Fuel cell health state prediction method, system and device based on space-time modeling and graph neural network and storage medium

Abstract

The application discloses a fuel cell health state prediction method, system and device based on space-time modeling and a graph neural network and a storage medium, and belongs to the technical field of industrial automation and energy system optimization. The application effectively overcomes the limitations of traditional data-driven and physical modeling methods in aspects of label scarcity, insufficient multi-channel correlation mining and limited nonlinear modeling capability by constructing a sparse label expansion mechanism, a multi-channel correlation feature extraction system and a nonlinear dynamic mapping strategy of a hybrid architecture, thereby providing reliable technical support for fuel cell health state evaluation and predictive maintenance, and helping to improve equipment reliability and system energy efficiency.

Description

Technical Field

[0001] This invention belongs to the field of industrial automation and energy system optimization technology, specifically relating to a method, system, device and storage medium for predicting the health status of fuel cells based on spatiotemporal modeling and graph neural networks. Background Technology

[0002] As a crucial energy conversion device in distributed integrated energy systems, fuel cells directly impact the system's energy supply efficiency and stability. However, fuel cells face significant health degradation issues during long-term operation, such as electrode active material degradation and membrane electrolyte aging, leading to power output reduction and decreased system reliability. Existing research largely focuses on health status assessment of rotating equipment (such as bearings), but research on non-rotating equipment (such as fuel cells) remains significantly insufficient.

[0003] Traditional physical models rely on mechanistic equations to describe electrochemical reaction processes, requiring precise acquisition of parameters such as membrane resistance and gas diffusion coefficient. However, in actual operation, these parameters are significantly affected by environmental factors such as temperature fluctuations and humidity changes, making it difficult to model them accurately in real time.

[0004] While data-driven methods can fit degradation and aging trends through historical operating conditions, their prediction accuracy is insufficient under complex and unsteady conditions (such as sudden load changes and aging phase transitions) due to the scarcity of tag data and limitations in sensor deployment.

[0005] Furthermore, fuel cell operating parameters are highly coupled (such as the dynamic relationship between current density and temperature and pressure), making it difficult for traditional single-point feature extraction methods to capture the implicit interaction mechanisms between multiple channels, thus limiting the ability to identify fault evolution paths.

[0006] Existing technologies face three core contradictions in assessing the health status of fuel cells: First, key degradation indicators (such as membrane impedance and electrode activity decay coefficient) lack direct measurement methods, and data tag acquisition is costly; second, the correlation between multi-source heterogeneous data generated by sensor networks is not fully explored, resulting in insufficient system-level state characterization capabilities; and third, traditional linear modeling methods under nonlinear coupling effects are difficult to adapt to complex operating conditions, limiting the model's generalization ability. Summary of the Invention

[0007] The purpose of this invention is to provide a method, system, device, and storage medium for predicting the health status of fuel cells based on spatiotemporal modeling and graph neural networks, in order to solve the technical problem that existing fuel cell health status assessments are unreliable due to the scarcity of labeled data, insufficient mining of multi-channel coupling relationships, and limited nonlinear modeling capabilities.

[0008] To achieve the above objectives, the present invention employs the following technical solution:

[0009] This invention discloses a method for predicting the health status of fuel cells based on spatiotemporal modeling and graph neural networks, comprising the following steps:

[0010] A hybrid model architecture consisting of a dual GCN layer, a frequency fusion layer, and a fully connected layer was constructed, and the extracted time step sequence features of the fuel cell sensor were used as the input data for the dual GCN layer.

[0011] The sparse impedance label data of fuel cells was extended using a Gaussian process regression model to obtain continuous impedance label data.

[0012] The electrochemical impedance data of fuel cells was optimized by using a primary-secondary separation strategy and continuous impedance tag data to obtain optimized electrochemical impedance data.

[0013] An adjacency matrix is ​​constructed based on time step sequence features, and information fusion is performed through graph convolution to obtain fused feature data.

