A method and system for monitoring the status of ion-exchange membrane electrolyzers based on current harmonic stratification sensing.

Abstract

This invention belongs to the technical field of intelligent monitoring of chlor-alkali chemical equipment and health management of water electrolysis hydrogen production systems. Specifically, it relates to a method and system for monitoring the status of ion-exchange membrane electrolyzers based on current harmonic layered sensing. The method first acquires time-domain current signals and process parameters. Then, it performs signal preprocessing and frequency domain transformation on the acquired time-domain current signals, dividing them into at least two characteristic frequency bands. Harmonic characteristic parameters are extracted from each characteristic frequency band, and a dynamic harmonic characteristic health baseline is constructed. The characteristic deviation value between the real-time harmonic characteristic parameters and the health baseline is calculated. The characteristic deviation value is compared with a preset threshold to determine the fault type, achieving fault unit cell location and status assessment. Finally, early warning information and maintenance suggestions are output, and the dynamic harmonic characteristic health baseline is updated. This solution overcomes the shortcomings of existing technologies, such as delayed early warning, inaccurate location, and significant operating condition interference, achieving early and accurate early warning, rapid fault location, and quantitative performance assessment.

Description

Technical Field

[0001] This invention belongs to the field of intelligent monitoring of chlor-alkali chemical equipment and health management technology of water electrolysis hydrogen production system, specifically involving a method and system for monitoring the status of ion membrane electrolyzers based on current harmonic layer sensing. Background Technology

[0002] Ion-exchange membrane electrolyzers are core reaction equipment in chlor-alkali and hydrogen energy chemical industries, and their operating status directly determines current efficiency, energy consumption, and production safety. Currently, mainstream monitoring methods in industrial settings have several intractable drawbacks. First, fault early warning is severely delayed. Existing technologies for monitoring the operating status of ion-exchange membrane electrolyzers rely solely on steady-state parameters such as unit cell voltage, electrolyte temperature, inlet and outlet pressure, and gas purity. These methods can only identify severe, overt faults such as membrane perforation and electrode burnout, failing to detect latent degradation such as membrane aging, localized pinholes, flow channel blockage, and poor electrode contact, easily leading to unplanned shutdowns. Second, monitoring coverage has blind spots. A single electrolyzer contains dozens to hundreds of unit cells. Existing single-point total cell voltage monitoring methods cannot pinpoint locally faulty units, leading to reactive measures only after the fault has spread. Furthermore, harmonic resources are severely wasted. For a long time, the field has treated current harmonics generated by rectified power supply as power quality interference, focusing only on harmonic mitigation and suppression without deeply exploring the strong coupling relationship between harmonic amplitude, phase, energy distribution, and the electrochemical impedance, membrane state, and electrode interface reactions within the electrolyzer, thus missing out on highly efficient sensing carriers. In addition, there are shortcomings such as extremely poor adaptability to operating conditions and the inability to quantitatively assess the state.

[0003] Therefore, there is an urgent need for a monitoring method and system that can achieve early warning of latent faults, precise location of faulty units, adaptive operation, and quantifiable status. Summary of the Invention

[0004] To address the aforementioned shortcomings in the existing technology, this invention provides a method and system for monitoring the state of an ion-exchange membrane electrolyzer based on current harmonic layer sensing, thereby solving the problems mentioned in the background technology. (I) A method for monitoring the status of an ion-exchange membrane electrolyzer based on current harmonic stratification A method for monitoring the state of an ion-exchange membrane electrolyzer based on current harmonic stratification sensing includes the following steps: Step 1: Acquire time-domain current signals in the main power supply circuit and unit cell branch circuits of the ion-exchange membrane electrolyzer, and simultaneously acquire process parameters. The time-domain current signal is a raw electrical signal continuously acquired in the time dimension, reflecting the change in the current magnitude of the ion-exchange membrane electrolyzer's power supply circuit over time. It is the basic data for recording the current waveform, fluctuations, and transient characteristics from a time perspective. The process parameters include current density, cell temperature, and electrolyte concentration. Rogowski coil non-invasive acquisition is preferred, with a sampling rate of not less than 100kHz, a resolution of not less than 16 bits, and an acquisition duration covering at least three rectified ripple cycles to ensure signal integrity.

