Method for estimating and / or predicting physical parameters at predefined local target regions of an electrochemical energy converter, method for generating at least one training data set, method for analyzing at least one state of an energy conversion system, computer program product for estimating and / or predicting physical parameters, computer program product for analyzing at least one state, energy conversion device and energy conversion system

Abstract

A method is provided for estimating at least one physical parameter or at least one target quantity (47) at local target areas (46) of an electrochemical energy converter (12), which comprises at least one functional module (14) and at least one measuring device (16), wherein the at least one measuring device (16) performs spatially resolved measurements at local measuring areas (44), the method comprising: - Measurement of at least one physical parameter or at least one output variable (45) of the electrochemical energy converter (12) by means of the at least one measuring device (16) at local measuring ranges (44), wherein the at least one output variable (45) is a physical parameter for the function of the electrochemical energy converter (12) and is accessible to direct measurement by means of the at least one measuring device (16); - Estimation of at least one physical parameter or at least one target quantity (47) at predefined local target areas (46) based on at least one machine learning model and the at least one output quantity (45), wherein predefined local target areas (46) are different from local measurement areas (44), and wherein the at least one target quantity (47) is a physical parameter for the function of the electrochemical energy converter (12) and the at least one target quantity (47) is not accessible to direct measurement.

Description

[0001] The invention relates to a method for estimating and / or predicting at least one physical parameter or at least one target variable at predetermined local target areas of an electrochemical energy converter, which comprises at least one functional module with a plurality of electrochemical functional units, and at least one measuring device, wherein the at least one measuring device performs spatially resolved measurements at local measuring areas.

[0002] Furthermore, the invention relates to a method for generating at least one training data set for training a machine learning model for carrying out the method for estimating and / or predicting at least one physical parameter or physical target quantity at predetermined local target areas of the electrochemical energy converter.

[0003] Furthermore, the invention relates to a method for training at least one machine learning model.

[0004] Furthermore, the invention relates to a method for analyzing at least one state, in particular an operating state, of an energy conversion system comprising at least one electrochemical energy converter.

[0005] Furthermore, the invention relates to a computer program product for estimating and / or predicting at least one target variable at the predetermined local target area of ​​the electrochemical energy converter.

[0006] Furthermore, the invention relates to a computer program product for analyzing at least one state, in particular an operating state, of an energy conversion system comprising at least one electrochemical energy converter.

[0007] Furthermore, the invention relates to an energy conversion device.

[0008] Furthermore, the invention relates to an energy conversion system.

[0009] From EP 4 249 427 A1 a method for supporting robust hydrogen production from one or more renewable energy sources is known.

[0010] From DE 10 2023 130 159 A1 is a plant for generating an energy carrier from electric current, with a process model for predicting a corrected energy quantity for the plant, which includes a trained algorithm for determining a correction value based on previous values ​​of a calculated target energy quantity and an actual energy quantity.

[0011] From DE 10 2023 205 814 A1 a method and a computing unit are known for determining a state of at least one electrochemical cell of an electrochemical energy converter.

[0012] WO 2023 / 107710 A2 discloses a system and method for determining the state of a battery.

[0013] From “Monitoring of operational conditions of fuel cells by using machine learning”, Andip Babanrao Shrote et al., an approach to monitoring fuel cell stacks during testing is known, which is based on machine learning.

[0014] From “Lifelong performance monitoring of PEM fuel cells using machine learning models”, Lukas Klass, an approach is known to monitor the operation of fuel cell stacks on a test bench and thus ensure the proper execution of the tests.

[0015] In existing electrochemical energy converters, such as electrolysis cells and / or electrolysis cell stacks and / or fuel cells and / or fuel cell stacks and / or electrical energy storage devices, especially batteries, physical parameters that characterize the function of the electrochemical energy converter can be measured using measuring devices. Such physical parameters can include, for example, electric current density and / or electric current density distribution. Measurement ranges where physical parameters are measured often deviate from target ranges. Likewise, the parameter values ​​measured at measurement ranges deviate from the parameter values ​​actually present at the local target ranges. Target ranges are, in particular, ranges that are essential for the function of an electrochemical energy converter.This includes, for example, an active area on a membrane and / or a catalytic layer where an electrochemical reaction takes place. Such target areas are not accessible to direct measurement, which means that physical parameters measured at a measurement area may differ from the actual parameters at the target areas, thus preventing a realistic quantification of the actual values ​​of parameters at the target areas.

[0016] The object of the present invention is to provide a method that allows, in a simple manner, an improved estimation and / or prediction of functional parameters of an electrochemical energy converter, which are not accessible to direct measurement.

[0017] This problem is solved according to the invention in a method of the type mentioned at the outset by the fact that the method comprises: - Measurement of at least one physical parameter or at least one output variable of the electrochemical energy converter by means of the at least one measuring device at local measuring ranges, wherein the at least one output variable is a physical parameter for the function of the electrochemical energy converter and is accessible to direct measurement by means of the at least one measuring device; - Estimation and / or prediction of at least one physical parameter or at least one target quantity at predetermined local target areas by means of a computational device based on at least one machine learning model and the at least one output quantity, wherein predetermined local target areas are different from local measurement areas, and wherein the at least one target quantity is a physical parameter for the function of the electrochemical energy converter and the at least one target quantity is not accessible to direct measurement.

[0018] The electrochemical energy converter is in particular an electrolysis cell and / or an electrolysis cell stack and / or a fuel cell and / or a fuel cell stack and / or an electrical energy storage device, in particular a battery.

[0019] First, at least one output parameter of the electrochemical energy converter is measured using at least one measuring device. The measurement is performed with spatial resolution at local measurement points.

[0020] Of particular interest are physical parameters at predefined local target areas that differ from local measurement areas. Physical parameters at local target areas of the electrochemical energy converter are, in particular, target quantities that represent physical parameters for the function of the electrochemical energy converter at these local target areas. These parameters are not accessible to direct measurement. Therefore, using the method according to the invention, it is possible to estimate and / or predict target quantities based on at least one machine learning model and at least one output quantity using a computational device. This contributes to an improved quantification of the values ​​of physical parameters at the predefined local target areas. Thus, it is possible to provide realistic values ​​of physical parameters at local target areas that are not accessible to direct measurement.

[0021] Machine learning is a branch of artificial intelligence. It involves learning from existing data and / or datasets and adapting to new data. This process generates insights, which can then be used to derive trends, predictions, and estimates. Machine learning is frequently employed to process large datasets and extract influencing factors and patterns that can affect desired predictions and estimates. These influencing factors and patterns are often inaccessible to the user but contribute to more precise predictions.

[0022] It is necessary to train machine learning models. Training data forms the basis for this training and can be determined experimentally, for example. During training, a learning algorithm and / or training algorithm maps predefined training data onto a mathematical model. After training, the solution path found, particularly in the form of at least one training data set, is stored in the machine learning model. The machine learning model trained in this way can make predictions and / or estimates for new data and generate recommendations and / or decisions. During training, the learning algorithm and / or training algorithm builds a model and adjusts the parameters so that the model's results solve the given task as effectively as possible. The relationship of the training data to the given task is specifically represented by the machine learning model.In the present method according to the invention, the objective is in particular to estimate and / or predict at least one target parameter at the specified local target area.

[0023] In the method according to the invention, the estimation and / or prediction of at least one target variable at predetermined local target areas of the electrochemical energy converter is based on at least one machine learning model and at least one output variable, which is measured with spatial resolution by means of at least one measuring device at local measuring areas of the electrochemical energy converter.

