DE102021125478A1 - DETERMINATION OF AN AGING VALUE FOR BATTERIES USING CURRENT-VOLTAGE TIME SERIES IN TIME DOMAIN AND LOAD DOMAIN - Google Patents

Abstract

Various examples of the disclosure relate to techniques for determining a aging value of a battery, e.g., capacity. Time series are taken into account that include values for operating load parameters of the battery at points in time that are defined in a time domain and a load domain.

Description

TECHNICAL AREA

Various examples of the disclosure relate to the characterization of batteries. Various examples relate to determining a aging value of a rechargeable battery. Various examples relate to the pre-processing of measured operational observables of a battery, such as measured current values or measured voltage values, in order to obtain an appropriate input to the machine-learned algorithm.

BACKGROUND

Rechargeable batteries are used in a wide variety of application scenarios. Examples include traction batteries for electric vehicles, static energy storage, etc.

In a wide variety of application scenarios, it can be helpful in connection with monitoring battery operation to obtain the most accurate possible characterization of the state of health of rechargeable batteries. The aging value of a battery quantifies its state of health.

Various techniques are known for determining an aging value of batteries. For example, some techniques are based on classical modeling of battery behavior. For example, physico-chemical models of the battery can be used to determine battery aging as a function of load factors. Empirical models can also be used. An example is e.g. Schmalstieg, Johannes, et al. "A holistic aging model for Li (NiMnCo) 02 based 18650 lithium-ion batteries." Journal of Power Sources 257 (2014): 325-334. Typically, with such a classic modeling, a significant effort is required for the parameterization of the corresponding models. For example, complex laboratory tests may be necessary.

Machine-learned algorithms are also known to determine an aging value of the battery. See eg CN111832220A - or see Ungurean, Lucian; Micea, Mihai V. Online state of health prediction method for lithium-ion batteries, based on gated recurrent unit neural networks International Journal of Energy Research 44(8) 6767-6777 - or see Xiang, Ming, et al. "State-of-health prognosis for lithium-ion batteries considering the limitations in measurements via maximal information entropy and collective sparse variational gaussian process." IEEE Access 8 (2020): 188199-188217 - or see Fan, Yaxiang, et al. "A novel deep learning framework for state of health estimation of lithium-ion battery." Journal of Energy Storage 32 (2020): 101741. It has been observed that such techniques are sometimes imprecise or require particularly computationally intensive implementation

SHORT SUMMARY

Therefore, there is a need for improved techniques for determining battery aging values.

This object is solved by the features of the independent patent claims. The features of the dependent claims define embodiments.

A computer-implemented method for determining an aging value of a battery includes obtaining measured current values of an operating current of the battery and measured voltage values of an operating voltage of the battery. The current values indicate the operating current during an observation period. Likewise, the voltage values indicate the operating voltage during the same observation period.

A first time series of first current values and first voltage values is determined based on the measured current values and the measured voltage values. In this case, the first time series includes the first points in time of the first current values and the first voltage values in the observation period. These first points in time are arranged according to a first sampling scheme defined in the time domain.

The sampling scheme can define an arrangement of the first points in time in the time domain.

In general, a sampling scheme can define the arrangement of corresponding points in time in the corresponding domain. For example, the sampling scheme could define a relative spacing of the corresponding time points in the corresponding domain from each other. For example, it would be possible for the sampling scheme to define the exact positions of the corresponding time points of the corresponding domain. For example, it would be conceivable that the sampling scheme defines a tolerated variance of the corresponding points in time in relation to reference points in time. The sampling scheme could determine a number of data points of a time series (ie the sampling points). For example, the sampling scheme could determine the sampling points depending on a length of a data structure used as input to an ML algorithm.

For example, it would be conceivable for the first sampling scheme to define an equidistant arrangement of the first points in time in the time domain.

An equidistant arrangement in the time domain means that the first points in time all have the same time distance from one another, for example each 5 minutes apart, to give just one example.

A second time series of second current values and second voltage values is also determined based on the measured current values and the measured voltage values. The second time series is different from the first time series.

The second time series includes second points in time of the second current values and the second voltage values in the observation period. The second points in time are arranged according to a second sampling scheme in a domain associated with an operating load parameter of the battery (load domain). It would again be conceivable for the second points in time to be arranged equidistantly in the load domain.

The operating load parameter can therefore denote a load size of the battery during active operation, that is to say in particular it can be different from the mere passage of time. Examples of operating load parameters are the charge throughput or the power throughput of operating the battery.

Based on the first time series and the second time series, it is then possible to determine the aging value of the battery. This can be done using a machine learned algorithm (ML algorithm).

