Air-cooled electric pile fault early warning management and control method based on prediction model

Abstract

The application discloses a kind of based on prediction model air-cooled electric pile fault early warning management and control method, it is related to fuel cell technical field, including the following steps: according to the response data of electric pile reconstruction, obtain full-band impedance spectrum, separate out ohm polarization impedance and charge transfer impedance from full-band impedance spectrum, based on ohm polarization impedance and charge transfer impedance, carry out fault dominant type determination, obtain current fault dominant type label, reversible recession health index and irreversible recession health index based on reversible recession health index and irreversible recession health index establish double-channel predictor, as prediction model, carry out prediction on different time scales by double-channel predictor, obtain recoverable window period and life end prediction value, based on fault dominant type label, recoverable window period and life end prediction value, carry out fault response mode determination, then, execute corresponding differential management and control instruction.

Description

Technical Field

[0001] This invention relates to the field of fuel cell technology, and in particular to a method for early warning and control of faults in air-cooled fuel cell stacks based on predictive models. Background Technology

[0002] In actual operation of fuel cell stacks, membrane drying caused by the migration of moisture within the membrane and active area decay caused by catalyst aging often coexist and are coupled with each other. Both of these degradation phenomena manifest as a decrease in output performance. Among them, membrane drying has reversible recovery characteristics, while active area decay constitutes a permanent capacity loss. Accurately identifying the two types of degradation processes and implementing differentiated early warning is the core prerequisite for improving fuel cell stack durability and operation and maintenance economy.

[0003] Existing fault diagnosis methods typically extract impedance amplitude in a single frequency band or rely on fixed thresholds for alarms, failing to simultaneously acquire independent evolution information of ohmic polarization and charge transfer impedance. Consequently, it is difficult to distinguish between reversible moisture imbalance and irreversible catalyst aging at the mechanistic level. This can easily lead to misjudging short-term membrane dryness as a serious fault and forcing a reduction in load, or to a delayed response to the actual degradation of the catalyst layer. More importantly, existing prediction models do not construct differentiated recursive frameworks for the different time constants of the two types of degradation mechanisms. The lack of coordinated estimation capability between the short-term moisture recovery window and the long-term lifetime endpoint results in a lack of refined classification basis for control strategies, which can easily lead to over-intervention and reduced availability, or accelerated irreversible damage propagation due to response lag, directly affecting the actual service life of the fuel cell stack. Summary of the Invention

[0004] The technical problem solved by this invention is that existing technologies cannot separate reversible and irreversible decay processes online, and the lack of multi-timescale collaborative prediction capabilities leads to mismatch in fault response strategies.

[0005] To address the aforementioned technical problems, this invention provides the following technical solution: a fault early warning and control method for air-cooled fuel cell stacks based on predictive models, comprising the following steps: Step S1: Reconstruct the impedance spectrum based on the response data of the fuel cell stack to obtain the full-band impedance spectrum, and separate the ohmic polarization impedance and charge transfer impedance from the full-band impedance spectrum. Step S2: Based on ohmic polarization impedance and charge transfer impedance, determine the fault dominance type and obtain the current fault dominance type label, reversible degradation health index and irreversible degradation health index. Step S3: Establish a dual-channel predictor based on the reversible and irreversible health indices as a prediction model. Perform predictions at different time scales through the dual-channel predictor to obtain the predicted values ​​of the recoverable window period and the end of life. Step S4: Based on the fault dominant type label, recoverable window period, and end-of-life prediction value, determine the fault response mode and execute the corresponding differentiated control instructions.

[0006] As a preferred embodiment of the predictive model-based air-cooled fuel cell stack fault early warning and control method described in this invention, step S1 specifically includes: Step S101: A multi-frequency small current disturbance signal is superimposed onto the DC bus of the fuel cell stack through a DC / DC converter; The data acquisition device collects the fuel cell stack response data, and the discrete voltage data points and discrete current data points in the fuel cell stack response data are arranged in the order of acquisition time to form the original voltage response time domain sequence and the original current response time domain sequence. By using a sliding time window to extract the original voltage response time-domain sequence and the original current response time-domain sequence, the voltage response time-domain sequence and the current response time-domain sequence are obtained.

[0007] As a preferred embodiment of the predictive model-based air-cooled fuel cell stack fault early warning and control method described in this invention, step S1 further includes: Step S102: Perform DC removal processing on the voltage response time domain sequence and the current response time domain sequence respectively to obtain the DC removed voltage response time domain sequence and the DC removed current response time domain sequence. Perform Fast Fourier Transform on the voltage response time-domain sequence and the current response time-domain sequence after DC removal, respectively, to obtain the voltage frequency domain sequence and the current frequency domain sequence. The ratio of the voltage frequency domain sequence to the current frequency domain sequence is calculated to obtain the first full-band impedance spectrum; Invalid frequency points are removed from the first full-band impedance spectrum to obtain the second full-band impedance spectrum.

[0008] As a preferred embodiment of the predictive model-based air-cooled fuel cell stack fault early warning and control method described in this invention, step S1 further includes: Step S103: Input the second full-band impedance spectrum into the relaxation time distribution deconvolution algorithm for processing to obtain two characteristic peaks; In the two characteristic peaks, the characteristic peak located in the first frequency band is assigned to the ohmic polarization impedance, and the characteristic peak located in the second frequency band is assigned to the charge transfer impedance, thus obtaining the real-time sequence of the ohmic polarization impedance and the real-time sequence of the charge transfer impedance.

[0009] As a preferred embodiment of the predictive model-based air-cooled fuel cell stack fault early warning and control method described in this invention, step S2 specifically includes: Step S201: Input the real-time sequence of ohmic polarization impedance and the real-time sequence of charge transfer impedance into the Randles-type equivalent circuit model for processing to obtain the real-time absolute amplitude of impedance under the current operating condition. Step S202: By using a sliding time window, calculate the rate of change of ohmic polarization impedance and charge transfer impedance relative to time, and obtain the rate of change of ohmic polarization impedance and the rate of change of charge transfer impedance.

[0010] As a preferred embodiment of the predictive model-based air-cooled fuel cell stack fault early warning and control method described in this invention, step S2 further includes: Step S203: Determine the fault-dominant type based on the rate of change of ohmic polarization impedance and the rate of change of charge transfer impedance, and obtain the current fault-dominant type label, reversible degradation health index, and irreversible degradation health index. When the real-time absolute impedance amplitude is greater than the preset impedance abnormality threshold, the fault dominant type determination is performed. When the real-time absolute impedance amplitude is less than or equal to the preset impedance abnormality threshold, the fault dominant type determination is not performed, and the fault dominant type label of the previous output remains unchanged. The determination of the dominant fault type specifically includes: When the rate of change of ohmic polarization impedance is greater than the ohmic polarization slope threshold and the rate of change of charge transfer impedance is less than the activation polarization slope threshold, the current fault dominant type is determined to be reversible decay, the current fault dominant type is labeled as reversible, and the reversible decay health index is calculated according to the first preset mapping relationship based on the amount by which the rate of change of ohmic polarization impedance exceeds the ohmic polarization slope threshold. The irreversible decay health index is assigned the value of 1 minus the reversible decay health index. When the rate of change of charge transfer impedance is continuously greater than the activation polarization slope threshold and the rate of change of ohmic polarization impedance is less than or equal to the ohmic polarization slope threshold, the current fault dominant type is determined to be irreversible decay. The current fault dominant type is labeled as irreversible. Based on the extent to which the rate of change of charge transfer impedance exceeds the activation polarization slope threshold, the irreversible decay health index is calculated according to the second preset mapping relationship. The reversible decay health index is assigned the value of 1 minus the irreversible decay health index. Otherwise, keep the fault-dominant type label from the previous output unchanged.

