Hydrogen production system fault diagnosis method based on kan network of kolmogorov-arnold time attention

Abstract

The application discloses a hydrogen production system fault diagnosis method based on a KAN network of Kolmogorov-Arnold time attention, and relates to the technical field of fault diagnosis.The application is aimed at solving the problems of low fault diagnosis accuracy and poor real-time performance of the existing hydrogen production system fault detection method.The application comprises the following steps: obtaining actual current, actual voltage and actual temperature of each working time point in the working process of a to-be-detected electrolytic cell, and performing pretreatment; inputting the actual current and actual temperature of the working time of the to-be-detected electrolytic cell after the pretreatment into a hydrogen production system fault detection model to obtain predicted voltage; obtaining the residual error of the actual working voltage of the to-be-detected electrolytic cell by using the predicted voltage and the pretreated actual voltage; and judging whether the current hydrogen production system has a fault or not by using the residual error of the actual working voltage of the to-be-detected electrolytic cell.The application is used for hydrogen production system fault detection.

Description

Technical Field

[0001] This invention relates to the field of fault diagnosis technology, and in particular to a fault diagnosis method for hydrogen production systems based on Kolmogorov-Arnold time attention KAN networks. Background Technology

[0002] With the continuous growth of global energy demand and the increasing severity of environmental pollution, hydrogen energy, as a clean and sustainable energy source, is gradually becoming an important part of the energy transition. Hydrogen energy, due to its high energy density, zero emissions, and broad application prospects, shows enormous potential, especially in transportation, industry, and energy storage. The production, storage, and transportation of hydrogen are key links in hydrogen energy applications, with hydrogen production technology, as the core of the hydrogen energy industry chain, directly determining the hydrogen supply capacity. Currently, there are several main hydrogen production technologies, including water electrolysis, natural gas reforming, and coal gasification. Water electrolysis, in particular, has received widespread attention in recent years because it can use electricity to decompose water into hydrogen and oxygen. With the rapid development of renewable energy sources (such as wind and solar power), water electrolysis is considered a green, environmentally friendly, and efficient method of hydrogen production. However, with the large-scale application of hydrogen electrolysis units, equipment malfunctions are becoming increasingly prominent, affecting the stability, reliability, and efficiency of hydrogen production. Therefore, improving the fault diagnosis capabilities of hydrogen production systems is particularly important for ensuring system safety, reducing maintenance costs, and improving production efficiency.

[0003] Fault diagnosis technology plays a crucial role in hydrogen electrolysis units. By monitoring the equipment's operating status in real time, fault diagnosis technology can promptly detect and identify anomalies in the system, thereby preventing the escalation of faults and enabling timely corrective measures. This not only helps improve the operating efficiency of the hydrogen electrolysis unit and extend the equipment's lifespan but also ensures system safety, reduces downtime, and enhances production stability.

[0004] Currently, in fault diagnosis of hydrogen electrolysis units, physical model-based fault diagnosis and signal processing-based fault diagnosis methods are commonly used. However, physical model-based fault diagnosis methods rely on accurate mathematical models and system laws, which are computationally intensive and require a high level of prior knowledge of the system. Signal processing-based fault diagnosis methods solve this problem; however, they mainly infer fault conditions by analyzing signal changes in sensor data, and their accuracy and real-time performance are often limited. Therefore, current fault diagnosis methods for hydrogen electrolysis units still suffer from low accuracy and poor real-time performance. Summary of the Invention

[0005] The purpose of this invention is to address the problems of low accuracy and poor real-time performance in existing hydrogen production system fault detection methods, and to propose a fault diagnosis method for hydrogen production systems based on KAN networks with Kolmogorov-Arnold temporal attention.

[0006] A fault diagnosis method for hydrogen production systems based on Kolmogorov-Arnold time attention KAN networks is as follows:

[0007] S1. Obtain the actual current, actual voltage, and actual temperature at each working time point during the operation of the electrolytic cell under test, and preprocess the actual current, actual voltage, and actual temperature of the electrolytic cell under test during its working time to obtain the preprocessed actual current, actual voltage, and actual temperature of the electrolytic cell under test during its working time.

