Automatic method for estimating the state of charge of a battery cell

0Associated
Cases 
0Associated
Defendants 
0Accused
Products 
0Forward
Citations 
0
Petitions 
1
Assignment
First Claim
1. A method comprising automatically estimating stateofcharge of a battery's cell, wherein automatically estimating stateofcharge comprises, at each instant k of a plurality of instants, acquiring a measured voltage yk and a measured current ik at an instant k, wherein said measured voltage yk , is a voltage across said cell's terminals, and wherein said measured current ik is a current selected from the group consisting of a current that charges said cell and a current that discharges said cell, using a state model, and calculating, at said instant k, a prediction of said cell's stateofcharge SOCk, wherein said state model relates said prediction to a product Fk·SOCk−1 and to a parameter measured at a preceding instant k−1 that precedes said instant, wherein said product is a product of said cell's stateofcharge SOCk−1at said preceding instant k−1 and a state transition matrix Fkat said instant k, wherein said prediction is corrupted by state noise that is characterized by a statenoise covariance matrix Qk, and wherein said parameter that is measured at a preceding instant k−1 is selected from the group consisting of said measured voltage and said measured current, wherein automatically estimating stateofcharge further comprises predicting a covariance of an error in said prediction based at least in part on said statenoise covariance matrix Qk−1 at said preceding instant k−1 and a measurementnoise covariance matrix Rk−1 at said preceding instant k−1, calculating a prediction ŷk of said measured voltage ykat said instant k using an observation model that relates said measured voltage ykto a product Hk·SOCk of an observability matrix Hk at said instant k and said prediction of said cell's stateofcharge at said instant k, wherein said prediction of said measured voltage ŷk is corrupted with measurement noise that is characterized by said measurement noisecovariance matrix at said instant k, and correcting said prediction of said stateofcharge SOCk based at least in part on a difference between said measured voltage yk and said prediction ŷk of said measured voltage yk, wherein automatically estimating said stateofcharge further comprises setting said statenoise covariance matrix Qk using the relationship Qk=[N0G0, k(N0)]−1 and setting said measurementnoise covariance matrix Rk to be the identity matrix, wherein N0is a preset integer that is greater than unity, and wherein G0, k(N0) is given by
1 Assignment
0 Petitions
Accused Products
Abstract
An automatic estimation procedure for estimating a battery's stateofcharge includes predicting the stateofcharge using a state model that includes a statetransition matrix. An observation model then uses an observability matrix in connection with predicting a measured value of a parameter. State noise, the covariance of which is characterized by a statenoise covariance matrix contaminates this prediction. Similarly, measurement noise, the covariance of which is characterized by a measurement covariance matrix, contaminates this prediction. The values of these covariance matrices are then adjusted during the estimation procedure.
12 Citations
No References
Method for calculating power capability of battery packs using advanced cell model predictive techniques  
Patent #
US 7,969,120 B2
Filed 12/30/2009

Current Assignee
LG Chem Limited

Sponsoring Entity
LG Chem Limited

Method and apparatus for charging a battery  
Patent #
US 6,329,793 B1
Filed 05/22/2000

Current Assignee
Midtronics Incorporated

Sponsoring Entity
Midtronics Incorporated

Electric motor and gear drive assembly for driving a vehicle wheel  
Patent #
US 20050023056A1
Filed 07/31/2003

Current Assignee
ArvinMeritor Technology LLC

Sponsoring Entity
ArvinMeritor Technology LLC

Simple optimal estimator for PbA state of charge  
Patent #
US 20050046388A1
Filed 08/28/2003

Current Assignee
GM Global Technology Operations LLC

Sponsoring Entity
GM Global Technology Operations LLC

Fuel cell health management system  
Patent #
US 20030211372A1
Filed 05/16/2003

Current Assignee
ESTCO BATTERY MANAGEMENT INC.

Sponsoring Entity
ESTCO BATTERY MANAGEMENT INC.

State of charge prediction method and apparatus for a battery  
Patent #
US 6,441,586 B1
Filed 03/23/2001

Current Assignee
GM Global Technology Operations LLC

Sponsoring Entity
General Motors Corporation

Method and apparatus for fast battery charging using neural network fuzzy logic based control  
Patent #
US 5,714,866 A
Filed 09/08/1994

Current Assignee
National Semiconductor Corporation

Sponsoring Entity
National Semiconductor Corporation

Motor vehicle performance monitor and method  
Patent #
US 5,570,087 A
Filed 02/18/1994

Current Assignee
Jerome H. Lemelson

Sponsoring Entity
Jerome H. Lemelson

Nonlinear Adaptive Observation Approach to Battery State of Charge Estimation  
Patent #
US 20130006454A1
Filed 06/28/2011

