Method and apparatus for predicting a change profile of an aging state of a device battery based on user individualized load distribution

Abstract

The invention relates to a method for predicting a battery aging state profile based on a user individualized load profile, comprising: - determining a current aging state value at an evaluation time, - providing at least one prediction value for the aging state from the current load profile by means of the current value, - providing a predicted aging state trajectory for the aging state based on the current aging state value and the at least one prediction value, - determining a further current aging state value at a later evaluation time, - providing at least one further aging state prediction value from the current load profile with the further current aging state value, and - performing the steps: - providing a further predicted aging state trajectory with the further current aging state value and the at least one further prediction value, - providing the further predicted aging state trajectory as the predicted aging state profile, if the deviation of the prediction value on the predicted aging state trajectory from the further current aging state value at the further aging state value evaluation time is greater than a predetermined threshold value.

Description

Technical Field

[0001] This invention generally relates to the prediction of the aging state of a device battery based on a user-personalized load distribution, which is derived from the variation curve of the device battery's operating parameters. Background Technology

[0002] Device batteries degrade during their lifespan and according to their use. For traction batteries in electric vehicles (EVs, HEVs), the degree of degradation depends decisively on driver behavior, external environmental conditions, and the type of vehicle battery. This aging (SoH) results in a continuous decrease in maximum storage capacity, and for EVs, it ranges from SoH = 100% at the time of delivery (BoL, Begin of Life) to approximately 80% at the end of life (EoL).

[0003] Physical aging state models (based on electrochemical battery model equations) can determine the current aging state of a battery based on historical operating state change curves. However, these aging state models have a model bias greater than 5%, making it difficult to predict the aging state change curves as accurately as possible. Other aging state models are data-based models or hybrid models (physical models combined with data-based models), but these aging state models can no longer be implemented on the vehicle's own Battery Control Unit (BCU), but rather on an external storage medium / central unit, such as the cloud. There, the model is parameterized or trained using operating parameters from multiple vehicles with the same battery type, which are in communication connection with the central unit (so-called connected cars).

[0004] Such models are also known, based on capacitance retention using coulomb counting. Predicting the aging state curve of the battery is an important technical parameter because it enables an economic assessment of the battery's and ultimately the vehicle's remaining value. The uncertainty of the prediction plays a crucial role here, as it quantifies the technical and economic risks, for example, in cases of specified aging or warranty breaches.

[0005] Different aging state prediction methods are based on inaccurate aging state models and therefore also on inaccurate numerical values, which impair the quality of the predicted values.

[0006] Therefore, the aging condition value is a key parameter used to describe the remaining battery capacity and to determine the remaining useful life (RUL). Summary of the Invention

[0007] According to the present invention, a method for predicting the change curve of the aging state of a battery according to claim 1 and a corresponding apparatus according to the parallel claims are provided.

[0008] Other design options are described in the dependent claims.

[0009] According to the first aspect, a method is provided for predicting the aging state change curve of a battery based on a user-personalized load distribution, the method comprising the following steps: - Determine the current aging state value at the time of evaluation. - Provide at least one predicted value for the aging state of the device battery based on the current values ​​and the current load distribution. - Provide a predicted aging trajectory for the battery's aging state based on the current aging state value and at least one predicted value. - Determine another current aging status value at a later evaluation time. - Starting from another current aging state value, provide at least one additional predicted value for the aging state based on the current load distribution, and - If the deviation between the predicted value at another current aging state value and the predicted value on the predicted aging state trajectory at the evaluation time of the other aging state value is greater than a predetermined threshold, perform the following steps: - Provide another predicted aging trajectory starting from another current aging state value and at least one other predicted value. - Provides another predicted aging trajectory as the predicted aging change curve.

[0010] Furthermore, when the deviation between the current aging state value and the predicted value on the predicted aging state trajectory at the evaluation time of the other aging state value is less than a predetermined threshold, and when the other predicted value in the farthest future is further into the future than the predicted value in the farthest future, another predicted aging state trajectory is provided starting from the other current aging state value and at least one other predicted value, wherein the other predicted aging state trajectory is provided as the predicted aging state change curve.

