Apparatus and computer-implemented method for determining a state of a fuel cell system

Abstract

Methods for operating a fuel cell system including a fuel cell stack and computer implementation, wherein data are provided that map (301) the input quantities of the fuel cell system and the positions of the cells in the fuel cell stack to the voltage of the cells, respectively, wherein a model is trained (302) based on the data to map the input quantities of the fuel cell system and the positions of the cells in the fuel cell stack to a predicted probability distribution of the voltage of the cells, wherein the current input quantities of the fuel cell system are determined (303), wherein for at least one cell in the fuel cell stack, at least one probability of the voltage of the cell and / or the total voltage of the fuel cell stack is determined (304) based on the probability distribution using the model, based on these current input quantities, wherein the state of the fuel cell system is determined (305) based on the probability.

Description

Technical Field

[0001] This invention relates to a device and a computer-implemented method for determining the state of a fuel cell system. Background Technology

[0002] A fuel cell system is a total system comprising multiple subsystems. This fuel cell system includes one or more fuel cell stacks and multiple subsystems that must exist to supply power to the fuel cell stacks.

[0003] Fuel cell stacks typically do not have individual actuators; that is, the fuel cell stack itself is a passive component or passive assembly.

[0004] To monitor the condition of the fuel cell stack, sensors can be installed that monitor parameters of the fuel cell stack, such as cell voltage. These sensors are expensive and must also be protected against malfunctions or erroneous measurements. Summary of the Invention

[0005] A computer-implemented method for determining the state of a fuel cell system including a fuel cell stack specifies the following: providing data that maps the input quantities of the fuel cell system and the cell positions of the fuel cell stack to the voltage of the cell, respectively; wherein a model is trained based on this data to map the input quantities of the fuel cell system and the cell positions of the fuel cell stack to a predicted probability distribution of the voltage of the cell; wherein the current input quantities of the fuel cell system are determined; and wherein, for at least one cell of the fuel cell stack, at least one probability of the voltage of that cell and / or the total voltage of the fuel cell stack is determined based on these current input quantities using the model according to the probability distribution; and wherein the state of the fuel cell system is determined based on this probability. This enables individual cell monitoring, operational strategy monitoring, and diagnostics, without requiring all the sensors used during model training during operation of the fuel cell system after training.

[0006] In one aspect, this state characterizes the safety of input quantities for the voltage of at least one cell and / or for the total voltage of the fuel cell stack, wherein probabilities are determined according to a probability distribution, and the operation of the fuel cell system is classified as safe if the probabilities satisfy a certain condition. Thus, aspects related to safety can be identified.

[0007] It can be specified that the probability of the voltage of at least one cell being less than a first threshold, or the voltage of at least one cell being greater than a second threshold, or the total voltage of the fuel cell stack being greater than a third threshold, can be determined. This can prevent voltage over- or under-voltage conditions related to safety during operation.

[0008] During training, these data can at least determine the expected value and variance of the probability distribution. This allows the construction of an objective function, particularly a likelihood function, that quantifies the degree to which the parameterized model describes these data. This parameterized model can be, for example, a Gaussian process model. Optimization of the objective function parameterizes this Gaussian process model.

[0009] It can be stipulated that the probability value of the battery's voltage is determined based on this probability distribution, and that the condition is checked to ensure that the value meets the specified criteria. In this way, individual batteries can be effectively monitored.

[0010] It can be stipulated that, for multiple cells in a fuel cell stack, a value for this probability is determined for each cell, wherein these values ​​are used to determine the common probability of the multiple cells, and wherein the common probability is checked to see if a condition is met. Thus, the total voltage is effectively monitored based on the probability of each cell.

[0011] It can be stipulated that the solution to the following optimization problem is determined, which is constrained by a function based on the common probabilities of multiple cells in a fuel cell stack, wherein the solution constrains at least one control quantity or at least one parameter for operating the fuel cell system. This ensures the reliable operation of the fuel cell system, depending on the circumstances.

[0012] The function can be limited based on the difference between the minimum and maximum voltages of these batteries.

[0013] This function can be limited based on the variance of the voltages of these batteries.

[0014] In one respect, this model can be used to determine the location of cells and / or cells in a fuel cell stack that have either a higher or lower probability of being in a safe operating state relative to other cells in the fuel cell stack. This allows for the identification and then avoidance of potential differences in cell voltage that are important for the operating strategy.

[0015] A first probability distribution of the real part of the impedance of the battery and / or the fuel cell stack and a second probability distribution of the imaginary part of the impedance of the battery and / or the fuel cell stack can be determined, wherein the state of the fuel cell system is determined based on the first probability distribution and the second probability distribution. This enables more detailed monitoring.

[0016] When a voltage is present, impedance measurements provide both the real and imaginary parts of the impedance as a result. This impedance provides additional characteristic quantities. For example, the moisture content of the membrane can be approximately deduced from this impedance.

