Fuel cell polarization curve fitting method, device, equipment, medium and product

Abstract

The present application relates to the technical field of fuel cell performance evaluation, and discloses a fuel cell polarization curve fitting method, device, equipment, medium and product, the method comprises the following steps: obtaining fuel cell original operation data, including sample points at continuous time, each sample point includes instantaneous current and instantaneous voltage; based on the continuous current gradient change, the steady state interval of the fuel cell operation is identified; the sample points at the steady state interval corresponding time are subjected to cluster analysis, a plurality of class clusters and cluster centers are obtained; a weighted data set is constructed based on the sample points and the cluster centers in the plurality of class clusters, wherein the weight of the cluster center is greater than the weight of the sample points in the plurality of class clusters; the current-voltage change relationship is fitted according to the weighted data set, and a polarization curve is obtained. The present application can construct a polarization curve that adapts to the actual running state of a vehicle.

Description

Technical Field

[0001] This invention belongs to the field of fuel cell performance evaluation technology, and particularly relates to a method, apparatus, equipment, medium and product for fitting polarization curves of fuel cells. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] The polarization curve is the relationship between the output voltage and current of a fuel cell system. It characterizes the battery's performance under different loads and is fundamental data for evaluating fuel cell performance, health status, and degradation characteristics. Most existing methods generate polarization curves based on bench tests or idealized assumptions. These methods struggle to reflect the actual operating status and performance changes of commercially available vehicles under real-world conditions. Actual onboard data is affected by multiple factors, including variations in operating conditions, environment, load, and control strategies. This results in issues such as high noise levels and an alternating distribution of steady-state and unsteady-state regions in the raw data signal. Directly fitting the raw data signal to the polarization curve makes it difficult to accurately generate the polarization curve for the fuel cell system. Summary of the Invention

[0004] In view of this, the present invention provides a method, apparatus, device, medium and product for fitting polarization curves of fuel cells, for constructing polarization curves that are adapted to the actual operating conditions of vehicles.

[0005] One aspect of the present invention provides a method for fitting polarization curves of a fuel cell, comprising the following steps:

[0006] Acquire raw operating data of the fuel cell, including sample points at continuous time points, with each sample point including instantaneous current and instantaneous voltage;

[0007] Based on continuous changes in current gradient, the steady-state range of the fuel cell operation is identified;

[0008] Cluster analysis was performed on the sample points at the corresponding time points in the steady-state interval to obtain multiple clusters and cluster centers;

[0009] A weighted dataset is constructed based on sample points and cluster centers in the multiple clusters, wherein the weight of the cluster centers is greater than the weight of the sample points in the multiple clusters;

[0010] The polarization curve is obtained by fitting the current-voltage relationship based on the weighted dataset.

[0011] Furthermore, after acquiring the raw operating data of the fuel cell, the raw operating data is preprocessed:

[0012] The raw operating data is effectively filtered to remove data from non-operating states of the fuel cell;

[0013] Perform time alignment and correction on the filtered valid data;

[0014] The corrected instantaneous current sequence data is then subjected to smoothing filtering.

[0015] Furthermore, identifying the steady-state operating range of the fuel cell includes:

[0016] The instantaneous current gradient sequence at consecutive moments is calculated based on absolute difference.

[0017] Based on a time window of preset length, a moving average is performed on the instantaneous current gradient sequence to obtain the average current gradient for each time window.

[0018] The time window in which the average current gradient is less than a set threshold is marked as a candidate steady-state interval;

[0019] Candidate steady-state intervals with a duration greater than a set threshold are selected as formal steady-state intervals.

[0020] In some embodiments, unsupervised clustering algorithms are used to perform clustering analysis on sample points at corresponding times in the steady-state interval.

[0021] In some embodiments, after obtaining multiple clusters, the average current and average voltage of the sample points in each cluster are calculated by minute to obtain an average sample point set, and a weighted dataset is constructed based on the average sample point set and the cluster centers.

[0022] In some embodiments, fitting the current-voltage change relationship based on the weighted dataset includes:

[0023] Based on the weighted dataset, the current-voltage relationship is fitted using a linear polynomial of a set order;

[0024] The weighted error function is calculated based on the weighted sum of the deviations between the actual and fitted values ​​of the samples.

[0025] With the minimum weighted error function as the optimization objective, the optimal parameters of the linear polynomial are solved to obtain the optimized polarization curve;

[0026] The optimized polarization curve is subjected to electrochemical consistency test and coefficient of determination test. If either test fails, the order of the linear polynomial is updated and the fitting is performed again.

