DE102022200538A1 - Method and device for initially providing an aging state model for energy stores based on active learning algorithms - Google Patents

Abstract

The invention relates to a method for initially providing an at least partially data-based aging state model (4) for an electrical energy storage device (3), with the following steps: - Providing (S1) a number of energy storage devices (3) on a test bench (1) for measurement depending on a respective load profile, the load profiles being different and characterizing a time course of at least one stressful operating variable for the energy storage device (3); - Operating (S2) the number of energy storage devices (3) with the respectively assigned load profile and acquisition of temporal performance of operating variables;- At a given evaluation time, determination (S4) of an aging state of a subset of the energy stores (3) as a label and generation of a training data set with the performance of operating variables and the specific label for each energy store (3) of the subset of the energy stores (3);- Selecting (S6) of the subset of the energy stores (3) with the respectively assigned load profile depending on an optimization method that is based on the total costs (CJ) of measuring the energy stores (3 ) on the test bench (1) and from a total amount of information (InfoJ) of the measurement of the energy storage device (3).

Description

technical field

The invention relates to methods and devices for initially providing an at least partially data-based aging state model for electrical energy stores of the same type, and in particular to methods for predicting the measurement costs and/or the measurement time.

Technical background

The power supply for the operation of mains-independent electrical devices and machines such. B. electrically driven motor vehicles, is usually done with electrical energy storage devices, such as. Device batteries or vehicle batteries.

Electrical energy stores degrade over their service life and depending on their load or use. This so-called aging leads to a continuously decreasing maximum performance and storage capacity. The aging state corresponds to a measure of the aging of energy storage devices. In the case of portable batteries as electrical energy stores, according to the convention, a new portable battery can have an aging status (in terms of its capacity, SOH-C) of 100%, which decreases noticeably over the course of its service life.

A measure of the aging of an electrical energy store (change in the state of aging over time) depends on an individual load on the energy store, i. H. in vehicle batteries of motor vehicles on usage behavior by a driver, external environmental conditions and the vehicle battery type.

In order to monitor the aging states of electrical energy stores in a large number of devices, operating variable data can be continuously recorded and transmitted in blocks to a device-external central unit as operating variable curves.

For the model-based determination of an aging state of an electrical energy store using operating variable data, it is necessary to provide an initial aging state model. For this purpose, an initial measurement, e.g. B. in a laboratory or a test bench, provided in order to generate training data for an aging model to be provided. For this purpose, the energy stores are operated in different ways, with energy having to be supplied or derived, in particular depending on the type of energy store, in order to simulate operating cycles of the energy stores. The energy required for this is considerable and scales with the number of energy stores to be initially measured. The initial measurement can also take a considerable amount of time, in particular if sufficient training data is still to be recorded for aged energy stores.

Disclosure of Invention

According to the invention, a method for initially providing an at least partially data-based aging state model for an energy storage type according to claim 1 and a corresponding device according to the independent claim are provided.

Further developments are specified in the dependent claims.

• - Bereitstellen einer Anzahl von Energiespeichern auf einem Prüfstand zur Vermessung abhängig von einem jeweiligen Lastprofil, wobei die Lastprofile unterschiedlich sind und einen zeitlichen Verlauf mindestens einer belastenden Betriebsgröße für den Energiespeicher charakterisieren;
• - Betreiben der Anzahl von Energiespeichern mit dem jeweils zugeordneten Lastprofil und Erfassen von zeitlichen Betriebsgrößenverläufen;
• - Zu jeweils einem vorgegebenen Auswertungszeitpunkt, Bestimmen eines Alterungszustands einer Untermenge der Energiespeicher als Label und Generieren eines Trainingsdatensatzes mit den Betriebsgrößenverläufen und dem bestimmten Label für jeden Energiespeicher der Untermenge der Energiespeicher;
• - Auswählen der Untermenge der Energiespeicher mit dem jeweils zugeordneten Lastprofil abhängig von einem Optimierungsverfahren, das von Kosten einer Vermessung der Energiespeicher auf dem Prüfstand und von einem Informationsmaß der Vermessung der Energiespeicher abhängt.

According to a first aspect, a method is provided for initially providing an at least partially data-based aging state model for an electrical energy store, with the following steps:

• - Providing a number of energy stores on a test bench for measurement depending on a respective load profile, the load profiles being different and characterizing a time course of at least one stressful operating variable for the energy store;
• - Operating the number of energy storage devices with the respectively assigned load profile and detecting time-related performance variables;
• - At a given evaluation time, determining an aging condition of a subset of the energy storage as a label and generating a training data set with the Operating quantity curves and the specific label for each energy store of the subset of energy stores;
• - Selecting the subset of the energy storage with the associated load profile depending on an optimization method that depends on the cost of a measurement of the energy storage on the test bench and an information measure of the measurement of the energy storage.

The aging condition of an electrical energy store is usually not measured directly. This would require a number of sensors inside the energy store, which would make the production of such an energy store cost-intensive and complex and would increase the space requirement. In addition, measurement methods suitable for everyday use for directly determining the state of aging in the energy storage devices are not yet available on the market.

For capacity reasons, the monitoring of energy stores in a large number of devices is therefore carried out in a device-external central unit. For this purpose, the devices can transmit operating variable profiles of operating variables of the energy stores to the central unit, with a current electrochemical state and/or aging state being determined in the central unit. Depending on the model used, time series of operating variables are continuously recorded as operating variable curves, such as a battery current, battery temperature, state of charge and/or battery voltage for a device battery as an energy store, and transmitted to the central unit in blocks and, if necessary, in compressed form. There, the performance variable curves are evaluated so that, based on one or more aging state models, a device-specific aging state and possibly other variables can be calculated/determined. In addition, the operating variables from the large number of energy stores can be evaluated using statistical methods in order to improve the applied aging state models, so that the determination and prediction of the aging state of the energy stores can be noticeably improved.