[0005] Step 2: Perform signal preprocessing on the acquired time-domain current signal, followed by frequency domain transformation, dividing it into at least two characteristic frequency bands, and extracting the harmonic characteristic parameters of each characteristic frequency band. The preprocessing includes wavelet soft thresholding denoising and trend term removal to eliminate on-site electromagnetic interference and baseline drift. The frequency domain transformation can use Hanning window Fast Fourier Transform (FFT). The division into at least two characteristic frequency bands specifically includes: a low-frequency characteristic band of 0-1kHz, a rectification characteristic band of 1-5kHz, and a high-frequency response band of 5-100kHz. The harmonic characteristic parameters include the harmonic amplitude, total harmonic distortion, phase shift, and energy percentage of each frequency band.

[0006] Step 3: Based on historical normal operating data of the ion membrane electrolyzer under healthy operating conditions, and combined with real-time process parameters as correction factors, construct a dynamic harmonic characteristic health baseline of each frequency band harmonic characteristic parameter under healthy operating conditions of the ion membrane electrolyzer and adapted to the current real-time process parameters, and calculate the characteristic deviation value between the real-time harmonic characteristic parameter and the health baseline. Furthermore, the rate of change of the characteristic deviation value needs to be calculated to characterize the fault development trend and assist in determining the severity level of the fault. The rate of change of the characteristic deviation value is determined by linearly fitting a sequence of historical characteristic deviation values ​​within a continuous time window, and using the slope obtained from the fitting as the magnitude of the change in the characteristic deviation value per unit time.

[0007] The harmonic characteristic parameters include the harmonic amplitude, total harmonic distortion, phase shift, and energy percentage for each frequency band.

[0008] The dynamic baseline is constructed as follows: Under healthy operating conditions of the electrolyzer, samples of current density, cell temperature, concentration parameters, and corresponding harmonic characteristic parameters are collected. A regression model is used for training to obtain the mapping from operating parameters to healthy harmonic characteristics. During online operation, real-time operating parameters are input into the model, and the healthy baseline value under the current operating conditions is output.

[0009] Step 4: Compare the characteristic deviation value with a preset threshold to determine the fault type, and locate the faulty unit cell and evaluate the electrolytic cell status by combining the harmonic characteristics of the unit cell branch circuit; the preset threshold includes a first threshold and a second threshold, and the graded early warning includes three levels: minor anomaly, moderate early warning, and severe alarm; the comparison rule between the characteristic deviation value and the threshold is: deviation ≤ first threshold, determined as minor anomaly; first threshold < deviation ≤ second threshold, determined as moderate early warning; deviation > second threshold, determined as severe alarm. The fault type determination rule is: If the increase in total harmonic distortion in the low-frequency characteristic band relative to the healthy baseline exceeds the first preset ratio, and the change in the second harmonic amplitude relative to the baseline exceeds the second preset ratio, it is determined to be aging or perforation of the ion membrane. If the total harmonic distortion in the high-frequency response band increases to more than twice the healthy baseline and the phase shift exceeds the preset range, it is determined to be an electrode failure or poor contact. The abnormal harmonic phase difference in the rectifier characteristic section indicates a rectifier system fault. If the harmonic energy dispersion in the low-frequency characteristic segment exceeds the upper limit of the healthy baseline, it is determined to be either a flow channel blockage or uneven electrolyte distribution.

[0010] The method for locating the faulty unit slot is as follows: the harmonic characteristics of the total slot current are correlated and compared with the harmonic characteristics of each branch circuit. By comparing the degree of harmonic anomaly of each branch circuit, the range is gradually narrowed down, and the faulty unit slot number is finally determined.

[0011] The specific implementation steps are as follows: The harmonic characteristic deviation of the total slot current is compared one by one with the harmonic characteristic deviation of each branch circuit to identify the abnormal branch whose deviation value is significantly higher than other branches. Within the abnormal branch, the harmonic characteristics of each unit slot are collected and the deviation is calculated to pinpoint the faulty unit slot with the largest deviation. Based on the preset correspondence between branches and unit slots, the specific number of the faulty unit slot is confirmed and output.

[0012] Step 5: Output early warning information and operation and maintenance suggestions, and iteratively update the dynamic harmonic characteristic health baseline based on real-time normal data.

[0013] Specifically, based on the comparison results of feature deviation values ​​and thresholds, three levels of early warning information—minor anomaly, moderate warning, and severe alarm—are output, and pushed simultaneously through on-site audio-visual displays, mobile devices, and the web. Combining the fault type and fault unit location results, targeted maintenance suggestions such as repair, replacement, troubleshooting, and shutdown are output. Real-time normal operating condition data is filtered and stored in a circular buffer. The baseline model is retrained according to preset sample size / time trigger conditions. After successful verification, the old model is replaced, achieving adaptive baseline updates.

