Method and virtual sensor for determining amount of degradation corrected using field data for electrochemical system, in particular fuel cell system

Abstract

The invention relates to a method (600) for determining an amount of degradation of an electrochemical system (13), in particular a fuel cell system or electrolysis system (13), corrected using field data, using a data-based correction model, and to a method (500) for training such a correction model. The invention also relates to a device (100), in particular a virtual sensor, and a digital process twinning, for determining a corrected amount of degradation and for quality monitoring of one or more electrochemical systems (13).

Description

Background Technology

[0001] When electrochemical cells (in high-temperature systems such as SOFC and SOEC or in low-temperature systems such as PEMFC or PEM-ELY) or battery complexes containing such cells—so-called battery stacks) are in operation, continuous degradation occurs, which may manifest, for example, as an increase in the effective surface resistance of the active regions or a reduction in the active regions, and thus continuously reduces the performance and efficiency of the system over time.

[0002] Under constant operating conditions, this continuous degradation can be seen in system performance, specifically in the voltage of the fuel cell (PEMFC and SOFC) and in the H2 volumetric flow rate of the electrolysis (PEM-ELY, alkaline electrolysis, AEM-ELY, and SOEC). However, in practice, these operating conditions (hereinafter referred to as operating states) change almost constantly, whether due to variations in performance requirements in mobile or stationary fuel cell applications or variations in the input power during electrolysis. Therefore, degradation can no longer be directly observed in performance.

[0003] To infer the progression of continuous degradation even under changing operating conditions, a method is needed to calculate a reference variable based on a combination of performance and operating conditions. This reference variable describes the progression of continuous degradation independently of the changing operating conditions. Here, the development of the reference variable (i.e., the degradation rate) can, of course, depend on the operating conditions, only that the corresponding current value should be independent of the current operating state, so that comparisons can be made between different time points and between different units. This reference variable can be a degradation parameter based on physical principles and independent of operating conditions, such as the percentage increase in surface resistance. However, it can also be the conversion of performance measured under potentially changing operating conditions in the field to performance under specified reference operating conditions. The reference variable obtained in this way allows for comparisons of degradation progression at different time points and between different systems.

[0004] Methodologically, there are, in principle, two approaches that allow the reference variable, i.e., the degradation, to be calculated as a function of performance and the operating state in which it exists.

[0005] On the one hand, data-based models can be used to calculate the degradation, for example, by calculating it as a comparison voltage of the system that is independent of operating conditions.

[0006] On the other hand, physical models can be used that allow for the correlation between performance and the existing operating state, using internal degradation parameters independent of operating conditions as reference values. These physical models must be inverted (using machine learning if necessary for real-time capability) to determine reference variables for a given combination of performance and operating conditions, thereby enabling the monitoring of degradation progress. Then, using the model, the determined reference variables can be reconstituted into performance under the given reference operating conditions.

[0007] The quality of conclusions from both methods is primarily based on the accurate depiction of the combined effects of degradation and operating conditions on performance through either a physics-based model or a data-based model. If these models are used to evaluate operating conditions or degradation states that are incorrectly parameterized in a physics-based model or not represented or adequately represented in the training data of a data-based model, erroneous predictions can be expected. This may manifest as jumps in the calculated degradation amount, which are not due to degradation but rather to changes in operating conditions. Such jumps occur because these models are typically, and due to practical limitations, only (can) use individual (method-dependent, fixed) operating points or limited intervals from the field. Figure 1 An example is shown for different operating states 70 ( And it depends on aging-related parameters. The degradation amount calculated using the inversion physical model of degradation amount (e.g., the cumulative runtime of the relevant system) is 70. The value of 50, and the aforementioned jump 55 that occurs here. Summary of the Invention

[0008] Advantages of the present invention Against this backdrop, the present invention relates to a method for determining a correction value for the degradation amount of an electrochemical system. Hereinafter, the corrected degradation amount, also referred to hereinafter as the corrected value, is determined as the sum of the value determined using a degradation amount model (which may also be referred to as the uncorrected value) and the correction value determined using a correction model. Hereinafter, the degradation amount model is also referred to as the degradation model.

