• 专利附录
  • 专利说明
  • 权利要求
  • 类似技术
  • 同族专利
  • 引用文献
  • 法律状态
  • 优先权
    • 专利摘要:
    In order to structure and at least partially automate a control-type observer for any battery type, a nonlinear model of the battery, in the form of a local model network consisting of a number of local, linear, time-invariant, dynamic, first of all data is obtained from the measurement data of a previously determined optimized design Models, which are valid in certain ranges of the input quantities, determined. For each determined local model (LMi) of the model network, a local observer is then determined. The control-technical observer (13) for estimating the SoC then results from a linear combination of the local observers.
    公开号:AT513189A2
    申请号:T50736/2013
    申请日:2013-11-06
    公开日:2014-02-15
    发明作者:
    申请人:Avl List Gmbh；
    IPC主号:
    专利说明:

    AV-3514 AT02
    Method for determining a control-technical observer for the SoC
    The subject invention relates to a method for determining a control-technical observer for estimating the SoC of a battery.
    In well-known battery management systems, e.g. for the battery of an electric or hybrid vehicle, the state of charge (SoC) of the battery or a battery cell is dependent on the measured variables in the vehicle, such as. Charging or discharging current, battery voltage and temperature determined. This is often done with a SoC control observer, who uses a nonlinear battery model, which models the nonlinear battery behavior of the battery voltage as a function of the charge and discharge current. The SoC observer then estimates the actual state of charge of the battery on the basis of this model and the battery voltage measured in the vehicle. This is e.g. from DE 10 2009 046 579 A1, which uses a simple equivalent electrical circuit as a model of the cell.
    One problem with this is that each model needs its own model. However, such models are difficult to parameterize (e.g., electrochemical models) and / or are only in a particular parameter range, e.g. only within a certain temperature range, trustworthy (e.g., equivalent circuit electrical models), and / or unsuitable for real-time applications due to their complexity, and thus for use in a battery management system.
    Another problem is the creation of the control-technical observer himself, who may have to fulfill different requirements for different models (battery types). Frequently, an extended Kalman filter is used for the nonlinear state estimation. Thereby a linearization of the nonlinear model takes place in each time step. Depending on the model structure used, however, this procedure leads to a high to very high computation effort, which makes implementation in a battery management system difficult or expensive.
    It is therefore an object of the present invention to specify a method with which it is possible to structure a control-technical observer for any type of battery and to create it at least partially automatically and which requires the least possible amount of computation.
    This object is achieved with the features of claim 1. First, a nonlinear model of the battery, in the form of a local model network, consisting of a number of local, linear, time-invariant, dynamic models, the model data of a previously determined optimized design plan via a data-based modeling method in each case in certain areas of the input variables are valid determined. For each local model of the model network determined in this way, a local observer is determined. The control-technical observer for estimating the SoC then results from a linear combination of the local observers. Such data-based modeling methods offer the advantage in this context that they are suitable for different battery types and also for real-time applications. In addition, the process can run as automated as possible, so that the effort for the creation of the SoC observer can be reduced.
    Once the SoC observer setup process is based on a design that is performed on a real battery by real-world measurements, the resulting readings and models will depend on the aging condition of the battery at the time of the measurements. Thus, the observer created for the SoC provides inaccurate estimates for the SoC for other aging conditions. In order to take into account the aging state of the battery in the estimation of the SoC, it is advantageous if at least one input variable or at least one valid function of the local model network, ie the nonlinear battery model, is scaled by at least one parameter for the aging state of the battery , In this way, regardless of the current state of aging of the battery, the SoC observer can provide good estimates for the SoC or determine the aging condition of the battery itself as an additional estimate. An advantage of this methodology is therefore that the change in the state of aging can be taken into account via a single parameter.
    The subject invention will be explained in more detail below with reference to Figures 1 to 8, which show by way of example, schematically and not by way of limitation advantageous embodiments of the invention. It shows
    1 shows the change in the SoC as a function of the load current,
    FIG. 2 an initial test plan for determining an optimized test plan, FIG. 3 an optimized test plan,
    4 shows a representation of the improvement of the information content and the test time through the optimized test plan,
    5 shows the integration of the SoC observer into a battery management system,
    6 is a block diagram of a linear model network and
    Figures 7 and 8 each show a comparison of the estimated SoC with a measured SoC.
    In a first step of the method according to the invention, a simple initial model of the battery or battery cell is determined. In the following, the terms battery 3/24 AV-3514 AT02 and battery cell in the context of the invention are considered equivalent. It is assumed that any known model of any battery cell. The known battery model is e.g. a model known from another battery, a linear battery model, or a non-linear battery model. A simple linear battery model may be 5 e.g. be determined by applying a current pulse to the battery or battery cell and thereby measuring the resulting voltage. The application of known impedance spectroscopy could also be used to create a simple battery model. From these data, a linear battery model, e.g. a simple current-voltage relationship can be determined. Although a 10 such jump attempt applies only to a specific operating point of the battery or battery cell and only for the specific current pulse, such a simple initial model is sufficient as a starting point for the methodology according to the invention, as described below.
