Hydrogen production electrolyzer sheet voltage and current fault detection method and system

Abstract

The application discloses a kind of hydrogen production electrolytic cell piece voltage current fault detection method and system, it is related to water electrolysis hydrogen production technical field, its technical scheme main point is: synchronous acquisition total current and each piece voltage formation time series data, identify total current dynamic rise and fall segment pairing constitutes hysteresis loop event, construct each piece voltage current hysteresis curve and extract hysteresis characteristic parameter, establish each piece multivariate Gaussian reference model and full stack consistency reference index in early health period, determine fault piece by Mahalanobis distance and full stack consistency index joint determination when on-line detection.The application does not need additional excitation source, utilizes power supply itself fluctuation to realize piece level fault on-line detection and classification, anti-working condition drift, reduce false alarm rate.

Description

Technical Field

[0001] This invention relates to the field of water electrolysis for hydrogen production technology, and more specifically, to a method and system for detecting voltage and current faults in hydrogen electrolyzer plates. Background Technology

[0002] Hydrogen electrolyzers typically consist of dozens to hundreds of individual cells connected in series. During operation, faults such as membrane electrode degradation, increased contact resistance, flow channel blockage, and internal bypass leakage current can directly cause abnormalities in the voltage and effective current of individual cells, potentially leading to hot spots, gas cross-contamination, or even equipment damage. Therefore, accurate detection of faults in individual cell voltage and current is crucial for ensuring the safe and stable operation of the electrolyzer.

[0003] In existing technologies, typical fault detection methods involve monitoring the voltage of each individual cell under constant current or steady-state conditions and triggering alarms by setting absolute or statistical range thresholds. However, with the large-scale promotion of hydrogen production from renewable energy sources, the power output of electrolyzers needs to frequently follow fluctuations in wind and solar power output, resulting in long-term non-steady-state voltage and current. The steady-state threshold method has significant drawbacks under such dynamic conditions: on the one hand, the overall voltage drift of all cells during changes in operating conditions may trigger false alarms; on the other hand, some early fault characteristics that only manifest during dynamic processes, such as transient anomalies caused by mass transfer limitations or micro-short circuits during current changes, show almost no change in steady-state voltage and cannot be detected in time. Therefore, existing steady-state threshold methods are insufficient to meet the fault detection requirements under dynamic conditions. In addition, some methods introduce electrochemical impedance spectroscopy or online harmonic injection to diagnose faults through impedance characteristics. However, these methods require dedicated AC excitation hardware, resulting in complex and costly systems. Furthermore, signal coupling is easily interfered with in strong electric environments, limiting their widespread application in practical engineering. Moreover, these methods typically require testing under specific operating conditions, making continuous online monitoring difficult.

[0004] Therefore, researching and designing a method and system for detecting voltage and current faults in hydrogen electrolyzers that can overcome the above-mentioned defects is an urgent problem to be solved. Summary of the Invention

[0005] To address the shortcomings of existing technologies, the present invention aims to provide a method and system for detecting voltage and current faults in hydrogen electrolyzers. This method fully utilizes the diagnostic information inherent in the power fluctuations that naturally occur during hydrogen production from renewable energy sources, eliminating the need for additional AC excitation signals or dedicated hardware. It overcomes the shortcomings of existing steady-state threshold methods, which suffer from false alarms and missed alarms due to overall voltage drift under dynamic operating conditions. It also avoids the problems of additional hardware, system complexity, and high cost associated with methods such as electrochemical impedance spectroscopy. This invention enables online detection of electrolyzer-level faults without additional disturbances and possesses the ability to resist overall operating condition drift.

[0006] The above-mentioned technical objective of the present invention is achieved through the following technical solution: In a first aspect, a method for detecting voltage and current faults in hydrogen electrolyzer plates is provided, comprising the following steps: The total current of the electrolytic cell and the voltage of each individual electrolytic cell are collected synchronously to form time-series data; Based on the timing data, the dynamic rise and fall segments of the total current are identified, and a rising segment and a falling segment that are adjacent in time are paired to form a hysteresis loop event. For each hysteresis loop event, a voltage-current hysteresis curve for each individual electrolytic cell is constructed, with the total current as the independent variable and the voltage of each individual electrolytic cell as the dependent variable. Hysteresis characteristic parameters of each single electrolytic cell are extracted from the hysteresis curve; In the early stage of healthy operation of the electrolyzer, a health benchmark model based on the hysteresis characteristic parameters is established for each single electrolyzer using multiple hysteresis loop events, and a benchmark index reflecting the dynamic consistency of single cells across the entire stack is established. During online detection, the hysteresis characteristic parameters of each single electrolytic cell in the newly captured hysteresis loop event are compared with their respective health benchmark models, and combined with the benchmark index, it is determined whether there is a faulty cell.

[0007] Furthermore, the hysteresis characteristic parameters include one or more of the following: hysteresis loop area, maximum voltage difference, and average voltage level; The hysteresis loop area is obtained by the closed-loop integration of voltage with respect to current. The maximum voltage difference is the maximum voltage difference between the rising trajectory and the falling trajectory under the same current value. The average voltage level is the average of the rising trajectory and the falling trajectory.

[0008] Furthermore, the health benchmark model is a multivariate Gaussian statistical model, which includes the mean vector and covariance matrix of the hysteresis feature parameters extracted by each of the single electrolytic cells in multiple hysteresis loop events.

[0009] Furthermore, the benchmark index is the coefficient of variation of the hysteresis loop area of ​​all the single electrolytic cells, or a consistency measure constructed based on other statistics of the hysteresis characteristic parameters of all the single electrolytic cells.