[0027] A second aspect of the present invention provides a fuel cell polarization curve fitting device, comprising:

[0028] The data acquisition module is configured to acquire raw operating data of the fuel cell, including sample points at continuous time points, each sample point including instantaneous current and instantaneous voltage;

[0029] The steady-state range determination module is configured to identify the steady-state range of the fuel cell operation based on continuous current gradient changes.

[0030] The clustering analysis module is configured to perform clustering analysis on sample points at corresponding times in the steady-state interval, and obtain multiple clusters and cluster centers;

[0031] The dataset construction module is configured to construct a weighted dataset based on sample points and cluster centers in the plurality of clusters, wherein the weight of the cluster centers is greater than the weight of the sample points in the plurality of clusters;

[0032] The polarization curve fitting module is configured to fit a polarization curve based on the weighted dataset.

[0033] A third aspect of the present invention provides an electronic device comprising one or more processors and a memory; wherein the memory stores one or more computer programs, the one or more computer programs comprising instructions that, when executed by the electronic device, cause the electronic device to perform the method.

[0034] A fourth aspect of the present invention provides a computer-readable storage medium storing instructions that, when executed on an electronic device, cause the electronic device to perform the method.

[0035] A fifth aspect of the present invention provides a computer program product comprising a computer program that, when executed by a processor, implements the method described herein.

[0036] One or more of the above technical solutions collect raw operating data from actual vehicle operating conditions. By screening steady-state intervals based on current gradient changes and performing cluster analysis on sample points within the steady-state intervals, abnormal data points can be effectively eliminated. At the same time, the types of steady-state operating conditions can be distinguished, and cluster centers that can reflect the typical electrical characteristics of similar steady-state operating conditions can be obtained. During the fitting process, by assigning higher weights to cluster centers and lower weights to sample points within clusters, the current-voltage relationship can be fitted, enabling the polarization curve to approximate these highly representative samples as a whole. In addition, by combining cluster sample points to supplement operating condition coverage, the accuracy and adaptability of the fitting results can be effectively balanced, and finally, a polarization curve that can reflect the true current-voltage characteristics of the fuel cell can be obtained. Attached Figure Description

[0037] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0038] Figure 1 A flowchart of the fuel cell polarization curve fitting method provided in an embodiment of this application is shown;

[0039] Figure 2 The diagram shows the program module architecture of the fuel cell polarization curve fitting device provided in this application embodiment. Detailed Implementation

[0040] Embodiments of this application will now be described in more detail with reference to the accompanying drawings. While some embodiments of this application are shown in the drawings, it should be understood that this application can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this application. It should be understood that the drawings and embodiments of this application are for illustrative purposes only and are not intended to limit the scope of protection of this application.

[0041] In the description of the embodiments of this application, the term "comprising" and similar terms should be understood as open-ended inclusion, i.e., "including but not limited to". The term "based on" should be understood as "at least partially based on".

[0042] As described in the background section, most existing methods generate polarization curves based on bench testing or idealized assumptions. For example, US Patent 8214174B2 and Chinese Patent CN104051755B both rely on empirical formulas for idealized polarization curves, calculating the curves by fitting and updating the formula coefficients. However, in actual vehicle operation, multiple factors such as changes in operating conditions, environment, load, and control strategies intertwine, causing significant differences between the characteristics of onboard data and data under ideal bench conditions. This makes it difficult for these methods to reflect the operating status and performance changes of market-ready vehicles under actual operating conditions.

[0043] Meanwhile, even when attempting to directly fit the data using onboard data, the inherent defects in the raw data signals make it difficult to accurately generate the polarization curve of the fuel cell system. The raw data not only contains short-term interference and outliers from the data transmission process, but more importantly, it is mixed with a large amount of unsteady-state data generated during cold starts, hot starts, shutdowns, and load changes, which is mixed with steady-state data such as cruise data. Existing technologies lack effective mechanisms to address this. For example, US8214174B2 directly uses raw current and voltage data for model parameter estimation without performing steady-state identification and unsteady-state removal; CN104051755B filters cold-start data, but can only eliminate this type of unsteady-state interference; and CN120245819B only uses threshold screening preprocessing for current, voltage, and power, failing to precisely remove unsteady-state fluctuations and noise under complex operating conditions. These problems all lead to significant fluctuations in fitting accuracy when directly using raw data, such as distortion in the first segment of the curve when data is scarce in the low-current activation region.

