JP2007123007A - Method and equipment for diagnosing fuel cell system - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a method for diagnosing a fuel cell which allows for accurate diagnosis even in the case of repeated start and shutdown operation of the fuel cell. <P>SOLUTION: The equipment for diagnosing a fuel cell system consists of an operation database 2 for a fuel cell system 1, a shutdown historical data extraction means 3 for extracting the shutdown historical data of the operation data at the shutdown operation time and/or during the shutdown period of the fuel cell system 1 from the operation database 2, and a status diagnostic means 4 for diagnosing the status of the fuel cell system 1 at the restarting time in accordance with the shutdown historical data and operation data at the restarting time. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to a diagnostic method and a diagnostic apparatus for a fuel cell system, and more particularly to a diagnostic method for a PEFC (solid polymer fuel cell).

  Conventional fuel cell system diagnostic techniques include dividing a large number of cells constituting a fuel cell stack into a plurality of cell groups, measuring the output voltage for each cell group, and determining a predetermined threshold value for the average voltage and voltage variation. Compared to the above, a method for diagnosing the state of a fuel cell has been proposed (see, for example, Patent Document 1).

  Patent Document 1 also proposes a method of estimating an output voltage when the amount of water in a cell is appropriate from a cell temperature-current-voltage characteristic map and comparing a deviation between a measured value and an estimated value with a threshold value. Yes.

  Therefore, even when the cell voltage changes due to changes in the cell temperature and current, it can be estimated by reflecting the variation.

  Furthermore, a technique has been proposed in which changes over time before and after a fuel cell stop are recorded, and the state of the fuel cell is evaluated by comparing this change over time with a healthy change in the fuel cell output over time. (For example, refer to Patent Document 2).

  A technique has also been proposed in which fuel cell measurement data and a reference value measured when the fuel cell operates normally are compared by a control unit that forms a neural network to diagnose performance degradation of the fuel cell (see, for example, Patent Document 3). .

JP 2004-127915 A (pages 7 to 10, FIGS. 1 to 4) Japanese Patent Laid-Open No. 08-096825 (pages 6-7, FIGS. 1-2) Japanese Patent No. 2516274 (FIGS. 1 to 3)

  However, when the fuel cell is operated, the start and stop may be repeated, and when the fuel cell is restarted after being stopped for a certain time, the behavior of the voltage may be different even under the same operating conditions as before the stop.

  Patent Document 1 does not disclose a diagnostic method for an event in which the behavior of a battery such as a voltage changes during restart after stopping.

  In Patent Document 2, the change over time before and after the stop of the fuel cell, that is, after the restart of the fuel cell is recorded, and the state of the fuel cell is compared by comparing this change over time with the healthy change in the cell output over time. Even if the evaluation is performed, the influence of the operation status during the stop operation of the fuel cell system and / or during the stop period is not taken into account, so abnormalities such as voltage at the time of restart cannot be accurately grasped, and changes over time after restart The food back control will not be accurate unless it is recorded considerably.

  Patent Document 3 basically targets changes over time in battery performance during continuous operation, and no matter how precisely the fuel cell measurement data is compared with the reference value measured during normal operation, the fuel cell system Since the influence of the operation state during the stop operation and / or the stop period is not taken into consideration, it is not always possible to accurately diagnose and grasp the state of the fuel cell system at the time of restart.

  An object of the present invention is to provide a fuel cell diagnostic method and apparatus capable of performing high-accuracy diagnosis when operating by repeatedly starting and stopping.

  In order to solve the above-described problems, the present invention provides a fuel cell system operation database and stop history data extraction that extracts stop history data that is operation data during a stop operation and / or a stop period of the fuel cell system from the operation database. Proposed is a fuel cell system diagnosis device comprising means and state diagnosis means for diagnosing the state of the fuel cell system at the time of restart based on stop history data and operation data at the time of restart.