[0014] The fused feature data is input into the frequency fusion layer and concatenated with the frequency information to form a multimodal fusion feature. The multimodal fusion feature is input into the fully connected layer, and the impedance prediction value is obtained through mapping calculation. Real-time operating data of the fuel cell and the target frequency value are input. Based on the impedance prediction value, the EIS curve is generated by traversing each target frequency value and the Nyquist plot is drawn. Combined with the optimized electrochemical impedance data, the dynamic assessment of the fuel cell health status is realized.

[0015] Furthermore, the time step sequence feature of the fuel cell sensor is extracted from the aging test data through a sliding window mechanism; each time step in the continuous time step sequence contains data from multiple effective sensor channels.

[0016] Furthermore, the specific steps for expanding the sparse impedance label data of fuel cells using a Gaussian process regression model to obtain continuous impedance label data are as follows:

[0017] For sparse impedance tag data of fuel cells, Gaussian process regression is used for expansion. The Gaussian process regression generates a continuous probability distribution function of impedance values ​​based on the mapping relationship between frequency values ​​and impedance values. When modeling, a squared exponential kernel or other non-specific type kernel function is selected to define the similarity between data points, so as to realize continuous interpolation from timestamp to impedance difference and obtain continuous impedance tag data.

[0018] Furthermore, when using the Gaussian process regression model to expand the sparse impedance label data of fuel cells, the Gaussian process regression generates a continuous probability distribution function based on the mapping relationship between frequency values ​​and impedance values, and achieves continuous interpolation from timestamps to impedance differences through a squared exponential kernel or other non-specific kernel functions; the Gaussian process regression model takes timestamps as input variables and impedance differences as output targets.

[0019] Furthermore, the primary and secondary separation strategy involves first averaging the electrochemical impedance data of the fuel cell to obtain an average impedance curve, and then separating the frequency influence and aging effect through a difference operation.

[0020] The formula for calculating the average impedance curve is as follows:

[0021]

[0022] in, and These are the real and imaginary parts of the fuel cell electrochemical impedance at a specific current ripple frequency f and operating time t, respectively. and These are the average values ​​of the real and imaginary parts of the fuel cell electrochemical impedance at a specific current ripple frequency f, respectively, and N is the total number of EIS measurements.

[0023] The formula used to separate the frequency effect and the aging effect through the difference operation is as follows:

[0024]

[0025] in, and These are the average values ​​of the real and imaginary parts of the electrochemical impedance of a fuel cell at a specific current ripple frequency f, respectively. and The real and imaginary parts of the fuel cell electrochemical impedance at specific current ripple frequency f and operating time t are compared to... and The deviation value, and These are the real and imaginary parts of the fuel cell electrochemical impedance at a specific current ripple frequency f and operating timestamp t, respectively.

[0026] Furthermore, the specific steps for constructing the adjacency matrix based on the time-step sequence features are as follows:

[0027] The Pearson correlation coefficient between the parameters of each sensor channel is calculated based on the time step sequence characteristics to quantify the degree of linear correlation between channels. Each sensor channel is a graph node, and the absolute value of the Pearson correlation coefficient is the edge weight between nodes to construct an adjacency matrix that represents the coupling relationship of multiple channels.

[0028] The adjacency matrix is ​​A; the element values ​​of the adjacency matrix A are real numbers in the range (-1, 1) calculated based on the Pearson correlation coefficient, and linear values ​​are retained.

[0029] Furthermore, the specific steps to obtain the fused feature data are as follows:

[0030] Based on the extracted time-step sequence features, the Pearson correlation coefficient between the parameters of each sensor channel is calculated to quantify the degree of linear correlation between channels. Using each sensor channel as a graph node and the absolute value of the Pearson correlation coefficient as the edge weight between nodes, an adjacency matrix representing the coupling relationship of multiple channels is constructed. This adjacency matrix and the time-step sequence features are input into a dual GCN layer, and the propagation and fusion of information from each channel are realized through graph convolution calculation, outputting the fused feature data.