[0024] In an advantageous embodiment of the method according to the invention, the method has at least one of the following: - local measuring areas are assigned to at least one measuring device; - local measuring ranges are assigned to at least one measuring device; - local measuring areas are directly assigned to at least one measuring device; - local measuring ranges are directly assigned to at least one measuring device; - Each measuring device is assigned a local measuring range; - One of the measuring devices is assigned a local measuring range.

[0025] Measurements can preferably be taken in the immediate vicinity of the at least one measuring device. The position of the at least one measuring device within the electrochemical energy converter advantageously defines a local measuring range.

[0026] It is advantageous if the procedure includes at least one of the following: - Predefined local target areas are assigned to at least one functional module of the electrochemical energy converter; - At least one functional module of the electrochemical energy converter is assigned predefined local target areas; - Predefined local target areas are directly assigned to at least one functional module of the electrochemical energy converter; - At least one functional module of the electrochemical energy converter is directly assigned to predefined local target areas; - Each functional module is assigned a predefined local target area; - A predefined local target area is assigned to a functional module; - Each electrochemical functional unit is assigned a predefined local target area; - A predefined local target area is assigned to an electrochemical functional unit.

[0027] Predefined local target areas are, in particular, areas that are especially relevant for the function of the electrochemical energy converter. Examples include an active area on a membrane and / or a catalytic layer where an electrochemical reaction takes place.

[0028] Predefined local target areas are arranged in particular within a functional module of the electrochemical energy converter.

[0029] A functional module comprises, in particular, a plurality of electrochemical functional units. An electrochemical functional unit is, in particular, a bipolar plate and / or an electrode array and / or an electrode and / or an end plate and / or a porous transport layer (PTL) and / or a catalyst-coated membrane (CCM) and / or a gas diffusion layer (GDL) and / or a proton exchange membrane (PEM).

[0030] It is advantageous if the at least one output quantity and the at least one target quantity are the same physical quantity. This allows for a direct comparison of the at least one output quantity with the at least one target quantity.

[0031] It is also fundamentally possible that the at least one output quantity and the at least one target quantity are different physical quantities.

[0032] In an advantageous embodiment of the method according to the invention, the electrochemical energy converter extends in one direction, wherein the at least one measuring device and the at least one functional module follow one another in this direction, and wherein the at least one functional module is effectively contacted by the at least one measuring device. This allows for the measurement of at least one output quantity in the electrochemical energy converter. Furthermore, this allows for an improvement in the resolution of the measured parameters.

[0033] Effective measurement means, in particular, that measurement data can be directly captured by the at least one measuring device.

[0034] The local measurement range and the specified local target range differ from each other, particularly with regard to the direction of extension of the electrochemical energy converter.

[0035] It is advantageous if the at least one measuring device comprises a first plate and a second plate, and at least one sensor device with at least one sensor is arranged between the first plate and the second plate, wherein at least the first and the second plate have a contact surface for effective contacting of at least one functional module of the electrochemical energy converter, wherein the at least one sensor device is effectively connected to the respective contact surface, and wherein the method comprises at least one of the following: - a segmental division of the first plate and the second plate into predefined area areas; - Each predefined area is assigned a measurement segment which is set up to record at least one physical parameter or at least one output variable; - Each measuring segment is assigned a sensor device.

[0036] In addition to spatial resolution along the direction of extension of the electrochemical energy converter, spatial resolution across the cross-section of the electrochemical energy converter is also possible. This spatial resolution across the cross-section corresponds, in particular, to measured values ​​acquired within a measurement segment of the at least one measuring device. For example, segment-wise spatial resolution across the cross-section is thus possible. This makes it possible to map influencing factors acting across the cross-section of the electrochemical energy converter. One such influencing factor is, for example, gravity.

[0037] The cross-section is, in particular, a plane perpendicular to the direction of extension of the electrochemical energy converter.

[0038] Preferably, the first plate and the second plate of the at least one measuring device are divided into segments. This allows, in particular, surface areas to be defined and measuring segments to be assigned to the defined surface areas.

[0039] Each measurement segment is assigned, for example, a sensor device.

[0040] Sensor devices are in particular evenly distributed over the cross-section of the at least one measuring device.

[0041] This makes it possible to determine a measured value, in particular an output value, for each measurement segment, thus enabling the assignment of individual measured values ​​to measurement segments. This allows for local spatial resolution, so that measured values, in particular at least one output value, can be displayed with local resolution across the cross-section.

[0042] The measurement segment-wise division can be transferred and / or projected onto the specified local target area, enabling the prediction and / or estimation of physical parameters or at least one target variable with respect to individual measurement segments. This allows for the acquisition and / or representation of a local, spatial resolution across the cross-section of the specified local target area.

[0043] The cross-section of the electrochemical energy converter has, in particular, a circular and / or rectangular contour.

[0044] It is advantageous if the procedure includes at least one of the following: - the at least one output variable is measured over time using the at least one measuring device; - at least one target variable is estimated and / or predicted over time using the calculation device.

[0045] This allows the representation and / or recording of a temporal sequence of at least one output variable and / or at least one target variable, so that, for example, information about the operating time of the electrochemical energy converter can be recorded.

[0046] Advantageously, the at least one output variable, which is a physical parameter for the function of the electrochemical energy converter and is accessible to direct measurement by means of the at least one measuring device, is at least one of the following: - an electrical current value; - an electrical current distribution value; - an electric current density distribution value; - an electrical voltage value; - a temperature value; - a temperature distribution value.

[0047] The values ​​listed above characterize, in particular, the function of the electrochemical energy converter. These are especially essential for the operation of the electrochemical energy converter.

[0048] It is advantageous if the at least one target variable, which is a physical parameter for the function of the electrochemical energy converter and is not accessible to direct measurement, is at least one of the following: - an electrical current value; - an electrical current distribution value; - an electric current density distribution value; - an electrical voltage value; - a temperature value; - a temperature distribution value.

[0049] The listed values ​​are particularly characteristic of the function of the electrochemical energy converter.

[0050] In contrast to the at least one output variable, the at least one target variable is not directly accessible to measurement by the at least one measuring device. Due to various influencing factors, such as flowing process media and / or components of the electrochemical energy converter, the value of the at least one target variable can deviate from the value of the at least one output variable. Components can be individual electrochemical functional units of the at least one functional module.

[0051] It is advantageous if the procedure includes at least one of the following: - at least one machine learning model accesses at least one training dataset; - at least one training dataset is provided to the at least one machine learning model.

[0052] The at least one machine learning model learns in particular from data, especially training data, in order to adapt to new datasets.

[0053] This makes it possible to gain insights and subsequently derive developments, predictions, and / or estimates from them. Machine learning models are trained, in particular, by inputting datasets, especially training datasets, to find and analyze patterns. Based on this, predictions and / or estimates can then be made.

[0054] In particular, the insights gained from the training data are integrated into the parameters of at least one machine learning model after training.

[0055] In particular, at least one physical parameter or at least one target variable can be estimated and / or predicted at predefined local target areas based on at least one machine learning model trained on training data sets.

[0056] Ideally, at least one machine learning model is based on at least one of the following: - artificial neural networks, especially multi-layer perceptron (MLP); - Regression algorithms; - Clustering algorithms, especially K-means clustering and / or hierarchical clustering.

[0057] Neural networks are used particularly in machine learning. They enable predictions and / or estimations based on datasets containing a large amount of data. This data cannot be described and analyzed using conventional analytical rules.

[0058] In particular, this can include measurement data, for example at least one output variable.

