CN-116165546-B - Battery pack state of charge estimation method considering monomer inconsistency

Abstract

The invention discloses a battery pack state of charge estimation method considering single-body inconsistency, which comprises the steps of firstly establishing a battery pack inconsistency compensation model, using virtual measurement noise to compensate the inconsistency degree, using variational Bayesian reasoning to obtain the joint posterior probability of a virtual noise mean value, a variance and a state quantity, using an unscented Kalman filtering algorithm to carry out recursion, and finally realizing accurate battery pack SOC estimation. The inconsistency compensation model provided by the invention can reduce the influence of monomer inconsistency on the SOC estimation precision while not obviously increasing the calculated amount, and has the characteristics of high estimation precision and good instantaneity.

Inventors

• HOU JING
• WANG XIN
• HAN SONGYAN
• YANG YAN
• REN HUANYU

Assignees

• 西北工业大学

Dates

Publication Date

20260508

Application Date

20221229

Claims (2)

1. 1. A battery state of charge estimation method taking into account cell inconsistencies, comprising the steps of: step 1, regarding the battery pack as a battery, and establishing a battery equivalent model; The method comprises the steps of selecting a second-order RC network as a battery equivalent circuit model, identifying each parameter of the battery model through a battery hybrid pulse power characteristic HPPC experiment, wherein the parameters comprise an ohmic internal resistance R 0 , two polarization resistors R 1 、R 2 and two polarization capacitors C 1 、C 2 of the battery; Step 2, establishing an inconsistency compensation model, and constructing a system state equation and a measurement equation; the state space equation for creating the inconsistency compensation model is as follows: Wherein k represents the moment, x k represents the system state variable x k ＝[SOC k ,U 1,k ,U 2,k ] T ,SOC k at moment k represents the state of charge of the battery at moment k, U 1,k 、U 2,k represents the terminal voltages of the two RC loops at moment k respectively, w k represents the process noise subject to the Gaussian distribution of zero mean covariance Q k , y k represents the system observation at moment k, here terminal voltage measurement U k ;ξ k ＝h * (x k )+v k , represents the system virtual measurement noise, the sum of measurement noise h * (x k for compensating inconsistency and the original system measurement noise v k is regarded as a whole to be calculated, the mean value and covariance of xi k are unknown, and the mean value is used Representation of covariance of A representation; f (·) and h (·) represent a nonlinear state transfer function and an observation function, respectively, expressed as follows: h(·)=U k ＝U OC (SOC k )-U 1,k -U 2,k -I k R 0 wherein, I k is the charge-discharge current of the battery pack, R 0 is the ohmic internal resistance of the battery pack, R 1 and R 2 respectively represent the polarized internal resistances in two loops, τ 1 ＝R 1 C 1 、τ 2 ＝R 2 C 2 respectively represent the time constants of two RC loops in the model, U k is the terminal voltage value at k moment and is also the output value of the model, eta is the coulomb efficiency, deltat is the sampling period, Q max is the rated capacity of the battery, U OC (SOC k ) is the relation curve of the open-circuit voltage and the SOC of the battery; step 3, adopting the variable dB leaf reasoning to approach the posterior distribution of the mean and variance of the system state and the virtual measurement noise Further obtaining a state estimation result, wherein y 1:k represents all measured values from the beginning to the k moment; Step 3-1, deducing original joint posterior probability distribution by a Bayesian filtering algorithm; according to the Chapman-Kolmogorov equation, a joint prediction distribution of the system state, the virtual noise mean and the variance is obtained: And recursively obtaining posterior distribution of the joint probability according to a Bayesian formula by using the measurement value at the next moment and the predictive probability distribution, wherein the posterior distribution of the joint probability is as follows: Step 3-2, approximating the joint posterior probability by means of variable decibel leaf reasoning; from the variational Bayesian inference method, the product of three marginal distributions is used to approximate the posterior joint probability distribution, as follows: step 3-3, obtaining three marginal distribution characteristics by minimizing KL divergence; Obtaining marginal distribution by minimizing KL divergence of product of true joint posterior distribution and three marginal distributions Q x (x k ) are respectively: Wherein, the I.e. assuming that the mean of the virtual metrology noise obeys the mean value Variance is Is used for the distribution of the gaussian distribution of (c), Is its variance regulating coefficient, the variance of the virtual measurement noise is subject to inverse gamma distribution, it is assumed to be a diagonal matrix, expressed as D is the dimension of the measured variable, Two parameters of inverse gamma distribution; is the mean and variance of the state variables; And obtaining by deduction calculation: logq x (x k ) obeys a gaussian distribution with mean value m k and covariance P k * , the recursive formula is: Obeying the mean value of eta k and the variance of eta The recurrence formula is: satisfy the sum of the products of the new d independent Inv-Gamma distributions, the parameters of the new Inv-Gamma distribution are: from the Inv-Gamma distribution properties, the noise variance estimate is measured as: thereby, the statistical characteristics of the mean and the variance of the virtual noise are obtained; step 4, combining the apodization Bayesian reasoning with unscented Kalman filtering, and providing a battery pack SOC estimation under inconsistent conditions based on an apodization Bayesian unscented Kalman filtering method, wherein the specific steps are as follows: Initializing system parameters, including a variable dB leaf initial parameter, a system initial state and covariance, wherein x 0 、m 0 、P 0 、η 0 、κ 0 、α i,0 、β i,0 is the initial state of the system; step 4-2, a prediction step; 1) Unscented transforms, computing sigma points and corresponding weights: Wherein W i c and W i m are weight coefficients of state mean and covariance, d is a dimension of state quantity, λ x is a composite coefficient, 0<a x <1,b x =2; 2) Calculating a predicted value of the state quantity and a predicted value of the state covariance: 3) Calculating a predicted value of the variable decibel leaf parameters: α k|k-1,i ＝μα k-1|k-1,i β k|k-1,i ＝μβ k-1|k-1,i η k|k-1 ＝η k-1|k-1 +1 The step 4-3 is an updating step, which is realized through N times of loop iteration and comprises the following steps: step 4-3-1, initializing, and letting Α k|k,i ＝α k|k-1,i +1, for n=0:n-1, iterate the following steps: step 4-3-2, calculating a measurement predicted value; First for And performing unscented transformation again to obtain sigma points under one-step prediction: the predicted metrology values are then calculated as follows: Step 4-3-3, a variable decibel leaf iterative process; 1) The covariance of the predicted metrology P yy and the cross covariance of the predicted metrology values and state vectors P xy are calculated: 2) Calculating Kalman gain, state estimation value and covariance: 3) Calculating a virtual noise mean eta and a parameter beta of virtual noise variance inverse gamma distribution: step 4-3-4, namely circularly iterating the steps 4-3-2 to 4-3-3 until the parameters converge or the maximum iteration times are reached; step 4-3-5, converging the obtained estimated value eta k|k , X k|k 、P k|k is used as the final estimated value of k time, and is sent to the step 4-2 to continue the filtering of the subsequent time until the state estimation of each time is completed.
2. 2. The method for estimating a state of charge of a battery pack in consideration of cell inconsistency according to claim 1, wherein N is 2 to 10.