[0031] The present invention also discloses a fuel cell health status prediction system based on spatiotemporal modeling and graph neural networks, comprising: a data acquisition unit, used to build a hybrid model architecture including a dual GCN layer, a frequency fusion layer and a fully connected layer, and to use the extracted time step sequence features of the fuel cell sensor as the input data of the dual GCN layer;

[0032] The data processing unit uses a Gaussian process regression model to expand the sparse impedance tag data of the fuel cell to obtain continuous impedance tag data.

[0033] The electrochemical impedance data of fuel cells was optimized by using a primary-secondary separation strategy and continuous impedance tag data to obtain optimized electrochemical impedance data.

[0034] The association analysis unit is used to construct an adjacency matrix based on time step sequence features and perform information fusion through graph convolution to obtain fused feature data.

[0035] The model prediction unit is used to input the fused feature data into the frequency fusion layer and concatenate it with the frequency information to form a multimodal fused feature; input the multimodal fused feature into the fully connected layer and obtain the impedance prediction value through mapping calculation; input the real-time operation data of the fuel cell and the target frequency value, generate the EIS curve and draw the Nyquist plot based on the impedance prediction value through each target frequency value, and realize the dynamic assessment of the health status of the fuel cell by combining the optimized electrochemical impedance data.

[0036] The present invention also discloses a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the fuel cell health status prediction method based on spatiotemporal modeling and graph neural networks as described above.

[0037] The present invention also discloses a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the fuel cell health status prediction method based on spatiotemporal modeling and graph neural networks as described above.

[0038] Compared with the prior art, the present invention has the following beneficial effects:

[0039] This invention discloses a method for predicting the health status of fuel cells based on spatiotemporal modeling and graph neural networks. It achieves sparse label expansion through a Gaussian regression model, reducing reliance on densely sampled data and improving model generalization ability under non-steady-state conditions. A system-level state representation system is constructed based on graph neural networks, overcoming the limitations of traditional single-point feature extraction methods and effectively identifying implicit interaction mechanisms and fault cascading response modes in fuel cell operation. Simultaneously, a hybrid architecture integrating graph topology and deep fully connected networks enables high-precision prediction of degradation trends under strongly nonlinear conditions, providing a reliable technical solution for fuel cell health status assessment. Attached Figure Description

[0040] Figure 1 This is a block diagram of the fuel cell health status modeling and prediction method based on spatiotemporal modeling and graph neural networks of the present invention;

[0041] Figure 2 This is an electrochemical impedance spectroscopy (EIS) curve of the fuel cell of the present invention;

[0042] Figure 3 This invention presents the sensor correlation adjacency matrix A in the form of a heatmap.

[0043] Figure 4 The results of the primary and secondary separation in the processing of fuel cell impedance dataset are as follows: a - real part of the original impedance; b - imaginary part of the original impedance; c - real part of the averaged impedance; d - imaginary part of the averaged impedance; e - real part of the impedance difference; f - imaginary part of the impedance difference.

[0044] Figure 5 This is a structural diagram of a fuel cell health status prediction model based on spatiotemporal modeling and graph neural networks. Detailed Implementation

[0045] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0046] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0047] This invention discloses a method for predicting the health status of fuel cells based on spatiotemporal modeling and graph neural networks, comprising the following steps:

[0048] a) Construct a hybrid model architecture, which includes a dual GCN layer, a frequency fusion layer, and a fully connected layer; wherein, the frequency fusion layer uses frequency information as an additional input dimension to achieve joint modeling of the relationship between frequency and channel topology;

[0049] b): Extract time-step sequence features from sensor operation data using a sliding window mechanism, and use them as model input;

[0050] c): Gaussian process regression (GPR) is used to expand sparse labeled data. The Gaussian process regression generates a continuous probability distribution function based on the mapping relationship between frequency value and impedance value, and achieves continuous interpolation from timestamp to impedance difference through squared exponential kernel or other non-specific kernel function.

