JP7690926B2 - System and method for estimating durability for fuel cell system - Google Patents

Description

The technology disclosed in this specification relates to a system and method for estimating the durability of a fuel cell system.

Patent Document 1 describes a technology for reusing batteries installed in vehicles in other products. In this technology, operation data of the battery both before and after reuse is collected, and a machine learning model is created that estimates the durability (lifespan) of the battery by machine learning using the collected operation data as learning data.

JP 2020-162309 A

As with the above-mentioned technology, the durability of industrial products can be estimated by using machine learning techniques. In this regard, since fuel cell systems are installed in vehicles and then reused in other products, it is desirable to accurately estimate their durability. When using machine learning techniques, a large amount of learning data is required. For example, for existing fuel cell systems that have already been provided to the public, actual operation data from a large number of fuel cell systems used by the public can be obtained as learning data. In contrast, for new fuel cell systems that are improvements on the existing fuel cell systems, such learning data cannot be obtained sufficiently, and it is difficult to estimate durability using machine learning techniques based only on the results of limited durability tests performed before being provided to the public. This specification provides a technology that can estimate the durability of fuel cell systems for which sufficient learning data does not exist using machine learning techniques.

The technology disclosed in this specification is embodied in a system for estimating durability of a fuel cell system. In a first aspect, the system may include a first storage device that stores, as learning data, operation data when the first type of fuel cell system actually operates for one or more first type fuel cell systems; a first arithmetic device having a machine learning model that performs machine learning using the learning data stored in the first storage device and estimates the durability of the first type of fuel cell system; a second storage device that stores the results of a durability test of the first type of fuel cell system and the results of a durability test of a second type of fuel cell system different from the first type of fuel cell system for the same usage conditions; and a second arithmetic device that determines a conversion formula or conversion model that derives the results of a durability test of the second type of fuel cell system from the results of the durability test of the first type of fuel cell system using the results of the durability test stored in the second storage device. The second arithmetic device may estimate the durability of the second type of fuel cell system by applying the conversion formula or conversion model determined by the second arithmetic device to the durability of the first type of fuel cell system estimated by the machine learning model of the first arithmetic device.

In the above-mentioned system, the operation data from one or more first type fuel cell systems is stored as learning data. Then, a machine learning model that estimates the durability of the first type fuel cell system is created by performing machine learning using the learning data. Furthermore, a conversion formula or conversion model that derives the results of the durability test of the second type fuel cell system from the results of the durability test of the first type fuel cell system is created using the results of the durability test of the first type and the second type fuel cell systems. Then, the conversion formula or conversion model is applied to the results of the estimated durability of the first type fuel cell system. With this configuration, the durability of the second type fuel cell system can be derived by applying the conversion formula or conversion model based on the results of the durability test of the two types of fuel cell systems to the results of the durability estimated by the machine learning model of the first type fuel cell system. This makes it possible to estimate the durability of the second type fuel cell system based not only on the results of the durability test of the second type fuel cell system but also on the known operation data of the first type fuel cell system.

In a second aspect, the system according to the first aspect described above may further include a receiving device that receives operation data from the first type of fuel cell system provided to the public or an industrial product equipped with the same. The first storage device may be connected to the receiving device and store the operation data received by the receiving device. With this configuration, a large amount of operation data can be collected from a large number of first type of fuel cell systems provided to the public or industrial products equipped with the same.

In a third aspect, in the first or second aspect described above, the second computing device may determine the conversion formula or conversion model by machine learning using the results of the durability test stored in the second storage device. With this configuration, the conversion formula or conversion model can be determined with high accuracy. However, in another embodiment, the conversion formula or conversion model may have a simple structure, for example, expressed by a linear function.