[0059] Artificial neural networks are primarily inspired by the networks formed by biological neurons in the brain. Biological neurons are interconnected and organized in layers. They can summate multiple input signals and only pass a signal on to other neurons when the sum of the input signals reaches a threshold. Artificial neural networks are formed by artificial neurons. These artificial neurons, in particular, emulate selected properties of biological neurons using mathematical methods. Typically, an artificial neural network consists of several layers of artificial neurons. Signals, especially at least one output signal, travel from a first layer (an input layer) to a final layer (an output layer), potentially passing through several intermediate layers (hidden layers).Each layer can transform the signals at its inputs differently. An artificial neural network with many hidden layers can decompose a complex task, based on a large amount of data, into several simpler tasks, each executed in different layers of the machine learning model. To make the most accurate predictions and / or estimates possible, artificial neural networks must be trained and / or calibrated using training data.

[0060] A multi-layer perceptron (MLP) is an artificial neural network consisting of multiple layers of artificial neurons. The artificial neurons in a multi-layer perceptron (MLP) typically use non-linear activation functions, enabling the network to learn complex patterns in data and / or datasets. In particular, multi-layer perceptrons are well-suited for identifying non-linear relationships in data and / or datasets and for performing classification, regression, and / or pattern recognition.

[0061] Regression algorithms are based, in particular, on a statistical approach to analyzing the relationship between variables. The goal is to identify the most suitable function that characterizes the relationship between these variables. Furthermore, the goal is to find the best-fitting model that can be used to generate predictions and / or estimates. This is specifically a machine learning model used to predict the value of variables, especially at least one output variable, particularly when applying the machine learning model to new, unknown data, specifically at least one target variable. Such a machine learning model, in particular, models the relationship between the at least one output variable and the at least one target variable, thus enabling the estimation and / or prediction of numerical values ​​of the at least one target variable.

[0062] Clustering algorithms are used in particular to find similarity structures in large datasets.

[0063] In k-means clustering, cluster centers are randomly determined and the sum of the squared Euclidean distances of the objects to their nearest cluster center is minimized.

[0064] Hierarchical clustering is based on a distinction between a finest and a coarsest partition. The coarsest partition corresponds to the entirety of all elements, specifically all output and / or target variables, while the finest partition contains only a single element, specifically a single output and / or a single target variable. Clusters can then be formed by partitioning and grouping these elements. Once formed, clusters cannot be dissolved, nor can individual elements be exchanged.

[0065] Advantageously, the procedure includes at least one of the following: - which is at least one output variable and / or is stored in a storage device; - which is at least one training data set and / or is stored in the storage device; - the storage device provides at least one output variable to the computing device; - the storage device provides the computing device with at least one training data set; - the computing device accesses at least one output variable on the storage device; - The computing device accesses at least one training data set on the storage device.

[0066] This allows for a data-effective connection of the data and / or datasets necessary for estimating and / or predicting the at least one target variable, in particular the at least one output variable and / or the at least one training dataset. "Data-effective" means, in particular, that data, for example, at least one output variable and / or at least one training dataset, can be transferred and processed.

[0067] It is particularly advantageous if the calculating device has a time calculation unit that calculates the temporal progression of at least one of the following: - which has at least one output variable; - at least one target variable.

[0068] This allows for a time-resolved representation of the at least one output variable and / or the at least one target variable. In particular, this makes it possible to depict the development of the at least one output variable and / or the at least one target variable over the operating time of the electrochemical energy converter.

[0069] Advantageously, the time history is stored on the storage device.

[0070] As mentioned at the outset, the invention relates to a method for generating at least one training data set for training a machine learning model for carrying out the method for estimating and / or predicting at least one physical parameter or physical target quantity at predetermined local target areas of an electrochemical energy converter, which comprises at least one functional module with a plurality of electrochemical functional units, and at least one measuring device, wherein the at least one measuring device performs spatially resolved measurements at local measuring areas, wherein the method for generating at least one training data set comprises: - Applying at least one physical training parameter or at least one training target variable to predetermined local target areas of the electrochemical energy converter, wherein the at least one training target variable is a physical parameter for the function of the electrochemical energy converter; - Measurement of at least one physical training parameter or at least one training output variable by means of at least one measuring device at local measuring ranges of the electrochemical energy converter, wherein predetermined local target ranges are different from local measuring ranges, and wherein the at least one training output variable is a physical parameter for the function of the electrochemical energy converter.

[0071] As mentioned above, the training of the at least one machine learning model is based on at least one training dataset. For the estimation and / or prediction of at least one physical parameter or at least one target variable at predefined target ranges, it is particularly necessary to establish a relationship between physical parameters at the predefined local target range and physical parameters at the local measurement range. This relationship is specifically reflected in a training dataset used to train the at least one machine learning model. For this purpose, the predefined local target range of the electrochemical energy converter is first subjected to a physical training parameter or at least one training target variable.

[0072] This type of testing should be carried out particularly in the run-up to the commissioning of the electrochemical energy converter. During operation of the electrochemical energy converter, the at least one generated training data set can then be accessed.

[0073] It is advantageous to generate a training dataset for different predefined spatial target areas. For example, if a predefined spatial target area is assigned to the first functional unit of a functional module, a first training dataset is created for this first predefined spatial target area. If a predefined spatial target area is assigned to the second functional unit of a functional module, a second training dataset is created for this second predefined spatial target area. The same applies to any further possible predefined spatial target areas.

[0074] The at least one training target variable can be freely selected, and it is also possible to apply different training target variables to predefined local target areas in order to generate the most comprehensive training data set possible.

[0075] In particular, a physical training parameter or at least one training output variable is measured at local measurement areas using at least one measuring device. These local measurement areas differ from predefined local target areas, especially with respect to the direction of extension of the electrochemical energy converter.

[0076] To simplify the provision of the training dataset, it is advantageous if the procedure includes at least one of the following: - at least one training output variable measured by means of the at least one measuring device forms the at least one training data set; - at least one training target variable forms the at least one training data set; - Storing the at least one training output variable measured by means of the at least one measuring device in a training data set; - Storing at least one training target variable in a training data set; - Storing at least one training data set in the storage device; - at least one training data set is provided to the computing device; - The computing device accesses at least one training data set.

[0077] Ideally, at least one training target parameter should be at least one of the following: - an electrical current value; - an electrical current distribution value; - an electric current density distribution value; - an electrical voltage value; - a temperature value; - a temperature distribution value.

[0078] The values ​​listed above characterize, in particular, the function of the electrochemical energy converter. These are especially essential for the operation of the electrochemical energy converter.

[0079] For the same reason, it is advantageous if at least one training output variable is at least one of the following: - an electrical current value; - an electrical current distribution value; - an electric current density distribution value; - an electrical voltage value; - a temperature value; - a temperature distribution value.

[0080] It is advantageous to apply at least one physical training parameter or at least one training target variable to adjacent predefined areas, where the values ​​of this parameter or target variable differ between adjacent areas. This allows for consideration of influencing factors that adjacent areas have on one another.

[0081] It is advantageous to divide predefined local target areas into predefined surface areas segment by segment, with each surface area being subjected to at least one physical training parameter or at least one training target variable. This allows for local spatial resolution, enabling a locally resolved representation of training parameters relative to the cross-section of the electrochemical energy converter.

[0082] As mentioned at the outset, the invention relates to a method for training a machine learning model, wherein the training is carried out using training data that was generated using the method for generating at least one training data set.

[0083] The method according to the invention has the advantages already explained in connection with the other methods according to the invention.