The state of health (SOH: State of Health) is the key variable for specifying a remaining battery capacity or remaining battery charge for device batteries as energy storage devices. The aging status is a measure of the aging of the device battery. In the case of a device battery or a battery module or a battery cell, the aging status can be specified as the capacity retention rate (SOH-C). The capacity maintenance rate SOH-C, i. H. the capacity-related state of aging is specified as the ratio of the measured instantaneous capacity to an initial capacity of the fully charged battery and decreases with increasing aging. Alternatively, the aging state can be specified as an increase in the internal resistance (SOH-R) in relation to an internal resistance at the beginning of the service life of the device battery. The relative change in internal resistance SOH-R increases as the battery ages.

Due to the often physically difficult to describe electrochemical effects during the operation of an energy storage device, the use of a data-based model as or in connection with an aging state model has proven itself.

A possible aging model can be provided in the form of a hybrid aging model, which corresponds to a combination of a physical aging model with a data-based model. In a hybrid model, a physical aging state can be determined using a physical or electrochemical aging model and a correction value can be applied to it, which results from a data-based correction model, in particular by addition or multiplication. The physical aging model is based on electrochemical model equations, which characterize the electrochemical states of a nonlinear differential equation system with regard to aging reactions, continuously calculated according to a time integration method and mapped to the physical aging state for output, as SOH-C and/or as SOH-R. The calculations can typically be made in the central unit (cloud) at intervals of predetermined evaluation periods of z. B. be run once a week.

Furthermore, the correction model of the hybrid data-based aging status model can be designed with a probabilistic regression model or a probabilistic regression model based on artificial intelligence, in particular a Gaussian process model, and can be trained to correct the aging status obtained by the physical aging model. For this purpose, a data-based correction model for correcting the capacity-related aging state and possibly a further data-based correction model for correcting the resistance change-related aging state can be provided. Possible alternatives to the Gaussian process are further supervised learning methods, such as based on a random forest model, an AdaBoost model, a support vector machine or a Bayesian neural network.

When a new type of energy storage device is put into operation, it is necessary to be able to determine the aging condition using a aging condition model. Since there is generally no precise knowledge of the aging of the energy storage device when it is put into operation, it is necessary to initially specify an aging condition model that can indicate an aging condition at least approximately based on operating variable data. The initial specification of the state of health model therefore requires an initial training of the data-based model.

Training data for such an initial training are usually carried out in the laboratory or on a test stand and include a predetermined number of energy storage devices, selected in particular at random, which are each operated with load profiles that differ from one another. These load profiles include or are converted into cyclic current flows, temperature profiles and the like. In particular in the case of portable batteries as energy stores, these load profiles can include different charging and discharging current profiles at different temperatures and can in particular be specified and characterized in a compressed manner by histogram data. These load profiles are converted into time series of power supply and power drain and corresponding operating variables, such as battery current, battery voltage, battery temperature and state of charge, are recorded and stored as operating variable profiles.

At specified points in time, measurements of the aging state are carried out using suitable additional models or measurement methods in order to determine a label for the course of the operating variables. From this, training data sets are formed, which can be used to parameterize the aging model and/or to train the data-based model in the case of a hybrid or purely data-based aging model. Various aging state models or methods can be used to determine the label.

A basic model can be provided as a possible model or method for determining an aging state, according to which an SOH-C measurement is carried out by Coulomb counting or by forming a current integral over time during the charging process, which is determined by the rise in the state of charge between the beginning and is divided at the end of the relevant charging and/or discharging phase. In this case, the calibration is advantageously carried out on the no-load voltage characteristic in idle phases in order to also calculate the course of the state of charge in the central unit. Sufficiently reliable information about the aging state can be obtained, for example, when the vehicle battery is brought from a completely discharged state of charge to a fully charged state during a charging process from a defined relaxed state under reproducible load and environmental conditions. The maximum charge detected in this way can be related to an initial maximum charge capacity of the vehicle battery. Resistance-related aging states (SOH-R values) can also be calculated from voltage changes related to a current change. These are usually related to a defined time interval as well as defined environmental conditions and the energy flow direction of the system.

According to the above method, the aging states of a subset of the energy stores can now be determined as a label and a training data set with the performance variable curves and the specific label can be generated for each energy store in the subset of the energy stores. The selection of the subset of the energy stores with the respectively assigned load profile is carried out depending on an optimization method.

The optimization method for selecting the subset of the energy stores can have the goal of minimizing the costs of the measurement on the test bench and maximizing an information gain for the creation of the initial aging state model, which is determined by the information measure.

According to the above method, the number of energy stores measured or being measured is successively reduced, depending on the information gain that can be obtained by further measuring the relevant energy store for the further development of the aging model.