[0011] As an alternative or supplementary option, the last determined predicted aging state change curve is retained if the deviation between the predicted value at the time of evaluation of another current aging state value and the predicted value on the predicted aging state trajectory is less than a predetermined threshold, and if the other predicted value in the farthest future is not further into the future than the predicted value in the farthest future.

[0012] It is possible to specify at least one predicted value and at least one additional predicted value for the aging state of a device battery by means of a corresponding current load distribution, the current load distribution being derived, in particular, from the operating parameter variation curves between the current aging state value and the previously determined aging state value, based on an aging state model comprising a system of differential equations and evaluating the time series of the operating parameter variation curves of the current load distribution.

[0013] In particular, the current load distribution can include durations, wherein the predicted value is separated from or offset from the time of a previous predicted value or the time of the last identified current aging state value by the duration, or the other predicted value is separated from or offset from the time of another previous current predicted value or the time of the last identified current aging state value by the duration.

[0014] Therefore, the method can specify that the user-personalized load distribution includes durations, and that the predicted values ​​include multiple points separated by the durations or trajectories derived from the points. It has been shown that good results can be obtained with such equidistant patterns.

[0015] Furthermore, it can compare the most distant future forecast with the warranty forecast at the same testing point, and output an operational recommendation when the most distant future forecast deviates from the warranty forecast at the same testing point, indicating greater degradation of the device battery. This operational recommendation can include measures for more protective operation of the device battery, such as reducing fast charging operations.

[0016] Furthermore, the provision of the predicted aging trajectory can be implemented by interpolating the curve's change curve using the current aging value and at least one predicted value, starting from the current aging value and at least one predicted value. Similarly, the provision of another predicted aging trajectory can be implemented by interpolating the curve's change curve using another current aging value and at least one additional predicted value, starting from another current aging value and at least one additional predicted value. Interpolation can be based on a parametric model, such as a polynomial function.

[0017] One idea is to determine the battery's aging state for a specific number of subsequent prediction levels (Prädiktionshorizont) based on the user's personalized load distribution and the corresponding current aging state value using an aging state model.

[0018] By accurately predicting the battery's aging condition, drivers can be given important tips / operational recommendations regarding their usage behavior, thereby maximizing the lifespan of the vehicle battery up to EoL (Earning Over Time).

[0019] The method described here takes into account the user's personalized load distribution when modeling and predicting aging states. Here, time-varying curves of specific operating parameters (current, temperature, voltage, and / or state of charge at the single-pool level, module level, or group level) from the start-up time (BoL) are used to determine the aging state for multiple prediction levels from the aging state to the start-up time using the aging state model, and adjust it as necessary.

[0020] The aging state can be simulated or determined using conventional and known models for determining aging states. In particular, electrochemical aging state models can be used, which are in principle based on electrochemical battery models. Such electrochemical battery models can include a system of differential equations that, based on differential equations parameterized by model parameters, model the internal battery state, particularly the equilibrium state, and, where necessary, the kinetic state, using time integration methods, and provide time-varying curves of the device battery's operating parameters, i.e., the relationship between the device battery's current, voltage, temperature, and state of charge and the internal battery state. Such electrochemical battery models are known, for example, from published documents US20230305073A1, US20220179009A1, US20220334191A1, US20220099743A1, US2016 / 023,566, US2016 / 023,567, and US2020 / 150,185. The aging state can be derived from the internal battery state. Data-based models or hybrid models, which serve as models of aging state, can also be used to identify changes in aging state.

[0021] The basic idea is to describe a method that uses an aging state model to determine the battery's aging state for multiple subsequent prediction levels based on user-personalized load distribution from past prediction levels and corresponding current aging state values. The method is based on incorporating one or more past operating parameter variation curves into the aging state determination and thus into its prediction.

[0022] This disclosure has the following features or advantages: - It can achieve segmented extension of the predicted trajectory based on the current effective aging state value; - The observable quantity and length of the predictive level allow for the identification of predicted aging state values ​​with small inaccuracies / dispersion widths, which provides earlier and more accurate operational recommendations with extended remaining battery life (RUL), improved battery value, economic and ecological advantages and sustainability, which enables improvements in life cycle assessment (LCA). - The prediction level is derived from the conditions that trigger the event; - The method is independent of the aging state model used; - No computationally intensive prediction methods / models are needed; comparing with a threshold value is sufficient. -Simple predictions without calibration models are only needed to adjust the predicted aging trajectory; - The basis for determining aging state values ​​is the actual historical operating parameters; - It can incorporate the driver's personalized load distribution pattern into the determination of the predicted aging trajectory; - No discontinuity is generated in the aging state change curve because the aging state trajectory continues based on the effective aging state value.