  The present invention also provides an operation database of the fuel cell system, stop history data extraction means for extracting stop history data that is operation data during a stop operation and / or a stop period of the fuel cell system from the operation database, and an operation database. A state that has a voltage estimation model that estimates the normal state voltage from the stored information, compares the estimated voltage with the actually measured voltage, and diagnoses the state of the fuel cell system at the time of restart A diagnostic device for a fuel cell system comprising diagnostic means is proposed.

  According to the present invention, the state of the fuel cell system at the time of restart is diagnosed based on stop history data that is operation data during the stop operation and / or stop period of the fuel cell system and operation data at the time of restart. Even when the fuel cell is operated by repeatedly starting and stopping, the state of the fuel cell can be diagnosed with high accuracy.

  Next, with reference to FIGS. 1-9, the Example of the diagnostic method and diagnostic apparatus of the fuel cell system by this invention is described.

  FIG. 1 is a block diagram showing a system configuration of a first embodiment of a diagnostic apparatus for a fuel cell system according to the present invention.

  The fuel cell system diagnosis apparatus according to the first embodiment includes an operation database 2 of the fuel cell system 1, a stop history data extraction unit 3, and a state diagnosis unit 4.

  That is, it has an operation database 2 of the fuel cell system 1 and a stop history data extraction means 3 that extracts stop history data that is operation data during a stop operation and / or a stop period of the fuel cell system 1 from the operation database 2. The state diagnosis means 4 diagnoses the state of the fuel cell system 1 at the time of restart based on the stop history data and the operation data at the time of restart.

  FIG. 2 is a diagram showing an outline of a fuel cell system to be diagnosed.

  The fuel cell system 1 is a PEFC (solid polymer fuel cell) using city gas as fuel. The fuel gas is converted into hydrogen H 2 by the reformer 11 and sent to the fuel cell stack 12. In addition to the hydrogen supplied from the reformer 11, water and air are supplied to the fuel cell stack 12.

  In the fuel cell stack 12, hydrogen and oxygen O2 in the air react to generate electrical energy and water. Since this reaction is an exothermic reaction, the generated heat is recovered by an exhaust heat recovery device (not shown) to generate hot water. Since the generated electrical energy is direct current, it is converted into alternating current by the inverter 13 and supplied to the consumer. Various sensors are installed in the reformer 11 and the fuel cell stack 12 to measure temperature, voltage, current, flow rate, and the like.

  Next, the operation database 2 will be described. The operation database 2 stores measured values representing the state of the fuel cell system 1 such as the cell voltage and the stack temperature measured by the fuel cell system 1. The measured value representing the state of the fuel cell system 1 includes “anode gas humidity”, “cathode gas humidity”, “reformer temperature”, “battery current”, and “continuous operation time” in addition to “battery voltage” and “battery cell temperature”. These measured values are stored at 1 second intervals.

  Since the data is transmitted to the state diagnosis means 4 at the same time as storing, no measurement delay is caused by the operation database 2 at the time of diagnosis. A part of the data in the operation database 2 is transferred to the stop history data extracting means 3 to extract stop history data. The extracted stop history data is stored in the operation database 2. The stop history data will be described in the description of the stop history data extraction unit 3.

  Next, the stop history data extraction unit 3 will be described. The stop history data extraction unit 3 extracts stop history data of the fuel cell system 1 from the operation database 2. The stop history data is operation data and its feature amount during the stop operation and during the stop period of the fuel cell system.

  One of the reasons that the voltage of the fuel cell fluctuates upon restarting is considered to be that the electrode is activated by the operation of stopping the fuel cell and the environment in which the cell is placed during the stop period.

  As a method for stopping the fuel cell stack, for example, after inserting a simulated load for consuming hydrogen remaining in the stack, the service load is cut off, and then the supplied reformed gas is stopped. There are several methods, such as a method of removing residual hydrogen in the stack by purging with nitrogen while cutting.