[0011] As a preferred embodiment of the predictive model-based air-cooled fuel cell stack fault early warning and control method described in this invention, step S3 specifically includes: Step S301: Obtain the current current density and the current stack temperature; Based on the reversible decay health index, the irreversible decay health index, the current current density, and the current stack temperature, an extended Kalman filter is used to establish a dual-channel predictor. The dual-channel predictor includes a short-timescale channel and a long-timescale channel; Step S302: Through a short timescale channel, perform an extended Kalman filter recursive estimation once in each impedance update cycle to obtain the predicted trajectory of the reversible declining health index. Based on the predicted trajectory, the remaining time for the reversible decline health index to reach the preset membrane dry safety health index threshold is calculated, thus obtaining the recoverable window period.

[0012] As a preferred embodiment of the predictive model-based air-cooled fuel cell stack fault early warning and control method described in this invention, step S3 further includes: Step S303: Extrapolate and predict the irreversible decline health index through a long-term timescale channel to obtain the predicted trajectory of the irreversible decline health index. Based on the predicted trajectory, the remaining time for the irreversible decline health index to reach the preset end-of-life health index threshold is calculated, and the end-of-life predicted value is obtained. Obtain the fluctuation noise estimate of the short-timescale channel, and compensate the observation noise covariance matrix of the long-timescale channel based on the fluctuation noise estimate to obtain the compensated observation noise covariance matrix.

[0013] As a preferred embodiment of the predictive model-based air-cooled fuel cell stack fault early warning and control method described in this invention, step S4 specifically includes: Step S401: When the fault dominant type label is reversible, output a level 1 maintenance warning and execute step S402. When the fault dominant type label is irreversible, proceed to step S403; Step S402: When the recoverable window period is greater than the preset descent threshold, output a thermal management adjustment command; When the recoverable window period is less than the preset emergency threshold, a short-term load reduction recovery command is output.

[0014] As a preferred embodiment of the predictive model-based air-cooled fuel cell stack fault early warning and control method described in this invention, step S4 further includes: Step S403: When the predicted end-of-life time value is less than or equal to the preset end-of-life time threshold or the irreversible decline health index is greater than the preset irreversible health index threshold, a level three mandatory warning is triggered, and a mandatory power limiting command and a shutdown and replacement prompt command are output.

[0015] The beneficial effects of this invention are as follows: Existing equivalent circuit fitting methods rely on preset topology and initial value selection, resulting in strong subjectivity in frequency point selection. Under dynamic conditions, these methods are prone to fitting drift and struggle to stably track the independent changes of the two types of impedance. This invention introduces relaxation time distribution deconvolution, eliminating the need for preset circuit structures. By inverting the time constant distribution, it can stably separate ohmic polarization and charge transfer impedance within each impedance update cycle, overcoming the aforementioned inherent defects. This makes independent online tracking of membrane drying and aging degradation possible. Furthermore, based on the difference in the rates of change of the two types of impedance, the current dominant fault type is dynamically determined, and a dual-channel extended Kalman filter predictor is constructed to independently recursively deduce reversible and irreversible health at second-minute and hour-month scales, respectively. The index, designed to address the issue of time constants differing by several orders of magnitude between the two types of decay processes, breaks away from the conventional architecture of a single model that considers multiple scales. It fundamentally avoids the accumulation of biases caused by mutual interference between rapidly changing signals and slowly changing drifts within the same recursive framework. Simultaneously, it adaptively compensates for the covariance of long-term channel observation noise through short-term channel innovation sequences, addressing the inherent difficulty of covariance estimation mismatch caused by sparse observation samples in long-term prediction, thus improving the robustness of end-of-life prediction. Finally, control commands are executed in stages based on fault type, recoverable window period, and end-of-life prediction value, avoiding both premature power limiting leading to availability loss and delayed intervention causing irreversible damage propagation. This effectively extends the actual service life of the fuel cell stack while ensuring operational safety. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the steps of a method for early warning and control of air-cooled fuel cell stack faults based on a predictive model, provided in an embodiment of the present invention.

[0017] Figure 2 This is a schematic diagram of a reversible degradation hierarchical control process provided in one embodiment of the present invention. Detailed Implementation

[0018] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0019] Example, refer to Figure 1 This embodiment of the present invention provides a method for early warning and control of air-cooled fuel cell stack faults based on a predictive model, including the following steps: This embodiment focuses on three core aspects: full-band impedance spectrum reconstruction, dual-channel lifetime prediction and differentiated control, and provides a detailed description of the specific implementation steps of the above method; Step S1: Reconstruct the full-band impedance spectrum based on the response data of the fuel cell stack, and separate the ohmic polarization impedance and charge transfer impedance from the full-band impedance spectrum. Step S2: Based on ohmic polarization impedance and charge transfer impedance, determine the fault dominance type and obtain the current fault dominance type label, reversible degradation health index and irreversible degradation health index. Step S3: Establish a dual-channel predictor based on the reversible and irreversible health indices as a prediction model. Perform predictions at different time scales through the dual-channel predictor to obtain the predicted values ​​of the recoverable window period and the end of life. Step S4: Based on the fault dominant type label, recoverable window period, and end-of-life prediction value, determine the fault response mode and execute the corresponding differentiated control instructions.

[0020] In one embodiment, steps S1, S2, S3, and S4 are executed sequentially. Step S1 is an impedance separation step, where the stack's response data is reconstructed to obtain a full-band impedance spectrum, and ohmic polarization impedance and charge transfer impedance are separated from the full-band impedance spectrum. Step S2 is a fault diagnosis step, where the fault dominance type is determined based on the ohmic polarization impedance and charge transfer impedance, resulting in a current fault dominance type label, a reversible decay health index, and an irreversible decay health index. Step S3 is a prediction step, where a dual-channel predictor is established based on the reversible and irreversible decay health indices as a prediction model. Predictions are performed at different time scales using the dual-channel predictor to obtain the recoverable window period and end-of-life prediction values. Step S4 is a response step, where the fault response mode is determined based on the fault dominance type label, recoverable window period, and end-of-life prediction values, and then corresponding differentiated control instructions are executed. Step S1 introduces relaxation time distribution deconvolution to replace traditional equivalent circuit fitting. This allows for stable acquisition of independent evolution information for the two types of impedance within each impedance update cycle without pre-setting the circuit topology, overcoming the drift defect of equivalent circuit fitting under dynamic conditions and enabling online independent tracking of membrane drying and aging degradation. Step S2 uses the divergence direction of the two impedance change rates as a criterion to distinguish between reversible moisture imbalance and irreversible catalyst aging, two degradation processes with different mechanisms, avoiding misjudging short-term membrane drying as permanent failure or a delayed response to actual catalyst layer degradation. Step S3 constructs an independent dual-channel recursive architecture to address the difference of several orders of magnitude in the time constants of the two degradation processes, preventing fast-changing signals from drifting with slow-changing signals. The cumulative deviation caused by mutual interference under the same recursive framework enables the short-term moisture recovery window and long-term end-of-life prediction to be output in a coordinated and non-interfering manner. In step S4, information from three dimensions—fault type, prognostic time window, and end-of-life—is used as decision input to change the crude intervention method under the existing single-threshold alarm mode, avoiding premature power limiting that causes availability loss or delayed intervention that leads to irreversible damage. Steps S1, S2, S3, and S4 use impedance separation results, fault type and health index, and window period and end-of-life prediction value as inputs and outputs between steps, forming a complete closed loop from mechanism identification to prognostic assessment to graded response, which extends the actual service life of the fuel cell stack while ensuring operational safety.