Several effects can be achieved by using the first time series and the second time series: the first time series can depict characteristic features in the operating current and in the operating voltage, which essentially describe a calendar aging of the battery. The second time series can depict characteristic features in the operating current and in the operating voltage, which essentially describe stress aging of the battery due to charging or discharging processes. This is achieved by using the first and second sampling schemes defined in the time domain and the load domain. The aging value can be determined particularly reliably by the combined use of both time series. At the same time, complex pre-processing of the measured current values or the measured voltage values—for example filtering or recognition of specific signatures and extraction of local measurement points—can be omitted. This simplifies the preprocessing. By using the first time series of the second time series, the data basis can also be compressed before input into the ML algorithm. For example, it may be possible to sample the measured current values comparatively sparsely over the measured voltage values in order to obtain the first time series and the second time series. This means that a large number of measured current values and measured voltage values can be present for each current value or voltage value of the first time series and the second time series.

The aging value can be determined, for example, by a capacity of the battery and/or by an impedance of the battery.

Using the techniques described herein, the aging value can in particular indicate an absolute aging since the battery was put into service. This means that not only the relative aging can be determined during the observation period, but rather the absolute aging value of the battery. For example, the absolute capacity of the battery could be determined (rather than just the relative to determine the decrease in capacity during the observation period). The absolute impedance could be determined.

Further pre-processing of the measured current values or measured voltage values (in addition to determining the first and second time series) is not absolutely necessary. In particular, the first time series and the second time series can be used directly as input to the ML algorithm. It would be possible to concatenate (i.e. "join" together in a matrix structure) the first time series and the second time series to obtain such a data structure, which then serves as input to the machine-learned algorithm. This means that a particularly compact and efficient pre-processing can take place.

The data structure can, for example, include time stamps for the first points in time in the first time series and time stamps for the second points in time in the second time series. In this way, ML algorithms that work in the time domain can also achieve reliable results for the second time series—which is not arranged equidistantly in the time domain.

Examples of ML algorithms would be a recurrent neural network, such as a long-short-term memory (LSTM) network or a gated recurrent unit (GRU) network. See for example: Chung, Junyoung, et al. "Empirical evaluation of gated recurrent neural networks on sequence modeling." arXiv preprint arXiv:1412.3555 (2014).

The first time series and the second time series can be determined, for example, using an interpolation operation applied to the measured current values and the voltage values, respectively. For example, linear interpolation can be used.

It is not necessary in all variants that an interpolation operation is used. For example, an interpolation operation might or might not be used depending on the sampling scheme; If no interpolation operation is used, such measured current values and measured voltage values could be selected for the first and second time series that are as close as possible to a reference time defined by the respective sampling scheme. As an example: the sampling scheme could define an equidistant arrangement of the points in time in the respective domain, for example the time domain (for the first sampling scheme); but at the same time define a tolerated variance that is greater than or of the same order of magnitude as the distance between measured operating observables (e.g. operating current and operating voltage) of the battery. Then it can be superfluous to use an interpolation operation because the measurement data sample the observation period sufficiently densely - the closest values of the respective operating observables can be taken over. Even if the arrangement of the points in time is then not perfectly equidistant from one another, the variance in the distances between the points in time is within a tolerance defined by the respective sampling scheme.

Nevertheless, an interpolation operation can be comparatively less computationally intensive, so that the pre-processing of the measured current values and measured voltage values can be implemented particularly efficiently even when using the interpolation operation. In particular, for example, it would be possible to implement the interpolation operation on data processing equipment local to the battery. To put it more generally, it would be conceivable for the first time series and the second time series to be determined on such a data processing system that is arranged locally at the battery. The aging value can then be determined on a further data processing system that is not arranged locally at the battery. This means that the ML algorithm could be run on a central server, for example.

The interpolation operation can take into account a length of the data structure as a boundary condition. This means that for observation periods of different lengths, a distance between the points in time can be varied so that the entire observation period can be covered while the length of the data structure is maintained at the same time.

In this way, in particular, the data can be compressed, ie the data structure has a specific size, regardless of the number of measured current values or voltage values. For example, the interpolation operation could result in a reduction of the first current value and the first voltage value with respect to the measured current value and the measured voltage value by a factor of 1000 or more. The same also applies to the second current value and second voltage values. This makes it possible to use comparatively long observation periods. For example it would be conceivable that an observation period includes several charging cycles and several discharging cycles and several idle cycles of the battery. For example, the observation period could be longer than 24 hours.

At the same time, the observation period can also be significantly shorter than the previous battery life. This means that the observation period can be less than approx. 5% of a period since the battery was put into service. Nevertheless, using the techniques described herein, it may be possible to determine the absolute aging value of the battery. This is because the signatures of the operational observables during the battery's respective observation period also depend on the absolute aging value (not only on the relative aging). Through the appropriate choice of the domain in which the values of the operational observables are represented - especially the stress domain - corresponding signatures can be efficiently extracted from the measurement data set or preserved.

