Air pump pressure and flow dynamic matching adaptive data processing method and system

Abstract

This invention provides an adaptive data processing method and system for dynamic matching of gas pump pressure and flow rate, belonging to the field of data processing technology. The method includes: acquiring the power generation demand signal from the fuel cell controller, the pressure signal at the gas pump outlet, flow rate data, and the fuel cell stack inlet pressure feedback signal; performing time-domain alignment and scaling normalization on the power generation demand signal, gas pump outlet pressure signal, flow rate data, and fuel cell stack inlet pressure feedback signal to obtain a multi-dimensional operating condition dataset; organizing the multi-dimensional operating condition dataset into a high-dimensional state space of gas supply dynamics; and performing unsupervised clustering analysis on the high-dimensional state space to identify characteristic regions of multiple different gas path operating modes. This invention achieves dynamic and accurate matching of pressure and flow rate.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to an adaptive data processing method and system for dynamic matching of air pump pressure and flow. Background Technology

[0002] In practical applications such as hydrogen commercial vehicles and distributed fuel cell power generation, the energy conversion efficiency and operational stability of proton exchange membrane fuel cells depend on the precise control of the hydrogen supply system. As a key device for hydrogen delivery, the dynamic adaptability of the pump's output pressure and flow rate directly determines the sufficiency of electrochemical reactions within the fuel cell stack. In practical applications, it is necessary to continuously address complex challenges such as frequent fluctuations in power generation demand and dynamic changes in gas path resistance over operating time.

[0003] The hydrogen supply system of a certain fuel cell application project adopts a traditional control scheme based on offline calibration operating condition curves. By pre-setting fixed parameters for pump speed and valve opening corresponding to different power ranges, the hydrogen pressure and flow rate are initially regulated. However, the technology has a defect: the control logic only relies on the preset static operating condition mapping relationship and does not consider the sudden fluctuations in power demand and the dynamic changes in the gas path environment during real-time operation. As a result, the matching response of pump pressure and flow rate cannot keep up with the rhythm of operating condition changes, which in turn affects the stability of the electrochemical reaction of the fuel cell. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide an adaptive data processing method and system for dynamic matching of air pump pressure and flow rate, so as to achieve dynamic and accurate matching of pressure and flow rate.

[0005] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows: Firstly, an adaptive data processing method for dynamically matching air pump pressure and flow rate, the method comprising: The system collects the power demand signal from the fuel cell controller, the pressure signal from the gas pump outlet, the flow data, and the pressure feedback signal from the fuel cell stack inlet. The power generation demand signal, the gas pump outlet pressure signal, the flow data and the battery stack inlet pressure feedback signal are processed by time domain alignment and scaling normalization to obtain a multi-dimensional operating condition dataset. Based on the multi-dimensional operating condition dataset, the multi-dimensional operating condition dataset is organized into a high-dimensional state space of gas supply dynamics; unsupervised clustering analysis is performed on the high-dimensional state space to identify the characteristic regions of multiple different gas supply operation modes. Based on the high-dimensional data point set corresponding to the feature region, the average coordinates of all high-dimensional data points in the feature region are calculated as the feature centroid. A reasonable data radiation range is set with the feature centroid as the center, and valid data points within the reasonable data radiation range are retained. By extracting the statistical distribution characteristics of effective data points, and based on the statistical distribution characteristics and the modal self-correction factors corresponding to each feature region, all modal self-correction factors are fused to obtain the target hydrogen pressure setpoint and target hydrogen flow rate setpoint that match the real-time operating conditions. Based on the target hydrogen pressure setpoint and the target hydrogen flow rate setpoint, the corresponding target speed command of the gas pump motor and the target opening command of the hydrogen supply valve are obtained by analysis. The speed control command of the gas pump motor and the opening control command of the supply valve are sent to the corresponding control components to coordinate the adjustment of the speed of the gas pump motor and the opening of the supply valve, so as to realize the closed-loop dynamic matching control of the hydrogen supply pressure and flow rate.

[0006] Furthermore, the power generation demand signal, the gas pump outlet pressure signal, the flow data, and the battery stack inlet pressure feedback signal are subjected to time-domain alignment and scaling normalization to obtain a multi-dimensional operating condition dataset, including: It receives and aggregates raw time-series data streams of power generation demand signals, air pump outlet pressure signals, flow data, and battery stack inlet pressure feedback signals from different sources. Based on the power demand signal, and based on the power step change point preset in the power demand signal, the time-series data of the air pump outlet pressure signal, flow data and battery stack inlet pressure feedback signal are time-stamp matched and interpolated to obtain a time-domain aligned set of synchronization signals. For each dimension of the synchronization signal set, the maximum value, minimum value and statistical distribution within the preset historical time window are analyzed to dynamically determine the range of signal feature values ​​used for normalization processing. Based on the range of signal characteristic values, the real-time values ​​of each dimension of the signal are linearly mapped to a unified numerical range, resulting in a scaled multidimensional signal sequence. By organizing and storing scaled, multi-dimensional signal sequences in chronological order, a multi-dimensional working condition dataset is obtained.

[0007] Furthermore, based on the multi-dimensional operating condition dataset, the dataset is organized into a high-dimensional state space representing the dynamics of the gas supply. Unsupervised clustering analysis is performed on this high-dimensional state space to identify characteristic regions of multiple different gas supply operation modes, including: From the multi-dimensional operating condition dataset, sample points of consecutive time intervals are extracted in chronological order; the multi-dimensional data of each sample point are combined to construct a high-dimensional state vector of the gas supply dynamics at the corresponding time. The high-dimensional state vectors are arranged and organized in time order to form a high-dimensional state space for the gas supply dynamics. For high-dimensional state vectors in a high-dimensional state space, a preset dimensionality reduction algorithm is applied to process them in order to reduce the data dimensionality and extract the main features, thereby obtaining the corresponding set of low-dimensional feature vectors. For the set of low-dimensional feature vectors, a pre-defined unsupervised clustering algorithm is applied for analysis, and data points with similar operating characteristics are classified into the same cluster. Each cluster corresponds to an identified gas path operating mode cluster, so as to obtain the clustering analysis results. Based on the cluster analysis results, the high-dimensional state vector in the high-dimensional state space corresponding to each cluster is defined and marked as an independent feature region, thus completing the identification of feature regions for multiple different gas path operation modes.

[0008] Furthermore, based on the high-dimensional data point set corresponding to the feature region, the average coordinates of all high-dimensional data points within the feature region are calculated as the feature centroid. A reasonable data radiation range is set with the feature centroid as the center, and valid data points within the reasonable data radiation range are retained, including: Based on each identified feature region, extract all the original high-dimensional state vectors defined and labeled by each feature region to form a high-dimensional data point set corresponding to each feature region; Based on a high-dimensional data point set, the arithmetic mean of all high-dimensional data points in the set is calculated in each dimension, and the arithmetic mean is combined to form the feature centroid coordinates of each feature region. Centered on the feature centroid coordinates, and based on the distribution statistics of all points in the corresponding high-dimensional data point set, a multi-dimensional spatial data radiation range boundary is set for each feature region. Based on the corresponding feature centroid coordinates, the distance from each high-dimensional data point in the high-dimensional data point set to the feature centroid is calculated one by one. Based on the set data radiation range boundary, the distance from the point to the feature centroid is compared and filtered, and data points with a distance less than the corresponding boundary are retained as valid data points within the feature region.

[0009] Furthermore, by extracting the statistical distribution characteristics of effective data points, and based on the statistical distribution characteristics and the modal self-correction factors corresponding to each feature region, all modal self-correction factors are fused to obtain the target hydrogen pressure setpoint and target hydrogen flow rate setpoint that match the real-time operating conditions, including: For each feature region's effective data point set, extract the statistical distribution characteristics of all data points in the set in terms of pressure and flow. The statistical distribution characteristics include at least the mean, variance, and covariance. Based on the statistical distribution characteristics of each feature region, and combined with the preset performance optimization objective function, the modal self-correction factor corresponding to each feature region is calculated and obtained. The current working condition multidimensional data is obtained after processing. The working condition multidimensional data is matched with the identified feature regions to determine the target feature region to which the current working condition belongs. The modal self-correction factor corresponding to the target feature region is called, and the target hydrogen pressure setpoint and target hydrogen flow setpoint are dynamically matched with the current real-time operating conditions by fusion calculation based on real-time operating data.

[0010] Furthermore, based on the target hydrogen pressure setpoint and the target hydrogen flow rate setpoint, the corresponding target speed command for the gas pump motor and the target opening command for the hydrogen supply valve are obtained through analysis, including: Using the target hydrogen pressure setpoint and the target hydrogen flow rate setpoint as joint input conditions, a preliminary target speed command for the gas pump motor that meets the joint conditions is obtained by parsing. Using the target hydrogen pressure setpoint, the target hydrogen flow rate setpoint, and the initial target speed command of the gas pump motor as inputs, the target opening command of the hydrogen supply valve that meets the target supply conditions is obtained through collaborative analysis. Based on the target opening command of the hydrogen supply valve, the initial target speed command of the gas pump motor is dynamically compensated and finely adjusted to finally determine the target speed command of the gas pump motor.