[0051] d) A primary-secondary separation strategy is used to process EIS (Electrochemical Impedance Spectroscopy) curves: First, the EIS data is averaged to extract the main components. Then, the frequency effect and aging effect are separated by subtraction. The primary-secondary separation step is based on frequency. f The average impedance curve is calculated using the frequency as the independent variable, and the continuous mapping relationship from frequency to impedance value is modeled using Gaussian process regression.

[0052] e): The linear correlation between sensor channels is quantified by the Pearson correlation coefficient, an adjacency matrix is ​​constructed to characterize the multi-channel coupling relationship, and the adjacency matrix is ​​input into the dual GCN layer for information propagation and fusion;

[0053] f): Input real-time operating data and target frequency value, traverse the frequency values ​​to generate EIS curves and draw Nyquist plots, thereby realizing dynamic evaluation of fuel cell impedance characteristics.

[0054] Preferably, the sliding window mechanism in step b extracts a continuous time step sequence from the aging test data, each time step containing data from multiple effective sensor channels, and uses the time step sequence as the dimensional structure of the input matrix X.

[0055] Preferably, in step c, the Gaussian process regression model establishes an independent model for each discrete frequency point, with the timestamp as the input variable and the impedance difference as the output target, and the model fitting effect is improved through standardization.

[0056] Preferably, in step d, the frequency is based on f The formula used to calculate the average impedance curve is:

[0057]

[0058] in and These are the real and imaginary parts of the fuel cell electrochemical impedance at a specific current ripple frequency f and operating time t, respectively. and These are the average values ​​of the real and imaginary parts of the fuel cell electrochemical impedance at a specific current ripple frequency f, respectively, where N is the total number of EIS measurements; and the frequency effect and aging effect are separated by subtraction:

[0059]

[0060] in, and These are the average values ​​of the real and imaginary parts of the electrochemical impedance of a fuel cell at a specific current ripple frequency f, respectively. and The real and imaginary parts of the fuel cell electrochemical impedance at specific current ripple frequency f and operating time t are compared to... and The deviation value, and These are the real and imaginary parts of the fuel cell electrochemical impedance at a specific current ripple frequency f and operating timestamp t, respectively.

[0061] Preferably, in step e, the element values ​​of the adjacency matrix A are real numbers in the range (-1, 1) calculated based on the Pearson correlation coefficient, retaining linear values ​​to distinguish the strength of correlation between sensor channels.

[0062] Preferably, the specific implementation of generating the EIS curve in step f includes: after inputting real-time running data and target frequency values, traversing preset frequency points and calculating the corresponding real impedance ReZ and imaginary impedance ImZ, and finally drawing the Nyquist plot to characterize the electrochemical impedance characteristics of the fuel cell.

[0063] The present invention will now be described in further detail with reference to the accompanying drawings:

[0064] like Figure 1 As shown in the figure, the present invention provides a fuel cell health status modeling and prediction method based on spatiotemporal modeling and graph neural networks. This method can predict the impedance curve of the fuel cell based on real-time operating data, thereby supporting real-time assessment of the fuel cell's health status and providing guidance for predictive maintenance. This embodiment uses a proton exchange membrane fuel cell (PEMFC) as an example, and the specific steps are as follows:

[0065] Step 1: For the fuel cell device model to be evaluated, build a corresponding test platform and conduct systematic data collection and annotation to create a dataset for subsequent model training and performance evaluation. The specific implementation is as follows:

[0066] Step 1.1: Based on the fuel cell device to be evaluated, construct a hydrogen fuel cell test platform. This platform must have the ability to measure and control various physical parameters during the operation of the fuel cell to ensure the accuracy and repeatability of the experimental conditions.

[0067] The fuel cell system consists of five stacks connected in series, each stack having an active area of ​​100 cm². 2 The proton exchange membrane fuel cell (PEMFC) used achieves structural integration through a commercially available proton exchange membrane, diffusion layer, and machined flow distribution plate. Its nominal current density is set at 0.70 A / cm². 2 The maximum operating current density is 1 A / cm². 2 .