[0023] It can be specified that the aging state model is event-triggered and that the aging state is determined by charge measurement during the charging process. This is an event that occurs repeatedly under similar conditions, thus enabling a good determination of the aging state.

[0024] According to one implementation, the user-personalized load distribution and / or predicted values ​​can include curves showing the variation of battery operating parameters. These curves can be used from the moment of activation (BoL) to determine the aging state for multiple predicted levels based on the aging state at the moment of activation and to adjust it as necessary.

[0025] It is possible to specify, optionally by means of a load spectrum having an estimated current distribution, to determine warranty predictions in advance, with the battery subjected to loads of said load spectrum during typical driving cycles over its lifespan. This allows for accurate estimations before actual driving characteristics are observed.

[0026] It is possible to specify that information used during vehicle use as user characteristic patterns and / or load distribution patterns can be used to adjust predictions. This has the advantage of being able to very quickly identify changing, unusual, and undesirable aging conditions or usage behaviors and respond to them with user-specific operational recommendations.

[0027] Furthermore, the method can specify that the predicted values ​​be adjusted using information identified as user characteristic patterns and / or load distribution patterns during vehicle use, so as to generate, for example, a more robust aging trajectory.

[0028] The described method is applicable to all types of battery-equipped vehicles and to all applications of lithium-ion batteries and other storage technologies, both stationary and mobile, and to consumer applications, such as batteries in smartphones with cloud connectivity.

[0029] According to a second aspect, an apparatus is provided for predicting the aging state change curve of a battery based on a user-personally-defined load distribution. This apparatus is configured to perform the method described above. The aforementioned design of the method can also be a design of the apparatus. Attached Figure Description

[0030] The preferred embodiments of the present invention will now be explained in detail with reference to the accompanying drawings. Wherein: Figure 1 A flowchart is shown for a method to predict the change curve of battery aging state based on user-personalized load distribution. Figure 2 A graph illustrating the predicted aging trajectory following the first prediction after BoL is shown. Figure 3 A graph illustrating the predicted aging state trajectory in the presence of deviations in aging state values / trajectories is shown; and Figure 4 A graph illustrating the predicted aging state trajectory without bias in the aging state value / trajectory is shown. Detailed Implementation

[0031] Figure 1 An example of method 10 for predicting the change curve of battery aging state based on user-personalized load distribution is shown.

[0032] This method is executed on an external storage medium / central unit, such as in the cloud. To this end, in step 11, the user-personalized load distribution is recorded from the BoL and fed into an aging state model. This model can be a base model that determines the aging state through charge measurements during the charging process. The determination of the aging state using the base model is event-triggered, thus the model values ​​for the aging state are only available at irregular times, such as after the vehicle battery has been charged. Alternatively, an electrochemical aging state model based on a system of differential equations can be used, which evaluates the changes in operating parameters.

[0033] Aging state models used to determine the aging state of a battery can be provided in the form of physical, data-based, or hybrid aging state models—that is, a combination of physical (electrochemical) aging models and data-based models. For example, these calculations are performed weekly. Which model is used is not important to this approach.

[0034] In step 12, the current aging state value or potential aging state value is determined using an aging state model as early as after the first charging process or after the first week following the BoL. It is possible that multiple charging processes occur within a week, or no charging process occurs at all, thus determining multiple aging state values ​​or no aging state value. However, it is also possible that the EV remains inactive for several weeks and calendar aging is not (sufficiently) considered in the model, with the aging state value only determined weekly when the EV is in motion.

[0035] Therefore, in step 13, the following conditions are determined for this method, and the current valid aging state value should be output according to these conditions.

[0036] The condition can specify that: "After at least a pre-given time period of at least one week and at least, for example, a newly determined current aging state value, the last identified current aging state value shall be determined as the currently valid aging state value or the current value of the aging state."