[0017] This also enables better operation of fuel cell systems without continuous cell voltage monitoring (CVM) and without a matching CVM sensor. In machine learning methods, data from impedance measurements can be additionally used to support the omission of the CVM sensor. This significantly improves the informational value of the model.

[0018] In one aspect, for this training, a first measurement can be recorded at the fuel cell system, wherein a first measure of the information content, in particular a first entropy, is determined for the first measurement. In this aspect, a second measurement can be recorded at the fuel cell system, wherein a second measure of the information content, in particular a second entropy, is determined for the second measurement. It can be specified that data from the first measurement is provided either when the first measure is greater than the second measure, or data from the second measurement is provided otherwise. Thus, the training is performed more quickly.

[0019] It can be specified that the training data is subject to at least one barrier (Schranke), specifically a barrier for the minimum or maximum permissible voltage of the battery. Thus, the model learns to consider the probability distribution of the minimum or maximum permissible voltage of the battery during runtime. This further improves monitoring.

[0020] An apparatus for determining the state of a fuel cell system specifies that the apparatus is constructed to perform the method. Attached Figure Description

[0021] Other advantageous embodiments are derived from the following description and accompanying drawings. In the drawings: Figure 1 A schematic diagram of the equipment used to operate a fuel cell system is shown; Figure 2 A schematic diagram of the collaboration of a model used to operate a fuel cell system is shown; Figure 3 The steps in a method for operating a fuel cell system are shown. Detailed Implementation

[0022] The structure described below, especially Figure 2 The structure is cited exemplarily to illustrate the method described below. This method can be applied correspondingly to model structures and regulator structures.

[0023] exist Figure 1The diagram schematically illustrates an apparatus 100 for determining the state of a fuel cell system having a fuel cell stack. The apparatus 100 is configured to implement the method described below. The apparatus 100 includes a forward model configured to determine electrical power and voltage as functions of regulation. For example, the forward model includes a first model 101, a second model 102, and at least one third model 103. The fuel cell system includes a fuel cell stack and a supply system. The fuel cell system forms a total system, which in this example is at least partially modeled using at least one third model 103. In this example, the at least one third model 103 is also a chemical or physical model, particularly described by differential equations.

[0024] In this example, the following four third models 103 are shown: Model 103-1 is a part of the overall system used for air delivery and / or exhaust. Model 103-2 is a part of the overall system used for metering hydrogen from the storage tank system, discharging purified gas from the anode path, draining water from the anode path, and circulating hydrogen in the fuel cell system. Model 103-3 is a part of the overall system used for cooling the fuel cell system. Model 103-4 is used for the electrical part of the overall system, which transmits the electrical power of the fuel cell stack to the vehicle power grid or other power grid, for example by means of a DC / DC converter and other components, such as devices for short-circuiting, current measurement, and voltage measurement of the fuel cell stack and / or the battery pack and / or individual cells of the fuel cell stack.

[0025] The first model 101 is constructed as a physical model, which describes the physical relationships in the fuel cell system, for example, by means of differential equations.

[0026] The second model 102 is constructed as a data-based model that models the difference between the physical model and the actual characteristics of the fuel cell system.

[0027] To date, there is no accurate dynamic model that describes the characteristics of the entire fuel cell system. Although each part of the overall system can be well described by at least one third model 103, the dynamic coordination of these parts within the overall system is unknown or only poorly understood.

[0028] Forward models can predict, for example, the electrical power of a fuel cell system at time t+1, based on possible manipulations at time t and the preceding short time period T.

[0029] This modeling is based on a hybrid model with chemical and / or physical components as well as data-based components. The chemical and physical components consist of known parts of the overall system, for which a first model 101 and at least one third model 103 are defined in the form of differential equations. Examples of known differential equations describing the dynamic characteristics of the various parts of the overall system, in this case, the air system, the cooling system, the hydrogen system, and the electrical system, are well-known, for example, from the following literature: [1] Control Analysis of an Ejector Based Fuel Cell AnodeRecirculation System, Amey Y. Karnik, Jing Sun and Julia H. Buckland; [2] Model-based control of cathode pressure and oxygen excess ratio of a PEM fuel cell system, Michael A. Danzer, Jörg Wilhelm, Harald Aschemann, Eberhard P. Hofer; [3] Humidity and Pressure Regulation in a PEM Fuel Cell Using a Gain-Scheduled Static Feedback Controller, Amey Y. Karnik, Jing Sun, Fellow, IEEE, Anna G. Stefanopoulou, and Julia H. Buckland; [4] MODELING AND CONTROL OF AN EJECTOR BASED ANODE RECIRCULATIONSYSTEM FOR FUEL CELLS, Amey Y. Karnik, Jing Sun; [5] Flatness-based multi-parameter control design of fuel cell system, Daniel Zirkel; [6] Model predictive control of PEM fuel cell systems, Jens Niemeyer; [7] Control for efficient operation of PEM fuel cell systems, Christian Hähnel.