Current Assignee
Ford Global Technologies LLC

Sponsoring Entity
Ford Global Technologies LLC

BATTERY CONTROLLER  
Patent #
US 20150369873A1
Filed 02/05/2013

Current Assignee
Hitachi Automotive Systems Limited

Sponsoring Entity
Hitachi Automotive Systems Limited

AUTOMATIC METHOD FOR ESTIMATING THE STATE OF CHARGE OF A CELL OF A BATTERY  
Patent #
US 20170269164A1
Filed 11/26/2015

Current Assignee
Renault Sas

Sponsoring Entity
Renault Sas

AUTOMATIC METHOD FOR DETERMINING THE STATEOFCHARGE OF A BATTERY  
Patent #
US 20170276734A1
Filed 11/26/2015

Current Assignee
Renault Sas

Sponsoring Entity
Renault Sas

8 Claims
 1. A method comprising automatically estimating stateofcharge of a battery's cell, wherein automatically estimating stateofcharge comprises, at each instant k of a plurality of instants, acquiring a measured voltage yk and a measured current ik at an instant k, wherein said measured voltage yk , is a voltage across said cell's terminals, and wherein said measured current ik is a current selected from the group consisting of a current that charges said cell and a current that discharges said cell, using a state model, and calculating, at said instant k, a prediction of said cell's stateofcharge SOCk, wherein said state model relates said prediction to a product Fk·SOCk−1 and to a parameter measured at a preceding instant k−1 that precedes said instant, wherein said product is a product of said cell's stateofcharge SOCk−1at said preceding instant k−1 and a state transition matrix Fkat said instant k, wherein said prediction is corrupted by state noise that is characterized by a statenoise covariance matrix Qk, and wherein said parameter that is measured at a preceding instant k−1 is selected from the group consisting of said measured voltage and said measured current, wherein automatically estimating stateofcharge further comprises predicting a covariance of an error in said prediction based at least in part on said statenoise covariance matrix Qk−1 at said preceding instant k−1 and a measurementnoise covariance matrix Rk−1 at said preceding instant k−1, calculating a prediction ŷk of said measured voltage ykat said instant k using an observation model that relates said measured voltage ykto a product Hk·SOCk of an observability matrix Hk at said instant k and said prediction of said cell's stateofcharge at said instant k, wherein said prediction of said measured voltage ŷk is corrupted with measurement noise that is characterized by said measurement noisecovariance matrix at said instant k, and correcting said prediction of said stateofcharge SOCk based at least in part on a difference between said measured voltage yk and said prediction ŷk of said measured voltage yk, wherein automatically estimating said stateofcharge further comprises setting said statenoise covariance matrix Qk using the relationship Qk=[N0G0, k(N0)]−1 and setting said measurementnoise covariance matrix Rk to be the identity matrix, wherein N0is a preset integer that is greater than unity, and wherein G0, k(N0) is given by
 6. An apparatus comprising a batterymanagement system for managing a battery equipped with at least one cell, wherein said batterymanagement system comprises a processor configured to execute said method of claim 1.
 8. A manufacture comprising a nontransitory and tangible computerreadable medium having encoded thereon instructions for causing an electronic processor to execute said method recited in claim 1.
1 Specification
This application is the national stage of PCT/FR/2015/053239, which was filed on Nov. 26, 2015, which claims the benefit of the Nov. 28, 2014 priority date of French application FR1461615, the contents of which are herein incorporated by reference.
The invention relates to battery management, and in particular, to estimating a battery's stateofcharge.
As a battery becomes older, it becomes less able to hold charge. It is therefore useful to be able to determine the stateofcharge of a battery.
A known method for estimating the state of charge of a cell of a battery can be found in L. Plett, et al.: “Extended Kalman filtering for battery management systems of LiPBbased HEV battery packs”, Journal of Power Sources, 2004, page 252292. This article will be referred to herein as “Plett 2004.”
Known methods of such estimation involve matrices that represent state noise and measurement noise. In practice, it is difficult to set these matrices. This is particularly true for the state noise matrix since it is difficult to quantify the modelling error a priori.
In one aspect of the invention, the covariance matrices for state noise and measurement noise, Q_{k }and R_{k }are set on the basis of known matrices F_{k }and H_{k }of the state and observation models and an integer N_{0}. In the above method, the user only has to choose a value for the integer N_{0 }instead of having to set each of the coefficients of the two covariance matrices Q_{k }and R_{k}. This considerably simplifies setting the two covariance matrices.
Some practices of the invention include repeating the setting of the covariance matrices each time one of the known matrices F_{k }and H_{k }is modified. This tends to improve the estimate of the cell's stateofcharge.
In another aspect, the invention features a tangible and nontransitory data storage medium that has, encoded thereon, instructions for executing the above automatic estimating method when these instructions are executed by an electronic processor.
In another aspect, the invention includes a batterymanagement system for managing a battery according to the foregoing methods. Another aspect of the invention includes a motor vehicle that includes a batterymanagement system for managing a battery according to the foregoing methods.
The invention will be better understood upon perusal of the following description, given solely as a nonlimiting example, and referring to the drawings, in which:
In these figures, the same references are used to denote the same elements. In the remainder of this description, characteristics and functions well known to the person skilled in the art are not described in detail.
The battery 10 comprises first and second battery terminals 12, 14 for electrical connection and several electric cells electrically connected between the battery terminals 12, 14. The battery terminals 12, 14 connect to the electric motor 4.
The battery cells are grouped into several stages. These stages connect in series between the first and second terminals 12, 14. Only two such stages are shown for simplicity. The first stage comprises first and second cells 18, 19; the second stage comprises second and third cells 20, 21. Each stage comprises several branches connected in parallel. Each branch of a stage comprises one electric cell or several electric cells in series. In the illustrated embodiment, the first stage comprises two branches, with each branch having a single electric cell. The second stage is structurally identical to the first stage in the example shown in
The cell's first and second electrical connection terminals 30, 32 connect it electrically to the other cells and, ultimately, to the battery's terminals 12, 14. The cell 18 is also fixed mechanically, with no degree of freedom, to the battery's other cells to form a pack of cells.
The cell 18 receives electrical energy while it is being charged and loses electrical energy while it is being discharged, for example when it is powering the motor 4. The complete discharging of a cell followed by its complete recharging constitutes what is known as a charging/discharging cycle, or simply the “cycle of a cell.” Although the cell can be any type, the cells described in this embodiment are those used in a lithiumion polymer battery.
A cell 18 is characterized by its initial capacitance C_{n}^{ini}, its initial internal resistance RO^{ini}, its maximum current I_{max}, its maximum voltage U_{max}, its minimum voltage U_{min}, and a function OCV(SOC_{k}). The initial capacitance C_{n}^{ini }is the capacitance of the cell 18 when it is brand new. The capacitance of a cell governs the maximum quantity of electric energy that can be stored in that cell, which is usually expressed in amperehours. As the cell 18 sustains charging and discharging cycles, it ages. This decreases its capacitance. Throughout this disclosure, the capacitance of the n^{th }cell 18 at time k shall be denoted as C_{n,k}.