    Based on the simple linear output model of the battery, a model-based Design of Experiments (DoE) method is used to determine the optimum excitation for the battery or battery cell. In this case, the largest possible, preferably the entire, operating range of the battery cell are to be covered, high accuracy can be achieved even for highly dynamic stimuli and effects of the battery cell as charge-discharge hysteresis and relaxation (vibration behavior) are taken into account. Current experimental design methods for non-linear systems use e.g. Amplitude-modulated pseudo-to-20 binary signals (APRB signals) to excite system dynamics. However, these methods can not be applied to batteries in the present application because the SoC is directly dependent on the excitation signal (the load current of the battery cell), e.g. according to the relationship
    SoC (t) = SoC0 +},
    τ = 0 Q 25 with l (t) the instantaneous load current, Cn of the nominal cell capacity and ηι (Ι) the
    Coulomb's efficiency. Therefore, when using batteries, a different approach must be sought. The DoE must also ensure that the excitation signal represents a suitable and sufficient excitation of the battery cell and at the same time, that the entire operating range of the SoC is covered. In addition, the time for carrying out the battery tests is also preferably taken into account.
    An optimal model-based DoE is often determined by means of the so-called Fisher information matrix and a judgment criterion such as e.g. a D-optimal evaluation criterion. According to this evaluation criterion, the determinant Jfim of the Fisher information matrix IFim is to be maximized. In addition to the D-optimal assessment criterion, there are of course other known evaluation criteria that could also be used. This is well known and will not be explained here.
    In this case, the previously determined (for example from individual jump tests) simple output model y = f (ijj) can be used to determine the Fisher information matrix. The Fisher 5 information matrix IFim is known to be based on the partial derivative of the model output y according to the model parameters ψ, in the form
    LFIM
    1 y-ι dy (k, y /) dy (k, y /) T σ2 k = 1 δψ δψ For the present application, the excitation signal is derived from a certain number of design points, each by the load current and the SoC value 10 are given, given at the beginning. The goal of the following optimization is an optimal sequential order of such design points to meet the excitation requirements mentioned above. For each design point, the sign of the load current I (as charging or discharging) and the duration of the load current I depend on the SoC to be achieved and the previous SoC, as shown in FIG. Within the permissible parameter range 1 (the input space), starting from a certain current SoC 2, 3, a new SoC 4, 5 can only be achieved if the load current I is increased (charging) for a certain period of time or for one certain period of time is lowered (discharged). Consequently, the duration of these charging or discharging pulses and thus also the test time T depends directly on this chronologically sequential sequence and the respective assignment of load current and SoC value of the design points. Thus, the test time T can also be the subject of optimization and therefore of the optimum DoE. This can e.g. be achieved by a result function
    T T J FlM, opt 'inil is minimized. In this case, variables with the index "init" designate the corresponding sizes 25 for the initial design (starting point of the optimization) and "opt" the optimized values, as a result of the optimization. The weighting parameter 0 < α < 1 weighted between accuracy and test time. The larger α becomes, the larger the information content of the excitation signal, while the reduction of the test time is weighted more, the smaller α is selected. Such optimization problems can e.g. by a heuristic optimization method (e.g., a simulated annealing method), or other known methods, as is well known. 5/24 AV-3514 AT02
    FIG. 2 shows by way of example an initial DoE (initial test plan) consisting of a number of temporally sequential design points. The test plan results from a suitably chosen sequence of the desired design points (defined by load current and SoC). In order to detect the voltage hysteresis and the relaxation behavior of the battery cell, individual charging or discharging pulses can follow phases with load current l = 0. This initial design was the starting point for a simulated annealing algorithm for the determination of an optimized DoE (optimized experimental design), as shown in FIG. The test time results from the sequential sequence and the assignment of load current and SoC of the design points. FIG. 4 shows the improvement of the determinant JF of the Fisher information matrix IF by a factor of about five while shortening the test time T by about 7% when performing the optimization.
    Through this procedure, the optimized test plan essentially covers the entire operating range of the battery cell and essentially captures the entire non-linear dynamic behavior of the battery cell. The design is then run on the battery cell test bench, i. the battery cell is subjected to the individual design points in the determined temporal sequence according to the experimental design, and measurements are made on the battery cell in order to detect state variables of the battery. This can also be automated or semi-automated. From these measurements, a model of the battery cell or battery can then be determined automatically. In doing so, e.g. Current, voltage and temperature of the battery measured, whereby the initial design plan is assumed to be known at the beginning of the measurement.
    In order to create a nonlinear model of the battery, one can also use methods known per se for nonlinear model identification. In the following, this data-based modeling will be described on the basis of a local model network (LMN), as explained in more detail below with reference to FIG. An LMN interpolates between local models, each of which is valid in certain operating ranges (or ranges of input variables).