[0021] Step S1 specifically includes: Step S101: A multi-frequency small current disturbance signal is superimposed onto the DC bus of the fuel cell stack through a DC / DC converter; The data acquisition device collects the fuel cell stack response data, and the discrete voltage data points and discrete current data points in the fuel cell stack response data are arranged in the order of acquisition time to form the original voltage response time domain sequence and the original current response time domain sequence. By using a sliding time window to extract the original voltage response time-domain sequence and the original current response time-domain sequence, the voltage response time-domain sequence and the current response time-domain sequence are obtained.

[0022] Step S1 also includes: Step S102: Perform DC removal processing on the voltage response time domain sequence and the current response time domain sequence respectively to obtain the DC removed voltage response time domain sequence and the DC removed current response time domain sequence. Perform Fast Fourier Transform on the voltage response time-domain sequence and the current response time-domain sequence after DC removal, respectively, to obtain the voltage frequency domain sequence and the current frequency domain sequence. The ratio of the voltage frequency domain sequence to the current frequency domain sequence is calculated to obtain the first full-band impedance spectrum; Invalid frequency points are removed from the first full-band impedance spectrum to obtain the second full-band impedance spectrum.

[0023] Step S1 also includes: Step S103: Input the second full-band impedance spectrum into the relaxation time distribution deconvolution algorithm for processing to obtain two characteristic peaks; In the two characteristic peaks, the characteristic peak located in the first frequency band is assigned to the ohmic polarization impedance, and the characteristic peak located in the second frequency band is assigned to the charge transfer impedance, thus obtaining the real-time sequence of the ohmic polarization impedance and the real-time sequence of the charge transfer impedance.

[0024] In one embodiment, step S1 specifically includes: step S101, superimposing a multi-band small current disturbance signal onto the DC bus of the fuel cell stack through a DC / DC converter; The multi-band small current disturbance signal is generated by superimposing multiple sinusoidal signals of different frequencies. In the frequency domain, the multi-band small current disturbance signal includes discrete frequency components ranging from 0.1Hz to 10kHz, and covers at least three sub-bands: 0.1Hz~1Hz, 1Hz~100Hz, and 1kHz~10kHz. The amplitude distribution of each frequency component in the multi-band small current disturbance signal is as follows: the initial amplitude of each frequency component is equal, and the amplitude of each frequency component is adjusted stepwise to ensure that each frequency component corresponds to a specific frequency in the stack response signal. All components exhibit response amplitudes greater than or equal to a preset signal-to-noise ratio (SNR) threshold. This threshold is set at 20 dB. The basis for this setting is that, through offline calibration experiments, noise signals from the fuel cell were collected when the fuel cell was in an open-circuit state without any external disturbance signals. The impedance spectrum phase error and relaxation time distribution deconvolution reconstruction error were measured under different SNRs. Experiments showed that when the SNR was below 20 dB, the impedance spectrum phase error exceeded 5° and the peak fitting residual of the relaxation time distribution deconvolution reconstruction increased by more than 50%, failing to meet the relaxation time distribution requirements. To ensure the convergence accuracy of the deconvolution algorithm, the preset signal-to-noise ratio threshold is set to 20dB. The offline calibration experiment, involving step-by-step adjustments, is as follows: Under the rated operating conditions of the fuel cell stack (rated operating conditions refer to operation at the factory-calibrated rated current, rated operating temperature, rated gas metering ratio, and rated gas pressure), a stable state is achieved when the output voltage fluctuation does not exceed 1% of the rated voltage within 60 consecutive seconds. An amplitude distribution experiment is then performed under this stable state to ensure impedance measurements are conducted under stable operation under rated conditions, eliminating operating condition fluctuations. To mitigate interference with the amplitude of the response signal, the amplitude of each frequency component is adjusted sequentially, using its initial amplitude as a reference. The amplitude is increased or decreased in increments of 0.1% of the rated current. After each adjustment, the amplitude of the frequency component in the response signal is recorded. Adjustment stops when the response amplitude of the frequency component reaches 1.2 times the amplitude corresponding to a preset signal-to-noise ratio threshold of 20dB, and this amplitude is recorded as the initial allocated amplitude for that frequency component. This process is repeated for each frequency component to obtain its initial allocated amplitude. The initial allocated amplitudes of each frequency component are summed. If the sum is less than or equal to 5% of the rated current, the initial allocated amplitude is used as the final allocated amplitude of each frequency component. If the sum is greater than 5% of the rated current, the amplitudes of all frequency components are reduced proportionally according to the initial allocated amplitude of each frequency component until the sum of the absolute values ​​of the amplitudes of each frequency component equals 5% of the rated current. The reduced amplitude is then used as the final allocated amplitude of each frequency component. The data acquisition device collects the fuel cell stack response data, and the discrete voltage data points and discrete current data points in the fuel cell stack response data are arranged in the order of acquisition time to form the original voltage response time domain sequence and the original current response time domain sequence. The data acquisition device is a multi-channel synchronous acquisition card with a sampling frequency of 20kHz. The front end of the sampling channel of the acquisition card is equipped with an anti-aliasing low-pass filter with a cutoff frequency of 10kHz. By using a sliding time window to extract the original voltage response time-domain sequence and the original current response time-domain sequence, the voltage response time-domain sequence and the current response time-domain sequence are obtained.