  In Example 1, “simulated load insertion time”, “nitrogen purge time”, and “nitrogen flow rate” were extracted as condition parameters related to these stopping methods. However, the data to be extracted is not limited to these, and if the stopping method is different, operation conditions different from these may be extracted.

  In the first embodiment, “stop time”, “stop period average cell temperature”, “stop period final cell temperature”, “stop period average humidity”, and “stop period final humidity” are extracted as feature values of the operation data during the stop period. . However, in the case where a hydrogen concentration sensor and an oxygen concentration sensor are installed in the fuel cell system 1, “average hydrogen concentration” and “average oxygen concentration” may be extracted in addition to the feature amount.

  Next, the state diagnosis unit 4 will be described. The state diagnosing means 4 is a state diagnosing method using an adaptive resonance theory (ART), and includes a preprocessing unit 41, an ART network 42, a determination unit 43, and a control unit 44.

  The preprocessing unit 41 converts the operation data in the operation database 2 into input data of the ART network 42. The ART network 42 classifies multidimensional input data into a plurality of categories according to the degree of similarity. The determination unit 43 determines the state of the fuel cell system 1 from the result of the category classified by the ART network 42. The control unit 44 controls the data flow of the entire state diagnosis unit 4 and sets internal parameters of each component.

  Next, an outline of the operation of the state diagnosis unit 4 will be described. The state diagnosis means 4 is a learning type diagnosis means, and there are a learning mode and a determination mode as operation modes. In the first embodiment, the operation in each mode will be described in the case where the data obtained in advance is only normal data.

  The learning mode is an off-line process, and the purpose is to classify the data obtained at the normal time by the ART network 42 and determine the range of the category that occurs at the normal time.

  First, learning data is extracted from the operation database 2. The learning data includes a measurement item indicating the current driving state and driving history information. The learning data is input to the state diagnosis unit 4 as a set of N-dimensional data at the same time. In the first embodiment, M sets of data are used as learning data.

  In the learning mode, the M sets of learning data are repeatedly input K times into the ART network 42 through the preprocessing unit 41. The ART network 42 can perform appropriate data classification when learning data is repeatedly input. The data classification standard is stored as a weighting factor that is a parameter inside the ART network 42. The range of the category classified at the normal time is passed to the determination unit and used as an abnormality detection standard.

  The determination mode is online processing. The data of the same item as the learning data is input to the state diagnosis means 4 one by one online. The input data is input to the ART network 42 via the preprocessing unit 41 and is classified into a certain category.

  The determination unit 43 determines that the category is normal if the classified category is the same as the category that is generated when the category is normal. However, if a category that has not occurred normally, that is, a new category has occurred, it is determined that the state of the fuel cell system 1 has changed, and it is determined that there is an abnormality.

  In the first embodiment, since the voltage data and the operation history information are input at the same time, “the relationship between the operation history and the voltage is different from that in the normal state even if the voltage value is a value that can be taken in the normal state”. It is possible to detect even in an abnormal state.

  When there is abnormal operation data from the beginning, the abnormal operation data may also be used as learning data, and the category in which the abnormal operation data is classified may be used as a criterion for abnormality determination.

  Next, each component of the state diagnosis unit 4 will be described in detail.

The preprocessing unit 41 converts the operation data in the operation database 2 into input data of the ART network 42. Since the value of the operation data varies depending on the measurement item, all measurement items Xi (1 = 1, 2,..., N) are normalized to a certain range of values. Measured value Xi
α ≦ Xi ≦ 1-α (i = 1,2, ..., n)
Normalized with. However, 0 ≦ α <0.5. Here, α = 0.1.

  The preprocessing unit 41 adds 1-Xi, which is its complement, to the n-dimensional data Xi to create 2n-dimensional data. This process is a process called complement coding, and is a method generally used in data classification by ART.

  The ART network 42 is a model simulating a human pattern recognition algorithm, and classifies multidimensional data into several categories according to the degree of similarity. A specific configuration and algorithm of the ART network 42 will be shown below.

  FIG. 3 is a diagram illustrating an example of the configuration of an ART network.