[0008] S2. Input the actual current and actual temperature of the pre-treated electrolyzer during its operating time into the hydrogen production system fault detection model to obtain the predicted voltage;

[0009] S3. Obtain the residual r' of the actual operating voltage of the electrolytic cell under test using the predicted voltage and the pre-processed actual voltage. i ;

[0010] S4. Utilize the residual r' of the actual operating voltage of the electrolytic cell under test. i Determine if the current hydrogen production system is malfunctioning.

[0011] Furthermore, in S1, the actual current, actual voltage, and actual temperature at each working time point during the operation of the electrolytic cell under test are obtained, and the actual current, actual voltage, and actual temperature of the electrolytic cell under test during its working time are preprocessed, specifically as follows:

[0012] First, remove outliers from the actual current, actual voltage, and actual temperature of the electrolytic cell under test during its operating time.

[0013] Then, the actual current, actual voltage, and actual temperature of the electrolytic cell under test after removing outliers are smoothed.

[0014] Finally, the actual current, actual voltage, and actual temperature of the electrolytic cell under test after smoothing are normalized to obtain the actual current, actual voltage, and actual temperature of the electrolytic cell under test after preprocessing.

[0015] Furthermore, the step of removing outliers from the actual current, actual voltage, and actual temperature of the electrolytic cell under test during its operating time specifically involves:

[0016] Draw box plots corresponding to the actual operating voltage, actual temperature and actual current data of the electrolytic cell under test, and remove the data that falls outside the upper and lower edges.

[0017] Furthermore, the actual current, actual voltage, and actual temperature of the electrolytic cell under test, after removing outliers, are smoothed, specifically as follows:

[0018]

[0019] in, This is the smoothed value of the data over the electrolytic cell's operating time n, where i is the operating time label and γ is the smoothing parameter. These are the raw data values ​​for the electrolytic cell's working time n;

[0020] The original data values ​​are the actual current value, actual voltage value, or actual temperature value of the electrolytic cell.

[0021] Furthermore, in step S3, the residual r' of the actual operating voltage of the electrolytic cell under test is obtained using the predicted voltage and the pre-processed actual voltage. i Specifically:

[0022]

[0023] in, It is the predicted voltage for operating time i, U' el,i It is the actual voltage after preprocessing and working time i.

[0024] Furthermore, in S4, the residual r' of the actual operating voltage of the electrolytic cell under test is used. i To determine whether the current hydrogen production system is malfunctioning, the following steps are taken:

[0025] First, obtain the residual r' of the actual operating voltage of the electrolytic cell. i The standard deviation σ' is as follows:

[0026]

[0027] Then, determine whether there is a fault in the hydrogen production system based on σ':

[0028] If the residual r' is in the range [-3σ', 3σ'] i If the quantity is greater than the preset percentage, it indicates that the current hydrogen production system is normal; otherwise, it indicates that the current hydrogen production system is malfunctioning.

[0029] Furthermore, the fault detection model for the hydrogen production system is obtained through the following methods:

[0030] Step 1: Simulate and generate the working voltage, temperature, and current of the electrolyzer in the hydrogen production system.

[0031] Step 2: Compile the simulated electrolytic cell operating voltage, temperature, and current into a dataset. Divide the dataset into a training set and a test set. Use the training set to train a KAN network based on Kolmogorov-Arnold temporal attention to obtain the trained KAN network based on Kolmogorov-Arnold temporal attention.

[0032] Step 3: Test the trained KAN network based on Kolmogorov-Arnold temporal attention using the test set to obtain the predicted operating voltage;

[0033] Step 4: Use the predicted operating voltage obtained in Step 3 and the simulated operating voltage to determine whether the trained KAN network based on Kolmogorov-Arnold time attention should be saved. If it is saved, save the trained KAN network based on Kolmogorov-Arnold time attention as a fault detection model for the hydrogen production system; otherwise, return to Step 1.

[0034] Furthermore, the KAN network based on Kolmogorov-Arnold temporal attention includes: a temporal attention module enhanced with Kolmogorov-Arnold, a feature fusion module, and a KAN module;

[0035] The Kolmogorov-Arnold-enhanced time attention module extracts timing features using the electrolytic cell's operating temperature and current.

[0036] The feature fusion module is used to multiply the time-series features by the electrolytic cell operating temperature to obtain temperature features, and to multiply the time-series features by the electrolytic cell operating current to obtain current features.

