Hydrogen production system and method for controlling hydrogen production system

Abstract

To provide a water electrolysis system capable of detecting abnormality of a water electrolysis cell and discriminating the detected abnormality.SOLUTION: A hydrogen production system is configured to comprise a water electrolysis system including a water electrolysis cell and an abnormality discrimination system for discriminating the abnormal mode of the water electrolysis system. The abnormality discrimination system includes a data acquisition part for acquiring data of the temporal change of at least one of temperature, water supply amount, current, and voltage of the water electrolysis cell. The abnormality discrimination system includes an abnormality detection part for detecting abnormality of the water electrolysis cell based on the data acquired by the data acquisition part, and an abnormality mode discrimination part for discriminating the abnormal mode of the water electrolysis cell where the abnormality occurred. The abnormal mode discrimination part discriminates an abnormal mode from a machine learning model obtained by learning the behavior of current or voltage in a plurality of abnormal modes and the data acquired by the data acquisition part.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to a hydrogen production system and a control method for the hydrogen production system. [Background technology]

[0002] In a hydrogen production system, multiple water electrolysis stacks are connected in multiple series or multiple series and multiple parallel configurations. In such a hydrogen production system, a voltmeter is installed in each of the water electrolysis cells that are linked together to form a system. In the hydrogen production system, each water electrolysis cell is individually managed and abnormalities are detected. However, when a hydrogen production system becomes large-scale, the number of stacks of water electrolysis cells also becomes enormous, and installing a voltmeter for each water electrolysis cell increases installation costs, including maintenance costs. A method using a machine learning model has been proposed as a method for reducing the maintenance costs of a hydrogen production system (see, for example, Patent Document 1). In this method, a machine learning model trained in advance is used to generate anomaly prediction results. A maintenance plan is then created based on the generated anomaly prediction results, and the hydrogen production system is operated. This technique also creates a machine learning model by learning in advance the relationship between the measured state of the water electrolysis cell and the deterioration state of the water electrolysis cell. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] International Publication No. 2002 / 075767 Summary of the Invention [Problem to be solved by the invention]

[0004] The results of anomaly prediction using the above-mentioned machine learning model make it possible to predict the lifespan of a water electrolysis cell. However, the above-mentioned method cannot determine the abnormality mode (type of anomaly) of an anomaly actually detected in a hydrogen production system. For example, it cannot determine whether the detected anomaly is one that requires the hydrogen production system to be stopped immediately, or one that is causing gradual deterioration of the water electrolysis cell and does not require an emergency shutdown. This makes it difficult to implement appropriate control in response to a detected anomaly. Therefore, there is a need for a water electrolysis system that is capable of detecting an abnormality in a water electrolysis cell and discriminating between the detected abnormality modes.

[0005] In order to solve the above-mentioned problems, the present invention provides a water electrolysis system capable of detecting an abnormality in a water electrolysis cell and distinguishing the detected abnormality, and a method for controlling the water electrolysis system.

[0006] The above and other objects of the present invention and novel features of the present invention will become apparent from the description of this specification and the accompanying drawings. [Means for solving the problem]

[0007] The hydrogen production system of the present invention includes a water electrolysis system including a water electrolysis cell, and an abnormality determination system that determines an abnormality mode of the water electrolysis system. The abnormality determination system includes a data acquisition unit that acquires data on time changes in at least one of the temperature, water supply amount, current, and voltage of the water electrolysis cell. The abnormality determination system also includes an abnormality detection unit that detects an abnormality in the water electrolysis cell based on the data acquired by the data acquisition unit, and an abnormality mode determination unit that determines the abnormality mode of the water electrolysis cell in which an abnormality has occurred. The abnormality mode determination unit determines the abnormality mode based on a machine learning model obtained by training the behavior of current or voltage in multiple abnormality modes and the data acquired by the data acquisition unit.

[0008] The present invention also provides a method for controlling a hydrogen production system, which includes a water electrolysis system with a water electrolysis cell, and an anomaly determination system that determines an anomaly mode of the water electrolysis system. Data on time changes in at least one of the temperature, water supply amount, current, and voltage of the water electrolysis cell is acquired, and an anomaly in the water electrolysis cell is detected based on the acquired data. The anomaly mode of the water electrolysis cell in which an anomaly has occurred is determined based on the acquired data and a machine learning model obtained by learning the behavior of the current or voltage in multiple anomaly modes. Effect of the Invention

[0009] Effects of the Invention The present invention provides a water electrolysis system capable of detecting an abnormality in a water electrolysis cell and distinguishing the detected abnormality mode, and a method for controlling a water electrolysis system.

[0010] Problems, configurations and effects other than those described above will become apparent from the following description of the embodiments. [Brief description of the drawings]

[0011] [Figure 1] FIG. 1 is a block diagram showing a configuration of a hydrogen production system. [Diagram 2] FIG. 11 is a block diagram showing the configuration of a modified example of the hydrogen production system. [Diagram 3] FIG. 1 is a diagram showing a configuration of a water electrolysis cell. [Figure 4] 4 is a flowchart of a method for determining an abnormality in a hydrogen production system. [Diagram 5] 13 is a flowchart of a method for generating an anomaly determination model. [Figure 6] 11 is a diagram showing the voltage behavior of an abnormal stack when the water supply is stopped, and the voltage behavior of a normal stack. FIG. [Figure 7] 11 is a diagram showing a difference ΔV between the voltage of an abnormal stack and the voltage of a normal stack. [Figure 8] 11 is a diagram showing the current behavior of an abnormal stack when the water supply is stopped, and the current behavior of a normal stack. FIG. [Figure 9]FIG. 13 is a diagram showing a difference ΔI between a current of an abnormal stack and a current of a normal stack. [Figure 10] FIG. 13 shows the voltage behavior in an abnormal mode when the electrode surface is covered with bubbles. [Figure 11] FIG. 1 is a diagram showing voltage behavior in an abnormal mode caused by electrode peeling and electrode cracking of a water electrolysis cell. [Figure 12] FIG. 13 is a diagram showing the behavior of the temperature of the water electrolysis stack when the water supply is stopped. [Figure 13] FIG. 13 is a diagram showing the behavior of the voltage of the water electrolysis stack when the water supply is stopped. [Figure 14] FIG. 4 is a diagram showing behavior of the total amount of water supplied to the water electrolysis device. [Figure 15] FIG. 13 is a diagram showing the behavior of the current value of the water electrolysis stack when the water supply is stopped. [Figure 16] FIG. 13 is a diagram showing the temperature behavior of a water electrolysis stack as the temperature increases. [Figure 17] FIG. 13 is a diagram showing the voltage behavior of a water electrolysis stack with increasing temperature. [Figure 18] FIG. 4 is a diagram showing behavior of the total amount of water supplied to the water electrolysis device. [Figure 19] FIG. 1 is a diagram showing the behavior of the current value of a water electrolysis stack as the temperature increases. [Figure 20] FIG. 4 is a diagram showing a life prediction model generated by a life prediction model generating unit. [Figure 21] FIG. 1 is a diagram showing a method for installing a voltmeter in a water electrolysis apparatus. [Figure 22] FIG. 1 is a diagram showing a method for installing an ammeter in a water electrolysis apparatus. [Figure 23] 1 is a flowchart of a procedure for detecting an abnormality in a water electrolysis stack in a multi-series and multi-parallel stack group, determining an abnormality mode, and replacing the water electrolysis stack. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0012] Hereinafter, an embodiment of a hydrogen production system and a control method for a hydrogen production system according to the present invention will be described with reference to the drawings. The following description shows a specific example of the contents of the present invention, and the present invention is not limited to these descriptions. Various changes and modifications can be made by those skilled in the art within the scope of the technical ideas disclosed in this specification. Furthermore, in all the drawings for explaining the present invention, parts having the same functions are given the same reference numerals, and repeated explanations thereof may be omitted.