[0014] (II) A status monitoring system for ion-exchange membrane electrolyzers based on current harmonic stratification An ion-exchange membrane electrolyzer status monitoring system based on current harmonic stratification sensing, comprising the following four modules: The signal acquisition module is used to synchronously acquire the time-domain current signal and process parameters of the electrolytic cell power supply circuit; The edge computing processing module is connected to the signal acquisition module and is used for signal denoising, frequency domain transformation, harmonic layer feature extraction, dynamic baseline construction, and fault matching and location. The cloud-based management module connects to the edge computing processing module and is used for data storage, model iteration, lifetime prediction, and historical tracing. The early warning output module is used to output tiered early warning commands and operation and maintenance suggestions.

[0015] Beneficial effects 1. This invention breaks through the limitations of traditional monitoring that relies solely on steady-state parameters. By using the layered sensing of high-frequency current harmonics, it can capture early signs of latent degradation such as ion membrane aging, local pinholes, electrode catalyst shedding, and flow channel blockage, thus enabling early warning of latent faults.

[0016] 2. This invention achieves accurate location of faulty unit slots by deploying acquisition points in branch circuits and establishing harmonic correlation mapping between the main slot, branch, and unit slots. This solves the technical problem that traditional main slot voltage monitoring cannot locate local faulty units. It also constructs a dynamic harmonic characteristic healthy baseline and adjusts the baseline value in real time using current density, slot temperature, and electrolyte concentration as correction factors, thereby reducing the interference of operating condition fluctuations on monitoring accuracy.

[0017] 3. This invention improves current efficiency, reduces power consumption per unit product, extends the lifespan of the membrane and electrodes, and significantly reduces maintenance costs.

[0018] 4. This invention transforms harmonics from "interference terms" into "sensory characteristic terms," ​​overcoming technical bias and providing a completely new technical path for intelligent monitoring of electrolytic cells, demonstrating outstanding inventiveness. Attached Figure Description

[0019] Figure 1 A flowchart of a method for monitoring the condition of an ion-exchange membrane electrolyzer based on current harmonic stratification. Figure 2 This is a block diagram of an ion-exchange membrane electrolyzer status monitoring system based on current harmonic stratification. Detailed Implementation To enable those skilled in the art to better understand the present invention, the technical solution of the present invention will be further described below with reference to embodiments.

[0020] 1. Training the baseline model (taking random forest as an example) In this embodiment, the dynamic harmonic characteristic health baseline is trained using a random forest regression algorithm. The function of this baseline model is to output the harmonic characteristic values ​​that the electrolyzer should have when it is in a healthy state under any given real-time operating parameters. The model training process is divided into two stages: offline training and online application, as detailed below: During the period when the ion-exchange membrane electrolyzer is confirmed to be in healthy operating condition, data should be continuously collected. The collection duration should cover the entire operating range of the electrolyzer, including: current density, cell temperature, and electrolyte concentration. Continuous collection for more than 72 hours is recommended, with a sampling rate of 100 kHz and a collection resolution of 16 bits. Record the following simultaneously during data collection: Input feature X=[I,T,C], where I is the current density (kA / m²), T is the cell temperature (°C), and C is the electrolyte concentration (wt%). Output label Y=[THD L A2, THD H ,Δϕ ,E disp [,…], specifically including: Total Harmonic Distortion (THD) in the low-frequency characteristic range (0–1 kHz) L (%), 2nd harmonic amplitude A2 (A), harmonic energy dispersion E disp (dimensionless); The amplitudes of the 5th, 7th, 11th, and 13th harmonics in the rectification characteristic segment (1–5 kHz) and their phase differences Δϕ (°) relative to the fundamental wave. Total harmonic distortion (THD) in the high-frequency response range (5–100 kHz) H (%), variance of harmonic amplitude fluctuation σ 2 A(A²).

[0021] Each sampling yields one sample (X) i ,Y i After collection, approximately 10,000 to 50,000 valid samples were obtained (depending on data compression and filtering strategies).