[0011] Furthermore, the gas pump motor speed control command and the supply valve opening control command are sent to the corresponding control components to coordinately adjust the gas pump motor speed and the supply valve opening, thereby achieving closed-loop dynamic matching control of hydrogen supply pressure and flow rate, including: The target speed command for the air pump motor and the target opening command for the hydrogen supply valve are sent to the air pump motor driver and the hydrogen supply valve controller respectively via the communication link, so as to drive the corresponding components to perform adjustment. After the command is sent and the components are adjusted, the current air pump outlet pressure signal, flow data and battery stack inlet pressure feedback signal are collected in real time to obtain the actual feedback signal; The actual feedback signal is compared with the target hydrogen pressure setpoint and the target hydrogen flow rate setpoint to obtain the pressure and flow closed-loop deviation signal between the actual supply state and the target state. Based on the closed-loop deviation signal, the operating condition matching, modal self-correction factor call and setpoint fusion calculation are re-triggered and executed to obtain the adjusted new control command, thereby realizing the closed-loop dynamic matching control of hydrogen supply pressure and flow.

[0012] Secondly, the adaptive data processing system for dynamic matching of air pump pressure and flow includes: The data acquisition module is used to acquire the power generation demand signal of the fuel cell controller, the pressure signal at the outlet of the gas pump, the flow data, and the pressure feedback signal at the inlet of the fuel cell stack. The alignment module is used to perform time-domain alignment and scaling normalization on the power generation demand signal, the gas pump outlet pressure signal, the flow data and the battery stack inlet pressure feedback signal to obtain a multi-dimensional operating condition dataset. The module is used to organize the multi-dimensional operating condition dataset into a high-dimensional state space of gas supply dynamics based on the multi-dimensional operating condition dataset; and to perform unsupervised clustering analysis on the high-dimensional state space to identify the feature regions of multiple different gas path operating modes. The optimization module is used to calculate the average coordinates of all high-dimensional data points in the feature region based on the high-dimensional data point set corresponding to the feature region, and use the feature centroid as the center to set a reasonable data radiation range, and retain the effective data points within the reasonable data radiation range. The fusion module is used to extract the statistical distribution characteristics of effective data points, and based on the statistical distribution characteristics and the modal self-correction factors corresponding to each feature region, fuse all modal self-correction factors to obtain the target hydrogen pressure setpoint and target hydrogen flow setpoint that match the real-time operating conditions. The analysis module, based on the target hydrogen pressure setpoint and the target hydrogen flow rate setpoint, analyzes and obtains the corresponding target speed command of the gas pump motor and the target opening command of the hydrogen supply valve. The matching module sends the gas pump motor speed control command and the supply valve opening control command to the corresponding control components to coordinate the adjustment of the gas pump motor speed and the supply valve opening, thereby achieving closed-loop dynamic matching control of hydrogen supply pressure and flow.

[0013] Thirdly, a computing device, comprising: One or more processors; A storage device for storing one or more programs that, when executed by one or more processors, cause the one or more processors to implement the method.

[0014] Fourthly, a computer-readable storage medium storing a program that, when executed by a processor, implements the method.

[0015] The above-described solution of the present invention has at least the following beneficial effects: Because it employs a full-link technology approach, including multi-dimensional operating condition signal acquisition, time-domain alignment and scaling normalization preprocessing, high-dimensional state space construction and unsupervised clustering mode identification, feature centroid screening for effective data, mode self-correction factor fusion to generate target setpoints, collaborative analysis of control commands, and closed-loop dynamic adjustment, it overcomes the technical problems of traditional control schemes based on offline calibration operating condition curves that rely solely on static mapping relationships and cannot respond to sudden fluctuations in power demand and dynamic changes in the gas path environment during real-time operation. This results in lag in the response of gas pump pressure and flow matching and poor adaptability. It achieves dynamic and accurate matching of gas pump pressure and flow with real-time operating conditions, improves the adaptability and response speed of the fuel cell hydrogen supply system, ensures the stability of electrochemical reactions within the fuel cell stack, and enhances the energy conversion efficiency and service life of the fuel cell system. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the adaptive data processing method for dynamic matching of air pump pressure and flow provided in an embodiment of the present invention.

[0017] Figure 2 This is a schematic diagram of an adaptive data processing system for dynamic matching of air pump pressure and flow provided in an embodiment of the present invention. Detailed Implementation

[0018] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0019] like Figure 1 As shown, embodiments of the present invention propose an adaptive data processing method for dynamic matching of air pump pressure and flow rate, the method comprising the following steps: Step 1: Collect the power demand signal from the fuel cell controller, the pressure signal from the gas pump outlet, the flow data, and the pressure feedback signal from the fuel cell stack inlet. Step 2: Perform time-domain alignment and scaling normalization on the power generation demand signal, air pump outlet pressure signal, flow data and battery stack inlet pressure feedback signal to obtain a multi-dimensional operating condition dataset. Step 3: Based on the multi-dimensional operating condition dataset, organize the multi-dimensional operating condition dataset into a high-dimensional state space of gas supply dynamics; perform unsupervised clustering analysis on the high-dimensional state space to identify the characteristic regions of multiple different gas path operating modes. Step 4: Based on the high-dimensional data point set corresponding to the feature region, calculate the average coordinates of all high-dimensional data points in the feature region as the feature centroid, set a reasonable data radiation range with the feature centroid as the center, and retain the effective data points within the reasonable data radiation range. Step 5: By extracting the statistical distribution characteristics of effective data points, and based on the statistical distribution characteristics and the modal self-correction factors corresponding to each feature region, all modal self-correction factors are fused to obtain the target hydrogen pressure setpoint and target hydrogen flow rate setpoint that match the real-time operating conditions. Step 6: Based on the target hydrogen pressure setpoint and the target hydrogen flow rate setpoint, the corresponding target speed command of the gas pump motor and the target opening command of the hydrogen supply valve are obtained by parsing. Step 7: Send the gas pump motor speed control command and the supply valve opening control command to the corresponding control components to coordinate the adjustment of the gas pump motor speed and the supply valve opening, so as to realize the closed-loop dynamic matching control of hydrogen supply pressure and flow rate.

[0020] In this embodiment of the invention, the integrated technical means of acquiring multi-dimensional core operating condition signals, performing time-domain alignment and scaling normalization preprocessing, constructing a high-dimensional state-space model and identifying gas path operating modes through unsupervised clustering, screening effective data points and extracting statistical features to calculate modal self-correction factors, fusing to obtain target setpoints, analyzing collaborative control commands, and implementing closed-loop dynamic adjustment overcomes the technical problems of traditional control schemes that rely on offline calibration of static mapping relationships, cannot adapt to real-time power demand fluctuations and gas path state changes, and cause lag in the matching response of gas pump pressure and flow. Thus, it achieves the technical effects of dynamic and accurate matching of gas pump pressure and flow with real-time operating conditions, improving the adaptability and response speed of the hydrogen supply system, ensuring sufficient and stable electrochemical reaction of fuel cells, and optimizing system energy conversion efficiency.

[0021] In a preferred embodiment of the present invention, step 1 above may include: Step 1 involves acquiring the power demand signal from the fuel cell controller, the pressure signal at the gas pump outlet, flow data, and the pressure feedback signal at the fuel cell stack inlet. Specifically, this includes establishing a full-link data acquisition channel using dedicated data acquisition equipment compatible with the fuel cell controller, gas pump outlet monitoring module, flow detection device, and fuel cell stack inlet pressure sensing unit, ensuring the stability of the acquisition process and the integrity of signal transmission. For the power demand signal from the fuel cell controller, the dynamic power command output by the controller is received in real time. This command directly reflects the current power load demand of the fuel cell. The acquisition frequency is kept consistent with the controller's output frequency to avoid signal loss or delay. For the pressure signal at the gas pump outlet, a high-precision pressure sensor installed at a designated location in the gas pump outlet pipeline continuously senses the instantaneous changes in hydrogen pressure within the gas path. The sensor acquisition frequency is set to capture the minimum interval required to detect pressure fluctuations, ensuring the real-time nature and accuracy of the pressure data. For the flow data, a flow detection device integrated into the hydrogen delivery pipeline measures the flow rate per unit time in real time. The detection device for hydrogen flow rate in the pipeline needs to adapt to the dynamic range of flow rate changes within the gas path to ensure data measurement accuracy in different flow ranges. For the battery stack inlet pressure feedback signal, the actual pressure value before hydrogen enters the battery stack is collected in real time by a pressure sensing unit deployed at the battery stack inlet. The feedback signal directly reflects the final effect of hydrogen supply. During the acquisition process, it maintains time synchronization with other signals. All signal acquisition processes adopt continuous sampling to avoid the loss of operating condition information caused by intermittent acquisition. All types of raw signals acquired are transmitted to the data processing unit in real time through a dedicated transmission line, and anti-interference processing is performed on the signals during the transmission process.