[0037] The KAN module uses current and temperature characteristics to obtain the operating voltage of the electrolytic cell.

[0038] Furthermore, the KAN module obtains the electrolytic cell operating voltage using current and temperature characteristics, specifically:

[0039]

[0040] Among them, U el U is the operating voltage of the electrolytic cell. r It is a reversible voltage; I el For current characteristics, r1 is the fixed ohmic resistance of the electrolytic cell, and r2 is the temperature-dependent ohmic resistance; T el Temperature characteristics; k el To activate the overpressure parameter, k T3 This is the mass transfer overpressure parameter, where A is the surface area of ​​the electrolytic cell electrode, and k is the mass transfer overpressure parameter.t1 It is a parameter that reflects the reduction of activation overvoltage effect with increasing temperature, k t2 It is a parameter that reflects the effect of temperature change on the electrolyte conductivity and ion diffusion rate.

[0041] Further, in step four, the predicted operating voltage obtained in step three and the simulated operating voltage are used to determine whether the trained KAN network based on Kolmogorov-Arnold time attention should be saved. If it is saved, the trained KAN network based on Kolmogorov-Arnold time attention is saved as a fault detection model for the hydrogen production system; otherwise, return to step one, specifically:

[0042] Step 41: Utilize the simulated electrolytic cell operating voltage U el,i and predicted operating voltage Get U el,i and residuals:

[0043]

[0044] in, U is the predicted operating voltage for operating time i. el,i It is the operating voltage of the operating time i generated by simulation;

[0045] Step 42: Determine whether the number of residual data in the range [-3σ, 3σ] is greater than a preset percentage. If it is greater than the preset percentage, use the currently trained KAN model as the fault detection model for the hydrogen production system and save it; otherwise, return to Step 1.

[0046] Where σ is the residual r i The standard deviation.

[0047] The beneficial effects of this invention are as follows:

[0048] This invention proposes a fault diagnosis method for hydrogen production systems. First, it employs a Kolmogorov-Arnold time attention model to extract features from the electrolyzer's operating voltage and temperature. Then, it utilizes a KAN network model to obtain the predicted voltage of the electrolyzer. Based on the relationship between the electrolyzer's current, temperature, and voltage, this invention establishes a fault diagnosis model for the hydrogen production system. This model leverages the characteristic that the relationship between the electrolyzer's current, temperature, and voltage remains unaffected during normal operation, but changes during system malfunctions, enabling accurate subsequent determination of whether a fault has occurred in the hydrogen production system. Furthermore, this invention determines whether a fault has occurred in the hydrogen production system based on the residual between the actual operating voltage and the predicted voltage of the electrolyzer, providing accurate fault diagnosis results in real time and improving both accuracy and real-time performance. Attached Figure Description

[0049] Figure 1 This is a flowchart of the present invention;

[0050] Figure 2 This is a schematic diagram of the Pauta criterion. Detailed Implementation

[0051] Specific implementation method one: as follows Figure 1 As shown, the specific process of the hydrogen production system fault diagnosis method based on Kolmogorov-Arnold time attention KAN network in this embodiment is as follows:

[0052] S1. Use current sensor, voltage sensor and temperature sensor to obtain the actual current, actual voltage and actual temperature at each working time point during the operation of the electrolytic cell under test, and preprocess the actual current, actual voltage and actual temperature of the electrolytic cell under test during the working time to obtain the preprocessed actual current, actual voltage and actual temperature of the electrolytic cell under test during the working time.

[0053] The preprocessing specifically includes:

[0054] First, remove outliers from the actual current, actual voltage, and actual temperature of the electrolytic cell under test during its operating time:

[0055] Draw box plots for the actual operating voltage, actual temperature and actual current data of the electrolytic cell under test. Data falling outside the upper and lower edges are outliers.

[0056] Then, the actual current, actual voltage, and actual temperature of the electrolytic cell under test after removing outliers are smoothed, specifically as follows:

[0057]

[0058] in, This is the smoothed value of the data over the electrolytic cell's operating time n, where i is the operating time label and γ is the smoothing parameter. These are the raw data values ​​over the working time n of the electrolytic groove;

[0059] The raw data values ​​are the actual current value, the actual voltage value, or the actual temperature value.

[0060] Finally, the actual current, actual voltage, and actual temperature of the electrolytic cell under test after smoothing are normalized to obtain the actual current, actual voltage, and actual temperature of the electrolytic cell under test after preprocessing.