According to one embodiment, with a Gaussian process model as a probabilistic model, the predictive covariance from the input vectors of one or more of the energy stores for the evaluation processing times of an entire measurement period are determined, with the input vectors being determined from the load profiles.

des datenbasierten Modells ermittelt, die sich aus dem im Alterungszustandsmodell vorgesehenen probabilistischen Regressionsmodell (Gaußprozessmodell) ergibt. Die prädiktive Kovarianz hängt bei Gauß-Prozess-Modellen als probabilistischem Regressionsmodell nicht von bisher ungesehenen Alterungszuständen ab, sondern nur von den Eingangsgrößen x̅_J(t) = [x(t), m(t ), z, Phys[x(t)]) des Gauß-Prozess-Modells und kann daher als $${\displaystyle \sum \left({\overline{x}}_j\left(t+t_1\right),{\overline{x}}_j\left(t+t_2\right),\dots ,{\overline{x}}_j\left(t+t_n\right)\right)}$$

geschrieben werden, wobei x(t) einem oder mehreren Betriebsgrößenverläufen, m(t) Histogrammdaten als Lastprofil, z(t) mehrdimensionalen elektrochemischen Zuständen des physikalischen Alterungsmodells, wie z.B. die SEI-Dicke, der Menge an zyklisierbarem Lithium, der Menge an Aktivmaterial, elektrochemischen Konzentrationen und dergleichen, und Phys[x(t)] dem physikalischen (modellierten) Alterungszustand entsprechen.A method is thus proposed in which the aging model is successively trained and those batteries are removed from the measurement or not taken into account that make a small contribution to the further development of the aging model. For this purpose, when using a probabilistic regression model as a data-based model at each evaluation time, the predictive covariance $${\displaystyle \sum \left(SO{H}_j\left(t+t_1\right),SO{H}_j\left(t+t_2\right),...,SO{H}_j\left(t+t_n\right)\right)}$$

of the data-based model, which results from the probabilistic regression model (Gaussian process model) provided in the aging model. With Gaussian process models as a probabilistic regression model, the predictive covariance does not depend on previously unseen aging states, but only on the input variables x̅ _J (t) = [x(t), m(t ), z, Phys[x(t) ]) of the Gaussian process model and can therefore be used as $${\displaystyle \sum \left({\overline{x}}_j\left(t+t_1\right),{\overline{x}}_j\left(t+t_2\right),...,{\overline{x}}_j\left(t+t_n\right)\right)}$$

are written, where x(t) one or more operating variable curves, m(t) histogram data as a load profile, z(t) multidimensional electrochemical states of the physical aging model, such as the SEI thickness, the amount of cyclable lithium, the amount of active material, electrochemical concentrations and the like, and Phys[x(t)] correspond to the physical (modeled) aging state.

The predictive covariance is a matrix to which measures of information, such as the determinant or the maximum eigenvalue, can be applied. An information measure is thus obtained for each of the measured energy stores. The information measure for several energy stores J = (j1 ... jm) can be given as Info _J = det(Σ (x̅ _j1 (t + t ₁ ), ..., x̅ _j1 (t + t _n ), x̅ _jm (t + _{t 1} ) _, .

Furthermore, a cost measure is used and determined for each energy store to be measured. The cost measure can include the energy use or requirement during the laboratory or test bench measurements and the duration of the entire measurement of the energy storage. In addition to this, the cost measure can also take into account the total test bench costs, the occupation time of the test bench and the other use of materials. Thus, a cost measure C can be predictively provided for each of the number J of energy storage devices to be measured as follows: $$\begin{array}{c}{C}_j=C\left({\overline{x}}_j\left(t+t_1\right),{\overline{x}}_j\left(t+t_2\right),...,{\overline{x}}_j\left(t+t_n\right)\right)\\ C_j={\mathrm{<figure-callout\; id="1"\; label="C"\; filenames="DE102022200538A1\_0011.png,DE102022200538A1\_0013.png"\; state="\{\{state\}\}">C</figure-callout>}}_1+\mathrm{..}+C_m\end{array}$$

With the set of cost information points for many of the batteries to be measured, the optimization problem can now be solved and a Pareto front can be determined. The Pareto front represents the cost measure of the achievable information measure of the initial state of health model to be created.

In terms of a variant of an active learning method, the selection of the cost-information measure points can now provide for selecting those energy stores for which these cost-information measure points are as close as possible to or on the Pareto front.

• - basierend auf einer Zielfunktion und insbesondere mit einem Greedy-Algorithmus durchgeführt wird, wobei die Zielfunktion eine gewichtete Summe der Kosten für die Vermessung der Anzahl von Energiespeichern und eines gesamten Informationsmaßes durch die Vermessung der Untermenge der Energiespeicher darstellt,
• - durch Maximierung eines gesamten Informationsmaßes unter der Nebenbedingung, dass die gesamten Kosten des Vermessens der so ausgewählten Fahrzeugbatterien kleiner sind als vorgegebene Maximalkosten;
• - durch Minimierung von gesamten Kosten unter der Nebenbedingung, dass das gesamte Informationsmaß größer ist als ein vorgegebenes minimales Informationsmaß;
• - durch Minimierung von gesamten Kosten unter der Nebenbedingung, dass eine Wahrscheinlichkeit, dass die gesamten Kosten kleiner sind als vorgegebene Maximalkosten, größer ist als eine vorgegebene Wahrscheinlichkeit; oder
• - durch Maximierung eines gesamten Informationsmaßes unter der Nebenbedingung, dass eine Wahrscheinlichkeit, dass das Informationsmaß größer ist als ein vorgegebenes Mindest-Informationsmaß, größer ist als eine vorgegebene Wahrscheinlichkeit.

It can be provided that the optimization method

• - is carried out based on a target function and in particular with a greedy algorithm, the target function representing a weighted sum of the costs for measuring the number of energy stores and a total information measure by measuring the subset of the energy stores,
• - by maximizing an overall information measure under the constraint that the overall cost of metering the vehicle batteries so selected is less than a predetermined maximum cost;
• - by minimizing total costs under the constraint that the total information metric is greater than a predetermined minimum information metric;
• - by minimizing total cost under the constraint that a probability of total cost being less than predetermined maximum cost is greater than a predetermined probability; or
• - by maximizing an overall information measure under the constraint that a probability that the information measure is greater than a predetermined minimum information measure is greater than a predetermined probability.