[0022] In this embodiment of the invention, by employing the technical means of accurately acquiring four types of core operating parameters—the power generation demand signal of the fuel cell controller, the outlet pressure signal of the gas pump, the flow data, and the inlet pressure feedback signal of the fuel cell stack—the technical problem of traditional data acquisition being limited by single signal dimensions or missing key parameters, and thus failing to fully reflect the matching relationship between hydrogen supply status and power generation demand, is overcome. This achieves the technical effect of providing complete and core basic data support for time-domain alignment, modal identification, and dynamic matching control, ensuring the pertinence and reliability of data processing and control strategy formulation.

[0023] In a preferred embodiment of the present invention, step 2 above may include: Step 2.1: Receive and aggregate raw time-series data streams from different signal sources, including power demand signals, gas pump outlet pressure signals, flow data, and fuel cell stack inlet pressure feedback signals. Specifically, this involves connecting to four different signal sources: the fuel cell controller, the gas pump outlet pressure monitoring module, the hydrogen flow detection device, and the fuel cell stack inlet pressure sensing unit. The system receives raw time-series data streams transmitted from each signal source in real time. These streams include the power demand signal data stream sent by the fuel cell controller, the pressure signal data stream collected from the gas pump outlet pressure, the flow data stream recorded by the hydrogen flow detection device, and the fuel cell stack inlet pressure feedback signal. The pressure feedback signal data stream from the sensing unit continuously monitors the data transmission status during the reception of each raw timing data stream. By verifying the data frame format and checking the data transmission timing, it checks and filters out abnormal data such as packet loss, code errors, and data corruption that may occur during transmission, ensuring the integrity and initial validity of each raw timing data stream. After all valid raw timing data streams have been received, the data streams from different sources are aggregated to a unified data processing center and clearly classified and labeled according to signal type. The signal source, signal type, and transmission start time of each data stream are recorded in detail.

[0024] Step 2.2: Using the power generation demand signal as a reference, and based on the preset power step change points of the power demand signal, perform time-stamp matching and interpolation synchronization on the time-series data of the gas pump outlet pressure signal, flow data, and battery stack inlet pressure feedback signal to obtain a time-domain aligned synchronization signal set. Specifically, this includes: determining the power generation demand signal as the time-domain aligned reference signal, because the reference signal directly reflects the current power generation load demand of the fuel cell and is the core basis for driving the hydrogen supply system to adjust pressure and flow. The rhythm of change determines the adjustment direction of the gas path operation state. By using data analysis tools to analyze the time-series change pattern of the power generation demand signal in advance, accurately identify the power step change points in the signal where the power suddenly rises or falls, and accurately record the precise timestamp corresponding to each power step change point. To establish a unified reference standard, timestamp information was extracted from the time-series data of the air pump outlet pressure signal, flow data, and battery stack inlet pressure feedback signal. These timestamps were then compared one by one with the reference timestamp. For data segments in the air pump outlet pressure signal, flow data, and battery stack inlet pressure feedback signal where the timestamp did not coincide with the reference timestamp or where there was a time delay, a linear interpolation method was used. Based on the values ​​and time intervals of the adjacent valid data points before and after the data segment, signal values ​​corresponding to the reference timestamp were calculated and supplemented to fill the data gaps. After timestamp matching and interpolation supplementation, it was ensured that the air pump outlet pressure signal, flow data, battery stack inlet pressure feedback signal, and power generation demand signal had valid data corresponding to each key time node and regular time scale, ultimately resulting in a set of synchronization signals that were completely aligned in the time domain.

[0025] Step 2.3 involves analyzing the maximum, minimum, and statistical distribution of each dimension of the synchronization signal set within a preset historical time window, dynamically determining the range of signal characteristic values ​​used for normalization processing. Specifically, this includes configuring a fixed-duration preset historical time window for each dimension of the signal set, encompassing the power generation demand signal, gas pump outlet pressure signal, flow data, and battery stack inlet pressure feedback signal. The duration of the preset historical time window must fully cover one operating cycle of the fuel cell under typical operating conditions to ensure that the complete change characteristics of the signal in that dimension over a period of time can be captured. For each dimension of the signal, a comprehensive statistical analysis is performed on all included signal data within its corresponding preset historical time window to accurately calculate the maximum and minimum values ​​of the signal values ​​within that time period. Simultaneously, by analyzing the frequency of data distribution and the degree of data dispersion in different value intervals, the statistical distribution characteristics of the signal in that dimension within the historical time window are fully understood. Based on the calculated maximum and minimum values ​​and statistical distribution characteristics, the actual effective range of change of the signal in this dimension is comprehensively judged, and the range of signal characteristic values ​​that can fully cover the numerical fluctuations that the signal may occur in the current period and over a period of time is dynamically determined, so as to avoid the misjudgment or truncation of some effective signal data due to the use of a fixed and unchanging range of characteristic values.

[0026] Step 2.4: Based on the signal feature value range, the real-time values ​​of each dimension signal are linearly mapped to a unified numerical interval to obtain a scaled multidimensional signal sequence. Specifically, this includes: selecting a unified numerical interval applicable to all dimensions based on the unique signal feature value range determined for each dimension signal. This numerical interval must have good numerical compatibility and computational convenience, and be able to meet the computational requirements of subsequent high-dimensional state space modeling and cluster analysis. For each real-time signal value in the four dimensions of power generation demand signal, air pump outlet pressure signal, flow data, and battery stack inlet pressure feedback signal, a linear mapping transformation method is used to accurately transform the corresponding original signal feature value range to the preset unified numerical interval. During the mapping transformation process, the relative magnitude relationship and change trend between the values ​​within each dimension signal are strictly maintained, without changing the operating condition change information contained in the signal itself. Through this linear mapping process, the four dimensions of signals, which originally had different dimensions and large differences in numerical span, are all converted into scaled and dimensionless standardized data, ultimately obtaining a multidimensional signal sequence composed of these four standardized data.

[0027] Step 2.5 involves structurally organizing and storing the scaled multidimensional signal sequences in chronological order to obtain a multidimensional operating condition dataset. Specifically, this includes arranging the scaled multidimensional signal sequences sequentially according to the original acquisition time of all signal data, ensuring that each time point corresponds to a complete set of standardized data encompassing four dimensions: power generation demand, gas pump outlet pressure, flow rate, and battery stack inlet pressure. A structured data organization method is used to establish a clear data index for each set of data, clearly labeling the timestamp, meaning, and correspondence of each dimension, ensuring good readability and traceability. The organized multidimensional signal sequences are then stored in a dedicated data storage unit. This unit must possess high-speed read / write capabilities and stable storage performance, supporting rapid data retrieval and batch processing during subsequent high-dimensional state-space model construction. Through structured organization and storage operations, a complete, standardized, and high-quality multidimensional operating condition dataset is obtained. This dataset can be directly used for high-dimensional state-space construction of gas supply dynamics and identification and analysis of gas supply operating modes.

[0028] In this embodiment of the invention, by employing techniques such as aggregating original time-series data from multiple sources, using power generation demand signals as a benchmark to achieve timestamp matching and interpolation synchronization of multiple signals based on power step change points, dynamically determining the normalization range based on a preset historical time window, linearly mapping each dimension of signals to a unified interval, and structured storage, the technical problems of poor data adaptability and low modeling accuracy caused by timestamp deviations, inconsistent dimensions, and fixed normalization ranges of different source signals in traditional data processing are overcome. This achieves the technical effect of obtaining a multi-dimensional operating condition dataset that is time-domain aligned, scale-consistent, and dimensionless, providing high-quality data support for the construction of high-dimensional state-space models and the identification of operating modes, and improving the reliability of overall data processing and control strategies.

[0029] In a preferred embodiment of the present invention, step 3 above may include: Step 3.1: Extract sample points from the multi-dimensional operating condition dataset in chronological order, representing consecutive time intervals. Combine the multi-dimensional data of each sample point to construct a high-dimensional state vector representing the gas supply dynamics at the corresponding moment. This includes: defining the complete data range included in the multi-dimensional operating condition dataset; setting reasonable consecutive time intervals based on the frequency and characteristics of operating condition changes during fuel cell operation; ensuring that the time intervals can fully capture short-term operating condition fluctuations while avoiding data redundancy due to excessively short intervals; extracting sample points from the multi-dimensional operating condition dataset in chronological order according to the set time intervals; ensuring that each sample point fully contains valid data in four dimensions: normalized power demand signal, gas pump outlet pressure signal, flow data, and battery stack inlet pressure feedback signal; integrating the four dimensions of data for each extracted sample point in a fixed order; and merging the multi-dimensional information reflecting the gas supply state at the same moment into a whole to construct a high-dimensional state vector that comprehensively represents the complete operating state of the gas supply dynamics at that moment, ensuring that the data in each dimension of the vector accurately corresponds to the key parameters of gas supply operation.

[0030] Step 3.2: Arrange the high-dimensional state vectors in chronological order and organize them into a high-dimensional state space for the gas supply dynamics. Specifically, this includes: arranging all the constructed high-dimensional state vectors in strict accordance with the time sequence of their corresponding sample points, with each time node corresponding to a unique high-dimensional state vector. Using time as the vertical axis and the parameters of each dimension of the high-dimensional state vector as the horizontal axis, all the arranged high-dimensional state vectors are systematically organized into a structured high-dimensional state space. The high-dimensional state space can fully present the trajectory of the gas supply system's operating state changes at different time nodes, clearly reflecting the impact of factors such as fluctuations in power generation demand and changes in gas supply resistance on the gas supply dynamics.