In order to enable the simplest possible monitoring for characterizing the battery, it would be conceivable, for example, for a beginning and an end of the observation period to be determined independently of the operation of the battery. This means that it may not be necessary to align the observation period specifically in relation to, for example, a charging process or a discharging process of the battery. This is because by choosing the appropriately long observation period - which is practically made possible by the compression associated with the time series, as described above - a significant signature is statistically always mapped to the observation period.

In other examples, however, it would also be possible to monitor the operation of the battery and to determine the start and end of the observation period depending on the monitoring of the operation of the battery.

For example, it would be conceivable that the observation period is always started after the completion of a charging or discharging process. For example, it would be conceivable for the observation period to end during a rest phase. For example, it would be conceivable to count the number of charging and/or discharging processes and to end the observation period after a certain number. These are just examples of how to implement the dependency between monitoring and the choice of observation period.

Techniques were described above in which measured current values and measured voltage values influence corresponding time series. As a general rule, other operational observables of the battery may alternatively or additionally be considered. Examples include, for example, the temperature of the humidity.

For example, it would be possible for measured temperature values of an operating temperature of the battery to be obtained during the observation period. The first time series can then also include first temperature values at the first points in time and the second time series can also include second temperature values at the second points in time, the first and second temperature values being determined based on the measured temperature values.

By taking other operating observables such as temperature into account, the accuracy when determining the aging value can tend to be further increased.

As described above, the absolute aging value can be determined using operating observables during an observation period. The aging value can, for example, be determined absolutely at the end of the observation period (this corresponds to an aging value at the actual time). A prediction of the absolute aging value could also be made, i.e. a forecast for the future. For example, it would be conceivable for the measured current values and the measured voltage values (or generally measured operating observables) to be augmented in order to obtain such expanded measured current values and expanded measured voltage values. These can show a prediction of the battery load for the future, beyond the observation period. This means that the observation period can be extended in the future. The first time series and second time series can then be determined based on such extended measured current values and extended measured voltage values. This enables a prediction of the state of the battery.

For example, it would be conceivable that a frequency or frequency of charging processes or discharging processes is determined and then this frequency or frequency is also assumed for the future is used to achieve the augmentation in this way. For example, a statistical load profile (load collective) could be determined based on the measured current values and the measured voltage values. Then the augmentation could take place taking into account such a statistical load profile.

A computer program or a computer program product or a computer-readable storage medium includes program code that can be loaded and executed by a processor. This causes the processor to execute a method for determining a battery aging value. The method includes obtaining measured current values of an operating current of the battery and measured voltage values of an operating voltage of the battery. The current values indicate the operating current during an observation period. Likewise, the voltage values indicate the operating voltage during the same observation period.

A first time series of first current values and first voltage values is determined based on the measured current values and the measured voltage values. In this case, the first time series includes the first points in time of the first current values and the first voltage values in the observation period. These first points in time are arranged according to a first sampling scheme defined in the time domain.

A device includes a processor. The processor can load and execute program code. This causes the processor to execute a method for determining a battery aging value. The method includes obtaining measured current values of an operating current of the battery and measured voltage values of an operating voltage of the battery. The current values indicate the operating current during an observation period. Likewise, the voltage values indicate the operating voltage during the same observation period. A first time series of first current values and first voltage values is determined based on the measured current values and the measured voltage values. In this case, the first time series includes the first points in time of the first current values and the first voltage values in the observation period. These first points in time are arranged according to a first sampling scheme defined in the time domain.

The features set out above and features described below can be used not only in the corresponding combinations explicitly set out, but also in further combinations or in isolation without departing from the protective scope of the present invention.

character list

• 1 12 schematically illustrates a system comprising multiple batteries and a server according to various examples.
• 2 12 schematically illustrates details related to the batteries according to various examples.
• 3 12 schematically illustrates details related to the server according to various examples.
• 4 Figure 12 is a flow diagram of an example method.
• 5 FIG. 12 schematically illustrates data processing for determining an aging value of a battery according to various examples.
• 6 FIG. 12 schematically illustrates two time series indicating current values at points in time, each equidistantly arranged in a time domain and a charge throughput domain, according to different examples.

DETAILED DESCRIPTION

The properties, features and advantages of this invention described above, and the manner in which they are achieved, will become clearer and more clearly understood in connection with the following description of the exemplary embodiments, which are explained in more detail in connection with the drawings.