[0025] Wherein, the length of the sliding time window is greater than or equal to twice the period corresponding to the lowest frequency component; the lowest frequency component is 0.1Hz; in this embodiment, the length of the sliding time window is set to 20 times the period corresponding to the lowest frequency component, that is, 20 seconds; Step S1 further includes: Step S102, performing DC removal processing on the voltage response time domain sequence and the current response time domain sequence respectively to obtain the DC removed voltage response time domain sequence and the DC removed current response time domain sequence; Specifically, the DC removal process involves calculating the arithmetic mean of all sampling points in the voltage response time-domain sequence and the arithmetic mean of all sampling points in the current response time-domain sequence, subtracting the arithmetic mean of the voltage response time-domain sequence from the value of each sampling point in the voltage response time-domain sequence, and subtracting the arithmetic mean of the current response time-domain sequence from the value of each sampling point in the current response time-domain sequence, to obtain the DC-removed voltage response time-domain sequence and the DC-removed current response time-domain sequence. Perform Fast Fourier Transform on the voltage response time-domain sequence and the current response time-domain sequence after DC removal, respectively, to obtain the voltage frequency domain sequence and the current frequency domain sequence. Both the voltage frequency domain sequence and the current frequency domain sequence include several discrete frequency points, each corresponding to a discrete frequency value. When performing a fast Fourier transform, the number of transform points is padded with zeros to a power of 2. The frequency resolution of both the voltage and current frequency domain sequences is equal to the ratio of the sampling frequency of the acquisition card to the number of transform points, and the frequency resolution is less than or equal to the lowest frequency component. In this embodiment, the number of transform points is padded with zeros to 262144, which is 2 to the power of 18, and the frequency resolution is 20000 / 262144≈0.076Hz, which is less than the lowest frequency component of 0.1Hz. The ratio of the voltage frequency domain sequence to the current frequency domain sequence is calculated to obtain the first full-band impedance spectrum; invalid frequency points are removed from the first full-band impedance spectrum to obtain the second full-band impedance spectrum. Specifically, invalid frequency point removal involves, for each frequency point in the first full-band impedance spectrum, if the amplitude of the current frequency domain sequence at that frequency point is lower than a preset signal-to-noise ratio threshold, then that frequency point is determined to be invalid data and removed from the first full-band impedance spectrum to obtain the second full-band impedance spectrum. The second full-band impedance spectrum includes ohmic polarization impedance and polarization impedance. Polarization impedance is the impedance component in the second full-band impedance spectrum other than ohmic polarization impedance, including charge transfer impedance and diffusion impedance. Step S1 also includes: Step S103, inputting the second full-band impedance spectrum into the relaxation time distribution deconvolution algorithm for processing to obtain two characteristic peaks; The specific processing includes: a relaxation time distribution deconvolution algorithm based on Tikhonov regularization, which inverts the second full-band impedance spectrum based on the integral convolution relationship between polarization impedance and time constant distribution function to obtain a continuous distribution function with respect to time constant; and extracts two characteristic peaks from the continuous distribution function. The process of extracting two characteristic peaks from the continuous distribution function specifically involves calculating the second derivative of the continuous distribution function, locating the local minimum point of the second derivative as the peak boundary, calculating the integral area within the coverage area of ​​each characteristic peak, and selecting the two peaks with the largest integral areas as the two characteristic peaks. Each execution of steps S101 to S102 is recorded as one impedance update cycle. The Tikhonov regularization parameter is determined using the L-curve criterion, and the Tikhonov regularization parameter is recalculated after every 10 impedance update cycles. The process of determining the regularization parameter using the L-curve criterion specifically involves plotting the curves of the solution norm and residual norm under different regularization parameters in a double logarithmic coordinate system, and using the regularization parameter corresponding to the curve inflection point as the optimal value. In this embodiment, the sliding time window is set to 20 seconds based on the lowest frequency of 0.1 Hz, and the impedance update cycle corresponds to approximately 20 seconds. After offline calibration, the water migration time constant within the fuel cell membrane is 200 seconds, which is 10 times the impedance update cycle. Therefore, the Tikhonov regularization parameter is recalculated every 10 impedance update cycles to balance tracking performance and computational load. When the rate of change of current density or temperature exceeds a preset threshold, a recalculation is triggered immediately to handle sudden changes in operating conditions. The preset thresholds include a temperature rate of change threshold of 1℃ / s and a current rate of change threshold of 5% / second of the rated current. These thresholds were determined based on offline calibration experiments. Experimental results show that when the temperature rate of change is greater than 1℃ / s, the output voltage fluctuation exceeds 3% of the rated voltage, and the impedance spectrum measurement repeatability error exceeds 5%. When the current rate of change is greater than 5% / second of the rated current, the stack dynamic response convergence time exceeds 5 seconds, exceeding a single impedance update cycle. Under these conditions, effective impedance measurement cannot be completed in a stable state, thus requiring immediate recalculation of the regularization parameter. In the two characteristic peaks, the characteristic peak located in the first frequency band is corresponding to the ohmic polarization impedance, and the characteristic peak located in the second frequency band is corresponding to the charge transfer impedance, thus obtaining the real-time sequence of the ohmic polarization impedance and the real-time sequence of the charge transfer impedance. The real-time sequence of ohmic polarization impedance and the real-time sequence of charge transfer impedance are obtained by extracting the peak resistance values ​​of two characteristic peaks in the first and second frequency bands in each impedance update cycle, and appending the peak resistance values ​​extracted in the current cycle to the end of the sequence of the previous cycle in time order to obtain the updated real-time sequence of ohmic polarization impedance and the real-time sequence of charge transfer impedance. The first frequency band is from 1 kHz to 10 kHz, and the second frequency band is from 0.1 Hz to 100 Hz. The characteristic peak corresponding to the first frequency band is the ohmic polarization impedance, and the characteristic peak corresponding to the second frequency band is the charge transfer impedance. The correspondence between the first and second frequency bands and the ohmic polarization impedance and charge transfer impedance is determined through offline calibration experiments. Specifically, the impedance spectrum of the entire frequency band is collected for the fuel cell stack at at least two preset membrane water contents and at least two preset catalyst aging degrees, and the characteristic peaks of relaxation time distribution are extracted. The experimental results show that the peak value in the range of 1 kHz to 10 kHz changes only with the membrane water content, and the peak value in the range of 0.1 Hz to 100 Hz changes only with the catalyst aging degree. Therefore, the first and second frequency bands are respectively assigned to the ohmic polarization impedance and the charge transfer impedance. Existing impedance measurement methods typically fit based on a pre-defined equivalent circuit model, relying on topological assumptions and initial value selection. Under dynamic conditions, these methods are prone to fitting non-convergence or drift, and cannot distinguish the differential contributions of membrane drying and catalyst aging to impedance under online conditions. Step S1 uses full-band multi-sinusoidal superposition perturbation and sliding time window extraction to synchronously acquire broadband impedance information in each impedance update cycle. Combined with a Tikhonov-regularized relaxation time distribution deconvolution algorithm, two characteristic peaks with clear physical correspondence can be obtained without pre-defined circuit topology. These peaks quantify membrane water content (ohmic polarization impedance) and catalyst aging degree (charge transfer impedance), respectively, enabling online independent tracking of both reversible and irreversible decay mechanisms. This provides input parameters with physical mechanism support for subsequent fault type identification.

[0026] Step S2 specifically includes: Step S201: Input the real-time sequence of ohmic polarization impedance and the real-time sequence of charge transfer impedance into the Randles-type equivalent circuit model for processing to obtain the real-time absolute amplitude of impedance under the current operating condition. Step S202: By using a sliding time window, calculate the rate of change of ohmic polarization impedance and charge transfer impedance relative to time, and obtain the rate of change of ohmic polarization impedance and the rate of change of charge transfer impedance.

[0027] Step S2 also includes: Step S203: Determine the fault-dominant type based on the rate of change of ohmic polarization impedance and the rate of change of charge transfer impedance, and obtain the current fault-dominant type label, reversible degradation health index, and irreversible degradation health index. When the real-time absolute impedance amplitude is greater than the preset impedance abnormality threshold, the fault dominant type determination is performed. When the real-time absolute impedance amplitude is less than or equal to the preset impedance abnormality threshold, the fault-dominant type determination is not performed, and the fault-dominant type label of the previous output remains unchanged. The determination of the dominant fault type specifically includes: When the rate of change of ohmic polarization impedance is greater than the ohmic polarization slope threshold and the rate of change of charge transfer impedance is less than the activation polarization slope threshold, the current fault dominant type is determined to be reversible decay, the current fault dominant type is labeled as reversible, and the reversible decay health index is calculated according to the first preset mapping relationship based on the amount by which the rate of change of ohmic polarization impedance exceeds the ohmic polarization slope threshold. The irreversible decay health index is assigned the value of 1 minus the reversible decay health index. When the rate of change of charge transfer impedance is continuously greater than the activation polarization slope threshold and the rate of change of ohmic polarization impedance is less than or equal to the ohmic polarization slope threshold, the current fault dominant type is determined to be irreversible decay. The current fault dominant type is labeled as irreversible. Based on the extent to which the rate of change of charge transfer impedance exceeds the activation polarization slope threshold, the irreversible decay health index is calculated according to the second preset mapping relationship. The reversible decay health index is assigned the value of 1 minus the irreversible decay health index. Otherwise, keep the fault-dominant type label from the previous output unchanged.