[0044] To address the aforementioned issues, one or more embodiments of the present invention provide a method for reconstructing the polarization curve of a fuel cell based on real on-board operating data. Using the original on-board operating data as a foundation, the method eliminates outlier data through steady-state interval identification and cluster analysis to obtain real steady-state characteristic data. Furthermore, it extracts high-confidence cluster centers to construct a weighted dataset, enabling the polarization curve fitting process to prioritize the core operating point while also considering the overall operating trend, ultimately generating a polarization curve that adapts to the characteristics of the actual vehicle.

[0045] Figure 1 A flowchart illustrating an example method for fitting polarization curves in a fuel cell is shown, including the following steps:

[0046] S101. Obtain raw operating data of the fuel cell, including sample points at continuous time points, each sample point including instantaneous current and instantaneous voltage.

[0047] S102. Based on the continuous change of current gradient, identify the steady-state range of the fuel cell operation, filter out the effective data segments with stable current, and eliminate non-steady-state data interference caused by current fluctuations.

[0048] S103. Perform cluster analysis on the sample points corresponding to the steady-state interval to obtain multiple clusters and cluster centers. Eliminate voltage anomalies and discrete noise in the steady-state interval and extract high-confidence cluster centers that represent typical steady-state conditions.

[0049] S104. Construct a weighted dataset based on the sample points and cluster centers in the multiple clusters, wherein the weight of the cluster center is greater than the weight of the sample points in the multiple clusters. Through weight differentiation design, the contribution of high-confidence cluster centers to the fitting results is strengthened, while taking into account the working condition coverage of the cluster sample points, thus balancing the fitting accuracy and adaptability.

[0050] S105. Fit the current-voltage change relationship based on the weighted dataset to obtain the polarization curve.

[0051] The above technical solution collects raw operating data from actual vehicle conditions. By filtering steady-state intervals based on current gradient changes and performing cluster analysis on sample points within these intervals, it effectively eliminates outlier data points. Simultaneously, it distinguishes between different types of steady-state operating conditions, obtaining cluster centers that reflect the typical electrical characteristics of similar steady-state conditions. During the fitting process, by assigning higher weights to cluster centers and lower weights to sample points within clusters, the current-voltage relationship is fitted, allowing the polarization curve to approximate these highly representative samples as a whole. Furthermore, by combining cluster sample points to supplement operating condition coverage, the accuracy and adaptability of the fitting results are effectively balanced, ultimately yielding a polarization curve that reflects the true current-voltage characteristics of the fuel cell.

[0052] In step S101, the raw operating data of the fuel cell is collected through an on-board remote data transmission system (such as a vehicle networking platform), including timestamps, instantaneous current, and instantaneous voltage. In some embodiments, raw operating data of the fuel cell from multiple vehicles can be obtained through the on-board remote data transmission system, and timestamps, instantaneous current, and instantaneous voltage can be obtained for each vehicle. At the same time, vehicle information tags are used to distinguish the raw operating data of the multiple vehicles.

[0053] After acquiring the raw runtime data, preprocessing is performed, including effective data filtering, time synchronization from multiple data sources, and filtering, to output a raw time-series dataset in a unified format. The specific process is as follows:

[0054] (1) Valid data screening: The original operating data is screened to remove data from non-operating states such as fuel cell system startup, standby, and purging, retaining only valid data that shows continuous normal operation after startup. The normal operating state of the fuel cell system is confirmed by the condition "State=State_running". By removing non-operating state data through valid data screening, invalid data from special stages such as startup and standby can be avoided from interfering with the fitting baseline, ensuring that all data used in the fitting originates from the normal operating state of the fuel cell.

[0055] (2) Timing consistency correction: Time alignment and correction are performed on the selected valid data to ensure the timing consistency between current and voltage, laying the foundation for subsequent collaborative analysis.

[0056] (3) Current signal filtering: Smoothing filtering is performed on the corrected instantaneous current sequence data. Based on the current data characteristics of different vehicles, an appropriate filtering algorithm can be selected, such as median filtering, moving average filtering, Kalman filtering, and Savgol filtering. Through filtering, abnormal points and short-term interference generated during data transmission can be eliminated, and purified current data can be output, providing high-quality input for subsequent analysis.