[0031] Step 3.3: Apply a preset dimensionality reduction algorithm to the high-dimensional state vectors in the high-dimensional state space to reduce the data dimensionality and extract the main features, thereby obtaining the corresponding set of low-dimensional feature vectors. Specifically, this includes processing a large number of high-dimensional state vectors contained in the constructed high-dimensional state space. Each of these high-dimensional state vectors integrates data from four dimensions: power generation demand, gas pump outlet pressure, flow rate, and battery stack inlet pressure, which fully reflects the gas path state. However, there is significant information overlap between different dimensions. Directly using these vectors for subsequent analysis would lead to problems such as excessive redundant information, long computation time, and interference with clustering accuracy. Given the characteristics of multi-dimensional linear correlation and the coexistence of core features and redundant information in working condition data, dimensionality reduction algorithms are a common solution for processing multi-dimensional working condition data in engineering. The core advantage is that it can efficiently remove redundant information while preserving the core change patterns of the data. Before applying principal component analysis, all high-dimensional state vectors are preprocessed in a targeted manner: Since principal component analysis is sensitive to data distribution, the vector data of each dimension must first be centered to eliminate the influence of differences in numerical benchmarks of different dimensions on the algorithm. At the same time, outliers are checked and removed again to ensure the consistency of input data. This step is the basis for accurate feature extraction by principal component analysis and avoids outlier data from interfering with the calculation of principal components.

[0032] The core processing logic of principal component analysis (PCA) is to transform multiple linearly correlated original dimensions into a few linearly independent principal components through mathematical transformation. Principal components are essentially linear combinations of the original parameters, capable of condensing key information scattered across different dimensions. During processing, the covariance matrix of the high-dimensional state vector is first calculated. eigenvalues ​​and eigenvectors are obtained through matrix decomposition. The magnitude of the eigenvalue represents the amount of original information contained in the corresponding principal component. Considering the engineering requirements of gas path operation, we set a variance contribution rate threshold, retaining principal components with high eigenvalue rankings and cumulative variance contribution rates reaching the threshold, while discarding minor components with small variances. For example, after processing, the original 4-dimensional data may retain 2 to 3 principal components. One principal component can cover most of the information on changes in operating conditions. These principal components constitute the dimensions of the low-dimensional feature vector. Through such principal component analysis, each high-dimensional state vector is transformed into a low-dimensional feature vector. The information that was originally scattered in the four dimensions of power, pressure, flow rate, and inlet pressure is condensed into a few principal components. Redundant correlations such as the inevitable increase in flow rate when power increases are removed. The core features of the gas path, such as the pressure response speed when power fluctuates suddenly and the matching relationship between flow rate and inlet pressure under stable operating conditions, are fully preserved. The final set of low-dimensional feature vectors has a significantly reduced dimensionality, which simplifies the computation of unsupervised clustering and makes modal identification more efficient.

[0033] Step 3.4 involves analyzing the low-dimensional feature vector set using a pre-defined unsupervised clustering algorithm. Data points with similar operational characteristics are grouped into the same cluster, with each cluster corresponding to an identified gas path operation mode cluster to obtain clustering analysis results. Specifically, for the low-dimensional feature vector set, considering the core requirement of automatically distinguishing diverse operating conditions and the lack of pre-defined category labels for gas path operation modes, density clustering is adopted as the pre-defined unsupervised clustering algorithm. The core advantage of unsupervised clustering algorithms is that they do not require pre-setting the number of clusters, can automatically divide clusters by identifying the density distribution of data points, and can effectively remove isolated outlier data points. It is particularly suitable for the distribution characteristics of dense data in stable operating conditions, scattered data in transitional operating conditions, and isolated data in occasional outlier conditions in gas path operation. Compared with clustering algorithms such as K-means clustering, which require a pre-defined number of clusters, density clustering does not require manual intervention in the number of operating mode modes, and can more realistically restore the natural state division of gas path operation, avoiding mode identification bias caused by improper pre-defined cluster numbers. Before applying the density clustering algorithm, two key parameters need to be determined based on the distribution characteristics of the low-dimensional feature vectors. The neighborhood radius and minimum sample size are specifically set as follows: By calculating the pairwise distances between all low-dimensional feature vectors and statistically analyzing the concentrated intervals of the distance distribution, the maximum distance of the majority vector cluster area is set as the neighborhood radius, ensuring that vectors of the same operating condition mode can be included in the same neighborhood. The average distribution density of low-dimensional feature vectors within the neighborhood radius is used as a reference to set the minimum sample size, ensuring that only a sufficiently representative set of vectors can obtain an effective cluster. During algorithm execution, each vector in the low-dimensional feature vector set is traversed, and the neighborhood range is defined based on the set neighborhood radius, with each vector as the center. The number of other vectors contained in each neighborhood is counted. If the number of vectors contained in the neighborhood of a certain vector reaches or exceeds the minimum sample size, then the vector is determined to be a core point. Core points represent the typical characteristics of a certain type of operating condition mode. For example, vectors in a low-power stable operating state will form a dense cluster of core points. Based on the core points, all vectors in the neighborhood are merged into an initial cluster. Then, through the neighborhood expansion of the core points, the clusters of adjacent core points are merged to form a complete cluster structure. Vectors whose number of samples in their neighborhood is less than the minimum number of samples and are not covered by the neighborhood of any core point are identified as noise points.

[0034] The core of cluster analysis is the similarity of the gas path operation features represented by the low-dimensional feature vectors. The similarity is quantified by calculating the Euclidean distance between the vectors: the closer the distance, the more similar the corresponding gas path operation states. The similar operating characteristics here specifically include: key indicators such as the trend of power demand change, the matching relationship between pressure and flow, and the stability of gas path operation condensed in the low-dimensional feature vectors. All low-dimensional feature vectors corresponding to low power demand, stable pressure and flow, and small fluctuation amplitude will be grouped into the same cluster due to their close proximity, and the cluster corresponds to the low-power stable operating mode. All vectors corresponding to high power demand, synchronous pressure and flow fluctuation, and large fluctuation amplitude will obtain another cluster, corresponding to the high-power fluctuating operating mode. The vectors of the start-stop transition operating mode will obtain relatively dispersed but consistent core features due to the gradual change of operating state from start-up to stability. Through the complete execution of the density clustering algorithm, the set of low-dimensional feature vectors is finally divided into multiple non-overlapping clusters. Each cluster corresponds to a gas path operating mode that has been accurately identified. At the same time, meaningless noise data is removed, and clustering analysis results that can clearly distinguish different gas path operating modes are obtained.

[0035] Step 3.5: Based on the clustering analysis results, define and mark the high-dimensional state vectors in the high-dimensional state space corresponding to each cluster as an independent feature region, thus completing the identification of feature regions for multiple different gas path operation modes. Specifically, this includes: based on the clustering analysis results, tracing all high-dimensional state vectors in the original high-dimensional state space corresponding to each data cluster; analyzing the distribution range and feature boundaries of high-dimensional state vectors within the same cluster; combining the physical characteristics and operating condition change patterns of the gas path operation; defining the specific distribution area of ​​each data cluster in the high-dimensional state space; clearly marking each defined distribution area; assigning it a corresponding gas path operation mode identifier; and defining it as an independent feature region. Through this series of operations, the accurate identification of feature regions for multiple different gas path operation modes is completed, and each feature region corresponds to a specific gas path operation state.

[0036] In this embodiment of the invention, the technical means of extracting sample points at equal time intervals from a multi-dimensional operating condition dataset to construct a high-dimensional state vector, organizing it into a high-dimensional state space in chronological order, extracting core features through a dimensionality reduction algorithm, and then classifying similar operating feature data points using an unsupervised clustering algorithm, and finally defining and marking the high-dimensional vectors corresponding to each cluster as independent feature regions, overcomes the technical problems of traditional control schemes that rely on static mapping, cannot accurately distinguish complex operating states of the gas path, and are difficult to adapt to dynamic changes in operating conditions. Thus, it achieves the technical effects of clearly identifying multiple gas path operating mode feature regions, providing accurate modal basis for targeted control strategy formulation, and improving the matching adaptability of air pump pressure and flow.

[0037] In a preferred embodiment of the present invention, step 4 above may include: Step 4.1: Based on each identified feature region, extract all original high-dimensional state vectors defined and labeled by each feature region to form a high-dimensional data point set corresponding to each feature region. Specifically, this includes: first, retrieving the completed gas path operation mode feature region identification results, which contain the high-dimensional state space range and labeling information corresponding to each feature region; then, matching the boundary range of each feature region with the original high-dimensional state vectors in the high-dimensional state space through data index association to identify all original high-dimensional state vectors contained in each feature region; for each feature region, automatically extracting all original high-dimensional state vectors that meet its boundary definition conditions. These vectors all correspond to the actual operating condition data of the same type of gas path operation mode; classifying and integrating the extracted vectors according to feature regions to construct a dedicated high-dimensional data point set for each feature region; and adding corresponding mode identifiers to each data point set to ensure that data from different modes are not confused.