    In this case, each i-th local model LM, of the LMN can consist of two parts, namely a validity function Φ, and a model parameter vector θ ,. The model parameter vector Θ, comprising all the parameters for the ith model and the validity function Φ, defines the scope of the ith local model (within the input space). A local estimate of the output yt (k) as the output of the i-th local model LM, at time k, results from yt (k) = xT {k) 6i, where x (k) denotes a regression vector, the current and past one - and outputs contains, eg the cell current l (k) and its last four past values l (k-1) ... l (k-4), the last SoC (k-1) and the cell temperature 6/24 AV-3514 AT02 turT (k -1). The global model output y (k), e.g. the cell voltage U (k), then results from a linear combination with a weighting of the M local model outputs by the validity function 4> j in the form
    M jw = 2 >, (* > Ä «·; = i 5 The nonlinear system behavior of the battery is thus described by the LMN via locally valid, linear models which are each subdivided into a certain subrange (defined via current, SoC, Temperature, etc.) of the entire workspace, eg the model output could be represented by y (k) = U (k) and the regression vector by x (k) = [U (k-1), U (k-2), .. ., U (kn), l (k), l (k-1) ..... I (km), T (k-1), SoC (k-1), 1] T, where 10 The model parameter vector θ, then would be, for example, [au, i, au, 2, au, n, bi, o, bi, i, ... bi, m, bT, i, bSoc, i , c0]. U designates the voltage of the battery, I the current and T the temperature, in the respective system order (k), (k-1), ... (kn) from the determined and conducted test plan as measured values The result of the search scheme is thus used for the model identificati on, wherein the model parameters or the model parameter vector Θ, by a well-known optimization, e.g. by least squares method, which minimizes the error between estimated output and measured output, or similar optimization methods. As a result, a non-linear model of the battery is obtained, which has the output y (k), e.g. U (k), as a function of the input quantities, e.g. Current I and temperature T, as well as predicted values of these quantities
    From the thus determined non-linear model of the battery in the form of the LMN, a control-technical observer for the SoC can be determined, which is described here using the example of a fuzzy observer. The use of a fuzzy observer allows a simple and automated parameterization of the non-linear state observer. Due to the stationary design of the local filters, a real-time implementation is possible because the computational effort in each time step is low (in contrast to the ex-extant Kalman filter, for example). Each local model of the LMN is a linear, time-invariant, dynamical system. Therefore, using standard calf filter theory, a local observer is created for each local linear state space model. A global filter then results from a linear combination of the local filters. The nonlinear observer design therefore includes the description of the nonlinear system as local, linear state models. Essentially, the state vector z (k-1) of the system is supplemented by the state SoC (k-1), which previously served as input for the determination of the LMN. This results from a combination of the LMN with the relative SoC model of previously in the form SoC {k) -SoC {k- ) + -i (k), where Ts denotes the sampling time. For the i-th
    Q5 local model is obtained according to the state space model, the state vector Zi (k) = AiZ (k-1) + BjU (k), with the system matrix A ,, the input matrix B, and the input vector u (k), wherein the state vector z (k-1) the state SoC (k-1) and the system matrix A, and the input matrix B, the previously determined model parameters. From this, the state vector for the system state follows again by weighting the local states with the validity state.
    M 10 function O > i to z (k) = ^ Φ; (k-V) {Aiz (k-X) + Βμ (Κ)} and the global model output y {k) = Cz (k) with C = [1 0 ... 0 0].
    In the example, the state vector z (k) contains outputs from above and the state of charge according to z (k) = [U (k-1), U (k-2),..., U (kn), SoC (k). 1)] T and the input vector u (k) contains the current and time-delayed inputs (and the offset term of the local 15 models) in the form u (k) = [l (k), l (k-1),. .., l (km), T (k-1), 1],
    An observer is necessary because the relative SoC model over a longer period of time is unreliable and gives false results and furthermore the initial state of charge must be known exactly, which is usually not the case. The observer can adapt this state based on measured real input and output variables and thus compensate for errors caused by the wrong initialization or by the unreliability of the model.
    For this purpose, a local Kalman filter with the gain matrix K i can be determined for each local model of the LMN in order to estimate the local state. The estimated state z (k) containing the estimated SoC then follows from equations 25
    M z * (k) = Atz (k -1) + Btu (k) and y (k) = ^ Φ t {k - ) Cz * (k).
    The determination of the Kalman gain matrix K, is e.g. on the well-known discrete algebraic Riccati equation (DARE) with the covariance matrices describing the measurement noise 8/24 AV-3514 AT02 and the process noise. These covariance matrices may be e.g. determined by known methods from the measured values recorded on the test bench.