[0057] In step S102, using the preprocessed dataset as input, a combination of sliding window and threshold judgment is employed to identify time periods when the fuel cell current is relatively stable during vehicle operation. This provides highly reliable steady-state data for subsequent cluster analysis and polarization curve fitting. The identification method is as follows:

[0058] S1021. For the preprocessed filtered current data, calculate the instantaneous current gradient sequence at consecutive time points based on absolute difference. Let the filtered current sequence be... (Sampling time is) The instantaneous current gradient between adjacent sampling points is defined as follows:

[0059]

[0060] in, Sampling points The instantaneous current gradient at a given point.

[0061] S1022. Based on a preset time window, perform a moving average on the instantaneous current gradient sequence to obtain the average current gradient for each time window. Based on a preset time window of length W (e.g., 5 seconds, which can be adjusted according to actual needs), perform a moving average on the instantaneous current gradient to calculate the smoothed average current gradient ΔI(t). k The formula is:

[0062]

[0063] In the formula, W is the length of the time window (which can be expressed as the number of seconds or the corresponding number of sampling points). It can effectively reduce the impact of a single outlier on gradient determination and improve the stability of gradient calculation.

[0064] S1023. Mark the time window where the average current gradient is less than a set threshold as a candidate steady-state interval. Then, smooth the average current gradient... Compare with a preset threshold ε, if within a certain time window If the current change is less than ε, the current change within the window is determined to be stable, and this window segment is marked as a candidate steady-state interval.

[0065] S1024. Select candidate steady-state intervals whose duration exceeds a set threshold as official steady-state intervals. To eliminate misjudgments caused by short-term fluctuations, the temporal continuity of the candidate steady-state intervals needs to be verified. Let the timestamp of the start time of the candidate steady-state interval be... The end time timestamp is Calculate its continuous duration and compared with the preset continuous steady-state determination threshold. Compare them.

[0066] like If so, the candidate segment is designated as the formal steady-state current region; if If the current-voltage relationship is not stable, it is still considered to be in a non-steady-state region. This step can effectively eliminate instantaneous fluctuations during sudden load changes, operating condition switching, and current-voltage mismatch caused by system response lag, ensuring that the identified steady-state region has reliable electrochemical characteristics.

[0067] The process for determining the formal steady-state interval described above can be summarized as follows:

[0068]

[0069] Instantaneous current and voltage data at each moment within the formal steady-state range are used as input data for the next stage of cluster analysis.

[0070] The above process, through verification of current gradient changes and duration, achieves steady-state range identification based on current stability, effectively eliminating non-steady-state data interference caused by current fluctuations. However, it should be noted that even when the current is in a stable state, the fuel cell may still experience voltage surges or abnormal drifts due to factors such as local reaction anomalies, instantaneous sensor errors, and brief fluctuations in hydrogen / air supply. Directly using these data points containing voltage anomalies for polarization curve fitting will still lead to local distortion of the curve.

[0071] In step S103, unsupervised clustering is performed on the instantaneous current-instantaneous voltage pairs corresponding to the steady-state interval. As an example, based on the DBSCAN clustering algorithm, unsupervised clustering is performed on the second-level current-voltage pairs within the steady-state interval to obtain multiple clusters, and the cluster centers are calculated. The clustering process is as follows:

[0072] Step S1031: Based on the instantaneous current and instantaneous voltage at the corresponding time in each steady-state interval, construct a second-level current-voltage set to form the input matrix X.

[0073]

[0074] The input matrix X contains N sample points, with each row corresponding to a sample point at a given time step. The current-voltage data at each time step is represented as follows: , Current values ​​in the order of seconds. The voltage value is on the order of seconds.

[0075] Step S1032: Based on a pre-defined neighborhood size, perform unsupervised clustering on the input matrix to obtain multiple clusters. The unsupervised clustering process is as follows:

[0076] For each sample point Find all sample points that are within a neighborhood of a given size. For any sample point... The formula for its neighborhood is expressed as:

[0077]

[0078]

[0079] in, Represents sample points The neighborhood of ε, that is, all sample points within the range of Euclidean distance ε. Let ε represent any sample point in the input matrix X, where ε is the neighborhood radius. Represents sample points and sample points The Euclidean distance between them. Representing the neighborhood The more sample points in a neighborhood, the higher the sample density in that region. This represents the threshold for the minimum number of neighborhood points. Then the sample points The core point is denoted as the core point, and other sample points in the neighborhood are denoted as reachable points.

[0080] The core point and all reachable points within its neighborhood are grouped into the same cluster. Then, starting from other core points within that cluster, the neighborhood is recursively expanded until all associative sample points are included in the cluster, ultimately resulting in K clusters. Sample points not associated with any cluster are considered noise points or marginal points and are removed. Noise points are isolated points with a sample number in their neighborhood far below a threshold, such as outliers from instantaneous voltage fluctuations. Marginal points are discrete points with a small number of samples in their neighborhood but cannot be associated with the core point, such as transitional data from steady-state to transient states.