  The components of the ART network 42 include an F0 layer 421, an F1 layer 422, an F2 layer 423, and an Orienting Subsystem 424, and the four parties are coupled to each other. The F1 layer 422 and the F2 layer 423 are coupled via a weighting factor, and the weighting factor represents a prototype (prototype °) of a category into which input data is classified.

  Next, the operation of the ART network 42 will be described. The outline of the algorithm when input data is input to the ART network 42 is as shown in the following steps 1 to 5.

  Step 1: The F0 layer 421 is normalized so that the magnitude of the input vector becomes 1, and noise is removed.

  Step 2: An appropriate category candidate is selected by comparing the input data input to the F1 layer 422 with a weighting factor.

  Step 3: The validity of the category selected by Orienting Subsystem 424 is evaluated by comparison with parameter ρ. If it is determined to be valid, the input data is classified into the category, and the process proceeds to step 4. If it is not judged to be valid, the category is reset, and a suitable category candidate is selected from the other categories (the process of step 1 is repeated). Increasing the value of parameter ρ makes the category classification finer, and decreasing ρ makes the classification coarse.

  Step 4: When all existing categories are reset in Step 2, it is determined as a new category, and a new weighting factor representing a representative vector of the new category is generated.

  Step 5: When the input data is classified into category J, the weighting factor corresponding to category J is updated.

  Details of the ART algorithm are described in, for example, “GA Carpenter and S. Grossberg:“ ART2: Self-Organization of stable category recognition codes for analog input patterns ”, Applied Optics, Vol 26, No 23, (1987)”. ing.

Next, the determination unit 43 will be described. The determination unit 43 determines the state of the fuel cell system 1 from the result of the category classified by the ART network 42. In the first embodiment, since the learning data is only normal data, when a category (new category) other than the category generated when the learning data is classified occurs, the state of the device may be abnormal. high. However, since the influence of measurement errors and the like is also considered, the criterion for detecting an abnormality was calculated as the new category occurrence frequency every 10 seconds and defined as “the time when the occurrence frequency is X or more continues for Y seconds or more”. In the first embodiment, X = 0.2 and Y = 30. However, for example, when the occurrence frequency is high, it may be determined that there is an abnormality in a shorter time (example: X = 0.5. Y = 10).
Next, the control unit 44 will be described. The control unit 44 has three functions: (1) an internal parameter setting function, (2) an operation mode switching function, and (3) a data flow control function.

  The internal parameter setting function is a function for setting internal parameters of the preprocessing unit 41, the ART network 42, and the determination unit 43. For example, parameters for determining the normalization upper and lower limit values of the preprocessing unit 41 and the determination criteria of the determination unit 43 are set.

  The operation mode switching function switches the operation mode of the state diagnosis means to either the learning mode or the determination mode.

  The data flow control function controls the data flow in each operation mode of the learning mode and the determination mode. The data flow in each mode is as shown in the description of the operation of the state diagnosis means 4.

  Next, diagnosis results and effects of the present invention will be described.

  FIG. 4 is a diagram illustrating an example of trend data of battery voltage.

  In this case, there are three stop periods, and the first stop history data (stop history A) is different from the second stop history data (stop history B), but the second stop history and the third stop history. Are the same. Here, the stop history data refers to “simulation load insertion time”, “nitrogen purge time”, “nitrogen flow rate” which is a stop condition, and “stop time” “stop period” which is a feature amount of operation data during the stop period. The average cell temperature, the final cell temperature during the stop period, the average humidity during the stop period, and the final humidity during the stop period.

  Both the stop histories A and B are data that have been experienced in the past, and are used as learning data. In the example of FIG. 4, the battery voltage is increased from V0 to V1 at the time of restart after the first and third stop periods. Further, at the time of restart after the second stop period, the battery voltage is increased from V0 to V2. In the case of such operation data, in the conventional diagnosis method, it was diagnosed that the battery voltage fluctuates between V0 and V2 because it is a behavior that can occur during normal operation.