The present invention is explained in more detail below on the basis of preferred embodiments with reference to the drawings. In the figures, the same reference symbols designate the same or similar elements. The figures are schematic representations of various embodiments of the Invention. Elements depicted in the figures are not necessarily drawn to scale. Rather, the various elements shown in the figures are presented in such a way that their function and general purpose can be understood by those skilled in the art. Connections and couplings between functional units and elements shown in the figures can also be implemented as an indirect connection or coupling. A connection or coupling can be implemented wired or wireless. Functional units can be implemented in hardware, software, or a combination of hardware and software.

Techniques related to the characterization of rechargeable batteries are described below. The techniques described herein can be used in connection with a wide variety of types of batteries, for example in connection with lithium-ion-based batteries such as lithium-nickel-manganese-cobalt oxide batteries or lithium-manganese oxide batteries.

The batteries described herein can be used in different areas of application, for example for batteries used in devices such as motor vehicles or drones or portable electronic devices such as mobile phones. It would also be conceivable to use the batteries described herein in the form of stationary energy stores.

The techniques described herein allow characterization of the battery based on condition monitoring. Condition monitoring can include ongoing monitoring of the load on the battery and/or determination of the absolute condition of the battery against a reference—typically start-up—and/or prediction of the condition of the battery. This means that the health of the battery can be tracked or determined by monitoring the load and/or predicted for some prediction interval in the future. In particular, an aging estimation of the state of health (SOH) of the battery can be made. This means that an aging value can be determined, typically the capacity of the battery (i.e. amount of charge that can be stored).

As a general rule, the SOH decreases as the battery ages. Accelerated aging can occur when the capacity of the battery decreases and/or when the impedance of the battery increases.

Various of the battery characterization examples described herein may be implemented, at least in part, on the server side. This means that at least some of the logic involved can be performed on a central server separate from the battery or battery-powered device. For this purpose, in particular, a communication connection can be established between the server and one or more management systems of the battery. By implementing at least part of the logic on the server, particularly accurate and computationally intensive models or algorithms can be used. This makes it possible to determine the aging value particularly precisely. In addition, it may be possible to collect and exploit data for an ensemble of batteries, for example in connection with machine-learned algorithms.

Various examples described herein may perform condition monitoring during battery usage based on measurement data for one or more operational observables of the battery. This means that condition monitoring is performed at a particular point in time during the life of the battery - with a reduced SOH. The battery can then be used in the field. In this way it may be possible, in particular, to also take into account the previous aging behavior of the battery. This also makes it possible to carry out the condition monitoring in a particularly flexible manner. Laboratory tests can be omitted.

As a general rule, different measurement data can be taken into account. For example, measured current values of a battery current (charge or discharge current) could be used. Alternatively or additionally, measured voltage values of the operating voltage (terminal voltage) could be used. Further examples relate to, for example, measured temperature values of the battery temperature.

According to various examples of the disclosure, corresponding time series of different battery observables are considered, e.g., a time series of current values and/or a time series of voltage values and/or a time series of temperature values.

Such time series then serve as input to an ML algorithm. The ML algorithm can be trained during a training phase in order to subsequently determine an aging value of the battery in an inference phase based on such a time series. For example, in the training phase, corresponding training time series could be measured as training data. At the same time, as a basic truth, the aging value could be determined using a laboratory method. Laboratory methods are known for determining, for example, the capacity of a battery particularly precisely (eg Coulomb counting or based on the open-circuit voltage characteristic; impedance spectroscopy; incremental capacity analysis; differential voltage analysis). Alternative models could also be used, such as Kalman filters. Based on such reference measurements or reference models, a label could then be determined which indicates the target value prediction of the ML algorithm. Based on a difference of the actual prediction of the ML algorithm, which is based on the training time series and the label, and using a suitable loss function, the parameter values of the ML algorithm can then be adjusted in the training phase. For example, a gradient descent method (backward propagation) may be employed to adjust the machine-learned algorithm's parameter values, taking into account a difference between the machine-learned algorithm's prediction based on the training time series and the label. Different instances of the ML algorithm can be used for different battery types. The training can be done separately for each instance, based on training data and labels that are obtained specifically for a battery of the respective battery type.

As a general rule, different ML algorithms can be used in the examples disclosed herein. In particular, such ML algorithms can be used which operate on time series as input. Examples would be, for example, recurrent neural networks. Examples include, for example, LSTM networks or GRU networks. In one example, the machine-learned algorithm could have two GRU layers followed by one or more fully-connected layers.

