Method and device for providing a charging profile for a device battery for a battery-operated device using active learning with field

Abstract

The invention relates to a method for determining an optimized charging profile for use in charging a device battery of the same type of a battery-operated device, comprising the following steps: associating a plurality of different test charging profiles with the device battery; operating the device battery for a predetermined observation duration, wherein the device battery is charged a plurality of times on the basis of the respectively associated test charging profiles; determining an average or aggregate charge time information and an aging state change for each device cell after the observation duration; training at least one data-based probabilistic charging profile model by means of a training data set that associates, for each device cell, the averaged / aggregated charging time information and the respective associated aging state change with the respective associated values of the electrochemical design parameters; determining an optimized charging configuration file according to the charging configuration file model; the determined optimized charging profile is provided to control the charging process of all device batteries.

Description

Technical Field

[0001] The present invention relates to device batteries, in particular to energy converters or energy storage systems of electric drive vehicles, such as vehicle battery systems, and particularly to adapting or selecting a charging profile for such device batteries. Background Art

[0002] Charging and discharging of device batteries cause degradation. The degree of degradation depends on the charging and discharging parameters. While the degradation caused by the discharging process is essentially determined by the use of the device battery which is usually not affected, the stress applied to the device battery during the automated charging process is determined by parameters which describe the manner and method of how the charging process is carried out. The parameters of the charging process thus determine the aging behavior of the device battery.

[0003] Charging of device batteries is usually carried out based on a preset charging profile. The charging profile is usually illustrated by a characteristic curve which determines the correlation between the charging current or the maximum charging current and the state of charge. The state of charge is expressed in % as the share of the stored charge relative to the maximum charge. Thus, the charging profile is determined such that it optimizes the charging time, the maximum charging current and the acceptable aging behavior, such that a compromise can be achieved between the load on the device battery and the required charging time.

[0004] Generally, before production starts, the charging profile of the device battery is determined based on a large number of laboratory measurements, wherein the charging profile is created such that the correlation between degradation and the charging current, the battery temperature and the state of charge of the device battery can be taken into account.

[0005] Since each user behavior loads the device battery differently, the aging state is determined not only by calendar aging, but also by the periodic load caused by the trend of the charging and discharging current, the state of charge and the battery temperature. However, since nowadays only charging profiles are preset for specific device battery types for normal and fast charging, the user can only restrictedly influence the impact of the charging process on the aging behavior of the device battery. Summary of the Invention

[0006] According to the present invention, there is provided a method for determining an optimized charging profile for device batteries of the same type according to claim 1 and a corresponding device according to the co-pending independent claims.

[0007] Further design options are described in the dependent claims.

[0008] According to a first aspect, there is provided a method, in particular at least partially computer-implemented method, for determining an optimized charging profile to be used when charging device batteries of the same type of a battery-operated device, the method having the following steps:

[0009] - Associate multiple different test charging profiles with the device battery, where the test charging profiles respectively specify the maximum allowable charging current related to the system state and are associated with at least one electrochemical design parameter, in particular the limit value of the anode overpotential. Among them, the test charging profiles are selected from the designed charging profiles according to a preset Pareto curve. The designed charging profiles specify the combination of the modeled average / total charging time and the modeled aging state change obtained when using the relevant test charging profiles. In this combination, the cost of each modeled aging state change is minimized according to a preset cost function.

[0010] - Run the device battery for a predetermined observation duration, where the device battery is charged multiple times based on the respectively associated test charging profiles and, if necessary, discharged during operation.

[0011] - After the observation duration, obtain the average or total charging time information and the aging state change for each device battery.

[0012] - Train at least one data-based probability charging profile model with a training data set. For each device battery, the training data set associates the corresponding average / total charging time information and the corresponding aging state change with at least one respectively associated electrochemical design parameter.

[0013] - Determine an optimized charging profile according to the charging profile model.

[0014] - Provide the determined optimized charging profile to control the charging process of all device batteries.

[0015] The aging of the device battery is significantly related to the load generated by its usage type, such as the level of charging and discharging current, high temperature, frequency of fast charging cycles, etc. Therefore, compliance with the warranty conditions depends on the usage type of the device battery, so the goal is to keep the load on the device battery as low as possible during use.

[0016] In the above method, the load on the device battery is reduced by adapting the preset charging profile. So far, the charging profile is usually determined based on the tests of the battery manufacturer, and for medium loads of the spare battery, compliance with the warranty conditions can be achieved through the medium load in addition to other operating conditions. For device batteries, one or more charging profiles are correspondingly preset and fixed, and the charging profiles are selected according to requirements, such as normal charging or fast charging. The selected charging profile is usually applied to all device batteries of the same type during the charging process.

[0017] A normal charging profile can be determined based on preset electrochemical design parameters, such as limit values of anode overpotential, and other or additional electrochemical state parameters, and the maximum charging current is described according to the system or charging state, i.e., the filling of the device battery relative to the total storage capacity, or the charging time or battery temperature. The trend of the maximum charging current or the target charging current as a function is the charging curve. Such a charging profile limits the maximum charging current and thereby avoids over-stressing the device battery by avoiding falling below at least one preset electrochemical design parameter. The anode overpotential is the difference between the actual potential of the anode and its thermodynamic equilibrium potential. It corresponds to the energy that must be used to initiate the electrochemical reaction at the anode. A charging profile can be created by means of an adjustment model based on a preset electrochemical cell model, by simulating the battery current at each preset charging state and at a preset safety value or limit value of at least one electrochemical design parameter, such as anode overpotential. See also EP4046228A1.

