Fault diagnosis method for filter of alkaline electrolysis system

Abstract

The invention relates to the technical field of water electrolysis hydrogen production, and discloses an alkaline electrolysis system filter fault diagnosis method, which comprises the following steps: acquiring the front and back pressure difference of a filter, the outlet flow of a circulating pump and the motor power of the circulating pump in real time, filtering, calculating the change rate of each parameter and the standard deviation in a preset time window, and constructing a multi-dimensional feature vector; threshold logic is adopted for preliminary judgment, and mild, moderate and severe blockage, filter element breakage and sensor faults are distinguished. Meanwhile, a similar working condition template is retrieved from a multi-working condition feature template library covering 10%-100% of rated current by taking a current running current value as an index, and a weighted Euclidean distance is calculated to determine an intelligent matching result. According to the method, the front and back pressure difference of the filter is collected in real time, and the multi-dimensional feature vector containing the change rate and the standard deviation is constructed, so that the system can accurately distinguish light blockage, medium blockage, heavy blockage, filter element breakage and sensor faults, and false report and missing report of traditional single pressure difference monitoring are avoided.

Description

Technical Field

[0001] This invention relates to the field of water electrolysis for hydrogen production technology, and in particular to a method for diagnosing filter failures in an alkaline electrolysis system. Background Technology

[0002] The alkaline water electrolysis hydrogen production system decomposes water into hydrogen and oxygen through an electrolyzer. The electrolyte is continuously transported between the electrolyzer and the storage tank by a circulating pump to maintain the uniformity of electrolyte concentration, temperature, and reaction rate. A filter is installed at the inlet of the circulating pump to intercept solid particles and precipitates in the electrolyte, preventing them from entering the pump body or the electrolyzer.

[0003] In existing technologies, filter condition monitoring primarily employs differential pressure sensors installed before and after the filter to determine the degree of filter element clogging based on real-time differential pressure values. When the differential pressure exceeds a preset threshold, the system triggers an alarm or shuts down. Some solutions add manual inspection to the differential pressure monitoring, where operators judge the filter condition based on experience by observing flow meter readings or pump vibrations.

[0004] However, in the process of implementing the technical solution of this application, the inventors of this application discovered that the above-mentioned technology has at least the following technical problems: single differential pressure monitoring cannot distinguish between filter element blockage, filter element rupture or sensor failure, which causes the system to still treat the filter element as blockage when it is ruptured, or to generate false alarms when the sensor fails, resulting in the diagnostic results not matching the actual filter status. Summary of the Invention

[0005] To overcome the above deficiencies, this invention provides a method for diagnosing filter faults in an alkaline electrolysis system. This method aims to improve the problem in the prior art where single differential pressure monitoring cannot distinguish between filter element blockage, filter element rupture, or sensor failure, causing the system to treat a ruptured filter element as a blocked one, resulting in diagnostic results that do not match the actual filter condition.

[0006] This invention provides the following technical solution: a method for diagnosing filter faults in an alkaline electrolysis system, comprising the following steps:

[0007] S1. Real-time acquisition of pressure difference across the filter, circulation pump outlet flow rate, and circulation pump motor power;

[0008] S2. The collected pressure difference before and after the filter, the outlet flow rate of the circulating pump, and the power of the circulating pump motor are filtered to eliminate high-frequency noise interference.

[0009] S3. Based on the filtered data, calculate the rate of change of the pressure difference before and after the filter, the rate of change of the circulating pump outlet flow rate, the rate of change of the circulating pump motor power, as well as the standard deviation of the pressure difference before and after the filter, the standard deviation of the circulating pump outlet flow rate, and the standard deviation of the circulating pump motor power within a preset time window, and construct a multi-dimensional feature vector containing the above rates of change and standard deviations.

[0010] S4. Use threshold logic to perform preliminary fault identification on the multidimensional feature vector and obtain the threshold logic identification result;

[0011] S5. Using the current operating current value as an index, retrieve the operating condition group whose deviation from the current operating current value is within a preset range from the pre-built multi-condition feature template library, calculate the weighted Euclidean distance between the multi-dimensional feature vector and the retrieved various condition templates, and determine the intelligent matching and discrimination result based on the minimum weighted Euclidean distance.

[0012] S6. Perform decision fusion on the threshold logic discrimination result and the intelligent matching discrimination result. When the two are consistent, output the fault status directly. When the two are inconsistent, calculate the threshold logic confidence and the intelligent matching confidence respectively, and select the result with higher confidence as the final fault diagnosis conclusion.

[0013] S7. Output the corresponding control strategy based on the final fault diagnosis conclusion.

[0014] Preferably, the threshold logic discrimination includes:

[0015] When the pressure difference across the real-time filter is greater than the first pressure threshold, the outlet flow rate of the circulating pump is less than the first flow rate threshold, and the motor power of the circulating pump is greater than the first power threshold, it is determined to be a mild blockage.

[0016] When the pressure difference across the real-time filter is greater than the second pressure threshold, the outlet flow rate of the circulation pump is less than the second flow threshold, and the motor power of the circulation pump is greater than the second power threshold, it is determined to be a moderate blockage.

[0017] When the pressure difference across the real-time filter is greater than the third pressure threshold, the outlet flow rate of the circulating pump is less than the third flow rate threshold, and the motor power of the circulating pump is greater than the third power threshold, it is determined to be a severe blockage.

[0018] When the pressure difference across the real-time filter is less than the fourth pressure threshold, the outlet flow rate of the circulating pump is greater than the fourth flow threshold, and the standard deviation of the circulating pump motor power is greater than the first power fluctuation threshold, the filter element is determined to be ruptured.

[0019] When the data from a single sensor is abnormal, but the coupling relationship of the other parameters conforms to the normal operating rules, it is determined to be a sensor malfunction.

[0020] Preferably, the construction process of the multi-condition feature template library includes:

[0021] Within the range of 10% to 100% of the rated current, the system acquires multi-dimensional feature vectors under normal, slightly blocked, moderately blocked, heavily blocked, filter element broken, and sensor fault states.

[0022] The collected multi-dimensional feature vectors, along with their corresponding operating current values ​​and status labels, are classified and stored to form a multi-condition feature template library covering the entire operating current range.

[0023] Preferably, the dynamic update process of the multi-condition feature template library includes:

[0024] Once the fault status is confirmed, the multi-dimensional feature vector within the fault time period is automatically extracted and labeled with status tags, and added as a new template sample to the multi-condition feature template library.