The technology disclosed in this specification is embodied in a method for estimating the results of durability tests on a fuel cell system. The method may include the steps of: storing operating data for one or more first type fuel cell systems when the first type fuel cell systems are actually operated in a first storage device as learning data; causing a machine learning model, which is a machine learning model that performs machine learning using the learning data stored in the first storage device and is possessed by a first arithmetic device, to estimate the durability of the first type fuel cell system; storing in a second storage device results of a durability test of the first type fuel cell system and results of a durability test of a second type fuel cell system different from the first type fuel cell system, for the same usage conditions; causing the second arithmetic device to determine a conversion equation or conversion model that derives results of a durability test of the second type fuel cell system from the results of the durability test of the first type fuel cell system, using the results of the durability test stored in the second storage device; and causing the second arithmetic device to estimate the durability of the second type fuel cell system by applying the conversion equation or conversion model determined by the second arithmetic device to the durability of the first type fuel cell system estimated by the machine learning model of the first arithmetic device. According to this method, the durability of the second type of fuel cell system can be derived by applying a conversion formula or conversion model based on the results of the durability tests of the two types of fuel cell systems to the results of the durability estimated by the machine learning model of the first type of fuel cell system. This makes it possible to estimate the durability of the second type of fuel cell system based not only on the results of the durability tests of the second type of fuel cell system, but also on the known operating data of the first type of fuel cell system.

FIG. 1 is a block diagram showing a configuration of an estimation system. FIG. 4 is a diagram for explaining a learning model of the first arithmetic device. 11 is a flowchart showing the procedure of an example of an estimation method.

(Example) The estimation system 10 of this example will be described with reference to Figures 1 to 3. This estimation system 10 estimates the durability of a fuel cell system. As an example, the fuel cell system is mounted on a vehicle. However, it is not limited to vehicles, and the fuel cell system may be mounted on other industrial products other than vehicles.

As shown in FIG. 1, the estimation system 10 includes a receiving device 12, a first storage device 14, a second storage device 16, a first calculation device 18, a second calculation device 20, and a communication device 22. The receiving device 12 is a device that receives operation data of a fuel cell system (hereinafter referred to as an existing FC system 2a) installed in each of a plurality of vehicles 2 (hereinafter simply referred to as vehicles 2) provided to the public. As an example, a part or all of the receiving device 12 can be configured with a mobile line network provided by a mobile phone operator. The receiving device 12 wirelessly receives operation data of the existing FC system 2a transmitted from the plurality of vehicles 2 using any communication protocol. The receiving device 12 is connected to the first storage device 14 so as to be able to communicate with it using a wired and/or wireless line. The receiving device 12 transmits the stored operation data to the first storage device 14. The operation data that the receiving device 12 transmits to the first storage device 14 is not limited to data of the existing FC system 2a of multiple vehicles 2, but may be data of the existing FC system 2a of a single vehicle 2. However, the more vehicles 2 that provide operation data, the more preferable it is.

In addition to the existing FC system 2a described above, the vehicle 2 is equipped with multiple sensors, a memory, and a communication device. Each sensor measures various parameters (voltage, current, voltage speed, temperature (temperature inside the tank, outside air temperature, etc.), gas flow rate, gas pressure, etc.) related to the existing FC system 2a of the vehicle 2 at predetermined time intervals. The operation data is time-series data in which the measured parameter values are associated with time information together with control signals for controlling the fuel cell system (e.g., target current, etc.). The memory stores this operation data. The communication device reads this operation data from the memory and periodically transmits the operation data to the receiving device 12 together with vehicle identification information unique to each vehicle 2. Although not limited to this, the communication device may be, for example, a mobile device capable of wireless communication, such as a mobile phone owned by the user of the vehicle 2.

The first storage device 14 is communicatively connected to the first arithmetic device 18. The first storage device 14 has a memory. The first storage device 14 receives the operation data transmitted from the receiving device 12 and stores the received operation data as learning data. The first storage device 14 transmits the stored learning data to the first arithmetic device 18. The first arithmetic device 18 executes an estimation process. The first arithmetic device 18 has a memory and a CPU. The first arithmetic device 18 has a learning model 19 stored in the memory. The first arithmetic device 18 causes the CPU to execute various processes based on the learning model 19 and/or other programs stored in the memory. The first arithmetic device 18 causes the learning model 19 to machine-learn the learning data received from the first storage device 14. The first arithmetic device 18 estimates the durability results of the existing FC system 2a using the machine-learned learning model 19. The communication device 22 is a communication network that connects the first arithmetic device 18 and the second arithmetic device 20 to each other. The first calculation device 18 is communicatively connected to the second calculation device 20 via the communication device 22. The first calculation device 18 transmits the durability results estimated by the learning model 19 to the second calculation device 20. The communication device 22 may be configured, for example, by a communication instruction program incorporated in each of the first calculation device 18 and the second calculation device 20, or a primary storage device provided in one or both of the first calculation device 18 and the second calculation device 20.