[0009] Electrochemical systems can be, in particular, fuel cell systems or electrolyzer systems comprising one or more fuel cells or electrolyzers, especially high-temperature systems, such as those comprising one or more solid oxide electrolyzers (SOECs) and / or solid oxide fuel cells (SOFCs), or low-temperature systems, such as those comprising one or more proton exchange membrane electrolyzers (PEM-ELYs), proton exchange membrane fuel cells (PEMFCs), anion exchange membrane electrolyzers (AEM-ELYs), or anion exchange membrane fuel cells (AEMFCs). Furthermore, the electrochemical system can be a battery, i.e., an electrochemical energy storage system, such as a lithium-ion battery, a sodium-ion battery, or a solid-state battery.

[0010] The calibration model is a data-based model, preferably a regression model, trained using field data and based on values ​​determined using a degradation model (especially for this field data). Here, field data should be understood as detected data on the operating variables of at least one, preferably multiple (hereinafter also referred to as a group) electrochemical systems, particularly measurements of the performance of the system or these systems, and detected values, particularly measurements, regarding operating conditions, especially under these measured performance conditions, i.e., regarding the operating state of the system or these systems. In the context of this invention, system performance can generally be understood as a measure of the system's performance during operation, such as the voltage or electrical power provided by the system, particularly in the case of a fuel cell, or in the case of electrolysis, the H2 volumetric flow rate.

[0011] Advantageously, the method according to the invention enables the measurement of the degradation state of an electrochemical system based on a determined, corrected amount of degradation, wherein this determination, or indirect measurement based on calibration using field data, possesses particularly high accuracy and reliability. Therefore, the method according to the invention can advantageously be used as a virtual sensor for degradation. Particularly reliable calibration is achieved here, especially when using field data from multiple electrochemical systems, as this subsequently covers a wider range of operating states and the distribution of differences, particularly due to manufacturing variations. The invention particularly addresses the problem that, due to the combined effects of operating conditions and degradation on performance, this degradation amount is often impossible to measure directly, making it impossible to directly quantify errors and derive simple calibration schemes for the degradation model used. In other words, there is no "ground truth" for the degradation amount that allows for the direct determination of the calibration of the degradation model. Advantageously, the invention effectively avoids this problem, providing numerous beneficial advantages for the operation of electrochemical systems due to the most accurate possible understanding of the degradation state, particularly the following: • A more accurate understanding of the amount of degradation allows for the identification of anomalies or critical states, such as when the rate of change of degradation suddenly changes, especially when it increases. • A more accurate understanding of the amount of degradation allows for more precise prediction of the system's behavior in the near future, for example, for quickly adjusting operating conditions to achieve desired performance. • The most accurate understanding of the amount of degradation allows for optimal distribution of the load in a complex consisting of batteries or stacks, or systems with multiple stacks. • A more accurate understanding of the amount of degradation allows for the derivation of limit values ​​or models that include other parameters for targeted and degradation-related planned maintenance interventions (“Smart / Predictive Maintenance”). • A precise understanding of the amount of degradation is a necessary foundation for developing algorithms that predict future degradation evolution. • A thorough understanding of the amount of degradation allows for the calculation of degradation rates and their correlation with operating conditions and manufacturing data, which can be used to identify key degradation drivers. This can serve as the basis for optimizing manufacturing and operating conditions.

[0012] Preferably, the trained calibration model is further trained, particularly retrained, using additional field data not previously used for training. This facilitates continuously better and more robust determination of calibration values, and consequently, improved determination of the corrected degradation values. This additional field data can be generated and measured, in particular, through the continued operation of preferred multiple electrochemical systems, for example, through continuous, particularly sequential, field testing or batch field runs of these systems. Alternatively or additionally, this group can be expanded to include other, particularly similar, systems that provide additional field data. For example, this further training can be performed periodically or when a minimum amount of additional field data not previously used for training is available. Alternatively, the calibration model can be retrained using field data, particularly all detected field data. This means that all trainable parameters are reset to their initial values, and the calibration model is then trained using a training dataset that includes not only the training data already used for training but also data not yet used for training, particularly all available training data. The advantage of this is that all the training data used is used equally for training, and in particular, the training is not affected by other training data used later.