[0025] Periodically recalculate the feature mean and variance of template samples under the same working condition and state in the multi-working-condition feature template library to optimize the templates.

[0026] Preferably, the confidence calculation in the decision fusion includes:

[0027] For clogging-related faults, the threshold logic confidence is calculated based on the ratio of the real-time pressure difference before and after the filter to the corresponding pressure threshold and the ratio of the circulating pump outlet flow rate to the corresponding flow rate threshold.

[0028] For filter element rupture type faults, the threshold logic confidence is calculated based on the ratio of the real-time pressure difference across the filter to the fourth pressure threshold and the ratio of the circulating pump outlet flow rate to the second flow rate threshold.

[0029] For sensor faults, the threshold logic confidence level is calculated based on the degree of deviation between abnormal sensor data and normal coupling relationship;

[0030] The confidence level of intelligent matching is calculated based on the minimum weighted Euclidean distance between the multidimensional feature vector and the most similar state template.

[0031] Preferably, the specific process of retrieving the operating condition group whose deviation from the current operating current value is within a preset range is as follows:

[0032] Using the current operating current value as the index key, retrieve all operating condition groups whose operating current values ​​deviate from the current operating current value within ±5% from the multi-operating condition feature template library.

[0033] Preferably, the calculation process for the weighted Euclidean distance is as follows:

[0034] After assigning weights to each dimension component of the multidimensional feature vector, the sum of the weighted squared differences between the current multidimensional feature vector and the corresponding dimension component of each state template is calculated, and then the square root is taken to obtain the weighted Euclidean distance.

[0035] Preferably, the preset time window is the most recent N sampling periods of continuous acquisition, where N is an integer greater than 1.

[0036] Preferably, the control strategy includes:

[0037] A maintenance reminder will be issued when there is a minor blockage;

[0038] When moderate congestion occurs, reduce the system load and issue a cleaning command;

[0039] Emergency shutdown and alarm triggered in case of severe blockage or filter element rupture;

[0040] When a sensor fails, switch to a backup sensor and record the fault log.

[0041] Preferably, the method is performed by a processing unit installed in the alkaline water electrolysis hydrogen production system, the processing unit being connected to pressure sensors before and after the filter, a flow meter at the outlet of the circulating pump, and a power acquisition module for the circulating pump motor.

[0042] The present invention has the following beneficial effects:

[0043] 1. This invention collects real-time differential pressure before and after the filter, the outlet flow rate of the circulating pump, and the power of the circulating pump motor to construct a multi-dimensional feature vector containing the rate of change and standard deviation. Combined with threshold logic and weighted Euclidean distance matching of a multi-condition feature template library, the system can accurately distinguish between mild blockage, moderate blockage, severe blockage, filter element rupture, and sensor failure, avoiding false alarms and missed alarms of traditional single differential pressure monitoring.

[0044] 2. In this invention, a dual criterion mechanism of threshold logic and intelligent matching is adopted in parallel, and decision fusion is performed based on confidence when the results are inconsistent, so as to ensure high consistency of diagnostic conclusions and low misjudgment rate under different operating conditions, and significantly enhance the system's stable operation capability in complex environments.

[0045] 3. In this invention, by identifying mild blockages early and issuing maintenance reminders, automatically reducing load when there is moderate blockage, and shutting down in case of severe blockage or rupture, combined with a dynamic self-learning template library update mechanism, the system has the ability to predict the development trend of faults, reduce unplanned downtime, and ensure long-term unattended operation of alkaline electrolysis hydrogen production.

[0046] 4. In this invention, the multi-condition feature template library covers the full range of rated current from 10% to 100%, and new samples are automatically added after fault confirmation, and the mean and variance of the templates are optimized regularly, so that the diagnostic model can continuously evolve as the equipment performance degrades and the electrolyte characteristics change, and maintain high diagnostic accuracy in the long term. Attached Figure Description

[0047] Figure 1This is a flowchart of a method for diagnosing filter faults in an alkaline electrolysis system proposed in this invention. Detailed Implementation

[0048] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0049] Example 1:

[0050] Reference Figure 1 In the first embodiment of the present invention, a method for diagnosing filter failures in an alkaline electrolysis system is provided, comprising the following steps:

[0051] S1. Real-time acquisition of pressure difference across the filter, circulation pump outlet flow rate, and circulation pump motor power;

[0052] S2. The collected pressure difference before and after the filter, the outlet flow rate of the circulating pump, and the power of the circulating pump motor are filtered to eliminate high-frequency noise interference.

[0053] S3. Based on the filtered data, calculate the rate of change of the pressure difference before and after the filter, the rate of change of the circulating pump outlet flow rate, the rate of change of the circulating pump motor power, as well as the standard deviation of the pressure difference before and after the filter, the standard deviation of the circulating pump outlet flow rate, and the standard deviation of the circulating pump motor power within a preset time window, and construct a multi-dimensional feature vector containing the above rates of change and standard deviations.

[0054] S4. Use threshold logic to perform preliminary fault identification on the multidimensional feature vector and obtain the threshold logic identification result;

[0055] S5. Using the current operating current value as an index, retrieve the operating condition group whose deviation from the current operating current value is within a preset range from the pre-built multi-condition feature template library, calculate the weighted Euclidean distance between the multi-dimensional feature vector and the retrieved various condition templates, and determine the intelligent matching and discrimination result based on the minimum weighted Euclidean distance.

[0056] S6. Perform decision fusion on the threshold logic discrimination result and the intelligent matching discrimination result. When the two are consistent, output the fault status directly. When the two are inconsistent, calculate the threshold logic confidence and the intelligent matching confidence respectively, and select the result with higher confidence as the final fault diagnosis conclusion.

[0057] S7. Output the corresponding control strategy based on the final fault diagnosis conclusion.

[0058] Specifically, the processing unit receives differential pressure signals from pressure sensors before and after the filter, flow signals from the circulating pump outlet flow meter, and power signals from the circulating pump motor power acquisition module. The processing unit executes a low-pass filtering algorithm to remove frequencies higher than the cutoff frequency from the differential pressure, flow, and power signals. In each sampling period, the processing unit calculates the rate of change of differential pressure as the difference between the differential pressures of two adjacent sampling periods divided by the sampling interval; the rate of change of flow rate as the difference between the flow rates of two adjacent sampling periods divided by the sampling interval; and the rate of change of power as the difference between the power rates of two adjacent sampling periods divided by the sampling interval. Within a preset time window, the processing unit calculates the standard deviation of differential pressure as the square root of the variance of the differential pressure samples within the window; the standard deviation of flow rate as the square root of the variance of the flow rate samples within the window; and the standard deviation of power as the square root of the variance of the power samples within the window. The processing unit combines the differential pressure, flow rate, power, rate of change of differential pressure, rate of change of flow rate, rate of change of power, standard deviation of differential pressure, standard deviation of flow rate, and standard deviation of power into a multi-dimensional feature vector.