[0037] Once the conditions are met, this value is output in step 14, that is, after event (a) "reaching a predetermined duration" (with or without EV use) and event (b) "reaching the number of charging cycles". This determines the frequency at which the currently valid aging state value or the current value of the aging state is determined.

[0038] In step 15, it is checked whether the current aging status value is the first value since the battery was enabled.

[0039] In step 16, at least one predicted value for the predicted state of aging of the battery is provided. More precisely, the predicted value may include multiple predicted state of aging values ​​that are spaced apart from each other by duration, or a trajectory derived from the predicted state of aging values ​​(step 17).

[0040] This is referred to below. Figure 2 To explain this, we show a graph 25 illustrating the predicted aging trajectory following the first prediction after BoL 26.

[0041] Figure 25 shows the time-varying curves of the aging state trajectory plotted from BoL 26 to EoL 27. For the battery, the warranty-aging trajectory 28 from BoL 26 to EoL 27 can be determined based on a pre-given, artificially generated, reference operating parameter variation curve. As a predicted value 29, the predicted aging state SoH is shown here, with control points 1 / 1p, 1 / 2p, 1 / 3p, and 1 / 4p. The aging state is observed here within the range from 100% SOH at BoL 26 to 80% SOH at EoL 27.

[0042] To this end, for example, starting from the first aging state value identified after BoL that satisfies the conditions mentioned above, based on the user-personalized load distribution up to that point (with respect to duration t1) – the temporal variation curves of operating parameters for discharging, quiescent state, and charging – this load distribution is used to model and predict several other consecutive predicted aging state values ​​by means of a physical / data-based / hybrid (especially electrochemical) aging state model, which can evaluate the time series of operating parameters in order to determine the resulting aging state value.

[0043] The identified aging state value is the current aging state value 1 / g. Predicted change curves of the operating parameters, for example four curves within the corresponding time period H1 = t1, are used to predict the aging state value using the aging state model. Each of these time periods is of equal length and represents the prediction level H1. Operating parameter change curves are assumed for each predicted time period, corresponding to the operating parameter change curves up to the current aging state within time period t1. Therefore, the predicted time period describes a time period that is contiguous with and ends with the last currently valid aging state value, i.e., the duration until the aforementioned conditions are met.

[0044] Using these four predicted values ​​1 / 1p, 1 / 2p, 1 / 3p, 1 / 4p, the current effective aging state value 1 / g, and the aging state start value at BoL (SoH = 100%), an interpolation curve based on a polynomial function, referred to as the aging state trajectory 29, was plotted.

[0045] In the next optional step 18, the forecast value 29 for the furthest future is compared with the pre-given warranty forecast value 28.

[0046] In the next optional step 19, an operational recommendation is output when there is a deviation between the most distant future predicted value 29 and the warranty predicted value 28. This may happen when a certain limit or threshold for said deviation is exceeded. Figure 2 In the current situation, the two values ​​or trajectories are close enough together that there is no deviation.

[0047] Then, return to the branch before step 11, thus starting the next iteration of the method at time t2.

[0048] The process is now similar to that described above; however, instead of starting from BoL, the current load distribution (the detected actual operating parameter change curves) is fed to the aging state model from the last, determined current aging state value 1 / g, thereby determining another current aging state value 2 / g. The prediction of the number (four) of these values ​​is based on the load distribution over the time period t1 from BoL to the current aging state value 1 / g. This usage pattern, based on the historical change curves of the operating parameters from BoL to time t1, provides possible future change curves of the operating parameters, reflecting possible vehicle user behavior or possible battery usage.

[0049] Accordingly, in step 15, the current value of the aging state is determined, for example, 2 / g (see...). Figure 3 This is not the first value since the battery was enabled, and the branch proceeds to step 20.

[0050] In step 20, the current aging state value is compared with the predicted value 1 / 1p. This is achieved by using... Figure 3 Let me explain.

[0051] Figure 3 This is a diagram illustrating the method in the presence of a deviation between the aging state value and the predicted trajectory.

[0052] The actual operating parameter change curve of the current load distribution (during time period t2) does not have to be forced to be the same as the historical operating parameter change curve (during time period t1), thus enabling the generation of a deviation Δ between the first predicted value 1 / 1p after the current aging state value 1 / g and another current aging state value 2 / g.