  FIG. 5 is a diagram showing a result of classifying operation data including the battery voltage data of FIG. 4 into categories.

  The vertical axis represents the category number, and the lower part of the broken line is a category generated when normal operation data (learning data) is classified. In FIG. 5, the driving data up to the third stop period is the same as the category that occurred during learning, but after the third stop period, the category that did not occur when the learning data was classified (new Category) has occurred. This means that after the third stop period, a behavior that does not exist in the learned data has occurred.

  The second stop history data and the third stop history data are exactly the same, but the value of the battery voltage is different. That is, the characteristics of the battery have changed and there is a high possibility that some abnormality has occurred.

  In the first embodiment, as described above, even if the value of the battery voltage is a value that can be taken so far, a change in the characteristics of the battery can be captured in consideration of the relationship with the stop history data.

  It is known that the effect of increasing the restart voltage is reduced with the operation time. In the first embodiment, “continuous operation time” is included in the input data to the state diagnosis means 4.

  Therefore, even if the operation time is about the same, if the degree of voltage drop differs due to some abnormality, the ART network 42 can be classified into different categories and detect that the state has changed.

  As described above, even if the voltage behavior itself is the same as normal, if the relationship with the stop history data or continuous operation time is different from normal, the change can be detected, and the abnormality is detected with high accuracy. It is possible.

  FIG. 6 is a block diagram showing a system configuration of a second embodiment of the diagnostic apparatus for a fuel cell system according to the present invention.

  The second embodiment includes an operation database 2 of the fuel cell system 1, a stop history data extraction unit 3, and a state diagnosis unit 4b.

  That is, the operation database 2 of the fuel cell system 1, the stop history data extracting means 3 for extracting stop history data that is operation data during the stop operation and / or the stop period of the fuel cell system 1 from the operation database 2, and the operation database 2 has a voltage estimation model 45 that estimates the voltage in the normal state from the information stored in 2, compares the estimated voltage with the actually measured voltage, and the state of the fuel cell system 1 at the time of restart The state diagnosis means 4b for diagnosing the above.

  The second embodiment differs from the first embodiment only in the state diagnosis means 4b. Therefore, the state diagnosis means 4b will be described.

  The state diagnosis unit 4b includes a voltage estimation model 45 and an abnormality diagnosis unit 46. The voltage estimation model 45 estimates a voltage in a normal state from information stored in the operation database 2.

  The abnormality diagnosis unit 46 compares the voltage estimated by the voltage estimation model 45 with the actually measured voltage to diagnose an abnormality.

  In the voltage estimation model 45, the voltage value in a normal state is estimated from information stored in the operation database 2. The information to be used is the stop history data extracted by the stop history data extracting means 3 and the measurement items indicating the current operating state.

  The stop history data includes “stop time”, “stop period cell temperature”, “stop period humidity”, and the like as described in the first embodiment.

  The measurement items indicating the current operation state are “battery cell temperature” “anode gas humidity” “cathode gas humidity” “reformer temperature” “battery current” “continuous operation time”. In Example 2, these 10 items were used as input items of the voltage estimation model, and the normal value of the voltage was estimated from these values. The model used for estimation is a neural network.

  FIG. 7 is a diagram illustrating an example of the structure of a neural network.

  This neural network consists of three layers: an input layer, an intermediate layer, and an output layer, and the nodes in each layer are connected by weighting factors. The number of nodes in the input layer is 10, which is the same as the number of input items, and the number of nodes in the output layer is one.

  Next, the operation of the voltage estimation model 45 will be described. In this neural network model, when input data is input to the input layer, calculation is performed using an internal weighting coefficient, and an estimated value of the voltage is output from the output layer.

  However, since the neural network is a learning model, the value of the internal weighting coefficient is determined by learning data given in advance. The back propagation method was used for learning. Specifically, it is as follows.