According to the above approaches for the optimization method, a weighting between costs and information gain can be specified by a target function, for example Z _j =Info _J +αC _J with a user-defined weighting parameter α. Now the best energy stores can be determined, in particular by applying a greedy algorithm. Greedy algorithms first choose the best energy storage and then add the second best, given that the best energy storage is already being measured. As an alternative to preselecting the above objective function, a constrained optimization problem can also be solved by limiting the maximum cost and by limiting the minimum information gain.

As an alternative to weighting the objective function with the weighting parameter, a limited optimization problem can also be solved by maximizing the information measure Info _J in order to select a subset of energy stores to be selected. This takes place under the secondary condition that the sum of the costs of measuring the selected energy stores is less than a predefined user parameter that indicates the maximum costs that may be.

As an alternative to weighting the target function with the weighting parameter, a limited optimization problem can also be solved by minimizing the costs in order to select a subset of energy stores to be selected. This takes place under the secondary condition that the information measure Info _J is greater than a specified minimum information gain.

If probabilistic regression models are used, then those energy storage devices can be selected for which the probability that the sum of the costs for a measurement of the selected energy storage device with the highest information dimensions is less than a predefined parameter is greater than a predefined parameter. Thus, a discrete cost-information gain point of the Pareto front can be provided, which is model-based linked to recommendations for action and specifically specifies energy storage for selection for further operation.

According to the method, concrete measurements of the aging state are now made for the selected energy stores and not made for the unselected energy stores. In this way, a training data set is obtained together with the associated performance variable curves

The data-based aging state model can be designed with a probabilistic data-based model, with an input vector of the data-based model for one of the energy storage devices, which comprises at least one operating variable profile, operating characteristics from the operating variable profiles, an internal state of the energy storage device and/or a physically modeled aging status, towards which modeling aging condition of the relevant energy store or a correction variable for correcting a physically modeled aging condition of the relevant energy store is mapped, the information measure for an energy store being determined as the determinant of the predictive covariance.

The predictive covariance can be determined from the input vectors of one or more of the energy stores for the evaluation times of an entire measurement period.

When sufficient new information is available, the data-based aging model is retrained. The training can be supplemented by automated hyperparameter tuning, e.g. B. via gradient-based methods or black box methods, such as Bayesian optimization. The method is terminated when a certain accuracy requirement is not met, such as <1.5 SOHC on a relevant, previously provided validation data set, and other robustness requirements, such as evaluated via cross-validation, are met.

It can be provided that at each evaluation time only the energy stores of the subset of energy stores are measured as a label to determine an aging state and the remaining energy stores continue to be operated according to the load profile.

In an alternative embodiment, only the energy stores of the subset of energy stores can be measured as labels at each evaluation time to determine an aging state and can continue to be operated according to the load profile, while the remaining energy stores are removed from the test bench.

Thus, according to the selected energy stores, the remaining energy stores, which lead to insufficient information gain, can be excluded from further measurement and removed from the test bench. As a result, the measurement can only be continued with a reduced number of selected energy stores in the test stand. This method can be carried out successively in order to reduce the number of energy storage devices in the measurement and thereby reduce the costs for using the test bench.

According to a further aspect, a device for carrying out one of the above methods is provided.

character list

• 1 eine schematische Darstellung eines Prüfstands zur Vermessung einer Vielzahl von Fahrzeugbatterien zum Erstellen eines initialen Alterungszustandsmodells;
• 2 eine schematische Darstellung eines hybriden Alterungszustandsmodells;
• 3 ein Flussdiagramm zur Veranschaulichung eines Verfahrens zum optimierten Erstellen eines initialen Alterungszustandsmodells;
• 4 eine Darstellung einer Pareto-Front aus Kosten-Informationsmaß-Punkten für die verschiedenen Fahrzeugbatterien in der Vermessung; und
• 5 eine Darstellung eines Verlaufs von Alterungszuständen verschiedener Fahrzeugbatterien bei unterschiedlichen Lastmustern.

Embodiments are explained in more detail below with reference to the accompanying drawings. Show it:

• 1 a schematic representation of a test bench for measuring a plurality of vehicle batteries to create an initial state of health model;
• 2 a schematic representation of a hybrid aging state model;
• 3 a flow chart to illustrate a method for the optimized creation of an initial state of health model;
• 4 a representation of a Pareto front of cost-information measure points for the various vehicle batteries in the survey; and
• 5 a representation of a course of aging states of different vehicle batteries with different load patterns.

Description of Embodiments

1 shows a schematic arrangement of a test stand with a test stand unit 2 which is connected to a large number of connected vehicle batteries 3 as electrical energy stores. The test stand unit 2 controls the vehicle batteries 3 according to a predetermined load pattern that characterizes different current curves and/or temperature curves. The current curves and/or temperature curves cause different loads and are therefore different cyclic aging of the vehicle batteries 3, for example by specifying an ampere-hour throughput, specifications for charging profiles, in particular maximum charging currents depending on the state of charge when charging, maximum charging currents when discharging, maximum and average Charging and discharging currents and corresponding temperature conditions. The load patterns are specified for each of the vehicle batteries 3 and correspond to stressful or less stressful operating modes of the vehicle batteries 3. Operating variable profiles of the vehicle batteries 3 over time are continuously recorded and temporarily stored. In the case of vehicle batteries, the performance variable curves include a battery voltage, a battery current, a battery temperature and a state of charge.

