Method for monitoring electrochemical system and electrochemical system
By adopting machine learning degradation models and reference models in fuel cell systems, monitoring and predicting system degradation, the problem of instability in the operating parameters of fuel cell system is solved, and reliable monitoring of degradation and safe operation mode is realized, which improves the System stability and security.
Patent Information
- Application Number
- CN202380032784.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Priority Date
- 2022-04-07
- Filing Date
- 2023-04-05
- Publication Date
- 2024-11-15
AI Technical Summary
Existing fuel cell system monitoring methods are difficult to effectively predict and monitor system degradation, especially in terms of wear, aging and functional regression caused by sediment, resulting in instability of operating parameters and potential safety risks.
Using machine learning-based degradation model and reference model, the degradation rate and expected residual service life of the fuel cell system are determined through the parameters and calculation unit evaluation of sensor detection, and the safe operation mode is activated when a predetermined threshold is reached, limiting the power range.
Reliable prediction and monitoring of fuel cell system degradation is achieved, the risk of functional regression is reduced, the stability and safety of the system is improved, and the degradation and reference models can be quickly created and evaluated.
Smart Images

Figure CN118974982A_ABST
Abstract
Description
Background Art
[0001] A method for monitoring a fuel cell system, in particular a high-temperature fuel cell system, has been proposed, wherein in at least one method step the functionality of the fuel cell system is determined based on at least one machine learning process. Summary of the invention
[0002] The present invention is based on a method for monitoring an electrochemical system, in particular a high-temperature electrochemical system, preferably a fuel cell system, wherein in at least one method step the functionality of the electrochemical system is determined based on at least one machine learning process.
[0003] In particular, the electrochemical system is configured to convert chemical energy into electrical energy or to convert electrical energy into chemical energy. The electrochemical system may be, for example, a fuel cell system, which in particular enables the conversion of a fuel, such as hydrogen, into electrical energy, or an electrolysis cell system, which in particular enables the conversion of electrical energy into chemical energy, preferably for storage. Accordingly, the high-temperature electrochemical system may be, for example, a high-temperature fuel cell system or a high-temperature electrolysis cell system.
[0004] The invention is described below by way of example for a fuel cell system without being restricted thereto. Thus, the invention can also be provided, for example, for an electrolysis cell system instead of a fuel cell system.
[0005] It is proposed to determine the functionality with the aid of a degradation model that is at least partially separated from a reference model of the fuel cell system. The method is preferably configured to determine the degradation of the fuel cell system, in particular to predict the degradation in advance. The degradation preferably represents a particularly gradual decline in the functionality of the fuel cell system, for example due to wear, aging processes, deposits, etc. The functionality is preferably evaluated according to at least one target parameter determined using the reference model and / or the degradation model. The at least one target parameter can be an operating parameter of the fuel cell system, for example a voltage provided by the fuel cell system, or an abstract characteristic number of the fuel cell system, wherein in particular a plurality of operating parameters of the fuel cell system are combined into the abstract characteristic number. Preferably, the at least one target parameter is output to an operator of the fuel cell system and / or a higher-level control facility. Before being output, the at least one target parameter can optionally be further processed with respect to the actual state of degradation, with respect to the degradation rate, with respect to the expected remaining service life of the fuel cell system, with respect to the time point at which the next maintenance should be performed, etc. Additionally or alternatively, the at least one target parameter is used by a control or regulation unit of the fuel cell system to control or regulate the fuel cell system, for example, so that when the functional loss of the fuel cell system exceeds a predetermined threshold value, a safe operating mode of the fuel cell system is started, which limits the power range.
[0006] The reference model preferably maps at least one reference input parameter to a reference value of at least one target parameter. The degradation model preferably maps at least one degradation input parameter to a degradation value of at least one target parameter. The word parts "reference" or "degradation" in the reference input parameter and the degradation input parameter merely indicate the designation of the corresponding input parameter assigned to the corresponding model. In principle, every parameter available during the operation of the fuel cell system can be used as a reference input parameter and / or as a degradation input parameter. In particular, the available parameters can be used only as reference input parameters, only as degradation input parameters or not only as reference input parameters but also as degradation input parameters. At least one reference input parameter and / or at least one degradation input parameter can be detected by means of at least one sensor of the fuel cell system or can be determined by a calculation unit of the fuel cell system, in particular based on a variable detected by means of at least one sensor. At least one reference input parameter and / or at least one degradation input parameter can be in particular a physical variable or an abstract characteristic number. The calculation unit preferably evaluates the reference model and / or the degradation model in order to determine the target parameter.
