Fuel cell vehicle hydrogen storage cabin hydrogen leakage identification method based on data driving, leakage identification electronic equipment and storage medium

Abstract

The invention provides a hydrogen leakage identification method for a hydrogen storage cabin of a fuel cell vehicle based on data driving, leakage identification electronic equipment and a storage medium, relates to the technical field of fault monitoring and diagnosis of a fuel cell system, and solves the technical problems of low identification precision and poor real-time performance under the working conditions of a closed environment and multiple leakage factors in the prior art. According to the technical scheme, the method is characterized in that a leakage data set covering multiple working conditions is constructed through simulation; performing sliding window segmentation and standardized preprocessing on the data; extracting multi-scale features of the leakage time sequence by using a time sequence convolutional network; enhancing the key time sequence features by using a multi-head self-attention mechanism; carrying out pooling dimension reduction on the enhanced features; and inputting the compressed feature vector into an XGBoost classifier to realize multi-label identification of the leakage level and the leakage position. According to the method, time sequence modeling, an attention mechanism and an integrated classifier are fused, and the accuracy, robustness and real-time performance of hydrogen leakage recognition under complex working conditions are remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of fuel cell system fault monitoring and diagnosis, and particularly relates to a hydrogen leakage identification method for a hydrogen storage tank of a fuel cell vehicle based on data driving, a leakage identification electronic device and a storage medium. BACKGROUND

[0002] As a new clean energy, hydrogen energy has the advantages of high energy density and zero carbon emission, and shows broad application prospects in transportation, industry and energy. In particular, in fuel cell vehicles, hydrogen energy is gradually replacing traditional fossil energy and promoting the transformation of the transportation system to low-carbon intelligence.

[0003] However, hydrogen itself has the characteristics of easy diffusion and wide explosion range, and once a leakage occurs in the hydrogen storage system, it is easy to cause safety accidents and threaten personal and property safety. At present, the traditional hydrogen leakage detection method mostly relies on physical sensors (such as hydrogen concentration sensors and pressure sensors), which are limited by the response speed, stability, deployment density and cost of the sensors, and their identification ability for leakage behavior in complex environments (especially in closed spaces and different ventilation conditions) is limited, which is difficult to meet the detection requirements of high precision, high robustness and real-time of the vehicle-mounted system.

[0004] In recent years, with the rapid development of data-driven methods and intelligent diagnosis technology, hydrogen leakage identification methods based on historical data and machine learning algorithms have gradually become a research hotspot. For example, some scholars use Fisher discriminant analysis (FDA) and related vector machine (RVM) methods to model the hydrogen pressure drop data, and some studies fuse CEEMDAN-CNN-LSTM methods to locate the leakage point. However, existing researches mostly focus on hydrogen refueling stations and open environments, and the research on leakage behavior identification under the influence of multiple variables (leakage level, position, direction) in closed spaces (such as fuel cell vehicle hydrogen storage tanks) and different ventilation conditions is still insufficient. In addition, traditional models mostly use single structures (such as only using CNN or LSTM), which have bottlenecks in feature extraction ability, processing long-term dependence, noise interference resistance and model generalization ability. SUMMARY

[0005] The present application aims to overcome the problems of low identification accuracy and poor real-time performance of existing technologies in closed environments and multiple leakage factors (level, position, direction) working conditions, and proposes a hydrogen leakage identification method for fuel cell hydrogen storage tanks that fuses time series modeling, attention mechanism and integrated classifier.

[0006] The present application achieves the above technical purpose by the following technical means.

[0007] A hydrogen leakage identification method for a hydrogen storage tank of a fuel cell vehicle based on data driving, comprising the following steps:

[0008] S1: dataset construction and preprocessing: based on CFD simulation, multi-dimensional hydrogen leakage time series data covering different ventilation conditions, different leakage levels, different leakage directions and different leakage positions are generated to form a dataset; the time series data is divided by a sliding window and standardized, and corresponding leakage level labels and leakage position labels are labeled; the processed dataset is divided into a training set and a test set;

[0009] S2: time series feature extraction: the time series data of the training set and the test set are input into a time series convolution network TCN for feature extraction, the time series convolution network TCN includes multiple one-dimensional causal convolution layers, and multi-scale time series dependency relationships are captured by setting an increasing inflation rate of inflation convolution, and a high-dimensional time series feature sequence is output;