[0038] Step 4.2: Based on the high-dimensional data point set, calculate the arithmetic mean of all high-dimensional data points in the set in each dimension, and combine the arithmetic means to form the feature centroid coordinates of each feature region. Specifically, this includes: for each constructed high-dimensional data point set, clarifying the dimensions contained in each high-dimensional state vector in the data point set, namely, power generation demand, pump outlet pressure, flow rate, and battery stack inlet pressure, and calculating the arithmetic mean for each dimension: taking the pump outlet pressure dimension as an example, extracting the pump outlet pressure values ​​of all high-dimensional state vectors in the data point set, and then... The numerical values ​​are summed and then divided by the total number of vectors in the data point set to obtain the arithmetic mean of that dimension. Following the same method, the arithmetic mean of the power generation demand dimension, the flow rate dimension, and the battery stack inlet pressure dimension are calculated in sequence. After completing the calculation of all dimensions, the arithmetic mean of the four dimensions is combined in a fixed order of power generation demand, air pump outlet pressure, flow rate, and battery stack inlet pressure to obtain a complete multidimensional coordinate. This multidimensional coordinate is the characteristic centroid coordinate of the corresponding characteristic region. The coordinate can accurately represent the typical state of the gas path operation mode corresponding to the characteristic region.

[0039] Step 4.3: Using the feature centroid coordinates as the center and based on the distribution statistics of all points in the corresponding high-dimensional data point set, set a multi-dimensional spatial data radiation range boundary for each feature region. Based on the corresponding feature centroid coordinates, calculate the distance from each high-dimensional data point in the high-dimensional data point set to the feature centroid one by one. Specifically, using the feature centroid coordinates as the central reference point of the multi-dimensional space, first analyze the distribution characteristics of all data points in the corresponding high-dimensional data point set: calculate the deviation of the data point in each dimension from the mean of the centroid of that dimension, statistically analyze the dispersion of the data in each dimension, and analyze the aggregation of multi-dimensional data points in space. For example, data points of low-power stable modes will be closely distributed around the centroid, while data points of transition modes will be relatively loosely clustered. This determines the overall range of data points distributed around the centroid. Combining the requirements for data validity in engineering practice, set the allowable fluctuation range of each dimension based on the dispersion of the data in each dimension, and then integrate the allowable fluctuation range of each dimension into the multi-dimensional spatial data radiation range boundary. For example, if the centroid mean of the power generation demand dimension is 50 kilowatts and the dispersion of the data in this dimension is small, then the allowable fluctuation range of this dimension is set to 47 to 53 kilowatts to ensure that the boundary can delineate a reasonable range of power demand under this mode. Similarly, allowable fluctuation ranges matching the mode characteristics are set for the gas pump outlet pressure, flow rate, and battery stack inlet pressure, and finally, a multi-dimensional boundary is obtained, just like drawing a reasonable region box for the normal data of the mode in four-dimensional space.

[0040] After the boundaries are set, it is necessary to quantify the degree of deviation of each data point from the typical modal state. Here, the Euclidean distance method, which is most commonly used in engineering to handle multidimensional parameter deviation problems, is adopted as the multidimensional spatial distance calculation method. The core advantage of the multidimensional spatial distance calculation method is that it can integrate the deviation of multiple dimensions to obtain a unified overall deviation value, avoiding the one-sidedness of single-dimensional judgment. For example, a data point may deviate slightly in the power dimension, but deviate very little in the pressure and flow dimensions. A single-dimensional judgment may be misjudged as abnormal, while the Euclidean distance can reflect the overall deviation through comprehensive calculation, which is more consistent with the gas path operation. In practice, the calculation of multi-parameter synergistic effects involves four steps for each high-dimensional data point in the high-dimensional data point set, using the feature centroid coordinates as a reference: First, extract the specific values ​​of the data point in each of the four dimensions, and simultaneously extract the mean values ​​of the feature centroid in the corresponding four dimensions. Second, calculate the dimensional deviation of the data point from the centroid in each dimension, i.e., the data point's dimensional value minus the centroid's dimensional mean. Third, square the deviation value in each dimension. Finally, sum the squared deviation values ​​of the four dimensions, and then take the square root of the sum. The result is the multidimensional spatial distance from the data point to the feature centroid coordinates.

[0041] Step 4.4: Based on the set data radiation range boundary, the distances from points to the feature centroid are compared and filtered. Data points with distances less than the corresponding boundary are retained as valid data points within the feature region. Specifically, this involves comparing the spatial distance of each high-dimensional data point to the feature centroid with the multi-dimensional data radiation range boundary set for the feature region one by one. If the spatial distance of a data point is less than the set radiation range boundary, it indicates that the data point deviates little from the typical state of the gas path operation mode corresponding to the feature region and belongs to the normal operating condition data under the mode, so it is retained as a valid data point. If the spatial distance of a data point is greater than or equal to the radiation range boundary, it indicates that the data point may be abnormal fluctuation data in the gas path operation and deviates too much from the core characteristics of the mode, so it is judged as an invalid data point and removed. During the filtering process, the system records the filtering results and corresponding distance values ​​of each data point for easy data traceability. Through this filtering operation, each feature region ultimately retains only valid data points that can accurately reflect the core characteristics of the corresponding gas path operation mode, removing various interfering abnormal data, and providing high-quality data support for extracting reliable statistical distribution characteristics and calculating accurate mode self-correction factors.

[0042] In this embodiment of the invention, by employing the technical means of extracting the original high-dimensional state vector corresponding to each feature region to form a high-dimensional data point set, calculating the arithmetic mean of each dimension to obtain the feature centroid coordinates, setting the multi-dimensional radiation range boundary according to the statistical characteristics of data distribution and calculating the distance from the data point to the centroid, and filtering effective data points with a distance less than the boundary, the technical problems of discrete abnormal data and insufficient data distribution in the feature regions obtained by clustering, resulting in insufficient accuracy of the correction parameters calculated based on data, are overcome. This achieves the goal of filtering out effective data points that accurately reflect the core characteristics of the corresponding gas path operation mode, providing high-quality data support for the accurate calculation of the subsequent modal self-correction factor and the accurate acquisition of the target set value, and improving the technical effect of dynamic matching control of air pump pressure and flow.

[0043] In a preferred embodiment of the present invention, step 5 above may include: Step 5.1: For each feature region's effective data point set, extract the statistical distribution characteristics of all data points in the set along the pressure and flow dimensions. The statistical distribution characteristics should include at least the mean, variance, and covariance. Specifically, for each feature region's effective data point set after filtering, focus on the two core dimensions that directly affect the sufficiency of the battery stack's electrochemical reaction: the gas pump outlet pressure and the hydrogen flow rate. For each feature region's effective data point set, first extract the gas pump outlet pressure values ​​of all data points. Summate these values ​​and divide by the total number of data points to obtain the mean of each feature region in the pressure dimension. The mean reflects the typical operating level of pressure under this mode. Then, calculate the deviation of each pressure value from the pressure mean, square all deviations, sum them, and divide by the total number of data points to obtain the variance of the pressure dimension. The magnitude of the variance directly reflects the degree of fluctuation in pressure values ​​under this mode. Following the same method, the mean and variance of the flow dimension are calculated separately. The flow mean represents the typical value of the flow under this mode, and the flow variance reflects the fluctuation of the flow. Finally, the covariance of pressure and flow is calculated: first, the pressure deviation of each data point is multiplied by the flow deviation of the corresponding data point, then all products are summed and divided by the total number of data points. The covariance result can reflect the correlation and change law between pressure and flow. For example, when the covariance is positive, it means that when the pressure increases, the flow usually increases synchronously. Through the operation, the mean, variance and covariance of each feature region in the pressure and flow dimensions are completely extracted, and the statistical distribution characteristics of pressure and flow under this mode are fully understood.

[0044] Step 5.2: Based on the statistical distribution characteristics of each feature region and combined with the preset performance optimization objective function, calculate and obtain the modal self-correction factor corresponding to each feature region. Specifically, the preset performance optimization objective function is formulated around the core operating requirements of the fuel cell. The core objective is to ensure that the electrochemical reaction in the fuel cell stack is sufficient and stable, while improving the energy conversion efficiency. Specifically, this includes ensuring that the hydrogen supply is accurately matched with the power generation demand, reducing unnecessary fluctuations in pressure and flow, and reducing gas path energy consumption. For each feature region, the extracted statistical distribution features are used as input conditions and substituted into the performance optimization objective function. If a feature region corresponds to a high-power fluctuating operating mode, its pressure and flow variances are large. After substituting these into the objective function, optimization directions that reduce fluctuations will be prioritized. If the pressure and flow covariance of a feature region is small, it indicates insufficient matching correlation between the two. The objective function will focus on strengthening the cooperative matching between the two. Through the calculation of the objective function, the specific parameters that need to be adjusted for the pressure and flow setpoints under this feature region are quantified. These specific parameters together constitute the mode self-calibration factor corresponding to each feature region. The mode self-calibration factor can adjust the baseline setpoints of pressure and flow according to the statistical characteristics of the mode, ensuring that the setpoints meet the optimization operation requirements of the mode. For example, for a low-power stable mode, the self-calibration factor will tend to keep the pressure and flow near the mean to reduce fluctuations.