[0013] In this specification, the word "to" is used to mean that the numerical values ​​before and after it are included as the upper and lower limits. In the numerical ranges described in stages in this specification, the upper or lower limit value described in one numerical range may be replaced with the upper or lower limit value described in another stage. The upper or lower limit value of the numerical ranges described in this specification may be replaced with the value shown in the examples.

[0014] <Hydrogen production system> The configuration of the hydrogen production system will be described below. The configuration of the hydrogen production system is shown in Fig. 1. The hydrogen production system 1 shown in Fig. 1 includes a water electrolysis system 10 and an abnormality determination system 20. The apparatus 10 includes a water electrolysis device 11, an electrolysis control unit 12, a storage unit 13, and a display unit 14. The electrolysis control unit 12 includes an apparatus control unit 121 and a control condition determination unit 122 as functional components. The abnormality determination system 20 includes a determination control unit 21 and a storage unit 22. The determination control unit 21 includes a data acquisition unit 211, an abnormality detection unit 212, a feature generation unit 213, and an abnormality mode determination unit 214.

[0015] [Water electrolysis system] The water electrolysis apparatus 11 includes a water electrolysis cell. The water electrolysis apparatus 11 includes a plurality of water electrolysis cells. The water electrolysis apparatus 11 may include a water electrolysis cell stack in which a plurality of water electrolysis cells are stacked. In the following description, unless there is a particular need to distinguish between them, the water electrolysis cell and the water electrolysis cell stack will be collectively referred to as the water electrolysis cell. Furthermore, the water electrolysis cells or the water electrolysis cell stack may be connected in series or parallel, or in multiple series or multiple series and multiple parallel configurations. The water electrolysis device 11 also includes auxiliary devices such as a temperature regulator for the water electrolysis cell and a water supply mechanism. Furthermore, the water electrolysis apparatus 11 includes various sensors such as a thermometer, a flow meter, a voltmeter, and an ammeter.

[0016] The electrolysis control unit 12 performs overall control of the water electrolysis system 10. The electrolysis control unit 12 is composed of, for example, a central processing unit (CPU), a read only memory (ROM), and a random access memory (RAM), all of which are not shown. The CPU is an example of a calculation unit related to the water electrolysis system 10. The CPU performs centralized control of the operation of each unit of the water electrolysis system 10. The CPU reads out program code of software related to various processes of the water electrolysis system 10, which is stored in a ROM (an example of a recording medium), and loads the program code into the RAM. The CPU then controls the water electrolysis system 10 in accordance with the loaded program. Note that the electrolysis control unit 12 may include another calculation device, such as an MPU (micro processing unit), as a calculation unit instead of a CPU.

[0017] The storage unit 13 is an example of a recording medium made of a non-volatile storage, etc. The storage unit 13 stores various processing programs executed by the CPU, programs such as an OS, information on the functions of the water electrolysis system 10 required to execute the programs, various data tables used for calculations by the electrolysis control unit 12, etc. The storage unit 13 is, for example, a hard disk drive (HDD), a solid state drive (SSD), a magnetic tape, etc.

[0018] The device control unit 121, which is a functional component of the electrolysis control unit 12, controls the operation of the water electrolysis device 11. For example, it controls the voltage applied to the water electrolysis cells of the water electrolysis device 11 and the amount of current supplied to the water electrolysis cells. The device control unit 121 also controls the operation of auxiliary devices such as a temperature regulator for the water electrolysis cell and a water supply mechanism. The device control unit 121 controls each auxiliary device to adjust the temperature of the water electrolysis cell, the temperature of water supplied to the water electrolysis cell, and the amount of water supplied. The device control unit 121 may control multiple water electrolysis cells individually. The device control unit 121 may also control multiple auxiliary devices individually. The control conditions used by the electrolysis control unit 12 are stored in the memory unit 13 described above. The device control unit 121 communicates with various sensors, such as a thermometer, a flow meter, a voltmeter, and an ammeter. The device control unit 121 acquires data from the various sensors. The device control unit 121 stores the acquired data, for example, in the above-mentioned storage unit 13. The device control unit 121 acquires information individually for each of the multiple sensors arranged in the water electrolysis apparatus 11, and stores the information in the storage unit 13. In addition, the device control unit 121 controls the operation of the display unit 14 .

[0019] The control condition determination unit 122, which is a functional component of the electrolysis control unit 12, determines whether to change the control conditions of the water electrolysis device 11 by the device control unit 121. The control condition determination unit 122 makes a determination when the abnormality determination system 20 detects an abnormality in the water electrolysis system 10. When the control condition determination unit 122 receives a notification of abnormality detection from the abnormality determination system 20, it checks the detected abnormality mode, the state of the water electrolysis cell in which the abnormality was detected, the life prediction model, etc. Specifically, the control condition determination unit 122 acquires various information including the abnormality mode and the life prediction model from the abnormality mode determination unit 214 and the life prediction model generation unit 215 of the determination control unit 21 of the abnormality determination system 20. Then, the control condition determination unit 122 extracts the abnormality mode, the life prediction model, the state of the water electrolysis cell, etc. from the acquired information.

[0020] The control condition determination unit 122 then determines whether to change the control conditions to be applied by the device control unit 121 based on the acquired abnormality mode, life prediction model, state of the water electrolysis cell, etc. If the control condition determination unit 122 determines to change the control conditions, it transmits new control conditions corresponding to the abnormality mode and state of the water electrolysis cell to the device control unit 121. The determination criteria for changing the control conditions and the new control conditions are stored in the memory unit 13. The control condition determination unit 122 reads out the determination criteria and various control conditions from the memory unit 13 and uses them.

[0021] The display unit 14 is composed of a display such as a liquid crystal display device. The display unit 14 displays various types of information on the display screen in accordance with instructions of a display signal input from v12. For example, the display unit 14 displays the operating status of the water electrolysis cell of the water electrolysis system 10, the control status of the water electrolysis cell and each auxiliary device, data acquired from various sensors, etc. in accordance with instructions from the electrolysis control unit 12. Furthermore, the display unit 14 displays the type of abnormality (abnormal mode) detected by the abnormality determination system 20, the position of the water electrolysis cell where the abnormality occurred, the state of the water electrolysis cell, the generation result of a life prediction model, etc. in accordance with instructions from the electrolysis control unit 12.

[0022] [Abnormality Detection System] The discrimination control unit 21 of the abnormality discrimination system 20 centrally controls the abnormality discrimination system 20. The discrimination control unit 21 is composed of, for example, a CPU, a ROM, a RAM, etc., which are not shown. The CPU is an example of a calculation unit related to the abnormality discrimination system 20. The CPU centrally controls the operation of each unit of the abnormality discrimination system 20. The CPU reads out program code of software related to various processes of the abnormality discrimination system 20 stored in a ROM (an example of a recording medium) and expands it into the RAM. The CPU then controls the abnormality discrimination system 20 in accordance with the expanded program. Note that the discrimination control unit 21 may include another calculation device, such as an MPU, as a calculation unit instead of a CPU.

[0023] The storage unit 22 is an example of a recording medium made of a non-volatile storage, etc. The storage unit 113 stores various processing programs executed by the CPU, programs such as an OS, information on the functions of the abnormality determination system 20 required to execute the programs, various data tables used for calculations by the determination control unit 21, etc. The storage unit 22 is, for example, an HDD, an SSD, a magnetic tape, etc.