[0084] As mentioned at the outset, the invention relates to a method for analyzing at least one state, in particular an operating state, of an energy conversion system comprising at least one electrochemical energy converter, wherein the at least one electrochemical energy converter comprises at least one functional module with a plurality of electrochemical functional units, and at least one measuring device, and wherein the at least one measuring device performs spatially resolved measurements at local measuring areas to detect at least one physical parameter or at least one output variable of the electrochemical energy converter, wherein the at least one output variable is a physical parameter for the function of the electrochemical energy converter, and wherein the method comprises: - Detection of at least one first state by means of a state detection device based on at least one physical parameter or at least one output variable; - Detecting at least one second state by means of the state detection device based on the at least one physical parameter or the at least one output variable, wherein the at least one second state is temporally subsequent to the at least one first state; - Determination of a current state, in particular a momentary current state, wherein the current state is at least a second state or constitutes at least a second state; - Calculation of the temporal evolution of the current state using a state calculation device; - Estimation and / or prediction of a target state using the state calculation device based on at least one machine learning model and the current state, in particular the temporal evolution of the current state; - Estimation and / or prediction of at least one limit value using the state calculation device based on at least one machine learning model and the target state; - Determination of a deviation between the actual state, in particular the current actual state, and the target state using a deviation calculation unit; - Issuance of a message, in particular an information message and / or warning message, by means of an output device, if the deviation reaches at least one limit value.

[0085] The energy conversion system consists in particular of an electrolyzer and / or a fuel cell system and / or a fuel cell power plant and / or an electrical energy storage system, in particular a battery system.

[0086] The state monitoring device can be used to record at least one state of the energy conversion system.

[0087] It is advantageous if at least one first state is based on at least one output variable that can be measured by means of at least one measuring device. The at least one first state is, for example, a state at a time when the operation of the energy conversion system was initiated.

[0088] Advantageously, at least one second state is based on the at least one output variable, which can be measured by the at least one measuring device. The at least one second state is, in particular, temporally subsequent to the at least one first state. The at least one second state is, in particular, detectable over time. This allows, for example, the detection of at least one second state at predetermined time intervals. Thus, it is possible to represent the temporal progression of at least one second state.

[0089] In particular, at least one second state constitutes an actual state. The actual state is, for example, the second state at the current time.

[0090] Advantageously, a time-based progression is calculated from current states.

[0091] Based on at least one machine learning model and the current state, in particular the temporal evolution of the current state, a target state of the energy transition system can preferably be estimated and / or predicted.

[0092] The target state is, in particular, a state that is predicted by the machine learning model based on the previous temporal progression and compared with the actual values.

[0093] It is particularly advantageous if an estimation and / or prediction of at least one limit value is performed based on the at least one machine learning model and the target state. Specifically, the at least one limit value is not predetermined and / or fixed, but is dynamically adjusted based on the data available to the at least one machine learning model.

[0094] It is advantageous if a deviation between the actual state and the target state is determined, so that a message, for example a warning message and / or information message, is issued when the determined deviation reaches at least a limit value.

[0095] This makes it possible to record actual states that deviate from the target state and to take appropriate corrective measures.

[0096] For example, this enables demand-based maintenance (predictive maintenance) of components in the energy transition system. This reduces potential downtime and / or maintenance periods. The main goal of predictive maintenance is to create the most precise possible maintenance schedule and prevent unexpected system failures. Knowing when which components and / or parts need servicing allows for better planning of maintenance resources such as spare parts or personnel. Furthermore, system availability can be increased by converting unplanned downtime into planned downtime. Additional benefits include a potentially longer system lifespan, increased system safety, fewer accidents with negative environmental impacts, and optimized spare parts management.

[0097] It is advantageous if the energy conversion system includes at least one operating device which is controlled by means of control parameters and / or regulated by means of control parameters, wherein at least one sensor device provides control parameters and / or control parameters, and wherein the method includes at least one of the following: - the detection of at least one first state by means of the state detection device is based on at least one control parameter and / or at least one regulation parameter; - the detection of at least one second state by means of the state detection device is based on at least one control parameter and / or at least one regulation parameter, wherein the at least one second state is temporally subsequent to the at least one first state.

[0098] In addition to the at least one output variable, which is measured, for example, by means of at least one measuring device, other parameters can also be considered when applying predictive maintenance. These include, for example, control parameters and / or regulation parameters that serve to control and / or regulate operating equipment.

[0099] Operating equipment includes, in particular, equipment necessary for the operation of the energy conversion system. This includes, in particular, devices for transporting and / or storing process media and / or devices for recording environmental parameters.

[0100] It is particularly advantageous if the control parameters and / or regulation parameters are at least one of the following: - physical parameters that characterize the function of at least one operating device; - Environmental parameters, especially external environmental parameters.

[0101] Physical parameters that characterize the function of at least one operating device include, for example, pressure parameters and / or level parameters and / or leakage parameters and / or temperature parameters and / or flow parameters and / or state parameters, in particular of valves (open / closed).

[0102] In particular, the physical parameters listed above relate to operating media and / or auxiliary media, especially hydrogen and / or oxygen and / or water.

[0103] Environmental parameters are in particular external environmental parameters, especially meteorological parameters and / or parameters that characterize the energy demand of an electrical network.

[0104] Based on environmental parameters, it is possible to control the operation of the energy conversion system in a targeted manner.

[0105] The energy conversion system, in particular an electrolyzer, requires electrical current for operation, so that possible operating times, for example when there is a particularly large amount of electricity based on renewable energy, especially photovoltaics, can be predicted and / or estimated using meteorological parameters and at least one machine learning model.

[0106] Alternatively, electricity can also be provided by means of an energy conversion system, in particular a fuel cell power plant and / or a fuel cell system. In this case, it is advantageous if operating times can be predicted and / or estimated based on energy demands and at least one machine learning model, so that appropriate preparatory measures for operation can be taken in advance.

[0107] It is advantageous if at least one machine learning model is based on at least one of the following: - artificial neural networks; - Regression algorithms; - Clustering algorithms, especially k-means clustering and / or hierarchical clustering.

[0108] The advantages of machine learning models have already been explained in connection with the inventive method for estimating and / or predicting at least one physical parameter or target variable at predetermined local target areas of the electrochemical energy converter. Reference is therefore made to the corresponding explanations.

[0109] As mentioned at the outset, the invention relates to a computer program product for estimating and / or predicting at least one target quantity, according to the inventive method for estimating and / or predicting at least one physical parameter or target quantity at predetermined local target areas of the electrochemical energy converter, wherein the computer program product is configured to perform at least one of the following steps: - Creating at least one measurement data set, which is formed from at least one physical parameter or at least one output quantity, which is measured by at least one measuring device; - Storing at least one measurement data set in a storage device; - Providing at least one measurement data set and at least one machine learning model to the computing device; - Creating at least one target data set, which is formed from at least one physical parameter or at least one target variable that is estimated and / or predicted by the computing device; - Storing at least one target data record in the storage device.

[0110] Using the computer program product according to the invention, the inventive method for estimating and / or predicting at least one physical parameter or target quantity at predetermined local target areas of the electrochemical energy converter can be carried out.

[0111] The computer program product is installed on the storage device, for example.

[0112] To easily estimate and / or predict at least one target variable, it is advantageous if the computer program is configured to perform at least one of the following steps: - Comparison of at least one measurement data set with at least one target data set; - Create at least one comparison data set, which is formed from the comparison of at least one measurement data set with at least one target data set.

[0113] For example, at least one comparison data set serves to further refine and / or train the machine learning model.

[0114] In particular, at least one comparison data set includes a temporal comparison, for example between physical parameters that are recorded at different times.

[0115] For simple monitoring, especially by a user, it is advantageous if the computer program product is configured to perform at least one of the following steps: - Visualization of at least one measurement data set; - Visualization of at least one target data set; - Visualization of at least one comparison data set.

[0116] At least one measurement data set and / or at least one target data set and / or at least one comparison data set can be visualized, for example, on an output device. This allows a user to visually evaluate the acquired data sets. Specifically, the visualization consists of graphs and / or tables and / or representations of values ​​from the at least one measurement data set and / or the at least one target data set and / or the at least one comparison data set projected onto an image of the electrochemical energy converter.