The test stand 1 is used to record data on the aging of the vehicle batteries 3 in order to provide a corresponding initial aging state model 4, which makes it possible to determine the current aging state of the vehicle battery in question with a predetermined minimum accuracy, depending on the operating variable curves recorded in real operation. The aging state model 4 to be created for this purpose can be at least partially data-based and have a data-based probabilistic regression model.

The measurement of the large number of device batteries 3 on the test stand 1 provides for selected vehicle batteries 3 at regular times using a suitable method to determine an aging assessment to determine the status. In connection with the corresponding associated operating variable curves, which are specified by the load pattern assigned to the vehicle battery 3 in question, this results in training data sets that can be used to train the data-based/hybrid state of health model 4 . The use of the test stand 1 and the measurement of the large number of vehicle batteries 3 causes costs that are caused on the one hand by the energy consumption, on the other hand by the occupancy time of the test stand and the other use of materials. These costs should be reduced when measuring the large number of vehicle batteries 3 without impairing the quality of the initially trained state of health model 4 .

2 shows a schematic example of the functional structure of an embodiment of a data-based aging model 9, which is designed in a hybrid manner. The aging state model 9 includes a physical aging model 5 and a data-based correction model 6.

The physical aging model 5 is a mathematical model based on non-linear differential equations. The evaluation of the physical aging model 5 of the aging state model 9 with operating variable profiles, in particular since the beginning of the service life of the vehicle battery, results in an internal state of the system of equations of the physical differential equations being set, which corresponds to a physical internal state of the vehicle battery. Since the physical aging model is based on physical and electrochemical laws, the model parameters of the physical aging model are quantities that specify physical properties.

The time series of the operating variables x(t) of the vehicle battery to be evaluated go directly into the physical aging state model 5, which is preferably designed as an electrochemical model and corresponding internal electrochemical states z(t), such as layer thicknesses (e.g. SEI thickness), change in the cyclable Lithiums due to anode/cathode side reactions, fast consumption of electrolyte, slow consumption of electrolyte, loss of active material in anode, loss of active material in cathode, etc. are modeled using nonlinear differential equations in a multidimensional state vector.

The physical aging model 5 thus corresponds to an electrochemical model of the vehicle battery 3 in question. This model determines internal physical battery states z(t) depending on the performance variable curves x(t) in order to obtain a physically based aging state SOHph = Phys[x(t)] of the dimension of at least one, depending on the above electrochemical states z(t), which are linearly or non-linearly mapped to a capacity retention rate (SOH-C) and/or an internal resistance increase rate (SOH-R) to define them as aging states (SOH-C and SOH -R) to provide.

However, the model values for the physical aging state SOHph provided by the electrochemical model are imprecise in certain situations, and it is therefore provided to correct them with a correction variable k. The correction variable k is provided by the data-based correction model 6, which is trained using training data sets that are determined by the test bench 1.

wobei µ ein m x 1-dimensionaler Vektor ist, der den prädiktiven Mittelwert angibt und Σ die m × m dimensionale prädiktive Kovarianzmatrix des Gaußprozessmodells ist. Die Formeln für den Mittelwert und die Varianz vom Gaußprozess sind wie folgt: $$\mu \left(x*\right)=k^T{C}_N^{-1}y$$

$${\displaystyle \sum \left(x*\right)}=c-k^T{C}_N^{-1}k$$

wobei N der Anzahl von Labels entspricht, x* m × d dimensional ist und eine Menge an m neuen Punkten im Eingangsraum angibt, wobei [x(t),x(t_hist),z,Phys[x(t)]) =x' und y jeweils einem N × 1 dimensionalen Vektor entsprechen, und wobei y jeweils ein gemessener oder als Label bestimmter Alterungszustand angibt, k entspricht einer N × m dimensionalen Matrix von Kernelauswertungen, die im Kernel des Gauß Prozesses kodierten Korrelationen zwischen den m neuen Punkten und den N gemessenen Punkten angeben. C entspricht einer N × N dimensionalen Matrix von Kernelauswertungen zwischen den N gemessenen Punkten und c einer m × m Matrix mit Kernelauswertungen zwischen den m neuen Punkten. T kennzeichnet das Transponierte. Siehe auch Bishop, „Pattern Recognition and Machine Learning“, 2006. The data-based correction model preferably corresponds to a Gaussian process model $GP=GP[x\left(t\right),x\left(t_{Hist}\right),\mathrm{e.g},PHys\left[x\left(t\right)\right]))\sim \mathcal{N}\left(\mu ,\sum \right),$

where µ is an mx 1-dimensional vector denoting the predictive mean and Σ is the m × m dimensional predictive covariance matrix of the Gaussian process model. The formulas for the mean and variance of the Gaussian process are as follows: $$\mu \left(x*\right)=k^T{C}_N^{-1}y$$

$${\displaystyle \sum \left(x*\right)}=c-k^T{C}_N^{-1}k$$

where N is the number of labels, x* is m × d dimensional and specifies a set of m new points in the input space, where [x(t),x(t _hist ),z,Phys[x(t)]) = x' and y each correspond to an N × 1 dimensional vector, and where y indicates an aging state measured or determined as a label, k corresponds to an N × m dimensional matrix of kernel evaluations, the correlations between the m new points encoded in the kernel of the Gaussian process and indicate the N measured points. C corresponds to an N×N dimensional matrix of kernel scores between the N measured points and c an m × m matrix with kernel scores between the m new points. T denotes the transposed. See also Bishop, "Pattern Recognition and Machine Learning", 2006.