[0059] The processing unit compares the differential pressure, flow rate, and power in the multi-dimensional feature vector with preset thresholds. When the differential pressure is greater than the first pressure threshold, the flow rate is less than the first flow rate threshold, and the power is greater than the first power threshold, a mild clogging result is output. When the differential pressure is greater than the second pressure threshold, the flow rate is less than the second flow rate threshold, and the power is greater than the second power threshold, a moderate clogging result is output. When the differential pressure is greater than the third pressure threshold, the flow rate is less than the third flow rate threshold, and the power is greater than the third power threshold, a severe clogging result is output. When the differential pressure is less than the fourth pressure threshold, the flow rate is greater than the fourth flow rate threshold, and the power standard deviation is greater than the first power fluctuation threshold, a filter element rupture result is output. When one of the differential pressure, flow rate, or power exceeds the normal range, while the other two parameters meet the normal coupling relationship, a sensor fault result is output.

[0060] The processing unit reads the current operating current value and uses this current value as the key to search the multi-condition feature template library for all operating condition groups with operating current value deviations within ±5%. The processing unit extracts template vectors corresponding to normal state, light blockage state, moderate blockage state, heavy blockage state, filter element rupture state, and sensor fault state from the retrieved operating condition groups. The processing unit applies a preset weight to each dimension of the multi-dimensional feature vector, calculates the weighted Euclidean distance with each template vector, and selects the state corresponding to the template with the smallest distance as the intelligent matching and discrimination result.

[0061] The processing unit compares the threshold logic discrimination result with the intelligent matching discrimination result. If they are the same, the state is directly taken as the final fault diagnosis conclusion. If they are different, the threshold logic confidence score and the intelligent matching confidence score are calculated. For blockage-type faults, the threshold logic confidence score is equal to the ratio of differential pressure to the corresponding pressure threshold plus the ratio of flow rate to the corresponding flow rate threshold. For filter element rupture-type faults, it is equal to the ratio of the fourth pressure threshold to the differential pressure plus the ratio of flow rate to the second flow rate threshold. For sensor-type faults, it is equal to the degree to which the abnormal parameter deviates from the normal coupling relationship. The intelligent matching confidence score is equal to the reciprocal of the minimum weighted Euclidean distance. The processing unit selects the discrimination result with the higher confidence score as the final fault diagnosis conclusion.

[0062] The processing unit executes control strategies based on the final fault diagnosis conclusion. For minor blockage, a maintenance reminder signal is sent. For moderate blockage, a signal to reduce system load and a cleaning command signal are sent. For severe blockage or filter element rupture, an emergency stop signal and an alarm signal are sent. In case of sensor failure, a signal to switch to a backup sensor is sent and a fault log is recorded.

[0063] Furthermore, the threshold logic judgment includes:

[0064] When the pressure difference across the real-time filter is greater than the first pressure threshold, the outlet flow rate of the circulating pump is less than the first flow rate threshold, and the motor power of the circulating pump is greater than the first power threshold, it is determined to be a mild blockage.

[0065] When the pressure difference across the real-time filter is greater than the second pressure threshold, the outlet flow rate of the circulation pump is less than the second flow threshold, and the motor power of the circulation pump is greater than the second power threshold, it is determined to be a moderate blockage.

[0066] When the pressure difference across the real-time filter is greater than the third pressure threshold, the outlet flow rate of the circulating pump is less than the third flow rate threshold, and the motor power of the circulating pump is greater than the third power threshold, it is determined to be a severe blockage.

[0067] When the pressure difference across the real-time filter is less than the fourth pressure threshold, the outlet flow rate of the circulating pump is greater than the fourth flow threshold, and the standard deviation of the circulating pump motor power is greater than the first power fluctuation threshold, the filter element is determined to be ruptured.

[0068] When the data from a single sensor is abnormal, but the coupling relationship of the other parameters conforms to the normal operating rules, it is determined to be a sensor malfunction.

[0069] Specifically, in each sampling period, the processing unit extracts real-time differential pressure, real-time flow rate, real-time power, and power standard deviation from the multi-dimensional feature vector. The first pressure threshold is set to 1.2 times the normal differential pressure under rated system conditions, the first flow rate threshold is set to 0.9 times the normal flow rate under rated system conditions, and the first power threshold is set to 1.1 times the normal power under rated system conditions. The processing unit performs a comparison operation; if the real-time differential pressure exceeds the first pressure threshold, the real-time flow rate is lower than the first flow rate threshold, and the real-time power exceeds the first power threshold, the threshold logic judgment result is marked as a mild blockage status code.

[0070] The second pressure threshold is set to 1.8 times the normal differential pressure under rated system conditions, the second flow threshold is set to 0.7 times the normal flow rate under rated system conditions, and the second power threshold is set to 1.3 times the normal power under rated system conditions. The processing unit continues to perform comparison operations. If the real-time differential pressure value exceeds the second pressure threshold, the real-time flow rate value is lower than the second flow rate threshold, and the real-time power value exceeds the second power threshold, then the threshold logic judgment result is marked as a moderate blockage status code.

[0071] The third pressure threshold is set to 2.5 times the normal differential pressure under rated system conditions, the third flow threshold is set to 0.5 times the normal flow rate under rated system conditions, and the third power threshold is set to 1.5 times the normal power under rated system conditions. The processing unit performs a comparison operation. If the real-time differential pressure value exceeds the third pressure threshold, the real-time flow rate value is lower than the third flow rate threshold, and the real-time power value exceeds the third power threshold, the threshold logic judgment result is marked as a severe blockage status code.

[0072] The fourth pressure threshold is set to 0.5 times the normal differential pressure under rated system conditions, and the first power fluctuation threshold is set to 2 times the normal power standard deviation under rated system conditions. The processing unit performs a comparison operation. If the real-time differential pressure value is lower than the fourth pressure threshold, the real-time flow rate value is higher than the fourth flow rate threshold, and the power standard deviation value exceeds the first power fluctuation threshold, then the threshold logic judgment result is marked as a filter element rupture status code, and the fourth flow rate threshold is set to 1.1 times the normal flow rate under rated system conditions.