[0007] The reference model and the degradation model preferably complement each other in the description of the fuel cell system, in particular in the determination of at least one target parameter. The degradation model preferably describes changes in at least one target parameter caused by the degradation of the fuel cell system, preferably all significant changes. Examples used as degradation input parameters or incorporated into at least one degradation input parameter include the cumulative operating time of the fuel cell system, the cumulative current throughput of the fuel cell system, the cumulative number of thermal cycles of the fuel cell system, or other parameters that describe or characterize the aging of the fuel cell system. For example, a combination of a cumulative aging value and an actual value of an operating parameter of the fuel cell system, such as the product of the actual value of the current throughput of the fuel cell system and the cumulative operating time of the fuel cell system, is used as a degradation input parameter and / or incorporated into at least one degradation input parameter. For example, a characterizing parameter of the fuel cell system, for example, detected after the production of the fuel cell system and before the fuel cell system is put into operation, is used as a degradation input parameter and / or incorporated into at least one degradation input parameter.
[0008] The reference model preferably describes the dependency of the target parameter on the actual values of the operating parameters and / or operating conditions of the fuel cell system. Examples of operating parameters used as reference input parameters or incorporated into the reference input parameters are, for example, the current generated by the fuel cell system, the fuel gas utilization of the fuel cell system, the fuel gas utilization of the fuel cell stack of the fuel cell system, the volume flow of the fuel gas, the volume flow of the oxygen-containing fluid, characterizing variables of the fuel cell system detected, for example, after the production of the fuel cell system and before the fuel cell system is put into operation, etc. Examples of operating conditions used as reference input parameters or incorporated into the reference input parameters are, for example, the ambient temperature, the ambient air pressure, etc.
[0009] The reference model is preferably at least substantially independent of the degradation of the fuel cell system. The reference model should preferably be considered substantially independent of the degradation if the change in the reference value due to the degradation under otherwise constant operating conditions is at least less than 75%, preferably less than 50%, particularly preferably less than 25%, and extremely preferably less than 10% of the change in the degradation value due to the degradation in the same time period. Preferably, at least one reference input parameter, in particular all reference input parameters, are independent of the degradation, for example because the at least one reference input parameter / all reference input parameters are kept constant by means of regulation. Alternatively, the dependence of at least one reference input parameter on the degradation within the reference model is at least partially compensated, in particular by the dependence of another reference input parameter on the degradation. Preferably, the dependence of at least one reference input parameter on the degradation of the fuel cell system is so low, in particular due to the compensation and / or for reasons inherent in the reference input parameter, that the reference model is substantially independent of the degradation. The reference model can describe, for example, a fuel cell system in a fully effective state, wherein the reference value represents an ideal value and is increasingly corrected by the degradation value as the degradation progresses. Alternatively, the reference model describes a minimum permissible effective state of the fuel cell system, wherein the reference value forms a base value which is increasingly supplemented by the degradation value as degradation progresses. Alternatively, the reference model describes the fuel cell system in a state with an average functionality, in particular in relation to the expected service life of the fuel cell system.
[0010] In at least one embodiment, the degradation model and the reference model are separated in the sense that the determination of the degradation value is performed independently of the determination of the reference value and / or the determination of the reference value is performed independently of the determination of the degradation value. The reference model and / or the degradation model are preferably created by a learning unit with the aid of at least one machine learning process. The learning unit can be the already mentioned computing unit or another external computing unit, wherein after the creation of the reference model and / or the degradation model, the reference model and / or the degradation model are transferred to a memory of the computing unit, in particular automatically or manually. Depending on the selection of the target parameter, the specific design of the fuel cell system and / or the selection of the machine learning process, the learning unit creates the reference model and the degradation model independently of each other, in a unilaterally dependent or mutually dependent manner. The learning unit is preferably configured to separate the dependency of at least one target parameter from at least one reference input parameter and at least one degradation input parameter, in particular as much as possible. It is conceivable that the reference model and the degradation model overlap with each other, for example with one or more common input layers when a neural network is used as a machine learning process.