The correction model 6 receives operating characteristics m(t) on the input side, which are determined from the curves of the operating variables x(t) and can also include one or more of the internal electrochemical states of the differential equation system of the physical model 5 . Furthermore, the correction model 6 can receive the physical aging state SOHph obtained from the physical aging model 5 on the input side. The operating features m(t) of the current evaluation period are generated in a feature extraction block 8 based on the operating variable profiles x(t). Furthermore, the internal states z(t) from the state vector of the electrochemical physical aging model 5 and advantageously the physical aging state SOHph are supplied to the correction model 6 . The feature vector m(t) is robust because it is independent of the model quality of the physical aging model or its state. Therefore, considering the feature vector m(t) is a useful addition to the internal states z(t) of the physical or electrochemical aging model.

Operating variables m(t) relating to an evaluation period can be generated from the operating variables x(t) in the central unit 2 for each vehicle fleet 3 or, in other embodiments, already in the respective motor vehicles. The evaluation period for determining the aging state can be a few hours (e.g. 6 hours) to several weeks (e.g. one month). A usual value for the evaluation period is one week.

The operating characteristics m(t) can include, for example, characteristics related to the evaluation period and/or accumulated characteristics and/or statistical variables determined over the entire previous service life. In particular, the operating characteristics can include, for example: electrochemical states, such as e.g. B. SEI layer thickness, change in cyclizable lithium due to anode/cathode side reactions, rapid uptake of electrolyte solvents, slow uptake of electrolyte solvents, lithium deposition, loss of active anode material and loss of active cathode material, information on impedances or the internal resistances, histogram characteristics, such as temperature versus state of charge, charging current versus temperature and discharging current versus temperature, in particular multi-dimensional histogram data relating to battery temperature distribution versus state of charge, charging current distribution versus temperature and/or discharging current distribution versus temperature, current throughput in ampere-hours, accumulated total charge (Ah), a Average increase in capacity during a charge process (especially for charge processes in which the increase in charge is above a threshold proportion [e.g. 20% ΔSOC] of the total battery capacity), the charge capacity and an extreme value (e.g. B. Maximum) of the differential capacity during a measured charging process with a sufficiently large swing in the state of charge (smoothed curve of dQ/dU: change in charge divided by change in battery voltage) or the accumulated mileage. These variables are preferably converted in such a way that they characterize the real usage behavior in the best possible way and normalize it in the feature space. The operating characteristics m(t) can be used in whole or in part for the method described below.

To determine a corrected aging state SOH to be output, the outputs SOHph, k of the physical aging model 5 and the data-based correction model 6, which is preferably designed as a Gaussian process model, are applied to one another. In particular, these can be added in a summation block 7 or otherwise also multiplied (not shown) in order to obtain the modeled aging state SOH to be output for a current evaluation period. In addition, the confidence of the Gaussian process can also be used as the confidence of the corrected aging value SOH of the hybrid model to be output. The confidence or the confidence value of the Gaussian process model thus characterizes the modeling uncertainty of the mapping of operating characteristic points to an aging state.

The initial training of the hybrid aging state model 9 takes place in the test bench 1. For this purpose, training data sets are created that assign the operating variable curves of a vehicle battery operated as a function of a load profile to an aging state determined empirically or model-based as a label.

For example, to determine an aging status as a label for training the hybrid or data-based aging status model, a basic model can be provided, according to which an SOH-C measurement is carried out by Coulomb counting or by forming a time-related current integral during the charging process, which is carried out by the hub the state of charge between the beginning and the end of the relevant charging and/or discharging phase. Sufficiently reliable information about the aging state can be obtained, for example, when the vehicle battery is brought from a completely discharged state of charge to a fully charged state during a charging process from a defined relaxed state under reproducible load and environmental conditions. The maximum charge detected in this way can be related to an initial maximum charge capacity of the vehicle battery. Resistance-related aging conditions (SOH-R values) can also be calculated from voltage changes related to a current change. These are usually related to a defined time interval as well as defined environmental conditions and the energy flow direction of the system.

An aging state can be determined as a label in a manner known per se by evaluating the performance variable curves with an additional aging model under defined load and environmental conditions of a label generation, such as e.g. B. constant temperature, constant current and the like. Other models can be used to determine the aging condition. The data-based correction model can be trained in a conventional manner based on the training data records and on the residuals of the modeled aging state.

In 3 a flowchart is shown that describes the sequence of the method for carrying out the measurement of the plurality of vehicle batteries 3 on the test stand 1 . The method is executed in the test bench control device 2 and leads to the provision of an initial data-based state of health model 4 that allows sufficient accuracy in determining the state of health based on operating variable profiles of vehicle batteries 3 in real operation.

In step S1, the test stand 1 is equipped with a large number of brand-new (or in a reference state) vehicle batteries 3 of the same type, and each of the vehicle batteries 3 is assigned a predefined load pattern. The load patterns are different and each represent a load in the range of a low to high load for the associated vehicle battery 3. The load patterns provide information from which temporal curves of a battery current in connection with temperature curves can be derived, which the vehicle batteries 3 in different ways stress.

In step S2, the plurality of on-vehicle batteries 3 are operated according to the predetermined load pattern.

In step S3 it is checked whether an evaluation time has been reached. The evaluation time can be provided at regular time intervals, such as at time intervals between one week and two months.