[0081] The formula for calculating the cluster center of the kth cluster is:

[0082]

[0083] in, Current-voltage pairs representing cluster centers , This represents the total number of sample points in this cluster.

[0084] Cluster analysis can automatically remove outliers far from the cluster centers, further improving data quality. At the same time, it can distinguish the main steady-state operating conditions of fuel cells and extract the most representative current-voltage feature points by calculating the cluster centers, providing high-quality data support for subsequent polarization curve fitting.

[0085] By performing filtering preprocessing on the data, identifying steady-state intervals based on current stability, and clustering analysis on current-voltage sample points within the steady-state intervals, a layer-by-layer screening of the original operating data was achieved. This retained only stable current-voltage sample points with steady-state characteristics that reflect the true performance of the system, effectively improving the representativeness and statistical stability of the input data. This approach abandons the traditional method of directly modeling based on the entire dataset, significantly reducing the impact of random fluctuations in non-steady-state data on the polarization curve morphology, thus ensuring the reliability and stability of subsequent analyses.

[0086] In step S104, based on the clustering results, the average current and average voltage are calculated minute by minute for continuous data within each sustained steady-state interval, resulting in an average sample point set. This average sample point set and the cluster centers are then integrated into a weighted dataset. For example, averaging the current and voltage over every 60 consecutive seconds yields the average current and average voltage for each minute of steady-state data, representing local operating characteristics. The cluster center set is defined as follows: The set of minute-level average points is as follows: , Where M is the total number of sample points in each cluster. To balance data confidence and coverage, weights are assigned to the cluster centers and the average number of sample points, respectively. and ,and .

[0087] Finally, a unified current-voltage pair dataset with weighted coefficients is constructed:

[0088]

[0089] In step S105, a polynomial is used to fit the current-voltage relationship to characterize the changing characteristics of the polarization curve. The fitting result is then verified. If the verification fails, the polynomial order is adjusted and the fitting is repeated. Specifically, this includes the following steps:

[0090] Step S1051: Based on the weighted dataset, fit the current-voltage relationship using a linear polynomial of a set order. To characterize the current-voltage relationship in the polarization curve, a d-order polynomial is introduced. The formula used to fit the polarization curve is:

[0091]

[0092] in Let I represent the current in the polarization curve, and d ≤ 5. As an example, the initial value of the linear polynomial order d is set to 3.

[0093] Step S1052: Calculate the weighted error function based on the weighted sum of the deviations between the actual sample values ​​and the fitted values. Specifically, to balance the fitting effect between high-confidence samples and the overall trend, weighted least squares (WLS) is used to estimate the parameter vector θ. Specifically, the weighted error function is defined... for:

[0094]

[0095] in, Take the current value Fitted values ​​at time, and These represent the weights of the cluster center and the minute-level average sample points, respectively.

[0096] Step S1053: Taking the minimization of the weighted error function as the optimization objective, solve for the optimal parameters of the linear polynomial to obtain the optimized polarization curve. Specifically, the optimal parameters are solved using the weighted least squares method. Make the weighted error function To achieve the minimum, the solution formula is as follows:

[0097]

[0098] The I matrix is ​​a feature matrix composed of the current values ​​of the cluster centers and the minute-averaged sample points, and its expression is shown below:

[0099]

[0100] The first K rows represent the current at each cluster center, and the last J rows represent the current of the sample points on a minute-by-minute basis, where n = K + J. The voltage vector V corresponds to the voltage value of each sample point in the feature matrix I, and their arrangement order is consistent with that in matrix I. The expression for V is shown below:

[0101]

[0102] The weight matrix W is a diagonal matrix used to distinguish the contributions of different samples (cluster centers and minute-averaged sample points) in weighted least squares fitting, and its expression is shown below:

[0103]

[0104] The optimal parameters after solving Substituting these values ​​into the polynomial formula yields the polarization curves that reflect the relationship between current and voltage for the current dataset.

[0105] Step S1054: Perform electrochemical consistency test and coefficient of determination test on the optimized polarization curve. If either test fails, update the order of the linear polynomial and refit.