According to various examples, it is possible to use two types of time series to form a data structure serving as input to the ML algorithm. These two types are listed below in TAB. 1 described Short description Exemplary details I Time series of values for battery observables at instants arranged in a time domain according to a sampling scheme For example, it would be possible to form a time series for current values (current value time series) based on measured current values by sampling a corresponding observation period in the period according to a corresponding sampling scheme (i.e. a sampling scheme that takes into account the spacing of the points in time in the time domain). In particular, an interpolation operation can be used. This means that a corresponding current value for this point in time can be determined in each case based on the interpolation of two measured current values which are arranged adjacent to a point in time in the time series. For example, linear interpolation could be used. In addition or as an alternative to such current values, the time series can also include voltage values or temperature values at the respective points in time. II Time series of values for battery observables at points in time arranged in a stress domain according to a corresponding sampling scheme This means that the observation period for which measured current values and/or voltage values etc. are obtained is not sampled equidistantly in the period. Rather, an operating load parameter - for example the charge that has flowed when charging or discharging the battery (charge throughput) or a power output or power consumption of the battery (power throughput) - can be determined and then points in time can be determined that are equidistant to one another in relation to such an operating load parameter are. This means that, for example, a specific value for the charge that has flowed can be determined and then a point in time for the time series is taken into account when this specific value for the charge that has flowed has been reached, starting from the previous point in time. This also corresponds to an interpolation operation, in this example in a charge domain.

TAB. 1: Two examples of time series based on which an input to a machine-learned algorithm can be determined. For example, it would be possible to determine a first time series and a second time series according to Examples I and II and then to concatenate these time series to thus obtain the input for the machine-learned algorithm.

By calculating the aging value using the ML algorithm using such time series according to TAB. 1 is determined, a particularly accurate prediction can be made. This is because the two time series according to Examples I and II from TAB. 1 depict both calendar aging and stress-induced aging.

At the same time, however, by determining the two time series, a significant compression of the size of the data used can be achieved before it is entered into the machine-learned algorithm. For example, a compression by a factor of no less than 500 or no less than 5000 can be achieved. This means that the number of entries in the time series is reduced by a corresponding factor compared to the measured values of the operational observables. However, by using equidistant points in time in the time domain or the load domain, it can be ensured at the same time that meaningful signatures for determining the aging value are retained.

Such techniques have the further advantage that no particularly complicated pre-processing of the measured values for the operating observables of the battery is necessary. For example, it may not be necessary to filter the measured current values or the measured voltage values. For example, it may not be necessary to extract characteristic signatures from the measured current values or the measured voltage values. Rather, the use of suitable sampling schemes in the time and stress domain can inherently ensure that characteristic signatures for the precise determination of the aging value are retained. This enables a particularly computationally efficient implementation of the techniques.

1 illustrates aspects related to a system 80. The system 80 includes a server 81 that is connected to a database 82. FIG. In addition, the system 80 includes communication links 49 between the server 81 and each of a plurality of batteries 91-96. The communication links 49 could be implemented over a cellular network, for example.

In general, different battery types can be used in the various examples described herein. This means that the batteries 91-96 can include several types. Batteries of different battery types may differ, for example, with regard to one or more of the following characteristics: shape of the cell (ie round cell, prismatic cell, etc.), cooling system (air cooling with active or passive concept, coolant in coolant hose, passive cooling elements, etc.), cell chemistry (e.g. electrode materials used, electrolytes, etc.), etc. Even between batteries 91-96 of the same type, there may be some variance associated with such characteristics. For example, it can happen that batteries 91-96 of one and the same type are mounted differently and therefore different cooling systems are used. Also can sometimes the same battery cells are arranged differently, so that an electrical and thermal system view of the ensemble of cells varies.

In the various examples described herein, battery type specific instances of ML algorithms can be used to characterize the batteries.

In 1 is illustrated by way of example that the batteries 91-96 can send status data 41 to the server 81 via the communication links 49 . For example, it would be possible for status data 41 to be indicative of one or more operating values of the respective battery 91-96, ie can indicate measurement data. Measurement data for different operational observables can be obtained, for example measured current or voltage values or also temperature values. The status data 41 could be event-driven or sent according to a predetermined time scheme.

It would also be possible for such measured values of one or more operating observables to be pre-processed locally on a data processing system that is assigned to the respective battery 91-96. For example, time series could be determined that contain corresponding values for the one or more operational observables at specific points in time. The time series can be compressed compared to the measurement data, ie have a smaller size. For example, an interpolation operation could be performed locally.

The status data 41 can be used in connection with the characterization of the batteries 91-96. For this purpose, the server 81 can determine a corresponding aging value based on an ML algorithm. Different instances of the ML algorithm can be used for different battery types, which means that the different instances could be trained for a specific battery type.