If an evaluation time has been reached (alternative: yes), the method continues with step S4, otherwise a jump is made back to step S2.

nicht von bisher nicht ermittelten Alterungszuständen ab, sondern nur von Eingangsgrößen x̅_J(t) = [x(t), m(t), z, Phys[x(t)]) ab. Man kann daher auch schreiben Σ(x̅_j(t + t₁), x̅_j(t + t₂), ...,x̅_j(t + t_n)). Dies stellt eine Matrix dar, die mit Informationsmaßen, wie der Determinante oder dem maximalen Eigenwert, bewertet werden kann. Der Informationsgewinn kann sich somit aus der Auswertung der sich aus den Lastmustern ergebenden Betriebsgrößenverläufe in einfacher Weise ergeben.In step S4, based on the previous training state of the state of health model, an information measure is now determined for each of the plurality of vehicle batteries 3, which indicates which information gain can be obtained in the course of the further measurement of the vehicle battery 3 in question. This measure of information can be expressed as a predictive covariance. In particular, when using a Gaussian process model as a probabilistic regression model, the predictive covariance depends $${\displaystyle \sum \left(SO{H}_j\left(t+t_1\right),SO{H}_j\left(t+t_2\right),...,SO{H}_j\left(t+t_n\right)\right)}$$

not from previously undetermined aging states, but only from input variables x̅ _J (t) = [x(t), m(t), z, Phys[x(t)]). One can therefore also write Σ(x̅ _j (t + t ₁ ), x̅ _j (t + t ₂ ), ...,x̅ _j (t + t _n )). This represents a matrix that can be evaluated with information measures such as the determinant or the maximum eigenvalue. The gain of information can thus result in a simple manner from the evaluation of the performance variable curves resulting from the load patterns.

angegeben werden.As a result, in step S4, the information measures about the expected information gains from the individual vehicle batteries 3 of the plurality of vehicle batteries 3. The combination of information measures (total information measure) of several vehicle batteries 3 J = (j1 ... jm) can be determined by the covariance matrix $$Inf{O}_j=\text{de}\left(\sum \left({\overline{x}}_{j1}\left(t+t_1\right),...,{\overline{x}}_{j1}\left(t+t_n\right),{\overline{x}}_{jm}\left(t+t_1\right),...,{\overline{x}}_{jm}\left(t+t_n\right)\right)\right)$$

be specified.

und als gesamte Kosten für die Vermessung einer bestimmten Menge von Fahrzeugbatterien 3 $$C_j=C_1+\mathrm{..}+C_m$$

wobei 1...m Indizes von Fahrzeugbatterien 3 einer Untermenge der Anzahl J entsprechen.A cost measure for each of the vehicle batteries 3 is calculated in a step S5. The cost measure can include the costs of using the test bench, including the expected energy consumption and the test time as well as other material usage. The cost measure can thus be provided predictively for each of the plurality of vehicle batteries 3 if the costs are added at the discretized evaluation times up to the end t _n of the measurement duration. Thus, a cost for each of the vehicle batteries 3 is obtained $$C_j=C\left({\overline{x}}_j\left(t+t_1\right),{\overline{x}}_j\left(t+t_2\right),...,{\overline{x}}_j\left(t+t_n\right)\right)$$

and as the total cost of measuring a specific quantity of vehicle batteries 3 $$C_j={\mathrm{<figure-callout\; id="1"\; label="C"\; filenames="DE102022200538A1\_0011.png,DE102022200538A1\_0013.png"\; state="\{\{state\}\}">C</figure-callout>}}_1+\mathrm{..}+C_m$$

where 1...m indices of vehicle batteries 3 correspond to a subset of the number J.

Thus, for each of the vehicle batteries 3, a measure of information is obtained about the probable gain of information and the corresponding costs that a further measurement of the vehicle battery 3 in question will cause on the test bench. A Pareto front results, for example as in 4 is shown schematically.

By selecting cost-information measure points that correspond to the costs and information measures of each of the vehicle batteries 3, those vehicle batteries 3 can be selected in step S6 for which a label in the form of a measured state of aging is to be determined at the current evaluation time.

The aging state can be determined using a suitable aging state model or a suitable measurement method for determining the aging state.

A measurement by Coulomb counting or by forming a current integral over time during the charging process can be used as a possible model or method for determining an aging state. In this case, the transferred charge is divided by the change in the state of charge between the beginning and the end of the relevant charging and/or discharging phase. In this case, the calibration is advantageously carried out on the no-load voltage characteristic in idle phases in order to also calculate the course of the state of charge in the central unit. Sufficiently reliable information about the aging state can be obtained, for example, when the vehicle battery is brought from a completely discharged state of charge to a fully charged state during a charging process from a defined relaxed state under reproducible load and environmental conditions. The maximum charge detected in this way can be related to an initial maximum charge capacity of the vehicle battery 3 .

The selection of the vehicle batteries 3 to be measured can be determined with an optimization method according to a target function that weighs the costs and the information gained against each other. The objective function can be of the form: Z _j = Info _J + αC _J with a user defined weight parameter α and a predetermined number of vehicle batteries can be determined for which the objective function yields the maximum values. To solve the optimization problem, greedy algorithms can be used, which first add the best vehicle battery 3 (highest result of the target function Z _J ) and then the next best one (with respect to Z _J ), assuming that the previously determined vehicle batteries 3 have already been measured are.

As an alternative to weighting the target function with the weighting parameter, a limited optimization problem can also be solved by maximizing the information measure Info _J by selecting a subset of the vehicle batteries from the large number of vehicle batteries to be selected. This takes place under the secondary condition that the sum of the costs of measuring the vehicle batteries selected in this way is less than a predefined user parameter, which indicates how high the costs may be at most.

As an alternative to weighting the target function with the weighting parameter, a limited optimization problem can also be solved by minimizing the costs by selecting a subset of the vehicle batteries from the large number of vehicle batteries to be selected. This takes place under the secondary condition that the total amount of information Info _J is greater than a specified minimum information gain about all selected vehicle batteries.

