite matrix X according to F-norm, namely: Wherein, the Representing the distance of the covariance matrix P from its nearest symmetric positive definite matrix X under F-norm, Representing the minimum value based on the F-norm in the case where the symmetric positive definite matrix X is equal to its transpose and greater than 0, Representing a transpose; s332, calculating a symmetrical part B and an antisymmetric part C of the covariance matrix P, namely: Wherein, the Representing covariance matrix Is provided with a symmetrical part of the (c) a symmetrical part, Representing covariance matrix Is an anti-symmetric part of (a); s333, performing polar decomposition on the symmetrical part B of the covariance matrix P, namely: Wherein, the Representing an orthogonal matrix of the matrix, Representing a positive definite symmetry matrix; S334, calculating the nearest real symmetric positive definite matrix of the distance covariance matrix Punique under F-norm The method comprises the following steps: Wherein, the Representing the nearest real symmetric positive definite matrix; s335, calculating covariance matrix P to real symmetric matrix Is the distance of (a), namely: Wherein, the Representing the covariance matrix P under F-norm to real symmetric matrix Is used for the distance of (a), Representing the characteristic value; s336, real symmetric matrix Substituting the sampled data into the UT sample of the next stage for calculation.
7. The method for jointly estimating SOC and SOH of a lithium ion battery according to claim 1, wherein step S4 specifically comprises the steps of: S41, initializing a network, determining the number of nodes of an input layer and the number of nodes of an output layer of the BP neural network according to an input learning sample, calculating the number of nodes of a hidden layer, initializing the connection weight between the input layer and the hidden layer, the connection weight between the hidden layer and the output layer, a hidden layer threshold value and an output layer threshold value, and setting expected output, namely: Wherein, the The number of hidden layer nodes is indicated, The number of output layer nodes is represented, The number of nodes in the input layer is indicated, The expression is that an integer between 1 and 10 is taken; s42, calculating the output of the hidden layer by adopting a bipolar S function, namely: Wherein, the The output of the hidden layer is indicated, The transfer function of the hidden layer is represented, The weights of the hidden layer and the input layer are represented, A value representing the output of the input layer, A threshold value representing a hidden layer; s43, calculating output of an output layer by adopting a linear transfer function, namely: Wherein, the Representing the output of the output layer(s), Representing the transfer function of the output layer, The weights of the hidden layer and the output layer are represented, The value representing the output of the hidden layer, A threshold value representing an output layer; s44, calculating the error of the output layer node according to the output and the expected output of the output layer in the step S43, namely: Wherein, the Representing the error of the output layer node, Representing a desired output; S45, calculating the error of the hidden layer node according to the error of the output layer node in the step S44 and the output of the hidden layer in the step S42, namely: Wherein, the Representing errors of hidden layer nodes; S46, calculating an error function, if the error function is lower than a set value or a maximum iteration number, finishing BP neural network training, otherwise, updating weights, and cycling the steps S42-S46, namely: Wherein, the The error function is represented by a function of the error, The number of samples is represented and the number of samples, The weight of the update is represented as such, Representing a learning rate; S47, based on the trained BP neural network, taking SOC data, lithium ion power battery terminal voltage and lithium ion power battery terminal current as inputs, and outputting state estimation of SOH by the BP neural network.
8. The method for jointly estimating SOC and SOH of a lithium ion battery according to claim 1, wherein step S5 specifically comprises the steps of: s51, inputting the SOC value estimated by the SOC state estimator in the step S3 into the SOH state estimator, and outputting an SOH estimated value; S52, inputting the SOH value estimated by the SOH state estimator in the step S4 into the SOC state estimator and the parameter identifier, and inputting the parameter identified by the parameter identifier in the step S2 into the SOC estimator to output an SOC estimated value.

Description

Combined estimation method for SOC and SOH of lithium ion battery Technical Field The invention relates to the technical field of lithium ion batteries, in particular to a combined estimation method of a lithium ion battery SOC and an SOH. Background The state of charge (SOC) and state of health (SOH) of a power battery of an electric vehicle are important parameters of the running state of the power battery, and are also key state variables for relevant control of the power battery in a battery management system, the SOC shows the charge storage state of the current battery and usually represents the ratio of discharge capacity to current available capacity for describing the change of the state in a short time at the moment, the SOH reflects the aging state of the battery under the full life cycle scale and usually represents the ratio of the current available capacity to initial capacity of the battery and is used for describing the degradation degree of the battery under different cycles. The accurate estimation of the method is beneficial to the health diagnosis of the battery and the timely replacement of the deteriorated battery. And SOH and SOC have strong coupling relation, and influence each other, single state estimation often can not reach sufficient precision. In the existing SOC estimation method, an estimation method based on an equivalent circuit model is widely used for lithium battery state estimation by a Kalman filtering algorithm according to tracking characteristics and instantaneity. Because of the nonlinear characteristics of the state equation and the observation equation established based on the equivalent circuit model, the traditional Kalman filter is not applicable any more, and the extended Kalman filter is widely applied to estimation application of the SOC in a first-order approximation form. However, the extended kalman filter takes only the first-order taylor expansion of the nonlinear observation equation, and is greatly limited in terms of accuracy. The Unscented Kalman Filter (UKF) performs point set sampling on the iteration state quantity by utilizing UT conversion, so that the accuracy of second-order extended Kalman filtering can be at least achieved, the operation time can not be greatly prolonged, the accuracy of third-order extended Kalman filtering can be achieved on the premise of Gaussian noise, the operation time can not be greatly prolonged, and the problem of insufficient estimation accuracy of the SOC under the EKF can be solved. However, the conventional UKF algorithm requires to perform cholesky decomposition on the covariance matrix in the iteration process, which requires that the covariance matrix must be guaranteed to be positive, however, in actual situations, the covariance matrix is easily caused to be non-positive due to initial value errors, noise disturbance, floating point errors of a calculation module and the like, so that iteration is stopped. To solve this problem, a square root unscented kalman filter algorithm (SRUKF) was proposed that can iterate using a cholesky decomposition factor of the covariance matrix (i.e., square root of the covariance matrix) instead of the covariance matrix, improving the numerical stability and ensuring the positive qualitation of the iteration matrix. However, a step of first-order updating of the cholesky factor occurs in the SRUKF algorithm, and the process still needs to be performed as cholesky decomposition, so that the problem that iteration is stopped due to matrix non-positive nature still exists, and the accuracy of an integer-order circuit equivalent model is limited, so that the accuracy of SOC estimation is insufficient. In the existing SOH estimation method, the estimation method based on the data model can obtain higher precision, the commonly used data driving mainly comprises methods of Gaussian process regression, support vector machines, related vector machines and the like, the conventional data driving SOH estimation is mostly based on parameters such as battery voltage, current and the like, and the representation of SOH by SOC is ignored. Disclosure of Invention Aiming at the defects in the prior art, the invention provides a combined estimation method of the SOC and the SOH of a lithium ion battery, which comprises the steps of firstly establishing a circuit model based on a first-order fractional equivalent circuit model of the lithium ion power battery, designing an aging experiment and an HPPC test experiment of the battery, extracting an SOC-OCV curve, respectively constructing a parameter identifier and an SOH state estimator by adopting a genetic algorithm and an unscented Kalman filter based on the improvement of an F-norm nearest symmetric matrix algorithm, simultaneously adopting a trained BP neural network as the SOH state estimator to estimate the SOH state, and finally coupling the established parameter identifier with the SOC state estimator and the SOH state estimator to