[0024] The data acquisition unit 211, which is a functional component of the discrimination control unit 21, acquires data from the water electrolysis system 10. In particular, the data acquisition unit 211 acquires data obtained by the device control unit 121 from various sensors (e.g., ammeter, voltmeter) arranged in the water electrolysis device 11. The device control unit 121 acquires, for example, the voltage applied to the water electrolysis cell and the amount of current supplied to the water electrolysis cell. In addition, the device control unit 121 acquires, for example, the temperature of the water electrolysis cell, temperature adjustment of water supplied to the water electrolysis cell, and the amount of water supplied. The data acquiring unit 211 acquires data for each sensor individually from various sensors arranged in the water electrolysis apparatus 11. For example, the data acquiring unit 211 acquires voltage values ​​and current values ​​individually from a plurality of voltmeters and ammeters arranged in the water electrolysis apparatus 11. The device control unit 121 stores the acquired various data in the storage unit 22.

[0025] The abnormality detection unit 212, which is a functional component of the determination control unit 21, detects an abnormality in the water electrolysis device 11. The abnormality detection unit 212 detects an abnormality in the water electrolysis device 11 based on data acquired by the data acquisition unit 211. Specifically, the abnormality detection unit 212 detects an abnormality in the water electrolysis device 11 based on a change over time in at least one of the current and the voltage acquired by the data acquisition unit 211. Furthermore, when the anomaly detection unit 212 detects an anomaly in the water electrolysis apparatus 11, it detects the position of the water electrolysis cell in which the anomaly has occurred. For example, the anomaly detection unit 212 detects the position of the water electrolysis cell in which the anomaly has occurred by comparing output values ​​from multiple sensors arranged in the water electrolysis apparatus 11. For example, the anomaly detection unit 212 compares the output values ​​of multiple voltmeters arranged in the water electrolysis apparatus 11 for each voltmeter and detects an anomaly in the output value of a specific voltmeter. In this way, the anomaly detection unit 212 detects the position of the water electrolysis cell in which the anomaly has occurred based on the arrangement of the voltmeter in which the anomaly output value has been detected. In addition, for example, the anomaly detection unit 212 compares the output values ​​of multiple ammeters arranged in the water electrolysis apparatus 11 for each ammeter and detects an anomaly in the output value of a specific ammeter. In this way, the anomaly detection unit 212 detects the position of the water electrolysis cell in which the anomaly has occurred based on the arrangement of the ammeter in which the anomaly output value has been detected.

[0026] The method by which the anomaly detection unit 212 detects an anomaly is not particularly limited. For example, the anomaly detection unit 212 detects an anomaly by simply comparing a current value or a voltage fluctuation with a threshold value (upper and lower limits). The anomaly detection unit 212 may also calculate the degree of change in time-series data, and detect an anomaly when the degree of change exceeds a set threshold value. The degree of change may be calculated using, for example, a singular spectrum transform method or a k-nearest neighbor method. It is preferable to set the threshold value to a level that indicates the possibility of an anomaly, rather than a dangerous value that immediately shuts off an abnormal stack line.

[0027] The feature generator 213, which is a functional component of the discrimination control unit 21, acquires various data of the water electrolysis cell in which the anomaly detection unit 212 has detected an anomaly. Hereinafter, the various data of the water electrolysis cell in which the anomaly detection unit 212 has detected an anomaly will be referred to as anomaly data. The feature generator 213 acquires a data set of the anomaly data. The data set includes, for example, at least one of time changes of a current and a voltage. The data set may also include at least one of time changes of a temperature of the water electrolysis cell and an amount of water supplied to the water electrolysis cell. The feature generator 213 acquires, for example, time series data for a predetermined period from the occurrence of an anomaly as the time change of each piece of data included in the data set. The feature generator 213 may also acquire, for example, time series data for a predetermined period going back from the occurrence of an anomaly as the time change of each piece of data included in the data set. The time going back is not particularly specified, but is preferably the same time range as the time series data previously learned by an anomaly discrimination model for discriminating an anomaly mode, which will be described later. Similarly, when acquiring time series data for a certain period of time after anomaly detection, it is preferable that the time range is the same as that of the pre-learned time series data.

[0028] Then, the feature generator 213 extracts features from the acquired data set. The feature is a value mathematically calculated from a characteristic behavior of each data. The feature generator 213 acquires, as the feature, for example, the gradient of the regression line of the abnormal data, the residual or sum of squares of the regression line, skewness, kurtosis, maximum value, minimum value, and the difference between them. The feature generation unit 213 stores the acquired data set and the extracted features in the storage unit 22.

[0029] The abnormality mode discrimination unit 214, which is a functional component of the discrimination control unit 21, discriminates the type of abnormality (abnormality mode) that has occurred in the water electrolysis apparatus 10. The abnormality mode discrimination unit 214 discriminates the abnormality mode using an abnormality discrimination model stored in the storage unit 22. The abnormality mode discrimination unit 214 also applies the feature amounts of the abnormality data extracted by the feature amount generation unit 213 to the abnormality discrimination model to discriminate the abnormality mode. The abnormality discrimination model used by the abnormality mode discrimination unit 214 is a machine learning model stored in advance in the storage unit 22. The machine learning model is created using a machine learning algorithm described below. The abnormal mode determination unit 214 then transmits the determined abnormal mode to the control condition determination unit 122 of the water electrolysis device 10. The abnormal mode determination unit 14 may transmit various data related to the abnormal mode together with the abnormal mode to the control condition determination unit 122. The various data related to the abnormal mode include the voltage, the amount of current, the temperature of the water electrolysis cell, the temperature of water supplied to the water electrolysis cell, and the amount of water supplied. The abnormal mode determination unit 214 may also transmit time-series data of these data for a predetermined period to the control condition determination unit 122.

[0030] [Modification of hydrogen production system] Next, a modified example of the hydrogen production system will be described. Fig. 2 shows the configuration of the modified hydrogen production system. The modified hydrogen production system shown in Fig. 2 differs from the hydrogen production system 1 shown in Fig. 1 described above only in the parts related to the functional configuration of the determination control unit 21 of the abnormality determination system 20. For this reason, detailed description of the same configuration as the hydrogen production system 1 shown in Fig. 1 described above will be omitted.

[0031] 2, the discrimination control unit 21 of the anomaly discrimination system 20 includes a life prediction model generation unit 215. The life prediction model generation unit 215 included in the discrimination control unit 21 as a functional component acquires the abnormality mode of the water electrolysis cell in which an abnormality has been detected from the abnormality mode discrimination unit 214. The life prediction model generation unit 215 may also acquire data on time series changes in voltage and current value of the water electrolysis cell in which an abnormality has been detected, in addition to the abnormality mode. The life prediction model generation unit 215 may also acquire data on time series changes in the water supply amount, temperature, and the like of the water electrolysis cell. The life prediction model generation unit 215 acquires these data from the storage unit 22. Then, the life prediction model generation unit 215 obtains a prediction formula corresponding to the abnormality mode from the storage unit 22. The prediction formula is pre-stored in the storage unit 22. Then, the life prediction model generation unit 215 substitutes various data (voltage, current, etc.) into the obtained prediction formula to generate a life prediction model for the water electrolysis cell in which an abnormality has been detected.

[0032] The life prediction model generation unit 215 outputs the generated life prediction model to the water electrolysis system 10. Specifically, the life prediction model generation unit 215 transmits the generated life prediction model to the control condition determination unit 122 of the electrolysis control unit 12. In the hydrogen production system 1 shown in FIG. 1, information on the abnormal mode and the like is transmitted from the abnormal mode discrimination unit 214 to the control condition judgment unit 122. In contrast, in the hydrogen production system 2 of the modified example shown in FIG. 2, only the life prediction model is transmitted from the life prediction model generation unit 215 to the control condition judgment unit 122. The control condition judgment unit 122 can judge whether or not to change the control conditions by the device control unit 121 based on only the life prediction model. For this reason, the hydrogen production system 2 does not need to transmit information on the abnormal mode and the like from the abnormal mode discrimination unit 214 to the control condition judgment unit 122. In addition, the hydrogen production system 2 may transmit the determined abnormal mode, various data, various information, and the like from the discrimination control unit 21 to the electrolysis control unit 12 or the control condition judgment unit 122 according to the life prediction model, as necessary. For example, the hydrogen production system 2 may transmit the abnormal mode and various data related to the abnormal mode from the abnormal mode discrimination unit 214 to the control condition judgment unit 122.