[0028] In one embodiment, step S2 specifically includes: step S201, inputting the real-time sequence of ohmic polarization impedance and the real-time sequence of charge transfer impedance into the Randles-type equivalent circuit model for processing to obtain the real-time absolute amplitude of impedance under the current operating condition. Among them, the Randles-type equivalent circuit model includes a topology structure in which the ohmic polarization impedance and the charge transfer impedance are connected in series and then in series with the diffusion impedance. Step S202: By using a sliding time window, calculate the rate of change of ohmic polarization impedance and charge transfer impedance relative to time, and obtain the rate of change of ohmic polarization impedance and the rate of change of charge transfer impedance. The specific process for calculating the rate of change of ohmic polarization impedance and the rate of change of charge transfer impedance is as follows: Calculate the time difference between the current moment and the previous moment as the sampling time interval; calculate the difference between the ohmic polarization impedance at the current moment and the ohmic polarization impedance at the previous moment to obtain the ohmic polarization impedance difference; calculate the ratio of this ohmic polarization impedance difference to the sampling time interval to obtain the rate of change of ohmic polarization impedance; calculate the difference between the charge transfer impedance at the current moment and the charge transfer impedance at the previous moment to obtain the charge transfer impedance difference; calculate the ratio of this charge transfer impedance difference to the sampling time interval to obtain the rate of change of charge transfer impedance. Step S2 also includes: Step S203, determining the fault-dominant type based on the rate of change of ohmic polarization impedance and the rate of change of charge transfer impedance, and obtaining the current fault-dominant type label, reversible degradation health index and irreversible degradation health index. The sum of the reversible decline health index and the irreversible decline health index is 1. When the real-time absolute impedance amplitude is greater than the preset impedance abnormality threshold, the fault dominant type determination is performed. The preset impedance anomaly threshold is 1.5 times the absolute amplitude of the initial impedance collected under rated operating conditions when the fuel cell stack leaves the factory. This setting is based on the fact that, through offline calibration experiments, the initial impedance amplitude was collected under rated operating conditions, and fault triggering tests were conducted at 1.2 times, 1.5 times, and 2.0 times the initial impedance amplitude, respectively. The experimental results show that when the ratio is lower than 1.5 times, fluctuations under normal operating conditions will trigger false alarms. When it is higher than 1.5 times, the impedance amplitude has exceeded 1.5 times the initial amplitude when the threshold is reached. At this time, the irreversible damage critical point corresponding to the irreversible loss of catalyst aging exceeding 20% ​​is reached. The impedance amplitude margin is less than 10% of the initial amplitude, which cannot reserve at least 140 seconds of response time for intervention actions such as thermal management adjustment or load reduction recovery. Therefore, the preset impedance anomaly threshold is set to 1.5 times the initial impedance amplitude. When the real-time absolute impedance amplitude is less than or equal to the preset impedance abnormality threshold, the fault-dominant type determination is not performed, and the fault-dominant type label of the previous output remains unchanged. The fault dominance type determination specifically includes: when the rate of change of ohmic polarization impedance is greater than the ohmic polarization slope threshold and the rate of change of charge transfer impedance is less than the activation polarization slope threshold, the current fault dominance type is determined to be reversible decay, the current fault dominance type is labeled as reversible, and the reversible decay health index is calculated according to the first preset mapping relationship based on the amount by which the rate of change of ohmic polarization impedance exceeds the ohmic polarization slope threshold. The irreversible decay health index is assigned the value of 1 minus the reversible decay health index. The ohmic polarization slope threshold is set to 0.5 mΩ / s. This setting is based on the fact that, through offline calibration experiments, the membrane drying process was simulated under rated operating conditions, and the rate of change of ohmic polarization impedance was measured to be in the range of 0.3~0.8 mΩ / s. During normal fluctuations, it does not exceed 0.5 mΩ / s. Therefore, the threshold is set to 0.5 mΩ / s to exclude normal fluctuations and effectively identify membrane drying. When the rate of change of charge transfer impedance is continuously greater than the activation polarization slope threshold and the rate of change of ohmic polarization impedance is less than or equal to the ohmic polarization slope threshold, the current fault dominant type is determined to be irreversible decay. The current fault dominant type is labeled as irreversible. Based on the extent to which the rate of change of charge transfer impedance exceeds the activation polarization slope threshold, the irreversible decay health index is calculated according to the second preset mapping relationship. The reversible decay health index is assigned the value of 1 minus the irreversible decay health index. The term "consistently greater than" refers to the measurement value of the charge transfer impedance change rate being greater than the activation polarization slope threshold for three consecutive impedance update cycles. This is used to avoid misjudgments caused by fluctuations in a single measurement. The activation polarization slope threshold is set to 0.3 mΩ / s. This setting is based on the fact that, through offline calibration experiments, the catalyst aging process was simulated by accelerated aging experiments under the rated operating conditions of the fuel cell stack. The charge transfer impedance change rate during the aging process was recorded to be within the range of 0.1~0.5 mΩ / s, and the normal fluctuation did not exceed 0.3 mΩ / s. Therefore, the activation polarization slope threshold is set to 0.3 mΩ / s to exclude normal fluctuations and effectively identify catalyst aging. Otherwise, keep the fault-dominant type label from the previous output unchanged; Both the first and second preset mapping relationships are continuous monotonic mappings that normalize the portion exceeding the threshold to the (0,1) interval. Existing fault diagnosis methods typically rely on single threshold alarms or simple logic judgments, which cannot distinguish between reversible membrane dryness and irreversible catalyst aging, easily leading to misjudgment or missed judgment. Step S2 obtains the absolute amplitude of the impedance by inputting the ohmic polarization impedance and charge transfer impedance into the Randle equivalent circuit model as a prerequisite for determining the dominant fault type. At the same time, the rate of change of the ohmic polarization impedance and the charge transfer impedance are calculated separately. Based on the differentiation direction of the rate of change of the ohmic polarization impedance and the rate of change of the charge transfer impedance, a bidirectional threshold comparison is performed, realizing the online distinction between reversible and irreversible degradation, avoiding misjudging short-term membrane dryness as permanent failure or lag in response to catalyst layer aging.

[0029] Step S3 specifically includes: Step S301: Obtain the current current density and the current stack temperature; Based on the reversible decay health index, the irreversible decay health index, the current current density, and the current stack temperature, an extended Kalman filter is used to establish a dual-channel predictor. The dual-channel predictor includes a short-timescale channel and a long-timescale channel; Step S302: Through a short timescale channel, perform an extended Kalman filter recursive estimation once in each impedance update cycle to obtain the predicted trajectory of the reversible decline health index. Based on the predicted trajectory, the remaining time for the reversible decline health index to reach the preset membrane dry safety health index threshold is calculated, thus obtaining the recoverable window period.

[0030] Step S3 also includes: Step S303: Extrapolate and predict the irreversible decline health index through a long-term timescale channel to obtain the predicted trajectory of the irreversible decline health index. Based on the predicted trajectory, the remaining time for the irreversible decline health index to reach the preset end-of-life health index threshold is calculated, and the end-of-life predicted value is obtained. Obtain the fluctuation noise estimate of the short-timescale channel, and compensate the observation noise covariance matrix of the long-timescale channel based on the fluctuation noise estimate to obtain the compensated observation noise covariance matrix.