A cell's initial internal resistance RO^{ini }is the value of the internal resistance of the cell 18 when it is brand new, and before it has had time to age. A cell's internal resistance is a physical quantity that is found in most circuit models of an electrical cell. As a cell ages, its internal resistance tends to increase. Throughout this disclosure, the internal resistance of the cell 18 at time k shall be denoted as RO_{k}.
The maximum current I_{max }is the maximum current that can be delivered by the cell 18 without the cell becoming damaged.
The maximum voltage U_{max }is the maximum voltage that can be present constantly between the cell's first and second terminals 30, 32 without damaging the cell. The minimum voltage U_{min }is the minimum voltage between the cell's first and second terminals 30, 32 when the cell 18 is completely discharged.
Throughout the discussion that follows, the values of I_{max}, U_{max}, U_{min }are regarded as constant physical quantities that do not vary over time.
OCV(SOC_{k}) is a predetermined function that returns the noload voltage of the cell 18 as a function of its stateofcharge SOC_{k}. The noload voltage is the voltage measurable between the cell's first and second terminals 30, 32 after the cell 18 has been electrically insulated from any electric load for two hours.
The cell's stateofcharge at time k is denoted SOC_{k}. It is equal to 100% when the quantity of electric energy stored in the cell 18 is equal to whatever its capacitance C_{n,k }permits. It is equal to 0% when the quantity of electric energy stored in the cell 18 is zero, that is, when no electric energy can be extracted from the cell 18 to energize an electric load.
The parameters C_{n}^{ini}, RO^{ini}, I_{max}, U_{max}, U_{min }and the function OCV(SOC_{k}) are known parameters of the cell. They are either provided by the cell's manufacturer or obtained by carrying out measurement son the cell.
The battery 10 also includes, for each cell, a voltmeter 34 and an ammeter 36. The voltmeter 34 measures the voltage between the cell's first and second terminals 30, 32. The ammeter measures the cell's charging or discharging current. To simplify
Unlike the different parameters of the cell 18 introduced above, the cell's stateofcharge SOC_{k }is not measurable. Thus, it needs to be estimated. The vehicle 2 thus also includes a batterymanagement system 40 carries out this function.
The batterymanagement system 40 determines the battery's stateofcharge and also its stateofhealth. To determine the stateofcharge and this stateofhealth, the batterymanagement system 40 estimates the stateofcharge and the stateofhealth of each cell of the battery 10. A cell's stateofhealth at time k, referred to herein as “SOHk,” represents how much that cell has aged. A suitable metric for SOHk is the ratio C_{n,k}/C_{n}^{ini}. To calculate the cell's stateofhealth, the batterymanagement system 40 must therefore also estimate its capacitance C_{n,k }at the present time k.
To perform these various estimates, the batterymanagement system 40 connects to each voltmeter and each ammeter of the battery 10. This permits the batterymanagement system 40 to acquire measurements of voltage and current associated with each cell.
As shown in
The parallel RC circuit 54 comprises a capacitor C_{D }connected in parallel to a resistor of value R_{D}. In the discussion that follows, these values are presumed to be known and constant. The voltage at time k across the terminals of the parallel RC circuit 54 is denoted as V_{D,k}. The value at time k of the voltage across the cell's first and second terminals 30, 32 is denoted as y_{k }and the charging or discharging current of the cell 18, at the same time, is denoted as ik. These will be referred to as “measured voltage” and “measured current.”
The first estimator 60 estimates the stateofcharge SOC_{k }and the voltage V_{D,k }based on the measured voltage y_{k }and the measured current ik. The first estimator 60 is implemented in the form of a Kalman filter. It thus uses a first state model 62, shown in
In
In the following, the temporal origin corresponds to the zero value of time k. In these conditions, the present time k is equal to kT_{e}, where T_{e }is the sampling period for the measurements of the battery's ammeters 36 and voltmeters 34. Thus, T_{e }is the period of time separating any two consecutive sampling times k and k−1 at which the batterymanagement system 40 receives a voltage and current measurement. The period T_{e }is typically a constant between 0.1 seconds and 10 seconds. Preferably, the period T_{e }is equal to 1±0.2 seconds or 1 second.
In the first state model 62, w_{k }is a state noise vector. In the embodiment described herein, the noise w_{k }is centered Gaussian white noise. This noise represents the uncertainty in the model used. The covariance matrix, at time k, of the noise w_{k }is denoted as Q_{k}. It is defined by Q_{k}=E(w_{k}·w_{k}^{T}), where E( . . . ) is the mathematical expectation function. The first state model 62 is likewise written in the form X_{k+1}=F_{k}x_{k}+B_{k}ik+w_{k}, where F_{k }is the state transition matrix at time k and B_{k }is the control vector at time k. The first state model 62 thus allows prediction of the stateofcharge SOC_{k+1 }at time k+1 from the preceding stateofcharge SOC_{k}.
The first observation model 64 allows prediction of the measured voltage y_{k }at time k from the stateofcharge SOC_{k}, the voltage V_{D,k}, and the measured current ik. In this model, vk is centered Gaussian white measurement noise. The covariance matrix of the noise vk at time k is denoted R_{k }in the following. In the particular case described here, matrix R_{k }is a matrix of a single column and a single row defined by the relation R_{k}=E(vk·vk^{T}). The noise vk is independent of the noise w_{k }and of the initial state vector x_{0}.
The first observation model 64 is nonlinear since the function OCV(SOC_{k}) is generally nonlinear. Because of this, the first estimator 60 implements the extended version of the Kalman filter. The extended version results in a linear observation model of the form y_{k}=H_{k}x_{k}+RO_{k2}·ik+vk. This is arrived at by linearizing the first observation model 64 in the neighborhood of the vector x_{k}. Linearizing typically includes developing the first observation model 64 into a Taylor's series in the neighborhood of the vector x_{k }and disregarding contributions of the derivatives starting with the second order. The matrix H_{k }is thus equal to the first derivative of the function OCV in the neighborhood of the stateofcharge SOC_{k}. This linearization of the first observation model 64 is typically carried out for each new value of the stateofcharge SOC_{k}.
The first estimator 60 needs to know the cell's capacitance C_{n,k3 }and the internal resistance RO_{k2 }in order to be able to estimate the stateofcharge SOC_{k+1}. These both vary as the cell 18 ages. To take this aging into account, the second estimator 66 estimates the internal resistance RO_{k2 }from the measured value y_{k2}, from the measured current ik_{2}, and from the stateofcharge SOC_{k2}. The third estimator 68 estimates the capacitance C_{n,k3 }from the current ik_{3 }and the stateofcharge SOC_{k3}.
A cell's internal resistance and capacitance vary more slowly than does its stateofcharge. Thus, in order to limit the computational burden of estimating a cell's stateofcharge of the cell without significantly degrading the resulting estimate's accuracy, the batterymanagement system executes the second and third estimators 66, 68 less frequently than it does the first estimator 60. In what follows, the execution times of the second and third estimators 66 and 68 are denoted respectively as k2 and k3 in order to distinguish them from the times k. The set of times k2 and the set of times k3 are subsets of the set of times k. Thus, between two successive times k2 and k2−1 and between two successive times k3 and k3−1 there elapse several periods T_{e }and several times k.