[0018] It is feasible to directly influence the further trend of the relevant device battery aging, for example, by avoiding critical states, by reducing the load on the device battery due to adapting the preset charging profile for the individual device battery. However, the charging time that the user has to wait to reach a preset charging state or a preset charging lift is predetermined by the preset charging profile. Thus, in a charging profile with a low maximum charging current that causes low aging, the charging duration is long, while a lower charging time can be achieved in a charging profile with a higher charging current that causes high aging. The combination of the aging state change and the charging time can be evaluated via a preset cost function.

[0019] The above method enables the selection of an optimized charging profile considering the evaluation of the charging profile of the device battery, the model points of the aging state change, and the average or total charging time information. With the help of the charging profile, a trade-off can be obtained between the reduction of the charging time and the maximization of the service life (reduction of aging caused by the charging process). For this purpose, different charging strategies / charging profiles are evaluated across devices in terms of aging impact and charging duration, and the best charging profile is selected in a model-based manner according to the charging profile model. Therefore, optimization based on real battery behavior can be performed via swarm intelligence, i.e., a large number of tested test charging profiles in field operation.

[0020] The background is that there is a Pareto-optimal relationship between the aging behavior deviated from the modeling and the modeled average / total charging time information. The actual situation / cost regarding the relationship between the aging behavior and the charging time information may deviate, for example, due to batch deviation, model limitations, and unforeseen effects in this field. Therefore, according to the charging profile of the actual device battery, a better cost value than that obtained through the ideal Pareto-optimized cost curve can be obtained for the cost of the cost function, and conversely, a worse cost value can be obtained.

[0021] It can be proposed that according to the parameterized electrochemical cell model of the device battery, in particular, multiple different test charging profiles are provided based on one or more electrochemical or physical parameters and one or more electrochemical design parameters, such as the limit value of the anode overpotential.

[0022] The charging curve of the test charging profile can be obtained by simulation based on electrochemical and / or physical variables, such as the anode overpotential, with the help of the electrochemical cell model, and is selected based on the Pareto-optimal relationship between the average / total charging time information and the change in the aging state. The Pareto-optimal relationship corresponds to a Pareto curve, which can be determined by simulation according to a preset cost function that takes into account the charging time information modeled for the corresponding charging profile and the change in the aging state modeled for the corresponding charging profile. Here, the charging curve / charging profile can be generated in a manner known per se such that the anode overpotential is not lower than the limit value of the anode overpotential unique to the battery type. The corresponding electrochemical design parameter or the limit value of the anode overpotential can be associated with the combination of the charging time information and the change in the aging state of the test charging profile respectively.

[0023] This can be done, for example, by adapting the limit value of the anode overpotential to the optimized value of the anode overpotential according to the aging state of the device battery, in particular, based on one or more electrochemical or physical parameters (such as equilibrium), or kinetic parameters or electrochemical states, such as the lithium that can be cycled, diffusion coefficient, etc., where the design can be carried out by considering the trade-off between minimizing the change in the aging state and minimizing the average / total charging time information. In this case, multiple Pareto-optimal charging profiles or the configuration of such charging profiles can be derived by presetting different limit values of the anode overpotential by means of adjustment techniques, such as simulation or adjustment to the limit value.

[0024] For example, it is known from the prior art to determine a suitable charging profile based on a battery model, for example from US 9,153,991 B2, US2017 / 033866 A1, DE 102020206272 A1, DE 102020210132A1, DE102019209165A1, DE 102020212579A1, US10,447,054B2, US 9,153,991 B2, US10,312,699B2, DE 102019216015A1 and Doyle, M., Fuller, T F and Newman, J., "Modeling of galvanostatic charge and discharge of the lithium / polymer / insertion cell", United States, https: / / doi.org / 10.1149 / 1.2221597.

[0025] According to the above method, first, different test charging profiles in the previously determined test charging profiles are associated with a plurality of device batteries. In the case of associating one or more of the test charging profiles with the device battery respectively, one test charging profile can be associated with a plurality of the device batteries. This achieves that due to the probability scheme, different usage behaviors of the device battery have a relatively small impact on the evaluation in terms of different charging profiles, where the different usage behaviors result in different aging behaviors, i.e., changes in the aging state, during the observation duration. The test charging profiles are selected from the Pareto-optimal charging profiles for various preset electrochemical design parameters, in particular the limit value of the anode overpotential.

[0026] The charging profile is selected in a Pareto-optimal manner from the correlation between the average / total charging time information and the average aging state change (per unit time), that is, selected from the set of charging profiles for which it is impossible to reduce the average / total charging time information while keeping the aging behavior the same or to deteriorate the aging state while keeping the average / total charging time information the same.

[0027] For example, 30 charging profiles can be selected discretely, where one test charging profile is associated with one device respectively.

[0028] During continuous operation, now, the charging profile is applied during the charging process of the corresponding device battery. Here, it is evaluated how the applied test charging profile affects the aging behavior of the relevant device battery within the observation duration.