[0010] S3: feature enhancement and compression: the high-dimensional time series feature sequence output by S2 is input into a multi-head self-attention mechanism module, and the key features related to the leakage sensitive area are dynamically enhanced by calculating the attention weights of different time steps; then, the attention-enhanced feature sequence is pooled along the time dimension to obtain a fixed-length feature vector;

[0011] S4: classification and recognition model training and testing: the fixed-length feature vector of the training set obtained in S3 is input into an XGBoost classifier for training to obtain a trained XGBoost leakage recognition model; the fixed-length feature vector of the test set is input into the trained XGBoost leakage recognition model to output corresponding leakage level prediction labels and leakage position prediction labels, and hydrogen leakage recognition is completed.

[0012] Further, in step S1, the different ventilation conditions include airtight space, natural ventilation and forced ventilation with different wind speeds; the different leakage levels include a plurality of preset leakage severity levels; the different leakage directions include up, down, left and right; and the different leakage positions include a plurality of preset points on the hydrogen storage cabin.

[0013] Further, in step S2, the time series convolution network TCN includes at least 4 one-dimensional causal convolution layers, and the inflation rate sequence thereof is set to 1, 2, 4 and 8; and the convolution kernel size is set to 5x1.

[0014] Further, in step S3, the multi-head self-attention mechanism adopts 4 heads, and the dimension of the key vector is set to 64; and the pooling operation is global average pooling.

[0015] Further, in step S4, the XGBoost classifier adopts gradient boosting decision trees during training, sets learning rate, maximum depth, number of decision trees, and multi-classification objective function; for the XGBoost classifier for identifying the leakage position, the learning rate is set to 0.1, the maximum depth is set to 8, and the number of decision trees is set to 161. For the XGBoost classifier for identifying the leakage level, the learning rate is set to 0.05, the maximum depth is set to 8, and the number of decision trees is set to 321.

[0016] Further, the training process of the XGBoost classifier in step S4 includes: splicing the preliminary diagnosis result generated by the first layer decision tree with the original fixed-length feature vector obtained in S3, and inputting the enhanced feature generated by the autoencoder to the lower layer decision tree for further diagnosis and optimization; the process is iteratively executed until the model accuracy converges, and the final XGBoost leakage identification model is output.

[0017] Further, the method further comprises:

[0018] S5: performance evaluation: based on the leakage level prediction label and the leakage position prediction label obtained in step S4, the corresponding true label is compared, the accuracy, recall rate and F1 score indicators are calculated, and the model recognition performance is evaluated.

[0019] A leakage identification electronic device comprises a processor and a memory, and the memory stores computer readable instructions, when the computer readable instructions are executed by the processor, the steps of the data-driven fuel cell vehicle hydrogen storage tank hydrogen leakage identification method are run.

[0020] A storage medium has a computer program stored thereon, and the computer program is executed by a processor to run the steps of the data-driven fuel cell vehicle hydrogen storage tank hydrogen leakage identification method.

[0021] The beneficial effects of the present application are:

[0022] 1. The data-driven fuel cell vehicle hydrogen storage tank hydrogen leakage identification method disclosed by the present application is based on a multi-dimensional data set constructed by CFD simulation, which covers closed space, natural ventilation, forced ventilation of various wind speeds, various leakage levels, directions and positions, so that the model training can cover typical and extreme working conditions that may be encountered in the actual operation of the fuel cell vehicle hydrogen storage tank, significantly improving the adaptability and robustness of the model in the actual complex and variable environment.

[0023] 2. The data-driven hydrogen leakage identification method for fuel cell vehicle hydrogen storage tank according to the present application effectively solves the problems of slow training and difficulty in capturing long-term dependencies of traditional RNN / LSTM by using TCN model (expanding convolution structure), and can accurately model the time sequence characteristics of the whole process from the outbreak to the diffusion of the leakage event; the multi-head self-attention mechanism (MHSA) is introduced, which can automatically focus on the most critical time sequence fragments and sensor channel features for leakage judgment, suppress irrelevant noise, and greatly enhance the discriminability of the features and the sensitivity of the model.