[0013] According to a specific design scheme, field data from a selected set of electrochemical systems is used as the other field data. This is particularly advantageous when field data from some or more electrochemical systems should not be used for training, for example, because these systems have been modified in terms of their hardware or software and are therefore no longer technically similar to other systems. Preferably, in this case, two different instances of the calibration model and / or degradation model are then used separately for further training and further determination of the calibration and degradation values. In particular, the previously trained calibration model, and, if necessary, the previously used degradation model, can be split into two instances, wherein the first instance is further trained using the other field data from the first set of electrochemical systems, and the second instance is further trained using the other field data from the second set of electrochemical systems. Here, these two sets of systems preferably each consist of only technically sufficiently similar systems.

[0014] Both the degradation model and the correction model are preferably designed to determine the value or correction value as a function of the operating value of the electrochemical system, preferably as a function of the performance of the electrochemical system (especially voltage in the case of a fuel cell or H2 volumetric flow rate in the case of electrolysis) and the operating state of the electrochemical system at that performance. Therefore, the correction model is preferably designed as an additive correction function of the degradation amount determined by the degradation model, or preferably includes such an additive correction function. Since the operating value, as the input value to the degradation model and the correction model, can be easily queried from the relevant system without complex calculations, i.e., especially the measured value, the corrected degradation amount can be determined immediately in real time and thereby indirectly measured. In this case, the fact that aging and thus degradation generally have a negative impact on performance is advantageously utilized. In other words, in the case of continuous aging / degradation due to component wear of the electrochemical system, the possible performance for a specific operating state will generally decrease. Here, as mentioned above, the operating value, especially the performance and operating state, can be incorporated as a measurement value into the degradation model and / or correction model, which is particularly advantageous when using the method described above as a virtual sensor.

[0015] The degradation model can be, in particular, a model based on one of the two methods described at the beginning, i.e., a data-based and / or physics-based model, to determine the degradation amount as a function of performance and the operating state existing under that performance. Preferably, a model is selected that has at least one reference operating state, or the model can be parameterized for at least one such reference operating state, for which the value of the degradation amount determined using the model can be considered sufficiently accurate such that correction is not required by correcting the value of a correction model. Preferably, the model is a degradation model of the steady-state operating state of the system.

[0016] Preferably, the degradation of the corresponding electrochemical system is a continuous function of aging-related parameters, which should be understood as variables of the electrochemical system that monotonically increase with time. For example, aging-related parameters correspond to the cumulative operating time of the electrochemical system, or, in the case of fuel cells or electrolyzers, to the cumulative current density or hydrogen production density. Preferably, the same parameters are selected for all electrochemical systems, and especially for all systems used in the group under consideration.

[0017] The subject of this invention is also a method for training such a calibration model, specifically a method for training a calibration model to determine calibration values ​​for values ​​of degradation in an electrochemical system (especially a fuel cell system). The calibration model is preferably a regression model, i.e., a model with parameters determined by a regression algorithm. For this training method, training data is used, comprising data tuples containing measured field data from at least one, preferably multiple, electrochemical systems. Each data tuple includes a measurement of the performance of the electrochemical system and a value of the detected operating state at that performance. In other words, each data tuple includes the operating point of the electrochemical system, characterized by measured performance at the detected operating state. Thus, when using field data from multiple electrochemical systems, the training data includes operating points of different electrochemical systems, wherein each operating point can be assigned to exactly one electrochemical system. Furthermore, each data tuple includes a separate calibration value for the degradation amount related to performance and operating state. Here, the individual correction value is determined by combining at least some of the detected field data and by combining a value determined using a degradation model, wherein the determined degradation value is determined using the model based on the at least some detected field data, and in particular, is determined as a function of the at least some detected field data. Preferably, to determine the individual correction value, field data from exactly one electrochemical system are used, such that each individual correction value, and therefore each data tuple, can also be uniquely assigned to exactly one electrochemical system.

[0018] Therefore, the method for training the calibration model according to the invention advantageously comprises two training steps. In the first step, individual calibration values ​​for each electrochemical system are determined based on the field data detected for the corresponding system, and are combined with the values ​​at the relevant operating points to form data tuples of training data for the calibration model. In the second step, the calibration model is trained using this training data, wherein the training data now includes field data for multiple, in particular all, electrochemical systems used, as well as the individual calibration values ​​determined in the first step.