[0106] The electrochemical consistency test verifies whether the polarization curve satisfies the fundamental electrochemical law that voltage monotonically decreases with current in classical polarization curves. Specifically, it verifies whether the slope of the optimized polarization curve at any point is less than 0. If so, the electrochemical consistency test is passed; otherwise, it is failed. That is, within the current range of the entire polarization curve, for any current point I, if the following formula is satisfied, the electrochemical consistency test is passed:

[0107]

[0108] If the electrochemical consistency test fails, the order of the polynomial is automatically changed, and the process returns to step S1051 until a result conforming to basic electrochemical laws is obtained. This avoids the distortion of the first segment of the curve that may be caused by the scarcity of data in the low-current activation region.

[0109] Furthermore, the optimized polarization curves were subjected to a coefficient of determination (R²) test, which measures the degree to which the fitted formula interprets the actual voltage-current data. The calculation formula is as follows:

[0110]

[0111] in, This represents the actual voltage of the i-th sample point in the weighted dataset. This represents the average actual voltage across all sample points in the weighted dataset. This represents the fitted value of the polarization curve when the current value of the i-th sample point is taken after optimization. When R² > 0.95, it indicates that the model can reflect the characteristics of the actual current-voltage polarization change well, has statistical significance, and the fitting result is effective. If R² ≤ 0.95, it is considered that the fitting accuracy is insufficient, the order of the polynomial is automatically changed, and the process returns to step S1051 until the result meets the requirements of the coefficient of determination test.

[0112] By characterizing the current-voltage relationship using a polynomial model and constructing an error function using weighted least squares, high weights are assigned to cluster centers and low weights to minute-level averages. This ensures that the fitting process prioritizes high-confidence core steady-state operating points while also considering the overall trend of the data across all operating conditions, effectively improving the accuracy of the polarization curve in representing the real operating characteristics of fuel cells. Furthermore, by verifying whether the fitted curve conforms to electrochemical characteristics in the activation, ohmic, and concentration regions, fitting results that violate fundamental electrochemical principles can be avoided. Introducing a coefficient of determination test to measure the model's interpretability of actual current-voltage data ensures fitting accuracy. When either the electrochemical characteristic test or the coefficient of determination test fails, the polynomial order is automatically adjusted for refitting, guaranteeing the scientific validity and reliability of the polarization curve. This approach does not rely on specific idealized polarization curve empirical formulas and requires no manual intervention to optimize model parameters and iterate the order. It enables continuous, monotonic, and interpretable polarization curve reconstruction across the entire current range, thus adapting to the operating data characteristics of different fuel cell vehicle models. Meanwhile, since the original operating data is directly collected from the actual operating conditions of the vehicle, it covers the current-voltage response characteristics of the fuel cell in real application scenarios, making the fitted curve more suitable for the actual operating state of the vehicle.

[0113] The method further includes step S106, which outputs the final polarization curve and related statistical results. Specifically, a visualization interface or chart can be provided to allow engineers to intuitively assess the performance and degradation trend of the fuel cell.

[0114] Experimental verification:

[0115] To verify the effectiveness of the method in improving the accuracy of polarization curve fitting, actual vehicle operating data from 200 hydrogen fuel cell systems of different models and fleets were used. Statistical analysis showed that the polarization curves reconstructed using the method of this invention for steady-state data extraction and cluster analysis generally had a coefficient of determination greater than 0.95 (R² > 0.95), indicating a very good fit and statistical significance. In contrast, traditional methods, which only eliminated abnormal operating states of the fuel cell engine such as start-up, standby, and purging, without further screening of stable current-voltage data points, generally had a coefficient of determination less than 0.8 (R² < 0.8), with some vehicles even showing R² < 0.3. This demonstrates that the traditional method lacks sufficient fitting accuracy and may even fail statistical testing.

[0116] Using data from a randomly selected vehicle as an example, we compare the fitting effects of fuel cell polarization curve fitting after different data processing steps. Most existing technologies only remove abnormal operating states of the fuel cell engine, such as start-up, standby, and purging, retaining only valid data from continuous normal operation after start-up. However, this data has a high degree of dispersion, R0 2=0.36, the root mean square error (RMSE) between the fitted value and the actual value is 22.3V. Further steady-state interval identification is performed on the effective data. Fitting a curve based on the effective data within the steady-state interval removes most of the deviation values ​​that are far from the fitted curve, thus improving the fitting result R0. 2 There was some improvement, and RMES also decreased further, but there were still current-voltage points that were significantly far from the fitted curve. Further clustering analysis was performed on the effective data in the steady-state region, and fitting was performed based on the method of this invention, resulting in a significant improvement in fitting accuracy. R0 2 =0.96, root mean square error RMES=3.5V.