[0011] The learning unit preferably creates a reference model and a degradation model by evaluating training data. The training data comprises a dependency of a target parameter on at least one, in particular a plurality of input parameters. The training data may be created by simulating a fuel cell system and / or detected at other fuel cell systems. The learning unit preferably analyzes the training data in order to identify at least one input parameter as a reference input parameter or a degradation input parameter. For example, the learning unit analyzes the training data with respect to a correlation between at least one input parameter and at least one target parameter. If the learning unit finds a correlation between at least one input parameter and a long-term trend of at least one target parameter, the learning unit preferably identifies the input parameter as a degradation input parameter. If the learning unit does not find a correlation between at least one input parameter and a long-term trend of at least one target parameter, the learning unit preferably identifies the input parameter as a reference input parameter. Alternatively, the input parameters are divided into degradation input parameters and reference input parameters by a programmer of the learning unit. Preferably, the degradation model is trained based on the training data, preferably at least one reference input parameter is removed from or not included in the training data, which is referred to as degradation training data in the following. Preferably, the reference model is trained based on training data, preferably from which at least one degradation parameter is removed or not contained, which are referred to below as reference training data.
[0012] By means of the design according to the invention, the degradation and / or degradation rate of the fuel cell system can be advantageously reliably determined. The risk that a target parameter trend which is not attributable to degradation and which is described in particular by a reference model is mistakenly identified as degradation can advantageously be kept at a low level. In particular, when determining the degradation and / or degradation rate by interpreting the target parameter, an advantageously high rationality can be achieved. In addition, the degradation model can be used directly, so that the modeled variation of the target parameter can advantageously be obtained directly without short-term changes. In addition, in particular while advantageously complex and / or flexible machine learning processes can be used for the reference model, advantageously simple-to-interpret machine learning processes, such as linear regression, can be used for the degradation model.
[0013] It is further proposed that a degradation model is created by means of a machine learning process and a reference model is created by means of another machine learning process. The machine learning process and the other machine learning process may be of the same type or may be machine learning processes of different types. The machine learning process and the other machine learning process may be performed by the same learning unit or by different learning units. For example, the degradation model is determined by means of linear regression or by means of a regularized variant of linear regression (such as by means of lasso regression or by means of ridge regression). It is also conceivable to use a higher dimensional model, in particular a square or cubic model, for the degradation model. For example, the reference model is determined by means of a Gaussian process, by means of a support vector machine, by means of a decision tree, by means of an integrated variant of the methods already mentioned (in particular by means of a random forest method or by means of a gradient boosting method), by means of a neural network, etc. Through the design scheme according to the present invention, the machine learning process can advantageously be specifically adapted to the determination of the reference value and to the determination of the degradation value of the target parameter.
[0014] It is further proposed that the degradation model and the reference model have independent dependencies on each other. The degradation model is preferably, in particular, only dependent on at least one degradation parameter and is in particular independent of at least one reference parameter. The reference model is preferably, in particular, only dependent on at least one reference parameter and is in particular independent of at least one degradation parameter. Preferably, the degradation model and the reference model are determined using a machine learning process that is inherently separable with respect to the input parameters. For example, the degradation model and the reference model are jointly determined from the training data by means of linear regression or by means of a regularized variant of linear regression, such as by means of lasso regression or by means of ridge regression. The degradation model preferably has coefficients, in particular one coefficient each, assigned to at least one degradation input parameter. In order to determine or predict the degradation value, at least one coefficient of the degradation model is evaluated by the calculation unit, in particular when the coefficient of the reference model is ignored. Alternatively, the degradation model and the reference model are jointly evaluated by the calculation unit in order to determine the degradation value, wherein at least one reference input parameter is set to a constant value. The degradation model and / or the reference model can be advantageously created and evaluated simply by the design according to the invention.