[0061] In this step, outliers may be caused by sensor malfunctions or external interference, while noisy data can mask the true operating status of the device, and duplicate data increases the burden of data processing. Data cleaning improves data quality and provides more accurate data support for subsequent analysis. Furthermore, data normalization is necessary to convert data of different dimensions and ranges to the same scale, facilitating input and processing by the neural network.

[0062] In the fault diagnosis process of electrolytic hydrogen production units, data extraction is necessary for both simulation and actual data. Various sensors, such as current, voltage, and temperature sensors, are installed on the unit to collect various parameter data in real time during operation. These sensors can accurately capture the operating status information of the unit under different operating conditions, including current magnitude, voltage stability, and temperature changes. Simultaneously, information such as the unit's operating conditions and environmental conditions, including load status and ambient temperature, also needs to be recorded. This data will serve as the basis for subsequent constant-pattern model establishment and fault diagnosis. During data extraction, to ensure the accuracy and reliability of the data, preprocessing is required. In this invention, the constant-pattern model refers to a KAN network based on Kolmogorov-Arnold time attention, which learns the mapping relationship between voltage, temperature, and current in the electrolyzer.

[0063] S2. Input the actual current and actual temperature of the pre-processed electrolyzer during its current working time into the hydrogen production system fault detection model to obtain the predicted voltage;

[0064] S3. Obtain the residual r' of the actual operating voltage of the electrolytic cell under test using the predicted voltage and the actual voltage after preprocessing. i Specifically:

[0065]

[0066] in, It is the predicted voltage for operating time i, U' el,i It is the actual voltage during the preprocessing working time i;

[0067] S4. Utilize the residual r' of the actual operating voltage of the electrolytic cell under test. i To determine if the current hydrogen production system is malfunctioning, the following steps are taken:

[0068] If the residual r' is in the range [-3σ', 3σ'] i If the quantity is greater than the preset percentage, it indicates that the current hydrogen production system is normal; otherwise, it indicates that the current hydrogen production system is malfunctioning.

[0069]

[0070] Where σ' is the residual r' of the actual operating voltage of the electrolytic cell. i The standard deviation is given by i, where i is the work time number and n is the total number of work time points.

[0071] Specific Implementation Method Two: The fault detection model for the hydrogen production system is obtained through the following method:

[0072] Step 1: Use Simulink in MATLAB to simulate and generate the working voltage, temperature, and current of the electrolyzer in the hydrogen production system.

[0073] Step 2: The electrolytic cell operating voltage, electrolytic cell temperature and electrolytic cell operating current generated in Step 1 are used to form a dataset. The dataset is divided into a training set and a test set in a 2:1 ratio. The training set is used to train a KAN network based on Kolmogorov-Arnold temporal attention to obtain a trained KAN network based on Kolmogorov-Arnold temporal attention.

[0074] The KAN network based on Kolmogorov-Arnold temporal attention includes: a temporal attention module enhanced with Kolmogorov-Arnold, a feature fusion module, and a KAN module;

[0075] The Kolmogorov-Arnold-enhanced time attention module is used to extract timing features based on the electrolyzer's operating temperature and current.

[0076] The feature fusion module is used to multiply the time-series features by the electrolytic cell operating temperature to obtain temperature features, and to multiply the time-series features by the electrolytic cell operating current to obtain current features.

[0077] The KAN module obtains the electrolytic cell operating voltage using current and temperature characteristics, specifically:

[0078] U el =F(T) el ,I el )

[0079]

[0080] Among them, U el U is the operating voltage of the electrolytic cell. r I is the reversible voltage of the electrolytic cell that changes with temperature; el The current characteristic is represented by r1, which is the fixed ohmic resistance of the electrolytic cell, reflecting the anode reactance of the electrolytic cell material itself; r2 is the temperature-dependent ohmic resistance, reflecting the change in electrolyte / membrane conductivity; T el Temperature characteristics; k el To activate the overpressure parameter, kT3 This is the mass transfer overpressure parameter, where A is the surface area of ​​the electrolytic cell electrode, and k is the mass transfer overpressure parameter. t1 It is a parameter that reflects the reduction of activation overvoltage effect with increasing temperature, k t2 It is a parameter that reflects the effect of temperature change on the electrolyte conductivity and ion diffusion rate.