[0117] This allows, for example, a visualization of the distribution of physical parameters with respect to the direction of extension and across the cross-section of the electrochemical energy converter. In particular, this enables a spatially resolved three-dimensional visualization of physical parameters.

[0118] As mentioned at the outset, the invention relates to a computer program product for analyzing at least one state, in particular an operating state, of an energy conversion system according to the inventive method for analyzing at least one state, in particular an operating state, of an energy conversion system comprising at least one electrochemical energy converter, wherein the at least one electrochemical energy converter comprises at least one functional module with a plurality of electrochemical functional units, and at least one measuring device, and wherein the at least one measuring device performs spatially resolved measurements at local measuring areas to detect at least one physical parameter or at least one output variable of the electrochemical energy converter, wherein the at least one output variable is a physical parameter for the function of the electrochemical energy converter, wherein the computer program product is configuredperform at least one of the following steps: - Creating and storing at least one initial state data record based on at least one physical parameter or at least one output variable; - Creating and storing at least one second state data set based on at least one physical parameter or at least one output variable; - Creating and storing at least one current state record, wherein the at least one current state record is formed from or is a current second state record; - Creating and saving the time history of at least one current state data record; - Creating and storing at least one target state data set based on at least one machine learning model and at least one actual state data set, in particular the temporal evolution of the at least one actual state data set; - Creating and storing at least one threshold data set based on at least one machine learning model and at least one target state data set; - Creating and saving a deviation data record based on a comparison of at least one actual state data record with at least one target state data record; - Create and store at least one comparison data set based on a comparison of at least one deviation data set with at least one limit data set; - Issuance of a message, in particular an information message and / or warning message, when a predefined value of the deviation data set reaches a predefined value of at least one limit value data set.

[0119] The computer program product according to the invention allows the inventive method for analyzing at least one state, in particular an operating state, of an energy conversion system to be carried out.

[0120] The computer program product is, for example, installed on a storage device.

[0121] It is particularly advantageous, once the computer program product is set up, to perform at least one of the following steps: - Create and store at least one control parameter data set based on at least one control parameter; - Create and store at least one control parameter data set based on at least one control parameter.

[0122] Advantageously, the computer program product is configured to perform at least one of the following steps: - Visualization of at least one initial state data set; - Visualization of at least one second state data set; - Visualization of at least one current state data set; - Visualization of the temporal progression of at least one current state data set; - Visualization of at least one target state data set; - Visualization of at least one limit value data set; - Visualization of at least one deviation data set; - Visualization of at least one comparison data set.

[0123] Advantages of the computer program product according to the invention have already been explained in connection with the method for analyzing at least one state, in particular an operating state, of an energy conversion system. Reference is therefore made to the corresponding explanations.

[0124] As mentioned at the outset, the invention relates to an energy conversion device which is configured to perform the inventive method for estimating and / or predicting at least one physical parameter or at least one target variable at predetermined local target areas of an electrochemical energy converter, wherein the energy conversion device comprises at least one of the following: - an electrochemical energy converter with at least one functional module comprising a plurality of electrochemical functional units, and at least one measuring device for spatially resolved measurement of the at least one physical parameter or the at least one output quantity at the local measuring range of the electrochemical energy converter, wherein the at least one output quantity is a physical parameter for the function of the electrochemical energy converter and is accessible to direct measurement by means of the at least one measuring device; - a computational device for estimating and / or predicting the at least one physical parameter or the at least one target quantity at predetermined local target areas based on the at least one machine learning model and the at least one output quantity, wherein predetermined local target areas are different from local measurement areas, and wherein the at least one target quantity is a physical parameter for the function of the electrochemical energy converter and the at least one target quantity is not accessible to direct measurement.

[0125] The electrochemical energy converter is in particular an electrolysis cell and / or an electrolysis cell stack and / or a fuel cell and / or a fuel cell stack and / or an electrical energy storage device, in particular a battery.

[0126] It is particularly advantageous if the electrochemical energy converter extends in one direction and the at least one measuring device and the at least one functional module follow one another in the direction of extension, and wherein the at least one measuring device effectively contacts the at least one functional module.

[0127] It is advantageous if the at least one measuring device comprises a first plate and a second plate and a plurality of measuring segments, wherein at least one sensor device with a plurality of sensors is arranged between the first plate and the second plate, wherein at least the first and the second plate have a contact surface for effective contacting of at least one functional module of the electrochemical energy converter, wherein the at least one sensor device is effectively connected to the respective contact surface, and wherein at least the first plate and the second plate are or are divided into predetermined surface areas, wherein measuring segments comprise at least one of the following: - a measuring segment is assigned to a predefined area; - A sensor device is assigned to each measuring segment.

[0128] The at least one measuring device is, in particular, a measuring device as disclosed in the German patent application with file number 10 2024 111 628.5, which has not been previously published. Reference is made in full to the patent application with file number 10 2024 111 628.5 with regard to an advantageous embodiment of the at least one measuring device.

[0129] The at least one measuring device may alternatively or additionally be a measuring device as known from patent specification DE 103 16 117 B3, to which full reference is made.

[0130] The energy conversion device according to the invention further exhibits the advantages already explained in connection with the method according to the invention for estimating and / or predicting at least one physical parameter or target variable at predetermined local target areas of the electrochemical energy converter. Reference is made to the corresponding explanations in this respect.

[0131] As mentioned at the outset, the invention relates to an energy conversion system which is equipped to carry out the method for analyzing at least one state, in particular an operating state, of the energy conversion system, wherein the energy conversion system comprises at least one of the following: - an energy conversion device according to the invention; - an operating device which is controlled by means of control parameters and / or regulated by means of control parameters; - a sensor device that provides control parameters and / or regulation parameters; - a condition monitoring device for recording at least one condition, in particular operating condition, of the energy conversion system; - a state calculation device; - a deviation calculation unit; - an output device.

[0132] The energy conversion system according to the invention further exhibits the advantages already explained in connection with the method according to the invention for estimating and / or predicting at least one physical parameter or target variable at predetermined local target areas of the electrochemical energy converter. Reference is made to the corresponding explanations in this respect.

[0133] The energy conversion system consists in particular of an electrolyzer and / or a fuel cell system and / or a fuel cell power plant and / or an electrical energy storage system, in particular a battery system.

[0134] The following description of preferred embodiments, in conjunction with the drawings, serves to further explain the invention. The drawings show: Fig. 1: a schematic representation of an energy conversion device according to the invention; Fig. 2: an exploded view of an embodiment of an electrochemical energy converter with a functional module and a measuring device; Fig. 3: a sectional view of the electrochemical energy converter made of Fig. 2; Fig. 4: an exploded view of another embodiment of an electrochemical energy converter with several functional modules and a measuring device; Fig. 5: a sectional view of the electrochemical energy converter made of Fig. 4; Fig. 6: a perspective view of an embodiment of an electrochemical energy converter according to the invention, wherein a first housing part of the electrochemical energy converter is hidden; Fig. 7: a top view of the measuring device; Fig. 8: a top view of the measuring device, with measured output quantities projected onto the measuring device; Fig. 9A: a schematic representation in which output quantities measured on the cross-section of the electrochemical energy converter are projected onto a local measuring area; Fig. 9B: a schematic representation in which estimated and / or predicted output quantities are projected onto the cross-section of the electrochemical energy converter at a predetermined local target area; Fig. 10: a schematic representation of parameter processing using the method according to the invention, in which output variables at the local measuring point and target variables at the predetermined local target point are projected onto a cross-section of the electrochemical energy converter; Fig. 11: a schematic representation of time-resolved parameters; Fig. 12: a schematic representation of the interaction of the energy conversion device and / or an energy conversion system as well as at least one machine learning model and one computer program product; Fig. 13: a schematic representation of an energy conversion system according to the invention.