[0022] Data preprocessing and outlier removal: For each feature (such as THD) L Calculate the mean and standard deviation, and delete sample points that deviate from the mean by more than 3 times the standard deviation (3σ criterion) to eliminate outliers introduced by sensor transient failures or communication interference; Trend term removal: After wavelet soft thresholding denoising of the time-domain current signal, a polynomial fitting method is used to remove low-frequency trend terms to ensure the accuracy of harmonic analysis; Normalization: Since I, T, and C have different dimensions and numerical ranges, Z-score normalization is used to prevent certain features from dominating model training. Where μ x σ is the mean of this feature in the training set. x The standard deviation is given. The standardized coefficients need to be saved so that the same transformation can be applied to real-time inputs during online applications. Then, all samples are randomly divided into a training set (80%) and a validation set (20%), ensuring that both are independent and identically distributed.

[0023] This embodiment uses the random forest regression algorithm, training an independent random forest model for each output dimension. The core hyperparameters of the model are set as follows: The number of decision trees n_estimators is set to 100, the maximum depth max_depth is 10, the minimum number of samples required for node split min_samples_split is 5, the minimum number of samples for leaf nodes min_samples_leaf is 2, the number of features randomly selected for each split max_features is sqrt(3) rounded to 2, self-sampling is adopted, and the random seed random_state is 42.

[0024] For each output variable, perform the following operations: Bootstrap sampling: Randomly select N samples with replacement from the training set (assuming a total of N samples) to form the training subset D of the t-th tree. t (t=1,2,…,100). Since sampling with replacement, D t The sample may contain duplicates, and approximately one-third of the original samples will not appear in D. t (Outside the bag, which can be used for internal verification.)

[0025] Decision tree growth: training subset D for each tree t The node splitting process is performed recursively. At each node, two features (max_features=2) are randomly selected from all three input features. For each selected feature, all possible splitting thresholds are tried (usually the average of the two adjacent values ​​of that feature in the node samples). The feature and threshold that minimize the sum of the mean squared errors (MSEs) of the two child nodes after the split are chosen. The MSE of node m is defined as: in Let m be the sample set. For the output value of the i-th sample, The condition for splitting the mean of the output values ​​of samples within node m is: in, This represents the mean square error of the parent node. Let be the sample set of the left child node. The sample set for the right child node. Let be the mean square error of the left child node. Mean square error of the right child node.

[0026] Stopping conditions: When the node depth reaches max_depth=10, or the number of node samples is less than min_samples_split=5, or the decrease in MSE due to splitting is less than 10. −6 When the split stops, the current node becomes a leaf node. The predicted value of a leaf node is the mean of the output values ​​of all samples within that node.

[0027] Integrated output: For a new input sample X new The prediction value of the random forest is the arithmetic mean of the prediction values ​​of all decision trees: in This represents the final predicted value from the random forest model. The new input sample (standardized real-time operating condition parameter vector). This represents the t-th decision tree with respect to the input. The predicted output value.

[0028] The model performance was evaluated using a reserved 20% validation set, employing grid search combined with 5-fold cross-validation to optimize key hyperparameters. Example of the search range: n_estimators: 50, 100, 200; max_depth: 5, 10, 15, 20; min_samples_split: 2, 5, 10.

[0029] The parameter combination with the smallest RMSE on the validation set is selected as the final model. In this embodiment, the optimal parameters are the aforementioned set values, and the baseline model is obtained.

[0030] 2. A specific method for iteratively updating the dynamic harmonic characteristic health baseline based on real-time normal data. During operation, the system continuously collects real-time data and filters out data samples belonging to normal operating conditions based on characteristic deviation values ​​and fault determination results. The filtered samples are stored in a circular buffer pool with a fixed capacity, employing a first-in, first-out (FIFO) strategy to retain normal samples from the most recent period as well as the initial training set.

[0031] When the preset update trigger conditions are met, the system retrieves all samples from the buffer pool and retrains the random forest regression model using the same hyperparameters as the initial training. After training, the system uses data from the buffer pool that has not been used for training in the recent past as the validation set to compare the prediction errors of the new and old models on the validation set. If the error of the new model is not greater than the preset tolerance range of the error of the old model, the new model replaces the old model for subsequent dynamic baseline generation; otherwise, the old model is retained and the event is recorded.

[0032] Through the aforementioned periodic retraining and validation mechanism, the random forest baseline model can adaptively track the slow characteristic drift of the electrolyzer caused by long-term operation, and always maintain high-precision monitoring.