[0024] 3. The data-driven hydrogen leakage identification method for fuel cell vehicle hydrogen storage tank according to the present application, XGBoost as a high-performance integrated classifier combines the powerful classification ability of decision tree and the generalization advantage of Boosting, and performs excellently in the multi-classification task of leakage level and position; at the same time, XGBoost can output feature importance evaluation (such as the importance of MHSA enhanced features and specific sensor channels), which improves the transparency and engineering reviewability of model decision, and facilitates fault tracing and model optimization.

[0025] 4. The data-driven hydrogen leakage identification method for fuel cell vehicle hydrogen storage tank according to the present application, the convolution structure inherent in TCN has certain translation invariance and noise resistance, the focusing on key features by MHSA further weakens the noise interference, and the integrated characteristics of XGBoost also reduce the sensitivity of the model to single data disturbance, so that the method can still maintain high identification accuracy when facing sensor minor faults, background noise or data fluctuations.

[0026] 5. The data-driven hydrogen leakage identification method for fuel cell vehicle hydrogen storage tank according to the present application, the model structure is TCN+MHSA+XGBoost, which has high inference efficiency and meets the demand of millisecond-level safety response of vehicle-mounted system; compared with the scheme relying on a large number of physical sensors, the present application has more relaxed requirements for sensor deployment density, reducing the system cost; the model is trained under multiple working conditions, and has strong adaptability, which is convenient for deployment and application in the actual hydrogen safety monitoring system of fuel cell vehicle. BRIEF DESCRIPTION OF DRAWINGS

[0027] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or prior art description will be briefly introduced as follows, and the drawings in the following description are some embodiments of the present application, and for those skilled in the art, other drawings can be obtained without creative labor on the premise of these drawings.

[0028] Figure 1 The flow chart of the hydrogen leakage identification method for fuel cell vehicle hydrogen storage tank according to the present application;

[0029] Figure 2 A three-dimensional schematic view of a fuel cell vehicle hydrogen storage tank leakage geometry model used in CFD simulation in an embodiment of the present application;

[0030] Figure 3 A core recognition algorithm (TCN+MHSA+XGBoost) structural flowchart of an embodiment of the present application; wherein part a shows a TCN feature extraction structure, and part b shows an MHSA feature enhancement and pooling compression structure;

[0031] Figure 4 A leakage recognition confusion matrix result graph of an embodiment of the present application; wherein a is a leakage position recognition confusion matrix, and b is a leakage level recognition confusion matrix. DETAILED DESCRIPTION

[0032] Embodiments of the present application are described in detail below with reference to the attached drawings, which show by way of example, embodiments in which like numerals indicate like elements or components, and in which the scope of the claims should not be construed as being limited to the embodiments set forth in the description below.

[0033] In the description of the present application, it needs to be understood that the orientation or positional relationship indicated by the terms "center", "longitudinal", "transverse", "length", "width", "thickness", "upper", "lower", "axial", "radial", "vertical", "horizontal", "inner", "outer" and the like is based on the orientation or positional relationship shown in the drawings, and is only for the convenience of describing the present application and simplifying the description, and therefore cannot be understood as indicating or implying that the device or element indicated must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as limiting the present application. In addition, the terms "first", "second" are only for the purpose of description, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the technical features indicated. Therefore, the features limited by "first", "second" can explicitly or implicitly include one or more of the features. In the description of the present application, the meaning of "multiple" is two or more, unless otherwise specifically limited.

[0034] In the present application, unless otherwise specifically defined and limited, the terms "mounting", "connection", "connection", "fixing" and the like should be understood broadly, for example, it can be fixedly connected, or it can be detachably connected, or it can be integrally connected; it can be mechanically connected, or it can be electrically connected; it can be directly connected, or it can be indirectly connected through an intermediate medium; it can be the communication inside two elements. For those skilled in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.

[0035] As Figure 1As shown, the data-driven-based hydrogen leakage identification method for hydrogen storage tank of fuel cell vehicle according to the present application comprises the following steps:

[0036] Step S1: data set construction and preprocessing, specifically:

[0037] The purpose of S1 is to create representative training and testing data. Using computational fluid dynamics (CFD) software, a three-dimensional geometric model of the hydrogen storage tank of the fuel cell vehicle is established, as shown in Figure 2 The following settings are made on the three-dimensional geometric model of the hydrogen storage tank of the fuel cell vehicle:

[0038] 5 scenarios: closed space, natural ventilation and forced ventilation, where the forced ventilation wind speed is 10 m / s, 20 m / s and 30 m / s, respectively.