The second storage device 16 is communicatively connected to the durability test device 4 and the second arithmetic device 20. The second storage device 16 has a memory. The second storage device 16 stores the results of durability tests of the various types of fuel cell systems received from the durability test device 4. The results of durability tests of the various types of fuel cell systems include the results of durability tests of the existing FC system 2a and the results of durability tests of the new FC system for the same usage conditions. The new FC system is different from the existing FC system 2a and is a fuel cell system that is an improvement of the existing FC system 2a. As an example, the output (kW) of the new FC system is about ±20% of the output (kW) of the existing FC system 2a. The second storage device 16 transmits the stored results of durability tests of the various types of fuel cell systems to the second arithmetic device 20. Here, in a fuel cell system, if the usage conditions experienced by the fuel cell system are different, the possible values of operating parameters that affect the deterioration of the fuel cell system (e.g., current, rate of change of voltage, outside air temperature, etc.) may be affected. Therefore, the results of the durability test of the existing FC system 2a and the endurance test of the new FC system, both of which were conducted under the same operating conditions, are transmitted to the second calculation device 20. In this specification, the same operating conditions refer to conditions in which the range of values that each operating parameter can take and the frequency of occurrence are approximately the same. Here, the existing FC system 2a is an example of a "first type of fuel cell system" in the technology disclosed in this specification, and the new FC system is an example of a "second type of fuel cell system" in the technology disclosed in this specification.

The durability test device 4 is a device that performs durability tests on the fuel cell system. The durability test device 4 is equipped with, for example, a chassis dynamometer, and can perform durability tests using a test vehicle equipped with a fuel cell system. The test vehicle, like the vehicle 2 described above, measures various parameters related to the fuel cell system (voltage, current, voltage speed, temperature (temperature inside the tank, outside air temperature, etc.), gas flow rate, gas pressure, etc.) at predetermined time intervals using sensors installed in the actual vehicle. The durability test device 4 records time-series data in which the measurement data is associated with time information, together with control signals (e.g., target current, etc.) for controlling the fuel cell system, as the results of the durability test. The durability test device 4 transmits the results of the durability test to the second storage device 16. Note that the specific configuration of the durability test device 4 is not particularly limited. Instead of using a test vehicle, the durability test device 4 may use a simulator device that can reproduce the operation of a fuel cell system installed in an actual vehicle.

The second calculation device 20 has a memory and a CPU. The CPU of the second calculation device 20 uses the results of the durability tests of the existing and new FC systems received from the second storage device 16 to determine a conversion model that derives the results of the durability test of the new FC system from the results of the durability test of the existing FC system 2a. The second calculation device 20 determines the conversion model by machine learning using the machine learning model stored in the memory. This makes it possible to accurately determine the conversion model. However, the conversion model does not need to be determined by machine learning, and may have a simple structure such as that expressed by a linear function. For example, it is assumed that the conversion formula is y = Ax. Here, x is an index value of the existing FC system 2a and is the difference between the value after the durability test and the initial value, y is an index value of the new FC system and is the difference between the value after the durability test and the initial value, and A is a constant and is determined from the ratio of x to y. If the durability of the new FC system is improved compared to the existing FC system 2a, A is a value smaller than 1.

The CPU of the second calculation device 20 applies the above-mentioned conversion formula or conversion model to the durability of the existing FC system 2a estimated by the learning model 19 of the first calculation device 18. This allows the second calculation device 20 to estimate the durability of the new FC system.

With reference to FIG. 2, the estimation by the learning model 19 and the application of the conversion model to the estimation will be described. As shown in FIG. 2, when the usage conditions of the existing FC system 2a to be estimated are input, the learning model 19 of the first calculation device 18 outputs an estimation result of the durability of the existing FC system 2a. The input parameters (usage conditions) include a current value, a temperature, a voltage change rate, and an operating time. Here, the output parameter (estimated durability result) is, for example, a voltage value of the fuel cell system when the current in the fuel cell system is a predetermined value. The higher the voltage value, the higher the durability of the fuel cell system is determined to be, and the lower the voltage value, the lower the durability of the fuel cell system is determined to be. The learning model 19 may be optimized using a part of the learning data read from the first storage device 14 so that when the input of the usage conditions is given, the error between the estimated output result and the operating data result is minimized. In addition, in order to ensure the generalization performance of the learning model 19, a part of the learning data may be reserved for verification. In this case, the reserved data may be used to avoid overlearning of the learning model 19.