    The covariance value which describes the process noise of the SoC in the covariance matrices can, however, also be considered and used as an adjustable tuning parameter for the non-linear SoC observer, as shown in FIGS. 7 and 8. The two FIGS. 7 and 8 show the function of the control-technical observer. In both figures it can be seen that the estimate of the SoC (in each case the dashed curve diagram) approaches the actual value of the SoC even with arbitrary output data. In Fig. 8, the weighting of SoC has been increased in the state, resulting in faster convergence but slightly increased error.
    The determined control-technical observer for estimating the SoC can be integrated into a battery management system 11, as shown in FIG. 5. The load current I of a battery 10 generates a battery voltage U and a battery temperature T, which are supplied to the battery management system 11 as measured variables. Therein, the extended state space model 12 determined according to the invention and the nonlinear observer 13, which estimates the current SoC, is implemented. The estimated SoC can then be reused accordingly, e.g. in the battery management unit 11 itself, or in a battery or vehicle control unit, not shown.
    In addition to the estimation of the state of charge SoC, the determination of the state of health (SoH) is important in order to be able to specify a criterion for the quality of the battery 10. For the SoH, however, there is still no universally valid definition. Often, the aging state SoH of a battery 10 is determined by a characteristic such as a value. As a ratio of the available capacity Cact of the battery compared to the known nominal capacity Cinit the battery indicated, so Cakt / Cinit- To enable efficient use of the battery 10 and thus to achieve the longest possible life, should preferably also the SoH, or a characteristic for it, as exactly as possible to be determined, eg from the battery management unit 11, since the current aging state SoH also influences the determination of the state of charge SoC. For an exact determination of the state of charge SoC an aged battery 10, it is therefore advantageous if the respective current aging state SoH is taken into account in the determination of the state of charge SoC. It will now be described how the above-described SoC observer can be supplemented to take into account or determine the state of aging SoH.
    In contrast to classical modeling approaches (chemical-physical models, equivalent-circuit models, etc.), the above-described, purely data-based (black-box) battery model in the form of the LMN, or the derived state space model 12, 9/24 AV- 3514 AT02 no physical information about the battery 10 itself. The model parameters contained in the model parameter vector Θ, however, are on the relationship γ ^) = χτ {k) 0i in
    Correlation with the input variables in the regression vector x (k) and with the output variable y (k). This interpretability of the local model network LMN can now be used to determine or extract influencing variables for the aging state SoH from the LMN and to use these factors to influence the battery model or estimate of the SoC. The goal here is also to adapt the dynamic quantities of the output variable y (k), e.g. the battery voltage U (k), from aging batteries 10 with the LMN, and derived from the state space model 12, and thus to allow an accurate estimation of the SoC. In addition, the aging state SoH can thereby be determined as a further variable, output and used further.
    In the simplest case, the parameter for the aging state SoH is the available capacity Cakt, or the ratio of the available capacity Cakt compared to the nominal capacity Cimt, ie Cakt / Cinjt. However, it can also be used as a parameter for the aging condition SoH current internal resistance Rakt or the ratio of actual internal resistance of the aged battery Rakt compared to the known internal resistance of a new battery Rmt, so Rimt / FW, or a combination thereof, are taken into account. A battery model in the form of the LMN can then be obtained by specifying the characteristic, e.g. Cakt / Cinit and / or Rakt / Rinit, adapted to simulate the behavior of the output variable y (k) at known input variables x (k) for any resulting aging states SoH, as indicated in Figure 9. In this case, the effect of this parameter (s) on at least one model parameter in the model parameter vector Θ, and possibly also on the validity functions Φ ,, and / or converted to at least one input variable x (k), whereby an adaptation of the static and dynamic behavior of the nonlinear Battery model is done. At the same time, with such a supplemented battery model from the observer derived therefrom for the state of charge SoC, as described above, the state of charge SoC can also be determined as a function of the current state of aging SoH.
    This will be explained below using the example above.
    Due to the interpretability of the local linear models LM, of the local model network LMN, the model parameters, e.g. au, i, au, 2,, au, n, bi, 0, bi, i, ... bi, m, bT, i, bSoc, i. c0, which determine the model parameter vector Θ, or the system matrix A, in the state space model 12, are changed accordingly in order to take into account the aging of the battery 10 in the battery model. Likewise, the input variables x (k) of the local model network LMN or the state variables in the state space model 12 can also be changed. 10/24 AV-3514 AT02
    Following are exemplary given possibilities to take into account the parameter (s) for the aging state SoH: Adaptation of the internal resistance: Via the model parameter bi, o, which is related to the actual current l (k) and which at an output quantity U (k) 5 has a resistance property, directly the internal resistance R of the Bat terie 10 can be influenced. Thus, the model parameter bi.oz.B. scaled according to the characteristic Rakt / Rimt. - Adaptation of the steady state gain: The stationary gain of the transfer function (as a quotient of input and output. The model parameters b |, i, ... b |, m, br, i, bs0c, i or directly via the input variables 10 assigned to the input variables ) be adjusted. - Capacity adjustment in the relative SoC model: the observer for the SoC takes into account the relationship between electricity and SoC over the relative
    T
    SoC model in the form SoC (k) = SoC (k-1) + -i (k). By adaptation or
    Q 15 Scaling of the capacity Cn, the change in capacity over the aging can be taken into account. The scaling can depend on the characteristics, e.g. Cakt / Cimt, possibly taking into account the current SoC.