[0045] Step 5.3: Obtain the processed multidimensional data of the current operating condition. Match the multidimensional data of the operating condition with the identified feature regions to determine the target feature region to which the current operating condition belongs. Specifically, this includes: acquiring multidimensional data of the current fuel cell operation in real time through the data acquisition channel, including power generation demand signal, gas pump outlet pressure signal, flow data, and fuel cell stack inlet pressure feedback signal; performing time-domain alignment and scaling normalization on these real-time data according to the processing standard to obtain standardized multidimensional data of the current operating condition; and retrieving information of all identified feature regions, including the feature centroid coordinates and multidimensional data radiation range boundaries of each feature region. A multidimensional spatial distance calculation method is adopted to calculate the spatial distance from the multidimensional data of the current working condition to the centroid coordinates of each feature region one by one. The spatial distance quantifies the deviation of the current working condition from the typical state of each modality. The distances from the current working condition data to the centroids of all feature regions are compared, and the feature region with the smallest distance is selected. At the same time, it is verified whether the current working condition data is within the boundary of the multidimensional data radiation range of the feature region. If it is within the range, the feature region is determined to be the target feature region to which the current working condition belongs; if it is not within the range, the data processing process is re-examined and it is confirmed whether there are any new unidentified modalities to ensure the accuracy of the target feature region matching.

[0046] Step 5.4 involves calling the modal self-calibration factor corresponding to the target feature region and, based on real-time operating data, fusing and calculating the target hydrogen pressure setpoint and target hydrogen flow setpoint dynamically matched to the current real-time operating conditions. Specifically, this includes: retrieving the modal self-calibration factor corresponding to the determined target feature region from a preset factor storage unit. The modal self-calibration factor contains optimized adjustment parameters for the modal pressure and flow setpoints; acquiring standardized multi-dimensional data of the current real-time operating conditions; focusing on extracting the current power generation demand, real-time values ​​of the gas pump outlet pressure and flow rate; and fusing the modal self-calibration factor with the real-time operating data: using the mean value of the pressure dimension of the target feature region as... The pressure benchmark setpoint is used as the reference value. Combined with the corresponding pressure adjustment parameters in the mode self-calibration factor, the pressure benchmark setpoint is dynamically corrected based on the deviation between the current power generation demand and the typical power of the mode. Following the same logic, the average value of the flow dimension is used as the flow benchmark setpoint. Combined with the flow adjustment parameters in the self-calibration factor and the real-time power deviation, the flow setpoint is corrected to obtain the flow setpoint. When the current real-time power is higher than the power average of the target characteristic region by a certain margin, the flow setpoint and pressure setpoint are appropriately increased through the self-calibration factor to ensure that the hydrogen supply can match the power increase demand. Through this fusion calculation, the target hydrogen pressure setpoint and target hydrogen flow setpoint are obtained to dynamically adapt to the current real-time operating conditions.

[0047] In this embodiment of the invention, by employing the technical means of extracting the pressure and flow statistical distribution characteristics of effective data points in each characteristic region, calculating the corresponding modal self-correction factor by combining the performance optimization objective function, and then matching the current operating condition to the target characteristic region and fusing the self-correction factor with real-time data to generate the target setpoint, the technical problem of poor adaptability caused by the traditional control scheme relying on static mapping and being unable to combine the core characteristics of the gas path operating mode with the real-time operating condition to dynamically adjust the pressure and flow setpoint is overcome. Thus, the technical effect of accurately matching the target hydrogen pressure and flow setpoint with the current real-time operating condition is achieved, improving the dynamic response capability and adaptability of the gas path supply regulation, and ensuring the full and stable electrochemical reaction of the fuel cell is achieved.

[0048] In a preferred embodiment of the present invention, step 6 above may include: Step 6.1: Using the target hydrogen pressure setpoint and target hydrogen flow rate setpoint as joint input conditions, a preliminary target speed command for the gas pump motor that meets the joint conditions is obtained through parsing. Specifically, this includes: pre-establishing a database of correlation characteristics between the gas pump motor speed and hydrogen pressure and flow rate. This database is constructed based on measured data of the gas pump under different operating conditions and contains a large number of pressure and flow output values ​​corresponding to different speeds, which can accurately reflect the actual working characteristics of the gas pump. The target hydrogen pressure setpoint and target hydrogen flow rate setpoint are used as joint input conditions and imported into the speed parsing module. The speed parsing module calls the data in the correlation characteristic database and uses a combination of data matching and interpolation calculation to filter out the range of gas pump motor speeds that simultaneously meet the target pressure and target flow rate requirements. Within the speed range, the optimal speed value is selected as the preliminary target speed command for the gas pump motor, taking into account the energy consumption economy and response speed requirements of the gas pump operation. For example, if the target pressure is 0.35 MPa and the target flow rate is 22 liters / minute, by matching the database data, it is found that the speed range of 1500 rpm to 1600 rpm can meet the requirements. After considering energy consumption and response speed, 1550 rpm is determined as the initial speed command, while ensuring that this speed is within the safe operating range of the air pump.

[0049] Step 6.2 involves using the target hydrogen pressure setpoint, target hydrogen flow rate setpoint, and preliminary target speed command of the gas pump motor as inputs to collaboratively analyze and obtain the target opening command of the hydrogen supply valve that meets the target supply conditions. Specifically, this includes constructing a fluid dynamics model of the hydrogen supply system. This model fully considers the resistance characteristics of the gas pipeline, the relationship between valve opening and flow rate, and the influence of gas pump speed on gas pressure, accurately simulating the gas pipeline operating state under different input conditions. The target hydrogen pressure setpoint, target hydrogen flow rate setpoint, and the preliminary target speed command of the gas pump motor obtained in Step 6.1 are input into this fluid dynamics model. The model first calculates the theoretical output pressure and flow rate of the gas pump based on the preliminary speed command. Then, combining the deviation of the target pressure and flow rate, it analyzes the influence of changes in gas pipeline resistance on the supply state. For example, when the gas pipeline resistance increases due to increased operating time, even if the gas pump speed reaches the preliminary command value, the actual flow rate may be lower than the target value. Based on these analysis results, the model calculates the target opening command of the hydrogen supply valve that can compensate for the deviation and ensure that the hydrogen supply meets the target conditions. This opening command can not only adapt to the target pressure and flow rate, but also work in conjunction with the initial speed command to ensure that the pressure and flow rate in the gas path stably reach the set target.

[0050] Step 6.3: Based on the target opening command of the hydrogen supply valve, dynamically compensate and fine-tune the initial target speed command of the gas pump motor to finally determine the target speed command of the gas pump motor. Specifically, this includes: pre-establishing a correlation model between the hydrogen supply valve opening and the gas pump operating back pressure; clarifying the gas path back pressure change law corresponding to different valve openings; decreasing the valve opening will increase the gas path back pressure, and increasing the valve opening will decrease the back pressure; and the back pressure change will directly affect the actual output performance of the gas pump, causing deviations in the pressure and flow output corresponding to the initial speed command; calculating the change in gas path back pressure under the opening; determining the value that the gas pump motor speed needs to compensate for; if the target valve opening is smaller than the initially matched opening, the gas path back pressure increases, and the gas pump... The actual output pressure at the initial speed will be too high and the flow rate too low. In this case, the speed needs to be reduced appropriately to compensate. If the target valve opening is too large, the back pressure will be reduced, and the actual output pressure of the air pump will be too low and the flow rate too high. In this case, the speed needs to be increased appropriately. The initial target speed command of the air pump motor is fine-tuned according to the compensation value. For example, if the initial speed is 1550 rpm, it is calculated that a compensation reduction of 20 rpm is needed. Therefore, the speed command is adjusted to 1530 rpm. After the fine-tuning is completed, the fluid dynamics model is used again to verify whether the combination of the adjusted speed and valve opening can accurately meet the target pressure and flow requirements. After confirming that there are no errors, the final target speed command of the air pump motor is determined to ensure that the coordinated operation of the air pump and valve can accurately adapt to the real-time working conditions.

[0051] In this embodiment of the invention, by employing a technical approach of obtaining a preliminary pump motor speed command by jointly inputting the target hydrogen pressure and flow rate setpoints, coordinating the analysis of the hydrogen supply valve opening command by combining the target value and the preliminary speed command, and then using the valve opening command to dynamically compensate and fine-tune the preliminary speed command, the technical problem of deviation between the pump output and the hydrogen supply target caused by the independent generation and lack of coordination and adaptation of speed and valve opening commands in traditional control is overcome. This achieves the technical effect of ensuring precise coordination between the pump motor speed and valve opening command, ensuring that the hydrogen pressure and flow rate stably reach the target value, and improving the control accuracy of the hydrogen supply system.