When the second calculation device 20 receives the estimation result of the existing FC system 2a from the first calculation device 18, it applies a conversion model or conversion formula to the input estimation result and outputs the durability result of the new FC system. The output parameter here is, for example, the voltage value of the fuel cell when the current in the fuel cell system is a predetermined value, as with the first calculation device 18 (i.e., the learning model 19).

The output parameter of the learning model 19 is not limited to the voltage value at a specific current value, but may be a voltage value when specific conditions are met, such as a voltage value at a specific temperature/gas supply amount. The output parameter may be an index value that represents the degree of deterioration of the fuel cell system. In a modified example, the output parameter may be, for example, the amount of gas leaking from a gas tank when hydrogen is pressurized under specific conditions.

In the estimation system 10 of this embodiment, ElasticNet is used as the learning model 19. However, the learning model 19 is not particularly limited, and other machine learning models such as regression other than ElasticNet (e.g., Ridge regression, Lasso regression, etc.) and neural networks may be used.

Conventionally, the durability of industrial products can be estimated by using machine learning techniques. In this regard, since fuel cell systems are installed in vehicles and then reused in other products, it is desirable to accurately estimate their durability. When using machine learning techniques, a large amount of learning data is required. For example, in the case of existing fuel cell systems that have already been provided to the public, actual operation data from a large number of fuel cell systems in use by the public can be obtained as learning data. In contrast, in the case of new fuel cell systems that are improvements on existing fuel cell systems, such learning data cannot be obtained in sufficient amounts, and it is difficult to estimate durability using machine learning techniques based only on the results of limited durability tests conducted before the systems are provided to the public.

In the estimation system 10 described above, operation data from one or more existing FC systems 2a is stored as learning data. Then, a learning model 19 is created to estimate the durability of the existing FC system 2a by performing machine learning using the learning data. Furthermore, a conversion formula or conversion model is created that derives the results of the endurance test of the new FC system from the results of the endurance test of the existing FC system 2a using the results of the endurance test of the existing FC system 2a using the results of the endurance test of the new FC system. Then, the conversion formula or conversion model is applied to the results of the estimated durability of the existing FC system 2a. With this configuration, the durability of the new FC system can be derived by applying the conversion formula or conversion model based on the results of the endurance test of two types of fuel cell systems (the existing and new FC systems) to the results of the durability estimated by the learning model 19 of the existing FC system 2a. This allows the durability of the new FC system to be estimated based not only on the results of the endurance test of the new FC system but also on the known operation data of the existing FC system 2a.

Next, the procedure of the durability estimation process (estimation method) executed by the estimation system 10 will be described with reference to FIG. 3. As shown in FIG. 3, first, in step S12, the estimation system 10 executes a first storage step of storing learning data. Specifically, in the first storage step, the estimation system 10 stores the operation data of the multiple existing FC systems 2a transmitted from the receiving device 12 in the first storage device 14 as learning data. Next, in step S14, the estimation system 10 executes a machine learning step of machine learning the learning data. Specifically, in the machine learning step, the estimation system 10 uses the learning data stored in the first storage device 14 in S12 to machine learn the learning model 19 of the first calculation device 18.

Next, in step S16, the estimation system 10 executes a first estimation process using a machine learning model. Specifically, in the first estimation process, the estimation system 10 causes the first calculation device 18 to estimate the durability results of the existing FC system 2a using a learning model 19 that has been machine-learned.

Next, in step S18, the estimation system 10 executes a second storage step of storing the results of the durability test. In the second storage step, specifically, the estimation system 10 stores the results of the durability test of the existing FC system 2a and the new FC system in the second storage device 16. Next, in step S20, the estimation system 10 executes a determination step of determining a conversion model using the results of the durability test. In the determination step, specifically, the estimation system 10 causes the second calculation device 20 to determine a conversion formula or conversion model that derives the results of the durability test of the new FC system from the results of the durability test of the existing FC system 2a, using the results of the durability test of the existing FC system 2a and the new FC system stored in the first storage device 14 in S18.