[0081] Step 3: Test the trained KAN network based on Kolmogorov-Arnold temporal attention using a test set to obtain the predicted operating voltage.

[0082] Step 4: Utilize the predicted operating voltage obtained in Step 3 and the simulated operating voltage U el Determine whether the trained KAN network based on Kolmogorov-Arnold time attention has been saved. If so, save the trained KAN network based on Kolmogorov-Arnold time attention as a fault detection model for the hydrogen production system; otherwise, return to step one, which is as follows:

[0083] Step 41: Utilize the simulated electrolytic cell operating voltage U el and predicted operating voltage Get U el and residuals:

[0084]

[0085] Step 4.2: Determine r within the range [-3σ, 3σ]. i If the number of data points exceeds a preset percentage, then use the currently trained KAN model as the fault detection model for the hydrogen production system and save it; otherwise, return to step one.

[0086] Where σ is the residual r i The standard deviation.

[0087] This invention employs the core idea of ​​establishing a normal mode, extracting the "invariants" of the normal mode of the electrolytic hydrogen production unit from various disturbance factors such as operating conditions and environmental conditions, as well as changes in various measurable parameters. This "invariant" can be a parameter or a mapping relationship, and it needs to satisfy two conditions: when the electrolytic hydrogen production unit is operating normally, despite significant changes in operating conditions and environmental conditions, the "invariant" remains essentially unchanged or changes very little; however, when the electrolytic hydrogen production unit malfunctions, the "invariant" will change significantly. To extract such an "invariant," feature selection of the collected data is first required. The electrolysis unit can be considered a nonlinear load, and studying its UI characteristics can improve gas production efficiency. The voltage and current characteristics of the electrolyzer are determined by the electrolyzer temperature, and due to its high nonlinearity, general curve fitting cannot represent them.

[0088] Due to the limited availability of actual data for electrolytic hydrogen production units, the trained constant-pattern model cannot accurately represent the desired outcome. This invention obtains a large amount of simulation data through a simulation device. Based on the Kolmogorov-Arnold time attention-based KAN network, a constant-pattern model domain is constructed by extracting simulation data. This domain closely approximates the correct constant-pattern model. The domain is then corrected using actual data to ensure accurate representation. The KAN neural network, based on the Kolmogorov-Arnold theorem, possesses powerful function approximation capabilities and can effectively handle complex nonlinear relationships. By training the KAN neural network, it learns the intrinsic mapping relationship between feature parameters, thereby extracting the "invariants" that satisfy the conditions. During training, an error backpropagation algorithm is used to optimize the network parameters, ensuring that the network's output "invariants" remain stable during normal operation and change significantly during fault occurrences. The constant-pattern model established in this way can accurately describe the characteristics of the electrolytic hydrogen production unit under normal operating conditions, providing a benchmark for subsequent fault diagnosis.

[0089] If a well-trained KAN network can effectively represent the mapping relationship between the input and output data, then the distribution of the residuals approximately follows a normal distribution N(0,σ). 2 N(0,σ) 2 The probability density function of ) is as follows Figure 2 As shown, δ is the standard score, and φ(δ) is the probability density function, where the probability of the residual being within [-3σ, 3σ] is 99.73%. Therefore, the Pauta criterion (i.e., the 3σ criterion) can be used to determine the threshold for fault detection. For a sample to be detected, the fitted value of the voltage is first calculated by the trained network, and then the residual is calculated between the predicted value and the actual value. If the absolute value of the residual does not exceed 3σ, it is detected as a normal sample; otherwise, it is detected as a fault sample.

[0090] This invention designs a KAN (Kolmogorov-Arnold Network) neural network based on Kolmogorov-Arnold time attention, which can extract normal and fault modes of hydrogen electrolysis units in complex multidimensional data environments, thereby achieving accurate fault diagnosis. The core idea of ​​the normal mode extraction method is to extract the "invariants" of the gas turbine's normal mode from various disturbance factors such as operating conditions and environmental conditions, as well as changes in various measurable parameters. These "invariants" can be, but are not limited to, a parameter or a mapping relationship. These "invariants" need to satisfy the following two conditions:

[0091] First, when the device is operating normally, even though various disturbances such as operating conditions and environmental conditions change significantly, this "invariant" remains basically unchanged or changes very little.