The second and third estimators 66, 68 are also each implemented in the form of a Kalman filter. The second estimator 66 uses a second state model 70, which is shown in
where N is a whole number greater
than one that is counted as will be described below. In the above relation and in the second observation model 72, the time k is equal to the time k2.
The second observation model 72 considers not only the stateofcharge SOC_{k}, the voltage V_{D,k}, and the current ik measured at time k=k2, but also the N previous estimates of the first estimator 60 and the N previous measured currents between the times k2 and k2−1. Considering the intermediate measurements and estimates between the times k2 and k2−1 improves the estimate of the internal resistance RO_{k2}.
The third estimator 68 uses a third state model 74, shown in
The observation model 76 permits estimation of a directly measurable physical quantity z_{k3}. The physical quantity z_{k3 }is the sum of the last N measured currents, which is defined by the summation:
In the above summation and in the third observation model 76, the time k is equal to the time k3. This physical quantity z_{k3 }takes into account not only the current ik−1 measured at time k−1 preceding time k3 but also the previous N currents measured between the times k3 and k3−1. In the above summation, N is a whole number greater than one, which is counted as shall be described further below. It is not necessarily equal to the N introduced in the second observation model 72. Taking into account intermediate measurements and estimates between the times k3 and k3−1 improves the estimate for the capacitance C_{n,k3}.
The method starts with a covarianceadjustment phase 100 that includes adjusting the different covariance matrices that are needed to execute the first, second, and third estimators 60, 66, 68. The covariance adjustment phase 100 includes three steps, one corresponding to each of the three estimators 60, 66, 68.
In a first covarianceadjustment step 102, the covariance matrices Q_{k }and R_{k }of the first estimator 60 are automatically adjusted with the aid of the following relations: Q_{k}=[N_{0}G_{0,k}(N_{0})]^{−1 }and R_{k}=I, where N_{0 }is a predetermined whole number that is greater than 1, I is the identity matrix, and G_{0,k}(N_{0}) is defined by:
The whole number N_{0 }is generally chosen during the design of the batterymanagement system 40 and then is set once and for all. Generally, N_{0 }is less than 100. In some embodiments, N_{0 }is between 5 and 15. In other embodiments, N_{0 }is chosen equal to 10.
The use of the preceding relations considerably simplifies the adjustment of the matrices Q_{0 }and R_{0 }as well as the adjustment of the matrices Q_{k }and R_{k}. In fact, the only parameter to be chosen is the value of the integer N_{0}.
During a second covarianceadjustment step 104, the covariances Q_{2,0 }and R_{2,0 }are also adjusted. In some embodiments, Q_{2,0 }is chosen to be equal to [(β·RO^{ini})/(3·N^{C}_{eol}·N_{S})]^{2}, where β is a constant chosen to be greater than or equal to 0.3 or 0.5 and preferably greater than 0.8 and generally less than three, N^{c}_{eol }is the predicted number of charging and discharging cycles of the cell 18 before it reaches its end of life, and N_{S }is the number of times that the internal resistance is estimated per charging and discharging cycle of the cell 18.
The constant β, expressed as a percentage divided by 100, represents the difference between the value of the initial internal resistance RO^{ini }and its end of life value. Typically, β is set by the user or measured experimentally. N_{c}^{eol }is a number of cycles that can be measured experimentally or obtained from data provided by the cell's manufacturer. N_{S }is set by the estimation method for estimating the stateofcharge as implemented by the computer 44. In this embodiment the internal resistance is estimated once per cycle. Consequently, N_{S}=1.
As an illustration, the covariance R_{2,0 }is chosen equal to (2ε_{m}U_{max}/300)^{2}, where ε_{m }is the maximum error of the voltmeter 34 expressed as a percentage.
Afterwards, the covariances Q_{2,k2 }and R_{2,k2 }are considered to be constant and equal respectively to Q_{2,0 }and R_{2,0}.
Some embodiments feature a step 106 in which the covariances are adjusted. For example, the covariance Q_{3,0 }is set to be equal to [γ·C_{n}^{ini}/(3·N^{C}_{eol}·N_{S})]^{2}, where γ, expressed as a percentage divided by 100, represents the difference between the initial capacitance C_{n}^{ini }and the capacitance of the cell 18 at the end of life. The value of γ is a constant chosen to be between 0.05 and 0.8, preferably between 0.05 and 0.3. In the embodiment described herein, γ=0.2.
In some embodiments, the covariance R_{3,0 }is chosen to be equal to [2·ε_{im}·I_{max}/300]^{2}, where ε_{im }is the maximum error of the ammeter 36 expressed in percentage.
Afterwards, the covariances Q_{3,k3 }and R_{3,k3 }are considered to be constant and equal respectively to Q_{3,0 }and R_{3,0}.
Once the covariance matrices have been adjusted, the estimation of the stateofcharge of the cell 18 can commence.
During a measurement phase 110, at each time k the voltmeter 34 and the ammeter 36 measure, respectively, the measured voltage y_{K }and the measured current ik. The batterymanagement system 40 then immediately acquires these measurements and records them in the memory 42. The measurement phase 110 is repeated at each time k.
In parallel, the first estimator 60 executes an SOCestimation phase 114 during which it estimates the cell's stateofcharge at time k.
The SOCestimation phase 114 begins with an SOCprediction step 116 during which the first estimator 60 calculates a stateofcharge prediction SÔC_{k/k−1 }and a voltage prediction V_{D,k/k−1 }to predict the cell's stateofcharge of the cell 18 and the voltage V_{D }at the terminals of the parallel RC circuit 54 shown in
As used herein, the index k/k−1 indicates that the prediction for time k considers only the measurements obtained between the times 0 and k−1. This is referred to as an “a priori” prediction. The index k/k indicates that the prediction considers all of the measurements done between the times 0 and k. This is referred to as an “a posteriori” prediction. The predictions SÔC_{k/k−1 }and V_{D,k/k−1 }are calculated with the help if the first state model 62 from the measured current ik_{−1 }and the capacitance C_{n,k3}. In the first state model 62, the state transition matrix F_{k−1 }is constant regardless of k and thus does not need to be reevaluated at each time k.
During a first covariancepredicting step 117, the first estimator 60 likewise calculates the prediction P_{k/k−1 }of a covariance matrix for the error in estimating the state vector x_{k}. In the embodiment described herein, the prediction is given by P_{k/k−1}=F_{k−1}P_{k−1/k−1}F_{k−1}^{T}+Q_{k−1}.
These various matrices F_{k−1}, P_{k−1/k−1 }and Q_{k−1 }have already been defined previously.
Then, during a linearizing step 118, the first estimator 60 constructs the matrix H_{k }by linearizing the first observation model 64 around the predictions SÔC_{k/k−1 }and V_{D,k/k−1}.
During a second covarianceupdating step 120, the covariance matrices Q_{k }and R_{k }are automatically updated. In the embodiment described herein, the second covarianceupdating step 120 is identical to the first covarianceadjustment step 102, but with the further consideration of the matrix H_{k }constructed during the linearizing step 118.
After having completed the second covarianceupdating step 120, the first estimator 60 proceeds with an SOCprediction correcting step 122. During the predictioncorrection step 160, the first estimator 60 corrects the predictions SÔC_{k/k−1 }and V_{D,k/k−1 }based on a difference between the measured voltage y_{k }and a predicted value of the voltage ŷ_{k}, the predicted value having been computed by the first observation model 64. This results in a predictiondiscrepancy.