Next, in step S22, the estimation system 10 executes an application step of applying a conversion model to the estimation by the learning model 19. Specifically, in the application step, the estimation system 10 applies the conversion formula or conversion model determined in S20 to the durability of the existing FC system 2a estimated by the learning model 19 of the first calculation device 18. Through the above series of steps S12 to S22, the durability of the new FC system can be estimated based not only on the results of the durability test of the new FC system, but also on the known operating data of the existing FC system 2a.

The estimation system 10 in this embodiment includes a receiving device 12 that receives operation data from a vehicle 2 equipped with an existing FC system 2a provided to the public. The first storage device 14 is also connected to the receiving device 12 and stores the operation data received by the receiving device 12. With this configuration, a large amount of operation data can be collected from a large number of vehicles 2 equipped with existing FC systems 2a provided to the public.

In the estimation system 10 of this embodiment, the operating data from the receiving device 12 is transferred to the first calculation device 18 via the first storage device 14. However, it does not necessarily have to go via the first storage device 14, and it may be sent directly from the receiving device 12 to the first calculation device 18. In this case, the memory of the first calculation device 18 may function as the first storage device 14. Similarly, the results of the durability test from the durability test device 4 do not necessarily have to go via the second storage device 16, and it may be sent directly from the durability test device 4 to the second calculation device 20.

Specific examples of the technology disclosed in this specification have been described in detail above, but these are merely examples and do not limit the scope of the claims. The technology described in the claims includes various modifications and variations of the specific examples exemplified above. The technical elements described in this specification or drawings exhibit technical utility alone or in various combinations, and are not limited to the combinations described in the claims at the time of filing. The technology exemplified in this specification or drawings can achieve multiple objectives simultaneously, and achieving one of those objectives is itself technically useful.

2a: Existing FC system 10: Estimation system 12: Receiving device 14, 16: Storage device 18, 20: Calculation device 19: Learning model

Claims (4)

A system for estimating durability of a fuel cell system, comprising:
a first storage device that stores, as learning data, operation data obtained when one or more first type fuel cell systems are actually operated, the first type fuel cell systems;
a first arithmetic device having a machine learning model that performs machine learning using the learning data stored in the first storage device, the machine learning model estimating durability of the first type of fuel cell system;
a second storage device that stores the results of a durability test of the first type of fuel cell system and the results of a durability test of a second type of fuel cell system different from the first type of fuel cell system, under the same operating conditions;
a second arithmetic unit that uses the results of the durability test stored in the second storage device to determine a conversion formula or a conversion model for deriving results of the durability test of the second type of fuel cell system from the results of the durability test of the first type of fuel cell system;
Equipped with
the second arithmetic device applies the conversion formula or the conversion model determined by the second arithmetic device to the durability of the first type of fuel cell system estimated by the machine learning model of the first arithmetic device, thereby estimating the durability of the second type of fuel cell system;
system. a receiving device for receiving the operation data from the first type of fuel cell system provided to the public or an industrial product equipped with the first type of fuel cell system;
The system of claim 1 , wherein the first storage device is coupled to the receiving device and stores the operational data received by the receiving device. The system according to claim 1 or 2, wherein the second computing device determines the conversion formula or the conversion model by machine learning using the results of the durability test stored in the second storage device. A method for estimating durability of a fuel cell system, comprising:
storing, in a first storage device, operation data obtained when one or more first type fuel cell systems are actually operated, as learning data;
A step of estimating the durability of the first type of fuel cell system using a machine learning model that performs machine learning using the learning data stored in the first storage device and that is included in a first computing device;
storing in a second storage device the results of a durability test of the first type of fuel cell system and the results of a durability test of a second type of fuel cell system different from the first type of fuel cell system, under the same operating conditions;
causing a second arithmetic device to determine a conversion formula or a conversion model for deriving a result of the durability test of the second type of fuel cell system from the result of the durability test of the first type of fuel cell system, using the result of the durability test stored in the second storage device;
A step of causing the second arithmetic device to estimate durability of the second type of fuel cell system by applying the conversion formula or the conversion model determined by the second arithmetic device to the durability of the first type of fuel cell system estimated by the machine learning model of the first arithmetic device;
A method for providing the above.

Images