    A characteristic, e.g. Cact / Cinit, can be used to scale input signals 20 (or states in the state space model) of the system. For example, in the form of a SoC correction in the battery model or state space model 12, e.g. via the model parameter bs0c, i- The scaling can depend on the characteristic, e.g. Cact / Cinit, for example, by SoCn0rm = SoC * Cinit / Cakt, where SoC is the estimated SoC and SoCn0rm is the scaled and returned 25 SoC.
    A characteristic, e.g. Cakt / Cinit, can be used to correct the stationary gain K, the transfer functions of the local linear models LM, and the battery current via the model parameters bi, i, ... bi, m, e.g. by Ki norm = K * Cakt / Cinit, where K is the steady state gain 30 resulting from the model and Ki norm is the scaled one.
    If the parameter is known or specified, the non-linear battery model LMN or the state space model 12 and thus the estimate of the SoC can be adapted to the current state of aging. The specific type of scaling depending on the 11/24 AV-3514 AT02
    Characteristic is not decisive here, but there are of course many possibilities how to scale.
    The parameter for the aging condition SoH, e.g. the capacitance Cakt or the internal resistance Rakt, but is usually not directly measurable and therefore not known. This characteristic, or the associated aging state SoH, should therefore be estimated by the nonlinear observer. The above-described non-linear observer 13 is therefore now extended by the estimation of the state of aging SoH, or a characteristic thereof.
    This combined SoH / SoC observer is based on the extended nonlinear battery model in the form of the local model network LMN and the derived extended state space model 12. The characteristic such as e.g. Cact, Cact / Cinit, Rakt or Rakt / Riniti influences directly or indirectly input variables and / or model parameters of this state space model, as described above.
    In choosing the architecture of the combined SoH / SoC observer, one can distinguish between a full observer for SoC and SoH and a cascaded observer.
    For the full observer, SoC and SoH (or SoH parameters) are estimated at the same time. An important point is the consideration of the different time constants for the observation of SoC and SoH, since the SoH usually changes much more slowly than the SoC.
    In the cascaded observer there is an inner observer (circle) which aims to determine the SoC as accurately as possible and which is described above. An external observer 14 building on the inner observer 13 then presents an overall estimator for the SoH characteristic, as shown in FIG. The advantage of the cascaded observer is that the different time constants for the two observers are already indirectly taken into account via the architecture. The SoH estimator can thus be operated by the separation into an inner and outer observer also with a much lower sampling time. The cascaded observer is therefore preferred.
    In Fig. 10, the observer of Fig. 5 is an observer 14 for the aging state SoH, or a characteristic thereof, as shown here e.g. Cakt, added in cascaded version. The determined extended state space model 12, supplemented by the above-described consideration of the parameter for the aging state SoH, and the non-linear observer 13 for the state of charge SoC are again provided. Added is a 12/24 AV-3514 AT02
    Observer 14 for the estimation of the characteristic of the state of aging, here the current capacity Cakt.
    To estimate the parameter, an observation error e with respect to the parameter is minimized. The observation error e is e.g. from the difference between the measured output variable y, in this case the battery voltage U, and the output variable y calculated by the nonlinear battery model or the state space model 12, here the estimated battery voltage U. In order to estimate the characteristic value, the observation error e is considered in a certain time window t = 0... T, whereby, however, only previous values are taken into account. The goal of the estimation / optimization in the observer 14 is now to minimize a certain quality measure with respect to the observation error e. As good measure can be used. Alternatively, e.g. the minimum of the error squares f e2 {t) dt = min J C, t = 0 can be aimed at as symmetrical as possible an error distribution, e.g. in the form T (ΛΟΟ- / 1Η :)) 2 8 = 0 de = min, or to minimize the offset of the estimation error, e.g. in
    Cakt of the form J h (e) e de = min, where the function is any distribution function, e.g. the
    Cact e _- ^ nax
    Normal distribution, can be.
    In this case, it is possible to proceed in such a way that, in the time window t = 0... T, the output variable Ü for different capacitances Cakt, e.g. between a given Cmin and a Cmax, calculated from the nonlinear battery model and the state space model 12, respectively, and the SoC is estimated, and from this, the observed error e associated with the respective measured output U associated with the respective time is determined. From this, the capacitance Cact can then be determined via the quality measure, which minimizes the observation error e according to the quality measure. Thus, by adapting Cact, a certain measure of the observational error e is minimized, which ultimately results in determining the capacitance Cakt which best explains the behavior of the battery 10.