The SOCprediction correcting step 122 includes first and second predictioncorrection operations. The first predictioncorrection operation includes calculating the prediction ŷ_{k}. The second predictioncorrection operation 126 includes the predictions SÔC_{k/k−1 }and V_{D,k/k−1 }and the matrix P_{k/k−1 }to obtain the corrected predictions SÔC_{k/k}, V_{D,k/k }and P_{k/k}.
During the first SOCprediction correcting operation 124, the prediction ŷ_{k }is calculated using the first observation model 64 with the stateofcharge being set equal to SÔC_{k/k−1 }and the value of the voltage V_{D,k }being set equal to V_{D,k/k−1}. For convenience, it is useful to refer to the difference between the measured voltage y_{k }and its predicted value ŷ_{k }as the discrepancy, E_{k}.
There are many methods for correcting the a priori estimates SÔC_{k/k−1 }and V_{D,k/k−1 }based on the discrepancy E_{k}. For example, during the second SOCprediction correcting operation 126, the estimates are corrected with the help of the Kalman gain K_{k}. The Kalman gain K_{k }is given by K_{k}=P_{k/k−1}H^{T}_{k}(H_{k}P_{k/k−1}H^{T}_{k}+R_{k})^{−1}. The a priori predictions are then corrected with the help of the following relation: x_{k/k}=x_{k/k−1}+K_{k}E_{k}.
The matrix P_{k/k−1 }is corrected with the help of the following relation: P_{k/k}=P_{k/k−1}−K_{k}H_{k}P_{k/k−1}.
The SOCestimation phase 114 repeats at each time k when a new estimate of the stateofcharge of the cell 18 needs to be done. During each new iteration, the state vector x_{k−1 }is initialized with the values obtained during the preceding iteration of the SOCestimation phase 114 for that cell 18.
In parallel, during a first comparisonstep 130, the computer 44 compares each new measurement of the current ik to a predetermined current threshold SH_{i}. As long as the measured current ik falls short of this threshold SH_{i}, the execution of the second estimator 66 is inhibited. Conversely, once the measured current ik surpasses this threshold SH_{i}, the second estimator 66 is immediately executed. In many embodiments, the threshold SH_{i }is greater than I_{max}/2. Preferably, it is greater than 0.8·I_{max }or even 0.9·I_{max}.
The second estimator 66 executes a resistanceestimation phase 140 during which it estimates the internal resistance RO_{k2 }at time k2, where the time k2 is equal to the time k at which the measured current ik crosses the threshold SH_{i}. The resistanceestimation phase 140 includes a resistanceprediction step 142, a second covariancepredicting step 144, and a predictioncorrection step 148.
During the resistanceprediction step 142, the second estimator 66 calculates the a priori prediction RÔ_{k/k−1 }of the internal resistance from the second state model 70.
Next, during the second covarianceprediction step 144, the second estimator 66 calculates the prediction P_{2,k2/k2−1 }of the covariance matrix of the error of estimation for the internal resistance. For example, this prediction is calculated with the help of the relation: P_{2,k2/k2−1}=P_{2,k2−1/k2−1}+Q_{2,0}. It is apparent that the second observation model 72 is a linear function of the state variable. Thus, it is not necessary to linearize it in the neighborhood of the prediction RÔ_{k2/k2−1 }to obtain the matrix H_{2,k2}. In this case, the matrix H_{2,k2 }is equal to −N.
During a resistanceprediction correction step 148, the second estimator 66 corrects the prediction RÔ_{k2/k2−1 }as a function of the difference between the measured physical quantity u_{k2 }and a corresponding predicted physical quantity û_{k2}. In this operation, N is a predetermined positive integer that, in some embodiments, is greater than ten, and in other embodiments, is greater than thirty. The second estimator 66 acquires the measured physical quantity u_{k2 }as the measured voltages y_{k }are measured and acquired.
In particular, the resistanceprediction correction step 148 includes a first resistancepredictioncorrection operation 150 and a second predictioncorrection operation 152.
During the first resistancepredictioncorrection operation 150, the computer 44 acquires the measured physical quantity u_{k2 }and calculates the predicted physical quantity û_{k2}. The acquisition of the measured physical quantity u_{k2 }is done by adding up the last N measurements of the measured value y_{k}. The predicted physical quantity û_{k2 }is calculated with the help of the second observation model 72. In this second observation model 72, the value RO_{k2 }is set to be equal to the previously calculated value RO_{k2/k2−1}.
Next, during the second predictioncorrection operation 152, the second estimator 66 corrects the prediction RÔ_{k2/k2−1 }as a function of the discrepancy E_{k2}. The discrepancy E_{k2 }is equal to the difference between the measured physical quantity u_{k2 }and the predicted physical quantity û_{k2}.
In the embodiment described herein, the second resistancepredictioncorrection operation 152 uses the same method as that implemented during the second SOCprediction correcting operation 126. Thus, the second resistancepredictioncorrection operation 152 is not described here in further detail. The new estimate RO_{k2/k2 }is then used during the following executions of the first estimator 60 in place of the previous estimate RO_{k2−1/k2−1}.
Triggering execution of the second estimator 66 only when the measured current ik is elevated improves the estimate of the internal resistance and does so with reduced computational load. In fact, a synergistic relationship arises because the measurement precision of the ammeter increases as the current ik increases.
Also in parallel with the measurement phase 110 and the SOCestimation phase 114, the method involves a second comparison step 160 during which, at each time k, the stateofcharge estimate SOC_{k }is compared to a predetermined upper threshold SH_{soc}. If the stateofcharge estimate SOC_{k }falls below this threshold SH_{soc}, the method continues immediately with a first counting step 162 and a third comparison step 164. Otherwise, the second comparison step 160 is repeated at the next time k. In the embodiment described herein, the threshold SH_{soc }lies between 90% and 100%.
During the first counting step 162, the computer 44 begins initializes a counter to be zero. It then increments the counter by 1 at each new measurement of the current ik since the start of the first counting step 162. Moreover, at each time k, the measured current ik and the stateofcharge estimate SOC_{k }are recorded and associated with the time k in a database.
In parallel with the first counting step 162, during a third comparison step 164, the computer 44 compares each new stateofcharge estimate SOC_{k }with a predetermined threshold SL_{soc}. As long as the stateofcharge estimate SOC_{k }remains higher than the predetermined threshold SL_{soc}, the first counting step 162 repeats at the following time k. Otherwise, as soon as the stateofcharge estimate SOC_{k }falls below the predetermined threshold SL_{soc}, the computer 44 immediately triggers execution of the third estimator 68 and stops incrementing the counter. Thus, as long as the stateofcharge estimate SOC_{k }does not surpass the predetermined threshold SL_{soc}, execution of the third estimator 68 is inhibited. In the embodiment described herein, the threshold SL_{soc }lies between 0% and 10%.
In those cases in which it executes, the third estimator 68 executes a capacitanceestimation phase 166 during which it estimates the capacitance C_{n,k3 }at time k3. Thus, the time k3 is equal to time k at which the execution of the third estimator 68 was finally triggered.
As was the case for the resistanceestimation phase 140, given that the third estimator 68 is not executed at each time k. The time k3−1 does not correspond to the time k−1. On the contrary, the times k3 and k3−1 are separated by an interval of time greater than or equal to NT_{e }where N is the number counted during the first counting step 162.