[0033] [Water electrolysis cell] Next, we will explain the water electrolysis cell of the hydrogen production system 1. The water electrolysis cell constitutes the water electrolysis apparatus 11. The configuration of the water electrolysis cell is shown in Figure 3. The water electrolysis cell 110 includes a cathode catalyst layer 111, an anode catalyst layer 112, a solid polymer membrane 113, a cathode current collector 114, an anode current collector 115, a power source 116, and separators 117 and 118.

[0034] The water electrolysis cell 110 includes a cathode current collector 114, a solid polymer membrane 113, and an anode current collector 115 stacked together. The cathode current collector 114 has a cathode catalyst layer 111 on the surface in contact with the solid polymer membrane 113. The anode current collector 115 has an anode catalyst layer 112 on the surface in contact with the solid polymer membrane 113. Therefore, the solid polymer membrane 113 is sandwiched between the cathode catalyst layer 111 and the anode catalyst layer 112. The cathode current collector 114 and the anode current collector 115 are, for example, porous bodies having electrical conductivity.

[0035] The solid polymer membrane 113 is ion-permeable, specifically, an ion exchange membrane. A known ion exchange membrane used in water electrolysis can be used as the solid polymer membrane 113. An example of the solid polymer membrane 113 is a solid polymer electrolyte membrane such as a perfluorocarbon resin into which an ion exchange group has been introduced.

[0036] Examples of the cathode catalyst layer 111 of the cathode current collector 114 include platinum, nickel, cobalt, silver, iron, and alloys of these metals. Examples of the anode catalyst layer 112 of the anode current collector 115 include iridium, platinum, nickel, cobalt, silver, iron, and alloys of these metals, as well as oxides of these metals, oxides of these alloys, and carbon.

[0037] The power supply 116 is a DC power supply and supplies power to the water electrolysis cell 110. The power supply 116 supplies power to the water electrolysis cell 110, thereby setting a hydrogen evolution electrode and an oxygen evolution electrode on the current collector. In Fig. 3, the cathode current collector 114 is the hydrogen evolution electrode, and the anode current collector 115 is the oxygen evolution electrode. The water electrolysis cell 110 electrolyzes water (H2O) using power supplied from a power source 116. The water electrolysis cell 110 decomposes water into hydrogen (H2) and oxygen (O2) through electrolysis. The water electrolysis cell 110 then recovers the hydrogen (H2) as an energy source.

[0038] The separators 117 and 118 have grooves formed therein through which water, hydrogen, oxygen, etc. flow. Water is supplied to the water electrolysis cell 110 from the separator 118. Furthermore, the water electrolysis cell 110 extracts unreacted water and oxygen produced by decomposition of water from the separator 118. Furthermore, the water electrolysis cell 110 extracts hydrogen produced by decomposition of water from the separator 117. Pure water is used as the water supplied to the water electrolysis cell 110, and in order to suppress deterioration of the membrane electrode assembly, it is preferable that the resistivity of the pure water is, for example, 10 MΩ·cm or more.

[0039] (Water electrolysis stack) The water electrolysis stack is formed by stacking the above-mentioned water electrolysis cells 110. In the hydrogen production system 1, the water electrolysis stacks are arranged in multiple series and multiple parallel configurations. This allows the hydrogen production system 1 to have a high output. Methods for connecting a large number of water electrolysis cells and a water electrolysis stack include the following methods.

[0040] The multi-series method is an arrangement in which all water electrolysis cells and water electrolysis stacks are connected in series in the same direction with both ends connected to a power source. By connecting the water electrolysis cells and water electrolysis stacks in multiple series, the current value of the power source can be reduced. Furthermore, from an industrial perspective, a multi-series connection does not require the installation of a transformer or the like for each stack. For this reason, a multi-series connection can reduce costs at a constant current and high voltage. Furthermore, a multi-series connection can save space. In addition, an arrangement in which a plurality of water electrolysis cells are connected in parallel, and a water electrolysis stack is connected in series, is called a multi-series / multi-parallel arrangement.

[0041] By simply connecting individual water electrolysis cells (water electrolysis stacks) in parallel, the voltage and current of each water electrolysis cell can be controlled individually. Therefore, even if a water electrolysis cell deteriorates due to long-term operation, a constant output can be maintained by increasing the voltage. Furthermore, when a water electrolysis cell malfunctions due to extreme deterioration or a short circuit, that cell can be disconnected from the system and then removed and replaced.

[0042] In the following description, an assembly of water electrolysis cells and water electrolysis stacks arranged in multiple series, parallel, and multiple series / multiple parallel configurations will be referred to as a stack group. Therefore, the stack group referred to in the following description may be any of the multiple series, parallel, and multi-series / multi-parallel connection configurations, unless there is a particular need to distinguish between them.

[0043] [Abnormality detection process] A specific method for determining an abnormality in a hydrogen production system will be described below with reference to a flowchart. A flowchart of the method for determining an abnormality in a hydrogen production system is shown in Fig. 4. The flowchart shown in Fig. 4 mainly describes the processing contents of the abnormality determination system 20 of the hydrogen production system 1.

[0044] First, in the discrimination control unit 21, the data acquisition unit 211 acquires data from the water electrolysis system 10 (step S10). Then, the data acquisition unit 211 stores the acquired data in the memory unit 22. The data acquisition unit 211 acquires at least one of the voltage and the amount of current of the water electrolysis cell.

[0045] Next, the anomaly detection unit 212 detects an anomaly in the water electrolysis apparatus 11 from the voltage and current amount acquired by the data acquisition unit 211 (step S11). For example, if the water electrolysis apparatus 11 is a stack group in which multiple water electrolysis cells are connected in series and parallel, the anomaly detection unit 212 detects an anomaly by checking the voltage of each water electrolysis cell. Furthermore, if it is difficult to install a voltmeter for each water electrolysis cell, the anomaly detection unit 212 detects an anomaly based on an increase or decrease in the value of the current flowing through each stack group connected in parallel.

[0046] Furthermore, in the case of the water electrolysis apparatus 11 in which a plurality of water electrolysis cells are arranged in a parallel direction, the anomaly detection unit 212 identifies the position of the water electrolysis cell in which an anomaly has occurred. Typically, a plurality of water electrolysis cells arranged in a parallel direction theoretically exhibit the same voltage behavior in the parallel direction. This also applies to the case of water electrolysis cells arranged in series. The voltage behavior of the water electrolysis cells arranged in series is the same in the parallel direction. However, the voltage behavior in the series direction including the water electrolysis cell in which an anomaly has occurred may be different from the voltage behavior of other normal water electrolysis cells in the parallel direction. For this reason, the anomaly detection unit 212 can identify the position of the water electrolysis cell in which an anomaly has occurred in the water electrolysis apparatus 11 by comparing the voltage behaviors of the water electrolysis cells arranged in the parallel direction.

[0047] Furthermore, the anomaly detection unit 212 may detect an anomaly in the water electrolysis cell using parameters such as the temperature of the water electrolysis cell, the amount of water supplied to the water electrolysis cell, and so on, in addition to the voltage and current values. For example, a stop in the water supply due to a breakdown in a supply device or a sudden change in the temperature of the water electrolysis cell due to an anomaly in a temperature regulator cannot be determined by the voltage and current values ​​alone. For this reason, the anomaly detection unit 212 may add the temperature of the water electrolysis cell, the amount of water supplied to the water electrolysis cell, and so on to the parameters for detecting an anomaly in the water electrolysis cell.