    In this case, boundary and secondary conditions can also be taken into account, e.g. monotonic drop in capacitance Cakt over time, as the battery 10 can only get older, but not younger. The combined observer thus generally aims to minimize the deviations of the estimated states from the actual values, taking into account marginal and constraints in the form of equality or inequality constraints. These constraints can be used to e.g. Capacity change rates to limit or penalize the direction of SoH correction, etc. Depending on 13/24 AV-3514 AT02
    Architecture and formulation of the quality measure and the restrictions can be used different, but known, optimization method.
    The thus determined observer 14 can then determine the SoH, or a characteristic of the SoH. 11, the characteristic value 5 Cakt estimated with the combined SoH / SoC observer relative to the rated capacity Cm of a battery is shown as a function of the number of full charge / discharge cycles Z (with crosses). Dots show the actually measured capacity drop. From this it can be seen that the state of aging SoH, or a parameter for it, can be estimated very well by the described method.
    After the observer for the SoH conditionalizes on past data, the SoH estimate can be much slower than the SoC estimate. The observer for the SoC always takes into account the current estimate for the SoH, or the parameter for it.
    In the case of a complete observer, the observer 14 for the aging state SoH would be integrated into the observer 13 for the SoC. 14/24
    权利要求:
    Claims (10)
    [1]
    AV-3514 AT02 Claims 1. A method for determining a control-related observer (13) for estimating the SoC of a battery (10), comprising the following method steps:
    [2]
    5 - predetermining an output excitation signal of the battery (10), wherein the excitation signal consists of a time sequence of a plurality of design points defined by the load current (I) and the SoC value, predetermining an output model of the battery (10) a model output (y) and model parameters (ψ),
    [3]
    10 - Determination of an optimized experimental design in the form of an optimized temporal sequence of design points using a model based Design of Experiments method based on the Fisher Information Matrix (IFim) defined by 1 ^ dy (k, ψ) dy (k, ψ) τ IFIM = -T- r -! - and a judgment criterion, σ k = l δψ δψ charging of the battery (10) with the individual design points in the temporal sequence according to the determined test plan, whereby measured values of state variables of the battery (10) determining a non-linear model of the battery (10) on the basis of the determined measured values and on the basis of a local model network consisting of a number of local, linear, time-invariant, dynamic models (LM,), each in certain ranges of the input variables Validity, whereby the model output is determined as a weighted linear combination of the outputs of the local models (LMi), converting the local models (LM,) of the model network into local, linear state space models, with a state vector containing the SoC,
    [4]
    25 - Creating a local observer for each local, linear state space model and creating the control technical observer (13) from a linear combination of local observers. Method according to claim 1, characterized in that the optimized comparison scheme is obtained by minimizing a result function J = a FIM'imt + (1 determines J FlM, opt ΤΜί, where JFim is the determinant A method according to claim 1, characterized in that each i-th local model (LM |) of the local model network (LMN) consists of a validity function (Φ |) and a model parameter vector θί, wherein the model parameter vector (0i) comprises all the model parameters for the ith model (LMj) and the validity function (Φ,) defines the range of validity of the i-th local model (LMj), and a local estimate of the output y ^ k) of FIG i th local model (LMj) at time k is determined from yt {k) = xT (k) 6i, where x (k) denotes a regression vector containing current and past inputs and outputs that the global model output y (FIG. k) from linear combination with a weighting of the local model outputs by the validity function (Φ |) according to M y (k) = 's £ j (& i (k) yi (k)) and that the model parameter vector (0,) is determined by means of a / = 1 15 optimization that minimizes the error between estimated output and measured output.
    [5]
    4. The method according to claim 3, characterized in that for each i-th local model (LMi) a state vector Zi (k) = AiZ (k-1) + BiU (k) is set, wherein the system matrix A, and the 20 and the entire system state is obtained by weighting the local states with the valid m ¯ function (,,) to z (k) = ^ Φ; (£ -1) { 4 · ζ (£ -l) + Biu (k)} and the global model output i = 1 to y (k) = Cz (k), with C = [1 0 ... 00], and that for each local State a local Kalman filter with the gain matrix K, is determined to estimate the local state.
    [6]
    A method according to claim 4, characterized in that the estimated state z {k) containing the estimated SoC is obtained from the equation, M,. ^) = ΣΦ ^ - ) {ζ * ψ) + κ γψ) -γ (Κ) § with z] {k) = A, z (k -1) + Btu (k) and / = 1 M m ^ Q ^ k-VCz'ik). i = 1 16/24 AV-3514 AT02
    [7]
    6. The method according to any one of claims 1 to 5, characterized in that at least one input variable of the local model network (LMN) is scaled by at least one characteristic (Cakt, Rakt) for the aging state (SoH) of the battery (10).