[0073] The processing unit maintains a database of normal operation patterns, which stores polynomial fitting coefficients between differential pressure, flow rate, and power at rated current. When the processing unit detects that the output of one of the differential pressure sensor, flow sensor, or power sensor exceeds its range or remains constant for more than a preset period, it marks that sensor as abnormal. The processing unit replaces the historical values ​​of the abnormal sensor with predicted values ​​calculated from the database based on the other two parameters. If the deviation between the predicted and actual values ​​is less than a preset error, the other two parameters satisfy the normal coupling relationship. When the processing unit detects a single sensor abnormality and the other two parameters satisfy the normal coupling relationship, it marks the threshold logic judgment result as a sensor fault status code and records the abnormal sensor identifier.

[0074] Furthermore, the construction process of the multi-condition feature template library includes:

[0075] Within the range of 10% to 100% of the rated current, the system acquires multi-dimensional feature vectors under normal, slightly blocked, moderately blocked, heavily blocked, filter element broken, and sensor fault states.

[0076] The collected multi-dimensional feature vectors, along with their corresponding operating current values ​​and status labels, are classified and stored to form a multi-condition feature template library covering the entire operating current range.

[0077] Specifically, the processing unit executes an offline acquisition program during the initial system operation phase. The processing unit controls the electrolysis system current source to increase from 10% to 100% of the rated current in 10% increments. At each current increment, the processing unit maintains current stability for at least 30 minutes.

[0078] Under normal conditions, with the filter element clean and all sensors functioning correctly, the processing unit continuously acquires multidimensional feature vectors for at least 1000 sampling cycles. The processing unit calculates the mean vector of this set of vectors as the normal state template and records the corresponding current value and normal state label.

[0079] For mild clogging, the processing unit artificially adds 5% of the filter area as blockage material at the filter inlet. The processing unit acquires multidimensional feature vectors for at least 500 sampling periods at the same current step size. The processing unit calculates the mean vector as a template for mild clogging and records the corresponding current value and mild clogging status label.

[0080] For moderate clogging, the processing unit adds 20% of the filter area as blockage material. The processing unit collects multidimensional feature vectors for at least 500 sampling periods, calculates the mean vector as a template for moderate clogging, and records the corresponding current value and moderate clogging status label.

[0081] For severe clogging, the processing unit adds blockage material covering 50% of the filter area. The processing unit collects multidimensional feature vectors for at least 300 sampling periods, stopping collection before the system triggers a protection shutdown. The processing unit calculates the mean vector as a template for severe clogging and records the corresponding current value and severe clogging status label.

[0082] For filter element rupture, the processing unit creates a 2mm diameter through-hole on the filter element. The processing unit acquires multidimensional feature vectors for at least 200 sampling periods at the same current step size. The processing unit calculates the mean vector as a template for the filter element rupture state and records the corresponding current value and filter element rupture state label.

[0083] For sensor fault states, the processing unit disconnects the signal line of one of the differential pressure sensor, flow sensor, or power sensor in sequence. The processing unit collects multi-dimensional feature vectors for at least 300 sampling periods under the same current step size. The processing unit calculates the mean vector for each type of sensor fault as the corresponding sensor fault state template and records the corresponding current value, sensor fault state label, and fault sensor type.

[0084] The processing unit stores all mean vectors, corresponding current values, and status tags in non-volatile memory. The processing unit establishes a hash index table, using the current value as the key, the status tag as the secondary key, and the mean vector as the value. The processing unit stores six sets of status templates for each current step, forming a multi-condition feature template library covering 10% to 100% of the rated current.

[0085] Furthermore, the dynamic update process of the multi-condition feature template library includes:

[0086] Once the fault status is confirmed, the multi-dimensional feature vector within the fault time period is automatically extracted and labeled with status tags, and added as a new template sample to the multi-condition feature template library.

[0087] Periodically recalculate the feature mean and variance of template samples under the same working condition and state in the multi-working-condition feature template library to optimize the templates.

[0088] Specifically, the processing unit continuously records multidimensional feature vectors from all sampling periods into a circular buffer during system operation. This buffer contains data from the most recent 24 hours. The processing unit receives an external acknowledgment signal, which includes a fault start timestamp, an end timestamp, and a fault status label. Based on the timestamps, the processing unit extracts all multidimensional feature vectors for the corresponding time period from the circular buffer.

[0089] The processing unit reads the average operating current value within the specified time period and rounds it to the nearest 10% of the rated current step. The processing unit calculates the mean vector of the extracted vectors. The processing unit locates the storage location corresponding to the current step and the confirmed fault status label in the multi-condition feature template library. The processing unit appends the calculated mean vector to the template sample list at that location.

[0090] The processing unit executes a periodic optimization program every 168 hours. The processing unit iterates through the template sample list for each current step and each state label in the multi-condition feature template library. The processing unit calculates the arithmetic mean for each dimension of all sample vectors in the list to obtain an updated mean vector. The processing unit also calculates the sample variance for each dimension as the updated variance value.

[0091] The processing unit removes sample vectors from the list whose Euclidean distance from the updated mean vector exceeds three standard deviations. The processing unit replaces the original template mean vector with the updated mean vector and stores the updated variance value. The processing unit updates the pointers to the corresponding keys in the hash index table, pointing to the optimized template sample list. The processing unit saves the updated multi-condition feature template library in non-volatile memory.

[0092] Furthermore, the confidence calculation in decision fusion includes:

[0093] For clogging-related faults, the threshold logic confidence is calculated based on the ratio of the real-time pressure difference before and after the filter to the corresponding pressure threshold and the ratio of the circulating pump outlet flow rate to the corresponding flow rate threshold.

[0094] For filter element rupture faults, the threshold logic confidence is calculated based on the ratio of the real-time pressure difference across the filter to the fourth pressure threshold and the ratio of the circulating pump outlet flow rate to the fourth flow rate threshold.

[0095] For sensor faults, the threshold logic confidence level is calculated based on the degree of deviation between abnormal sensor data and normal coupling relationship;

[0096] The confidence level of intelligent matching is calculated based on the minimum weighted Euclidean distance between the multidimensional feature vector and the most similar state template.