[0030] All these parts of the overall system have various control parameters that affect their dynamics. Below, exemplary portions of the overall system are listed with respect to the control parameters of the fuel cell system and descriptions of these control parameters, which can be used to influence or affect the dynamics. Furthermore, these parameters are also important, particularly for the degradation or aging of the individual components of the fuel cell stack, and for the energy consumption or power demands of the supply system supplying the fuel cell stack, especially due to parasitic losses. For example, the air compressor of the fuel cell system alone can consume 15% of the fuel cell stack's power. The fuel cell stack must calculate how much more of this power it can provide so that the fuel cell stack can output the desired net power as effective power.

[0031] 1) Air system lambda_cath: Air excess relative to the stoichiometry in the cathode path of the fuel cell system; mAir_cath: Air mass flow in the cathode path of the fuel cell system; p_cath: Pressure in the cathode path of the fuel cell system; T_cath: Temperature in the cathode path of the fuel cell system; fi_cath: Humidity in the cathode path of the fuel cell system.

[0032] This part of the fuel cell system is used for air supply and / or exhaust from the fuel cell stack.

[0033] In this example, the parameters lambda_cath and mAir_cath can be used interchangeably. If the fuel cell system allows setting the humidity of the incoming air, then humidity control can be provided.

[0034] 2) Hydrogen system lambda_anod: The amount of hydrogen molecules in excess relative to the stoichiometry in the anode path of the fuel cell system, that is, the amount of H2 in excess; mH2_anod: The mass flow of hydrogen molecules in the anode path of a fuel cell system, that is, the H2 mass flow; p_anod: Pressure in the anode path of the fuel cell system; dp_anod_cath: The differential pressure between the cathode path and the anode path in a fuel cell system; mN2_anod: Nitrogen mass flow, nitrogen concentration, or nitrogen molecule flow in the anode; mH2_addfromtank: H2 mass or H2 mass flow metered from the H2 tank of the fuel cell system or from an external source into the anode path; Purge_actuation: Used to control the discharge or removal of anode gas from the anode path; Drain_actuation: Used to control the discharge or removal of liquid water from the anode path; Purge & Drain_actuation: Combined control of valves or common valves used for Purge_actuation and Drain_actuation.

[0035] This part of the fuel cell system is used for hydrogen recycling and other functions of the fuel cell system.

[0036] In this example, the parameters lambda_anod and mH2_anod can be used interchangeably. For instance, if a hydrogen recirculation fan is present in the fuel cell system, the recirculation rate of that fan is related to mH2_anod.

[0037] The parameter mH2_addfromtank can additionally include temperature records. The parameter mH2_addfromtank can be supplemented by lambda_anod or by mH2_anod, or used in combination with them.

[0038] The parameter mN2_anod can be derived from model calculations or determined by sensors. The parameter mN2_anod can be used to trigger a purge action.

[0039] The parameter Purge_actuation can specify, discretely and at time intervals, the opening duration and / or opening interval of the valve used to purge or remove anolyte gas. Both can be variable.

[0040] The parameter Drain_actuation can discretely and at time intervals specify the opening duration and / or opening interval of the valve used to drain or remove liquid water. Both can be variable.

[0041] 3) Cooling system T_Stack_op: The operating temperature of the coolant in the fuel cell system, which is approximately the operating temperature of the fuel cell stack. Fan_actuation: Control of the ventilation system; dT_Stack: Temperature changes of the coolant on the fuel cell stack, such as heating, or temperature changes in the fuel cell system; m_Cool: The mass flow of coolant through the cooling path of a fuel cell stack or fuel cell system; dp_Cool: Pressure drop along the cooling path of a fuel cell stack or fuel cell system; Pump_actuation: Pump control used to generate coolant mass flow; Valve_actuation: Valve control used to generate coolant mass flow; p_Cool: Pressure in the coolant path of the stack.

[0042] This part of the fuel cell system is used for the circulation of coolant within the fuel cell system.

[0043] The parameter T_Stack_op can be extended, or more precisely, used for the membrane, a temperature-critical component of the fuel cell stack. For this purpose, the membrane temperature can be inferred, for example, using models of coolant temperature, stack exhaust air temperature, stack voltage, and stack current. This operating temperature can be modeled based on load, ambient temperature, and ventilation control—that is, based on Fan_actuation.

[0044] The parameter dT_Stack can be determined based on the temperature difference between the output and input temperatures of the coolant and can be set by means of the mass flow of the coolant, such as by using pumps and three-way valves in the cooling system used for fuel cell stacks or fuel cell systems.

[0045] Instead of the parameter p_Cool, differential pressure relative to the cathode and / or relative to the anode can be used.

[0046] 4) Electrical system Voltage: Current: Current density: Electric power: Short-circuit relays, short-circuit devices, and other electrical actuators as necessary.

[0047] The electrical parameters of a fuel cell stack—voltage, current, current density, and power—interact strongly with a power grid whose architecture may be very different from its own.

[0048] For example, the electrical power of a fuel cell stack can be transmitted to the grid using a DC-DC converter, such as a DC / DC converter, based on the voltage and / or current of the fuel cell stack. For instance, the DC / DC converter can set the current drawn from the fuel cell stack using a voltage gradient.