[0015] It is further proposed that the degradation model and the reference model are determined by alternating iteration steps. The method preferably comprises a reference iteration step. In the reference iteration step, the learning unit preferably creates a reference submodel by means of another machine learning process and reference training data, in which the target parameter is replaced by the current transfer value. In a first pass through the reference iteration step, the transfer value is set equal to the value of the target parameter contained in the training data. In the reference iteration step, the learning unit preferably determines the iteration value of the target parameter by means of the reference submodel determined in the reference iteration step. In the reference iteration step, the calculation unit preferably determines a new transfer value according to the deviation of the iteration value of the target parameter determined in the reference iteration step from the previous transfer value. The method preferably comprises a degradation iteration step. The new transfer value is transferred to the degradation iteration step as the current transfer value. In the degradation iteration step, the learning unit preferably creates a degradation submodel by means of a machine learning process and degradation training data, in which the target parameter is replaced by the current transfer value. In the degradation iteration step, the learning unit preferably determines the iteration value of the target parameter by means of the degradation submodel determined in the degradation iteration step. In the degradation iteration step, the calculation unit preferably determines a new transfer value based on the deviation of the iteration value of the target parameter determined in the degradation iteration step from the previous transfer value. The new transfer value is preferably transferred to the reference iteration step as the current transfer value. The reference iteration step and the degradation iteration step are preferably repeated until a deviation measure (e.g., mean square deviation) between the iteration value and the current transfer parameter is less than a predetermined threshold. In order to determine or predict the degradation value by means of the degradation model or to determine or predict the reference value by means of the reference model by the calculation unit, preferably all degradation sub-models or reference sub-models created in the corresponding iteration step are evaluated and their results are aggregated. Through the design according to the present invention, any machine learning process can be advantageously used to create the reference model and the degradation model.
[0016] It is further proposed that the degradation model is determined based on a reference model. Preferably, in a model determination step of the method, the reference model is determined by a learning unit with the aid of a training data set. In particular, the reference model is dependent on at least one reference input parameter and at least one degradation parameter. Preferably, in a further model determination step of the method, the degradation model is determined by the learning unit. In a further model determination step, the learning unit preferably uses degradation training data in which the target parameters are replaced by pseudo values of the target parameters to determine the degradation model.
[0017] The learning unit preferably determines a pseudo value with the aid of the reference model determined in the model determination step, wherein at least one degradation input parameter from the training data is used and a substitute value is used for the at least one reference input parameter. The substitute value is, for example, a constant value stored in a memory of the learning unit or an average value of the values of at least one reference input parameter contained in the training data. The calculation unit preferably determines the actual value of the degradation with the aid of the degradation model. The calculation unit determines, for example, only a prediction of the target parameter with the aid of the reference model. By means of the design according to the invention, the reference model and the degradation model can be advantageously created quickly and advantageously evaluated quickly.
[0018] It is further proposed that, in order to determine at least one operating parameter of the fuel cell system, the degradation model and the reference model have a common base value, in particular different from zero. For example, the degradation model and the reference model are determined as discussed in the previous paragraph. Preferably, the learning unit determines the base value, in particular additionally. The calculation unit preferably determines the base value as the target parameter extracted from the training data minus the degradation value determined from the training data using the degradation model and minus the reference value determined from the training data using the reference model. When determining the reference value, the learning unit preferably inserts a constant value, in particular zero, into the reference model instead of the degradation input parameter of the training data. The learning unit preferably determines the base value of multiple data sets, in particular all data sets, of the training data, and forms an average value from the obtained base values, which is stored as the base value in a memory of the calculation unit. Preferably, the calculation unit adds the stored base value, the prediction of the degradation value created using the degradation model and the prediction of the reference value created using the reference model to form a prediction of the target parameter, wherein the at least one degradation input parameter remains constant in the reference model. By means of the embodiment according to the invention, reference models and degradation models can be advantageously quickly created and advantageously quickly evaluated, wherein the degradation model can additionally be verified.