[0019] According to the preferred design, these individual correction values ​​are calculated as differences between the value of the degradation amount determined using the model and the function value of the fitted degradation function for the operating state and the measured performance detection values. The model determines the value of the degradation amount as a function of the performance of the electrochemical system and the operating state existing under that performance. Furthermore, the degradation function is determined as a fit of the value of the degradation amount determined using the model to a continuous function of the operating state and the measured performance detection values ​​(i.e., the detected field data for the relevant electrochemical system). Here, the degradation function is selected as a function of the aging-related parameters of the relevant system, for example, as a function of the system's cumulative operating time or cumulative current density as described above. Therefore, for this fitting, the value of the degradation amount determined using the model is plotted relative to the corresponding relevant value of the aging-related parameters, and then the fitting is performed for the value pairs (the determined value of the degradation amount of the system at operating point X, and the determined value of the aging-related parameters of the system at operating point X). Here, the required values ​​for aging-related parameters can be transmitted along with field data, especially as part of the detected operating status, or determined from the usage history recorded by the relevant system.

[0020] Preferably, the value pairs used for fitting the degenerate function are considered to different degrees in the fitting. Advantageously, this allows for the consideration of these value pairs to different degrees for the fitting, thereby considering different operating states, especially depending on the distance between the operating state of the corresponding value pair and one or more (as described above) reference operating states. For this purpose, a function of the distance can preferably be defined, which determines the degree to which the value correspondence is considered in the fitting. This function is particularly used as a weighting function, which can be a monotonically decreasing function, such as a linearly decreasing function, a polynomial decreasing function, or an exponentially decreasing function. Preferably, a metric for the distance is also defined so as to specify the influence of the value pairs according to the distance, and the function is designed to be a function of this metric. Depending on the number of operating conditions defining the operating states and thereby according to the number of values ​​for each operating state, the metric for the distance between the operating state and the reference operating state can be selected, for example as a simple, weighted, or normalized Cartesian distance or the maximum of the normalized distance. This advantageously allows for the consideration not only of these operating states as a whole, but also for the consideration of different operating conditions and their expected effects to different degrees in the fitting. As described above, the degradation function thus determined can be used to determine these individual correction values, which then constitute part of the training data used to train the correction model.

[0021] As described above, the method used to train the calibration model can be reapplied, preferably arbitrarily frequently, to the already trained calibration model, especially when other field data that has not yet been used to train the calibration model is available, particularly, as described above, using other field data only from a specific electrochemical system for further training.

[0022] The method according to the invention, namely, the method for determining the amount of degradation of an electrochemical system corrected using field data and the method for training a correction model, can in particular be designed as a computer-implemented method. Therefore, the invention further includes: a computer program comprising instructions that, when executed by a computer, cause the computer to perform a corresponding method according to the invention; and a computer-readable data carrier on which the computer program is stored. Finally, the invention also includes: a computer including such a computer-readable data carrier; and an evaluation and control unit including such a computer-readable data carrier, and further including means for performing the steps of the method according to the invention. In particular, the method according to the invention can also be used as part of a digital process twin, and can be implemented as part of such a process twin, particularly as a virtual sensor, especially for process and / or quality monitoring of one or more electrochemical systems. Attached Figure Description

[0023] Embodiments of the present invention are shown in the accompanying drawings and will be described in more detail in the following description.

[0024] in: Figure 1 This illustrates the non-physical jump problem in the model used to determine the amount of degradation, as described at the beginning; Figure 2 A group of electrochemical systems having embodiments of the device according to the invention is shown; and Figure 3 and Figure 4 A flowchart illustrating an embodiment of the method according to the present invention is shown; and Figure 5 The fit for determining the degradation function of a single electrochemical system to determine individual correction values ​​for the degradation value of the system is shown. Detailed Implementation

[0025] Figure 2 Multiple electrochemical systems 10 are shown, which, as described at the beginning, may be, for example, similar systems comprising SOEC, SOFC, PEM-ELY, or PEMFC. The method according to the invention can also be applied to a single electrochemical system 10, but preferably, it can be applied to multiple such systems 10. Figure 2Also shown is an apparatus 100 on which the method according to the invention is implemented, such as a computer, part of a cloud computing architecture, or other programmable device, for example as a virtual sensor of one or more electrochemical systems 10 or as part of a digital twin of the one or more electrochemical systems.