In 1 is also illustrated by way of example that the server 81 can send control data 42 to the batteries 91-96 via the communication links 49. In particular, it can be possible to control the operation of the battery as a function of the result of the characterization of the battery—ie, for example, as a function of the aging value. For example, it would be possible for the control data 42 to indicate one or more operating limits for the future operation of the respective battery 91-96. For example, the control data could indicate one or more control parameters for thermal management of the respective battery 91-96 and/or charging management of the respective battery 91-96. By using the control data 42, the server 81 can thus influence or control the operation of the batteries 91-96. In this way, further aging could be limited, for example, in the case of advanced aging of the battery. It would also be conceivable to determine a secondary use of the battery in another application scenario, such as changing from a traction battery to a battery for a stationary energy storage device.

In 1 the respective SOH 99 is also illustrated schematically for each of the batteries 91-96. As a general rule, the SOH 99 of a battery 91-96 may include one or more different characteristics depending on the implementation. Typical parameters of the SOH 99 can be, for example: electrical capacity, ie the maximum possible stored charge; and/or electrical impedance, ie the frequency response of the resistance or alternating current resistance as a ratio between electrical voltage and electrical current intensity. Other aging values are also conceivable.

Techniques for characterizing the batteries 91-96 that enable the SOH 99 to be determined as an aging value for each of the batteries 91-96 during the use of the batteries 91-96 are described below. This means that, for example, the electrical impedance and/or the electrical capacitance can be determined. This can be done using an ML algorithm on the server. The server 81 could then in turn provide corresponding information about the SOH 99 to the batteries 91-96, for example via the control data 42. A management system of the batteries 91-96 could then adapt an operating profile for the batteries, for example to prevent further degradation of the SOH 99 to avoid.

2 illustrates aspects related to batteries 91-96. Batteries 91-96 are coupled to a device 69, respectively. This device is powered by electrical energy from the respective battery 91-96.

The batteries 91-96 comprise or are associated with one or more management systems 61, eg a BMS or other control logic such as an on-board unit in the case of a vehicle. The manage Management system 61 is arranged locally at the respective battery 91-96 and is connected to it, for example, by a wired connection. The management system 61 can be implemented by software on a CPU, for example. Alternatively or additionally, an application-specific integrated circuit (ASIC) or a field-programmable gated array (FPGA) could be used, for example. The batteries 91-96 could communicate with the management system 61 via a bus system, for example. The batteries 91-96 also include a communication interface 62. The management system 61 can establish a communication connection 49 with the server 81 via the communication interface 62.

while in 2 management system 61 is drawn separately from batteries 91-96, in other examples it would also be possible for management system 61 to be part of batteries 91-96.

In addition, the batteries 91-96 include one or more battery packs 63. Each battery pack 63 typically includes a number of battery cells connected in parallel and/or in series. Electrical energy can be stored there.

Typically, the management system 61 can rely on one or more sensors in the one or more battery packs 63 . For example, the sensors may measure current flow and/or voltage in at least some of the battery cells. Alternatively or additionally, the sensors can also measure other variables in connection with at least some of the battery cells, for example to determine the temperature, volume, pressure, etc. of the battery and send them to the server 81 in the form of status data 41 . Management system 61 can also be set up to implement thermal management and/or charging management of the respective battery 91-96. In connection with thermal management, management system 61 could control cooling and/or heating, for example. For example, in the context of charge management, the management system 61 could control a charge rate or a depth of discharges. The management system 61 can therefore set one or more operating boundary conditions for the operation of the respective battery 91-96, for example based on the control data 42.

3 12 illustrates aspects related to the server 81. The server 81 includes a processor 51 and a memory 52. The memory 52 may include a volatile memory element and/or a non-volatile memory element. In addition, the server 81 also includes a communication interface 53. The processor 51 can establish a communication link 49 with each of the batteries 91-96 and the database 82 via the communication interface 53. FIG.

For example, program code may be stored in memory 52 and loaded by processor 51 . The processor 51 can then execute the program code. Execution of the program code causes processor 51 to perform one or more of the following processes, as described in detail in connection with the various examples herein: characterization of batteries 91-96; running an ML algorithm to determine a aging value for the batteries 91-96; controlling the operation of the batteries 91-96 via control data; receiving measured operational observables from batteries 91-96; storing a result of the characterization of a corresponding battery 91-96 in a database 82; Etc..

4 Figure 12 is a flow diagram of an example method. The method is used to characterize a battery. The method characterizes the aging of the battery using an ML algorithm. For example, the procedure could be run on a central server (cf. 1 , Server 81) are running. It would also be conceivable that only parts of the method are executed on a central server and other parts of the method are executed on a management system that is arranged locally at a battery. It would also be conceivable for the entire method to be executed on a local battery management system.

In 4 optional blocks are shown with dashed lines.

First, in block 5005, measurement data is obtained. This means that measured values are obtained for one or more operational observables of a battery. These are recorded during an observation period. For example, measured current values and/or measured voltage values could be obtained. Alternatively or additionally, measured temperature values of the battery can be obtained.