[0019] It is further proposed that the degradation model is separated from the superordinate model in order to determine the reference model. Preferably, after the degradation model has been separated from the superordinate model, the reference model is determined as a remaining residue of the superordinate model. Preferably, in the model determination step of the method, the superordinate model is determined by the learning unit with the aid of a training data set. In particular, the superordinate model is associated with at least one reference input parameter and at least one degradation input parameter. In a further model determination step, the learning unit preferably determines the degradation model using degradation training data, in which the target parameter is replaced by a pseudo value of the target parameter. The learning unit preferably determines the pseudo value with the aid of the superordinate model determined in the model determination step, wherein at least one degradation input parameter from the training data is used and a substitute value is used for at least one reference input parameter. Preferably, the reference model is determined with the aid of reference training data, in which the target parameter is determined by the difference between the target parameter contained in the reference training data and the degradation value determined by the learning unit with the aid of the degradation model and the degradation training data. By means of the embodiment according to the invention, reference models and degradation models can be advantageously quickly created and advantageously quickly evaluated, wherein the degradation model can additionally be verified.
[0020] Furthermore, an electrochemical system, in particular a high-temperature electrochemical system, preferably a fuel cell system, is proposed, which has at least one computing unit for carrying out the method according to the invention.
[0021] Here also, the invention is described below by way of example with reference to a fuel cell system, without being restricted thereto. Thus, the invention can also relate, for example, to an electrolysis cell system instead of a fuel cell system.
[0022] A "computing unit" or "learning unit" should be understood as a unit having information input, information processing and information output. The computing unit or learning unit advantageously has at least one processor, a memory, input and output devices, other electrical components, an operating program, a regulation routine, a control routine and / or a calculation routine. The components of the computing unit or learning unit are preferably arranged on a common circuit board and / or advantageously arranged in a common housing. The computing unit or learning unit is preferably specially established and / or specially programmed for performing the method according to the invention. The fuel cell system preferably comprises at least one fuel cell device. The fuel cell device preferably comprises at least one fuel cell, in particular at least one fuel cell stack. At least one fuel cell is preferably configured as a high-temperature fuel cell, for example as a solid oxide fuel cell (SOFC) or a molten carbonate fuel cell (MCFC). At least one fuel cell is preferably configured to convert a fuel gas, in particular hydrogen and / or at least one hydrocarbon, in the case of conveying oxygen, in order to generate electric current. The fuel cell device preferably comprises components for processing and / or preparing a fuel gas and / or an oxygen-containing fluid (in particular ambient air). Examples of such components are at least one blower, at least one reformer, at least one afterburner, at least one heat exchanger, at least one exhaust gas recirculation device and / or other components that appear meaningful to a person skilled in the art. The fuel cell system preferably comprises at least one sensor, which is arranged in the fuel cell device in order to detect at least one operating parameter of the fuel cell system, in particular the fuel cell device. The fuel cell system preferably comprises at least one output unit, such as a display, an indicator light, a loudspeaker, etc., in order to output a target parameter and / or a variable derived from the target parameter. The output unit, the calculation unit and / or the learning unit can be arranged at the fuel cell device, in particular in the housing of the fuel cell device or at the housing of the fuel cell device, or can be constructed separately from the fuel cell device and communicate with the fuel cell device (in particular the sensor) and / or each other via a data network, in particular via the Internet or via a dedicated data network. By means of the design according to the invention, a fuel cell system can be provided, the degradation of which can be advantageously and reliably monitored.
[0023] In this case, the method according to the invention and / or the fuel cell system according to the invention are not to be restricted to the above-described applications and embodiments. In particular, the method according to the invention and / or the fuel cell system according to the invention may have a number of individual elements, components and units and method steps that differs from the number mentioned herein for fulfilling the mode of action described herein. In addition, in the case of value ranges specified in the disclosure, the values within the mentioned limits are also to be regarded as disclosed and as being freely usable. BRIEF DESCRIPTION OF THE DRAWINGS
[0024] Other advantages are derived from the following description of the drawings. Embodiments of the present invention are shown in the drawings. The drawings, the specification and the claims contain many features in combination. Those skilled in the art will also appropriately consider these features individually and combine the features into other meaningful combinations.