[0049] A short-circuit relay can be installed to short-circuit the fuel cell stack, that is, to short-circuit two terminals. This can be used, for example, for freeze-start, in which electrical power is temporarily not supplied to the grid and is instead converted into heat.

[0050] Parameters derived from this, such as resistance or efficiency, can also be modeled.

[0051] Impedance measurements, such as impedance spectroscopy, can also be integrated into the electronic systems of fuel cell systems. This impedance measurement can be configured to determine the real and imaginary parts of the impedance of one or more cells in the fuel cell stack and output these parts for further calculations.

[0052] These parameters are all variables. Not all possible variables are ultimately listed. Among these variables, there can be both model-based values ​​and measured values. In addition to or in place of absolute parameters, difference parameters or differences relative to reference values ​​can also be used. Alternatively, only a subset of the possible variables can be used as parameters for modeling.

[0053] The device 100 includes a control unit 104 configured to control a fuel cell system or a subsystem for operating a fuel cell stack using various control parameters. The device 100 may include a measuring device 106, particularly sensors for recording parameters at the fuel cell system. In this example, the device includes: at least one computing device 108 configured to implement the steps of the method described below; and at least one memory 110 for modeling. The at least one computing device 108 may be: a local computing device in a vehicle; a computing device on a server or in the cloud; or particularly a computing device distributed across multiple servers or across the vehicle and at least one server.

[0054] A fuel cell system includes a fuel cell stack. In this example, the fuel cell stack includes n cells 112, which are... Figure 1 The diagram is presented schematically and numbered according to its position i in the fuel cell stack, where i=1, ... n.

[0055] in accordance with Figure 2 It describes the collaboration of models used to operate fuel cell systems.

[0056] In this example, for the fuel cell system, the desired operating quantity y_req is defined as the input quantity. Preferably, this operating quantity is the electrical power, voltage, efficiency, or waste heat of the fuel cell system, especially thermal power. The fuel cell system should be manipulated using at least one control quantity u_req to ensure that the fuel cell actually provides this operating quantity. The at least one control quantity u_req is a target value for the control of the fuel cell system by the control device 104. In this example, the desired operating quantity y_req is mapped to the at least one control quantity u_req through a control strategy. This strategy may be mapping the desired operating quantity y_req to the at least one control quantity u_req through a pre-given linear or nonlinear function or through a pre-given table.

[0057] Due to dead time, inertia, hysteresis, aging effects, or deviations between the actuator and the target value, a control quantity different from the target value may occur. On one hand, this control quantity can be recorded by a sensor, for example, as an actual set control quantity u_act. On the other hand, at least one set control quantity u_pred can be determined as a prediction using at least one third model 103. In this example, for at least a portion of the fuel cell system, particularly for the fuel cell stack or for at least one subsystem of the subsystem used to supply the fuel cell stack, a predicted x[subsy]_pred of at least one set control quantity u_pred for at least that portion of the fuel cell system is determined based on a pre-given control quantity x[subsy]_req for that portion of the fuel cell system, and the at least one set control quantity u_pred is defined based on the predicted x[subsy]_pred. In this example, the parameter x[subsy]_req is combined in a vector that defines the control quantity u_req. Each of the tuning values ​​listed above can be used as a parameter x[subsy]_req for the corresponding portion of the fuel cell system. If multiple tuning values ​​are set for a subsy, the parameter x[subsy]_req is a vector that includes these tuning values. For illustrative purposes, only the selected parameter is described below.

[0058] exist Figure 2 In Model 103-1, that is, for the air system, the parameter is represented by [subsy] = A; in Model 103-2, that is, for the hydrogen system, the parameter is represented by [subsy] = H; in Model 103-3, that is, for the cooling system, the parameter is represented by [subsy] = C; and in Model 103-4, that is, for the electrical system, the parameter is represented by [subsy] = E.

[0059] All actual control quantities, or only a portion thereof, can be determined or measured using the model based on the corresponding pre-given control quantities.

[0060] Regardless of whether the controlled quantity is measured (u_act) or modeled (u_pred), it can be: the pressure difference between the anode and cathode of the fuel cell system; the temperature difference between the first temperature of the coolant upon its entry and the second temperature of the coolant upon its exit from the fuel cell stack; the humidity of the air, especially upon its exit from the fuel cell stack; the pressure of the air, hydrogen, and / or coolant; the operating temperature; the air mass flow; the hydrogen molecular mass flow; the coolant mass flow; or electrical characteristics, especially the current, current density, or voltage at the fuel cell system. The fuel cell system represents the overall system.