The parameters of the Kalman filter of the third estimator 68 are initialized with the previous values of these parameters obtained at the end of the previous iteration at time k3−1 of the capacitanceestimation phase 166.
The capacitanceestimation phase 166 includes a capacitanceprediction step 170, a third covariancematrix predicting step 172, and a capacitanceprediction correction step 174.
The capacitanceprediction step 170 includes calculating the prediction C_{n,k3/k3−1 }with the help of the third state model 74.
The third covariancematrix predicting step 172 includes calculating a prediction P_{3,k3/k3−1 }of the covariance matrix for the estimation error associated with estimating the capacitance.
During the third covariancematrix predicting step 172 and the capacitanceprediction correction step 174, the matrix of observability H_{3,k3 }is equal to [(SOC_{k}−SOC_{k−N})]·3600/(NT_{e}), where N is the number of times k that have elapsed between the time when the estimated stateofcharge dropped below the threshold SH_{soc }and the time when the estimated stateofcharge has dropped below the threshold SL_{soc}. The value N of is equal to the value counted during the first counting step 162.
The capacitanceprediction correction step 174 includes correcting the predictions C_{n,k3/k3−1 }and P_{3,k3/k3−1}. This involves first and second capacitanceprediction correction operations 176, 178.
The first capacitanceprediction correction operation 176 includes acquiring the measured physical quantity z_{k3 }and calculating the prediction {circumflex over (z)}_{k3 }of the quantity z_{k3}. The acquisition of the quantity z_{k3 }includes calculating the sum of the last N currents measured between the times k−1 and k−N. The prediction {circumflex over (z)}_{k3 }is obtained from the observation model 76.
During the second capacitanceprediction correction operation 178, the third estimator 68 corrects the prediction C_{n,k3/k3−1 }as a function of the difference between the measured quantity z_{k3 }and the predicted quantity {circumflex over (z)}_{k3 }in order to obtain the a posteriori estimate of the capacitance C_{n,k3/k3}. This correction is done for example as described during second SOCprediction correcting operation 126.
Next, the capacitance C_{n,k3/k3 }is sent to the first estimator 60, which proceeds to use it to estimate the stateofcharge of the cell 18 at the times that follow.
Triggering execution of the third estimator 68 only after the cell 18 has already been, for the most part, discharged promotes improved estimation while also reducing the computational load associated with carrying out the estimation.
At the end of capacitanceestimation phase 166, during an SOHcalculating step 180, the computer calculates the stateofhealth SOH_{k3 }at time k3 based on a capacitance ratio: SOH_{k3}=C_{n,k3}/C_{n}^{ini}.
As illustrated by the method of
The scheduling method of
A cell whose stateofcharge gives it only a medium priority level only has to have its stateofcharge refreshed at frequency of a fraction of this fundamental frequency, for example, at a frequency f_{e}/3.
A cell whose stateofcharge gives it only a low priority level only has to have its stateofcharge refreshed at frequency of that is an even smaller fraction of this fundamental frequency, for example, at a frequency f_{e}/10.
In the illustrated example, there is a quota in place for elevated and medium priority cells. Statesofcharge will thus only be refreshed for limited numbers of cells whose states of charge place them at either the high or medium priority levels.
To schedule the times at which the estimates of the stateofcharge of each of the cells need to be refreshed, the computer begins by assigning, during a priorityassignment step 198, a priority level to each cell.
The priorityassignment step 198 starts with a first prioritizing operation 200 during which the batterymanagement system 40 acquires the measured voltage y_{k }across the terminals of each cell.
Next, during a second prioritizing operation 202, if the measured value y_{k }is above an upper threshold SH_{y }or, on the other hand, below a lower threshold SL_{y}, and if the quota for elevated priority has not yet been met, the computer 44 then assigns to this cell the elevated priority level.
The upper threshold SH_{y }is greater than or equal to 0.9. U_{max }and, preferably, greater than 0.95·U_{max}. The lower threshold SL_{y}, is greater than or equal to U_{min }and less than 1.1. U_{min}, or 1.05·U_{min}. It is important to refresh frequently the estimate of the stateofcharge of the cells whose voltage is close to U_{max }or on the other hand close to U_{min}. In fact, an error in the estimate of the cell's stateofcharge in such a situation may lead to a degradation of the electrical and mechanical properties of that cell.
Next, for the other cells, during a third prioritizing operation 204 the computer 44 calculates the difference in voltage between the present measured value y_{k }and a previous value y_{k−x}, where X is a predetermined whole number greater than or equal to one and generally less than 5 or 10. Here, X=1.
During a fourth prioritizing operation 205, the computer 44 identifies twin cells. Cells are considered to be “twins” if, at the same time k, they have the same voltage difference and the same measured value y_{k}.
To identify such twin cells, the computer 44 compares the voltage difference and the measured value y_{k }of one cell to the voltage differences and the measured values y_{k }of the other cells at the same time. For those cells that are twins to a particular cell, the particular cell's identifier, as well as those of its twins, are grouped into a set. This set is then recorded in the memory 42.
In the embodiment described herein, the above comparison is carried out for each of the cells of the battery 10 whose identifier has not already been incorporated into one of the sets of twin cells so recorded. Once a group of cells has been identified as a set of twin cells, a priority level is assigned to only cell in that set. Thus, the fifth prioritizing operation 206 and the following scheduling step 208 and SOCestimation step 210 are performed only for the cells not having any twin cell and for a single cell of each set of twin cell
During a fifth prioritizing operation 206, the computer 44 sorts the cells in decreasing order of the absolute value of the difference calculated during the third prioritizingoperation 204. It then assigns, to the first cells of the resulting sorted list, any remaining places available for cells with elevated priority levels. It then assigns the remaining cells to any remaining places available for cells with medium priority level to the next cells in the sorted list. Finally, any remaining cells are assigned the lowest priority level.
Once a priority level has been assigned to each cell, the computer 44 carries out a scheduling step 208 during which it schedules times for refreshing the state of charge estimates for the various cells. This scheduling is carried out subject to the constraint that cells with higher priority levels must have their estimates refreshed more frequently than cells that have a lower priority level.
During the scheduling step 208, the computer 44 schedules the refresh times for estimating statesofcharge for the various cells as a function of their respective priority levels. The scheduling step 208 is carried out so as yield the correct estimation frequency for a cell based on that cell's priority level.
One way to schedule refresh times is to have the computer 44 reserve the times at which estimates of elevatedpriority level cells need to be refreshed. Next, the computer 44 reserves the times at which cells of medium priority level need to have their stateofcharge estimates refreshed. In doing so, the computer 44 avoids scheduling conflicts with refresh times that have already been reserved. Finally, the computer repeats this procedure for those cells assigned a low priority level.
To illustrate this, assume that an elevated priority level has been assigned to the cell 18, a medium priority level has been assigned to the cells 19 and 20, and a low priority level has been assigned to the cell 21. Furthermore, assume that during a period T_{e }the computer executes the SOCestimation phase 114 of the method of