[0048] The anomaly detection unit 212 can also identify an anomaly based on the vibration pattern of the water electrolysis cells. The vibration pattern of the water electrolysis cells is obtained, for example, by installing a vibration sensor in each water electrolysis cell. The location where the vibration sensor is installed is not particularly limited as long as it is on the surface of the water electrolysis cell. From the viewpoint of measuring vibration waves, the vibration sensor is preferably installed on a surface parallel to the membrane electrode assembly. There is no particular limit to the type of the vibration sensor to be installed, and examples of the vibration sensor include frequency change type, piezoelectric type, electrodynamic type, servo type, eddy current type, capacitance type, and optical type vibration sensors.

[0049] Next, the feature generator 213 acquires a data set of abnormality data for the water electrolysis cell in which an abnormality has occurred (step S12). The feature generator 213 acquires the data set of abnormality data from the storage unit 22. Then, the feature generator 213 extracts features from the acquired data set of abnormal data. For example, the feature generator 213 acquires, as features, the gradient of the regression line of the abnormal data, the residual or sum of squares of the regression line, skewness, kurtosis, maximum value, minimum value, and the difference between them.

[0050] Next, the abnormality mode discrimination unit 214 discriminates the abnormality mode based on the feature extracted by the feature generation unit 213 (step S14). The abnormality mode discrimination unit 214 calculates the degree of abnormality by comparing with learning data previously stored in the storage unit 22, and discriminates the abnormality mode. The learning data is created, for example, by learning data determined to be normal in advance as normal data. The learning data can be created, for example, by the k-nearest neighbor method, which is one of the machine learning techniques.

[0051] Next, the abnormality mode determination unit 214 outputs the determined abnormality mode to the water electrolysis system 10. (Step S15). For example, the abnormal mode determination unit 214 outputs the determined abnormal mode to the control condition determination unit 122 of the water electrolysis system 10. The abnormal mode determination unit 214 may also transmit to the control condition determination unit 122, together with the determined abnormal mode, various data related to the abnormal mode, such as the voltage, the amount of current, the water electrolysis cell temperature, the supply water temperature, and the amount of supply water, as well as time-series data of these data for a predetermined period of time. After the abnormal mode is output to the water electrolysis system 10, the process according to this flowchart ends.

[0052] After the abnormality mode discrimination process in step S14, the discrimination control unit 21 may cause the life prediction model generation unit 215 to generate a life prediction model. The life prediction model generation unit 215 generates a life prediction model based on the abnormality mode discriminated by the abnormality mode discrimination unit 214. The life prediction model generation unit 215 also generates a life prediction model based on time-series changes in the voltage, current value, water supply amount and temperature of the water electrolysis cell, etc., as well as the abnormality mode. The life prediction model generation unit 215 then outputs the generated life prediction model to the control condition determination unit 122 of the water electrolysis system 10.

[0053] [Anomaly detection model] Next, the abnormality discrimination model used by the abnormality mode discrimination unit 214 to discriminate the abnormality mode will be described. The abnormality discrimination model is stored in advance in the storage unit 22. The abnormality discrimination model may be generated by the abnormality discrimination system 20 of the hydrogen production system 1, or may be generated by an external computing device separate from the hydrogen production system 1. Preferably, the abnormality discrimination model is generated using an external computing device such as a personal computer, a workstation, a server computer, or a general-purpose computer. In the following description, a case where an external arithmetic device is used as a method for generating an abnormality discrimination model used to discriminate an external abnormal mode will be described. In the following example, the arithmetic device includes, for example, a control unit and a storage unit, and the control unit has a learning unit that generates a learning model. The learning unit is a functional configuration of the control unit, and generates a learning model for discriminating an abnormal mode. Furthermore, the learning unit may learn and update the learning model based on the learning data stored in the storage unit. Furthermore, the learning unit may perform reinforcement learning to update the learning model.

[0054] A flowchart of the method for generating an anomaly discrimination model is shown in FIG. First, the learning unit acquires abnormal data in which an abnormal mode has been identified as a data set (step S20). The data set of the abnormal mode and the abnormal data is prepared in advance and stored in the storage unit. There is no particular limit to the number of abnormal data acquired by the learning unit, but it is preferable that the number is 50 or more. Even if the abnormal data is the same abnormal mode, there is variation in its behavior. For this reason, by acquiring 50 or more pieces of abnormal data, the learning unit can generate an abnormality discrimination model with high accuracy in abnormality discrimination. In addition, the learning unit may acquire abnormal data measured by the hydrogen production system 1 each time and re-learn. By re-learning, a discrimination model with high accuracy in anomaly detection and discrimination can be generated.

[0055] Next, the learning unit extracts features of the abnormal data from the data set of the abnormal data (step S21). For example, the learning unit extracts features such as the gradient of the regression line of the abnormal data, the residual of the regression line, the sum of squared residuals, skewness, kurtosis, the maximum value, the minimum value, and the difference between them. Note that the features are not limited to these, and it is preferable to calculate features suitable for the abnormal data each time as necessary. The learning unit may also recognize the shape of the time-series data of the abnormal data as an image, and mathematically process the features from the shape pattern. In this case, the learning unit uses, for example, a recurrence plot or the like.

[0056] Next, the learning unit performs learning using the extracted features (step S22). The learning unit selects a machine learning algorithm according to the abnormal mode. Then, the learning unit inputs data (training data) such as features into the selected machine learning algorithm and performs rule calculations. The learning unit uses, for example, a decision tree, a support vector machine, a neural network, a logistic regression, a random forest, or XGBoost as the machine learning algorithm.

[0057] Next, the learning unit saves the learning model created using the machine learning algorithm and verifies the accuracy of this learning model (step S23). The learning unit applies test data to the created learning model and evaluates the prediction accuracy of the learning model. The learning unit repeats creating a learning model using the above-mentioned machine learning algorithm and evaluating the created learning model until a desired accuracy is obtained. At this time, the learning unit selects the machine learning algorithm with the highest accuracy while verifying the accuracy of anomaly discrimination. Then, the learning unit creates an anomaly determination model by repeating the creation and evaluation of the learning model until a desired accuracy is obtained (step S24). After the anomaly determination model is created, the process according to this flowchart ends. The learning unit stores the created anomaly determination model in the storage unit 22 of the anomaly determination system 20. The created machine learning model (anomaly determination model) is applied to the abnormality mode determination unit 214 as the anomaly determination model in the flowchart shown in FIG. 4 above.

[0058] <Specific example of abnormality determination process> A specific example of the abnormality determination process according to the flowchart shown in FIG. 4 will now be described. [Anomaly detection] Anomaly detection by the anomaly detection unit 212 in step S11 of the flowchart shown in FIG. 4 will be described using a specific example. FIGS. 6 to 9 show an example of anomaly detection in a water electrolysis apparatus in which water electrolysis stacks are connected in 2 series×3 parallel. Below, an anomaly detection method will be described using an example in which water supply is stopped from one of six water electrolysis stacks. In the water electrolysis apparatus, the water electrolysis stack from which water supply has been stopped is designated Stack1, and the normal water electrolysis stacks are designated Stack2, Stack3, Stack4, Stack5, and Stack6. Stack1 and Stack2, Stack3 and Stack4, and Stack5 and Stack6 are connected in series (2 series). The two connected water electrolysis stacks in series are then connected in parallel (3 parallel).

[0059] (Detection of abnormalities based on voltage behavior) Fig. 6 shows the behavior of the voltage (VStack1+Stack2) of the two series including the abnormal stack 1 where the water supply has been stopped, and the voltage (VStack3+Stack4, VStack5+Stack6) of the two series consisting of only normal stacks. Fig. 7 also shows the difference ΔV between the voltage (VStack1+Stack2) of the two series including the abnormal stack 1 and the voltage (VStack3+Stack4, VStack5+Stack6) of the two series consisting of only normal stacks as ΔVStack12-Stack34, ΔVStack12-Stack56, and ΔVStack34-Stack56, respectively.