[0053] In step 21, this deviation Δ is evaluated.

[0054] If this deviation Δ is less than a pre-given threshold SW, then a predicted value for the aging state is determined according to step 22 using another current aging state value. This is done as described previously in steps 16 and 17.

[0055] Then, according to step 23, a comparison is made between the potential end time of the predicted aging state trajectory (each end time corresponds to the farthest future predicted value, upon which the aging state trajectory is determined) used for this currently valid aging state value and the end time of the already existing predicted aging state trajectory, the current aging state value being compared with the already existing predicted aging state trajectory.

[0056] This is thanks to Figure 4 Let me explain. Figure 4 A graph illustrating the predicted aging state trajectory without deviation of the aging state value / trajectory is shown. In this example, to observe the judgment in step 23, the starting point is the third time t3, that is, the third iteration of the method. Accordingly, there exists a currently valid aging state value 3 / g, which lies on the predicted aging state trajectory or prediction curve of the previously valid aging state value 2 / g. Therefore, the deviation Δ is less than the pre-given threshold SW.

[0057] The predicted aging trajectory (not shown here for simplicity) for the currently valid aging state value 3 / g includes four aging state values ​​3 / 1p, 3 / 2p, 3 / 3p, and 3 / 4p over four equally long time intervals t3.

[0058] If the end time of the predicted aging state trajectory 30 used for the currently valid aging state value is further into the future than the end time 31 of the predicted aging state trajectory that already exists from the value 2 / g, then the predicted aging state trajectory used for the currently valid aging state value 3 / g is retained and the process branches to optional step 18. This variant does not... Figure 4 As shown in the image.

[0059] If the end time of the predicted aging state trajectory 30 used for the currently valid aging state value 3 / g is in time prior to the end time 31 of the already existing predicted aging state trajectories 2 / 1p, 2 / 2p, 2 / 3p, 2 / 4p (e.g., in...) Figure 4 As shown in the diagram, the predicted aging trajectory is not generated starting from this currently valid aging state value. The method continues with the next identified currently valid aging state value. That is, the method returns to step 11.

[0060] If the deviation Δ in step 21 is greater than a pre-given threshold SW, then steps 16 and 17 are performed as described above, and optionally steps 18 and 19 are also performed. Therefore, based on this value 2 / g, aging state prediction is performed using the current load distribution, and the predicted aging state trajectory is determined using points 2 / 1p, 2 / 2p, 2 / 3p, and 2 / 4p. Figure 3 ).

[0061] As an alternative to the deviation of another currently valid aging state value 2 / g from the predicted value 1 / 1p or from the predicted trajectory—since the two points do not necessarily coincide in time (t1 does not have to be equal to t2)—it is also possible to adjust the predicted aging state trajectory when another current aging state value 2 / g is outside the confidence interval of multiple predicted values ​​or the first (earliest) predicted value 1 / 1p of the predicted aging trajectory.

[0062] In another design approach, the aging trajectory can be finally adjusted only after confirming the deviation Δ following the next valid aging value, thereby eliminating potential outliers. If user characteristic patterns / load distribution patterns are identified every two weeks during vehicle use, for example, based on even / odd weeks or repeated distributions, this understanding can be used to generate a more robust aging trajectory.

[0063] For each battery, a warranty-aging trajectory 28 from BoL 26 to EoL 27 is determined based on artificially generated operating parameter variation curves. The warranty-aging trajectory 28 is determined in advance, for example, based on operating parameter variation curves, such as load spectrum, i.e., the estimated current distribution, and, where necessary, temperature distribution (with the battery subjected to loads using said load spectrum during typical driving cycles over its service life). Battery manufacturers design batteries based on these characteristic curves. If the battery is subjected to loads using a similar load spectrum in actual driving operation, the actual aging of the battery is determined according to the warranty-aging trajectory variation curves.