  First, M sets of input data and voltage data are prepared in advance as learning data from past operation data. With the weighting factor of the neural network set to an initial value, the first set of input data is input to the input layer. When the weighting factor is an initial value, there is a deviation between the voltage deviation data corresponding to the input data and the value actually output from the output layer of the neural network.

  Therefore, the weighting coefficient is corrected so that this deviation becomes small. When this operation is repeated N times for all M sets of data, the neural network can learn the relationship between the driving history information and the voltage.

  Details of this learning method are described in, for example, “The MIT Press, Neurocomputing Foundations of Research, 1988, pp 318-362”.

  In this way, when unlearned input data is input to the neural network that has learned the relationship between the driving history information and the correct voltage from past driving data, the voltage value at that time can be estimated.

  In the second embodiment, a neural network is used as the voltage estimation model, but other statistical models such as a multiple regression equation and an ARX model may be used.

  Next, the abnormality diagnosis unit 46 will be described. The abnormality diagnosis unit 46 receives the measured value of the battery voltage from the operation database 2 and the estimated value of the voltage from the voltage estimation model 45. The abnormality diagnosis unit 46 compares both values and diagnoses the presence or absence of abnormality. In the second embodiment, if the deviation between the two is larger than a predetermined reference value, it is determined that there is an abnormality.

  Since the degree of abnormality differs depending on the deviation value, it is possible to provide a reference value for each degree of abnormality (light failure, serious failure, etc.) and make a diagnosis including the degree of abnormality. In addition, when the value of deviation differs depending on the type of abnormality, it is possible to provide a reference value for each type of abnormality and make a diagnosis including the type of abnormality.

  If the battery voltage is estimated from the stop history data and the current operation data, the battery voltage can be estimated with high accuracy, and the abnormality can be accurately diagnosed by comparing this estimated value with the actually measured battery voltage. Become.

  FIG. 8 is a block diagram showing a system configuration of Embodiment 3 of the diagnostic apparatus for the fuel cell system according to the present invention.

  The third embodiment includes an operation database 2 of the fuel cell system 1, a stop history data extraction unit 3, a state diagnosis unit 4, and a voltage correction amount calculation unit 5.

  That is, the operation database 2 of the fuel cell system 1, the stop history data extracting means 3 for extracting stop history data that is operation data during the stop operation and / or the stop period of the fuel cell system 1 from the operation database 2, and the stop history Voltage correction amount calculation means 5 for calculating a voltage correction value at the time of restart from data, and a state diagnosis means for diagnosing the state of the fuel cell system 1 at the time of restart based on the voltage correction value and the operation data at the time of restart It consists of four.

  The third embodiment differs from the first embodiment in the input data to the voltage correction amount calculation means 5 and the state diagnosis means 4. Therefore, input data to the voltage correction amount calculation means 5 and the state diagnosis means 4 will be described.

  The voltage correction amount calculation means 5 calculates the voltage correction amount from the stop history data extracted by the stop history data extraction means and the “continuous operation time” recorded in the operation database 2. The voltage correction amount is a change in voltage due to the stop, and has the largest value at the time of startup. However, this effect tends to decrease over time.

  Therefore, in the third embodiment, the voltage correction amount is estimated from the stop history data and the “continuous operation time”. Here, the amount of voltage correction was estimated using a neural network. The input data includes four items of “stop time”, “stop period cell temperature”, “stop period humidity”, and “continuous operation time”, which are stop history data, and the output is a voltage correction amount. Since the learning and estimation method using the neural network has been described in the second embodiment, the description thereof will be omitted.

  Input data to the state diagnosis means 4 includes “battery cell temperature”, “anode gas humidity”, “cathode gas humidity”, “reformer temperature”, “battery current” indicating the state of the fuel cell system 1, and “corrected battery”. Voltage ". These are classified into categories by the ART network 42, and the state is determined. The “corrected battery voltage” is a value obtained by removing the influence of the stop operation and the stop period from the actually measured “battery voltage”.