[0060] As shown in FIG. 6, the voltage behavior of the two series connections of only normal stacks (VStack3+Stack4, VStack5+Stack6) and the voltage behavior of the two series connections including an abnormal stack (VStack1+Stack2) each show a similar tendency to increase. 7, the voltage difference ΔV (ΔVStack34-Stack56) between the two series connected with only normal stacks remains almost constant. In contrast, the voltage difference ΔV (ΔVStack12-Stack34, ΔVStack12-Stack56) between the two series connected with an abnormal stack and the two series connected with only normal stacks increases.

[0061] In general, the voltages of the water electrolysis stacks connected in parallel are the same. However, the voltage behavior of the two series including the water electrolysis stack in which an abnormality has occurred is slightly different from that of the normal water electrolysis stack alone. Therefore, while ΔVStack34-Stack56 is constant, ΔVStack12-Stack34 and ΔVStack12-Stack56 increase. In this way, the two series stack group including the water electrolysis stack in which an abnormality has occurred has a difference in the time series change of ΔV from the other normal series stack groups. Therefore, the abnormality determination system 20 detects an abnormality in the abnormal stack line when the difference in the time series change of ΔV exceeds a certain threshold value. Specifically, although it depends on the number of water electrolysis stacks in series and the size of the stack group, in the case of a stack group configured with two water electrolysis stacks connected in series, if a voltage difference of, for example, ±0.03V or more occurs, the abnormality determination system 20 can determine an abnormality in the water electrolysis cell.

[0062] Moreover, by comparing ΔV between the parallel stack groups, it is possible to identify an abnormal series line including an abnormal stack. For this reason, the abnormality determination system 20 calculates each voltage difference ΔV between all parallel stack groups that are in operation. Then, by comparing the behavior of ΔV between each of the parallel stack groups, the abnormality determination system 20 can identify the series stack group including the water electrolysis stack in which an abnormality has occurred.

[0063] (Detection of abnormalities based on current behavior) Fig. 8 shows the behavior of the current (IStack1) of the two series including the abnormal stack 1 where the water supply has been stopped, and the current (IStack3, IStack5) of the two series consisting of only normal stacks. Fig. 9 also shows the difference ΔI between the current (IStack1) of the two series including the abnormal stack 1 and the current (IStack3, IStack5) of the two series consisting of only normal stacks as ΔIStack1-Stack3, ΔIStack1-Stack5, and ΔIStack3-Stack5, respectively.

[0064] As shown in Figure 8, the current behavior of the two series connections with only normal stacks (IStack3+Stack4, IStack5+Stack6) tends to increase, but the current behavior of the two series connections with an abnormal stack (IStack1+Stack2) tends to decrease. As shown in Fig. 9, the current difference ΔI (ΔIStack3-Stack5) between the two series with only normal stacks remains almost constant. In contrast, the current difference ΔI (ΔIStack1-Stack3, ΔIStack1-Stack5) between the two series with an abnormal stack and the two series with only normal stacks decreases.

[0065] In general, the current of the parallel water electrolysis stacks is the same. However, the current behavior of the two series including the water electrolysis stack in which an abnormality has occurred is different from that of the normal water electrolysis stack. Therefore, while ΔIStack34-Stack56 is constant, ΔIStack12-Stack34 and ΔIStack12-Stack56 tend to decrease immediately after the water supply is stopped. In this way, the two series stack group including the water electrolysis stack in which an abnormality has occurred has a difference in the time series change of ΔI from the other normal series stack groups. Therefore, the abnormality determination system 20 detects an abnormality in the abnormal stack line when the difference in the time series change of ΔI exceeds a certain threshold value. Specifically, although it depends on the number of water electrolysis stacks in series and the size of the stack group, in the case of a stack group configured with two water electrolysis stacks connected in series, if a current difference of, for example, ±0.5A or more occurs, the abnormality determination system 20 can determine an abnormality in the water electrolysis cell.

[0066] Moreover, by comparing ΔI between the parallel stack groups, it is possible to identify an abnormal series line including an abnormal stack. For this reason, the abnormality determination system 20 calculates each current difference ΔI between all parallel stack groups that are in operation. Then, by comparing the behavior of ΔI between the parallel stack groups, the abnormality determination system 20 can identify the series stack group including the water electrolysis stack in which an abnormality has occurred.

[0067] [Abnormality determination] Next, specific examples will be used to explain the acquisition of abnormality data by the feature generator 213 in step S12 and the extraction of features by the feature generator 213 in step S13 in the flowchart shown in Fig. 4. Specific examples of abnormal modes are described below, including the occurrence of bubbles on the electrode surfaces, water drying up in the water electrolysis cell, electrode peeling, water supply stoppage, and temperature control abnormality in the water electrolysis cell.

[0068] (bubbles, water drying) FIG. 10 shows an example of voltage behavior in an abnormal mode when the electrode surface is covered with bubbles. When hydrogen is produced in a water electrolysis cell, oxygen or hydrogen bubbles generated on the electrode surface may cover the electrode surface. In this case, the time-series data of the abnormal data has a characteristic voltage behavior. As shown in FIG. 10, the abnormal mode caused by the generation of bubbles on the electrode surface shows a monotonous increase in voltage. In a normal water electrolysis cell, the voltage rises gradually (less than 0.03 V) as shown in FIG. 6 above.

[0069] Therefore, the feature generator 213 extracts feature amounts from the time-series changes in voltage acquired as a data set of abnormal data. Specifically, the feature generator 213 approximates the graph of the monotonically increasing voltage shown in Fig. 10 with a regression line, and extracts the gradient of the regression line, the sum of squared residuals, and the absolute value of the difference between the maximum and minimum voltage values ​​as feature amounts. This allows the abnormality mode discrimination unit 214 to discriminate the abnormal mode of bubble generation based on the above-mentioned feature amounts extracted by the feature amount generation unit 213.

[0070] Furthermore, when water depletion occurs in the water electrolysis cell, the voltage also exhibits a monotonically increasing behavior, similar to the above-mentioned bubble generation. Therefore, the feature amount generation unit 213 can extract the gradient of the regression line of the graph of the time series change in voltage, the sum of squared residuals, and the absolute value of the difference between the maximum and minimum voltage values ​​as feature amounts. However, the feature amount values ​​extracted by the feature amount generation unit 213 differ between the above-mentioned bubble generation and water depletion. Therefore, the abnormality mode discrimination unit 214 can discriminate between the abnormality mode of bubble generation and the abnormality mode of water depletion based on the above-mentioned feature amounts extracted by the feature amount generation unit 213.

[0071] (Electrode peeling) Figure 11 shows an example of voltage behavior in an abnormal mode caused by electrode peeling and electrode cracking in a water electrolysis cell. Electrode peeling occurs when, for example, an electrode layer including a catalyst layer peels off from an electrolyte membrane (solid polymer membrane) in a water electrolysis cell. Electrode cracking occurs when damage such as cracks occurs in the electrode layer including the catalyst layer. When an abnormality such as electrode peeling or cracking occurs, the time series data of the abnormal data has a characteristic voltage behavior. As shown in Figure 11, in the abnormal mode of electrode peeling or cracking, the voltage rises suddenly when peeling or cracking occurs, and then the voltage drops gradually.