[0029] It can be proposed that the change in the aging state can be determined as the difference in the aging state at the start and end of the observed observation duration. Alternatively, instead of the entire change in the aging state deltaSOH during the observation duration, it is also possible to determine only the change in the aging state accumulated during the charging process (preferably during a preset, constant number of charging processes), i.e., to determine the sum of the reduction in the capacity-related aging state SOH-C only during the charging process (e.g., by forming the difference in the aging state before and after each charging process).

[0030] Furthermore, the average / total charging time information can be determined as the average of the charging times of the charging processes in which a charging state change greater than a preset threshold is achieved. The change in the aging state can also be determined by, in a model-based manner, eliminating or subtracting from the entire change in the aging state determined during the observation duration the aging effects or changes in the aging state not caused by the charging process. The total charging time information can be determined by accumulating the historical charging times up to this point. Furthermore, it can be proposed that only a representative, filtered part, i.e., for example, the charging times between SOC_min = 20% and SOC_max = 80%, be included in the total charging time. Thus, filtering can be performed to increase comparability or representativeness / correlation.

[0031] Therefore, the change in the aging state deltaSOH of the device battery can be determined after a predetermined calendar duration after the start of the observation. This is based on the fact that all device batteries have the same SoH value, i.e., the same aging state, at the start of the duration, for example, an SoH value of 100%. Furthermore, the average / total charging time information related to the charging profile is determined. The charging time can be taken into account in each charging process or only in the charging processes in which the determined charging lift with, for example, delta SOH > 80% is exceeded.

[0032] Furthermore, the average / total charging time information can be presented herein in absolute terms or relative to the change in the charging state at a specific charging state level.

[0033] The current aging state can be simulated or determined with the aid of conventional and per se known models for determining the aging state. In particular, an electrochemical aging state model can be used, which in principle is based on an electrochemical cell model. Such an electrochemical cell model can include a system of differential equations that models the internal cell state, in particular the equilibrium state and possibly the kinetic state, by means of a time integration method based on differential equations parameterized by model parameters, and provides the relationship between the operating variables of the device battery, namely the battery current, battery voltage, battery temperature, and state of charge of the device battery, and the internal cell state. Such an electrochemical cell model is known, for example, from the printed documents US2016 / 023,566, US 2016 / 023,567, and US2020 / 150,185. The aging state can be derived from the internal cell state.

[0034] The initially parameterized aging state model and the initially parameterized battery model are used for the above determination of the aging state change and the charging time information, and these models can be continuously improved subsequently.

[0035] Then, a cross-device evaluation can be performed in a central unit remote from the device, with the aim of: evaluating the true state points of the average / total charging time information and the aging state change of the device battery in terms of the original design of the device battery by means of a test charging profile. The background is that it is considered that not all aging effects can be explained by the charging profiles used by means of simulation to create the charging profiles.

[0036] For this purpose, a charging profile model is first preset, which associates at least one corresponding electrochemical design parameter or a limit value of the anodic overpotential with the measured aging state change and the average / total charging time information. The charging profile model can be set as a multi-output probability regression model in a data-based manner, in particular as a multi-output Gaussian process model. This has the advantage that when modeling together in one model, the correlation between the output variables results in a higher accuracy of the model output.

[0037] For this purpose, a cost function can be proposed, which calculates a cost value based on the charging profile and the aging state change, such as a weighted sum of the reciprocal of the aging state change and the charging time.

[0038] For each device battery, there is now a state point composed of the aging state change and the average / total charging time information above at least one associated electrochemical design parameter (limit value of the anodic overpotential), such that the charging profile model can be modeled as the association of at least one electrochemical design parameter (limit value of the anodic overpotential) with the average / total charging time information (if necessary, the change in the state of charge per percentage) and the aging state change.

[0039] This achieves: estimating in the solution space the cost of using a specific charging profile for a specific device battery. For example, a Gaussian process model or a multi-output Gaussian process model can be used as a data-based charging profile model that maps at least one design parameter to an aging state change or a state vector. It can also be proposed that the mapping further includes charging time information that is expected in reality and can be different from the design.

[0040] Additionally, other characteristics of the device battery, such as electrochemical design characteristics, in particular known battery state variables, can be used or considered in the Gaussian process model.

[0041] With the probabilistic charging profile model, in addition to the average / total charging time information and the aging state change as model outputs, a confidence level can also be estimated, which indicates the reliability of the probabilistic charging profile model at the evaluation point.

[0042] Furthermore, it can be proposed that an optimized charging profile is determined by finding the value of the electrochemical design parameter or the limit value of the anode overpotential for the modeled state point found to be the cost minimum. In particular, it can be determined for which value of the at least one electrochemical design parameter or for which limit value of the anode overpotential the cost minimum is obtained. From this, the optimized charging profile applied to part or all of the device battery can be determined.

[0043] Thus, with the aid of a preset charging profile model, an optimized charging profile can be determined. The charging profile model together with the cost function shows that the higher the aging behavior or the greater the aging state change and the greater the average / total charging time information, the higher the evaluated cost is for the user. A higher decrease in the aging state means a greater decrease in the residual value of the vehicle, and a longer charging time means that the user has to wait longer during the charging process and thus loses comfort. Therefore, for each device, with the aid of the charging profile model, the actual cost can be indirectly evaluated and provided based on the cost function related to the aging state change and the charging time information.