In this figure, the times k to k+11 have been plotted along the xaxis. Above each of these times k, two boxes symbolize the fact that the computer 44 may, at each time k, execute the SOCestimation phase 114 of the method of
Finally, during an SOCestimation step 210, for each cell assigned a priority level the computer 44 executes the SOCestimation phase 114 at the scheduled time for this cell. Outside of these scheduled times, the computer 44 inhibits the full execution of the SOCestimation phase 114 for this cell. Likewise, the computer 44 also inhibits execution of the SOCestimation phase 114 for the twin cells to which no priority level has been assigned.
In parallel, during a twinsuppression step 212, for each twin cell that has no priority level assigned to it, the stateofcharge estimate for that cell is set equal to the most recent stateofcharge estimate carried out for a twin cell of this cell. Thus, the SOCestimation phase 114 is executed for only one of the twin cells in a set of twin cells. This makes it possible to reduce the computing power needed to determine the stateofcharge of the battery without degrading the estimate.
Optionally, in parallel with step 210, at each time k the computer 44 also executes an SOCprediction step 214 during which it predicts the stateofcharge for each of the cells that were not processed during the SOCestimation step 210 at time k. The SOCprediction step 214 consists in executing only the SOCprediction step 116, without executing the SOCprediction correcting step 122, for all the cells for which at the same time the full estimate of the SOCestimation phase 114 was not executed. In fact, the SOCprediction step 116 imposes a much smaller computational load than the SOCprediction correcting step 122 and thus it may be executed, for example, at each time k. Thus, when the SOCprediction step 214 is carried out, one will have, at each time k, a new estimate of the stateofcharge for each of the cells of the battery.
The prioritizing step 198 and the scheduling step 208 are repeated at regular intervals in order to update the priority level assigned to each of these cells. This will update the refresh frequencies for estimating the statesofcharge of these cells. This method makes it possible to limit the computing computational load associated with such estimates without significantly degrading the estimate.
In fact, the method of
During the execution of the methods of
It will likewise be noted that, whenever the computer 44 executes the SOCestimation phase 114 for a cell, it retrieves the necessary information for this execution from values obtained at the end of the previous execution of this phase for the same cell. This is the case in particular for the state variables, for example. It will be noted, however, that the time of previous execution is then not necessarily the time k−1, but it can be the time k−3 or k−10 depending on the priority level assigned to the cell.
Many other embodiments of the estimation method are possible.
For example,
The combined estimator 230 estimates the capacitance C_{n,k4 }and the internal resistance RO_{k4 }at the same time. It does so by implementing a Kalman filter that uses a fourth state model 232 and a fourth observation model 234. These are shown in
During the fourth comparisonstep 240, the computer 44, at each time k, compares the measured value y_{k }with an upper threshold SH_{y2}. Typically, the upper threshold SH_{y2 }is greater than or equal to 0.8·U_{max }or 0.9·U_{max}. Only if the measured value y_{k }drops below this threshold SH_{y2 }will the second counting step 242 and the fifth comparison step 244 actually execute.
During the second counting step 242, the computer 44 initializes a counter at zero and then increments this counter by 1 at each new time k. Moreover, at each of these times k the measured current ik, the measured voltage y_{k}, the stateofcharge SOC_{k}, and the estimated voltage V_{D,k }are recorded in a database and associated with this time k.
In parallel with the second counting step 242, at each time step k, the fifth comparison step 244 compares the new measured value y_{k }to a lower voltage threshold SL_{y2}. This lower threshold SL_{y2 }is less than or equal to 1.2·U_{min }or 1.1·U_{min }and greater than or equal to U_{min}.
Once the measured value y_{k }drops below the lower threshold SL_{y2}, the second counting step 242 stops incrementing the counter. The combined estimator 230 then begins execution. On the other hand, as long as the measured value y_{k }remains above the lower threshold SL_{y2}, the combined estimator 230 never has to execute.
The combined estimator 230 executes an estimation phase 246. As before, the times k_{4 }and k_{4−}1 are separated by an interval of time greater than or equal to NT_{e}, where N is the value of the counter as incremented during the second counting step 242. The functioning of the combined estimator 230 can be inferred from the functioning previously described second and third estimators 66, 68. Thus, it will not be described here in further detail.
Other circuit models and thus other state models can be used to estimate the cell's stateofcharge. Some practices omit the parallel RC circuit 54 from the circuit model. On the other hand, other practices use a more complex circuit model that has several parallel RC circuits electrically connected in series with each other. The state model of the cell 18 should thus be modified as a consequence in order to correspond to the new circuit model of the cell. However, all that has been described above applies with no difficulty to such a modified state model. Examples of modified state models are described in the patent application WO2006057468, the contents of which are herein incorporated by reference.
The parameters R_{D }and C_{D }of the circuit model 50 can also be estimated instead of being considered to be predetermined constant parameters. For this purpose, these two parameters R_{D }and C_{D }are introduced into the state vector x_{k}, which then becomes [SOC_{k}, V_{D,k}, R_{D,k }and C_{D,k}]^{T}. In that case, the state model is modified to incorporate the following two equations R_{D,k−1}=R_{D,k }and C_{D,k+1}=C_{D, k}.
In some embodiments, the cell's temperature is included in the state vector x_{k}. In such embodiments, the temperature is estimated concurrently with the stateofcharge.
Embodiments also include those in which the cell has one or more supplemental sensors to measure other physical quantities. This leads to modified observation models, examples of which are given in WO2006057468.
Other possible circuit models to simulate the electrical cell are also presented in part 2 of Plett 2004, in chapter 3.3.
The automatic continual adjustment of the covariance matrices R_{k }and Q_{k }can be done in a different way. One alternative method is covariance matching as described in Mehra, R. K: “On the identification of variances and adaptive Kalman Filtering,” Automatic Control, IEEE Transaction on, Volume 15, No. 2, pages 175184, April 1970. This method is applied after an initial setup of the matrices R_{0 }and Q_{0}, for example, as described during the first covarianceadjustment step 102.
In another practice, the matrices Q_{0}, R_{0}, Q_{k }and R_{k }are not adjusted as described in connection with the first covarianceadjustment step 102 and the covarianceupdating step 120. Instead, these matrices are adjusted by carrying out a conventional method. In a simplified case, the matrices are constant. In some practices, the matrix R_{0 }is set up using data provided by the manufacturer of the sensors or based on tests performed on the sensors, and the matrix Q_{0 }is set up by consecutive tests.
Theory Algorithms and Software”, Wiley Interscience, 2001.