[0010] Furthermore, the determination of whether a faulty chip exists specifically involves: Calculate the Mahalanobis distance of the newly extracted hysteresis feature parameters of each single electrolytic cell relative to the health benchmark model, and mark the single electrolytic cell as having an abnormal feature when the Mahalanobis distance exceeds a predetermined threshold; Simultaneously, the overall dynamic consistency index of the current hysteresis loop event is calculated. If the overall dynamic consistency index exceeds the threshold set based on the benchmark index, the single electrolytic cell marked as having a characteristic anomaly is confirmed as a real faulty cell.

[0011] Furthermore, the method also includes: When the number of single electrolytic cells whose Mahalanobis distance exceeds the predetermined threshold is less than a preset threshold, and the overall dynamic consistency index of the stack does not exceed the set threshold, an early warning state is entered without immediately confirming it as a real fault.

[0012] Furthermore, after determining the existence of the faulty chip, the process also includes: Based on the deviation direction of the hysteresis characteristic parameters of the faulty chip, the faults can be distinguished as mass transfer obstruction faults, ohmic resistance increase faults, or internal micro-short circuit and bypass leakage current faults.

[0013] Furthermore, the method also includes: When the hysteresis characteristic parameters of all the single electrolytic cells show a synchronous shift in the same direction, and the magnitude of the shift exceeds the preset overall shift magnitude threshold, while the operating conditions of the electrolytic cell remain unchanged, it is determined that the total current sensor has drifted or failed.

[0014] Furthermore, the method also includes: When no natural fluctuation event that meets the preset natural fluctuation amplitude threshold is captured within a continuous time period, the hydrogen production power supply is controlled to superimpose a low-frequency current disturbance with an amplitude smaller than the preset disturbance amplitude threshold on the DC bus to actively stimulate a measurable hysteresis loop.

[0015] Secondly, a voltage and current fault detection system for hydrogen electrolyzer plates is provided, comprising: The data acquisition module is used to synchronously acquire the total current of the electrolytic cell and the voltage of each individual electrolytic cell to form time-series data; The event recognition module is used to identify dynamic rise and fall segments of the total current based on the timing data, and to pair a rising segment and a falling segment that are adjacent in time to form a hysteresis loop event. The curve construction module is used to construct a voltage-current hysteresis curve for each of the hysteresis loop events, with the total current as the independent variable and the voltage of each of the individual electrolytic cells as the dependent variable. The feature extraction module is used to extract the hysteresis feature parameters of each single electrolytic cell from the hysteresis curve; The benchmark establishment module is used to establish a health benchmark model for each single electrolytic cell based on the hysteresis characteristic parameters using multiple hysteresis loop events during the initial stage of healthy operation of the electrolytic cell, and to establish a benchmark index that reflects the dynamic consistency of single cells across the entire stack. The online detection module is used to compare the hysteresis characteristic parameters of each single electrolytic cell in a newly captured hysteresis loop event with their respective health benchmark models during online detection, and to determine whether there is a faulty cell by combining the benchmark index.

[0016] Compared with the prior art, the present invention has the following beneficial effects: 1. The present invention provides a method for detecting voltage and current faults in hydrogen electrolyzers. By synchronously collecting time-series data of the total current of the electrolyzer and the voltage of each individual electrolyzer, dynamic rise and fall segments of the total current are identified and paired to form hysteresis loop events. Based on this, voltage and current hysteresis curves of each individual electrolyzer are constructed and hysteresis characteristic parameters are extracted. Then, multiple hysteresis loop events in the early stage of healthy operation are used to establish a health benchmark model based on hysteresis characteristic parameters and a benchmark index reflecting the dynamic consistency of individual electrolyzers across the entire stack. Finally, during online detection, the hysteresis characteristic parameters of each individual electrolyzer in the newly captured hysteresis loop events are compared with their respective health benchmark models, and the presence of a faulty electrolyzer is determined by combining the benchmark index. This invention fully utilizes the diagnostic information inherent in the power fluctuations that naturally occur during the hydrogen production process from renewable energy sources. It eliminates the need for additional AC excitation signals or dedicated hardware, overcoming the shortcomings of existing steady-state threshold methods that suffer from false alarms and missed alarms due to overall voltage drift under dynamic operating conditions. It also avoids the problems of additional hardware, system complexity, and high cost associated with methods such as electrochemical impedance spectroscopy. This invention enables online detection of electrolyzer-level faults without additional disturbances and has the ability to resist overall operating condition drift.

[0017] 2. This invention limits the hysteresis characteristic parameters to one or more of the following: hysteresis loop area, maximum voltage difference, and average voltage level. The hysteresis loop area is obtained by the closed-loop integration of voltage versus current and characterizes dynamic irreversible loss. The maximum voltage difference is the maximum voltage difference between the rising and falling trajectories under the same current value. The average voltage level is the average of the rising and falling trajectories. These three characteristic parameters comprehensively characterize the electrochemical response of a single electrolytic cell in a dynamic process from different dimensions: the hysteresis loop area quantifies the magnitude of irreversible loss, the maximum voltage difference reflects the degree of voltage hysteresis, and the average voltage level reveals changes in the operating voltage baseline. This invention enables subsequent health benchmark modeling and fault detection based on multi-dimensional information, significantly improving the sensitivity of fault detection and the resolution of fault type differentiation compared to traditional methods that rely solely on a single voltage amplitude.