[0044] In addition to the quantification of uncertainty and the resulting quantile modeling, the advantage of probabilistic modeling is that it is also possible to evaluate regions of the charging profile model that have not been evaluated or observed so far. The cost assessment of the cost function including the confidence level assessment can be uniquely associated with each possible charging profile.

[0045] It can be proposed that within the scope of active learning, one or more values are selected for at least one electrochemical design parameter, for which values the charging profile model illustrates the maximum uncertainty estimated by the model, where one or more additional test charging profiles are determined for one or more values of at least one electrochemical design parameter, where one or more device batteries are selected and one or more additional test charging profiles are associated with the device batteries, where one or more of the selected device batteries are measured during continuous operation or on a test bench in order to further train the aging state model and / or reparameterize the electrochemical cell model.

[0046] In other words: The probabilistic charging profile model can be analyzed and it can be determined for which values of at least one electrochemical design parameter or for which limiting values of the anode overpotential the information / uncertainty of the charging profile model is particularly high. In particular, within the scope of the active learning method, values of at least one electrochemical design parameter or limiting values of the anode overpotential can be determined for which a particularly low confidence is obtained from the evaluation of the Gaussian process model or a particularly high information gain is obtained when measuring the corresponding state points.

[0047] Then, the resulting values of at least one electrochemical design parameter can be used to generate one or more additional test charging profiles and associated with a part of the device batteries for further measurement during continuous operation. Then, the operating variable trends obtained for the device batteries can be used to improve the underlying aging state model and electrochemical cell model.

[0048] For this purpose, accelerated aging tests and performance tests can be performed on a test bench with the relevant device batteries according to the active learning method described above, for which device batteries an excessive uncertainty is obtained in the charging profile model, in order to improve the aging state model for determining the aging state and the electrochemical cell model in a known manner by training or parameterization by means of measurement (measuring the operating points of battery voltage, battery current, battery temperature and state of charge and determining the aging state, for example by Coulomb counting) in static and dynamic operating conditions. In particular, the electrochemical cell model can be updated or reparameterized by fitting. Then, the above method for selecting the optimized charging profile can be re-executed based on the historical operating variable trends existing for each device battery.

[0049] For additional test charging profiles determined by an active learning method, comfort and safety conditions can be queried. Thus, the additional test charging profile is only implemented and applied in one or more devices if the comfort and safety conditions are met. The comfort and safety conditions can evaluate the charging profile or an optimized limit value of the anode overpotential according to a comfort function. The purpose of the comfort function is that in the case of an unfavorable additional test charging profile, the user is not annoyed or dissatisfied due to too long a charging time or excessive battery aging or the formation of other drawbacks. The comfort conditions can use the comfort function or a data-based comfort model in order to obtain a comfort value based on aging state changes and charging time information.

[0050] Thus, the comfort function can be defined such that the probability that the distance of the aging state change or the charging time information calculated with the aid of the charging profile model or the cost value obtained with the aid of a cost function from the corresponding point on the modeled Pareto curve is greater than a preset distance. As long as the probability is less than a threshold probability preset by the comfort and safety conditions, the additional test charging profile is accepted.

[0051] Alternatively, the comfort function can be a data-based probability model (e.g., a single-output Gaussian process) that calculates the distance between the measured charging time and the Pareto-optimal charging time of the aging state change resulting from the same measurement. The model is in turn suitable for calculating the probability that the distance is greater than a preset distance difference. If the probability is less than a preset threshold probability, the charging profile is accepted.

[0052] The additional test charging profile is only implemented in the vehicle if the comfort value of the additional test charging profile or the corresponding electrochemical design parameter (limit value of the anode overpotential) exceeds a preset minimum value of the comfort value (e.g., compliance with a performance commitment). Otherwise, the value of the electrochemical design parameter (limit value of the anode overpotential) is sought according to a charging profile model that has the greatest possible uncertainty while meeting the comfort conditions.

[0053] In addition to the above, the comfort and safety conditions can specifically take into account physically meaningful or critical limits of battery operation, such as the state of charge (e.g., SOC at the end of charging), temperature, current, and voltage, as well as cumulative or integral variables, such as the AH throughput per unit time. Furthermore, the safety and comfort conditions can also include the environmental conditions of the battery or battery cell, such as pressure (force / area) or humidity.

[0054] The additional test charging profile can be transmitted to the device such that the test charging profile can subsequently be used to perform the charging process. Detection of the trend of the operating variables then enables reparameterization or retraining of the aging state model and the electrochemical cell model.

[0055] It can be proposed to perform, in a central unit remote from the device, obtaining the average / total charging time information and the aging state change during the observation duration, obtaining the cost value, selecting one of the test charging profiles, and providing the selected test charging profile or the newly created charging profile.

[0056] By retraining the aging state model and the electrochemical cell model for determining the aging state to obtain the charging profile according to the device battery as follows, the charging profile model can be continuously improved at a preset or regular time interval, where the device battery operates with the additional test charging profile and there is a high uncertainty in the model evaluation for the device battery in the charging profile model.