[0025] Figure 1 A schematic diagram showing a fuel cell system according to the present invention; and
[0026] Figure 2 A schematic flow chart of the method according to the invention is shown. DETAILED DESCRIPTION
[0027] Figure 1 A fuel cell system 12 is shown. The fuel cell system 12 preferably comprises a fuel cell device 20 for generating electric current by an electrochemical reaction of a fuel gas with oxygen. The fuel cell system 12 preferably comprises at least one sensor 22 for detecting at least one operating parameter of the fuel cell device 20. The fuel cell system 12 comprises a computing unit 18 for executing the method 10, which comprises: Figure 2 . The calculation unit 18 is configured to determine a target parameter, in particular as a function of at least one operating parameter detected by means of at least one sensor 22. The degradation model and the reference model are preferably stored in a memory of the calculation unit 18, with the calculation unit 18 determining the target parameter with the aid of the degradation model and the reference model. The degradation model and / or the reference model may be created by the calculation unit 18 or by an external learning unit 26. In the following, a "learning unit" may refer to the calculation unit 18 or the external learning unit 26. The calculation unit 18 is preferably configured to interpret the target parameter regarding the degradation of the fuel cell device 20. The fuel cell system 12 preferably includes an output unit 24 for outputting the target parameter regarding the degradation of the fuel cell device 20 and / or interpreting the target parameter regarding the degradation of the fuel cell device 20.
[0028] Figure 2A method 10 for monitoring a fuel cell system 12, in particular a fuel cell device 20, is shown. The method 10 preferably includes a machine learning process 14. The method 10 preferably includes a further machine learning process 16. The method 10 preferably includes an evaluation step 34. In the evaluation step 34 of the method 10, the functionality of the fuel cell system 12 is determined by a computing unit 18 based on at least one of the machine learning processes 14, 16. The functionality is determined by the computing unit 18 with the aid of a degradation model that is at least partially separated from a reference model of the fuel cell system 12. The degradation model is created by the learning unit with the aid of the machine learning process 14, and the reference model is created with the aid of the further machine learning process 16. The computing unit 18 preferably determines a degradation value of a target parameter with the aid of the degradation model. The computing unit 18 preferably determines a reference value of the target parameter with the aid of the reference model.
[0029] The method 10 preferably comprises a data acquisition step 28. In the data acquisition step 28, training data are detected, with the aid of which the machine learning process 14, 16 is created. The training data preferably comprises a data set consisting of input parameters and values of target parameters. The input parameters are preferably equal to operating parameters that can be detected by means of the sensor 22 or to quantities or features derived therefrom. The input parameters can be selected in particular as any combination of operating parameters in the process of feature engineering. Particularly preferably, the target parameter is equal to a voltage associated with the current generated by the fuel cell device 20. The training data can be created by simulating the fuel cell device 20 or by detection at other fuel cell devices.
[0030] The method 10 preferably comprises a preprocessing step 30 for preprocessing the training data. The training data are preferably filtered, in particular with regard to the operating time of the fuel cell device 20. For example, only those data sets of the training data are further used, which are detected or created within a predetermined nominal load range of the fuel cell device 20. For example, only those data sets of the training data are further used, in which the input parameters fluctuate by less than a predetermined tolerance value and in particular describe the stationary state of the fuel cell device 20. Optionally, a fuel cell device 20 of the type to be considered is preselected, which is to be described by the degradation model and the reference model.
[0031] The learning unit or the programmer of the learning unit preferably assigns the input parameters to the reference model as reference input parameters, especially if the influence is relatively low, or to the degradation model as degradation input parameters, especially if the influence is relatively high, depending on the influence of the input parameters on the target parameters. Preferably, the calculation unit 18 creates reference training data including reference input parameters and target parameters and degradation training data including degradation input parameters and target parameters from the training data. The learning unit performs a machine learning process 14 for creating a degradation model, preferably using the degradation training data set. The learning unit performs another machine learning process 16 for determining the reference model, preferably using the reference training data set.