If probabilistic models are used for the state of health model, then the secondary condition can also be specified as a probability greater than a specified probability that the costs are less than the specified parameter. Alternatively, the constraint can also be specified as a probability greater than a specified probability that the entire information measure Info _J is greater than a specified minimum information measure.

Subsequently, in step S7, the selected vehicle batteries 3 are used in order to carry out a highly precise measurement of the aging status as a label. For the large number of vehicle batteries, there are 3 progressions of the aging states, as shown, for example, in 5 is shown. Training data sets are available with which the aging state model 4 can be trained in connection with the operating variable curves resulting from the load pattern and the determined aging state as a label. The training data records newly determined in this way are used in step S8 for further training of the data-based probabilistic regression model. In addition, automated hyperparameter tuning, e.g. B. via a gradient-based method or a black box method, such as a Bayesian optimization performed.

In a subsequent step S9, it is checked whether the trained aging state model has sufficient accuracy of z. B. exceeds a maximum error of 1.5% SOHC. If this is the case (alternative: yes), the method continues with step S2, otherwise (alternative: no) the measurement of the vehicle batteries 3 is ended.

In an alternative embodiment, the selected vehicle batteries can continue to be measured as the sole vehicle batteries, while the other vehicle batteries 3 that have not been selected are removed from the measurement or from the test bench, in order to free up test bench spaces. The number of vehicle batteries remaining in the measurement is thus gradually reduced, so that the overall costs of the measurement, which scale considerably with the number of vehicle batteries 3, can be significantly reduced.

Claims (12)

Method for initially providing an at least partially data-based aging state model (4) for an electrical energy store (3), with the following steps: - providing (S1) a number of energy stores (3) on a test bench (1) for measurement depending on a respective load profile, wherein the load profiles are different and characterize a time course of at least one stressful operating variable for the energy store (3); - Operation (S2) of the number of energy storage devices (3) with the respective associated load profile and detection of temporal performance variable profiles; - At a given evaluation time, determining (S4) an aging state of a subset of the energy storage devices (3) as a label and generating a training data set with the operating variable profiles and the specific label for each energy storage device (3) of the subset of the energy storage devices (3); - Selecting (S6) the subset of the energy storage devices (3) with the respectively assigned load profile depending on an optimization method that is based on the total costs (C _J ) of a measurement of the energy storage devices (3) on the test bench (1) and on a total information measure (Info _J ) the measurement of the energy storage (3) depends. procedure after claim 1 , wherein the optimization method aims to minimize the total costs (C _J ) of the measurement on the test bench (1) and to maximize an information gain determined by the information measure (Jnfo _J ) for the creation of the initial aging state model (4), the Costs, in particular of energy use or demand during the measurement on the test bench (1) and/or a period of time for the entire measurement of the energy storage device to be measured (3) and/or total test bench costs, which include at least one occupancy time of the test bench (1) and/or take into account the use of materials. procedure after claim 2 , wherein the optimization method - is carried out based on an objective function and in particular with a greedy algorithm, the objective function being a weighted sum of the total costs (C _J ) for measuring the number of energy storage devices (3), and a total information measure (Jnfo _J ) represents the energy store (3) by measuring the subset, - by maximizing an overall information measure (Info _J ) under the constraint that the total costs (C _J ) of measuring the energy stores (3) selected in this way are smaller than predetermined maximum costs; - by minimizing total costs (C _J ) under the constraint that the total information measure (Jnfo _J ) is greater than a predetermined minimum information measure; - by minimizing total cost (C _J ) under the constraint that a probability of total cost (C _J ) being less than predetermined maximum cost is greater than a predetermined probability; or - by maximizing an overall information measure (Info _J ) under the constraint that a probability that the overall information measure (Jnfo _J ) is greater than a predetermined minimum information measure is greater than a predetermined probability. Procedure according to one of Claims 1 until 3 , wherein the data-based aging state model (4) is designed with a probabilistic data-based model, wherein for one of the energy stores (3) an input vector of the data-based model, which has at least one performance variable profile (x(t)) and/or at least one performance feature (m(t )) from the at least one performance variable profile, an internal state of the energy store (3) and/or a physically modeled aging state, to the aging state to be modeled of the relevant energy store (3) or a correction variable for correcting a physically modeled aging state of the relevant energy store (3 ) is mapped, the information measure () for an energy store (3) being determined as the determinant of the predictive covariance. procedure after claim 4 , wherein the predictive covariance is determined from the input vectors of one or more of the energy stores (3) for the evaluation times of an entire measurement period. procedure after claim 5 , With a Gaussian process model as a probabilistic model, the predictive covariance is determined from the input vectors of one or more of the energy stores (3) for the evaluation times of an entire measurement period, the input vectors being determined from the load profiles. Procedure according to one of Claims 1 until 6 , wherein at each evaluation time only the energy stores (3) of the subset of the energy stores (3) are measured as a label to determine an aging state and the remaining energy stores (3) continue to be operated according to the load profile. Procedure according to one of Claims 1 until 6 At each evaluation time, only the energy stores (3) of the subset of the energy stores (3) are measured as a label to determine an aging state and continue to be operated according to the load profile, while the remaining energy stores (3) are removed from the test bench. Procedure according to one of Claims 1 until 8th , wherein the aging state model (4) is trained with the determined training data sets. Device for carrying out one of the methods according to one of Claims 1 until 9 . Computer program product, comprising instructions which, when the program is executed, cause at least one data processing device to carry out the steps of the method according to one of the Claims 1 until 9 to execute. Machine-readable storage medium, comprising instructions which, when executed by at least one data processing device, cause the latter to carry out the steps of the method according to one of Claims 1 until 9 to execute.

Images