[0072] Therefore, the feature amount generating unit 213 extracts feature amounts from the time-series change in voltage acquired as a data set of abnormal data. Specifically, the feature amount generating unit 213 extracts, as feature amounts, the gradient of the regression line of the voltage rise up to the voltage peak position shown in Fig. 11, the sum of squared residuals, and the absolute value of the difference between the maximum and minimum voltage values. In addition, the feature amount generating unit 213 extracts, as feature amounts, the gradient of the regression line of the voltage behavior when the voltage decreases from the peak voltage position, the sum of squared residuals, and the absolute value of the difference between the maximum and minimum voltage values. This allows the abnormality mode discrimination unit 214 to discriminate the abnormality modes of electrode peeling and electrode cracking based on the above feature amounts extracted by the feature amount generation unit 213.

[0073] (Water supply stopped) An abnormal mode that occurs when the water supply to one water electrolysis stack is stopped in a water electrolysis apparatus in which water electrolysis stacks are connected in 2 series x 3 parallel configuration is described below. Figures 12 to 15 show a data set of abnormal data. Figure 12 shows an example of the behavior of the temperature of the water electrolysis stack 1 when the water supply is stopped. Figure 13 shows an example of the behavior of the voltage of the water electrolysis stack 1 when the water supply is stopped. Figure 14 shows an example of the behavior of the total amount of water supplied to the water electrolysis apparatus (total of stacks 1 to 6). Figure 15 shows an example of the behavior of the current value of the water electrolysis stack 1 when the water supply is stopped.

[0074] The electrode area of ​​the water electrolysis stack was 45 cm2. The measurement conditions were a stack temperature of 40°C, a total water supply rate of 1.0 to 1.3 lmin-1, and a total load current of 10 A, 20 A, or 30 A. The stack temperature was controlled by the temperature of the water supplied to the water electrolysis stack.

[0075] As shown in Fig. 14, when the water supply to stack 1 is stopped, the amount of water supplied to the entire water electrolysis apparatus decreases. Furthermore, as shown in Fig. 12, at all load currents, the stack temperature decreases when the water supply to stack 1 is stopped. Furthermore, as shown in Fig. 13, at all load currents, the stack voltage increases when the water supply to stack 1 is stopped. On the other hand, the current value of stack 1 shown in Fig. 15 does not show any change in behavior associated with the stop of the water supply to stack 1.

[0076] Therefore, the feature amount generating unit 213 extracts feature amounts from the time-series changes in voltage acquired as a data set of abnormal data. Specifically, the feature amount generating unit 213 extracts the gradient of a regression line that approximates the voltage behavior shown in Fig. 13 and the sum of squares of the residuals of the regression line as feature amounts. In addition, the feature amount generating unit 213 extracts the data on the water supply amount and the behavior of each parameter of the stack voltage, stack current, and stack temperature as feature amounts. This allows the abnormality mode discrimination unit 214 to discriminate an abnormality mode of water depletion in the water electrolysis cell based on the above feature values ​​extracted by the feature value generation unit 213.

[0077] Meanwhile, a shortage of water supply to the water electrolysis cells may be due to an abnormality in the supply pump or a water leak in the piping, cells, stack, etc. Abnormalities caused by these factors can also be confirmed by sensors installed in each device. Therefore, by quickly detecting an abnormality in the water supply shortage, it is possible to detect abnormalities in peripheral devices and appropriately control the water electrolysis device.

[0078] (Temperature control abnormality) An abnormal mode will be described below in a water electrolysis apparatus in which water electrolysis stacks are connected in 2 series x 3 parallel configuration, in which the temperature of one water electrolysis stack rises from 40°C to approximately 50°C. Figures 16 to 19 show data sets of abnormal data. Figure 16 shows an example of the behavior of the temperature of the water electrolysis stack 1 when the temperature rises. Figure 17 shows an example of the behavior of the voltage of the water electrolysis stack 1 when the temperature rises. Figure 18 shows an example of the behavior of the total amount of water supplied to the water electrolysis apparatus (total of stacks 1 to 6). Figure 19 shows an example of the behavior of the current value of the water electrolysis stack 1 when the temperature rises.

[0079] As shown in Fig. 16, the stack temperature rises at all load currents. Also, as shown in Fig. 17, the stack voltage drops as the temperature of stack 1 rises at all load currents. Furthermore, as shown in Fig. 19, the stack current rises as the temperature of stack 1 rises. On the other hand, the behavior of the total water supply amount shown in Fig. 18 does not change in accordance with the rise in temperature of stack 1.

[0080] Therefore, the feature generator 213 extracts feature amounts from the time series changes in voltage, current, and stack temperature acquired as the data set of anomaly data. Specifically, the feature generator 213 extracts the gradient of a regression line approximating the voltage behavior shown in Fig. 17, the current behavior shown in Fig. 16, and the temperature behavior shown in Fig. 19, and the sum of squares of the residuals of the regression line as feature amounts. This allows the abnormality mode determination unit 214 to determine the abnormality mode of the temperature abnormality of the water electrolysis cell based on the feature amounts extracted by the feature amount generation unit 213.

[0081] (Short circuit) In the case of an abnormality mode caused by a short circuit of the electrodes of a water electrolysis cell, it is not possible to obtain abnormal data such as the voltage and current value of the short-circuited water electrolysis cell. However, the abnormality can be detected by observing the fluctuation in the voltage of the series stack group including the short-circuited water electrolysis cell. Specifically, an abnormality caused by a short circuit appears as a sudden drop in the stack voltage. For this reason, the feature generator 213 extracts feature values ​​from the time-series change in the voltage of the series stack group acquired as a data set of abnormal data. In addition, an abnormality caused by a short circuit also appears as a sudden increase in the stack current, so that it can be detected and feature values ​​can be extracted in the same manner as for the stack voltage. This allows the abnormality mode discrimination section 214 to discriminate an abnormality mode of an electrode short circuit based on the above-mentioned feature amount extracted by the feature amount generation section 213.

[0082] [Lifespan prediction model] Fig. 20 is an example of a lifetime prediction model generated by the lifetime prediction model generating unit 215. The lifetime prediction model shown in Fig. 20 is a result of predicting the voltage behavior by fitting the behavior of the stack voltage that decreases with the increase in the water electrolysis stack temperature shown in Fig. 16 to Fig. 19 with a prediction formula for each abnormality mode. As shown in Figure 20, the voltage of the water electrolysis stack decreases over time. Therefore, the timing for replacing the water electrolysis stack can be estimated from a voltage drop prediction model. As a result, the running costs and maintenance costs of the hydrogen production system can be reduced. The life prediction model generating unit 215 obtains a prediction formula corresponding to the abnormality mode from the storage unit 22, and generates the above-mentioned life prediction model by fitting various data (voltage, current, etc.) to the prediction formula. The prediction formula may be, for example, a simple polynomial approximation using Taylor expansion or Maclaurin expansion, or an individual deterioration formula corresponding to the type of abnormality may be used.

[0083] [How to install a voltmeter and ammeter] Next, a method for installing a voltmeter and an ammeter in the water electrolysis apparatus will be described. Figures 21 and 22 show examples of water electrolysis equipment in which water electrolysis stacks are connected in 2 series x 3 parallel. Figure 21 shows an example of a method for installing a voltmeter, and Figure 22 shows an example of a method for installing an ammeter. 21 and 22, the water electrolysis apparatus has six water electrolysis stacks, of which Stack1 and Stack2, Stack3 and Stack4, and Stack5 and Stack6 are connected in series (two in series), and each of the two connected water electrolysis stacks in series is connected in parallel (three in parallel).

[0084] In conventional water electrolysis systems, a voltmeter is installed in each water electrolysis stack. The individually installed voltmeter constantly measures the voltage of the water electrolysis stack to detect any abnormalities. However, installing a voltmeter in each water electrolysis stack results in huge installation costs. In addition, when the water electrolysis system is in operation, the maintenance costs for each voltmeter are huge.