[0068] The testing platform needs to monitor and record the following core parameters in real time:

[0069] Temperature parameters: Cooling water temperature (20℃~80℃), gas temperature (20℃~80℃);

[0070] Flow parameters: Cooling water flow (0~10 L / min), hydrogen flow (0~30 L / min), air flow (0~100 L / min);

[0071] Pressure and humidity: gas pressure (0~2 bars), gas humidity (0~100% RH);

[0072] Electrical parameters: fuel cell current (0~300 A), single cell voltage (U1~U5) and total voltage (Utot).

[0073] Step 1.2: Data acquisition and annotation, including two parts: aging test and electrochemical impedance spectroscopy (EIS) measurement, to obtain the operating characteristic data of fuel cell under different operating periods.

[0074] Step 1.2.1: Aging Test: At the rated current density (0.70 A / cm²) 2 The fuel cell is continuously operated, and monitoring data during operation is collected and stored. The data recording format is an xlsx file, and the specific recording parameters are shown in Table 1.

[0075] Table 1 Aging Test Record Data

[0076]

[0077] Step 1.2.2: Electrochemical Impedance Spectroscopy (EIS) Measurement:

[0078] An EIS measurement is performed after every 160 hours of aging testing. The measurement frequency range is 50 mHz to 10 kHz. It is recommended to select 78 frequency points according to a logarithmic distribution to obtain the Nyquist plot of the fuel cell. Figure 2 As shown. The real impedance (ReZ) and imaginary impedance (ImZ) at each frequency point need to be recorded and stored as an xlsx file, containing three columns of data: frequency, ReZ, and ImZ.

[0079] Step 1.3: Based on the electrochemical reaction process of fuel cells, the following basic assumptions are established: the degradation of electrode active materials is assumed to be a slow, gradual, non-abrupt process; there may be correlations between multi-channel parameters; and the electrochemical reaction gives the fuel cell nonlinear electrical characteristics.

[0080] Step 2: Systematically process and enhance the experimental data, including three steps: primary and secondary data separation mechanism, sparse label expansion, and correlation matrix construction. The specific implementation method is as follows:

[0081] Step 2.1: Separation of Primary and Secondary Characteristics: By separating the primary components and detailed features in the EIS curve and modeling them separately using Gaussian process regression (GPR), characteristic information of the fuel cell at different levels can be extracted, such as... Figure 4 As shown. It mainly consists of the following two steps:

[0082] Step 2.1.1: Based on frequency f Using [variable name] as the independent variable, all EIS curves are averaged to obtain the impedance under the overall trend (divided into real part ReZ and imaginary part ImZ). The method for calculating the average EIS curve is as follows:

[0083]

[0084] in, and These are the real and imaginary parts of the fuel cell electrochemical impedance at a specific current ripple frequency f and operating time t, respectively. and These represent the average values ​​of the real and imaginary parts of the fuel cell electrochemical impedance at a specific current ripple frequency f, respectively, where N is the total number of EIS measurements. The average impedance curve can then be obtained. and .

[0085] Subsequently, by subtracting the EIS curves at each time point from the average curve, the effects of frequency and aging were separated:

[0086]

[0087] in, and These are the average values ​​of the real and imaginary parts of the electrochemical impedance of a fuel cell at a specific current ripple frequency f, respectively. and The real and imaginary parts of the fuel cell electrochemical impedance at specific current ripple frequency f and operating time t are compared to... and The deviation value, and These are the real and imaginary parts of the fuel cell electrochemical impedance at a specific current ripple frequency f and operating timestamp t, respectively.

[0088] Step 2.1.2: Based on the above average curve, Gaussian process regression (GPR) is used for fitting to achieve a continuous mapping from frequency to impedance value. The kernel function used is the squared exponential kernel (SE kernel), which has the following form:

[0089]

[0090] in, The GPR model is used to train and obtain or approximate the value, but when the number of t is small, To ensure convergence, we need to switch from automatic training to manual specification. The model input is frequency. f The output is the impedance value (containing both real and imaginary parts). In actual fitting, it is recommended to use standardization (such as Z-score) to improve modeling performance.