[0018] 3. This invention calculates the Mahalanobis distance of newly extracted hysteresis characteristic parameters of each individual electrolytic cell relative to the health baseline model. When the Mahalanobis distance exceeds a predetermined threshold, the cell is marked as having an abnormal characteristic. Simultaneously, the overall dynamic consistency index of the current hysteresis loop event is calculated. Only when the overall dynamic consistency index exceeds a threshold set based on the baseline index is the single electrolytic cell marked as having an abnormal characteristic confirmed as a true fault cell. If the number of single electrolytic cells with a Mahalanobis distance exceeding the predetermined threshold is less than a preset threshold and the overall dynamic consistency index does not exceed the set threshold, an early warning state is entered instead of immediate confirmation as a true fault. The Mahalanobis distance measures the degree to which each cell deviates from its own health baseline from an individual perspective, while the overall dynamic consistency index measures the dispersion between all cells from a global perspective. The combined use of both can effectively distinguish between true local faults and overall deviations caused by overall operating condition drift. When only a few cells exhibit abnormal characteristics while the overall consistency of the entire cell is normal, it may be due to sensor transient noise or very early weak degradation. In this case, an early warning state is entered instead of an immediate alarm, thus significantly reducing the false alarm rate while maintaining sensitivity.

[0019] 4. After confirming the actual faulty chip, this invention further distinguishes between mass transfer obstruction faults, increased ohmic resistance faults, or internal micro-short circuit and bypass leakage current faults based on the deviation direction of the faulty chip's hysteresis characteristic parameters. Specifically, when the hysteresis loop area significantly increases, the maximum voltage difference increases, and the average voltage level rises slightly, it is determined to be a mass transfer obstruction fault. When the average voltage level rises significantly while the hysteresis loop area remains basically unchanged, it is determined to be an increased ohmic resistance fault. When the hysteresis loop area significantly decreases and the average voltage level decreases synchronously, it is determined to be an internal micro-short circuit or bypass leakage current fault. Mass transfer obstruction leads to an increase in additional polarization during the dynamic process, thereby causing the hysteresis loop to expand. Increased ohmic resistance mainly affects the voltage baseline and has a smaller impact on dynamic hysteresis. Internal micro-short circuits cause the effective operating current to be lower than the total current measurement value, thereby causing the hysteresis loop to compress and shift downward. This classification method based on physical mechanisms does not require additional sensors or complex model identification. It can achieve accurate fault type identification using only the extracted hysteresis characteristic parameters, providing maintenance personnel with a clear direction for troubleshooting and shortening fault diagnosis time.

[0020] 5. This invention determines whether the total current sensor has drifted or failed by monitoring whether the hysteresis characteristic parameters of all individual electrolytic cells show synchronous shifts in the same direction and whether the shift amplitude exceeds a preset overall shift amplitude threshold, while also considering whether the operating conditions of the electrolytic cell have changed. This invention utilizes a unique manifestation of total current sensor failure: when the sensor experiences gain drift or zero-point shift, the characteristic vectors of all individual electrolytic cells exhibit a highly consistent, unidirectional overall shift. However, since the relative relationships between the cells remain unchanged, the overall dynamic consistency index remains at a normal level. This characteristic is the opposite of local faults, allowing this invention to clearly distinguish between sensor faults and electrolytic cell body faults without additional hardware, avoiding false alarms for all individual electrolytic cells due to sensor malfunctions, and further improving system reliability. Attached Figure Description

[0021] The accompanying drawings, which are included to provide a further understanding of embodiments of the invention and form part of this application, do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart from Embodiment 1 of the present invention; Figure 2 This is a system block diagram in Embodiment 2 of the present invention. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the embodiments and accompanying drawings. The illustrative embodiments and descriptions of this invention are only for explaining this invention and are not intended to limit this invention.

[0023] Example 1: A method for detecting voltage and current faults in hydrogen electrolyzer plates, such as... Figure 1 As shown, it includes the following steps: S1: Synchronously collect the total current of the electrolytic cell and the voltage of each individual electrolytic cell to form timing data; S2: Based on time series data, identify dynamic rise and fall segments of the total current, and pair a rising segment and a falling segment that are adjacent in time to form a hysteresis loop event; S3: For each hysteresis loop event, construct the voltage-current hysteresis curve for each individual electrolytic cell with the total current as the independent variable and the voltage of each individual electrolytic cell as the dependent variable; S4: Extract the hysteresis characteristic parameters of each single electrolytic cell from the hysteresis curve; S5: In the early stage of healthy operation of the electrolyzer, a health benchmark model based on hysteresis characteristic parameters is established for each single electrolyzer using multiple hysteresis loop events, and a benchmark index reflecting the dynamic consistency of single cells across the entire stack is established. S6: During online detection, the hysteresis characteristic parameters of each single electrolytic cell in the newly captured hysteresis loop event are compared with their respective health benchmark models, and combined with the benchmark index, it is determined whether there is a faulty cell.

[0024] In step S1, the total current of the electrolytic cell is synchronously acquired by a DC current sensor and a multi-channel chip voltage acquisition module at a sampling frequency of not less than 1Hz. and the voltage of each individual electrolytic cell ,in , This represents the total number of individual electrolytic plates connected in series in the electrolytic cell. For each single electrolytic cell, the serial number is [number]. This corresponds to a single, independent electrolytic cell. The acquired raw data is arranged in chronological order to form time-series data. and , Indicates the sampling time.