[0057] Based on the aging state model, according to the previously detected operating variable trend of the device battery with the assigned test charging profile, the aging state change and the average / total charging time information of each device battery, as well as the associated test charging profile / value of the electrochemical design parameter or the limit value of the anode overpotential, can be recalculated, and an updated probabilistic charging profile model can be created, which associates the corresponding aging state change and the corresponding average / total charging time information with the limit value of the anode overpotential. Description of the Drawings

[0058] The embodiments will be explained in more detail below with reference to the accompanying drawings. Among them:

[0059] Figure 1 A schematic diagram of a system is shown, which is used to provide the operating variables of the driver and the vehicle individual to individually observe the effects of different charging profiles of the vehicle battery in a central unit outside the vehicle;

[0060] Figure 2 A view showing an exemplary charging profile for performing the vehicle battery charging process is shown;

[0061] Figure 3 A flowchart showing a method for determining the optimal charging profile of a vehicle in a fleet is shown;

[0062] Figure 4 A view showing the trend of the correlation between the average charging time and the aging state change of different charging profiles is shown;

[0063] Figure 5 A view showing the measurement of the charging time and the corresponding aging state change at the vehicle battery, where different charging profiles are preset for the vehicle battery; and

[0064] Figure 6A view showing a modeled cost function that serves as a basis for selecting an optimized charging profile. Detailed Description

[0065] Below, a method according to the present invention is described in terms of a vehicle battery, which is used as an equipment battery in a large number of motor vehicles that are the same type of equipment. The method can be partially run in a central unit and is used to select a suitable charging profile for the vehicle battery.

[0066] The examples described herein represent a large number of fixed or mobile devices with network-independent energy supplies, such as vehicles (electric vehicles, electric bicycles, etc.), facilities, machine tools, household appliances, Internet of Things devices, etc., which are connected to a central unit (cloud) outside the device via corresponding communication connections (such as LAN, Internet).

[0067] Figure 1 A system 1 is shown for collecting fleet data in a central unit 2 to select a favorable charging profile. Figure 1 A fleet 3 with a plurality of motor vehicles 4 is shown.

[0068] In Figure 1 One of the motor vehicles 4 is shown in more detail. Each motor vehicle 4 has a vehicle battery 41 as a rechargeable electrical energy storage, an electric drive motor 42, and a control unit 43. The control unit 43 is connected to a communication module 44, which is suitable for transmitting data between the corresponding motor vehicle 4 and the central unit 2 (the so-called cloud).

[0069] The motor vehicle 4 monitors the operation of the vehicle battery 41 by means of the control unit 43. Here, information on the change in the aging state and the average / total charging time of the charging process is obtained with the help of the respectively associated charging profile, and the information on the change in the aging state and the average / total charging time is transmitted to the central unit 2 at a preset time point.

[0070] The motor vehicle 4 also sends an operating variable F to the central unit 2, which at least describes a variable that affects the aging state of the vehicle battery 41. In the case of a vehicle battery, the operating variable F can be a time series of battery current, battery voltage, battery temperature, and state of charge (SOC) at the group, module, and / or unit level. The operating variable F is detected at a fast time grid of 1 Hz to 100 Hz and can be transmitted to the central unit 2 regularly in an uncompressed and / or compressed form.

[0071] The central unit 2 has a data processing unit 21 and a database 22, in which the method described below can be executed, and the database is used to store data points, model parameters, states, etc.

[0072] The algorithm can be implemented in the central unit 2, which selects one or more charging profiles suitable for use by the vehicle 4 from a set of test charging profiles. The algorithm can be executed regularly to select an appropriate charging profile.

[0073] The charging profiles for the vehicle battery 41 preset a maximum charging current according to the state of charge. The normal charging profile has a maximum charging current that decreases as the state of charge increases, for example as Figure 2 shown, and the maximum charging current is described according to the state of charge, i.e., the available charge stored in the vehicle battery 41 or with respect to the charging time. The charging profile is preset and determined such that the anodic overpotential is not lower than a preset limit value.

[0074] Figure 3 The method for determining the charging profile to be associated with the vehicle battery 41 is illustrated according to a flow chart. This method can be executed in conjunction with the central unit 2.

[0075] The method for determining the optimized charging profiles for a large number of vehicle batteries 41 first proposes in step S1: parameterizing an electrochemical cell model based on the received operating variable trend F of the vehicle battery 41, particularly by fitting. Such an electrochemical cell model can include a system of differential equations, which are based on differential equations parameterized by model parameters, and model the internal battery state (particularly the equilibrium state and possible kinetic states) by means of a time integration method, and provide the relationship between the operating variable trend of the device battery, i.e., the battery current, battery voltage, battery temperature, and state of charge of the device battery, and the internal battery state. Such an electrochemical cell model is known, for example, from the printed literature US2016 / 023,566, US2016 / 023,567, and US2020 / 150,185.

[0076] Then, in step S2, the electrochemical cell model realizes: based on electrochemical and / or physical parameters and one or more design parameters, such as the limit value of the anodic overpotential, in the trade-off between the influence of aging and the required average / total charging time information, deriving one or more test charging profiles as Pareto optimization state points (Pareto state points) from the average / total charging time information and the aging state change, and providing them as a family of charging curves or charging profiles. Finding this family of curves from the possible test charging profiles can be performed in a manner known per se, taking into account the anodic overpotential derived from the battery chemistry in the central unit 2 and the preset design parameters, under the preset average / total charging time information. The Pareto curve is obtained by minimizing a cost function, which provides cost values based on the aging state change and the average / total charging time information.