[0117] Furthermore, after removing non-normal operating conditions, the total number of remaining valid data points was 41,316 (1 point corresponds to 1 second). After further processing with steady-state interval identification and cluster analysis using the method of this invention, the total number of remaining valid data points was 34,831, meaning only about 16% of the data was removed, but the fitting results were significantly improved. It can be seen that the data processing steps proposed in this invention can serve as reliable basic data for health status assessment and performance degradation analysis, ensuring that the model output not only accurately reflects the actual operating performance of the fuel cell system but also possesses consistency and engineering application value across vehicle models and scenarios.

[0118] Since the same vehicle does not experience significant performance degradation within a single day, the rationality can be verified by comparing the polarization curves from two days prior and subsequent days. Real-world operating data from 100 fuel cell vehicles were selected, and the polarization curve results generated over two days were compared. Based on the method of this invention, the voltage difference between the polarization curves from two days prior and subsequent days was generally less than 2%. Based on the traditional method, the voltage difference between the polarization curves from two days prior and subsequent days was mostly greater than 5%, with some vehicles experiencing a voltage difference of up to 30%. The fact that the same vehicle does not experience such significant performance degradation within a single day indicates that the traditional method has a significant systematic error. This is because past methods heavily relied on data from the activation region, but many vehicles have very little or no activation region data, which is entirely non-steady-state data. Substituting the coefficients of the activation region calculated from this data into the model leads to systematic bias in the calculated results.

[0119] The polarization curve achieved by this invention not only highly conforms to the electrochemical laws of fuel cells, but also has good adaptability and robustness to different vehicle types and operating conditions (including city buses, heavy trucks, light trucks and tractors, etc.). It can maintain stable curve reconstruction performance under different control strategies, temperature and humidity conditions and load fluctuations, providing a reliable data foundation for health status assessment and performance degradation analysis.

[0120] The above one or more embodiments provide a method for constructing polarization curves reflecting the current performance of fuel cells based on real on-board operating data. By constructing a comprehensive process including data preprocessing, stable current-voltage sample point screening, polarization curve fitting, and verification, representative stable current-voltage sample points can be accurately extracted based on real vehicle operating data from the market without additional experiments or sensors. This generates polarization curves that accurately reflect the current performance of fuel cell systems in market-sold vehicles. Furthermore, it effectively solves the problem of data non-steadiness caused by environmental interference, changes in operating conditions, and load variations affecting the accuracy of polarization curve fitting. It is adaptable to various vehicle types (including tractors, buses, heavy trucks, and light trucks) and diverse operating conditions (such as frequent start-stop, hilly road conditions, urban congestion, and highway cruising), exhibiting strong robustness, versatility, and practical value. It can provide real-time and accurate polarization curves for the health management, life prediction, and activation strategies of various fuel cell models, serving as crucial data support.

[0121] Based on the above method, one or more embodiments of the present invention also provide a fuel cell polarization curve fitting device, such as... Figure 2 As shown, the system includes: a running data acquisition module 201, configured to acquire raw operating data of the fuel cell, including sample points at continuous time intervals, each sample point including instantaneous current and instantaneous voltage; a steady-state interval determination module 202, configured to identify the steady-state interval of the fuel cell operation based on continuous current gradient changes; a clustering analysis module 203, configured to perform clustering analysis on the sample points at corresponding time intervals in the steady-state interval to obtain multiple clusters and cluster centers; a dataset construction module 204, configured to construct a weighted dataset based on the sample points and cluster centers in the multiple clusters, wherein the weight of the cluster centers is greater than the weight of the sample points in the multiple clusters; and a polarization curve fitting module 205, configured to fit a polarization curve based on the weighted dataset.

[0122] One or more embodiments of the present invention also provide an electronic device that can be used to implement the methods in the above embodiments. The electronic device includes one or more processors, one or more memories coupled to the processors, and a communication module coupled to the processors.

[0123] The memory in this embodiment of the invention is used to store various types of data to support, for example... Figure 1 The execution of the method shown.

[0124] It is understood that the memory can be volatile memory or non-volatile memory, or it can include both volatile and non-volatile memory. The memory in this embodiment of the invention is capable of storing, for example... Figure 1The computer programs corresponding to each step in the method shown are as follows. The operating system contains various system programs, such as the framework layer, core library layer, and driver layer, used to implement various basic business functions and handle hardware-based tasks. Application programs can contain various other applications.