[0091] Step 2.2: Sparse Label Expansion: Based on the assumption of slowly varying fuel cell impedance, multiple independent Gaussian process regression models are constructed to achieve time continuity mapping of impedance values ​​at each frequency point. Specific method: A separate GPR model (divided into real and imaginary parts) is built for each frequency point (78 in total), implemented using a squared exponential kernel (SE kernel) to minimize noise terms. The training data consists of the impedance values ​​at that frequency point during each EIS test. The model uses timestamps as input variables and outputs the impedance difference at different measurement times for the corresponding frequency point (i.e.,...). and This generates time-continuous impedance label data. In practical modeling, it is recommended to introduce standardization to enhance model fitting.

[0092] Step 2.3: Correlation Matrix Construction: Calculate the Pearson correlation coefficient for the effective sensor channels (redundant columns may need to be removed) to generate the adjacency matrix A, as shown below. Figure 3 As shown, this adjacency matrix characterizes the correlation between channels. Note that the values ​​in this adjacency matrix range from (-1, 1) rather than being discrete, to reflect the strength of the correlation between different sensor channels; an absolute value threshold can also be set (generally recommended). r |>0.3) to define the "correlation" relationship.

[0093] Step 3: Construct a neural network prediction model to predict fuel cell impedance and guide maintenance strategies. The specific steps are as follows:

[0094] Step 3.1: Establish Figure 5 The model architecture shown consists of two GCN layers, a frequency fusion layer, and a fully connected layer.

[0095] The model input consists of an input matrix X and frequencies. f The input matrix X employs a sliding window mechanism to extract sensor data from 10 time steps of the aging test data. Each time step contains 22 valid channels (assuming the number of filtered data channels is 22); frequency f Input through the frequency fusion layer.

[0096] The model output includes the real impedance ReZ and the imaginary impedance ImZ.

[0097] The model structure is as follows:

[0098] 1. Dual GCN layers

[0099] The propagation formula is:

[0100]

[0101] Among them, W (l) For the firstl Layer trainable matrix, , , , N X is the dimension of the adjacency matrix A (equal to the number of effective channels, i.e., 22), and X is the input matrix with dimension (22×10).

[0102] 2. Frequency fusion layer

[0103] Implementation: Add a frequency scalar to the end of the vector output by the dual GCN.

[0104] 3. Fully connected layer:

[0105] The model consists of four layers: an input layer, an output layer, and two hidden layers. The data dimensions are as follows: (22×10)→1024→512→256→2. ELU is used as the activation function to enhance the model's non-linear expressiveness, and a bias term is enabled.

[0106] Step 3.2: The training process is as follows:

[0107] The input data frame used for training consists of 10 steps of 22-channel sensor aging test data and the frequency values ​​to be evaluated.

[0108] The data labels used for training are determined by both frequency values ​​and timestamps. The frequency values ​​are used to select the appropriate model from the 78 GPR model groups, while the timestamps are input into the model to obtain ReZ and ImZ training labels. Other training strategies are shown in Table 2.

[0109] Table 2 Other training strategies

[0110]

[0111] Step 4: Fuel Cell Health Prediction and Maintenance Strategy Guidance

[0112] Step 4.1: Health Status Assessment: Input real-time sensor data and target frequency. After traversing the target frequency, the ReZ-ImZ curve can be plotted, and the Nyquist plot can be generated (the result is similar to...). Figure 2 ), and examine the impedance characteristics of the fuel cell.

[0113] Step 4.2: Develop a maintenance strategy: The maintenance can be guided by the fuel cell management strategy given in Table 3 by drawing the Nyquist diagram.

[0114] Table 3 Fuel Cell Management Strategy

[0115]

[0116] An embodiment of the present invention provides a terminal device. This terminal device includes a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps in the various method embodiments described above. Alternatively, when the processor executes the computer program, it implements the functions of each module / unit in the various device embodiments described above.

[0117] The computer program can be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention.

[0118] The terminal device may be a desktop computer, laptop, handheld computer, or cloud server, etc. The terminal device may include, but is not limited to, a processor and a memory.