[0077] In practice, the average / total charging time information can typically correspond to the average of the charging durations until a preset minimum charging lift, such as deltaSOC = 80%, respectively, and can be normalized for a sufficiently high charging state lift and described, for example, with respect to full charge equivalent.

[0078] Now, the change in the aging state caused by applying the corresponding charging profile to the vehicle battery 41 during a specified observation duration can be determined by simulation. A correlation is obtained between the average / total charging time information (or the average of the charging times for each change in the charging state) and the change in the aging state after a predetermined simulation run duration (observation duration), for example, between 1 and 3 years.

[0079] The change in the aging state can be determined as the difference between the aging state at the start of simulating the vehicle battery 41 with the test charging profile and the aging state at the end of simulating the vehicle battery 41 with the test charging profile. Alternatively, the change in the aging state can be determined as the sum of the differences between the aging state at the start and the end of the charging process during the predetermined observation duration, with the corresponding test charging profile. In an ideal case, for a specific charging profile, a correlation is obtained between the average / total charging time information T and the aging state SOH or the change in the aging state, as shown, for example, in Figure 4 as shown. The Pareto curve shown can be determined by Pareto optimization. The different effects of the charging profile on the aging of the vehicle battery 41 and the resulting charging times are identified.

[0080] In a subsequent step S3, a plurality of test charging profiles are selected from the Pareto curve and associated with the electric vehicles 4 of the fleet 3 as operating strategies for charging, respectively. This association can be carried out by uploading the charging profile to the control unit 43 of the relevant vehicle 4 and applying it there during each charging process. For example, 30 test charging profiles can be selected, which are used in 70 vehicles, respectively, to obtain a sufficient database. A test charging profile for normal charging and a test charging profile for fast charging can be provided for each vehicle 4.

[0081] Each vehicle 4 or vehicle battery 41 is operated on-site in step S4 and monitored during an observation duration, for example, between 6 months and 3 years, and it is determined how the application of the charging profile affects the aging behavior of the corresponding vehicle battery 41. To this end, the trends of the operating variables are continuously recorded, and the aging state is determined by means of an aging state model based on the evaluation of the time series of the operating variables. The effect is evaluated by the change in the aging state during the observation duration and by the associated average / total charging time information caused by the test charging profile associated with the relevant vehicle battery 41. To this end, the change in the aging state (overall or only during the charging process) is determined starting from the aging state at the time of commissioning, such as SoH = 100%, during the preset observation duration. The average / total charging time information and the change in the aging state are transmitted to the central unit 2 after the observation duration.

[0082] Based on different charging profiles associated with the vehicle, the trend of the average / total charging time information T of the aging state SOH after the observation duration is obtained, for example, as shown in Figure 5 . It is recognized that there are systematic regions or patterns in which there is an over-performance or under-performance of the relationship between the average / total charging time information and the aging state SOH after the observation duration with respect to the theoretical optimum value of the upper Pareto curve P.

[0083] In Figure 5 , each individual real state point F in the figure represents the average / total charging time information and the charging state change of a specific vehicle battery 41 at the current time point or at the time point when the observation duration ends.

[0084] By means of a preset cost function, the efficiency of the test charging profile associated with each vehicle can be evaluated. The cost function describes the cost for using the relevant test charging profile according to the aging state or (the actual change in the aging state and the actual average / total charging time information T), in particular, the greater the change in the aging state during the observation duration and the greater the average / total charging time information during the observation duration, the greater the cost value of the test charging profile of the relevant vehicle battery 41. For example, such a cost function can be represented by the figure in Figure 6 . Therefore, the cost function takes into account the user cost due to long charging or waiting at the charging station and the cost of the residual value loss of the vehicle battery 41 due to the increase in the change in the aging state.

[0085] Now, in step S5, it is possible to model the average / total charging time information and the corresponding aging state change for all vehicle batteries with the aid of a data-based probabilistic charging profile model, which can be configured as a probabilistic regression model, in particular as a multi-output Gaussian process model. To this end, the limit value of the anode overpotential (and thus the resulting charging profile) is associated with the aging state change dSOH and the average / total charging time information T as labels in order to obtain a training data set and a validation data set, where the test charging curve is modeled with the aid of the limit value. The training data set and the validation data set are now used to: train the probabilistic charging profile model. This enables a global representation in order to obtain a true continuous association between the aging state change dSOH and the average / total charging time information and the charging profile.

[0086] By evaluating the trained multi-output Gaussian process model in combination with a cost function, it is now possible in step S6 to: determine the uncertainty for the course of the charging profile model. The confidence can indicate: how informative the previously selected charging profile from the family of curves of the test charging profile is with respect to the underlying cost function.

[0087] By evaluating the cost values of the modeled state points for the course of the charging profile model, based on the probabilistic charging profile model, it is thus possible to determine the modeled state point with the lowest cost, where the course of the probabilistic charging profile model can be illustrated by the modeled state points of curve L for various limit values of the anode overpotential. To this end, the limit value of the anode overpotential can be changed, and the state points resulting from the charging profile model and their cost values can be determined.

[0088] According to Figure 5 the view, this is feasible, based on which the average / total charging time information and the aging state change can be associated with each charging profile. Thus, it is possible to clearly associate the cost with each charging profile (illustrated by the limit value of the anode overpotential) with the aid of the charging profile model, including a measure of the uncertainty of the corresponding cost estimate. The measure of uncertainty is obtained in a manner known per se by evaluating the charging profile model modeled as a Gaussian process model. Thereby, it is possible to determine the limit value of the anode overpotential that corresponds to the state point composed of the aging state change and the average / total charging time information. The selection is made according to the minimum cost value, which can be associated with the corresponding limit value of the anode overpotential.