[0052] In a preferred embodiment of the present invention, step 7 above may include: Step 7.1: The target speed command for the air pump motor and the target opening command for the hydrogen supply valve are sent to the air pump motor driver and the hydrogen supply valve controller respectively via communication links to drive the corresponding components to perform adjustments. Specifically, this includes: confirming the completeness and validity of the final determined target speed command for the air pump motor and the target opening command for the hydrogen supply valve, ensuring that the command parameters are within the safe operating range of the equipment, selecting an industrial Ethernet with high real-time performance and anti-interference capabilities as the communication link. This link can adapt to the complex electromagnetic environment in mobile scenarios such as hydrogen-powered commercial vehicles, avoiding delayed or lost command transmission. The target speed command for the air pump motor is then sent according to… The communication protocol of the gas pump motor driver is converted to ensure that the command can be directly recognized by the driver. At the same time, the target opening command of the hydrogen supply valve is converted into a signal format compatible with the valve controller. The converted speed command is sent separately to the gas pump motor driver and the opening command is sent separately to the hydrogen supply valve controller through the communication link to avoid command confusion. After receiving the speed command, the driver immediately adjusts the internal power output module to drive the gas pump motor to operate at the target speed. After receiving the opening command, the valve controller drives the valve core to rotate through the stepper motor, so that the valve reaches the target opening and completes the initial adjustment of the hydrogen supply.

[0053] Step 7.2: After the command is sent and the components are adjusted, the current gas pump outlet pressure signal, flow data, and battery stack inlet pressure feedback signal are collected in real time to obtain the actual feedback signal. Specifically, this includes: starting the data acquisition system in the first sampling cycle after the gas pump motor and hydrogen supply valve begin to perform adjustment actions. The current gas pump outlet pressure signal is collected by a high-precision pressure sensor installed in the gas pump outlet pipeline. The measurement accuracy of this sensor needs to reach 0.01 MPa to meet the requirements of precise control. Real-time flow data is collected by a flow detection device integrated in the hydrogen delivery main pipeline. This device needs to be able to adapt to a wide flow range of 5 to 50 liters / minute to ensure the accuracy of data under different power conditions. The current battery stack inlet pressure feedback signal is collected by a pressure sensing unit at the battery stack inlet end. This signal directly reflects the final effect of hydrogen supply. The three types of signals collected are filtered by the signal conditioning module to remove noise caused by pipeline vibration and electromagnetic interference, and then converted into standardized digital signals. Finally, the actual feedback signal set containing pressure, flow, and inlet pressure is obtained and transmitted to the data processing center in real time.

[0054] Step 7.3 compares the actual feedback signal with the target hydrogen pressure setpoint and the target hydrogen flow rate setpoint to obtain the pressure and flow rate closed-loop deviation signals between the actual supply state and the target state. Specifically, this includes: retrieving the determined target hydrogen pressure setpoint and target hydrogen flow rate setpoint from the data storage unit; comparing the pump outlet pressure value in the actual feedback signal with the target hydrogen pressure setpoint time-by-time to calculate the difference, which is the pressure closed-loop deviation signal; and comparing the flow rate data in the actual feedback signal with the target hydrogen flow rate setpoint time-by-time to obtain the flow rate closed-loop deviation signal. During the calculation process, if the deviation value at a certain moment exceeds the preset normal fluctuation threshold, the deviation signal at that moment is marked as an abnormal deviation, and the corresponding operating condition information is recorded. This is combined with the battery stack inlet pressure value in the actual feedback signal to help determine the cause of the deviation. If the deviation between the inlet pressure and the outlet pressure is too large, it indicates that the gas path resistance may have changed, which needs to be considered in subsequent adjustments. Finally, a complete pressure and flow rate closed-loop deviation signal containing normal deviation, abnormal deviation, and auxiliary judgment information is obtained.

[0055] Step 7.4: Based on the closed-loop deviation signal, re-trigger and execute the operating condition matching, modal self-correction factor invocation, and setpoint fusion calculation to obtain the adjusted new control command, realizing closed-loop dynamic matching control of hydrogen supply pressure and flow rate. Specifically, this includes: inputting the closed-loop deviation signal to the control logic trigger module; if the deviation signal is a normal deviation, triggering the lightweight adjustment process: focusing on dimensions with larger deviations, re-execute the operating condition matching step. At this time, the operating condition matching uses the current actual feedback signal and deviation signal as the core basis to quickly locate the subtle deviation direction between the current operating condition and the target feature region; invoking the modal self-correction factor of the target feature region and proportionally adjusting the self-correction factor according to the deviation magnitude; and then applying the adjusted self-correction factor... The factors are fused with the current real-time operating data to obtain the fine-tuned new gas pump motor speed command and the new hydrogen supply valve opening command. If the deviation signal is an abnormal deviation, the complete adjustment process is triggered: the entire process from multi-dimensional data acquisition, time domain alignment, modal identification to operating condition matching is re-executed to ensure accurate identification of operating condition changes caused by sudden power fluctuations or gas path resistance changes; based on the newly identified operating conditions, the target feature area is determined, the corresponding self-correction factor is called and combined with the abnormal deviation analysis results to generate a new control command. After the new control command is generated, the adjustment effect is first verified. After confirming that the deviation can be controlled within the normal range, it is sent to the corresponding driver and controller for execution to achieve continuous closed-loop dynamic matching control of hydrogen supply pressure and flow.

[0056] In this embodiment of the invention, the closed-loop dynamic adjustment technique, which uses a communication link to send target control commands to drive the gas pump motor and hydrogen supply valve to perform regulation, collects actual pressure and flow rate and inlet pressure feedback signals in real time, compares the target value to obtain the closed-loop deviation signal, and re-triggers the operating condition matching mode self-correction based on the deviation and calculates the adjusted new control command by fusing the set value, overcomes the technical problems of traditional static control schemes lacking a real-time feedback correction mechanism and being unable to cope with sudden fluctuations in power generation and dynamic changes in gas path resistance that cause continuous deviations between hydrogen supply and target state. Thus, it achieves the technical effects of continuously and dynamically correcting control commands, ensuring that hydrogen supply pressure and flow rate always accurately match the target requirements, and improving the stability of fuel cell electrochemical reaction and energy conversion efficiency.

[0057] like Figure 2 As shown, embodiments of the present invention also provide an adaptive data processing system for dynamic matching of air pump pressure and flow rate, comprising: The data acquisition module is used to acquire the power generation demand signal of the fuel cell controller, the pressure signal at the outlet of the gas pump, the flow data, and the pressure feedback signal at the inlet of the fuel cell stack. The alignment module is used to perform time-domain alignment and scaling normalization on the power generation demand signal, the gas pump outlet pressure signal, the flow data and the battery stack inlet pressure feedback signal to obtain a multi-dimensional operating condition dataset. The module is used to organize the multi-dimensional operating condition dataset into a high-dimensional state space of gas supply dynamics based on the multi-dimensional operating condition dataset; and to perform unsupervised clustering analysis on the high-dimensional state space to identify the feature regions of multiple different gas path operating modes. The optimization module is used to calculate the average coordinates of all high-dimensional data points in the feature region based on the high-dimensional data point set corresponding to the feature region, and use the feature centroid as the center to set a reasonable data radiation range, and retain the effective data points within the reasonable data radiation range. The fusion module is used to extract the statistical distribution characteristics of effective data points, and based on the statistical distribution characteristics and the modal self-correction factors corresponding to each feature region, fuse all modal self-correction factors to obtain the target hydrogen pressure setpoint and target hydrogen flow setpoint that match the real-time operating conditions. The analysis module, based on the target hydrogen pressure setpoint and the target hydrogen flow rate setpoint, analyzes and obtains the corresponding target speed command of the gas pump motor and the target opening command of the hydrogen supply valve. The matching module sends the gas pump motor speed control command and the supply valve opening control command to the corresponding control components to coordinate the adjustment of the gas pump motor speed and the supply valve opening, thereby achieving closed-loop dynamic matching control of hydrogen supply pressure and flow.

[0058] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. An adaptive data processing method for dynamic matching of air pump pressure and flow rate, characterized in that, The method includes: The system collects the power demand signal from the fuel cell controller, the pressure signal at the outlet of the gas pump, the flow data, and the pressure feedback signal at the inlet of the fuel cell stack. The power generation demand signal, the gas pump outlet pressure signal, the flow data and the battery stack inlet pressure feedback signal are processed by time domain alignment and scaling normalization to obtain a multi-dimensional operating condition dataset. Based on the multi-dimensional operating condition dataset, the multi-dimensional operating condition dataset is organized into a high-dimensional state space of gas supply dynamics; unsupervised clustering analysis is performed on the high-dimensional state space to identify the characteristic regions of multiple different gas supply operation modes. Based on the high-dimensional data point set corresponding to the feature region, the average coordinates of all high-dimensional data points in the feature region are calculated as the feature centroid. A reasonable data radiation range is set with the feature centroid as the center, and valid data points within the reasonable data radiation range are retained. By extracting the statistical distribution characteristics of effective data points, and based on the statistical distribution characteristics and the modal self-correction factors corresponding to each feature region, all modal self-correction factors are fused to obtain the target hydrogen pressure setpoint and target hydrogen flow rate setpoint that match the real-time operating conditions. Based on the target hydrogen pressure setpoint and the target hydrogen flow rate setpoint, the corresponding target speed command of the gas pump motor and the target opening command of the hydrogen supply valve are obtained by analysis. The speed control command of the gas pump motor and the opening control command of the supply valve are sent to the corresponding control components to coordinate the adjustment of the speed of the gas pump motor and the opening of the supply valve, so as to realize the closed-loop dynamic matching control of the hydrogen supply pressure and flow rate.