[0119] The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.

[0120] The memory can be used to store the computer program and / or module. The processor implements various functions of the terminal device by running or executing the computer program and / or module stored in the memory and calling the data stored in the memory.

[0121] If the modules / units integrated into the terminal device are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0122] This invention employs a primary-secondary separation strategy to process EIS curves. It extracts the main components through averaging and then separates the frequency influence and aging effect through subtraction. Gaussian process regression (GPR) is used to expand sparse labeled data, generating a continuous probability distribution function based on the mapping relationship between frequency and impedance values, and interpolating the timestamp to impedance difference. Pearson correlation coefficients are used to quantify the linear correlation between sensor channels, constructing an adjacency matrix to characterize multi-channel coupling relationships, which is then input into a dual GCN layer for information propagation and fusion. A sliding window mechanism extracts continuous time-step sequence features from sensor operating data as model input. A frequency fusion layer adds frequency information as an additional input dimension to the model output vector, which is then output as impedance values ​​via a fully connected layer. Ultimately, this invention, by constructing a sparse label expansion mechanism, a multi-channel correlation feature extraction system, and a hybrid architecture nonlinear dynamic mapping strategy, effectively overcomes the limitations of traditional data-driven and physical modeling methods in terms of label scarcity, insufficient multi-channel correlation mining, and limited nonlinear modeling capabilities. This provides reliable technical support for fuel cell health status assessment and predictive maintenance, contributing to improved equipment reliability and system energy efficiency.

[0123] The above content is only for illustrating the technical concept of the present invention and should not be construed as limiting the scope of protection of the present invention. Any modifications made to the technical solution based on the technical concept proposed in this invention shall fall within the scope of protection of the claims of this invention.

Claims

1. A method for predicting the health status of a fuel cell based on spatiotemporal modeling and graph neural networks, characterized in that, Includes the following steps: A hybrid model architecture consisting of a dual GCN layer, a frequency fusion layer, and a fully connected layer was constructed, and the extracted time step sequence features of the fuel cell sensor were used as the input data for the dual GCN layer. The sparse impedance label data of fuel cells was extended using a Gaussian process regression model to obtain continuous impedance label data. The primary and secondary separation strategy and continuous impedance tag data were used to optimize the electrochemical impedance data of fuel cells, resulting in optimized electrochemical impedance data. An adjacency matrix is ​​constructed based on time step sequence features, and information fusion is performed through graph convolution to obtain fused feature data. The fused feature data is input into the frequency fusion layer and concatenated with the frequency information to form a multimodal fusion feature. The multimodal fusion feature is input into the fully connected layer, and the impedance prediction value is obtained through mapping calculation. Real-time operating data of the fuel cell and the target frequency value are input. Based on the impedance prediction value, the EIS curve is generated by traversing each target frequency value and the Nyquist plot is drawn. Combined with the optimized electrochemical impedance data, the dynamic assessment of the fuel cell health status is realized. The specific steps for expanding the sparse impedance tag data of fuel cells to obtain continuous impedance tag data using a Gaussian process regression model are as follows: For sparse impedance tag data of fuel cells, Gaussian process regression is used to extend it. The Gaussian process regression generates a continuous probability distribution function of impedance value based on the mapping relationship between frequency value and impedance value. When modeling, the square exponential kernel or other kernel function is selected to define the similarity between data points, so as to realize continuous interpolation from timestamp to impedance difference and obtain continuous impedance tag data. When using the Gaussian process regression model to expand the sparse impedance tag data of fuel cells, the Gaussian process regression generates a continuous probability distribution function based on the mapping relationship between frequency values ​​and impedance values, and achieves continuous interpolation from timestamps to impedance differences through a squared exponential kernel or other kernel functions; the Gaussian process regression model takes timestamps as input variables and impedance differences as output targets. The primary and secondary separation strategy involves first averaging the electrochemical impedance data of the fuel cell to obtain an average impedance curve, and then separating the frequency effect and aging effect through a difference operation. The formula for calculating the average impedance curve is as follows: in, and These are the real and imaginary parts of the fuel cell electrochemical impedance at a specific current ripple frequency f and operating time t, respectively. and These are the average values ​​of the real and imaginary parts of the fuel cell electrochemical impedance at a specific current ripple frequency f, respectively, and N is the total number of EIS measurements. The formula used to separate the frequency effect and the aging effect through the difference operation is as follows: in, and These are the average values ​​of the real and imaginary parts of the electrochemical impedance of a fuel cell at a specific current ripple frequency f, respectively. and The real and imaginary parts of the fuel cell electrochemical impedance at specific current ripple frequency f and operating time t are compared to... and The deviation value, and These are the real and imaginary parts of the fuel cell electrochemical impedance at a specific current ripple frequency f and operating timestamp t, respectively.