[0033] Example 1: The system and method embodiments of the present invention will be described in detail below with reference to the accompanying drawings. This embodiment uses a 30,000 t / a ion-exchange membrane caustic soda unit in a chlor-alkali plant as an application scenario. This embodiment is only used to explain the present invention and does not constitute a limitation on the scope of protection of the present invention.

[0034] I. System Configuration The system in this embodiment includes a signal acquisition module, an edge computing processing module, a cloud management and control module, and an early warning output module.

[0035] 1. Signal Acquisition Module The signal acquisition module includes a Rogowski coil current sensor, process parameter transmitters, and unit cell voltage acquisition units. One Rogowski coil current sensor (range ±50kA, accuracy 0.5%) is installed in the main DC power supply circuit of the electrolytic cell. The 120 unit cells are divided into 12 groups of 10 units each, with one Rogowski coil (range ±5kA, accuracy 0.5%) installed on each group's branch busbar. All Rogowski coil outputs are converted back to current signals by an integrator. The process parameter transmitters include a temperature transmitter (PT100, accuracy ±0.1℃) and an alkali concentration transmitter (online refractometer, accuracy ±0.1wt%) installed near each branch. Current density is calculated using the total current and the number of unit cells. The unit voltage acquisition unit uses a multiplexer (NI PXIe-2527) to connect to the voltage lead of each unit slot, with a range of 0-5V and an accuracy of ±1mV. Simultaneously, a high-frequency current transformer (frequency response 0.1-200kHz) is non-invasively connected to the voltage lead loop of each unit slot to sense the high-frequency harmonic current of the unit slot, enabling independent acquisition of the harmonic characteristics of each unit slot. For synchronous acquisition, all sensors are connected to the high-speed acquisition card of the edge computing processing module via shielded cables, with a sampling rate of 100kHz, 16-bit resolution, and GPS clock synchronization.

[0036] 2. Edge computing processing module The edge computing processing module includes an edge gateway, a high-speed acquisition card, and built-in harmonic analysis, baseline calibration, and fault diagnosis units. Hardware: The edge gateway is an Advantech MIC-7700 (Intel Core i7-9700TE, 16GB RAM, 256GB SSD), running Linux real-time operating system; the high-speed acquisition card is an ADLINK PCIe-9852 (4 channels, 200kHz, 16-bit). Harmonic Analysis Unit: Developed based on Python 3.9, integrating PyWavelets, NumPy, and SciPy. It implements wavelet soft thresholding denoising, trend term removal, Hanning window FFT, frequency band division, and harmonic feature parameter extraction. Baseline Calibration Unit: Loads a pre-trained random forest model (ONNX format), standardizes the input operating parameters, and predicts the healthy harmonic feature baseline. Fault Diagnosis Unit: Calculates the feature deviation value and deviation change rate, compares it with a preset threshold, matches fault judgment rules, performs unit slot location, and maintains a circulating buffer pool (capacity 10,000 samples).

[0037] 3. Cloud-based management module The cloud server (2-way Xeon Silver 4214, 64GB RAM, 4TB RAID10) runs Ubuntu 20.04LTS and deploys MySQL 8.0, Docker containers (model training engine), and Grafana dashboards. It communicates with the edge gateway via MQTT to achieve data storage, model iteration, lifetime prediction, and historical traceability.

[0038] 4. Early warning output module The on-site audible and visual alarm (three-color light + buzzer) is connected to the edge gateway, the 4G DTU module pushes messages to the mobile APP of the operation and maintenance personnel, and the web dashboard displays the trend curve and unit slot heat map in real time.

[0039] II. Method Implementation Process (a) Offline training phase (building a dynamic baseline) Step 1: Collect historical normal operating condition data Under healthy operating conditions of the electrolyzer, data was continuously collected for 72 hours during the first month after the new membrane was put into operation. The signal acquisition module recorded the current signals and process parameters of the main circuit and 12 branch circuits at a sampling rate of 100kHz. Every 10 seconds was a time window, and the harmonic analysis unit of the edge computing processing module processed the data within the window: wavelet soft thresholding noise reduction (db4, 4 layers), trend term removal, Hanning window FFT (65536 points), and divided the data into 0-1kHz low frequency band, 1-5kHz rectification band, and 5-100kHz high frequency band, extracting the harmonic amplitude, total harmonic distortion (THD), phase shift, energy percentage, and energy dispersion of each frequency band. A total of 25,920 samples were obtained, each containing input X = [current density I (kA / m²), cell temperature T (°C), concentration C (wt%)] and output Y = [low-frequency THD]. L (%), 2nd harmonic amplitude A2(A), high-frequency THD H (%), Phase difference of the 5th harmonic in the rectifier section Δφ5 (°), Low-frequency energy dispersion E disp ].