[0061] The controllable quantities may define, for example: the pressure difference between the anode and cathode of the fuel cell system; the temperature difference between the first temperature of the coolant upon its entry and the second temperature of the coolant upon its exit from the fuel cell stack; the humidity of the air, especially upon its exit from the fuel cell stack; the pressure of the air, hydrogen, and / or coolant; the operating temperature; or the air mass flow in the air supply and / or exhaust portions of the fuel cell system. The controllable quantities may define the hydrogen molecular mass flow in portions of the fuel cell system used for hydrogen circulation within the fuel cell system. The controllable quantities may define the coolant mass flow in portions of the fuel cell system used for cooling the fuel cell system. The controllable quantities may define an operating temperature approximately equal to the coolant temperature. The controllable quantities may define electrical characteristics of the electrical components of the fuel cell system, such as the current, current density, or voltage of one of the fuel cells or the fuel cell system.

[0062] Preferably, at least one pre-defined control quantity u_req defines a target value for pressure, operating temperature, air mass flow, hydrogen molecular mass flow, coolant mass flow, or electrical characteristic quantity, particularly the current or voltage of the fuel cell system. In this example, the control quantity xA_req at time point t defines a target value for pressure or air mass flow in the portion of the overall system used for air delivery and / or exhaust. In this example, the control quantity xH_req at time point t defines a target value for hydrogen molecular mass flow in the portion of the overall system used for hydrogen circulation within the fuel cell system. In this example, the control quantity xC_req at time point t defines a target value for coolant mass flow in the portion of the overall system used for cooling the fuel cell system. The control quantity may also define an operating temperature that is approximately the coolant temperature. In this example, the control quantity xE_req at time point t defines a target value for an electrical characteristic quantity of the electrical portion of the overall system, such as the current or voltage of one of the fuel cells or the fuel cell system. In this example, the pre-given manipulation quantity u_req is a vector u_req = (xA_req, xH_req, xC_req, xE_req). T Correspondingly, the manipulators appearing in this example are defined by vectors. If all the manipulators appearing can be measured, then the manipulators appearing are u_act = (xA_act, xH_act, xC_act, xE_act). T For the case where all occurrences of manipulation are modeled, the occurrences of manipulation are u_pred = (xA_pred, xH_pred, xC_pred, xE_pred). T Preferably, a hybrid approach is used, in which the control quantities that can be measured using sensors that are generally available at the fuel cell system are measured while other control quantities are modeled.

[0063] The operating quantity y_act of the fuel cell system is determined using the first model 101 based on at least one occurrence of a control quantity. In this example, the occurrence of the operating quantity is a scalar, but it is also possible to determine a vector with multiple values ​​of different operating quantities using the first model 101. In this example, the Kulikovsky fuel cell model is used for the first model 101 as a static model. The Kulikovsky model is analytically derived from the basic set of differential equations describing the kinetics of the cathode catalyst layer. The model uses the following input quantities: cathode mass flow, cathode λ (Lambda), cathode input pressure, cathode output pressure, air humidity at the cathode inlet, air humidity at the cathode outlet, current or current density, coolant inlet temperature, and coolant outlet temperature.

[0064] Using the second model 102, a prediction is made of the deviation dy_pred between the operating quantity y_act determined by the first model 101 and the actual value of the operating quantity at the fuel cell system, based on the at least one occurring manipulation quantity.

[0065] In this example, the second model 102 is a data-based model that should predict the deviation dy_pred between the first model 101 and the actually measured characteristics of the fuel cell system using a Gaussian process. During training, the second model 102 can be randomly initialized first and trained iteratively.

[0066] In this example, the second model 102 has already been trained.

[0067] Based on the runtime y_act determined by the first model 101 and the prediction for the deviation dy_pred, the runtime y_pred is determined at the correction device 202. This means that the prediction of the runtime by the physical model is corrected by using the data-based model to predict the deviation.

[0068] In principle, what matters is the voltage of each cell in the fuel cell stack. The distribution varies within the bandwidth dU. The bandwidth dU is measured by the battery voltage. The bandwidth dU is derived specifically from the dielectric distribution, flow state, and aging state. For n batteries, the distribution of the voltage across the n batteries... Maximum voltage at With the batteries in the n batteries Minimum voltage at The bandwidth between Where j and k represent different batteries.

[0069] In this example, the bandwidth dU should be as small as possible. Here, the medium refers to one or more of the fluids mentioned above. The voltage of a fuel cell stack with n cells... The voltage of each battery composition: To protect the components of the fuel cell system, especially to prevent damage to the cells, the voltage of these cells can be specified. Greater than the minimum threshold : minimum threshold It can be positive, negative, or zero. See below for reference. Figure 3The described computer-implemented method for determining the state of a fuel cell system specifies that, in a step for machine learning, additional sensors are deployed at the fuel cell system. These sensors are required to operate the fuel cell system after machine learning, or not all of these sensors are required to operate the fuel cell system after machine learning.

[0070] These sensors can be used to record labeled data tuples for machine learning. The data is from (i, u). Here... This represents the input quantity, where i represents the battery position and u represents the battery voltage. Examples include current density, air mass flow rate, air pressure, hydrogen molecular mass flow rate, and fuel cell stack temperature. For situations where all controllable quantities are measurable, one can use... u_act = (xA_act, xH_act, xC_act, xE_act) T For cases where all manipulatory variables are modeled, the following can be used: u_pred = (xA_pred, xH_ pred,xC_pred, xE_ pred) T It can be stipulated that: [Regarding...] The data uses partially measurable and partially modeled parameters. These data can be recorded in a fleet or on a test bench. The data can be determined for multiple different environmental conditions, driving modes, driving characteristics, or performance profiles.