Description

Battery pack state of charge estimation method considering monomer inconsistency Technical Field The invention belongs to the technical field of batteries, and particularly relates to a battery pack state of charge estimation method. Background The state of charge (SOC) of the battery pack reflects the residual available electric quantity of the battery pack, is an important index for evaluating the current performance of the battery pack, is a precondition for realizing other functions such as battery equalization, safety control and fault diagnosis, and is important for guaranteeing the use safety of the electric automobile and prolonging the cycle life of the power battery pack. However, SOC is an internal state of a battery, is generally difficult to directly measure and acquire, and is subject to many factors such as operating conditions, temperature, aging degree, etc. to change in real time, so accurate on-line estimation of SOC of the battery is very difficult. Particularly, for a battery energy storage system of an electric automobile, because the capacity, the voltage and the like provided by a single lithium ion battery cannot meet the requirements of the electric automobile, a battery pack formed by connecting tens to thousands of single batteries in series and parallel is generally required to provide energy. However, due to the limitation of the production process, the inconsistency of the operation conditions of each single battery and the difference of aging histories, the power battery pack inevitably has certain inconsistency. If these inconsistencies are ignored, treating the power battery as a "large cell" may severely affect the state estimation accuracy of the battery. If the state estimation and the parameter identification are performed for each single body, the calculation amount is too large to realize. The current power battery pack SOC estimation method related to single-body inconsistency mainly comprises a method based on a large battery model, a method based on a representative battery model and a method based on an average-difference model. The large battery model takes the battery pack as a large single battery to carry out SOC estimation, and the influence of inconsistent factors is ignored, so that the accuracy of SOC estimation is poor, but the method is simpler and the calculated amount is smaller. The representative battery model is to select one or several representative single batteries, and obtain the SOC estimation of the whole battery pack by calculating the SOC of the single batteries, but the difference of representative battery selection criteria may cause significant difference of the SOC estimation of the battery pack. The method based on the average-difference model firstly establishes an average model of all the battery monomers, then establishes a difference model to obtain the difference of each monomer compared with the average model, and thus obtains the SOC estimation of each monomer. The method has high precision, but the model is complex, the calculated amount is large, and the real-time performance is difficult to ensure. Therefore, how to realize accurate real-time estimation of the SOC of the battery pack under the consideration of consistency is still a problem to be solved. Disclosure of Invention In order to overcome the defects of the prior art, the invention provides a battery pack state of charge estimation method considering single inconsistency, which comprises the steps of firstly establishing a battery pack inconsistency compensation model, using virtual measurement noise to compensate the inconsistency degree, then using variational Bayesian reasoning to obtain the joint posterior probability of the mean value and the variance of the virtual noise and the state quantity, and then using an unscented Kalman filtering algorithm to carry out recursion, thus finally realizing accurate battery pack SOC estimation. The inconsistency compensation model provided by the invention can reduce the influence of monomer inconsistency on the SOC estimation precision while not obviously increasing the calculated amount, and has the characteristics of high estimation precision and good instantaneity. The technical scheme adopted by the invention for solving the technical problems comprises the following steps: step 1, regarding the battery pack as a battery, and establishing a battery equivalent model; The method comprises the steps of selecting a second-order RC network as a battery equivalent circuit model, identifying each parameter of the battery model through a battery hybrid pulse power characteristic HPPC experiment, wherein the parameters comprise an ohmic internal resistance R 0, two polarization resistors R 1、R2 and two polarization capacitors C 1、C2 of the battery; Step 2, establishing an inconsistency compensation model, and constructing a system state equation and a measurement equation; the state space equation for creating the inconsistency