[0031] In one embodiment, step S3 specifically includes: step S301, obtaining the current current density and the current stack temperature; The process of obtaining the current current density includes: acquiring the current load current signal through a current sensor, reading the current value in the current load current signal, and dividing the current value by the membrane electrode active area to obtain the current current density; the membrane electrode active area is the design parameter when the fuel cell stack leaves the factory; the current fuel cell stack temperature is acquired in real time through a temperature sensor. Based on the reversible decay health index, the irreversible decay health index, the current current density, and the current stack temperature, an extended Kalman filter is used to establish a dual-channel predictor. The dual-channel predictor is built based on the extended Kalman filter. The building process includes taking the reversible and irreversible health indices acquired at the current time as the two components of the initial state vector of the EKF. Both the reversible and irreversible health indices are dimensionless values, ranging from 0 to 1, and their sum is 1. The initial value of the state covariance matrix is ​​set. Specifically, the state covariance matrix is ​​a diagonal matrix, and its diagonal elements are determined according to the uncertainty of the initial estimated values ​​of each state variable in the initial state vector. The uncertainty is characterized by the statistical variance of the initial estimated values, which is obtained by statistically analyzing the initial health indices measured multiple times through offline calibration experiments. Based on the current current density and current stack temperature, a pre-set noise covariance (MAP) table is consulted to set the initial values ​​of the process noise covariance matrix and the observation noise covariance matrix. The noise covariance (MAP) table is a lookup table pre-generated through offline calibration experiments, used to establish the mapping relationship between the current current density, current stack temperature, and the optimal noise covariance matrix. The optimal noise covariance matrix includes the process noise covariance matrix and the observation noise covariance matrix. In the offline calibration experiments, the optimization objective is to minimize the statistical variance of the EKF output residual sequence. Different combinations of the current current density and current stack temperature are iterated to determine the corresponding optimal noise covariance matrix. The compensated observation noise covariance matrix is ​​based on the initial set value in the MAP table, with the fluctuation noise estimate superimposed on it. This is used to update the observation noise covariance matrix of the long-timescale channel at the current moment, and is used for subsequent extended Kalman filter recursive calculations. State transition equations are defined for the short-term and long-term channels. Specifically, the short-term and long-term channels are two independently operating extended Kalman filters. The short-term channel uses the reversible declining health index as the state variable and the prediction time domain is from seconds to minutes. Its state transition equation adopts a linear random walk model. The long-term channel uses the irreversible declining health index as the state variable and the prediction time domain is from hours to months. Its state transition equation also adopts a linear random walk model. Furthermore, the state transition equation of the long-term channel is forced to constrain the state variable to be monotonically non-decreasing. The monotonically non-decreasing constraint is implemented as follows: if the predicted value of the current step in the extended Kalman filter recursion is less than the estimated value of the previous step, then the estimated value of the previous step is used as the predicted value of the current step. Define the observation equation that maps the state vector to the observation, specifically including: For the first The state vector at time t includes two components: a reversible decline health index and an irreversible decline health index; the observation equation is... ;in, This represents the vector of measured values ​​of the reversible and irreversible health indices obtained at the current moment. Represents a 2×2 identity matrix. Indicates observation noise; After completing the above settings, following the prediction-update recursive process of the standard extended Kalman filter, state estimation and covariance update are performed once in each impedance update cycle. After each update, the estimated values ​​of the reversible decay health index and the irreversible decay health index are normalized to obtain the corresponding first and second estimates. The sum of the first and second estimates is made to 1, thus obtaining the dual-channel predictor. The dual-channel predictor includes a short-timescale channel and a long-timescale channel; Step S302: Through a short timescale channel, perform an extended Kalman filter recursive estimation once in each impedance update cycle to obtain the predicted trajectory of the reversible degradation health index; based on the predicted trajectory, calculate the remaining time for the reversible degradation health index to reach the preset membrane dry safety health index threshold to obtain the recoverable window period. The remaining time for the reversible decline health index to reach the preset membrane dry safety health index threshold is calculated. Specifically, on the predicted trajectory of the reversible decline health index, the search is performed forward along the time axis from the current moment to locate the time point when the predicted value of the reversible decline health index first reaches or exceeds the preset membrane dry safety health index threshold; the time difference between this time point and the current moment is calculated to obtain the recoverable window period. If the predicted values ​​of the reversible decay health index at all points on the predicted trajectory are lower than the preset membrane dry safety health index threshold, then the recoverable window period is determined to be infinite; where the preset membrane dry safety health index threshold is the reversible decay health index threshold corresponding to the upper limit of ohmic polarization impedance safety determined by offline calibration. Step S3 also includes: Step S303, extrapolating and predicting the irreversible decline health index through a long-term timescale channel to obtain the predicted trajectory of the irreversible decline health index; and calculating the remaining time for the irreversible decline health index to reach the preset end-of-life health index threshold based on the predicted trajectory to obtain the end-of-life predicted value. Specifically, the remaining time for the irreversible decline health index to reach the preset end-of-life health index threshold is calculated by searching forward along the time axis from the current moment on the predicted trajectory of the irreversible decline health index to locate the time point when the predicted value of the irreversible decline health index first reaches or exceeds the preset end-of-life health index threshold; calculating the time difference between this time point and the current moment to obtain the end-of-life predicted value; if the predicted values ​​of the irreversible decline health index at all points on the predicted trajectory are lower than the preset end-of-life health index threshold, then the end-of-life predicted value is determined to be infinite; wherein, the preset end-of-life health index threshold is the end-of-life value of the irreversible decline health index determined through offline calibration. Obtain the fluctuation noise estimate of the short timescale channel, and compensate the observation noise covariance matrix of the long timescale channel based on the fluctuation noise estimate to obtain the compensated observation noise covariance matrix. The acquisition process includes recording the difference between the reversible decay health index obtained at the current moment and the reversible decay health index estimated at the previous moment during the recursive estimation of extended Kalman filtering in the short time scale channel, thus obtaining the innovation sequence; calculating the variance of the values ​​of the innovation sequence within a preset time window to obtain the variance estimate, which is used as the fluctuation noise estimate; wherein, the preset time window is 10 impedance update cycles, which is used to achieve a balance between noise estimation accuracy and response speed; The compensation process includes superimposing the fluctuation noise estimate onto the diagonal elements of the observation noise covariance matrix of the long-timescale channel to obtain the compensated observation noise covariance matrix. The compensated observation noise covariance matrix is ​​used to replace the observation noise covariance matrix of the long-timescale channel at the current moment and participates in the Kalman gain calculation and state covariance update in the extended Kalman filter update step. It is only valid within the current impedance update cycle. In the next cycle, the initial value is reset through the MAP table and the compensation is performed again based on the latest fluctuation noise estimate. Existing prediction models do not construct differentiated recursive frameworks for the different time constants of reversible and irreversible decline. When a single model makes cross-scale predictions, rapidly changing signals and slowly changing drifts interfere with each other, resulting in a lack of coordinated estimation capabilities for short-term recovery windows and long-term life ends. Step S3 constructs a dual-channel extended Kalman filter predictor that includes short-term and long-term time-scale channels. It independently recursively derives reversible and irreversible health indices at the second-minute and hour-month scales, respectively. It also uses the innovation sequence of the short-term time-scale channel to adaptively compensate for the observation noise covariance of the long-term time-scale channel, thereby achieving coordinated prediction of reversible and irreversible decline. This provides predictive values ​​for recoverable window periods and life ends for differentiated management.

[0032] Step S4 specifically includes: Step S401: When the fault dominant type label is reversible, output a level 1 maintenance warning and execute step S402. When the fault dominant type label is irreversible, proceed to step S403; Step S402: When the recoverable window period is greater than the preset descent threshold, output a thermal management adjustment command; When the recoverable window period is less than the preset emergency threshold, a short-term load reduction recovery command is output.

[0033] Step S4 also includes: Step S403: When the predicted end-of-life time value is less than or equal to the preset end-of-life time threshold or the irreversible decline health index is greater than the preset irreversible health index threshold, a level three mandatory warning is triggered, and a mandatory power limiting command and a shutdown and replacement prompt command are output.