[0125] As an example, a processor can be an integrated circuit chip with signal processing capabilities, such as a general-purpose processor, a digital signal processor (DSP), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc., where a general-purpose processor can be a microprocessor or any conventional processor, etc.

[0126] Specifically, according to embodiments of this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program including functions for executing... Figure 1 The program code for the method shown. In such an embodiment, the computer program can be downloaded and installed from a network via a communication component, and / or installed from a removable medium. When the computer program is executed by the central processing unit, it performs the various functions defined in the apparatus of this application.

[0127] in, Figure 1 The computer program instructions corresponding to the method shown may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to operate 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 the process. Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0128] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for fitting polarization curves of a fuel cell, characterized in that, Includes the following steps: Acquire raw operating data of the fuel cell, including sample points at continuous time points, with each sample point including instantaneous current and instantaneous voltage; The instantaneous current gradient sequence at consecutive moments is calculated based on absolute difference. Based on a time window of preset length, a moving average is performed on the instantaneous current gradient sequence to obtain the average current gradient for each time window. The time window in which the average current gradient is less than a set threshold is marked as a candidate steady-state interval; Candidate steady-state intervals whose duration exceeds a set threshold are selected as formal steady-state intervals. Cluster analysis was performed on the sample points corresponding to the formal steady-state interval to obtain multiple clusters and cluster centers; sample points not associated with any cluster were considered noise points or marginal points and were removed. For each cluster, the average current and average voltage are calculated per minute for the sample points to obtain an average sample point set. A weighted dataset is then constructed based on the average sample point set and the cluster centers; wherein the weight of the cluster centers is greater than the weight of the sample points in the multiple clusters. The polarization curve is obtained by fitting the current-voltage relationship based on the weighted dataset.

2. The fuel cell polarization curve fitting method as described in claim 1, characterized in that, After acquiring the raw operating data of the fuel cell, the raw operating data is further preprocessed: The raw operating data is effectively filtered to remove data from non-operating states of the fuel cell; Perform time alignment and correction on the filtered valid data; The corrected instantaneous current sequence data is then subjected to smoothing filtering.

3. The fuel cell polarization curve fitting method as described in claim 1, characterized in that, Cluster analysis is performed on sample points at corresponding times in the steady-state interval based on an unsupervised clustering algorithm.

4. The fuel cell polarization curve fitting method as described in claim 1, characterized in that, Fitting the current-voltage relationship based on the weighted dataset includes: Based on the weighted dataset, the current-voltage relationship is fitted using a linear polynomial of a set order; The weighted error function is calculated based on the weighted sum of the deviations between the actual and fitted values ​​of the samples. With the minimum weighted error function as the optimization objective, the optimal parameters of the linear polynomial are solved to obtain the optimized polarization curve; The optimized polarization curve is subjected to electrochemical consistency test and coefficient of determination test. If either test fails, the order of the linear polynomial is updated and the fitting is performed again.

5. A fuel cell polarization curve fitting device, characterized in that, include: The data acquisition module is configured to acquire raw operating data of the fuel cell, including sample points at continuous time points, each sample point including instantaneous current and instantaneous voltage; The steady-state interval determination module is configured to calculate the instantaneous current gradient sequence of consecutive time moments based on absolute difference; perform a moving average on the instantaneous current gradient sequence based on a preset time window to obtain the average current gradient of each time window; mark the time window with the average current gradient less than a set threshold as a candidate steady-state interval; and select the candidate steady-state interval with a duration greater than the set threshold as the formal steady-state interval. The clustering analysis module is configured to perform clustering analysis on the sample points corresponding to the formal steady-state interval, and obtain multiple clusters and cluster centers; sample points that are not associated with any cluster are considered noise points or marginal points and are removed. The dataset construction module is configured to calculate the average current and average voltage of the sample points within each cluster by minute to obtain an average sample point set, and construct a weighted dataset based on the average sample point set and cluster centers; wherein the weight of the cluster centers is greater than the weight of the sample points in the multiple clusters; The polarization curve fitting module is configured to fit a polarization curve based on the weighted dataset.

6. An electronic device, the electronic device comprising one or more processors; and a memory; wherein, The memory stores one or more computer programs, the one or more computer programs including instructions, characterized in that, when the instructions are executed by the electronic device, the electronic device performs the method according to any one of claims 1-4.

7. A computer-readable storage medium storing instructions, characterized in that, When the instructions are executed on an electronic device, the electronic device causes the electronic device to perform the method according to any one of claims 1-4.

8. A computer program product, the computer program product comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1-4.

Images