[0026] The device 100 and the electrochemical system 10 are designed to transmit operating data of the electrochemical system 10 to the device 100, particularly via a suitable wireless communication module such as a mobile radio or a WLAN module. The device 100 is designed to receive this operating data and preferably also transmit the data to the systems 10, for example, in response to a request to transmit the operating data or to control commands to change the operating state of one or more of the systems 10.

[0027] In Figure 2 Taking the systems 10 or devices 100 shown as examples, such as systems containing SOFC or PEM-ELY, the following embodiments of the invention describe the method according to the invention. Here, the method according to the invention can be applied to the entire group, or it can be applied only to a portion 11 of the group, that is, only to selected systems in the group.

[0028] First, an embodiment of a method for training a calibration model is described, wherein the calibration model is used to determine a calibration value for the degradation amount of electrochemical system 13. The electrochemical system 13 may be in... Figure 2 The system shown is one of the systems 10 or another system (in which case, the field data of the other system is not used to determine or train correction values ​​or correction models), wherein, in the latter case, the other system is preferably of the same type, i.e., especially having the same characteristics as... Figure 2 The system 10 shown in the figure has the same structure type. Figure 3 A flowchart of the relevant method steps 500 is shown.

[0029] That is, an exemplary method 500 is described, which allows by including a system The field data of the group was used to determine the degradation amount using the additive correction function. The correction model (part of it) , The corrected degradation amount is calculated according to the following formula. , The degradation determined by the model (hereinafter also referred to as the "uncorrected model") without incorporating such field data. Compared to simple values, it better approximates the actual degradation amount. This method 500 employs two assumptions: 1. When considered in isolation, if a suitable, continuous, monotonically increasing aging-related parameter is selected... As the independent variable (defined for all systems), 10 for each electrochemical system The correct amount of degradation Both represent a continuous function. Examples include the cumulative operating time, cumulative current density, or hydrogen production density of a fuel cell or electrolyzer. 2. There exists at least one known reference operating state. For this reference operating state, the uncorrected degradation model provides good predictions, making It can be considered accurate enough that no correction is needed.

[0030] For example, especially for SOFCs, a degradation model, such as a numerical, particularly data-driven inversion model, based on methods from Zaccaria, V., Tucker, D., Traverso, A., 2016. A distributed real-time model of degradation in a solid oxide fuel cell, part I: Model characterization. Journal of Power Sources 311, 175–181. Doi:10.1016 / j.jpowsour.2016.02.040, can be used as an application-specific parameterized model. Here, for example, the following approach can be taken: First, select a model with suitable parameters that can be adapted to the operating conditions. To predict the performance of the system under consideration For example, according to the method of the publications of Zaccaria et al. discussed above: In the next step, select at least one parameter in the model. This parameter depicts the decisive impact of degradation on performance, such as surface resistivity. Alternatively, it can be embedded in the model. This can be interpreted as a specific degradation factor, such as a dimensionless pre-factor preceding the surface resistance. In this way, the model allows for the following mapping: Then, perform a large number of simulations, in which... and The values ​​vary within the expected range on-site. This model is then used to generate the dataset. Finally, this dataset was used to train a regression model, which has the following form: The model can then be used as a degradation model according to the present invention.

[0031] The method 500 preferably includes the following six steps 501-506: 1. First, in step 501 of the first method, for each working point detected by the device 100 and measured on-site, Each working point includes a value representing its running status. and the performance values ​​of the electrochemical system existing under this operating condition. The degradation model is used to determine the relevant values ​​of the degradation amount. : As described above, for this purpose, field data (i.e., detected operating points) from all systems 10 can be used, or field data (i.e., detected operating points) from only selected systems 10 of the considered group (i.e., a portion 11 of that group) can be used. In particular, for this method, field data from the second portion 12 of systems 10 that exhibit abnormal behavior or can only provide a small amount of field data close to the aforementioned reference operating state can be disregarded.