[0025] In this embodiment, the alkaline hydrogen production electrolyzer consists of 100 individual electrolytic plates connected in series. The rated DC current is 2000A, and the sampling frequency is set to 10Hz, meaning that 10 sets of total current and individual plate voltage data are collected per second. The multi-channel plate voltage acquisition module adopts an isolated voltage acquisition circuit with high common-mode rejection ratio. Synchronous sampling is achieved through multiplexing and a 16-bit analog-to-digital converter to ensure the consistency and alignment accuracy of the voltage acquisition time of each individual electrolytic plate with the total current acquisition time.

[0026] During the data acquisition process, readings from the temperature and pressure sensors are acquired simultaneously and recorded as temperature. and pressure It is stored together with the current and voltage timing sequence.

[0027] After obtaining the raw time-series data, filtering and outlier removal are performed. Median filtering is used with a window length of 5 sampling points to suppress spike noise introduced by electromagnetic interference and sensor jitter. Outlier removal is performed using the following criteria: for each sampling channel, the mean of the data is calculated within a sliding window. and standard deviation If the absolute value of the difference between the value at a certain sampling time and the mean of that window exceeds... If the value is not found, it is considered an outlier and replaced with the median value of the sliding window.

[0028] After filtering and outlier removal, the preprocessed time series data is obtained, denoted as... and .

[0029] In step S2, firstly, the rate of change of the total current with time is calculated in real time, that is, the derivative of the current with respect to time, denoted as . In practical discrete sampling systems, the central difference method is used for approximate calculation, and the expression is: ; in, Indicates the first Each sampling time, The sampling interval is specified in this embodiment. This corresponds to a sampling frequency of 10Hz. This differential formula uses the current values ​​from two consecutive sampling points to calculate the instantaneous rate of change at the current moment, effectively suppressing the influence of noise.

[0030] Next, a threshold for the rate of change of current is set. and a shortest duration When multiple consecutive sampling points occur within a certain time period All greater than And the duration of this period exceeds When this time period is reached, it is marked as a dynamic segment. The start point of the dynamic segment is the first sampling time that meets the condition, and the end point is the last sampling time that meets the condition. According to... The sign of the symbol divides the dynamic segment into an ascending segment and a descending segment: if If so, it is the rising segment, indicating that the total current is increasing; if If the current decreases, it indicates a decreasing segment, meaning the total current is decreasing.

[0031] In this embodiment, The value is 10A / s. The value is set to 5 seconds. That is, only when the absolute value of the current change rate continuously exceeds 10A / s and this state is maintained for at least 5 seconds is it considered a valid dynamic segment. This setting can filter out pseudo-dynamics caused by power supply ripple or small random fluctuations, and only retain power fluctuations with substantial diagnostic significance.

[0032] After completing the dynamic segment labeling, a pairing operation is performed: a temporally adjacent rising segment and its immediate following falling segment are considered as a group, or a falling segment and its immediate following rising segment are considered as a group. Each group constitutes a complete hysteresis loop event. Two pairing directions are allowed because actual wind and solar power output fluctuations include both peak-shaped fluctuations (rising first, then falling) and valley-shaped fluctuations (falling first, then rising), both of which can induce voltage and current hysteresis characteristics. Each hysteresis loop event corresponds to a continuous current rise and fall cycle, recording the total current value and the voltage value of each individual electrolytic cell at all sampling moments within that cycle.

[0033] After pairing, each hysteresis loop event is treated as an independent data unit, storing its corresponding current sequence and the voltage sequences of all individual electrolytic cells, and assigning a unique event number. During the 72-hour health operation phase of this embodiment, approximately 1200 valid hysteresis loop events were accumulated. These events covered different fluctuation amplitudes and rates of change, providing sufficient samples for establishing a reliable health benchmark model.

[0034] In step S3, firstly, for each sampling time in each hysteresis loop event, based on the synchronously acquired temperature... and pressure The measured voltage of each individual electrolytic cell Effective voltage reduced to a uniform reference temperature and reference pressure The reference temperature is denoted as... The reference pressure is denoted as In this embodiment, take That is, 60℃. The correction is based on electrochemical thermodynamics; for alkaline or proton exchange membrane electrolyzers, the change in reversible voltage with temperature and pressure can be approximately described by the Nernst equation. For ease of engineering implementation, a linearized correction formula is adopted: ; in, This is a temperature correction factor, measured in volts per Kelvin. In this embodiment, it is calibrated according to the characteristics of the electrolyzer. ; The pressure correction factor is calibrated as follows: This invention uniformly converts the measured voltages under different operating conditions to standard conditions, eliminating the overall voltage drift caused by fluctuations in operating conditions, so that the subsequent hysteresis curve can truly reflect the dynamic behavior of a single electrolytic cell.

[0035] After voltage correction is completed, for each single electrolytic cell in each hysteresis loop event... , with total current The horizontal axis is used to correct for voltage. Using the vertical axis as the ordinate, plot the voltage and current trajectories for the rising and falling segments of this event, respectively. The rising segment trajectory is formed by the sum of all rising sampling times during this event. The rising and falling segments are formed by connecting points in chronological order, while the falling segment is formed by connecting points at the falling sampling time. Together, these two segments constitute the voltage-current hysteresis curve of the single electrolytic cell under this hysteresis loop event. Due to the electrochemical hysteresis effect and mass transfer resistance inside the electrolytic cell, the rising and falling segments usually do not overlap, but instead form an approximately closed loop curve, i.e., a hysteresis loop.

[0036] In a typical hysteresis loop event in this embodiment, the total current rapidly decreases from 1600A to 800A and then rises back to 1600A. The correction voltages for the rising and falling segments at the same current value differ; for example, at a current of 1200A, the correction voltage for the rising segment is higher than that for the falling segment, thus... A hysteresis loop of a certain area is defined on a plane. The shape and size of this hysteresis loop reflect the degree of irreversible loss and response characteristics of a single electrolytic cell during this dynamic process.