[0135] In Fig. Figure 1 is a schematic representation of a preferred embodiment of an energy conversion device according to the invention, which is designated overall by reference numeral 10. The same reference numerals are used below for identical or similar components.

[0136] The energy conversion device 10 comprises an electrochemical energy converter 12 with one or more functional modules 14 and a measuring device 16.

[0137] The energy conversion device 10 also includes a calculation device 18 and a storage device 20.

[0138] The calculation device 18 preferably includes a time calculation unit 19 for calculating the temporal progression of parameters.

[0139] The electrochemical energy converter 12, the computing device 18 and the storage device 20 are interconnected in a data-effective manner, making it possible to transfer data between the electrochemical energy converter 12, the computing device 18 and the storage device 20.

[0140] The energy conversion device 10 has interfaces 22, whereby data and / or signals, in particular control signals and / or regulation signals, are provided and / or received via the interfaces.

[0141] In the Fig. 2 and Fig. Figure 3 is a schematic representation of an embodiment of an electrochemical energy converter 12.

[0142] As already mentioned, the electrochemical energy converter comprises a measuring device 16 and a functional module 14.

[0143] Functional module 14 comprises a plurality of electrochemical functional units 24.

[0144] The functional module 14 is in particular an electrode arrangement 26 and / or an electrode with a bipolar plate 28.

[0145] Electrochemical functional units of the functional module 14 are, for example, a membrane electrode arrangement 30, a gas diffusion layer 32 and / or an intermediate layer 34.

[0146] For example, a functional module 14, in particular the electrode arrangement 26, comprises several electrochemical functional units 24, in particular a membrane electrode arrangement 30, each with a gas diffusion layer 32 arranged on both sides of the membrane electrode arrangement 30 and an intermediate layer 34.

[0147] The measuring device 16 preferably comprises the bipolar plate 28 or preferably acts as a bipolar plate 28. It is therefore possible to replace a bipolar plate 28 of the electrochemical energy converter 12 with a measuring device 16, so that the functions of the bipolar plate 28 are integrated into the measuring device 16.

[0148] Alternatively, it is also possible to integrate the measuring device 16 into the electrochemical energy converter 12.

[0149] The electrochemical energy converter 12 further comprises a housing 36 with a first housing part 38 and a second housing part 40.

[0150] It is intended that the first and second housing parts 38, 40 are mechanically connected to each other by means of connecting elements 42.

[0151] The functional module 14 with electrochemical functional units 24 and the measuring device 16 are arranged between the first housing part 38 and the second housing part 40.

[0152] The electrochemical energy converter 12 extends in a direction 12a.

[0153] The measuring device 16 and the functional module 14 follow each other in the extension direction 12a.

[0154] The measuring device 16 and the functional module 14 are in a measuring-effective connection.

[0155] The measuring device 16 is set up to measure at least one physical parameter or at least one output quantity 45 with spatial resolution at a local measuring area 44.

[0156] Preferably, the local measuring area 44 of the measuring device 16 is directly assigned.

[0157] The at least one output quantity 45 is accessible to direct measurement by means of the measuring device 16.

[0158] The at least one output variable 45 is, for example, a physical parameter for the function of the electrochemical energy converter 12. In particular, the at least one output variable 45 is an electric current value and / or an electric current distribution value and / or an electric current density distribution value and / or an electric voltage value and / or a temperature value and / or a temperature distribution value.

[0159] Functional module 14 includes a predefined local target area 46.

[0160] Specified local target areas 46 are in particular directly assigned to an electrochemical functional unit 24 of the functional module 14.

[0161] The 46 physical parameters occurring at local target areas are not accessible to direct measurement.

[0162] Specified local target areas 46 are, in particular, areas that are essential for the function of an electrochemical energy converter 12. These include, for example, an active area of ​​the membrane-electrode assembly 30 where an electrochemical reaction takes place.

[0163] With respect to the extension direction 12a of the electrochemical energy converter 12, the positions of the local measuring range 44 and the specified local target range 46 are different from each other.

[0164] The output quantities 45 measured by the measuring device 16 at the local measuring range 44 deviate, sometimes considerably, from the target quantities 47 actually prevailing at predetermined local target ranges 46. This deviation arises in particular due to the influence of various influencing factors. Examples of influencing factors are flowing process media and / or components of the electrochemical energy converter 12, which are arranged between the local measuring range 44 and the local measuring range 46 and whose material and / or surface properties can lead to a deviation between output quantities 45 and target quantities 47.

[0165] In the present embodiment, the gas diffusion layer 32 and the intermediate layer 34 are arranged between the measuring device 16, to which the local measuring range 44 is assigned, and the membrane electrode assembly 30, to which the local target range 46 is assigned. The gas diffusion layer 32 and the intermediate layer 34 must be penetrated and / or passed through by physical parameters of the electrochemical energy converter 12. This can lead to distortions between the measured output quantities 45 at the local measuring range 44 and the actual target quantities 47 occurring at the specified local target range 46.

[0166] Of particular interest for the function of the electrochemical energy converter 12, however, are the actual physical parameters occurring at the specified local target area 46.

[0167] As already explained, physical parameters that actually occur at the specified local target area 46 are not accessible to direct measurement. For this reason, these parameters are estimated and / or predicted using the method according to the invention, based on at least one machine learning model and at least one output variable 45, which is measured by the measuring device 16, in order to achieve the most realistic representation possible by means of the computing device 18.

[0168] The method according to the invention and the at least one machine learning model are described below.

[0169] The measuring device 16 is specifically designed as a so-called "segmented plate". The measuring device 16 is, in particular, a measuring device as disclosed in the German patent application with file number 10 2024 111 628.5, which has not been previously published. Reference is made in full to the patent application with file number 10 2024 111 628.5 regarding an advantageous embodiment of the measuring device 16.

[0170] Alternatively or additionally, the measuring device 16 is designed as is known from patent specification DE 103 16 117 B3, to which full reference is made.

[0171] An installation situation of the electrochemical energy converter 12 is in Fig. Figure 6 is shown. For better clarity, the first housing part 38 of the electrochemical energy converter 12 has been hidden.

[0172] The measuring device 16 comprises a plurality of sensor devices 48, each of which includes at least one sensor 49. Sensors 49 are, for example, resistance measuring devices 50, and / or temperature-sensitive elements 52.

[0173] The measuring device 16 further comprises a first plate 54 and a second plate 56.

[0174] Furthermore, the measuring device 16 has a plurality of measuring segments 58.

[0175] The first plate 54 and the second plate 56 are joined together by material bonding and are in particular spaced apart, and in particular arranged parallel to each other.

[0176] Sensor devices 48 are arranged between the first plate 54 and the second plate 56.

[0177] Sensor devices 48 are intended for measuring various quantities, for example, measuring an electric current and / or an electric voltage and / or a temperature.

[0178] For this purpose, sensor devices 48 are effectively connected, in particular electrically conductively, to the first plate 54 and the second plate 56.

[0179] The first plate 54 and the second plate 56 are preferably made of an electrically conductive material.

[0180] The first plate 54 and the second plate 56 each have a partially circular contour, with a tab 62 arranged on opposite sections 60 for easier assembly. The tabs 62 are preferably integrally connected to the plates 54 and 56.

[0181] The first plate 54 and the second plate 56 alternatively have a rectangular contour, in particular a square contour.

[0182] The first plate 54 and the second plate 56 each have a contact surface 64, in particular a closed one, for measuring purposes, in particular electrical contacting of the functional module 14 of the electrochemical energy converter 12.