2. The adaptive data processing method for dynamic matching of air pump pressure and flow rate according to claim 1, characterized in that, The power generation demand signal, air pump outlet pressure signal, flow data, and battery stack inlet pressure feedback signal are time-domain aligned and scaled to obtain a multi-dimensional operating condition dataset, including: It receives and aggregates raw time-series data streams of power generation demand signals, air pump outlet pressure signals, flow data, and battery stack inlet pressure feedback signals from different sources. Based on the power demand signal, and based on the power step change point preset in the power demand signal, the time-series data of the air pump outlet pressure signal, flow data and battery stack inlet pressure feedback signal are time-stamp matched and interpolated to obtain a time-domain aligned set of synchronization signals. For each dimension of the synchronization signal set, the maximum value, minimum value and statistical distribution within the preset historical time window are analyzed to dynamically determine the range of signal feature values ​​used for normalization processing. Based on the range of signal characteristic values, the real-time values ​​of each dimension of the signal are linearly mapped to a unified numerical range, resulting in a scaled multidimensional signal sequence. By organizing and storing scaled, multi-dimensional signal sequences in chronological order, a multi-dimensional working condition dataset is obtained.

3. The adaptive data processing method for dynamic matching of air pump pressure and flow rate according to claim 2, characterized in that, Based on the multi-dimensional operating condition dataset, the multi-dimensional operating condition dataset is organized into a high-dimensional state space of gas supply dynamics. Unsupervised clustering analysis was performed in a high-dimensional state space to identify characteristic regions of multiple different gas path operating modes, including: From the multi-dimensional operating condition dataset, sample points of consecutive time intervals are extracted in chronological order; the multi-dimensional data of each sample point are combined to construct a high-dimensional state vector of the gas supply dynamics at the corresponding time. The high-dimensional state vectors are arranged and organized in time order to form a high-dimensional state space for the gas supply dynamics. For high-dimensional state vectors in a high-dimensional state space, a preset dimensionality reduction algorithm is applied to process them in order to reduce the data dimensionality and extract the main features, thereby obtaining the corresponding set of low-dimensional feature vectors. For the set of low-dimensional feature vectors, a pre-defined unsupervised clustering algorithm is applied for analysis, and data points with similar operating characteristics are classified into the same cluster. Each cluster corresponds to an identified gas path operating mode cluster, so as to obtain the clustering analysis results. Based on the cluster analysis results, the high-dimensional state vector in the high-dimensional state space corresponding to each cluster is defined and marked as an independent feature region, thus completing the identification of feature regions for multiple different gas path operation modes.

4. The adaptive data processing method for dynamic matching of air pump pressure and flow rate according to claim 3, characterized in that, Based on the high-dimensional data point set corresponding to the feature region, the average coordinates of all high-dimensional data points within the feature region are calculated as the feature centroid. A reasonable data radiation range is then defined with the feature centroid as the center, and valid data points within this range are retained, including: Based on each identified feature region, extract all the original high-dimensional state vectors defined and labeled by each feature region to form a high-dimensional data point set corresponding to each feature region; Based on a high-dimensional data point set, the arithmetic mean of all high-dimensional data points in the set is calculated in each dimension, and the arithmetic mean is combined to form the feature centroid coordinates of each feature region. Centered on the feature centroid coordinates, and based on the distribution statistics of all points in the corresponding high-dimensional data point set, a multi-dimensional spatial data radiation range boundary is set for each feature region. Based on the corresponding feature centroid coordinates, the distance from each high-dimensional data point in the high-dimensional data point set to the feature centroid is calculated one by one. Based on the set data radiation range boundary, the distance from the point to the feature centroid is compared and filtered, and data points with a distance less than the corresponding boundary are retained as valid data points within the feature region.

5. The adaptive data processing method for dynamic matching of air pump pressure and flow rate according to claim 4, characterized in that, By extracting the statistical distribution characteristics of effective data points, and based on the statistical distribution characteristics and the modal self-correction factors corresponding to each feature region; By integrating all modal self-correction factors, target hydrogen pressure setpoints and target hydrogen flow rate setpoints that match real-time operating conditions are obtained, including: For each feature region's effective data point set, extract the statistical distribution characteristics of all data points in the set in terms of pressure and flow. The statistical distribution characteristics include at least the mean, variance, and covariance. Based on the statistical distribution characteristics of each feature region, and combined with the preset performance optimization objective function, the modal self-correction factor corresponding to each feature region is calculated and obtained. The current working condition multidimensional data is obtained after processing. The working condition multidimensional data is matched with the identified feature regions to determine the target feature region to which the current working condition belongs. The modal self-correction factor corresponding to the target feature region is called, and the target hydrogen pressure setpoint and target hydrogen flow setpoint are dynamically matched with the current real-time operating conditions by fusion calculation based on real-time operating data.

6. The adaptive data processing method for dynamic matching of air pump pressure and flow rate according to claim 5, characterized in that, Based on the target hydrogen pressure setpoint and the target hydrogen flow rate setpoint, the corresponding target speed command for the gas pump motor and the target opening command for the hydrogen supply valve are obtained through analysis, including: Using the target hydrogen pressure setpoint and the target hydrogen flow rate setpoint as joint input conditions, a preliminary target speed command for the gas pump motor that meets the joint conditions is obtained by parsing. Using the target hydrogen pressure setpoint, the target hydrogen flow rate setpoint, and the initial target speed command of the gas pump motor as inputs, the target opening command of the hydrogen supply valve that meets the target supply conditions is obtained through collaborative analysis. Based on the target opening command of the hydrogen supply valve, the initial target speed command of the gas pump motor is dynamically compensated and finely adjusted to finally determine the target speed command of the gas pump motor.

7. The adaptive data processing method for dynamic matching of air pump pressure and flow rate according to claim 6, characterized in that, The gas pump motor speed control command and the supply valve opening control command are sent to the corresponding control components to coordinately adjust the gas pump motor speed and the supply valve opening, thereby achieving closed-loop dynamic matching control of hydrogen supply pressure and flow rate, including: The target speed command for the air pump motor and the target opening command for the hydrogen supply valve are sent to the air pump motor driver and the hydrogen supply valve controller respectively via the communication link, so as to drive the corresponding components to perform adjustment. After the command is sent and the components are adjusted, the current air pump outlet pressure signal, flow data and battery stack inlet pressure feedback signal are collected in real time to obtain the actual feedback signal; The actual feedback signal is compared with the target hydrogen pressure setpoint and the target hydrogen flow rate setpoint to obtain the pressure and flow closed-loop deviation signal between the actual supply state and the target state. Based on the closed-loop deviation signal, the operating condition matching, modal self-correction factor call and setpoint fusion calculation are re-triggered and executed to obtain the adjusted new control command, thereby realizing the closed-loop dynamic matching control of hydrogen supply pressure and flow.

8. An adaptive data processing system for dynamic matching of air pump pressure and flow rate, wherein the system implements the method as described in any one of claims 1 to 7, characterized in that, include: The data acquisition module is used to acquire the power generation demand signal of the fuel cell controller, the pressure signal at the outlet of the gas pump, the flow data, and the pressure feedback signal at the inlet of the fuel cell stack. The alignment module is used to perform time-domain alignment and scaling normalization on the power generation demand signal, the gas pump outlet pressure signal, the flow data and the battery stack inlet pressure feedback signal to obtain a multi-dimensional operating condition dataset. The module is used to organize the multi-dimensional operating condition dataset into a high-dimensional state space of gas supply dynamics based on the multi-dimensional operating condition dataset. Unsupervised clustering analysis was performed in a high-dimensional state space to identify the characteristic regions of multiple different gas path operation modes; The optimization module is used to calculate the average coordinates of all high-dimensional data points in the feature region based on the high-dimensional data point set corresponding to the feature region, and use the feature centroid as the center to set a reasonable data radiation range, and retain the effective data points within the reasonable data radiation range. The fusion module is used to extract the statistical distribution characteristics of effective data points and, based on the statistical distribution characteristics and the modal self-correction factors corresponding to each feature region, perform modal self-correction. By integrating all modal self-correction factors, target hydrogen pressure setpoints and target hydrogen flow rate setpoints that match real-time operating conditions are obtained; The analysis module, based on the target hydrogen pressure setpoint and the target hydrogen flow rate setpoint, analyzes and obtains the corresponding target speed command of the gas pump motor and the target opening command of the hydrogen supply valve. The matching module sends the gas pump motor speed control command and the supply valve opening control command to the corresponding control components to coordinate the adjustment of the gas pump motor speed and the supply valve opening, thereby achieving closed-loop dynamic matching control of hydrogen supply pressure and flow.

9. A computing device, characterized in that, include: One or more processors; A storage device for storing one or more programs, which, when executed by one or more processors, cause the one or more processors to implement the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a program that, when executed by a processor, implements the method as described in any one of claims 1 to 7.

Images