[0040] Step 2: Data Preprocessing After removing outliers (approximately 500) other than 3σ, the input features are Z-score standardized: μ is calculated. I =4.0, σ I =0.8; μ T =87, σ T =3;μ C =32, σ C =1.2. The remaining 25,420 samples were randomly divided into a training set (20,336 samples) and a validation set (5,084 samples).

[0041] Step 3: Train the dynamic baseline model The training process is as described above.

[0042] Step 4: Threshold setting First threshold = 20%, second threshold = 50%; in the fault judgment rules, the first preset ratio = 20%, the second preset ratio = 30%, the phase offset preset range = ±5°, and the upper limit of energy dispersion = 1.5 times the healthy baseline.

[0043] (II) Online Monitoring Phase Step S1: Real-time data acquisition After the system starts up, the signal acquisition module continuously acquires the current signals and process parameters of the main circuit and 12 branch circuits at a sampling rate of 100kHz. Every 10 seconds is a cycle, and the edge computing processing module reads the data within this window.

[0044] Step S2: Harmonic Layering Processing The harmonic analysis unit performs the following on the data in each window: wavelet soft threshold denoising, trend term removal, Hanning window FFT, frequency band division, extraction of real-time harmonic feature vector Y_real, and calculation of average process parameters within the window: current density I_avg = 4.2 kA / m²; cell temperature T_avg = 86.3 ℃; electrolyte concentration C_avg = 32.1 wt%.

[0045] Step S3: Dynamic baseline generation and deviation calculation I_avg, T_avg, and C_avg are Z-score normalized (using μ and σ saved during training), and μ is obtained from the input data preprocessing stage. I =4.0, σ I =0.8; μ T =87, σ T =3;μ C =32, σ C =1.2, according to get: ; ; The obtained (I′,T′,C′) are combined into a vector X′, which is then input into the random forest model in the baseline calibration unit. Based on the functional form of the healthy baseline: in, The number of decision trees in the random forest is 100. Let M represent the prediction function of the t-th decision tree. t c is the number of leaf nodes in the t-th tree. t,m R is the predicted value of leaf node m. t,m It is the feature space region corresponding to the leaf node.

[0046] The specific output format of the health baseline: Low-frequency THD, a characteristic parameter of the healthy baseline, was obtained: THD L,base =3.2%; Second harmonic amplitude: A 2,base =0.52A; High-frequency THD: THD H,base =1.1%; Phase difference of the 5th harmonic in the rectifier section: Δ ϕ5,base =2.3∘。 Low-frequency energy dispersion: E disp,base =0.45.

[0047] Then, as described above, after preprocessing the acquired signal, it undergoes frequency domain transformation and frequency band division, and then... Where A1 is the fundamental frequency (50 Hz) amplitude, A h Let h be the amplitude of the h-th harmonic. The summation range covers all harmonics in the low-frequency band (h=2,3,…h=2,3,… up to frequency ≤ 1 kHz). Calculate the real-time measured characteristic parameters to obtain the THD. L，real =6.9%; directly read the amplitude at 100 Hz from the spectrum to obtain A. 2，real =0.76 A; Calculated using the same formula, but only considering harmonics in the 5-100 kHz frequency band (excluding the fundamental frequency), the measured THD is... H，real =1.2%; the phase angle ϕ at 250 Hz (5th harmonic) is extracted from the spectrum. 5,real Using the fundamental phase as a reference, the relative phase difference is calculated as follows: Δ ϕ5，real =ϕ 5,real −5ϕ1 (ϕ1 is the fundamental phase), the measured Δ ϕ5，real =2.5∘; In the low-frequency band (0-1 kHz), harmonic energy is grouped by frequency, and the ratio of the standard deviation to the mean of the energy proportion of each group is calculated to characterize the uniformity of energy distribution. The measured E... disp，real =0.48 (dimensionless).

[0048] The fault diagnosis unit calculates the characteristic deviation value of each harmonic parameter based on: The THD calculated above L,base A 2,base THD H,base Δ ϕ5,base E disp,base Substitute THD L,real A 2,real THD H,real Δ ϕ5,real E disp,real Substitute THD was calculated L Deviation ≈ 115.6%; A2 deviation ≈ 46.2%; THD H Deviation ≈ 9.1%; Δϕ5 Deviation ≈ 8.7%; E disp With a deviation of approximately 6.7%, the maximum deviation was found to be 115.6% (corresponding to THD). L If the threshold exceeds 50%, a severe alarm will be triggered.