[0037] Each hysteresis loop event generates a unique voltage-current hysteresis curve for each single electrolytic cell.

[0038] In step S4, firstly, the hysteresis loop area is calculated. The area of ​​a hysteresis loop is defined as the area between the rising and falling segments of the trajectory. The area enclosed by the plane reflects the energy consumed by the single electrolytic cell due to irreversible losses during a complete current rise and fall cycle, such as the hysteresis components in activation polarization, mass transfer polarization, and ohmic polarization. In discrete sampling systems, the hysteresis loop area is obtained through numerical integration, and the calculation formula is: ; In practical implementation, the sampling points of the rising and falling segments are sorted in ascending order of total current. The trapezoidal rule is used to integrate the trajectories of the rising and falling segments respectively, and then the difference between the two is taken. Let the rising segment trajectory be... It consists of 1 sampling point, arranged in ascending order of total current. ,in , For the first The total current value of each sampling point in the rising segment. For the first Electrolytic wafers in the first batch The corrected voltage for each rising segment sampling point. The falling segment trajectory is composed of... It consists of 1 sampling point, arranged in descending order of total current. ,in , For the first The total current value of each sampling point in the falling segment. For the first Electrolytic wafers in the first batch Corrected voltage for each falling segment sampling point.

[0039] First, integrate the rising segment trajectory along the current direction: ; in, No. The total current value of each rising segment sampling point; For the first Electrolytic wafers in the first batch Corrected voltage for each rising segment sampling point.

[0040] Similarly, the trajectory of the falling segment is integrated along the current direction. However, since the current direction of the falling segment is opposite to that of the rising segment, to ensure that the integration direction is consistent, the falling segment trajectory needs to be rearranged in ascending order of the total current before integration, or it can be calculated directly using the following formula: ; in, For the first Electrolytic wafers in the first batch Corrected voltage at each falling segment sampling point For the first The total current value of the sampling points in the falling segment.

[0041] Hysteresis loop area That is, the absolute value of the difference between the integral value of the rising segment and the integral value of the falling segment: ; In a typical case of mass transfer obstruction fault in this embodiment, the hysteresis loop area of ​​the 37th single electrolytic cell is... The voltage was 2.35 V·A, while the baseline area of ​​the chip during the healthy phase was 1.72 V·A, representing an area increase of approximately 36%, indicating that the chip incurred additional irreversible losses during the dynamic process.

[0042] Secondly, calculate the maximum voltage difference. The maximum voltage difference is defined as the maximum difference between the rising trajectory voltage and the falling trajectory voltage under the same total current value. Since the sampling points for the rising and falling segments are usually not at exactly the same current value, linear interpolation is needed to map the rising and falling segments onto a unified current grid. Let the current grid be the minimum value of the total current from this hysteresis loop event. To the maximum value Uniformly distributed There are 1 point, with a step size of 1. For each grid current The corresponding correction voltages are obtained by interpolation from the rising and falling segments, respectively. and The maximum voltage difference is: ; In this embodiment, the number of current grid points A value of 100 is sufficient to cover the current variation range and ensure interpolation accuracy. The maximum voltage difference reflects the maximum voltage hysteresis of the single electrolytic cell during dynamic processes; this value typically increases significantly for mass transfer obstruction faults. For example, the 37th cell mentioned above under fault conditions... The voltage is 0.12V, while the healthy baseline value is 0.06V, which is double the value.

[0043] Finally, calculate the average voltage level. The average voltage level is taken as the average of the rising and falling hysteresis loop trajectories, used to characterize the overall voltage level of the single electrolytic cell during this dynamic event. The specific calculation method is as follows: for each current point on the aforementioned unified current grid... Calculate the average of the rising trajectory voltage and the falling trajectory voltage, and then take the arithmetic mean of the average values ​​of all current points: ; The average voltage level reflects the baseline operating voltage of a single electrolytic cell, and this value changes significantly when there is an increase in ohmic resistance or an internal micro-short circuit. For example, in the case of an internal micro-short circuit fault, the 88th single electrolytic cell... A drop of 0.15V indicates that the effective operating current has been shunted by the bypass.

[0044] The above three feature parameters , and These are combined into a three-dimensional column vector, which is the feature vector of the single electrolytic cell. This feature vector comprehensively characterizes the dynamic behavior of a single electrolytic cell during a hysteresis loop event. Each hysteresis loop event generates a feature vector for each single electrolytic cell, and multiple events form a multidimensional feature sample set for that cell.

[0045] In step S5, the initial healthy operation refers to the stage where the electrolytic cell has completed commissioning, all performance indicators are stable, and no known faults have occurred. In this embodiment, the electrolytic cell operated continuously for 72 hours under fluctuating power supply, during which approximately 1200 effective hysteresis loop events were accumulated. These events covered different fluctuation amplitudes and rates of change, which can fully characterize the dynamic response diversity of the electrolytic cell in a healthy state. Each hysteresis loop event generates a feature vector for each single electrolytic cell; therefore, for the first... A single electrolytic wafer has K=1200 feature vector samples, denoted as... .

[0046] First, a multivariate Gaussian statistical baseline model is established for each single electrolytic cell. This model assumes that, under healthy conditions, the eigenvectors of a single electrolytic cell follow a multidimensional normal distribution, whose distribution characteristics are completely described by the mean vector and covariance matrix. For the ... Single electrolytic cell, mean vector The calculation formula is: ; in, It is a three-dimensional column vector whose three components are the mean of the hysteresis loop area, the mean of the maximum voltage difference, and the mean of the average voltage level, respectively. This mean vector represents the typical values ​​of the characteristic parameters of the chip in a healthy state.