[0097] Specifically, the processing unit initiates a confidence calculation procedure when the threshold logic judgment result is inconsistent with the intelligent matching judgment result. The processing unit first identifies the fault category to which the threshold logic judgment result belongs.

[0098] When the threshold logic determination result is mild blockage, the processing unit calculates the differential pressure confidence factor as the real-time differential pressure value divided by a first pressure threshold, and the flow rate confidence factor as the first flow rate threshold divided by the real-time flow rate value. The processing unit adds the differential pressure confidence factor and the flow rate confidence factor to obtain the threshold logic confidence level for mild blockage. When the threshold logic determination result is moderate blockage, the processing unit calculates the differential pressure confidence factor as the real-time differential pressure value divided by a second pressure threshold, and the flow rate confidence factor as the second flow rate threshold divided by the real-time flow rate value. The processing unit adds the differential pressure confidence factor and the flow rate confidence factor to obtain the threshold logic confidence level for moderate blockage.

[0099] When the threshold logic determination result is severe clogging, the processing unit calculates the differential pressure confidence factor as the real-time differential pressure value divided by the third pressure threshold, and the flow rate confidence factor as the third flow rate threshold divided by the real-time flow rate value. The processing unit adds the differential pressure confidence factor and the flow rate confidence factor to obtain the threshold logic confidence level for severe clogging. When the threshold logic determination result is filter element rupture, the processing unit calculates the differential pressure confidence factor as the fourth pressure threshold divided by the real-time differential pressure value, and the flow rate confidence factor as the real-time flow rate value divided by the fourth flow rate threshold. The processing unit adds the differential pressure confidence factor and the flow rate confidence factor to obtain the threshold logic confidence level for filter element rupture.

[0100] When the threshold logic determines a sensor fault, the processing unit loads the polynomial fitting coefficients corresponding to the current operating current value from the normal operation pattern database. Using the other two normal parameters as input, the processing unit calculates the predicted value of the abnormal parameter. The processing unit calculates the deviation as the absolute value of the difference between the real-time value and the predicted value of the abnormal parameter, divided by the predicted value. The processing unit uses this deviation as the threshold logic confidence level for the sensor fault.

[0101] The processing unit extracts the minimum weighted Euclidean distance from the intelligent matching discrimination results. The processing unit calculates the intelligent matching confidence score as the reciprocal of the minimum weighted Euclidean distance multiplied by a normalization coefficient. This coefficient is set to ensure the confidence score range aligns with the threshold logical confidence score. The processing unit compares the threshold logical confidence score with the intelligent matching confidence score. If the threshold logical confidence score is higher, the final fault diagnosis conclusion adopts the threshold logical discrimination result. If the intelligent matching confidence score is higher, the final fault diagnosis conclusion adopts the intelligent matching discrimination result. The processing unit records the selected discrimination result, the two confidence scores, and the current multidimensional feature vector to a divergence log file for subsequent template library optimization.

[0102] Furthermore, the specific process for retrieving operating condition groups whose deviation from the current operating current value is within a preset range is as follows:

[0103] Using the current operating current value as the index key, retrieve all operating condition groups whose operating current values ​​deviate from the current operating current value within ±5% from the multi-operating condition feature template library.

[0104] Specifically, the processing unit reads the real-time current value from the electrolysis system current sensor during each diagnostic cycle. The processing unit divides this current value by the system's rated current to obtain the normalized current percentage. The processing unit calculates a lower threshold as the normalized current percentage minus 0.05 and an upper threshold as the normalized current percentage plus 0.05. The processing unit loads a hash index table from the multi-condition feature template library, which uses discrete current percentages as primary keys, with each 10% of the rated current as an discrete step size. The processing unit iterates through all primary keys in the index table, filtering out entries whose primary key values ​​fall between the lower and upper thresholds.

[0105] For each selected primary key, the processing unit extracts six subordinate state templates, including normal state template, slightly blocked state template, moderately blocked state template, heavily blocked state template, filter element rupture state template, and sensor fault state template. Each template contains a mean vector and a corresponding variance vector. The processing unit organizes all selected state templates into a temporary operating condition group list, which records the source current percentage and state label for each template. The processing unit allocates contiguous storage space in memory, serializes the temporary operating condition group list, and passes it to the subsequent weighted Euclidean distance calculation module. If no operating condition group meets the deviation range of ±5%, the processing unit uses all state templates of the most recent single current step as candidate operating condition groups and records a deviation warning to the system log.

[0106] Furthermore, the calculation process for the weighted Euclidean distance is as follows:

[0107] After assigning weights to each dimension component of the multidimensional feature vector, the sum of the weighted squared differences between the current multidimensional feature vector and the corresponding dimension component of each state template is calculated, and then the square root is taken to obtain the weighted Euclidean distance; the preset time window is the most recent N sampling periods of continuous acquisition, where N is an integer greater than 1.

[0108] Specifically, after receiving the temporary operating condition group list, the processing unit assigns fixed weight coefficients to the nine dimensions of the multidimensional feature vector. Weights w1 to w2 correspond to the static components of differential pressure, flow rate, and power, and are set to 0.15; weights w4 to w6 correspond to the rate of change components and are set to 0.12; weights w7 to w9 correspond to the standard deviation components and are set to 0.10. The processing unit normalizes all weights so that the sum equals 1.

[0109] The processing unit extracts nine component values ​​from the multidimensional feature vector of the current diagnostic cycle. The processing unit performs iterative calculations for each state template in the temporary operating condition group list. First, the processing unit subtracts the template's differential pressure component from the current feature vector's differential pressure component, multiplies the difference by a weight w1, and then squares it. The processing unit repeats the same operation for the flow rate component, power component, differential pressure rate of change component, flow rate of change component, power rate of change component, differential pressure standard deviation component, flow rate standard deviation component, and power standard deviation component, obtaining nine weighted squared terms. The processing unit sums these nine weighted squared terms one by one to obtain a total. The processing unit takes the square root of the total to obtain the weighted Euclidean distance of the template. The processing unit associates the calculation result with the template's state label and source current percentage, storing it in a distance result array. After traversing all templates in the list, the processing unit selects the minimum distance value and its corresponding template's state label from the distance result array as the intelligent matching and discrimination result.