[0032] The method 10 optionally includes a verification step 32. In the verification step 32, the value of the target parameter is predicted by means of the determined degradation model and / or the determined reference model and the value is compared with the test data. If the test data and the predicted value are sufficiently consistent, the degradation model and the reference model are stored in the memory of the calculation unit 18. If the test data and the predicted value are not sufficiently consistent, the machine learning process 14 and the further machine learning process 16 are preferably repeated based on other machine learning models. Alternatively, a plurality of machine learning models known to the learning unit are executed in sequence, and it is determined in the verification step 32: based on which machine learning model the most accurate degradation model and the most accurate reference model are created. When the machine learning process 14 and the further machine learning process 16 are executed by means of linear regression or other machine learning models that are inherently separable with respect to the input parameters, the degradation model and the reference model have dependencies on these input parameters that are independent of each other. When the machine learning process 14 and the further machine learning process 16 are executed, the degradation model and the reference model are determined by alternating iterative steps. When the machine learning process 14 and the further machine learning process 16 are executed, the degradation model is determined based on the reference model. When executing the machine learning process 14 and the further machine learning process 16, the learning unit determines a base value common to the degradation model and the reference model, in particular different from zero, which the calculation unit 18 uses to determine the target parameter. When executing the machine learning process 14 and the further machine learning process 16, the learning unit separates the degradation model from the superordinate model in order to determine the reference model.
[0033] In an evaluation step 34, the calculation unit 18 evaluates the reference model and / or the degradation model based on the operating parameters determined by means of the at least one sensor 22 in order to obtain an actual value of the target parameter or a prediction of the target parameter. The calculation unit 18 preferably interprets the target parameter with respect to the degradation and / or degradation rate of the fuel cell device 20. The method 10 preferably comprises an output step 36 in which the output unit 24 outputs the target parameter and / or its interpretation with respect to the degradation and / or degradation rate.
[0034] In an alternative embodiment not shown in greater detail, the electrolytic cell system may be similar to Figure 1 The fuel cell system 12 shown in FIG. 1 is constructed as shown in FIG. Accordingly, the electrolytic cell system 12 may preferably include an electrolytic cell device for producing hydrogen by electrochemical reaction with water under the auspices of electrical energy. The electrolytic cell system may also include at least one sensor for detecting at least one operating parameter of the electrolytic cell device. The electrolytic cell system may also include a sensor for performing a similar operation to the above. Figure 2 A computing unit of the method. Accordingly, the computing unit can be configured to determine the target parameter, in particular based on at least one operating parameter detected by means of at least one sensor. A degradation model and a reference model can preferably be stored in the memory of the computing unit 18, and the computing unit determines the target parameter with the aid of the degradation model and the reference model. The degradation model and / or the reference model can be created by the computing unit or an external learning unit. In the following, a "learning unit" can refer to a computing unit or an external learning unit. The computing unit can preferably be configured to interpret the target parameter with respect to the degradation of the electrolysis cell device. Accordingly, the electrolysis cell system can also preferably include an output unit for outputting the target parameter and / or an interpretation of the target parameter with respect to the degradation of the fuel cell device.
[0035] Accordingly, according to an alternative embodiment, it is possible to Figure 2 A method for monitoring an electrolytic cell system, in particular an electrolytic cell device, is performed. The method may preferably include a machine learning process. The method may also preferably include another machine learning process 16. The method may preferably include an evaluation step. In the evaluation step of the method, the functionality of the electrolytic cell system may be determined by a computing unit based on at least one machine learning process. The functionality may be determined by the computing unit with the aid of a degradation model that is at least partially separated from a reference model of the electrolytic cell system. The degradation model may be created by the learning unit with the aid of a machine learning process 14, and the reference model may be created with the aid of another machine learning process. Preferably, the computing unit determines a degradation value of a target parameter with the aid of a degradation model. The computing unit preferably determines a reference value of the target parameter with the aid of a reference model.
[0036] For alternative embodiments, the method may also preferably include a data acquisition step. In the data acquisition step, training data may be detected, with the aid of which a machine learning process is created. The training data may preferably include a data set consisting of input parameters and values of target parameters. The input parameters may preferably be equal to operating parameters detectable by sensors or quantities or features derived from the operating parameters. The input parameters may be selected in particular as any combination of operating parameters during feature engineering. Particularly preferably, the target parameter will be equal to the amount of hydrogen produced by the electrolysis cell device. The training data may be created by simulating the electrolysis cell device or by detecting at other electrolysis cell devices.