[0183] Sensor devices 48 are electrically connected to the contact surfaces 64 by means of a respective conductor device 66 and / or connection device 68.

[0184] The first plate 54 and the second plate 54 each also have a border region 70 and a central region 72. The central region 72 is enclosed by the border region 70. In particular, the border region 70 completely surrounds the central region 72.

[0185] Measuring segments 58 are in particular assigned to the respective central areas 70 of the plates 54, 56.

[0186] Central areas 72 are each specifically assigned sensor devices 48.

[0187] Central areas 72 form, or are in particular measurement-effective areas 73. This means that measurement-effective areas 73 are set up to measure physical parameters.

[0188] The local measuring range 44 preferably coincides with the measuring effective range 73.

[0189] The plates 54, 56 are bonded together by means of an adhesive 74. The adhesive 74 is in particular made of an electrically insulating material and / or a thermally conductive material.

[0190] The adhesive 56, 58 is or forms, for example, a filler 60 that fills a space 76 between the first plate 54 and the second plate 56 and thus contributes to the stability of the measuring device 16.

[0191] Furthermore, measuring segments 58 are each effectively connected, in particular electrically conductive, to the respective contact surfaces 64 of the plates 54, 56.

[0192] For example, measuring segments 58 are arranged in a structured grid. An arrangement in a structured grid means, in particular, that a grid of measuring segments 58 has a regular topology. For example, it can be an arrangement in a curved grid ( Fig. 7).

[0193] Alternatively or additionally, measuring segments 58 can also be arranged in a regular grid, so that a subdivision into axially parallel, rectangular areas is made.

[0194] Furthermore, the first plate 54 and the second plate 56 are or will be divided into predefined surface areas 78, with each measuring segment 58 being assigned to a predefined surface area 78. This results in a spatial division of the plates 54 and 56 into predefined surface areas 78 and an allocation of measuring segments 58 to predefined surface areas 78. In particular, exactly one measuring segment 58 is assigned to exactly one predefined surface area 78.

[0195] Predefined surface areas 78 of the first plate 54 are connected to corresponding predefined surface areas 78 of the second plate 56 by means of sensor devices 48, specifically by electrical conductivity, via the respective connecting device 66 and / or connection device 68. Each predefined surface area 78 of the first plate 54 is thus assigned a corresponding corresponding predefined surface area 78 of the second plate 56. This enables coupling, particularly electrical coupling, between the first plate 54 and the second plate 56, so that, for example, when the measuring device 16 is used within the electrochemical energy converter 12, a current flow through the measuring device 16 is enabled and measured there.

[0196] The current flow occurs particularly in the extension direction 12a of the electrochemical energy converter 12.

[0197] The electrochemical energy converter 12 is operated in particular with supplied operating fluid and / or supplied process media. Operating fluids and / or process media are, for example, under pressure.

[0198] Another embodiment of an electrochemical energy converter of the energy conversion device 10 according to the invention is described in the Fig. 4 and Fig. 5 is shown schematically and labelled there with the reference symbol 12'.

[0199] The electrochemical energy converter 12' is largely identical in structure to the electrochemical energy converter 12. The electrochemical energy converter 12' is in particular an electrolysis stack and / or a fuel cell stack, which is composed of several electrolysis cells and / or fuel cells connected in series with the components described above.

[0200] In the inventive method for estimating and / or predicting the at least one physical parameter or the at least one target quantity 47 at the predetermined local target area 46 of the electrochemical energy converter 12, at least one physical parameter or at least one output quantity 45 is first measured by means of the measuring device 16.

[0201] A distribution of these measured output quantities 45 is in Fig. 8 is represented by means of a projection onto the measuring device 16. Different hatching patterns represent different numerical values ​​of output quantities 45.

[0202] As already mentioned, physical parameters or target variables 47 are not directly accessible to measurement at the specified local target area 46. In a further step of the method according to the invention, the at least one target variable 47 is therefore estimated and / or predicted by means of the computational device 18 based on the at least one machine learning model and the at least one measured output variable 45.

[0203] The output quantities 45 measured by the measuring device 16 at the local measuring area 44 deviate, as explained above, from the target quantities 47 actually prevailing at specified local target areas 46, sometimes considerably. A description of these deviations is given in Fig. 9A and Fig. 9B is shown. Output quantities measured at local measuring ranges are compared ( 45 ( Fig. 9A) and target variables actually occurring at specified local target areas 47 ( Fig. 9B). Different numerical values ​​are represented by different hatching patterns. The aim of the method according to the invention is to realistically predict and / or estimate the at least one target variable 47 at the specified local target area 46 using the at least one machine learning model and the at least one measured output variable 45.

[0204] In particular, it is also possible to record the time course of at least one output variable 45 and / or at least one target variable 47. This allows the representation and / or recording of a temporal sequence of the at least one output variable and / or the at least one target variable, so that, for example, information about the operating time of the electrochemical energy converter can be recorded ( Fig. 11).

[0205] The temporal sequence is represented in particular by a time point t0 and a time point t1, where time point t1 follows time point t0. Possible changes in the at least one output variable 45 and / or the at least one target variable 47 can thus be represented in particular in their temporal sequence.

[0206] At least one machine learning model is, in particular, an artificial neural network 80.

[0207] The artificial neural network 80 is or is formed from artificial neurons 82 and is formed from several layers of artificial neurons 82.

[0208] The artificial neural network 80 has an input layer 84 and an output layer 86. Between the input layer 84 and the output layer 86, the artificial neural network 80 has several intermediate layers 88.

[0209] The measured output quantities 45 run from the input layer 84 to the output layer 86, passing through several intermediate layers 88.

[0210] Intermediate layers 88 are in particular hidden layers.

[0211] Each of layers 84, 86, and 88 can transform signals, especially output variables, differently at their respective inputs. This makes it possible to decompose a complex task, based on a large amount of data, into several simpler tasks, each executed in different layers 84, 86, and 88 of the machine learning model.

[0212] In order to make the most precise prediction and / or estimation of at least one target variable 47, the artificial neural network 80 must be trained and / or calibrated using training data.

[0213] For this purpose, the artificial neural network 80 accesses at least one training data set 90.

[0214] The at least one training data set 90 is stored, for example, in the storage device 20 and / or in the artificial neural network 80, for example via learned parameters of the at least one machine learning model.

[0215] The at least one training data set 90 is generated by the method according to the invention by applying at least one training target variable 90a to the predetermined local target area 44, wherein the at least one training target variable 90a is a physical parameter for the function of the electrochemical energy converter 12.

[0216] In a further step of the procedure for generating the at least one training data set 90, at least one training output variable 90b is measured at the local measuring area 44 using the measuring device 16.

[0217] Thus, it is possible to establish a relationship between the known and / or specified training target variables 90a and the measured training output variables 90b.

[0218] The training output variable 90b and the training target variable 90a together form, for example, at least one training data set 90.

[0219] Alternatively or additionally, the at least one training output variable 90b measured by means of the measuring device 16 forms the at least one training data set 90.

[0220] Alternatively or additionally, at least one training target variable 90a forms the training data set 90.

[0221] The computing unit 18 is preferably provided with at least one training data set 90.

[0222] For different predefined local target areas 46, separate training data sets 90 are or will be generated.

[0223] Separate training data sets 90 are preferably generated for different electrochemical functional units 24 of the electrochemical energy converter 12, so that, for example, the influence of different materials and / or manufacturers can be taken into account.

[0224] Preferably, the local target area 46 is subdivided into predefined area areas. Predefined area areas of local target areas 46 correspond in particular to predefined area areas 78. Predefined area areas of local target areas 46 are then subjected to training target variables and measured at the corresponding area area 78 of the measuring device 16 and the local measuring area 44 assigned to the measuring device 16.