[0110] The preset time window N value is set to 30. The processing unit maintains a circular buffer with a capacity of 30 consecutive sampling periods of the original multidimensional feature vector data. In each new sampling period, the processing unit pushes the latest data to the end of the buffer and removes the oldest data. When calculating the standard deviation, the processing unit reads all 30 pressure difference samples from the buffer, calculates the variance, and takes the square root as the standard deviation of the pressure difference. The processing unit performs the same operation on the flow rate and power samples to obtain the standard deviations of the flow rate and power. The processing unit only enables standard deviation calculation when the buffer is full; when it is not full, it uses the available sample size for calculation and marks the window as not saturated.

[0111] Furthermore, the control strategies include:

[0112] A maintenance reminder will be issued when there is a minor blockage;

[0113] When moderate congestion occurs, reduce the system load and issue a cleaning command;

[0114] Emergency shutdown and alarm triggered in case of severe blockage or filter element rupture;

[0115] When a sensor fails, switch to a backup sensor and record the fault log.

[0116] Specifically, after outputting the final fault diagnosis conclusion, the processing unit loads the control strategy mapping table, which uses the fault status code as the key and the control command sequence as the value.

[0117] When the final fault diagnosis is a mild blockage status code, the processing unit sends a maintenance reminder message to the human-machine interface via the CAN bus. This message includes the fault code, current differential pressure value, flow rate value, and power value. Simultaneously, the processing unit illuminates a yellow indicator light on the panel, which remains on for 5 seconds and then turns off, repeating this cycle three times.

[0118] When the final fault diagnosis result is a moderate blockage status code, the processing unit sends a load reduction command to the electrolysis system power controller, reducing the current setpoint to 70% of the rated value. The processing unit sends a start signal to the cleaning pump driver via the RS485 interface, running the cleaning pump for 30 seconds before stopping. The processing unit records the load reduction timestamp and the cleaning operation log to flash memory.

[0119] When the final fault diagnosis result is a severe blockage status code or a filter element rupture status code, the processing unit sends an emergency shutdown command to the system main control unit, cutting off the DC power supply to the electrolytic cell. The processing unit activates the audible and visual alarm, emitting a 100dB buzzer and flashing a red indicator light until manually reset. The processing unit writes the shutdown reason, real-time parameters, and timestamp to the accident record area of ​​the non-volatile memory.

[0120] When the final fault diagnosis result is a sensor fault status code, the processing unit identifies the faulty sensor. If the differential pressure sensor is faulty, the processing unit switches the multiplexer to the backup differential pressure sensor channel. If the flow sensor is faulty, it switches to the backup flow meter channel. If the power sensor is faulty, it switches to the backup power acquisition module. The processing unit records the switching operation, the faulty sensor identifier, the switching timestamp, and the current reading of the backup sensor to a fault log file, which is stored in an SD card partition. The processing unit sends a sensor switching completion notification to the human-machine interface.

[0121] Furthermore, the method is executed by a processing unit installed in the alkaline water electrolysis hydrogen production system, which is connected to pressure sensors before and after the filter, a flow meter at the outlet of the circulating pump, and a power acquisition module for the circulating pump motor.

[0122] Specifically, the method is further executed by a processing unit installed in the alkaline water electrolysis hydrogen production system, which is connected to pressure sensors before and after the filter, a flow meter at the outlet of the circulating pump, and a power acquisition module for the circulating pump motor.

[0123] Specifically, the processing unit employs an industrial programmable logic controller (PLC), fixed within the electrolysis system control cabinet. In practice, signals from pressure sensors installed before and after the filter, the flow meter at the circulating pump outlet, and the power acquisition module at the pump motor end are all connected to the PLC's analog input module. The PLC is responsible for real-time, high-speed acquisition of these raw signals and performing preliminary signal conditioning, such as filtering and dimension conversion. The PLC uploads the processed real-time data packets (containing differential pressure, flow rate, power, and system current values) to a host industrial control computer running MATLAB software via industrial Ethernet and standard communication protocols such as TCP / IP. Within the MATLAB environment, the core fault diagnosis algorithm runs. This algorithm first performs in-depth preprocessing and feature extraction on the received data, calculating its rate of change and standard deviation to construct a multi-dimensional feature vector. Subsequently, the algorithm executes two tasks in parallel: first, it performs rapid discrimination based on preset threshold logic; second, it uses the current system current value as an index to query the local multi-condition feature template database for intelligent matching and diagnosis. MATLAB fuses the results from these two sources with confidence levels to arrive at a final diagnostic conclusion. It then sends the fault type, confidence level, and corresponding control commands (such as alarms, suggested load reduction, or shutdown) back to the PLC via the same communication link. Finally, the PLC, based on the received commands, drives audible and visual alarms, controls relays, or interacts with the higher-level distributed control system (DCS), thus completing a full closed-loop control from state perception to diagnostic decision-making and execution feedback.

[0124] Example 1 provides the basic process framework of the alkaline electrolysis system filter fault diagnosis method of the present invention. Example 2 below further describes the complete execution process of the method in detail with actual operation scenarios and parameter examples.

[0125] Example 2:

[0126] Reference Figure 1 During system operation, the differential pressure signal across the filter, the flow signal at the pump outlet, and the motor power signal are collected in real time. These data are then input into the processing unit for signal preprocessing, feature extraction, and fault identification.

[0127] 1. Feature vector extraction and basic threshold logic discrimination

[0128] In the specific implementation process, the processing unit first filters the raw sensor signal to eliminate high-frequency noise interference. Then, based on the filtered real-time data, the system calculates the rate of change (first derivative) and fluctuation amplitude (standard deviation) of pressure difference, flow rate, and power. Finally, a multi-dimensional feature vector is constructed. :

[0129] ;

[0130] in:

[0131] : represents the pressure difference between the pressure sensors before and after the filter at time t, in bars;

[0132] : where m is the flow rate reading of the flow meter at time t. 3 / h;

[0133] : represents the operating power of the alkali circulation pump motor at time t, in kW;

[0134] : represents the rate of change of the pressure difference before and after the filter at time t, reflecting how fast the pressure difference changes;

[0135] : represents the rate of change of electrolyte flow rate in the pipeline at time t, reflecting how fast the flow rate changes;

[0136] : represents the rate of change of the power of the alkali circulation pump motor at time t, reflecting how fast the power consumption changes;

[0137] : The standard deviation of the pressure difference within a certain time window, representing the fluctuation amplitude of the pressure difference signal;

[0138] : The standard deviation of the flow rate within a certain time window, representing the fluctuation amplitude of the flow rate signal;

[0139] : is the standard deviation of motor power within a certain time window, characterizing the fluctuation amplitude of the power signal.