[0032] 2. Then, in the second step, for each system 10, the selected parameters are... These values ​​of degradation are plotted above, causing the production Point set , in, Indicates that it is for the corresponding system The number of working points detected from the field data, and therefore the number of determinable values ​​representing the amount of degradation. Figure 5 This is illustrated by example for system 10. Here, the parameters The value is, for example, transmitted by system 10 to device 100 as part of field data, or calculated by device 100 based on the data transmitted by system 10, particularly by integration, for example, in the case of deriving cumulative operating time or cumulative current from a detected curve of operating time or current. For example, using =100 systems 10, and each system 10 contains =20 (for all) There are 2000 work points in total. In practice, the number of available work points varies across different systems. It may be different.

[0033] 3. In the subsequent third step 503, define the metric. This metric defines the work point measured in the field. Compared with the above reference operating state Calculation of the distance between each pair: This metric can take into account the different expected effects of various operating parameters. Examples are simple Cartesian distance, weighted Cartesian distance and / or normalized Cartesian distance, or the maximum value of normalized distance.

[0034] 4. In step 504, using this metric Define a function that describes the degree to which the working point should be considered in the fit provided in subsequent step 505. This function can be used as... The monotonically decreasing function will weight the working point Described as or as A monotonically increasing function is used to account for the uncertainty of the operating point. Instructions are For example, for this function, you can choose a linear function or a quadratic function (for...). Or choose the inverse quadratic function (for...) The choice of this metric and the weight together determine the impact of each working point on the fit.

[0035] 5. Now, in step 505, perform the fitting to determine 10 for each system. function , such as in Figure 5 The example is shown as a fitting function 330 for system 10. Therefore, the systems to be fitted can preferably be respectively... The range is limited to a range with a sufficiently high density of working points near the reference working point, where the aforementioned metrics can be used. Alternatively, a weighted metric can be preferably used. or , as a metric for determining density. Here, these to be fitted The range should be limited such that, on the one hand, the density of working points near the reference working point is not too small, and on the other hand, the number of remaining, more distant working points for the fit is not too small. Thus, the ideal choice depends on the data considered. For these ranges, then 10... Perform fitting when using a weighting function or an uncertainty function: or ,in Here, the fitting function 330 could be, for example, its weights. spline functions, or those using uncertainty. The heteroscedastic Gaussian process is used. For the fitted range, discrete individual correction values ​​are then obtained according to the following formula: , As in Figure 5 The example shown is for a single correction value 301. These discrete individual correction values ​​are called individual correction values ​​because they are determined only when combined with field data from each individual system 10.

[0036] Because for all The operating conditions measured on-site are also known. and performance Therefore, individual correction values ​​for all systems They are combined separately in the data tuple containing the work points: ,in, .

[0037] Therefore, each data tuple includes the value of the operating point, i.e., the performance value of system 10 and the value of the operating conditions existing at that performance, as well as a separate correction value related to the amount of degradation. These data tuples then constitute the training data used to train the correction model in the subsequent sixth step 506. .

[0038] 6. Now, use the training data obtained in this way in step 6.506. In order to use regression algorithms To train the calibration model This allows the trained correction model to perform the required correction. As running status and performance The function is used to predict. .

[0039] In principle, It can be any regression algorithm, such as linear regression, random forest, or explainable boosting machine.

[0040] Therefore, by utilizing the correction model trained in this way, the amount of degradation corrected using field data can be determined for the corresponding operating point using the method 600 of the present invention for determining the amount of degradation corrected using field data. The steps of the method are described exemplarily below and Figure 4 It is schematically shown in the flowchart.

[0041] In the first step 601, the value of the degradation amount is determined using the (uncorrected) degradation model. In particular, it was identified as a relevant work site. The function. In the second step 602, the correction value is determined using the trained correction model. In particular, this correction value was determined as the relevant operating point. The function, wherein the second step 602 may be performed before, after, or simultaneously with the first step 601 in time. In the third step 603, the two values ​​are added to form the corrected value of the corrected degradation. : .

[0042] The described training method 500 can be applied periodically, especially when multiple systems 10 continuously generate data during field testing or batch operation. Here, it can be expected that the generated calibration model will continuously improve and become more robust. When modifying system 10 in subsequent development, separate calibration models should preferably be determined for groups of systems 10 that are sufficiently technically similar.