[0092] Second, this "invariant" will change significantly when the device malfunctions.

[0093] Once an "invariant" satisfying the above two conditions is extracted, sensitive and robust gas turbine fault detection can be achieved by detecting changes in this "invariant." This invention seeks "invariants" in the normal mode of the equipment.

[0094] In fault diagnosis of hydrogen electrolysis units, the KAN network can learn the normal operating mode of the system from a large amount of sensor data and promptly detect anomalies when faults occur. A significant advantage of the KAN network is its ability to effectively handle situations with scarce data, especially in some industrial environments where sensor data may be missing or noisy. Through its adaptive modeling characteristics, the KAN network can better extract useful information from incomplete data. This makes the KAN network exhibit strong robustness in fault diagnosis of hydrogen electrolysis units. Furthermore, when dealing with nonlinear and time-varying fault modes, the KAN network can automatically adjust its internal structure and parameters, ensuring that the model maintains high diagnostic accuracy. This provides strong technical support for real-time monitoring and fault early warning of hydrogen electrolysis units. Compared with traditional methods, the KAN network not only provides more accurate fault diagnosis results but also offers higher real-time performance, adapting to rapidly changing industrial environments.

[0095] Example: To verify the beneficial effects of the present invention, the following experiments were conducted:

[0096] The inputs and outputs of the constant mode model of the hydrogen production system (based on a KAN network with Kolmogorov-Arnold temporal attention) are shown in Table 1:

[0097] Table 1

[0098]

[0099] Since there was no fault data, the detection performance of the test model on normal data was tested, and the final detection results are shown in Table 2. The fault diagnosis method for hydrogen production systems proposed in this invention can effectively detect the normal operating status of hydrogen production systems. Compared with traditional methods, the introduced Kolmogorov-Arnold time attention KAN structure performs excellently in fault diagnosis. This indicates that the present invention has high accuracy and robustness in fault detection.

[0100] Table 2

[0101]

Claims

1. A fault diagnosis method for hydrogen production systems based on Kolmogorov–Arnold temporal attention KAN networks, characterized in that... The specific process of the method is as follows: S1. Obtain the actual current, actual voltage, and actual temperature at each working time point during the operation of the electrolytic cell under test, and preprocess the actual current, actual voltage, and actual temperature of the electrolytic cell under test during its working time to obtain the preprocessed actual current, actual voltage, and actual temperature of the electrolytic cell under test during its working time. S2. Input the actual current and actual temperature of the pre-treated electrolyzer during its operating time into the hydrogen production system fault detection model to obtain the predicted voltage; The fault detection model for the hydrogen production system was obtained through the following method: Step 1: Simulate and generate the working voltage, temperature, and current of the electrolyzer in the hydrogen production system. Step 2: Compile the simulated electrolytic cell operating voltage, temperature, and current into a dataset. Divide the dataset into a training set and a test set. Use the training set to train a KAN network based on Kolmogorov–Arnold temporal attention to obtain the trained KAN network based on Kolmogorov–Arnold temporal attention. The KAN network based on Kolmogorov–Arnold temporal attention includes: a temporal attention module enhanced by Kolmogorov–Arnold, a feature fusion module, and a KAN module; The Kolmogorov–Arnold-enhanced time attention module extracts timing features using the electrolytic cell's operating temperature and current. The feature fusion module is used to multiply the time-series features by the electrolytic cell operating temperature to obtain temperature features, and to multiply the time-series features by the electrolytic cell operating current to obtain current features. The KAN module obtains the electrolytic cell operating voltage using current and temperature characteristics, specifically: in, This is the operating voltage of the electrolytic cell. It is a reversible voltage; Current characteristics, For the fixed ohmic resistor of the electrolytic cell, It is a temperature-dependent ohmic resistor; Temperature characteristics; To activate the overpressure parameters, It is a mass transfer overpressure parameter. The surface area of ​​the electrodes in the electrolytic cell. It is a parameter that reflects the reduction of activation overvoltage effect with increasing temperature. These are parameters that reflect the effect of temperature changes on the electrolyte conductivity and ion diffusion rate. Step 3: Test the trained KAN network based on Kolmogorov–Arnold temporal attention using the test set to obtain the predicted operating voltage; Step 4: Use the predicted operating voltage obtained in Step 3 and the simulated operating voltage to determine whether the trained KAN network based on Kolmogorov-Arnold time attention should be saved. If it is saved, save the trained KAN network based on Kolmogorov-Arnold time attention as a fault detection model for the hydrogen production system; otherwise, return to Step 1. S3. Obtain the residual of the actual operating voltage of the electrolytic cell under test using the predicted voltage and the pre-processed actual voltage. ; S4. Utilize the residual of the actual operating voltage of the electrolytic cell under test. Determine if the current hydrogen production system is malfunctioning.