    [8]
    7. The method according to any one of claims 3 to 6, characterized in that by at least one parameter (Cakt, Rakt) for the aging state (SoH) of the battery (10) at least one model parameter of the local model network (LMN) is scaled.
    [9]
    8. The method according to claim 6 or 7, characterized in that the observer (13) for the SoC by an observer (14) for the characteristic (Cakt, Rakt) is extended.
    [10]
    9. The method according to claim 8, characterized in that by the observer 10 (14) for the SoH an observation error (e) between the measured output (y) and estimated output (y) of the nonlinear battery model using a predetermined measure of quality for the observation error ( e) is minimized with respect to the characteristic (Cakt, Rakt). 17/24
    类似技术:
    公开号 | 公开日 | 专利标题
    AT513189B1|2014-11-15|Method for determining a control-technical observer for the SoC
    DE102006018208B4|2017-11-23|A method and apparatus for detecting a charged state of a secondary battery based on a neural network calculation
    DE102013216200B4|2017-03-09|Online battery capacity estimate
    DE112015005201T5|2017-07-27|ASSESSMENT UNIT FOR A REMAINING STORED ENERGY AMOUNT, METHOD FOR EVALUATING A REMAINING STORED ENERGY AMOUNT OF A MEMORY BATTERY AND A COMPUTER PROGRAM
    DE102010038646A1|2012-02-02|Method and device for estimating the performance of at least one battery unit of a rechargeable battery
    DE102004035858A1|2006-02-16|State and parameter estimator with integral and differential component for electrical energy storage
    EP1120663A2|2001-08-01|Determining the capacity of an accumulator
    EP2502086A1|2012-09-26|Method and device for error-compensated current measurement of an electrical accumulator
    DE102013206895A1|2013-10-24|Battery controller with monitoring logic for model-based battery control
    EP1952169A1|2008-08-06|Method for determining storage battery operating conditions
    EP2318853A1|2011-05-11|Method for calculating the charge state of a battery
    EP3766120B1|2021-11-24|Characterisation of lithium plating in rechargeable batteries
    DE102014220913A1|2016-04-21|Method and device for determining a change in resistance of an energy storage device and vehicle
    DE102017103617A1|2018-08-23|Method for estimating the aging state of a battery system
    DE102015011745B3|2016-09-22|Charge state determination in a motor vehicle battery
    EP3117228B1|2018-01-03|Energy accumulator emulator and method for emulation of an energy accumulator
    DE102014210603A1|2015-12-17|A method of estimating an electric capacity of a secondary battery
    DE102013106083B4|2022-02-10|Method and device for determining a parameter of a model of a technical device
    DE102012016279A1|2013-03-07|Method for parameterization of impedance-based model of e.g. lithium ion battery, involves determining parameter by adjusting response with time response of system with impedance elements e.g. resistors and constant phase elements
    DE102007031303A1|2009-01-08|Method and device for determining a state variable of a motor vehicle battery that correlates with the battery state of charge using a self-learning battery model
    DE102019210212A1|2021-01-14|Method for estimating a respective parameter value of several model parameters of a model of a device, such as the battery system and the motor vehicle
    AT521643B1|2020-09-15|Method and battery management system for determining a state of health of a secondary battery
    DE102012213233B4|2015-02-12|Model-based determination of the state of charge of a rechargeable electrical energy source
    DE102019218591A1|2021-06-02|Method for estimating the state of an energy storage device
    DE102019126245A1|2021-04-01|System and method for determining the functional status and / or health status of an electric battery
    同族专利:
    公开号 | 公开日
    EP2948785B1|2016-12-21|
    CN105008946A|2015-10-28|
    US20150362559A1|2015-12-17|
    AT512003A3|2014-05-15|
    EP2948785A1|2015-12-02|
    JP6404832B2|2018-10-17|
    WO2014114564A1|2014-07-31|
    KR20150111961A|2015-10-06|
    US10338146B2|2019-07-02|
    AT513189A3|2014-07-15|
    KR102142745B1|2020-08-10|
    AT513189B1|2014-11-15|
    AT512003A2|2013-04-15|
    CN105008946B|2017-12-22|
    JP2016513238A|2016-05-12|
    引用文献:
    公开号 | 申请日 | 公开日 | 申请人 | 专利标题
    CN1199050C|1998-05-28|2005-04-27|丰田自动车株式会社|Means for estimating charged state of battery and method for estimating degraded state of battery|
    US6534954B1|2002-01-10|2003-03-18|Compact Power Inc.|Method and apparatus for a battery state of charge estimator|