[0071] The computing device 108 is configured to train a model using this data in machine learning. The model is trained to process the inputs of the fuel cell system... The location of cell i in the fuel cell system is mapped to the voltage of that cell. The probability distribution of the prediction.

[0072] In this example, the probability distribution of battery i is defined by its expected value and variance. For a Gaussian process, for example, using... ( ,i) and ( ,i).

[0073] This model can be used to predict input. Safety. For example, if the voltage of each battery... Less than the threshold The probability satisfies the condition If so, the operation of the fuel cell system is considered safe. This condition is, for example: in This is the threshold voltage at battery i. This condition... and that threshold This condition can be given in advance by experts. It can be specified to a suitable value, or constrained as a function, which depends on environmental conditions or operating status, for example. The same threshold can be used for multiple batteries. and the same conditions .

[0074] It is also possible to specify a threshold that depends on the battery. For example, edge cells, that is, cells i=1 and i=2 and cells i=n and i=n-1, often have problems maintaining their voltage. For example, for n cells... Specified threshold, where Is with Different thresholds. Threshold They can have the same value or different values. Threshold They can have the same value or different values ​​from each other.

[0075] In this case, the threshold Unlike threshold .

[0076] The bandwidth or variance of the voltage at battery i can also be specified.

[0077] The computing device 108 is configured, for example, to determine the common probability that the fuel cell system is safe to operate based on the model. For example, for n batteries, this common probability... It was identified as: in This represents the corresponding input quantity of the n batteries. ,i. For example, the common probability It is an n-dimensional Gaussian process, that is, an n-dimensional normal distribution.

[0078] The computing device 108 is configured, for example, to determine the position of battery i based on the model and thereby determine the battery among the n batteries that has a higher probability of being in a safe operating state relative to the other batteries among the n batteries.

[0079] The computing device 108 is configured, for example, to determine the position of battery i based on the model and thereby determine the following battery among the n batteries, relative to the other batteries among the n batteries or relative to a threshold. This indicates a higher probability that the battery is in an unsafe operating state. This threshold... This threshold can be predetermined by experts. In this example, the threshold is... For example, regarding thresholds It is determined as described. Unlike the threshold... The determination of this threshold The value is specified for unsafe operating conditions, under which the voltage of each battery is... Greater than this threshold It is also possible to specify a threshold that depends on the battery. .

[0080] Therefore, the probability of the battery being in an unsafe operating state is: .condition It can be limited by experts as described above, or it can have other values.

[0081] It is important that the entire stack, that is, all the cells, is operating safely. This is true when each individual cell is operated within safe operating limits. Furthermore, the bandwidth dU of the cell voltage across all cells should be less than a limit.

[0082] The bandwidth or variance of the bandwidth dU can also be specified.

[0083] Voltage of fuel cell stack It can be used for additional supplementary limit values.

[0084] Impedance values ​​can be used as additional criteria. These impedance values, in addition to other parameters in the model, can be used to calculate voltages. Applicable individually for all i. Alternative solutions include: .

[0085] For example, for the threshold d of bandwidth dU, the additional condition is: .

[0086] In this example, based on the voltage of each battery i Greater than the threshold The probability of the common probability is checked to see if it satisfies the condition. In this example, the condition The parameter δ is used to define the acceptable risk. This parameter δ can be defined, for example, by an expert. In this example, it is determined by the common probability for the n batteries. Applicable to: , It can then identify a safe operating status.

[0087] The computing device 108 can be configured to determine the solution to an optimization problem based on the common probability of the n batteries. The term "to be limited to" is: These will be referred to below as quality indicators. It is the objective function, and can be, for example, defined as the difference between the minimum and maximum voltages falling on the n batteries: Quality Indicators For example, it can be limited to the variance of voltage: Alternatively, regarding the single voltage of the n cells in the fuel cell stack. ,..., The objective function is constrained to be: in It is the n x n covariance matrix of the predicted n voltages, and det is the determinant.

[0088] Alternatively, instead of a determinant, the largest eigenvalue can be used: Alternatively, the trace of the covariance matrix can be used.

[0089] For example, regarding the single voltage of n cells in a fuel cell stack. ,..., The objective function for these single voltages ,..., Expected value The case is limited to: It is also possible that, for the solution to this optimization problem, the Pareto Front is a combination of these alternatives.

[0090] It can be specified that the real and imaginary parts of the impedance of one or more cells in a fuel cell stack be evaluated. For example, the real and imaginary parts can be determined using the impedance spectrum of a single cell in the fuel cell stack or the cells themselves. The cell's impedance value can be used as input to the model. Individual impedance values ​​of each cell in the fuel cell stack can also be used as input to the model. Separate thresholds can be specified for the real and imaginary parts.