  Therefore, the state diagnosing means 4 can perform diagnosis with data excluding the influence of the stop history of the fuel cell system 1.

  In the third embodiment, since the influence of the stop period, which has been difficult to diagnose in the conventional method, can be removed in advance by the voltage correction amount calculation means 5, the state of the fuel cell system can be diagnosed with high accuracy.

  In the second embodiment, since only the correction amount of the battery voltage due to the influence of the stop period is calculated, when the battery deteriorates over time, the category gradually changes, and the state of deterioration over time can be captured.

  On the other hand, as an abnormality diagnosis system, there is a case where it is desired to detect only an abnormality excluding aged deterioration. In this case, when “total battery usage time” is input to the voltage correction amount calculation means 5, it is possible to calculate the voltage correction amount considering aging degradation.

  The amount of voltage decrease due to aging may be known in advance, for example, x [V] / 1000 hours. In this case, the total voltage correction amount may be calculated by adding the voltage drop due to aging deterioration to the voltage correction amount obtained by the neural network that does not input the “total battery usage time”. When the influence of aging deterioration is corrected in this way, only abnormal events excluding aging deterioration can be detected.

  FIG. 9 is a block diagram showing a system configuration of embodiment 4 of the diagnostic apparatus for the fuel cell system according to the present invention.

  The fourth embodiment includes an operation database 2 of the fuel cell system 1, a stop history data extraction unit 3, a state diagnosis unit 4 b, and a voltage correction amount calculation unit 5. This voltage correction amount calculation means 5 is the same as that of the third embodiment.

  That is, the operation database 2 of the fuel cell system 1, the stop history data extracting means 3 for extracting stop history data that is operation data during the stop operation and / or the stop period of the fuel cell system 1 from the operation database 2, and the stop history The voltage correction amount calculation means 5 for calculating the voltage correction value at the time of restart from the data, and the voltage estimation model 45 for estimating the voltage in the normal state from the information stored in the operation database 2, the estimated voltage Is compared with the voltage correction value, and the state diagnosis means 4b for diagnosing the state of the fuel cell system 1 at the time of restart.

  The fourth embodiment differs from the second embodiment in the input data to the voltage correction amount calculation means 5 and the state diagnosis means 4b. Therefore, here, input data to the state diagnosis means 4b will be described.

  Inputs to the state diagnosis means 4b are “battery cell temperature”, “anode gas humidity”, “cathode gas humidity”, “reformer temperature”, “battery current”, “continuous operation time”, which are the state quantities of the fuel cell system 1. “Corrected battery voltage”.

  The former state quantity is input to the voltage estimation model 45. The voltage estimation model 45 estimates the battery voltage by a neural network. The estimated battery voltage is sent to the abnormality diagnosis unit 46. The abnormality diagnosis unit 46 diagnoses the state of the battery based on the deviation between the estimated voltage and the “corrected voltage”. Since the diagnosis method is the same as that of the second embodiment, the description thereof is omitted.

  In the second embodiment, the estimated value of the battery voltage is obtained from the stop history data and the state quantity of the fuel cell system 1. On the other hand, in the fourth embodiment, the part for calculating the voltage correction amount from the stop history data and the part for calculating the estimated value of the voltage from the state quantity of the fuel cell system 1 are separated. Therefore, in the fourth embodiment, in addition to the effect of the second embodiment in which the battery state can be diagnosed with high accuracy, it is easy to verify the validity of the voltage correction amount calculated from the stop history data.

  Since no stop history data is input to the voltage estimation model, the model is simplified.