[0034] In one embodiment, step S4 specifically includes: step S401, when the fault-dominant type label is reversible, outputting a level-one maintenance warning and executing step S402; when the fault-dominant type label is irreversible, executing step S403; step S402, when the recoverable window period is greater than a preset descent threshold, outputting a thermal management adjustment command; the specific process of reversible degradation graded control in this embodiment is as follows: Figure 2 As shown; The preset slack-down threshold is 300 seconds, which allows for a full trial period of 120 seconds and an observation window of 180 seconds for thermal management adjustments. If the adjustment is ineffective, the load can be immediately reduced. At the same time, thermal management adjustments are initiated when the recoverable window period is greater than 300 seconds, so as to avoid affecting availability by directly reducing the load before the thermal management adjustment has been proven to be ineffective. The thermal management adjustment command is to increase the humidity of the infeed air until the ohmic polarization impedance falls back to the safe range; The adjustment action in the thermal management adjustment command has no fixed adjustment amount. The stopping condition is that the ohmic polarization impedance falls back to the safe range. During the execution, the ohmic polarization impedance is monitored in real time. When the ohmic polarization impedance falls back to the safe range, the execution of the thermal management adjustment command stops. If the ohmic polarization impedance fails to fall back to the safe range within 120 seconds after the execution of the thermal management adjustment command, the execution of the short-term load reduction recovery command will be initiated. The 120-second interval is used to promptly switch to load reduction intervention when thermal management adjustment is ineffective, avoiding waiting for more than 120 seconds, which would cause the ohmic polarization impedance to increase further and delay the start of load reduction recovery. The safe range is the range in which the ohmic polarization impedance is lower than the preset membrane dry safety impedance threshold. The preset membrane dry safety impedance threshold is the upper limit of the ohmic polarization impedance determined by offline calibration. The offline calibration process of the preset membrane dry safety impedance threshold is as follows: under the rated operating conditions of the fuel cell stack, the relative humidity of the air entering the stack is gradually reduced in increments of 5%. After the humidity stabilizes at each level, the ohmic polarization impedance and the output performance of the fuel cell stack are recorded. The ohmic polarization impedance value corresponding to the inflection point of irreversible decay of the output performance is the threshold. When the recoverable window period is less than the preset emergency threshold, a short-term load reduction recovery command is output. The preset emergency threshold is 60 seconds, which is used to trigger load reduction recovery in a timely manner when the recoverable window period is less than 60 seconds, so as to avoid the aggravation of membrane dryness due to failure to start load reduction for more than 60 seconds, and to avoid frequent triggering of load reduction during normal fluctuations. If the recoverable window period is between the preset slow reduction threshold and the preset emergency threshold, no new control instructions will be executed, the current operating status will continue to be maintained and the changes in the recoverable window period will be continuously monitored. The short-time load descent recovery command is as follows: reduce the load current to 60% of the rated current and continue operating for 20 seconds. After the ohmic polarization impedance returns to the safe range, restore the load current to the level before the short-time load descent recovery command was executed. Reducing the load current to 60% of the rated current ensures continuous output of the fuel cell stack while providing dynamic space for the recovery of moisture in the membrane. The continuous operation for 20 seconds covers the time constant of the membrane moisture recovery, allowing the ohmic polarization impedance to return to the safe range. Step S4 also includes: Step S403, when the predicted end-of-life value is less than or equal to the preset end-of-life time threshold or the irreversible decline health index is greater than the preset irreversible health index threshold, a level three mandatory warning is triggered, and a mandatory power limiting command and a shutdown replacement prompt command are output. The mandatory power limiting command restricts the load current to no more than 40% of the rated current; the shutdown and replacement prompt command provides a shutdown and replacement prompt for the fuel cell stack; limiting the load current to no more than 40% of the rated current avoids accelerated carbon corrosion caused by local reverse polarity and ensures the safe operation of the fuel cell stack under irreversible degradation. The preset end-of-life time threshold is the remaining lifespan warning time value determined through offline calibration, and the preset irreversible health index threshold is the mandatory warning upper limit value of the irreversible degradation health index determined through offline calibration; this fault response and control process ends. The existing control strategy is based on a single threshold alarm and does not provide graded responses based on the difference between reversible and irreversible degradation and the remaining recovery time. This results in premature power limiting for reversible degradation, causing availability loss, or delayed intervention for irreversible degradation, which accelerates the spread of damage. Step S4 makes a comprehensive judgment based on three dimensions of information: fault type label, recoverable window period, and end-of-life prediction value. For reversible degradation, thermal management adjustment or short-term load reduction recovery is performed according to the recoverable window period. For irreversible degradation, forced power limiting and shutdown replacement prompts are triggered, thus achieving differentiated and precise graded control.

[0035] To verify the effectiveness of this scheme, an experimental platform was built based on a certain type of air-cooled proton exchange membrane stack. Under rated operating conditions, namely rated current, rated operating temperature, rated gas metering ratio and rated gas pressure, comparative experiments were conducted using existing methods and this scheme. All indicators are average measured values ​​after 72 hours of continuous operation. Key indicator data for each stage were recorded. Specific experimental results are shown in Table 1.

[0036] Table 1; Table 1 compares the experimental data of our proposed solution with existing methods in four technical aspects. Existing methods exhibit performance bottlenecks in each aspect, specifically: the phase error of the equivalent circuit fitting under dynamic conditions exceeds 5°; the misclassification rate between reversible and irreversible degradation is 18.6%; the prediction error for the recoverable window period is ±80 seconds; the success rate of thermal management regulation under single-threshold response is 61.5%; and the stack availability is 82.4%. Experimental data show that our proposed solution outperforms existing methods in impedance separation accuracy, fault diagnosis accuracy, prediction accuracy, and control success rate, demonstrating significant performance improvements in each aspect. These experimental data validate the overall effectiveness of our proposed solution from impedance separation to hierarchical control.

[0037] This invention addresses the shortcomings of existing equivalent circuit fitting methods, which rely on preset topologies and initial value selection, suffer from strong subjectivity in frequency point selection, are prone to fitting drift under dynamic conditions, and struggle to stably track the independent changes of two types of impedance. This invention introduces relaxation time distribution deconvolution, eliminating the need for preset circuit structures. By inverting the time constant distribution, it can stably separate ohmic polarization and charge transfer impedance within each impedance update cycle, overcoming the aforementioned inherent defects and enabling independent online tracking of membrane drying and aging degradation. Based on this, it dynamically determines the dominant fault type according to the difference in the rates of change of the two types of impedance and constructs a dual-channel extended Kalman filter predictor to independently deduce reversible and irreversible health indices on second-minute and hour-month scales, respectively. To address the issue of time constants differing by several orders of magnitude between the two types of decay processes, this approach breaks away from the conventional architecture of a single model that considers multiple scales. It fundamentally avoids the accumulation of biases caused by mutual interference between rapidly changing signals and slowly changing drifts within the same recursive framework. Simultaneously, it adaptively compensates for the covariance of long-term channel observation noise through short-term channel innovation sequences, addressing the inherent difficulty of covariance estimation mismatch caused by sparse observation samples in long-term prediction. This enhances the robustness of end-of-life prediction. Ultimately, control commands are executed in stages based on fault type, recoverable window period, and end-of-life prediction value. This avoids both premature power limiting leading to availability loss and delayed intervention causing irreversible damage propagation, effectively extending the actual service life of the fuel cell stack while ensuring operational safety.

[0038] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. 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.

[0039] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the protection scope of the present invention.

[0040] Certain aspects of this invention include the process steps and instructions described herein in algorithmic form. It should be noted that the process steps and instructions of this invention can be implemented in software, firmware, and / or hardware, and when implemented in software, they can be downloaded, stored on various operating systems and operated from said platforms.

[0041] Those skilled in the art will understand that the structures shown in the figures are merely block diagrams of some structures related to the present application and do not constitute a limitation on the terminal device to which the present application is applied. Specific terminal devices may include more or fewer components than those shown in the figures, or combine certain components, or have different component arrangements.