[0091] In one respect, different models can be learned and / or used for different operating states of a fuel cell system. Operating states can be defined, for example, as startup, normal operation, shutdown, or cold start.

[0092] When recording data, it can be stipulated that measurements are recorded iteratively, and the data that provides the largest, and especially the most information, is selected. This speeds up the recording process because fewer measurements are needed for training of the same quality. For a Gaussian process as a model, for example, the entropy of these measurements can be determined and used as a measure of the uncertainty about the model.

[0093] It can be stipulated that safety barriers be used during training. These safety barriers can be pre-defined, for example, by a physical model or other models from the field of machine learning. For instance, the model or these models can be trained using the minimum permissible voltage of these batteries to avoid damage to the batteries caused by voltages below that minimum permissible voltage.

[0094] For example, dynamic effects can be considered by using the history of the input. For instance, a model with a nonlinear autoregressive exogeneous (NARX) structure can be used, which considers the input... The input space.

[0095] It can be stipulated that these data are labeled to enable the identification of input quantities. The key combinations. It can be specified that these combinations are output after identification for user inspection.

[0096] It can be stipulated that, instead of the active learning described so far, Bayesian optimization should be used to optimize the quality metrics. In this case, it's not about learning from input quantities to quality metrics. Instead of searching for a complete mapping, it only looks for the optimal operating point. This allows for faster determination of... The best working location.

[0097] Alternatively, it can be specified that: learning is used for machine learning of states and for the associated quality metrics. A common model. For example, a Gaussian process can be used for multidimensional outputs.

[0098] The total voltage of the fuel cell stack can be specified. Available, where the voltage of each battery is... Not available. In this case, it can be specified that the total voltage of the fuel cell stack is evaluated. The history or the total voltage of the fuel cell stack The evaluation is based on the deviation from the value predicted by the model.

[0099] See below for reference. Figure 3 The described method can be implemented in a vehicle or on a test bench. After training, the method can be implemented in a vehicle, wherein the steps of the method used to perform the training need not be performed again.

[0100] The method specifies in step 301 that data is provided, which includes multiple tuples representing the input quantities of the fuel cell system. The position of cell i in the fuel cell stack is mapped to the cell voltage u.

[0101] In one aspect, these data are provided iteratively. It can be specified that: in one iteration, a first measurement is recorded at the fuel cell system. It can be specified that: for this first measurement, a first measure of information content, in particular a first entropy, is determined. It can be specified that: in a second iteration, a second measurement is recorded at the fuel cell system. It can be specified that: for this second measurement, a second measure of information content, in particular a second entropy, is determined. It can be specified that: data from the first measurement is provided either when the first measure is greater than the second measure, or data from the second measurement is provided otherwise. Thus, the measurement with the most information content is selected.

[0102] In one respect, these data are provided such that they are limited by at least one barrier, particularly a safety barrier, which defines a minimum or maximum permissible voltage at the battery.

[0103] In one respect, this data is provided to enable the use of bandwidth dU.

[0104] In the subsequent step 302, the model is trained based on this data to: input the fuel cell system The location of cell i in the fuel cell stack is mapped to the voltage of that cell. The probability distribution of the prediction.

[0105] In this example, the expected value and variance of the probability distribution are learned from this data. For a Gaussian process, this would be used to determine the expected value of a normal distribution for battery i. ( i) and variance ( The trained model represents the probability distribution of multiple cells, in this case, the n cells of a fuel cell stack.

[0106] The trained model can then be used to determine the state of the fuel cell system. Therefore, steps 301 and 302 are no longer necessary.

[0107] In step 303, the current input of the fuel cell system is determined. It can be stipulated that only the input values ​​used in this training session will be used. Part of it.

[0108] In step 304, for at least one cell i of the fuel cell system, based on these current input quantities... The probability is determined based on this probability distribution using this model.

[0109] In one aspect, the voltage of the at least one battery is determined. Less than the threshold The probability of.

[0110] In one aspect, the voltage of the at least one battery is determined. Greater than the threshold The probability of.

[0111] It can be stipulated that, for multiple cells in a fuel cell system, preferably for the n cells, a probability is determined for each cell based on this probability distribution.

[0112] It can be stipulated that the common probability be determined.

[0113] In step 305, the state of the fuel cell system is determined based on at least one probability determined for battery i and / or based on the common probability.

[0114] This state can characterize the voltage for at least one battery. Input volume Security.

[0115] This state can characterize the total voltage for the fuel cell system. Input volume Security.

[0116] For example, if the probability and / or the common probability satisfy the condition If so, the operation of the fuel cell system is considered safe.

[0117] It can be stipulated that the voltage of at least one battery be checked. Less than the threshold probability Does it meet the conditions? .

[0118] It can be stipulated that the voltage of at least one battery be checked. Greater than the threshold probability Does it meet the conditions? .

[0119] It can be stipulated that: check the common probability. Does it meet the conditions? .

[0120] It can be defined that the probability of checking if the bandwidth dU of the voltage is less than the threshold d is... Does it meet the conditions? .