[0049] Simultaneously, maintain the deviation sequence for the past 7 days, calculate the deviation for the past 7 days using the method described above, and apply the characteristic deviation value as a function of time: Where t is time (hours), the fault time t=0 can be set, and the past time t<0; a is the intercept, and b is the slope (unit: % / hour), so b=4% / hour, therefore the deviation change rate is 4×24=96% / day.

[0050] Step S4: Fault Diagnosis and Location Tiered warning: The maximum deviation is compared with a threshold. If the maximum deviation is 115.6%, exceeding the second threshold by 50%, a severe alarm is triggered. Furthermore, the first and second preset ratios are pre-calibrated through laboratory accelerated aging tests. In this embodiment, the first preset ratio is 20%, and the second preset ratio is 30%. The low-frequency THD deviation of 115.6% > 20% and the A2 deviation of 46.2% > 30%, meeting the 'ion membrane aging or perforation' rule. Therefore, the fault type is determined to be ion membrane aging or local pinhole formation.

[0051] Unit slot positioning: Position the total slot THD L Deviation and THD of 12 branches L In this embodiment, the total low-frequency THD deviation of the entire slot is 115.6%, while the low-frequency THD deviation of branch 4 (corresponding to slots 31-40) is 210%. The deviations of other branches are all around 15%. Branch 4, with a deviation significantly higher than other branches, is considered an abnormal region, thus branch 4 is identified. Within branch 4, a high-frequency current transformer embedded in the slot voltage acquisition unit is used to obtain the low-frequency THD deviation of each slot. This high-frequency current transformer can sense the high-frequency harmonic current on each slot branch, thereby independently calculating the harmonic characteristic parameters of each slot using the method described above, and obtaining the THD of each slot from 31 to 40. L The deviations are as follows: No. 37 is 260%, No. 38 is 280%, and the rest are <50%. Therefore, the faulty unit slot is No. 37-38 in the 4th branch.

[0052] Electrolyzer condition assessment: based on THD L The empirical relationship between deviation and film attenuation coefficient is based on Attenuation coefficient = exp(−k × deviation) Where k is the attenuation coefficient, obtained through accelerated aging tests in the laboratory. Different membrane models need to be calibrated individually, and the k value varies for different models and materials of ion exchange membranes. In this example, k = 0.0025 is taken, and the membrane attenuation coefficient is calculated to be 0.68.

[0053] Finally, based on statistical experience from similar historical failures, a mapping table of "current low-frequency THD deviation → remaining life" was established by statistically analyzing the historical operating data of multiple electrolyzers, resulting in a remaining life prediction of approximately 15 days.

[0054] Step S5: Early Warning Output and Model Iteration Warning output: The warning output module immediately issues a severe alarm: the red light on site flashes and the buzzer sounds continuously; the mobile APP pushes a message: "[Severe alarm] Ion membrane aging / pinhole formation, fault unit: No. 37-38, planned maintenance recommended"; the corresponding unit's heat map on the Web dashboard turns red.

[0055] Model Iteration: Every day at midnight, samples marked as "normal" in the past 24 hours are added to a circulating buffer pool (capacity 10,000, first-in, first-out). When the number of new samples added to the buffer pool is ≥2,000 or the last retraining is ≥7 days, the cloud management module retrieves all samples from the buffer pool and retrains the random forest model (new model). Validation is performed using normal data from the last 24 hours. If the new model's RMSE ≤ the old model's RMSE × 1.05, the new model is deployed to the edge gateway, atomically replacing the old model. III. Implementation Results The system and method of this embodiment have been running continuously in the industrial field for 6 months, achieving a latent fault warning advance of ≥30 days; improved unit-level fault location accuracy; reduced false alarm rate; improved current efficiency; reduced power consumption per ton of alkali; reduced unplanned downtime; and reduced operation and maintenance costs.