[0047] The formula for calculating the covariance matrix is: ; in, It is a 3×3 symmetric matrix, where the diagonal elements are the variances of the three characteristic parameters, and the off-diagonal elements are the covariances among the characteristic parameters. This matrix characterizes the correlation between the characteristic parameters and their respective fluctuation ranges under healthy conditions. For example, the hysteresis loop area is usually positively correlated with the maximum voltage difference, and this correlation is captured by the covariance matrix. In this embodiment, each single electrolytic cell independently establishes its own mean vector. Covariance Matrix A total of 100 baseline models were established.

[0048] Secondly, a benchmark index reflecting the dynamic consistency of all individual electrolytic cells is established. This index measures the degree of consistency in the dynamic behavior of each individual electrolytic cell. In this embodiment, the coefficient of variation of the hysteresis loop area of ​​all individual electrolytic cells is selected as the benchmark index, denoted as [reference needed]. For the first For each hysteresis loop event, first calculate the hysteresis loop area of ​​all single electrolytic cells under that event. arithmetic mean and standard deviation Where i = 1, 2, ..., N, and N = 100. The formula for calculating the coefficient of variation is: ; in, For the first The standard deviation of the area of ​​all hysteresis loops under each event. For the first The coefficient of variation is the arithmetic mean of the hysteresis loop areas of all chips under a given event. The coefficient of variation is dimensionless, eliminating the influence of the absolute size of the area and simply reflecting the degree of dispersion between chips. Under healthy conditions, the coefficient of variation is usually small because the performance of each chip is similar; when a local fault occurs, the hysteresis loop area of ​​the faulty chip will deviate from the normal range, leading to an increase in the coefficient of variation.

[0049] Based on all K hysteresis events during the initial stage of healthy operation The values ​​are analyzed, their statistical distribution is established, and a baseline threshold is determined. In this embodiment, all are calculated. mean and standard deviation The baseline threshold is the mean plus three standard deviations: ; According to calculations, in this embodiment , , This threshold serves as the basis for determining whether there is significant inconsistency across the entire stack during subsequent online detection. Alternatively, other statistical measures can be used to construct a consistency metric, such as the ratio of the interquartile range of all hysteresis loop areas to the median, or inter-piece dispersion based on Mahalanobis distance. This embodiment uses the coefficient of variation as an example for illustration.

[0050] In step S6, the online detection process begins with the extraction of new event features. Whenever a new hysteresis loop event is captured from the real-time data stream, this event is identified and paired in step S2, and after processing in steps S3 and S4, the feature vectors of each individual electrolytic cell are obtained. Let the feature vector of the i-th individual electrolytic cell under this event be denoted as... The three components of this eigenvector are the areas of the hysteresis loops. Maximum voltage difference and average voltage level .

[0051] Next, Mahalanobis distance is calculated and feature anomalies are identified. For each single electrolytic cell i, its new feature vector is calculated. The Mahalanobis distance relative to the health baseline model of this film Mahalanobis distance is a distance metric that considers the correlation between characteristic parameters and can eliminate the influence of differences in the dimensions and variances of different features. Its calculation formula is: ; in, The deviation vector between the new feature vector and the mean vector is normalized in the ellipsoidal space defined by the covariance matrix to obtain a dimensionless distance value. A predetermined threshold is set. This threshold is determined based on the statistical distribution of data from the healthy period. In this embodiment, the 95th percentile of the Mahalanobis distance of all single electrolytic plates during the healthy period is taken as the predetermined threshold, and calculated... When the Mahal distance of a certain area Greater than When this happens, the piece is marked as a feature anomaly. For example, in online inspection, the Mahalanobis distance of the 37th single electrolytic piece is 4.8, which exceeds 2.8, so the 37th piece is marked as a feature anomaly.

[0052] Meanwhile, for the newly captured hysteresis loop events, the full-heap dynamic consistency index is calculated. The calculation method for this index is exactly the same as that for the health period benchmark index in step S5: first, calculate the hysteresis loop area of ​​all single electrolytic plates under the current event. arithmetic mean and standard deviation Then calculate the coefficient of variation: ; This index reflects the dispersion of the hysteresis loop area across all individual electrolytic cells in the current dynamic event. Under healthy conditions, the performance of each cell is similar. Typically small; when a partial fault occurs, the hysteresis loop area of ​​the faulty chip deviates from the normal range, leading to... Increase. Set a judgment threshold, which is based on the baseline threshold established in step S5. Multiply by a safety factor To obtain, that is ,in In this embodiment, the safety factor... Set the threshold to 1.2. Therefore, the threshold is determined. .