[0089] This selection can be made based on the average cost or / and supplemented by determining a preset quantile of the cost values. For example, thus, the 1% quantile or the 50% quantile can be used according to the risk preference.

[0090] In step S7, an optimized charging profile can thus be determined, and the electrochemical cell model can be parameterized or simulated in terms of regulation with an optimized limit value of the anode overpotential associated with the optimized charging profile. By solving the regulation technical problem in a manner that takes into account specific limit values, models of regulation paths, and other variables (such as temperature), the optimized charging profile can be derived from the limit value of the anode overpotential in the above manner. The creation of the charging profile or charging curve or charging map can in particular be carried out in a model-predicted manner.

[0091] In addition to the anode overpotential, other electrochemical parameters or states can be used to derive the optimized charging profile, and the other electrochemical parameters or states can be associated with an operating strategy, such as temperature or derating limit.

[0092] In the subsequent step S8, the thus optimized charging profile is transmitted to the vehicles of fleet 3 such that the vehicles can subsequently be used to perform the charging process.

[0093] In step S9, a check can be performed after a preset duration. To this end, the aging state model for determining the aging state and the electrochemical cell model can be continuously improved at preset or regular time intervals by retraining the aging state model and the electrochemical cell model based on the operating variable data of the vehicle battery, where the vehicle battery is operated with a selected additional test charging profile.

[0094] By identifying one or more modeled state points with the greatest uncertainty in the charge profile model, additional test charge profiles for further training are identified. Thus, at least one modeled state point can be determined, the model output uncertainty of which is maximized according to the underlying probabilistic charge profile model. The corresponding limit value of the anode overpotential can be associated with the state point thus determined, in particular by means of modeling with the aid of an aging state model and an electrochemical cell model. The limit value of the anode overpotential can then be associated with the charge profile in the above-described manner, the charge profile being provided as an additional test charge profile. This additional test charge profile can be implemented in one or more vehicles for further measurements to take into account future charging processes. This is an active learning process, which is implemented in a model-based manner: the vehicle battery is measured with the additional test charge profile. In particular, the vehicle battery with the additional test charge profile can also be additionally fed to precise measurements in a laboratory, on a test bench or in a workshop in order to perform accelerated aging tests and performance tests on the test bench and, by means of the measurements (measurement of the operating points of the battery voltage, battery current, battery temperature and state of charge in static and dynamic operating conditions or determination of the aging state, for example by means of coulomb counting), improve the aging state model and the electrochemical cell model for determining the aging state in a manner known per se by training or parameterization. In particular, the electrochemical cell model can be updated or reparameterized by fitting.

[0095] The additional measurements are intended to more precisely characterize the underlying aging state model and electrochemical cell model. By reusing the above method, a new charge profile model can be created to determine updated and optimized charge profiles for implementation in vehicles.

[0096] For the additional test charge profiles determined by the active learning method, comfort and safety conditions can be queried. Thus, the additional test charge profiles are only implemented and applied in one or more vehicles if the comfort and safety conditions are met. The comfort and safety conditions can evaluate the charge profile or the optimized limit value of the anode overpotential according to a comfort function. The purpose of the comfort function is that the user is not annoyed or dissatisfied in the case where the additional test charge profile is unfavorable due to too long a charging time or excessive battery aging or other drawbacks. The comfort conditions can use the comfort function or a data-based comfort model in order to obtain a comfort value based on aging state changes and charging time information.

[0097] The comfort function can be defined such that it calculates, with the aid of the charging profile model, the probability that the distance of the change in the aging state or the charging time information or the cost value obtained with the aid of the cost function from the corresponding point on the Pareto curve is greater than a preset distance. As long as the probability is less than the threshold probability preset by the comfort and safety conditions, the additional test charging profile is accepted.

[0098] Alternatively, the comfort function can be a data-based probability model (e.g., a single-output Gaussian process) that calculates the distance between the measured charging time and the Pareto-optimal charging time of the change in the aging state resulting from the same measurement. The model is in turn suitable for calculating the probability that the distance is greater than a preset distance difference. If the probability is less than the preset threshold probability, the additional charging profile is accepted.

[0099] Thus, the comfort and safety conditions can use a data-based comfort model to obtain a comfort value. The optimized charging profile is only implemented in the vehicle if the comfort value of the charging profile or the corresponding limit value of the anode overpotential exceeds a preset minimum value of the comfort value (e.g., compliance with the performance commitment). Otherwise, the limit value of the anode overpotential is sought according to the charging profile model that has the smallest possible cost value according to the cost function while satisfying the comfort and safety conditions.

[0100] Here, the comfort and safety conditions specifically take into account physically meaningful or critical limitations of the battery operation, such as the state of charge (e.g., the SOC at the end of charging), temperature, current, and voltage, as well as cumulative or integrated variables, such as the AH throughput per unit time.

[0101] Subsequently, in step S10, the aging state model and the electrochemical cell model can be trained with the aid of the identified trends of the operating variables of the vehicle battery.