Claims

1. A method (500) for training a calibration model to determine calibration values ​​for values ​​of degradation in an electrochemical system (13), particularly a fuel cell or electrolysis system (13), wherein, The correction model is trained using training data, which comprises detected field data from at least one, preferably multiple, electrochemical systems (10, 11, 13) in the form of data tuples. Each data tuple includes a measured performance value of one of the electrochemical systems (10, 11, 13) and a detected operating state value of the system (10, 11, 13) at said performance, as well as a separate correction value for the amount of degradation related to said performance and said operating state. The separate correction value is determined by combining at least some of the detected field data and by combining values ​​determined using the degradation model.

2. The method (500) of claim 1, wherein The calibration model is trained using a regression algorithm with the training data.

3. The method (500) of claim 1 or 2, wherein, The relevant individual correction values ​​are calculated as differences between the value determined by the model for the degradation amount and the function value of the fitted degradation function for the operating state and the measured performance, respectively, wherein the model determines the value of the degradation amount as a function of the performance of the electrochemical system (10) and the operating state present under said performance, and wherein the degradation function is determined as a fit of a continuous function of the value of the degradation amount determined by the model for the operating state and the measured performance.

4. The method (500) of claim 3, wherein, The degradation function indicates the degradation of the electrochemical system (10) as a function of a continuous aging-related parameter, and the fitting of the degradation function is performed as a fitting of a pair of values ​​of the determined value of the degradation amount and the corresponding values ​​of the aging-related parameter, wherein the pair of values ​​is preferably included in the fitting with different weights, wherein the weights depend in particular on the distance between at least one reference operating state and the corresponding operating state belonging to the pair of values.

5. The method (500) of claim 4, wherein, Define a metric for the distance so as to specify the effect of the value on the fit, and in particular the effect of the various operating conditions of the running state on the fit.

6. A method (600) for determining a correction value of a degradation quantity of an electrochemical target system (13), in particular a fuel cell or electrolysis system (13), wherein The uncorrected value of the degradation amount of the system (13) can be determined using a degradation amount model of the system (13), especially the steady-state operating state, wherein the corrected value is determined by a correction model, wherein the correction model is trained using field data from at least one electrochemical system (10, 11, 13) and based on the value determined by the degradation amount model, especially by the method (500) according to any one of claims 1 to 5.

7. The method (600) of claim 6, wherein, The correction model is designed to output the correction value when the operating value of the target system (13), especially the performance value of the target system (13) and the operating state of the target system under the performance, is input.

8. The method (600) of claim 6 or 7, wherein Determine the corrected degradation amount of the electrochemical system (13), particularly the fuel cell or electrolysis system (13), wherein the degradation amount model is used to determine the uncorrected value of the degradation amount, wherein the uncorrected value is determined as the corrected value of the corrected degradation amount by adding it to the corrected value determined using the correction model according to claim 6 or 7.

9. The method (500, 600) according to any of the preceding claims, wherein The trained calibration model is further trained using other field data, wherein, preferably, only field data from a selected set of electrochemical systems (10) are used as other field data.

10. The method (500, 600) of any of the preceding claims, wherein, The degradation model determines the value of the degradation as a function of the performance of the electrochemical system (10), particularly the voltage and the operating state of the electrochemical system, and preferably, the correction model determines the correction value as a function of the performance, particularly the voltage and the operating state.

11. The method (500, 600) according to any of the preceding claims, wherein The model determines the value of the degradation amount to be independent of the current operating conditions.

12. A device (100) for determining a degradation quantity of a usage site data correction of an electrochemical system (10), in particular a fuel cell system or an electrolysis system (10), in particular a virtual sensor, wherein The device (100) is configured to perform the method (500, 600) according to any one of the preceding claims.

13. Digital process twinning, in particular for process and / or quality monitoring of one or more electrochemical systems (10), wherein The twin is configured to perform the method (500, 600) according to any one of claims 1 to 10.

14. A computer program comprising instructions that, when executed by a computer, cause the computer to perform the method according to any one of claims 1 to 11.

15. A computer-readable data carrier having a computer program as claimed in claim 15 stored thereon.

Images