[0053] Then, a joint determination of the faulty piece is performed. Based on the combined Mahalanobis distance marking results and the overall reactor dynamic consistency index, there are three scenarios. The first scenario is if at least one single electrolytic cell is marked as having a characteristic anomaly, and the current overall reactor dynamic consistency index... Greater than the judgment threshold If a fault occurs, the system is determined to have significant inconsistencies, and all single-electrolyte cells marked as having anomalous characteristics are confirmed as truly faulty cells. This is because when a fault occurs, not only do the characteristics of the faulty cell deviate from the baseline, but the differences between the faulty cell and the normal cells also lead to a deterioration in overall stack consistency. For example, cell number 37 is marked as having anomalous characteristics, and The value is greater than 0.18, therefore the 37th piece is confirmed as a genuine faulty piece. The second scenario is if only a few individual electrolytic cells are marked as having characteristic anomalies, but the current overall stack dynamic consistency index is higher than 0.18. Not exceeding the judgment threshold If the event occurs, an early warning state is entered. At this point, it is not immediately confirmed as a genuine fault; instead, the event is recorded and the trend of that chip is continuously tracked. This situation may be caused by sensor transient noise, sudden changes in operating conditions, or very early, subtle degradation, requiring more events for confirmation. For example, if in a certain event only the 50th chip has a Mahalanobis distance of 3.1 greater than 2.8, and... If the value is less than 0.18, the system enters an early warning state, notifying maintenance personnel to monitor the subsequent performance of the 50th electrolytic cell. In the third scenario, if no single electrolytic cell is marked as having abnormal characteristics, then all single electrolytic cells are considered to be in a healthy state under the current event, and no alarms or early warnings are triggered.

[0054] After confirming the actual faulty chip, the fault type is further determined based on the deviation direction of the hysteresis characteristic parameters of the faulty chip. If the hysteresis loop area... Significantly increased, maximum voltage difference Increase, while average voltage level A slight increase indicates a mass transfer obstruction fault. For example, the 37th piece... The value is 2.35 V·A, which is much greater than the reference value of 1.72. The voltage is 0.12V, which is greater than the reference voltage of 0.06V. A voltage increase of 0.08V is consistent with mass transfer impairment characteristics. A shutdown inspection revealed that the flow channel was blocked by debris. If the average voltage level... Significantly increased, while the hysteresis loop area If the resistance remains essentially unchanged or changes very little, the fault is classified as an increase in ohmic resistance, such as membrane degradation or increased contact resistance. If the hysteresis loop area... Significantly reduced and average voltage level If the current decreases synchronously, it is determined to be an internal micro-short circuit or bypass leakage current fault. For example, the 88th chip... The value of 1.05 is less than the baseline of 1.85. A voltage drop of 0.15V is consistent with bypass leakage current characteristics, and disassembly revealed localized damage to the diaphragm. If this abnormal characteristic is accompanied by voltage fluctuations, it can be further identified as a structural fault such as loose electrode plates.

[0055] Finally, the above judgment results are output to the human-machine interface or the upper-level monitoring system. The output includes the event number, timestamp, the serial number of the single electrolytic cell marked as having an abnormal characteristic and its Mahalanobis distance, the current dynamic consistency index of the entire stack, the judgment conclusion (i.e., a list of actual faulty cells and their fault types), and a list of warning cells. For single electrolytic cells that have entered the warning state, the system will continuously track the Mahalanobis distance change trend of that cell in subsequent events. If multiple consecutive events show abnormal characteristics, the judgment will automatically escalate to an actual fault determination.

[0056] Example 2: A voltage and current fault detection system for hydrogen electrolyzer plates. This system is used to implement the voltage and current fault detection method for hydrogen electrolyzer plates described in Example 1, such as... Figure 2 As shown, it includes a data acquisition module, an event recognition module, a curve construction module, a feature extraction module, a benchmark establishment module, and an online detection module.

[0057] The system comprises the following modules: a data acquisition module for synchronously acquiring the total current of the electrolytic cell and the voltage of each individual electrolytic cell to form time-series data; an event identification module for identifying dynamic rise and fall segments of the total current based on the time-series data, pairing a temporally adjacent rise segment and a fall segment to form a hysteresis loop event; a curve construction module for constructing a voltage-current hysteresis curve for each individual electrolytic cell for each hysteresis loop event, with the total current as the independent variable and the voltage of each individual electrolytic cell as the dependent variable; a feature extraction module for extracting hysteresis characteristic parameters of each individual electrolytic cell from the hysteresis curves; a benchmark establishment module for establishing a health benchmark model for each individual electrolytic cell based on hysteresis characteristic parameters using multiple hysteresis loop events during the initial healthy operation of the electrolytic cell, and establishing a benchmark index reflecting the dynamic consistency of the entire stack of individual cells; and an online detection module for comparing the hysteresis characteristic parameters of each individual electrolytic cell in newly captured hysteresis loop events with their respective health benchmark models during online detection, and combining the benchmark index to determine whether a faulty cell exists.

[0058] Working Principle: This invention synchronously collects time-series data of the total current of the electrolyzer and the voltage of each individual electrolyzer plate to identify dynamic rise and fall segments of the total current and pair them to form hysteresis loop events. Based on this, it constructs the voltage-current hysteresis curve of each individual electrolyzer plate and extracts hysteresis characteristic parameters. Then, it uses multiple hysteresis loop events in the early stage of healthy operation to establish a health benchmark model based on hysteresis characteristic parameters and a benchmark index reflecting the dynamic consistency of individual plates across the entire stack. Finally, during online detection, the hysteresis characteristic parameters of each individual electrolyzer plate in newly captured hysteresis loop events are compared with their corresponding health benchmark models, and the benchmark index is used to determine whether a faulty plate exists. This invention fully utilizes the diagnostic information contained in the power fluctuations that naturally occur in the renewable energy hydrogen production process, without the need for additional AC excitation signals or dedicated hardware. It overcomes the shortcomings of existing steady-state threshold methods, which suffer from false alarms and missed alarms due to overall voltage drift under dynamic operating conditions. It also avoids the problems of additional hardware, system complexity, and high cost associated with methods such as electrochemical impedance spectroscopy. It achieves online detection of electrolyzer plate-level faults without additional disturbances and has the ability to resist overall operating condition drift.