[0140] During the fault diagnosis phase, the system first uses threshold logic for rapid judgment:

[0141] Filter cartridge (slight) clogging assessment: If the real-time pressure difference... First pressure threshold And traffic <First flow threshold> Meanwhile, the motor power P > the first power threshold. The blockage was determined to be mild.

[0142] Filter cartridge (moderate) clogging assessment: If the real-time pressure difference... Second pressure threshold and traffic Second flow threshold Meanwhile, the motor power P > the second power threshold. The blockage was determined to be moderate.

[0143] Filter cartridge (severe) clogging assessment: If the real-time pressure difference... Third pressure threshold and traffic Third flow threshold Meanwhile, the motor power P > the third power threshold The blockage was determined to be severe.

[0144] Filter element rupture detection: If the real-time pressure difference Fourth pressure threshold and traffic Fourth flow threshold Meanwhile, power fluctuation standard deviation First power fluctuation threshold It was determined to be a rupture;

[0145] Sensor failure detection: If the data from a single sensor is abnormal, and – – If the coupling relationship still conforms to the normal operating rules, then it is determined to be a sensor malfunction.

[0146] 2. Intelligent matching and diagnostic mechanism based on multi-condition feature template library

[0147] To further improve the system's discrimination accuracy and adaptability, this invention introduces an intelligent matching and diagnostic mechanism based on a multi-condition feature template library, building upon the basic threshold logic discrimination. The core of this mechanism lies in constructing and dynamically maintaining a comprehensive multi-dimensional feature index database. By matching current data with historical state templates in the database in real time, accurate identification of fault types can be achieved.

[0148] (1) Construction and storage of multi-condition feature template library

[0149] During long-term operation, the system continuously collects multi-dimensional feature vectors under stable operating conditions with different current loads (such as 10%, 20%, 30%, ... 100% of the rated current). Each feature vector group... All of these are associated with a specific system state (label), which includes:

[0150] Normal;

[0151] Clogging (which can be further subdivided into mild, moderate, and severe);

[0152] Rupture;

[0153] Sensor fault.

[0154] These precisely labeled feature vector sets are based on the current values ​​acquired at the time of data collection. The status labels are categorized and stored in a central database, forming a vast multi-condition, multi-state feature template library. Essentially, this template library is a high-performance index table, whose core structure can be conceptualized as follows:

[0155] (2) Intelligent matching process in real-time diagnosis

[0156] During real-time monitoring, the system performs the following steps for high-precision diagnosis:

[0157] Data acquisition and preprocessing: Real-time acquisition of current current value Pressure difference ,flow ,power And calculate the feature vector at the current time. .

[0158] Operating condition retrieval: based on current current value Using this as the primary index key, all operating condition groups with similar current values ​​(with an allowable deviation of ±5%) are retrieved from the feature template library.

[0159] Template matching and similarity calculation: Extract feature templates corresponding to various states (normal, blocked, ruptured, sensor malfunction) from the retrieved operating condition groups. Calculate the current feature vector. Multi-dimensional similarity with each state template. Similarity is calculated using weighted Euclidean distance:

[0160] ;

[0161] Among them, the smaller the weighted Euclidean distance, the higher the similarity.

[0162] Preliminary state classification: Select the category corresponding to the state template with the smallest distance from the current feature vector as the preliminary diagnostic result of the intelligent matching module.

[0163] 3. Decision fusion and confidence assessment

[0164] The system does not use the matching result alone, but rather fuses it with the result of threshold logic judgment to ensure reliability:

[0165] Consistent situation;

[0166] If both the threshold logic and intelligent matching determine the same fault state (e.g., both are determined to be "severe blockage"), the system will immediately trigger the control strategy corresponding to that state with high confidence.

[0167] Inconsistencies;

[0168] Inconsistency: If the results differ, the system will calculate the confidence level of the comprehensive threshold logic output. (Based on a margin of preset thresholds) and the confidence level output by the intelligent matching module. (Based on minimum weighted Euclidean distance), and the result with the highest overall confidence is selected as the final diagnostic conclusion. Simultaneously, the system records this divergence for subsequent template library optimization.

[0169] Confidence level of comprehensive threshold logic output ;

[0170] 1. For blockage-related faults (judged based on high differential pressure and low flow rate), the confidence level is calculated using the following formula:

[0171] ;

[0172] Pressure differential confidence factor. The larger the ratio, the more severe the pressure differential exceedance and the stronger the evidence.

[0173] Traffic confidence factor. The higher the ratio, the more severe the traffic insufficiency, and the stronger the evidence.

[0174] : Confidence level of blockage-type faults.

[0175] 2. Confidence level for rupture-related faults (judgment criteria: low differential pressure, high flow rate) The following formula is used for calculation:

[0176] ;

[0177] Pressure differential confidence factor. The larger the ratio, the more severe the pressure differential and the stronger the evidence.

[0178] Traffic confidence factor. The higher the ratio, the more severe the traffic exceedance and the stronger the evidence.

[0179] 3. Confidence level for sensor fault types (The problem may be a fault in one of the differential pressure, flow, or power sensors), and the following formula can be used for calculation:

[0180] ;

[0181] The degree of deviation from the expected value. The greater the deviation of the ratio from 1, the stronger the evidence.

[0182] The degree of deviation in flow rate. The greater the deviation of the ratio from 1, the stronger the evidence.

[0183] The degree of power deviation. The greater the deviation of the ratio from 1, the stronger the evidence.

[0184] The intelligent matching module outputs confidence level.

[0185] 4. The confidence score output by the intelligent matching module is calculated based on the least weighted Euclidean distance; the smaller the distance, the higher the confidence score.

[0186] ;

[0187] : is the current feature vector The weighted Euclidean distance between the template and the template most similar to the template in the feature template library. The smaller the value, the more similar the current state is to a certain historical state, and the higher the confidence level. The higher the value, the better.

[0188] Decision rules when there is inconsistency

[0189] if > The final result equals the result of the threshold logic.

[0190] if > The final result is the result of the intelligent matching.

[0191] 4. Dynamic self-learning and updating of the template library

[0192] The system has the ability to continuously evolve: whenever a fault is confirmed (such as through manual confirmation or verification by other reliable means), the corresponding feature vector data within that time period will be automatically labeled and added to the feature template library as a new template sample.