    JP3935099B2|2003-04-15|2007-06-20|株式会社デンソー|Internal state detection system for power storage device for vehicle|
    AT556327T|2003-12-18|2012-05-15|Lg Chemical Ltd|DEVICE AND METHOD FOR ESTIMATING THE CHARGING STATUS OF A BATTERY USING A NEURONAL NETWORK|
    AT7710U3|2004-06-30|2006-07-15|Avl List Gmbh|METHOD FOR CREATING A MODEL OF A CONTROL SIZE FOR NONLINEAR, STATIONARY REAL SYSTEMS, FOR EXAMPLE, COMBUSTION ENGINES OR SUBSYSTEMS THEREOF|
    CA2588856C|2004-11-29|2012-11-13|Lg Chem, Ltd.|Method and system for battery state and parameter estimation|
    US7197487B2|2005-03-16|2007-03-27|Lg Chem, Ltd.|Apparatus and method for estimating battery state of charge|
    FR2949565B1|2009-09-02|2012-12-21|Inst Francais Du Petrole|IMPROVED METHOD FOR ESTIMATING NON-MEASURABLE CHARACTERISTICS OF AN ELECTROCHEMICAL SYSTEM|
    DE102009046579A1|2009-11-10|2011-05-12|SB LiMotive Company Ltd., Suwon|Improved parameter determination of an electrochemical energy store|
    TWI411796B|2009-12-22|2013-10-11|Ind Tech Res Inst|Apparatus for estimating battery's state of health|
    DE102010038646A1|2010-07-29|2012-02-02|Robert Bosch Gmbh|Method and device for estimating the performance of at least one battery unit of a rechargeable battery|
    DE102010040451A1|2010-09-09|2012-03-15|Robert Bosch Gmbh|Method for determining state quantity of motor vehicle, involves reading out correct discrete time constant parameter vector and identifying corresponding estimated state vector based on maximum probability computation|
    DE102010062856A1|2010-12-10|2012-06-21|Sb Limotive Company Ltd.|Method for determining operating parameters of a battery, battery management system and battery|
    AT511577B1|2011-05-31|2015-05-15|Avl List Gmbh|MACHINE IMPLEMENTED METHOD FOR OBTAINING DATA FROM A NON-LINEAR DYNAMIC ESTATE SYSTEM DURING A TEST RUN|
    DE102011077448A1|2011-06-14|2012-12-20|Robert Bosch Gmbh|Method for estimating state variables of an electrical energy store|
    US8706333B2|2011-06-28|2014-04-22|Ford Global Technologies, Llc|Nonlinear observer for battery state of charge estimation|
    US8880253B2|2011-06-28|2014-11-04|Ford Global Technologies, Llc|Nonlinear adaptive observation approach to battery state of charge estimation|US9602351B2|2014-06-06|2017-03-21|Microsoft Technology Licensing, Llc|Proactive handling of network faults|
    US9887878B2|2014-06-06|2018-02-06|Microsoft Technology Licensing, Llc|Dynamic scheduling of network updates|
    US9491054B2|2014-06-06|2016-11-08|Microsoft Technology Licensing, Llc|Network-state management service|
    US9381823B2|2014-07-17|2016-07-05|Ford Global Technologies, Llc|Real-time battery estimation|
    KR20170134193A|2016-05-26|2017-12-06|삼성전자주식회사|Method for estimating state of charge of battery and apparatus for impelmeing the method|
    CN110036304B|2016-12-21|2021-08-20|沃尔沃卡车集团|Battery management system and method for controlling battery management system|
    KR20190123172A|2018-04-23|2019-10-31|삼성에스디아이 주식회사|Method of estimating battery states|
    AT521643B1|2018-08-31|2020-09-15|Avl List Gmbh|Method and battery management system for determining a state of health of a secondary battery|
    CN112018465B|2020-10-13|2021-01-29|北京理工大学|Multi-physical-field-constrained intelligent quick charging method for lithium ion battery|
    CN113190969B|2021-04-07|2022-02-11|四川大学|Lithium battery model parameter identification method based on information evaluation mechanism|
    法律状态:
    优先权:
    申请号 | 申请日 | 专利标题
    ATA50046/2013A|AT512003A3|2013-01-23|2013-01-23|Method for determining a control-technical observer for the SoC|
    ATA50736/2013A|AT513189B1|2013-01-23|2013-11-06|Method for determining a control-technical observer for the SoC|ATA50736/2013A| AT513189B1|2013-01-23|2013-11-06|Method for determining a control-technical observer for the SoC|
    KR1020157022682A| KR102142745B1|2013-01-23|2014-01-17|Method for determining a control observer for the soc|
    CN201480011726.3A| CN105008946B|2013-01-23|2014-01-17|Method for determining the control technology observer for SoC|
    EP14700888.2A| EP2948785B1|2013-01-23|2014-01-17|Method for determining a control observer for the soc|
    PCT/EP2014/050905| WO2014114564A1|2013-01-23|2014-01-17|Method for determining a control observer for the soc|
    US14/762,719| US10338146B2|2013-01-23|2014-01-17|Method for determining a control observer for the SoC|
    JP2015554111A| JP6404832B2|2013-01-23|2014-01-17|Method of determining control technical observer for SoC|
    [返回顶部]