[0059] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0060] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0061] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0062] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0063] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for detecting voltage and current faults in hydrogen electrolyzer plates, characterized in that, Includes the following steps: The total current of the electrolytic cell and the voltage of each individual electrolytic cell are collected synchronously to form time-series data; Based on the timing data, the dynamic rise and fall segments of the total current are identified, and a rising segment and a falling segment that are adjacent in time are paired to form a hysteresis loop event. For each hysteresis loop event, a voltage-current hysteresis curve for each individual electrolytic cell is constructed, with the total current as the independent variable and the voltage of each individual electrolytic cell as the dependent variable. Hysteresis characteristic parameters of each single electrolytic cell are extracted from the hysteresis curve; In the early stage of healthy operation of the electrolyzer, a health benchmark model based on the hysteresis characteristic parameters is established for each single electrolyzer using multiple hysteresis loop events, and a benchmark index reflecting the dynamic consistency of single cells across the entire stack is established. During online detection, the hysteresis characteristic parameters of each single electrolytic cell in the newly captured hysteresis loop event are compared with their respective health benchmark models, and combined with the benchmark index, it is determined whether there is a faulty cell.

2. The method for detecting voltage and current faults in a hydrogen electrolyzer according to claim 1, characterized in that, The hysteresis characteristic parameters include one or more of the following: hysteresis loop area, maximum voltage difference, and average voltage level; The hysteresis loop area is obtained by the closed-loop integration of voltage with respect to current. The maximum voltage difference is the maximum voltage difference between the rising trajectory and the falling trajectory under the same current value. The average voltage level is the average of the rising trajectory and the falling trajectory.

3. The method for detecting voltage and current faults in a hydrogen electrolyzer according to claim 1, characterized in that, The health benchmark model is a multivariate Gaussian statistical model, which includes the mean vector and covariance matrix of the hysteresis feature parameters extracted from each of the single electrolytic cells in multiple hysteresis loop events.

4. The method for detecting voltage and current faults in a hydrogen electrolyzer according to claim 1, characterized in that, The benchmark index is the coefficient of variation of the hysteresis loop area of ​​all the single electrolytic cells, or a consistency measure constructed based on other statistics of the hysteresis characteristic parameters of all the single electrolytic cells.

5. The method for detecting voltage and current faults in a hydrogen electrolyzer according to claim 1, characterized in that, The determination of whether a faulty chip exists specifically involves: Calculate the Mahalanobis distance of the newly extracted hysteresis feature parameters of each single electrolytic cell relative to the health benchmark model, and mark the single electrolytic cell as having an abnormal feature when the Mahalanobis distance exceeds a predetermined threshold; Simultaneously, the overall dynamic consistency index of the current hysteresis loop event is calculated. If the overall dynamic consistency index exceeds the threshold set based on the benchmark index, the single electrolytic cell marked as having a characteristic anomaly is confirmed as a real faulty cell.

6. The method for detecting voltage and current faults in a hydrogen electrolyzer according to claim 5, characterized in that, The method also includes: When the number of single electrolytic cells whose Mahalanobis distance exceeds the predetermined threshold is less than a preset threshold, and the overall dynamic consistency index of the stack does not exceed the set threshold, an early warning state is entered without immediately confirming it as a real fault.

7. The method for detecting voltage and current faults in a hydrogen electrolyzer according to claim 1, characterized in that, After determining the presence of the faulty chip, the process also includes: Based on the deviation direction of the hysteresis characteristic parameters of the faulty chip, the faults can be distinguished as mass transfer obstruction faults, ohmic resistance increase faults, or internal micro-short circuit and bypass leakage current faults.

8. The method for detecting voltage and current faults in a hydrogen electrolyzer according to claim 1, characterized in that, The method also includes: When the hysteresis characteristic parameters of all the single electrolytic cells show a synchronous shift in the same direction, and the magnitude of the shift exceeds the preset overall shift magnitude threshold, while the operating conditions of the electrolytic cell remain unchanged, it is determined that the total current sensor has drifted or failed.

9. The method for detecting voltage and current faults in a hydrogen electrolyzer according to claim 1, characterized in that, The method also includes: When no natural fluctuation event that meets the preset natural fluctuation amplitude threshold is captured within a continuous time period, the hydrogen production power supply is controlled to superimpose a low-frequency current disturbance with an amplitude smaller than the preset disturbance amplitude threshold on the DC bus to actively stimulate a measurable hysteresis loop.

10. A voltage and current fault detection system for a hydrogen electrolyzer, characterized in that, include: The data acquisition module is used to synchronously acquire the total current of the electrolytic cell and the voltage of each individual electrolytic cell to form time-series data; The event recognition module is used to identify dynamic rise and fall segments of the total current based on the timing data, and to pair a rising segment and a falling segment that are adjacent in time to form a hysteresis loop event. The curve construction module is used to construct a voltage-current hysteresis curve for each of the hysteresis loop events, with the total current as the independent variable and the voltage of each of the individual electrolytic cells as the dependent variable. The feature extraction module is used to extract the hysteresis feature parameters of each single electrolytic cell from the hysteresis curve; The benchmark establishment module is used to establish a health benchmark model for each single electrolytic cell based on the hysteresis characteristic parameters using multiple hysteresis loop events during the initial stage of healthy operation of the electrolytic cell, and to establish a benchmark index that reflects the dynamic consistency of single cells across the entire stack. The online detection module is used to compare the hysteresis characteristic parameters of each single electrolytic cell in a newly captured hysteresis loop event with their respective health benchmark models during online detection, and to determine whether there is a faulty cell by combining the benchmark index.

Images