2. The method for fault diagnosis of hydrogen production systems based on Kolmogorov–Arnold temporal attention KAN networks according to claim 1, characterized in that: In step S1, the actual current, actual voltage, and actual temperature at each working time point during the operation of the electrolytic cell under test are obtained, and the actual current, actual voltage, and actual temperature of the electrolytic cell under test during its working time are preprocessed, specifically as follows: First, remove outliers from the actual current, actual voltage, and actual temperature of the electrolytic cell under test during its operating time. Then, the actual current, actual voltage, and actual temperature of the electrolytic cell under test after removing outliers are smoothed. Finally, the actual current, actual voltage, and actual temperature of the electrolytic cell under test after smoothing are normalized to obtain the actual current, actual voltage, and actual temperature of the electrolytic cell under test after preprocessing.

3. The method for fault diagnosis of hydrogen production systems based on Kolmogorov–Arnold temporal attention KAN networks according to claim 2, characterized in that: The process of removing outliers from the actual current, actual voltage, and actual temperature of the electrolytic cell under test during its operating time specifically involves: Draw box plots corresponding to the actual operating voltage, actual temperature and actual current data of the electrolytic cell under test, and remove the data that falls outside the upper and lower edges.

4. The method for fault diagnosis of hydrogen production systems based on Kolmogorov–Arnold temporal attention KAN networks according to claim 3, characterized in that: After removing outliers, the actual current, actual voltage, and actual temperature of the electrolytic cell under test are smoothed according to the following steps: in, This is a smoothed value of the data over the electrolytic cell's operating time n. It is a work time label. It is a smoothing parameter. These are the raw data values ​​for the electrolytic cell's working time n; The original data values ​​are the actual current value, actual voltage value, or actual temperature value of the electrolytic cell.

5. The method for fault diagnosis of hydrogen production systems based on Kolmogorov–Arnold temporal attention KAN networks according to claim 4, characterized in that: In step S3, the residual of the actual operating voltage of the electrolytic cell under test is obtained by using the predicted voltage and the pre-processed actual voltage. Specifically: in, It is the predicted voltage for operating time i. It is the actual voltage after preprocessing and working time i.

6. The method for fault diagnosis of hydrogen production systems based on Kolmogorov–Arnold temporal attention KAN networks according to claim 5, characterized in that: The residual voltage of the actual operating voltage of the electrolytic cell under test in step S4 is used in this step. To determine whether the current hydrogen production system is malfunctioning, the following steps are taken: First, obtain the residual of the actual operating voltage of the electrolytic cell. Standard deviation Specifically: Then, according to Determine if there is a malfunction in the hydrogen production system: If in residuals within the range If the quantity is greater than the preset percentage, it indicates that the current hydrogen production system is normal; otherwise, it indicates that the current hydrogen production system is malfunctioning.

7. The method for fault diagnosis of hydrogen production systems based on Kolmogorov–Arnold temporal attention KAN networks according to claim 6, characterized in that: In step four, the predicted operating voltage obtained in step three and the simulated operating voltage are used to determine whether the trained KAN network based on Kolmogorov-Arnold time attention should be saved. If it is saved, the trained KAN network based on Kolmogorov-Arnold time attention is saved as a fault detection model for the hydrogen production system; otherwise, return to step one, specifically: Step 41: Utilize the simulated electrolytic cell operating voltage and predicted operating voltage Get and residuals: in, It is the predicted operating voltage for operating time i. It is the operating voltage of the operating time i generated by simulation; Step 42, Determine if... If the number of residual data within the range is greater than a preset percentage, then the currently trained KAN model is used as the fault detection model for the hydrogen production system and saved; otherwise, return to step one. in, It is a residual The standard deviation.

Images