The SOCprediction correcting step 122 or the second capacitanceprediction correction operation 178 for correction of the prediction can be done differently. For example, in one preferred method, the correction of the prediction of the stateofcharge and the voltage V_{D,k }is done by minimizing a quadratic cost function J composed a first term that is linked to the error of prediction of the measured value, and a second term that is linked to the error of estimation of the state vector. This method is described in detail in chapter 10.5.2 of Y. BarShalom, et al.: “Estimation With Applications to Tracking and Navigation, Theory Algorithms and Software,” Wiley Interscience, 2001
In another practice, the first estimator 60 is not implemented as Kalman filter. Instead, the stateofcharge is estimated by simulating its development over time using an infinite impulse response filter whose coefficients are estimated by the recursive least squares method.
Other models of state can be used to estimate the cell's internal resistance and capacitance. For example, in some embodiments, a fifth state model 250 replaces the fourth state model 232. The fifth state model 250, shown in
In the fifth state model 250, N^{c}_{k }is equal to the number of charging/discharging cycles of the cell performed prior to the time k. This number of cycles is measured for example by counting the number of times that the stateofcharge of the cell drops below the upper threshold SH_{soc }then below the lower threshold SL_{soc}. The noise term w^{a}_{d,k }represents centered Gaussian white noise. The value of γ is the difference, expressed as a percentage divided by 100, between the initial capacitance C_{n}^{ini }of the cell and its end of life capacitance. This model accounts for the fact that, as a cell ages, its cell's internal resistance increases and its internal capacitance diminishes.
In a similar manner, the second state model 70 can be replaced by the state model: RO_{k2+1}=(α+βN^{C}_{k2}/N^{C}_{EOL})RO_{k2}+w_{2,k2}, where the different symbols of this model have already been defined previously.
The third state model 74 can be replaced by the state model: C_{n,k3+1}=(1−γN^{C}_{k3}/N^{C}_{EOL})C_{n,k3}+v_{3,k3 }where the different symbols of this model have already been defined previously.
Depending on the observation model used by the third estimator 68, the quantity z_{k3 }can be calculated differently. In some practices, z_{k3 }is equal to the sum of the last N currents measured between times k and k−N+1. In this case, when N is equal to 1, z_{k3}=ik_{3}.
What was described above for the initialization of the covariance matrices Q_{k }and R_{k }can also be applied to the initialization of the covariance matrices of the third estimator 68 and the combined estimator 230.
In alternative practices, the third estimator 68 is not implemented in the form of a Kalman filter. In one such alternative practice, the capacitance is estimated by simulating its development over time using an infinite impulse response filter whose coefficients are estimated by the recursive least squares method.
The methods of
In another practice, at each time k between the times k3 and k3−1 only the capacitanceprediction step 170 of calculation of a prediction C_{n,k }is executed, but capacitanceprediction correction step 174 of correction of this prediction is not executed. Thus, one obtains a new prediction of the capacitance of the cell at each of these times k while avoiding an excessive computational burden. In a similar fashion, at each time k between the times k4 and k4−1, only the step of calculating the predictions of the capacitance and the internal resistance is executed without executing the step of correcting these predictions. Thus, in these practices, the capacitance of the cell is predicted at each time k but this prediction is corrected only at the times k3 or k4. The algorithm for estimation of this capacitance is thus only partly executed between times k3 and k3−1 or k4 and k4−1 and fully executed only at time k3 or k4.
At each time k between times k3 and k3−1 or between times k4 and k4−1, the capacitance can be estimated by executing a first algorithm. Then at time k3 or k4 the capacitance is estimated by executing a second algorithm that is different from the first algorithm and that imposes a greater computational load. The first and second algorithms do not necessarily correspond, as described above, to respectively the capacitanceprediction step 170 and the estimation phases 166, 246 of a Kalman filter. They may also be two completely different estimation algorithms.
The capacitanceestimation phase 166 or the estimation phase 246 can also be triggered in response the stateofcharge crossing a threshold ofcharge, as described in connection with
In some practices, the estimation phases 166, 246 are triggered in response to current crossing a threshold. To execute this procedure, starting with the time at which the voltage or the stateofcharge of the cell has dropped below a predetermined upper threshold, at each time k the computer 44 calculates the accumulated charge transferred QC_{k }with the help of the relationship QC_{k}=QC_{k−1}+ikT_{e}. Once the accumulated charge QC_{k }crosses an upper threshold SH_{Q}, the computer 44 executes the estimation phases 166, 246. On the other hand, as long as the accumulated charge QC_{k }remains above the threshold SH_{Q}, the computer 44 continues to inhibit execution of the estimation phases 166, 246. In an alternative practice, the quantity QC_{k }may also be calculated on a sliding window containing the last N times k, where N is a predetermined constant.
Other practices omit triggering estimation of capacitance and/or the internal resistance in response to crossing of a threshold. Among these are practices that trigger these estimation steps at regular intervals. In cases where sufficient computational capacity is available, the regular interval is equal to T_{e}.
Other practices of the method of
Other practices either carry out the second prioritizing operation 202 differently or omit it altogether. Among the former are practices in which only a single one of the upper and lower thresholds is used. In other practices, the second prioritizing operation 202 is omitted altogether.
Although two priority levels are necessary, any number beyond that is arbitrary.
Other methods for assigning a priority level to the cells are possible. For example, in some practices, the priority level of a cell may be calculated with the help of a formula linking its priority level to its voltage difference and its voltage. In these practices, the comparison operations are omitted.
The method described for associating the refresh times with the cells as a function of their priority levels is only one example. Any other known method for ordering tasks as a function of the priority level of these tasks can be adapted to ordering of the refresh times for estimating statesofcharge of the cells.
The scheduling of the refresh times for the estimation of the stateofcharge of each of the cells as described in regard to
In an alternative practice, the computer 44 comprises several programmable sub computers each of them able to execute in parallel the method of estimation of
Alternative practices calculate the cell's stateofhealth of a cell using a ratio of resistances, such as SOH_{K}=RO_{K}/RO^{ini}.
Many kinds of battery 10 are possible, including a lead battery, a super capacitance, or a fuel cell. In such case, the state model and/or the observation model of the first estimator 60 are optionally adapted to accommodate the peculiarities of the relevant battery technology.
What has been specified above also applies to a hybrid vehicle, that is, the vehicle whose driving of the powered wheels is provided at the same time, or alternately, by an electric motor and a thermal internal combustion engine. The vehicle 2 can be a truck, a motorbike or a tricycle or, generally speaking, any vehicle capable of moving by driving the power wheels with the aid of an electric motor energized by a battery. For example, it may be a hoist.
The battery 10 can be recharged with the aid of an electrical outlet which allows it to be electrically connected to the electricity mains. The battery 10 can also be recharged by a thermal internal combustion engine.