[0085] In response to this, in a multi-series, multi-parallel water electrolysis stack, a voltmeter V is installed for each series stack line as shown in FIG. 21. Furthermore, an ammeter I is installed for each series stack line as shown in FIG. 22. By installing the voltmeter V or ammeter I in this way, it is possible to measure the voltage or current of the entire series stack line without installing a voltmeter for each water electrolysis stack. Furthermore, the series stack line including the water electrolysis stack in which an abnormality has occurred can be identified from the behavior of the difference in the measured voltage or current between the parallel stack groups.

[0086] 21 and 22, not only can the installation cost of the measuring instruments be reduced, but also the maintenance cost of the measuring instruments themselves can be reduced by reducing the number of measuring instruments. Furthermore, the stack maintenance cost can be reduced by detecting anomalies in the water electrolysis stack using the voltmeter V and the ammeter I. However, the more water electrolysis stacks are connected in series, the more difficult it becomes to detect the behavior of a water electrolysis stack in which an abnormality has occurred. Therefore, when there are many water electrolysis stacks, the number of installed measuring devices including a voltmeter V and an ammeter I may be increased appropriately according to the number of water electrolysis stacks connected in series.

[0087] [Water electrolysis stack replacement procedure] Next, an example of a procedure for replacing a water electrolysis stack in which an abnormality has been detected in a water electrolysis apparatus will be described. FIG. 23 is a flowchart showing the procedure for detecting an abnormality in a water electrolysis stack in a multi-series and multi-parallel stack group, determining the abnormality mode, and replacing the water electrolysis stack.

[0088] First, the voltage or current of each series stack line is measured, and the difference between the measured values ​​is calculated individually (step S30). Moreover, the voltage or current of each parallel stack group is measured, the difference between the measured values ​​is calculated individually, and the behavior is acquired (step S31). When the difference between the acquired voltage or current values ​​exceeds a certain threshold, an abnormality in the abnormal stack line is detected (step S32).

[0089] Next, after detecting an abnormality, the abnormal mode is identified from the acquired behavior shape and fluctuations in other parameters, and the lifespan is estimated using a lifespan prediction model to determine whether or not the abnormal stack line needs to be shut off (step S33). Then, each of the water electrolysis stacks constituting the blocked abnormal stack line is inspected, and the water electrolysis stack in which the abnormality has occurred is identified and removed from the water electrolysis apparatus (step S34). Furthermore, only the water electrolysis stack in which the abnormality has occurred is replaced with a normal water electrolysis stack, and the cut-off stack line is reconnected to the system (step S35).

[0090] As described above, the hydrogen production system of the present embodiment makes it possible to easily detect an abnormality in a water electrolysis cell and to distinguish an abnormal mode. Furthermore, in a multi-series / multi-parallel water electrolysis system, a series line including an abnormal stack can be identified by comparing the time changes in the total voltage or current of each series stack group connected in parallel. This makes it possible to reduce the installation and maintenance costs of measuring instruments without having to install measuring instruments in each stack.

[0091] The anomaly discrimination system may detect not only anomalies in the water electrolysis device and the water electrolysis cell, but also anomalies in auxiliary devices related to the water electrolysis device and the water electrolysis cell, such as a temperature control device and a water supply device. The anomaly detection of the configuration of auxiliary devices other than the water electrolysis cell related to the water electrolysis device, the determination of the abnormal mode, and the generation of the life prediction model are similar to the processing of the water electrolysis cell described above. For example, the anomaly discrimination system collects various data by monitoring the temperature with a temperature sensor and the flow rate of the water supply mechanism. Then, the anomaly discrimination system uses the collected various data as one of parameters for machine learning or the like to generate a normal detection model in advance. Then, the anomaly discrimination system can discriminate the abnormal mode of the auxiliary devices and the like using the anomaly detection model. Furthermore, the anomaly discrimination system can generate a life prediction model using the above various data.

[0092] The present invention is not limited to the above-described embodiments, and various modifications are possible. For example, the above-described embodiments have been described in detail to clearly explain the present invention, and the present invention is not necessarily limited to an embodiment having all of the configurations described. It is also possible to replace a part of the configuration of one embodiment with the configuration of another embodiment. It is also possible to add the configuration of another embodiment to the configuration of one embodiment. It is also possible to delete a part of the configuration of each embodiment, or to add or replace another configuration. [Explanation of symbols]

[0093] 1,2 Hydrogen production system, 10 Water electrolysis system, 11 Water electrolysis device, 12 Electrolysis control unit, 13,22 Memory unit, 14 Display unit, 20 Anomaly discrimination system, 21 Discrimination control unit, 110 Water electrolysis cell, 111 Cathode catalyst layer, 112 Anode catalyst layer, 113 Solid polymer membrane, 114 Cathode current collector, 115 Anode current collector, 116 Power source, 117,118 Separator, 121 Device control unit, 122 Control condition judgment unit, 211 Data acquisition unit, 212 Anomaly detection unit, 213 Feature generation unit, 214 Anomaly mode discrimination unit, 215 Life prediction model generation unit

Claims

1. 1. A hydrogen production system comprising: a water electrolysis system including a water electrolysis cell; and an abnormality determination system that determines an abnormality mode of the water electrolysis system, The abnormality determination system includes: a data acquisition unit that acquires data on time changes of at least one of a temperature, a water supply amount, a current, and a voltage of the water electrolysis cell; an anomaly detection unit that detects an anomaly in the water electrolysis cell based on the data acquired by the data acquisition unit; an abnormality mode discrimination unit that discriminates the abnormality mode of the water electrolysis cell in which an abnormality has occurred, The abnormality mode determination unit determines the abnormality mode based on a machine learning model obtained by learning the behavior of current or voltage in a plurality of the abnormality modes and the data acquired by the data acquisition unit. Hydrogen production system.

2. The abnormality mode determination unit determines at least one abnormality mode selected from the group consisting of foam generation, water depletion, electrode peeling, water supply stop, and temperature control abnormality. The hydrogen production system according to claim 1 .

3. the abnormality determination system includes a feature amount generation unit that extracts a feature amount of the data acquired by the data acquisition unit, The feature generating unit generates the feature using at least one of a graph shape of the data when it changes over time, a gradient of a regression line in an arbitrary range, a residual sum of squares indicating a deviation from the regression line, and a difference between a maximum value and a minimum value. The hydrogen production system according to claim 1 .

4. The hydrogen production system includes a multi-series / multi-parallel stack group in which a plurality of the water electrolysis cells connected in series are arranged in parallel, The detection unit identifies a series line including the water electrolysis cell in which an abnormality has occurred from the stack group. The hydrogen production system according to claim 1 .

5. The detection unit identifies a series line including the water electrolysis cell in which an abnormality has occurred based on a difference in voltage behavior between parallel stack groups. The hydrogen production system according to claim 4 .

6. The detection unit identifies a series line including the water electrolysis cell in which an abnormality has occurred based on a difference in current behavior between parallel stack groups. The hydrogen production system according to claim 4 .

7. The hydrogen production system includes a multi-series / multi-parallel stack group in which a plurality of the water electrolysis cells connected in series are arranged in parallel, A measuring device is provided for measuring the voltage or current of a stack group in which at least one water electrolysis cell is connected in series. The hydrogen production system according to claim 1 .

8. The anomaly determination system includes a life prediction model generation unit that generates a life prediction model for the water electrolysis cell in which an anomaly has occurred. The hydrogen production system according to claim 1 .

9. A control method for a hydrogen production system including a water electrolysis system including a water electrolysis cell and an abnormality determination system that determines an abnormality mode of the water electrolysis system, comprising: acquiring data on time changes of at least one of a temperature, a water supply amount, a current, and a voltage of the water electrolysis cell; Detecting an abnormality in the water electrolysis cell based on the acquired data; The abnormality mode of the water electrolysis cell in which an abnormality has occurred is determined based on the acquired data and a machine learning model obtained by learning the current or voltage behavior in the plurality of abnormality modes. A method for controlling a hydrogen production system.

Images