For example, block 5005 could include reading appropriate sensors of the battery. Alternatively or additionally, block 5005 could include receiving corresponding status data (cf 1 : include status data 41) via a communication interface. The measurement data could also be read from a memory or a database.

It would then be optionally possible for the measured values of the one or more operating observables to be augmented in order to obtain a prediction of a load on the battery for the future beyond a corresponding observation period. Such an augmentation could, for example, continue a statistical load during the observation period for a prediction period.

Based on such an augmentation, augmented measured values for the one or more operational observables can then be obtained. Such an augmentation serves to be able to predict an aging value for a future period. If no forecast is desired for a future time period, the augmentation in block 5010 can also be omitted.

Then, in block 5015, an input data structure for the ML algorithm is determined. The input data structure can include multiple time series. These time series can each have values of one or more operating observables of the battery at corresponding points in time that are characteristic of the respective time series. Such points in time can each be determined based on a respective sampling scheme (which samples the measurement data), the sampling scheme being determined in a domain associated with the corresponding time series. For example, a first time series could include first current values and first voltage values at first points in time in the observation period (and possibly also in the prediction period), which are arranged on a first sampling scheme specified in a time domain. A second time series could include second current values and second voltage values at second points in time, the second points in time being arranged according to a second sampling scheme predetermined in a stress domain.

Depending on the sampling scheme, an interpolation operation can be used in block 5015 to determine the first time series and the second time series. For example, a linear interpolation between measured current values could be performed in order to obtain a current value for the first time series at a (target) point in time that lies between the points in time of the two measured current values. If the deviation between this target time and the times of the two measured current values is sufficiently small, an interpolation operation can be dispensed with and one of the measured current values can be used directly in the time series (e.g. the closest one).

The interpolation operation could constrain a length of the data structure. Generally speaking, the length of the data structure can be predetermined by the scanning scheme. The first current values and the first voltage values can be reduced in relation to the measured current values and the measured voltage values by the interpolation operation. Likewise, the second current values and the second voltage values can be reduced in relation to the measured current values and the measured voltage values. This means that the data structure that serves as input to the ML algorithm can be smaller than the original measurement data. This allows for a shorter runtime of the ML algorithm.

In block 5020, the aging value can then be determined. This is based on the data structure obtained as a result of block 5015. The data structure serves as input to an ML algorithm. This means that no further processing needs to occur between block 5015 and block 5020.

The data structure includes a concatenation of multiple time series containing values for one or more operational observables at respective points in time determined according to a respective sampling scheme. The data structure can also contain time stamps for the corresponding points in time.

It would then optionally be possible in block 5025 to control the operation of the respective battery based on the aging value. For example, operating boundary conditions could be specified.

5 illustrates aspects related to the processing of measurement data 501 to determine an aging value 99 of a battery. Data processing off 5 For example, the procedure might look like 4 to implement.

First, measurement data 501 is obtained, for example measured current values and measured voltage values and measured temperature values. These are processed in a pre-processing algorithm 505 in order to subsequently obtain a data structure 510 which serves as input to an ML algorithm 515 . Based on the data structure 510, the ML algorithm 515 determines the SOH 99 as the aging value. For example, the SOH 99 could indicate the capacity in absolute terms (say: ampere-hours).

The data structure 510 (compare 4 : block 5015) is a matrix that indicates voltage values, current values and temperature values for different points in time (“t”). The timestamps in the fourth column of the matrix are useful when they are not equidistant in the time domain - which would typically be the case for a sampling scheme used by the preprocessing algorithm 505 that is defined in a load domain.

6 illustrates aspects related to the measurement data 501 in an observation period 585, as well as the data structure 510 and the pre-processing algorithm 505.

In 6 , above is a time dependence of the operating current of the battery, i.e. measured current values, is shown with the dashed line (in fact, the measured current values are discrete in the time domain; due to the high sampling rate of the measurement, in 6 , shown above a line for illustration). The observation period 585 is about 10 hours long and includes several charging and discharging processes.

Based on the measured current values, a first time series 581 with first current values (filled circles in 6 , above) and a second time series 582 with second current values (stars in 6 , above). The first time series 581 and the second time series 582 cover the same observation period 585; ie the times of the first and second current values alternate in the time domain. The first current values - i.e. the filled circles - all show an equidistant distance from each other in the time domain (i.e. in the scaling of the X-axis 6 , top and bottom) on. This is because the corresponding points in time for the first current values are determined based on a corresponding sampling scheme defined in the time domain. In the representation of 6 time increases linearly as a function of the X-axis position (this is the dotted-dashed line in 6 , below; showing this linear dependency).