[0056] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. 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 for monitoring the state of an ion-exchange membrane electrolyzer based on current harmonic stratification sensing, characterized in that, Includes the following steps: S1: Collect time-domain current signals in the main power supply circuit and unit cell branch circuit of the ion membrane electrolyzer, and simultaneously collect process parameters; S2: Perform signal preprocessing on the acquired time-domain current signal, then perform frequency domain transformation, divide it into at least two characteristic frequency bands according to frequency, and extract the harmonic characteristic parameters of each characteristic frequency band respectively; S3: Based on historical normal operating condition data, combined with real-time process parameters as correction factors, construct a dynamic harmonic characteristic health baseline, and calculate the characteristic deviation value between the real-time harmonic characteristic parameters and the health baseline. S4: Compare the characteristic deviation value with a preset threshold to determine the fault type, and locate the faulty unit cell and evaluate the electrolytic cell status by combining the harmonic characteristics of the unit cell branch circuit. S5: Output early warning information and operation and maintenance suggestions, and iteratively update the dynamic harmonic characteristic health baseline based on real-time normal data.

2. The method for monitoring the state of an ion-exchange membrane electrolyzer based on current harmonic stratification sensing according to claim 1, characterized in that, The time-domain current signal is acquired using a Rogowski coil non-invasive method with a sampling rate of not less than 100kHz and an acquisition resolution of not less than 16 bits; the process parameters include current density, tank temperature, and electrolyte concentration.

3. The method for monitoring the state of an ion-exchange membrane electrolyzer based on current harmonic stratification sensing according to claim 1, characterized in that, The preprocessing described in S2 includes wavelet soft thresholding denoising and trend term removal of the time-domain current signal; the division into at least two characteristic frequency bands by frequency specifically includes: a low-frequency characteristic band of 0-1kHz, a rectification characteristic band of 1-5kHz, and a high-frequency response band of 5-100kHz.

4. The method for monitoring the state of an ion-exchange membrane electrolyzer based on current harmonic stratification sensing according to claim 3, characterized in that, The harmonic characteristic parameters include the harmonic amplitude, total harmonic distortion, phase shift, and energy percentage for each frequency band.

5. The method according to claim 1, characterized in that, S3 also calculates the rate of change of the characteristic deviation value, which is used to characterize the fault development trend and assist in determining the severity level of the fault.

6. The method for monitoring the state of an ion-exchange membrane electrolyzer based on current harmonic stratification sensing according to claim 1, characterized in that, The preset thresholds in S4 include a first threshold and a second threshold. The graded warning includes three levels: minor anomaly, moderate warning, and severe alarm. The comparison rule between the feature deviation value and the threshold is as follows: if the deviation is less than or equal to the first threshold, it is determined to be a minor anomaly; if the first threshold is less than or equal to the deviation is less than or equal to the second threshold, it is determined to be a moderate warning; and if the deviation is greater than or equal to the second threshold, it is determined to be a severe alarm.

7. The method for monitoring the state of an ion-exchange membrane electrolyzer based on current harmonic stratification sensing according to claim 3, characterized in that, The fault type determination rules are as follows: if the total harmonic distortion of the low-frequency characteristic segment increases by more than a first preset ratio relative to the healthy baseline, and the change in the amplitude of the second harmonic relative to the baseline exceeds a second preset ratio, it is determined to be aging or perforation of the ion membrane; if the total harmonic distortion of the high-frequency response segment increases by more than twice the healthy baseline, and the phase shift exceeds a preset range, it is determined to be electrode failure or poor contact; if the harmonic phase difference of the rectification characteristic segment is abnormal, it is determined to be a rectification system fault. If the harmonic energy dispersion in the low-frequency characteristic segment exceeds the upper limit of the healthy baseline, it is determined to be either a flow channel blockage or uneven electrolyte distribution.

8. The method for monitoring the state of an ion-exchange membrane electrolyzer based on current harmonic stratification sensing according to claim 1, characterized in that, The method for locating the faulty unit slot in S4 is as follows: the harmonic characteristics of the total slot current are correlated and compared with the harmonic characteristics of each branch circuit to determine the faulty unit slot number.

9. A state monitoring system for an ion-exchange membrane electrolyzer based on current harmonic stratification sensing, characterized in that, include: The signal acquisition module is used to synchronously acquire the time-domain current signal and process parameters of the electrolytic cell power supply circuit; The edge computing processing module, connected to the signal acquisition module, is used for signal denoising, frequency domain transformation, harmonic layering feature extraction, dynamic baseline construction, and fault matching and location; the cloud management module, connected to the edge computing processing module, is used for data storage, model iteration, lifetime prediction, and historical tracing; and the early warning output module is used to output tiered early warning commands and operation and maintenance suggestions.

Images