2. The fuel cell health status prediction method based on spatiotemporal modeling and graph neural networks according to claim 1, characterized in that, The time step sequence feature of the fuel cell sensor is extracted from aging test data through a sliding window mechanism; each time step in the continuous time step sequence contains data from multiple effective sensor channels.

3. The fuel cell health status prediction method based on spatiotemporal modeling and graph neural networks according to claim 1, characterized in that, The specific steps for constructing an adjacency matrix based on time step sequence features are as follows: The Pearson correlation coefficient between the parameters of each sensor channel is calculated based on the time step sequence characteristics to quantify the degree of linear correlation between channels. Each sensor channel is a graph node, and the absolute value of the Pearson correlation coefficient is the edge weight between nodes to construct an adjacency matrix that represents the coupling relationship of multiple channels. The adjacency matrix is ​​A; the element values ​​of the adjacency matrix A are real numbers in the range (-1, 1) calculated based on the Pearson correlation coefficient, and linear values ​​are retained.

4. The fuel cell health status prediction method based on spatiotemporal modeling and graph neural networks according to claim 3, characterized in that, The specific steps to obtain the fused feature data are as follows: Based on the extracted time step sequence features, the Pearson correlation coefficient between the parameters of each sensor channel is calculated to quantify the degree of linear correlation between channels; with each sensor channel as a graph node and the absolute value of the Pearson correlation coefficient as the edge weight between nodes, an adjacency matrix representing the coupling relationship of multiple channels is constructed. The adjacency matrix and time step sequence features are input into a dual GCN layer. The propagation and fusion of information from each channel are achieved through graph convolution calculation, and the fused feature data is output.

5. A fuel cell health status prediction system based on spatiotemporal modeling and graph neural networks, characterized in that, The method for implementing any one of claims 1 to 4 includes: a data acquisition unit, used to build a hybrid model architecture including a dual GCN layer, a frequency fusion layer and a fully connected layer, and to use the extracted time step sequence features of the fuel cell sensor as input data for the dual GCN layer; The data processing unit uses a Gaussian process regression model to expand the sparse impedance tag data of the fuel cell to obtain continuous impedance tag data. The electrochemical impedance data of fuel cells was optimized by using a primary-secondary separation strategy and continuous impedance tag data to obtain optimized electrochemical impedance data. The association analysis unit is used to construct an adjacency matrix based on time step sequence features and perform information fusion through graph convolution to obtain fused feature data. The model prediction unit is used to input the fused feature data into the frequency fusion layer and concatenate it with the frequency information to form a multimodal fused feature; input the multimodal fused feature into the fully connected layer and obtain the impedance prediction value through mapping calculation; input the real-time operation data of the fuel cell and the target frequency value, generate the EIS curve and draw the Nyquist plot based on the impedance prediction value through each target frequency value, and realize the dynamic assessment of the health status of the fuel cell by combining the optimized electrochemical impedance data.

6. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the fuel cell health status prediction method based on spatiotemporal modeling and graph neural networks as described in any one of claims 1 to 4.

7. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the fuel cell health status prediction method based on spatiotemporal modeling and graph neural networks as described in any one of claims 1 to 4.

Images