[0193] Template optimization: The system regularly maintains the template library, such as recalculating the characteristic mean and variance of a certain state under a certain working condition, so that the template always represents the latest operating characteristics of the system.

[0194] Enhanced adaptability: In this way, the system can continuously adapt to slow processes such as electrolyzer performance degradation and electrolyte characteristics changes, as well as learn new fault modes that it has never encountered before, and its diagnostic capabilities are continuously enhanced as the operating time increases.

[0195] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method of diagnosing a filter malfunction in an alkaline electrolysis system, characterized by, The method comprises the following steps: S1, collecting the differential pressure before and after the filter, the outlet flow of the circulating pump and the power of the circulating pump motor in real time; S2, filtering the collected differential pressure before and after the filter, the outlet flow of the circulating pump and the power of the circulating pump motor to eliminate high-frequency noise interference; S3, based on the filtered data in step S2, calculating the change rate of the differential pressure before and after the filter, the change rate of the outlet flow of the circulating pump, the change rate of the power of the circulating pump motor, and the standard deviation of the differential pressure before and after the filter within a preset time window, the standard deviation of the outlet flow of the circulating pump within a preset time window and the standard deviation of the power of the circulating pump motor within a preset time window, and constructing a multi-dimensional feature vector containing the above change rates and standard deviations; S4, using threshold logic to preliminarily judge the multi-dimensional feature vector to obtain a threshold logic judgment result; S5, taking the current operating current value as an index, searching for a working condition group with a deviation within a preset range from the current operating current value in a pre-constructed multi-working condition feature template library, calculating the weighted Euclidean distance between the multi-dimensional feature vector and each type of state template searched out, and determining an intelligent matching judgment result according to the minimum weighted Euclidean distance; S6, decision fusion is performed on the threshold logic judgment result and the intelligent matching judgment result, when the threshold logic judgment result and the intelligent matching judgment result are consistent, the fault state is directly output, when the threshold logic judgment result and the intelligent matching judgment result are inconsistent, the threshold logic confidence and the intelligent matching confidence are calculated respectively, and the result with higher confidence is selected as the final fault diagnosis conclusion; S7, outputting a corresponding control strategy according to the final fault diagnosis conclusion.

2. The filter fault diagnostic method for an alkaline electrolysis system according to claim 1, characterized by, The threshold logic judgment comprises: when the real-time differential pressure before and after the filter is greater than a first pressure threshold, the outlet flow of the circulating pump is less than a first flow threshold, and the power of the circulating pump motor is greater than a first power threshold, it is determined that there is mild blockage; when the real-time differential pressure before and after the filter is greater than a second pressure threshold, the outlet flow of the circulating pump is less than a second flow threshold, and the power of the circulating pump motor is greater than a second power threshold, it is determined that there is moderate blockage; when the real-time differential pressure before and after the filter is greater than a third pressure threshold, the outlet flow of the circulating pump is less than a third flow threshold, and the power of the circulating pump motor is greater than a third power threshold, it is determined that there is severe blockage; when the real-time differential pressure before and after the filter is less than a fourth pressure threshold, the outlet flow of the circulating pump is greater than a fourth flow threshold, and the standard deviation of the power of the circulating pump motor is greater than a first power fluctuation threshold, it is determined that the filter element is broken; when a single sensor data is abnormal and the coupling relationship of the remaining parameters meets the normal operation rule, it is determined that there is a sensor fault.

3. The filter fault diagnostic method for an alkaline electrolysis system according to claim 1, characterized by, The construction process of the multi-working condition feature template library comprises: in the range of 10% to 100% of the rated current, the multi-dimensional feature vectors in the states of normal state, mild blockage state, moderate blockage state, severe blockage state, filter element broken state and sensor fault state are collected; the collected multi-dimensional feature vectors are stored together with the corresponding operating current values and state labels, forming a multi-working condition feature template library covering the full operating current range.

4. The filter fault diagnostic method for an alkaline electrolysis system according to claim 1, characterized by, The dynamic updating process of the multi-working condition feature template library comprises: When the fault state is confirmed, the multi-dimensional feature vector in the fault period is automatically extracted and labeled with the state label, and added to the multi-condition feature template library as a new template sample; The feature mean and variance of the template sample of the same state under the same condition in the multi-condition feature template library are recalculated periodically for template optimization.

5. The method of claim 1, wherein the method further comprises: The confidence calculation in the decision fusion includes: For the blockage fault, the threshold logic confidence is calculated based on the ratio of the real-time differential pressure before and after the filter to the corresponding pressure threshold and the ratio of the circulating pump outlet flow to the corresponding flow threshold; For the filter rupture fault, the threshold logic confidence is calculated based on the ratio of the real-time differential pressure before and after the filter to the fourth pressure threshold and the ratio of the circulating pump outlet flow to the second flow threshold; For the sensor fault, the threshold logic confidence is calculated based on the deviation of the abnormal sensor data from the normal coupling relationship; The intelligent matching confidence is calculated based on the minimum weighted Euclidean distance between the multi-dimensional feature vector and the most similar state template.

6. The method of claim 1, wherein the method further comprises: The specific process of retrieving the working condition group with a deviation of the current running current value within a preset range is: Using the current running current value as the index key, all working condition groups with a deviation of the running current value within ±5% of the current running current value are retrieved in the multi-condition feature template library.

7. The method of claim 1, wherein the method further comprises: The calculation process of the weighted Euclidean distance is: After assigning weights to each dimension component of the multi-dimensional feature vector, the sum of the weighted squared differences between the current multi-dimensional feature vector and each dimension component of the state template is calculated, and the weighted Euclidean distance is obtained by taking the square root.

8. The method of claim 1, wherein the method further comprises: The preset time window is the most recent N sampling periods continuously collected, and N is an integer greater than 1.

9. The filter fault diagnostic method for an alkaline electrolysis system according to claim 2, characterized by, The control strategy includes: Issuing a maintenance reminder when the blockage is mild; Reducing the system load and issuing a cleaning instruction when the blockage is moderate; Triggering an emergency shutdown and alarming when the blockage is severe or the filter is ruptured; Switching to a backup sensor and recording fault logs when the sensor fails.

10. The method of claim 1, wherein, The method is performed by a processing unit installed in an alkaline electrolytic water hydrogen production system, which is connected with a filter before and after pressure sensor, a circulating pump outlet flow meter and a circulating pump motor power acquisition module.

Images