[0102] Based on the aging state model, in step S11, the aging state or the change in the aging state of each vehicle battery and the associated test charging profile are re-determined according to the previously detected trends of the operating variables of the vehicle battery with the test charging profile, and an updated probability charging profile model is created in the above manner, which includes the average / total charging time information and the corresponding aging state or change in the aging state.

[0103] The method can be executed cyclically after a preset time period.

[0104] Therefore, a plausibility check can be performed in the "backtest", i.e., the Pareto-optimized charging profile is simulated again with the help of the improved aging state model and the electrochemical cell model. In addition, with the help of such adapted models, the optimized charging profile can be determined according to the method described above. This can be used to determine and predict the aging state of all vehicle batteries. Thereby, a comparison between reality and simulation can be realized to verify the new model.

Claims

1. A method, in particular an at least partially computer-implemented method, for determining an optimized charging profile to be used when charging device batteries (41) of the same type of a battery-operated device (4), comprising the following steps: - associating a plurality of different test charge profiles with the device battery (41), the test charge profiles each indicating a maximum permissible charge current that is dependent on the system state and being associated with at least one electrochemical design parameter, in particular a limit value of the anode overpotential, wherein the test charge profiles are selected from designed charge profiles according to a predefined Pareto curve, the designed charge profiles indicating a combination of modeled average / aggregate charge times and modeled changes in the state of aging that result when using the associated test charge profiles, in which combination the cost of each modeled change in the state of aging is minimized according to a predefined cost function, - operating (S4) the device battery (41) for a predetermined observation duration, wherein the device battery (41) is respectively charged a plurality of times based on the respectively associated test charging profile; - after the observation duration, determining (S4) average or aggregate charging time information and aging state change for each of the device batteries (41); - training (S5) at least one data-based probabilistic charging profile model with the aid of a training data set, said training data set associating, for each device battery (41), average / aggregate charging time information and respectively associated aging state changes with respectively associated values ​​of said electrochemical design parameters; - determining (S7) an optimized charging profile according to the charging profile model; - Providing (S8) the determined optimized charging profile to control the charging process of all device batteries (41).

2. A method according to claim 1, wherein a plurality of different test charging profiles are provided by means of an electrochemical battery model preset for the same type of device batteries (41) of the device battery (41) in the following manner: different values ​​of the at least one electrochemical design parameter are taken and charging profiles are determined respectively, and the charging profiles determine the charging current with respect to the charging state by simulation so that the anode overpotential always corresponds to the value of the at least one electrochemical design parameter.

3. The method according to claim 1 or 2, wherein when associating one or more of the test charging profiles with the device batteries (41) respectively, a test charging profile is associated with multiple device batteries in the device batteries (41).

4. The method according to claim 1 , wherein the change in the aging state is determined as the difference in the aging state at the beginning and at the end of the observation period, or wherein the change in the aging state is determined as the sum of the differences in the aging state at the beginning and at the end of the respective charging process. 5 . The method according to claim 1 , wherein the average charging time information is determined as an average of charging times of charging processes in which a charge state change greater than a predefined threshold value is achieved.

6. The method according to any one of claims 1 to 5, wherein the determination of the optimized charging profile is performed according to the charging profile model, having the following steps: - determining a modeled state point of aging state variation and average charging time based on a probabilistic charging profile model, said state point having a minimum cost value determined according to said cost function, - determining a charging profile from the values ​​of the electrochemical design parameters associated with the modeled state points, the charging profile having a minimum cost value with respect to a pre-set cost function or having a maximum negative difference in cost from a corresponding Pareto optimal point.

7. A method according to any one of claims 1 to 6, wherein one or more values ​​of at least one electrochemical design parameter are selected (S9) within the scope of active learning, for which the charging profile model describes the maximum uncertainty of the model estimate, wherein additional test charging profiles are determined for one or more values ​​of the at least one electrochemical design parameter, wherein one or more device batteries (41) are selected, and one or more additional test charging profiles are associated with the device batteries, wherein the selected one or more device batteries (41) are measured in continuous operation or on a test bench in order to further train the aging state model and / or reparameterize the electrochemical battery model (S10).

8. The method according to claim 7, wherein determining the one or more test charging profiles also takes into account comfort and safety conditions, wherein the test charging profile is associated with the device battery (41) and implemented within the scope of the active learning only when the comfort and safety conditions are met, wherein the comfort and safety conditions take into account a comfort function, wherein the comfort function describes: the probability that the deviation of the cost value from the cost of the Pareto curve is greater than a preset threshold under the corresponding aging state change or the corresponding charging time information, wherein the comfort and safety conditions define a threshold probability, so that the comfort and safety conditions are met only when the probability according to the comfort function is lower than the threshold probability.

9. The method according to any one of claims 1 to 8, wherein the data-based charging profile model is preset as a multi-output Gaussian process model for outputting the aging state change and the charging time information, wherein in particular the multi-output Gaussian process model also models at least two or more internal battery states, in particular the amount of cyclable lithium, as additional outputs.

10. An apparatus for performing the method according to any one of claims 1 to 9.

11. A computer program product comprising instructions which, when the program is executed by at least one data processing device, cause the data processing device to perform the steps of the method according to any one of claims 1 to 9.

12. A machine-readable storage medium comprising instructions which, when executed by at least one data processing device, cause the data processing device to perform the steps of the method according to any one of claims 1 to 9.

Images