[0064] The predicted aging trajectory described in this method follows this curve. Because the aging trajectory has a shorter prediction range (e.g., 4-8 weeks) compared to the warranty-aging trajectory (from BoL to EoL), it is only calibrated for this timeframe. This has the advantage of very quickly identifying changing, unusual, and undesirable aging conditions or usage behaviors and being able to respond to them with user-specific operational recommendations that prevent the battery from aging further. As the prediction level increases, the confidence interval widens—because the actual battery load is unknown and can only be estimated—the accuracy of the aging value decreases, by which time the battery may already be irreparably damaged, and determining the corresponding operational recommendations becomes increasingly difficult.

[0065] Using the described method, the battery aging state can be determined for an arbitrarily long prediction level by means of an aging state model based on the user's personalized historical load distribution and the corresponding current aging state value, wherein the aging state model can evaluate the time series of the operating parameter change curves of the load distribution.

Claims

1. A method for predicting the state of aging of a battery based on user-personalized load distribution, comprising the following steps: - Determine the current aging status value (1 / g) at the time of evaluation (12). - Using the current value (1 / g) to provide at least one predicted value (1 / 1p…1 / 4p) for the aging state of the device battery based on the current load distribution. - Provide the battery's aging state trajectory (29) based on the current aging state value (1 / g) and at least one predicted value (1 / 1p…1 / 4p) (16). - Determine another current aging status value (2 / g) at a later evaluation time (18). -Based on another current aging state value (2 / g), provide (22) at least one additional predicted value (2 / 1p…2 / 4p) for the aging state according to the current load distribution, and - When the deviation between the predicted value at another current aging state value (2 / g) and the predicted value on the predicted aging state trajectory (29) at the evaluation time (t_1 / pT) of the other aging state value (2 / g) is greater than a predetermined threshold, perform the following steps: - Provide another predicted aging trajectory (32) starting from another current aging state value (2 / g) and at least one other predicted value (2 / 1p…2 / 4p). - Provide another predicted aging state trajectory (32) as the predicted aging state change curve.

2. The method according to claim 1, wherein the deviation between the predicted value at the evaluation time (t_1 / pT) of another current aging state value (2 / g) and the predicted value on the predicted aging state trajectory (29) is less than the predetermined threshold and when the other predicted value in the farthest future (2 / 4p) is further into the future than the predicted value in the farthest future (1 / 4p), the following steps are performed: - Provide another predicted aging trajectory (32) starting from another current aging state value (2 / g) and at least one other predicted value (2 / 1p…2 / 4p). - Provide another predicted aging state trajectory (32) as the predicted aging state change curve.

3. The method according to claim 1 or 2, wherein the deviation between the predicted value at the evaluation time (t_1 / pT) of another current aging state value (2 / g) and the predicted value on the predicted aging state trajectory (29) is less than a predetermined threshold and the last determined predicted aging state change curve is maintained when the other predicted value (2 / 4p) in the farthest future is not further into the future than the predicted value (1 / 4p) in the farthest future.

4. The method according to any one of claims 1 to 3, wherein at least one predicted value (1 / 1p…1 / 4p) and at least one additional predicted value (2 / 1p…2 / 4p) for the aging state of the device battery are determined by means of a corresponding current load distribution, the current load distribution being derived, in particular, from a curve of operating parameter variation between the current aging state value and immediately preceding a previously determined aging state value, the aging state model comprising a system of differential equations and evaluating a time series of the operating parameter variation curve of the current load distribution.

5. The method of claim 4, wherein the current load distribution comprises a duration, and wherein the predicted value is separated from the time of a previous predicted value or from the time of the last identified current aging state value by the duration or is offset from each other by the duration, or another predicted value is separated from the time of another previous current predicted value or from the time of another last identified current aging state value by the duration or is offset from each other by the duration.

6. The method according to any one of claims 1 to 5, wherein a warranty prediction is optionally determined in advance by means of a load spectrum having an estimated current distribution, wherein the battery is subjected to load with said load spectrum during a typical driving cycle during its service life.

7. The method according to any one of claims 1 to 6, wherein the prediction level of the prediction value in terms of time is shorter than the prediction level of the warranty prediction value in terms of the time from the start of battery operation to the end of operation.

8. An apparatus for predicting a change curve of battery aging state based on user-personalized load distribution, said apparatus being configured to implement the method (10) according to any one of claims 1 to 7.

9. A computer program product comprising instructions that, 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 8.

10. A machine-readable storage medium comprising instructions that, 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