[0042] In the description of this specification, the use of terms such as "one embodiment," "some embodiments," "example," "specific example," or "possible design," etc., refers to a specific feature, structure, material, or characteristic described in connection with that embodiment or example, which is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in a suitable manner in any one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0043] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0044] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for early warning and control of air-cooled fuel cell stack faults based on a predictive model, characterized in that, Includes the following steps: Step S1: Reconstruct the full-band impedance spectrum based on the response data of the fuel cell stack, and separate the ohmic polarization impedance and charge transfer impedance from the full-band impedance spectrum. Step S2: Based on ohmic polarization impedance and charge transfer impedance, determine the fault dominance type and obtain the current fault dominance type label, reversible degradation health index and irreversible degradation health index. Step S3: Establish a dual-channel predictor based on the reversible and irreversible health indices as a prediction model. Perform predictions at different time scales through the dual-channel predictor to obtain the predicted values ​​of the recoverable window period and the end of life. Step S4: Based on the fault dominant type label, recoverable window period, and end-of-life prediction value, determine the fault response mode and execute the corresponding differentiated control instructions.

2. The method for early warning and control of air-cooled fuel cell stack faults based on predictive models as described in claim 1, characterized in that: Step S1 specifically includes: Step S101: A multi-frequency small current disturbance signal is superimposed onto the DC bus of the fuel cell stack through a DC / DC converter; The data acquisition device collects the fuel cell stack response data, and the discrete voltage data points and discrete current data points in the fuel cell stack response data are arranged in the order of acquisition time to form the original voltage response time domain sequence and the original current response time domain sequence. By using a sliding time window to extract the original voltage response time-domain sequence and the original current response time-domain sequence, the voltage response time-domain sequence and the current response time-domain sequence are obtained.

3. The method for early warning and control of air-cooled fuel cell stack faults based on predictive models as described in claim 2, characterized in that: Step S1 also includes: Step S102: Perform DC removal processing on the voltage response time domain sequence and the current response time domain sequence respectively to obtain the DC removed voltage response time domain sequence and the DC removed current response time domain sequence. Perform Fast Fourier Transform on the voltage response time-domain sequence and the current response time-domain sequence after DC removal, respectively, to obtain the voltage frequency domain sequence and the current frequency domain sequence. The ratio of the voltage frequency domain sequence to the current frequency domain sequence is calculated to obtain the first full-band impedance spectrum; Invalid frequency points are removed from the first full-band impedance spectrum to obtain the second full-band impedance spectrum.

4. The method for early warning and control of air-cooled fuel cell stack faults based on predictive models as described in claim 3, characterized in that: Step S1 also includes: Step S103: Input the second full-band impedance spectrum into the relaxation time distribution deconvolution algorithm for processing to obtain two characteristic peaks; In the two characteristic peaks, the characteristic peak located in the first frequency band is assigned to the ohmic polarization impedance, and the characteristic peak located in the second frequency band is assigned to the charge transfer impedance, thus obtaining the real-time sequence of the ohmic polarization impedance and the real-time sequence of the charge transfer impedance.

5. The method for early warning and control of air-cooled fuel cell stack faults based on predictive models as described in claim 1, characterized in that: Step S2 specifically includes: Step S201: Input the real-time sequence of ohmic polarization impedance and the real-time sequence of charge transfer impedance into the Randles-type equivalent circuit model for processing to obtain the real-time absolute amplitude of impedance under the current operating condition. Step S202: By using a sliding time window, calculate the rate of change of ohmic polarization impedance and charge transfer impedance relative to time, and obtain the rate of change of ohmic polarization impedance and the rate of change of charge transfer impedance.

6. The fault early warning and control method for air-cooled fuel cell stacks based on predictive models as described in claim 5, characterized in that: Step S2 also includes: Step S203: Determine the fault-dominant type based on the rate of change of ohmic polarization impedance and the rate of change of charge transfer impedance, and obtain the current fault-dominant type label, reversible degradation health index, and irreversible degradation health index. When the real-time absolute impedance amplitude is greater than the preset impedance abnormality threshold, the fault dominant type determination is performed. When the real-time absolute impedance amplitude is less than or equal to the preset impedance abnormality threshold, the fault-dominant type determination is not performed, and the fault-dominant type label of the previous output remains unchanged. The determination of the dominant fault type specifically includes: When the rate of change of ohmic polarization impedance is greater than the ohmic polarization slope threshold and the rate of change of charge transfer impedance is less than the activation polarization slope threshold, the current fault dominant type is determined to be reversible decay, the current fault dominant type is labeled as reversible, and the reversible decay health index is calculated according to the first preset mapping relationship based on the amount by which the rate of change of ohmic polarization impedance exceeds the ohmic polarization slope threshold. The irreversible decay health index is assigned the value of 1 minus the reversible decay health index. When the rate of change of charge transfer impedance is continuously greater than the activation polarization slope threshold and the rate of change of ohmic polarization impedance is less than or equal to the ohmic polarization slope threshold, the current fault dominant type is determined to be irreversible decay. The current fault dominant type is labeled as irreversible. Based on the extent to which the rate of change of charge transfer impedance exceeds the activation polarization slope threshold, the irreversible decay health index is calculated according to the second preset mapping relationship. The reversible decay health index is assigned the value of 1 minus the irreversible decay health index. Otherwise, keep the fault-dominant type label from the previous output unchanged.

7. The method for early warning and control of air-cooled fuel cell stack faults based on predictive models as described in claim 1, characterized in that: Step S3 specifically includes: Step S301: Obtain the current current density and the current stack temperature; Based on the reversible decay health index, the irreversible decay health index, the current current density, and the current stack temperature, an extended Kalman filter is used to establish a dual-channel predictor. The dual-channel predictor includes a short-timescale channel and a long-timescale channel; Step S302: Through a short timescale channel, perform an extended Kalman filter recursive estimation once in each impedance update cycle to obtain the predicted trajectory of the reversible declining health index. Based on the predicted trajectory, the remaining time for the reversible decline health index to reach the preset membrane dry safety health index threshold is calculated, thus obtaining the recoverable window period.

8. The method for early warning and control of air-cooled fuel cell stack faults based on predictive models as described in claim 7, characterized in that: Step S3 also includes: Step S303: Extrapolate and predict the irreversible decline health index through a long-term timescale channel to obtain the predicted trajectory of the irreversible decline health index. Based on the predicted trajectory, the remaining time for the irreversible decline health index to reach the preset end-of-life health index threshold is calculated, and the end-of-life predicted value is obtained. Obtain the fluctuation noise estimate of the short-timescale channel, and compensate the observation noise covariance matrix of the long-timescale channel based on the fluctuation noise estimate to obtain the compensated observation noise covariance matrix.

9. The method for early warning and control of air-cooled fuel cell stack faults based on predictive models as described in claim 1, characterized in that: Step S4 specifically includes: Step S401: When the fault dominant type label is reversible, output a level 1 maintenance warning and execute step S402. When the fault dominant type label is irreversible, proceed to step S403; Step S402: When the recoverable window period is greater than the preset descent threshold, output a thermal management adjustment command; When the recoverable window period is less than the preset emergency threshold, a short-term load reduction recovery command is output.

10. The method for early warning and control of air-cooled fuel cell stack faults based on predictive models as described in claim 9, characterized in that: Step S4 also includes: Step S403: When the predicted end-of-life time value is less than or equal to the preset end-of-life time threshold or the irreversible decline health index is greater than the preset irreversible health index threshold, a level three mandatory warning is triggered, and a mandatory power limiting command and a shutdown and replacement prompt command are output.

Images