[0225] Adjacent predefined area regions within predefined local target areas are subjected to different training target variables during the generation of the training dataset. In particular, this allows for the consideration of influencing factors that neighboring area regions have on one another.

[0226] Training target variables and / or training output variables include, in particular, an electric current value and / or an electric current distribution value and / or an electric current density distribution value and / or an electric voltage value and / or a temperature value and / or a temperature distribution value.

[0227] The at least one training data set 90 is generated in particular in advance of commissioning the energy conversion device 10 by means of the method according to the invention.

[0228] The at least one training data set 90 is used to train the at least one machine learning model, for example the artificial neural network 80, so that a trained model 92 is created.

[0229] Different training datasets can be used to generate different models of artificial neural networks 80', 80" in particular.

[0230] By means of a computer program product 94 according to the invention, it is possible to create at least one measurement data set 96 from the at least one output quantity 45, which is then stored in particular in the storage device 20.

[0231] The computing device 18 estimates and / or predicts at least one target variable 46. The computer program product 94 according to the invention is configured to create a target data set 98 from the at least one target variable 46.

[0232] The target data set 98 is stored in particular in storage device 20.

[0233] The target data set 98 can then be evaluated using, for example, a rating unit 104.

[0234] By means of a comparison unit 100 of the computing device 18, it is possible, with the aid of the computer program product 94 according to the invention, to compare the at least one measurement data set 96 with the at least one target data set 98 and thus to generate at least one comparison data set 102. This serves, for example, to further refine and / or train the machine learning model.

[0235] The at least one comparison data set 102 can then be evaluated, for example, using the evaluation unit 104.

[0236] The computer program product 94 according to the invention is also configured to provide a trained model 92 to the computing device 18 by means of a model provisioning unit 106.

[0237] By means of a visualization device 108, which has a plurality of visualization units 110, it is in particular possible to visualize the at least one measurement data set 96 and / or the at least one target data set 98 and / or the at least one comparison data set 102.

[0238] In particular, the visualization device 108 is or forms a user interface.

[0239] An energy conversion system 112 according to the invention comprises at least one energy conversion device 10 according to the invention.

[0240] The energy conversion system 112 is, for example, an electrolyzer and / or a fuel cell system and / or a fuel cell power plant and / or an electrical energy storage system, in particular a battery system.

[0241] Furthermore, the energy conversion system 112 includes at least one operating unit 114 which can be controlled by means of control parameters 116 and / or regulated by control parameters 118.

[0242] Control and / or regulation parameters 116, 118 are provided by means of at least one sensor device 120.

[0243] Control and / or regulation parameters 116, 118 are in particular physical parameters that characterize the function of the at least one operating device 114. For example, pressure parameters and / or level parameters and / or leakage parameters and / or temperature parameters and / or flow parameters and / or state parameters, in particular of valves (open / closed).

[0244] Control and / or regulation parameters 116, 118 can also be environmental parameters 117. Environmental parameters 117 are, in particular, meteorological parameters and / or parameters that characterize the energy demand of an electrical network.

[0245] Operating equipment 114 includes, in particular, equipment necessary for the operation of the energy conversion system 112. This includes, in particular, devices for transporting and / or storing process media and / or devices for recording environmental parameters.

[0246] The energy conversion system 112 is configured to carry out the inventive method for analyzing at least one state.

[0247] For this purpose, the energy conversion system 112 also has at least one state detection device 122 for detecting at least one state.

[0248] By means of the at least one state detection device 122, at least one first state and one second state are detected based on the at least one control parameter 116 and / or regulation parameter 118.

[0249] The second state is, in particular, temporally subsequent to at least one first state.

[0250] By means of a state calculation device 124 of the energy conversion system 112 it is particularly possible to determine an actual state of the energy conversion system 112 and to derive a temporal progression of the actual state from it.

[0251] Based on at least one machine learning model and the temporal evolution of the current state, the state calculation device makes it possible, in particular, to estimate and / or predict a target state of the energy conversion system 112. Specifically, at least one limit value can be estimated and / or predicted based on this.

[0252] A deviation calculation device 126 of the energy conversion system 112 is configured to determine a deviation between the actual state and the predicted and / or estimated target state. If the deviation reaches at least one limit value, an output device 128 of the energy conversion system is configured to issue an information message and / or warning message.

[0253] For example, this also enables demand-based maintenance (predictive maintenance) of components 10, 114, and 120 of the energy conversion system 112. This reduces potential downtime and / or maintenance periods. The main goal of predictive maintenance is to create the most precise possible maintenance schedule and prevent unexpected system failures. Knowing when which components and / or parts need servicing allows for better planning of maintenance resources such as spare parts or personnel. Furthermore, system availability can be increased by converting unplanned downtime into planned downtime. Additional benefits include a potentially longer service life for the systems, especially the energy conversion system 112, increased system safety, fewer accidents with negative environmental impacts, and optimized spare parts handling.

[0254] Another computer program product according to the invention is designed to perform at least one of the following steps based on the method for analyzing at least one state of the energy conversion system 112: - Creating and storing at least one initial state data record based on at least one physical parameter or at least one output variable; - Creating and storing at least one second state data set based on at least one physical parameter or at least one output variable; - Creating and storing at least one current state record, wherein the at least one current state record is formed from or is a current second state record; - Creating and saving the time history of at least one current state data record; - Creating and storing at least one target state data set based on at least one machine learning model and at least one actual state data set, in particular the temporal evolution of the at least one actual state data set; - Creating and storing at least one threshold data set based on at least one machine learning model and at least one target state data set; - Creating and saving a deviation data record based on a comparison of at least one actual state data record with at least one target state data record; - Create and store at least one comparison data set based on a comparison of at least one deviation data set with at least one limit data set; - Issuance of a message, in particular an information message and / or warning message, when a predefined value of the deviation data set reaches a predefined value of at least one limit value data set.

[0255] By means of the methods according to the invention, the computer program products according to the invention, the energy conversion device 10 according to the invention and the energy conversion system 112 according to the invention, it is thus possible to provide an improved estimation and / or prediction of functional parameters of the energy conversion device 10 and / or the energy conversion system 112, which are not accessible to direct measurement. Reference symbol list 10 Energy conversion device 12 electrochemical energy converters 12' electrochemical energy converter 12a Direction of extension 14 Functional module 16 Measuring device 18 Calculation device 19 Time calculation unit 20 Storage setup 22 Interface 24 electrochemical functional unit 26 Electrode arrangement 28 Bipolar plate 30 Membrane electrode array 32 Gas diffusion layer 34 Intermediate shift 36 cases 38 first housing part 40 second housing part 42 Connecting element 44 local measuring range 45 Initial size 46 specified local target area 47 Target size 48 sensor devices 49 Sensor 50 Resistance measuring device 52 temperature-sensitive element 54 first record 56 second record 58 measuring segments Section 60 62 tab 64 Contact surface 66 Management equipment 68 Connection device 70 Edge area 72 Central Area 73 measuring effective range 74 Adhesive 76 space 78 specified area 80 artificial neural network 80' artificial neural network 80" artificial neural network 82 artificial neurons 84 Input layer 86 Output layer 88 Intermediate shift 90 training data set 90a Training target size 90b Training output size 92 trained models 94 Computer program product 96 measurement data set 98 Target data set 100 comparison units 102 Comparison data set 104 assessment units 106 Model Deployment Unit 108 Visualization equipment 110 visualization units 112 Energy transition system 114 Operating equipment 116 control parameters 117 environmental parameters 118 control parameters 120 sensor device 122 Condition monitoring device 124 State Calculation Device 126 Deviation Calculation Device 128 Output device

Images