[0121] In this example, a safe operating state is identified if each of the checked conditions is met.

[0122] It can be stipulated that: the solution to the optimization problem is determined based on the probabilities for the n batteries. It is subject to limitations.

[0123] In this example, the following optimization problem is used with respect to the threshold c: This optimization problem can be correspondingly constrained for other thresholds. For threshold s, the optimization problem is, for example: For the threshold d, the optimization problem is as follows: It can be specified that, in optional step 306, the fuel cell system is operated using at least one control quantity or at least one parameter defined by the solution.

[0124] Then step 303 can be implemented. Thus, with the new current input... Let's continue with this method.

[0125] This method can be implemented using a first probability distribution of the real part of the impedance of the fuel cell in the fuel cell system and a second probability distribution of the imaginary part of the impedance. In this case, the state of the fuel cell system is determined based on the first probability distribution and the second probability distribution. For example, the conditions mentioned for safe operation must be met by both parts; otherwise, operation is classified as unsafe.

[0126] This method can be implemented using different models for different operating states of the fuel cell system.

Claims

1. A computer-implemented method for determining the state of a fuel cell system, the fuel cell system comprising a fuel cell stack, characterized in that, providing data mapping input quantities of the fuel cell system ( ) and positions of cells of the fuel cell stack to voltages of the cells, wherein the model is trained to map input quantities of the fuel cell system ( ) and positions of cells of the fuel cell stack to a predicted probability distribution of voltages of the cells, wherein current input quantities of the fuel cell system are determined, wherein for at least one cell of the fuel cell stack, at least one probability of a voltage of the cell and / or of a total voltage of the fuel cell stack is determined from the probability distribution using the model in dependence on the current input quantities ( ), wherein a state of the fuel cell system is determined in dependence on the probability. After this training, when the fuel cell system is running, the sensors used during the training of the model are eliminated.

2. The method according to claim 1, characterized in that, This state characterizes the input amount for the voltage of the at least one cell and / or for the total voltage of the fuel cell stack. The safety of the fuel cell system is determined based on a probability distribution, wherein if the probability satisfies a certain condition, the operation of the fuel cell system is classified as safe.

3. The method according to claim 2, characterized in that, Determine the probability that the voltage of at least one cell is less than a first threshold, or the voltage of at least one cell is greater than a second threshold, or the total voltage of the fuel cell stack is greater than a third threshold.

4. The method according to any one of the preceding claims, characterized in that, Based on this data, at least the expected value and variance of the probability distribution can be determined.

5. The method according to claim 4, characterized in that, The probability of determining the battery voltage is based on this probability distribution, and it is checked whether the value meets the conditions.

6. The method according to claim 4 or 5, characterized in that, For the multiple cells of the fuel cell stack, a value for each probability is determined, wherein these values ​​are used to determine the common probability of the multiple cells, and wherein the common probability is checked to see if a condition is met.

7. The method according to any one of the preceding claims, characterized in that, Determine the solution to the following optimization problem, which is constrained by the common probabilities of the multiple cells in the fuel cell stack and by a function, wherein the solution constrains at least one control quantity or at least one parameter for operating the fuel cell system.

8. The method according to claim 7, characterized in that, The function is defined based on the difference between the minimum and maximum voltages of these batteries.

9. The method according to claim 7, characterized in that, The function is defined based on the variance of the voltages of these batteries.

10. The method according to any one of the preceding claims, characterized in that, The model determines the location of the battery and / or the following batteries in the fuel cell stack, which have either a higher probability or a lower probability of being in a safe operating state relative to the other batteries in the fuel cell stack.

11. The method according to any one of the preceding claims, characterized in that, A first probability distribution of the real part of the impedance of the battery and / or the fuel cell stack and a second probability distribution of the imaginary part of the impedance of the battery and / or the fuel cell stack are determined, wherein the state of the fuel cell system is determined based on the first probability distribution and the second probability distribution.

12. The method according to any one of the preceding claims, characterized in that, A first measurement is recorded at the fuel cell system, wherein a first metric for information content is determined for the first measurement; a second measurement is recorded at the fuel cell system, wherein a second metric for information content is determined for the second measurement; wherein data from the first measurement is provided when the first metric is greater than the second metric, and data from the second measurement is provided otherwise.

13. The method according to claim 12, characterized in that, The first measure is the first entropy.

14. The method according to claim 12, characterized in that, The second measure is the second entropy.

15. The method according to any one of the preceding claims, characterized in that, Provide the following data, which are subject to at least one barrier.

16. The method according to claim 15, characterized in that, The at least one barrier is a barrier for the minimum or maximum permissible voltage of the battery.

17. An apparatus for determining the state of a fuel cell system, characterized in that, The device is configured to implement the method according to any one of claims 1 to 16.

18. A computer program, characterized in that, The computer program includes machine-readable instructions, which, when executed by a computer, operate according to any one of claims 1 to 16.

19. The computer program according to claim 18, characterized in that, The computer in question is a distributed computer.

Images