[0037] In the scope of alternative embodiments, the method can also preferably include a preprocessing step for preprocessing the training data. Preferably, the training data is filtered, in particular with respect to the operating time of the electrolysis cell device. For example, only data sets of training data that have been detected or created within a predetermined nominal load range of the electrolysis cell device are further used. For example, only the following data sets of training data are further used in which the input parameters fluctuate less than a predetermined tolerance value and in particular describe the static state of the electrolysis cell device. Optionally, an electrolysis cell device of the type to be considered is preselected, which should be described by the degradation model and the reference model.
[0038] In the scope of an alternative embodiment, the learning unit or a programmer of the learning unit can preferably assign the input parameters to a reference model as reference input parameters, in particular if the influence is relatively low, or to a degradation model as degradation input parameters, in particular if the influence is relatively high, depending on the influence of the input parameters on the target parameters. Preferably, the calculation unit can create reference training data including reference input parameters and target parameters and degradation training data including degradation input parameters and target parameters from the training data. The learning unit can perform a machine learning process for creating a degradation model, preferably using the degradation training data set. The learning unit can perform another machine learning process for determining the reference model, preferably using the reference training data set.
[0039] The method may optionally also include a verification step within the scope of alternative embodiments. In the verification step, the value of the target parameter can be predicted with the aid of the determined degradation model and / or the determined reference model and the value can be compared with the test data. If the test data and the predicted value are sufficiently consistent, the degradation model and the reference model can be stored in the memory of the calculation unit 18. If the test data and the predicted value are not sufficiently consistent, the machine learning process and another machine learning process can preferably be repeated based on other machine learning models. Alternatively, multiple machine learning models known to the learning unit can be executed in sequence, and it is determined in the verification step: based on which machine learning model the most accurate degradation model and the most accurate reference model are created. When the machine learning process and another machine learning process are performed with the aid of linear regression or other machine learning models that are inherently separable with respect to input parameters, the degradation model and the reference model can have independent dependencies on these input parameters. When performing the machine learning process and another machine learning process, the degradation model and the reference model can be determined by alternating iterative steps. When performing the machine learning process and another machine learning process, the degradation model can be determined based on the reference model. When performing a machine learning process and another machine learning process, the learning unit can determine a common, in particular different from zero, base value for the degradation model and the reference model, which the calculation unit can use to determine the target parameter. When performing a machine learning process and another machine learning process, the learning unit can extract the degradation model from the superordinate model in order to determine the reference model.
[0040] In the scope of an alternative embodiment, the calculation unit can also evaluate the reference model and / or the degradation model in the evaluation step 34 based on the operating parameters determined by means of the at least one sensor 22 in order to obtain an actual value of the target parameter or a prediction of the target parameter. The calculation unit can preferably interpret the target parameter with respect to the degradation and / or degradation rate of the electrolysis cell arrangement. The method can preferably also include an output step in which the output unit outputs the target parameter and / or the target parameter interpretation with respect to the degradation and / or degradation rate.
Claims
1. A method for monitoring an electrochemical system, in particular a high-temperature electrochemical system, preferably a fuel cell system, wherein in at least one method step, the functionality of the electrochemical system is determined based on at least one machine learning process (14, 16), characterized in that The functionality is determined with the aid of a degradation model which is at least partially separated from a reference model of the electrochemical system.
2. The method according to claim 1, characterized in that The degradation model is created by means of the machine learning process (14), and the reference model is created by means of a further machine learning process (16).
3. The method according to any one of the preceding claims, characterized in that The degradation model and the reference model have independent dependencies on each other.
4. The method according to any one of the preceding claims, characterized in that The degradation model and the reference model are determined by alternating iterative steps.
5. The method according to any one of the preceding claims, characterized in that The degradation model is determined based on the reference model.
6. The method according to any one of the preceding claims, characterized in that For determining at least one operating parameter of the electrochemical system, the degradation model and the reference model have a common base value, in particular a value different from zero.
7. The method according to any one of the preceding claims, characterized in that The degradation model is extracted from the upper-level model to determine the reference model.
8. An electrochemical system, in particular a high-temperature electrochemical system, preferably a fuel cell system, having at least one computing unit (18) for carrying out the method according to any of the preceding claims.