The second current values - that is, the stars in 6 , above - the second time series do not have an equidistant distance to each other in the time domain (i.e. in the scaling of the X-axis in 6 , above) on. This is because the respective times for the second current values are determined based on a respective sampling scheme defined in the charge throughput domain. Between two adjacent points in time of the second time series 582 (i.e. between two stars in 6 , above), the same charge always flows, i.e. the battery is charged or discharged by the same amount. this is in 6 , shown below by the non-linear correlation of charge throughput with progression of time by the dashed line.

Of course, the features of the embodiments and aspects of the invention described above can be combined with one another. In particular, the features can be used not only in the combinations described, but also in other combinations or taken on their own, without departing from the field of the invention.

QUOTES INCLUDED IN DESCRIPTION

This list of documents cited by the applicant was generated automatically and is included solely for the better information of the reader. The list is not part of the German patent or utility model application. The DPMA assumes no liability for any errors or omissions.

Patent Literature Cited

• CN 111832220A [0005]

Claims (18)

A computer-implemented method for determining an aging value (99) of a battery, the method comprising: - obtaining measured current values of an operating current of the battery (91-96) during an observation period (585) and measured voltage values of an operating voltage of the battery (91) during the observation period (585), - based on the measured current values and the measured voltage values: determining a first time series (581) of first current values and first voltage values, the first time series (581) comprising first points in time of the first current values and the first voltage values in the observation period (585), which according to are arranged in a first sampling scheme predetermined in a time domain, and - based on the measured current values and the measured voltage values: determining a second time series (582) of second current values and second voltage values, the second time series comprising second points in time of the second current values and of the second voltage values in the observation period (585), which are based on one in a arranged with a domain associated with an operational load parameter of the battery predetermined predetermined second sampling scheme, and - based on the first time series (581) and the second time series (582) and using a machine-learned algorithm (515), determining the aging value (99). procedure after claim 1 , wherein the aging value indicates an absolute value of a capacity or impedance of the battery (91-96). procedure after claim 1 or 2 , the method further comprising: - concatenating the first time series and the second time series to obtain a data structure (510), the data structure (510) serving as input to the machine-learned algorithm (515). procedure after claim 3 , wherein the data structure further includes first time stamps for the first points in time in the first time series and second time stamps for the second points in time in the second time series. A method according to any one of the preceding claims, wherein the first time series and the second time series are determined using an interpolation operation applied to the measured current values and voltage values, respectively. procedure after claim 3 or 4 , as well as after claim 5 , where the interpolation operation considers a length of the data structure as a constraint. procedure after claim 5 or 6 , wherein the first time series and the second time series are determined on a first data processing system, which is arranged locally at the battery, wherein the aging value is determined on a second data processing system, which is not arranged locally at the battery. Procedure according to one of Claims 5 until 7 , wherein the interpolation operation causes a reduction of the first current values and the first voltage values in relation to the measured current values and the measured voltage values, wherein the interpolation operation causes a reduction in the second current values and the second voltage values in relation to the measured current values and the measured voltage values. Method according to one of the preceding claims, wherein the observation period comprises a plurality of charging cycles and a plurality of discharging cycles and a plurality of rest cycles of the battery. A method according to any one of the preceding claims, wherein the observation period is less than 5% of a period since the battery was put into service. Procedure according to one of Claims 1 until 10 , wherein a beginning and an end of the observation period are determined independently of an operation of the battery. Procedure according to one of Claims 1 until 10 , the method further comprising: - monitoring an operation of the battery, and - determining a beginning and an end of the observation period depending on the monitoring of the operation of the battery. A method according to any one of the preceding claims, wherein the method further comprises: - obtaining measured temperature values of an operating temperature of the battery during the observation period, wherein the first time series also includes first temperature values and is determined based on the measured temperature values, wherein the second time series also includes second temperature values and is determined based on the measured temperature values. A method as claimed in any preceding claim, wherein the operational load parameter comprises a charge throughput of operation of the battery or comprises a power throughput of operation of the battery. A method according to any one of the preceding claims, wherein the method further comprises: - Augmenting the measured current values and the measured voltage values in order to obtain extended measured current values and extended measured voltage values, which depict a prediction of a load on the battery for the future beyond the observation period, the first time series and the second time series being based on the extended measured values current values and the extended measured voltage values. Method according to one of the preceding claims, wherein the first sampling scheme defines an equidistant arrangement of the first points in time in the time domain, wherein the second sampling scheme defines an equidistant arrangement of the second points in time in the domain associated with the operating load parameter of the battery. Method according to one of the preceding claims, wherein the first sampling scheme defines a variance for a first distance between the first points in time, wherein the second sampling scheme defines a variance for a second distance between the second points in time. A method according to any one of the preceding claims, wherein the method further comprises: - based on the aging value, controlling an operation of the battery.

Images