It is a block diagram which shows the system | strain structure of Example 1 of the diagnostic apparatus of the fuel cell system by this invention. It is a figure which shows the outline | summary of the fuel cell system used as a diagnostic object. It is a figure which shows an example of a structure of an ART network. It is a figure which shows an example of the trend data of a battery voltage. It is a figure which shows the result of having classified the operation data containing the battery voltage data of FIG. 4 into the category. It is a block diagram which shows the system | strain structure of Example 2 of the diagnostic apparatus of the fuel cell system by this invention. It is a figure which shows an example of a structure of a neural network. It is a block diagram which shows the system | strain structure of Example 3 of the diagnostic apparatus of the fuel cell system by this invention. It is a block diagram which shows the system | strain structure of Example 4 of the diagnostic apparatus of the fuel cell system by this invention.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Fuel cell system 11 Reformer 12 Fuel cell stack 13 Inverter 2 Operation database 3 Stop history data extraction means 4 State diagnosis means 4b State diagnosis means 41 Preprocessing part 42 ART network 43 Determination part 44 Control part 45 Voltage estimation model 46 Abnormality diagnosis unit 5 Voltage correction amount calculation means

Claims (10)

  A fuel cell system diagnosis method for diagnosing a state of a fuel cell system at the time of restart based on stop history data that is operation data during a stop operation and / or stop period of the fuel cell system and operation data at the time of restart. Calculate the voltage correction value at the time of restart from the stop history data that is the operation data during the stop operation and / or stop period of the fuel cell system,
A fuel cell system diagnosis method for diagnosing the state of the fuel cell system at the time of restart based on the voltage correction value and the operation data at the time of restart. Classifying the stop history data, which is operation data during the stop operation of the fuel cell system and / or the stop period, and the operation data at the time of restart into a plurality of categories according to the similarity of the data,
A fuel cell system diagnosis method for diagnosing the state of a fuel cell system from changes in classified categories. In the diagnostic method of the fuel cell system according to claim 3,
A method for diagnosing a fuel cell system, wherein the data classification method is adaptive resonance theory. Calculate the voltage correction value at the time of restart from the stop history data that is the operation data during the stop operation and / or stop period of the fuel cell system,
Based on the voltage correction value and the operation data at the time of restart, classify by a data classification method that classifies into a plurality of categories according to the similarity of data,
A fuel cell system diagnosis method for diagnosing the state of a fuel cell system from changes in classified categories.   At least one of simulated load insertion time, nitrogen purge time, stop time, stop cell temperature, stop humidity, stop hydrogen concentration, stop oxygen concentration and restart operation data during stop operation of the fuel cell system A fuel cell system diagnosis method for diagnosing the state of a fuel cell system at the time of restart based on the method. Fuel cell system operation database;
Stop history data extraction means for extracting stop history data that is operation data during a stop operation and / or a stop period of the fuel cell system from the operation database;
A fuel cell system diagnosis device comprising state diagnosis means for diagnosing the state of a fuel cell system at the time of restart based on stop history data and operation data at the time of restart. Fuel cell system operation database;
Stop history data extraction means for extracting stop history data that is operation data during a stop operation and / or a stop period of the fuel cell system from the operation database;
It has a voltage estimation model that estimates the voltage in the normal state from the information stored in the operation database, compares the estimated voltage with the actually measured voltage, and determines the state of the fuel cell system at the time of restart A diagnostic apparatus for a fuel cell system comprising state diagnostic means for diagnosing. Fuel cell system operation database;
Stop history data extraction means for extracting stop history data that is operation data during a stop operation and / or a stop period of the fuel cell system from the operation database;
Voltage correction amount calculating means for calculating a voltage correction value at restart from stop history data;
A diagnostic apparatus for a fuel cell system, comprising state diagnosis means for diagnosing the state of the fuel cell system at the time of restart based on the voltage correction value and the operation data at the time of restart. Fuel cell system operation database;
Stop history data extraction means for extracting stop history data that is operation data during a stop operation and / or a stop period of the fuel cell system from the operation database;
Voltage correction amount calculating means for calculating a voltage correction value at restart from stop history data;
It has a voltage estimation model that estimates the normal state voltage from the information stored in the operation database, compares the estimated voltage with the voltage correction value, and diagnoses the state of the fuel cell system at the time of restart A diagnostic device for a fuel cell system comprising state diagnostic means.

Images