[0039] 5 levels of leakage: Level 1 (trace leakage) to Level 5 (severe leakage), simulated by changing the leakage aperture or pressure difference.

[0040] 4 leakage directions: up, down, left and right.

[0041] 4 leakage locations: at four different pre-set location points on the surface of the hydrogen storage tank, such as Figure 2 Leakage point 1 to leakage point 4 as shown.

[0042] By combining the above parameters, a total of 400 different leakage condition samples are generated. For each sample, transient CFD simulation is performed to simulate the diffusion process after hydrogen leakage, and the sequence data (multi-channel time series) of the hydrogen concentration at the hydrogen storage tank and the key monitoring points (simulated sensor positions) over time are recorded. Each sample is labeled with its corresponding leakage level label and leakage location label.

[0043] The original time series data is preprocessed as follows:

[0044] Sliding window segmentation: long time series data is segmented into fixed length segments for model processing.

[0045] Standardization: standardize the data of each channel (such as Z-score standardization) to eliminate dimensional effects and speed up model convergence.

[0046] Finally, the processed 400 sample data sets are randomly divided into a training set (320 groups, accounting for 80%) and a test set (80 groups, accounting for 20%).

[0047] Step S2: Time series feature extraction, input the time series data of the training set and the test set into the time series convolution network TCN for feature extraction, the time series convolution network TCN contains multiple layers of one-dimensional causal convolution layers, and the multi-scale time series dependence is captured by setting the inflation rate of the inflation convolution to increase, and the high-dimensional time series feature sequence is output, specifically:

[0048] The time series convolution network (TCN) is used to automatically learn multi-scale leakage features from the preprocessed time series. Each time series sample of the training set and the test set is input into the TCN model. The TCN structure of the embodiment is as shown in part a of Figure 3

[0049] The number of layers of the TCN structure is 4 layers of one-dimensional causal convolution (Causal Convolution).

[0050] The inflation rate (Dilation Rate) of the TCN structure: increasing layer by layer, set to 1, 2, 4, 8. That is, the inflation rate of the first layer is 1, the inflation rate of the second layer is 2, the inflation rate of the third layer is 4, and the inflation rate of the fourth layer is 8. The inflation rate of the first layer and the inflation rate of the second layer can be considered as low inflation rate, which is used to capture the local mutation features (such as the initial burst signal of the leakage) of the adjacent time points, while the inflation rate of the third layer and the inflation rate of the fourth layer can be considered as high inflation rate, which is used to perceive the dependence relationship in a longer historical range (such as the diffusion trend of concentration under different ventilation conditions).

[0051] The convolution kernel size of the TCN structure is 5x1.

[0052] The activation function of the TCN structure is ReLU.

[0053] The output of the TCN structure: after 4 layers of convolution, each input time series sample is converted into a 256-dimensional high-dimensional time series feature sequence. This feature sequence contains deep time series patterns related to leakage in the original data.

[0054] Step S3: Feature enhancement and compression, input the high-dimensional time series feature sequence output by S2 into the multi-head self-attention mechanism module, dynamically enhance the key features related to the leakage sensitive area by calculating the attention weight of different time steps, and then perform pooling operation along the time dimension to compress the fixed-length feature vector, specifically:

[0055] S3 is mainly used to enhance the key features and reduce the data dimension to adapt to the subsequent classifier. The 256-dimensional time series feature sequence output by S2 is input into the multi-head self-attention mechanism (Multi-Head Self-Attention, MHSA) module, as shown in part b of Figure 3 ​​

[0056] The multi-head self-attention mechanism is: 4 attention heads (Head) are set, and the key vector (Key) dimension is 64. Each head independently calculates the attention weight between different time steps in the input sequence. This mechanism can automatically evaluate the importance of each time step feature, and assign higher weights to key time steps (such as time points where leakage features change significantly) or key feature channels that are highly related to leakage judgment, thereby dynamically enhancing these key features and suppressing secondary or noise features.

[0057] The feature sequence (still 256 dimensions) enhanced by the multi-head self-attention mechanism is input into the pooling layer. In this embodiment, a global average pooling (Global Average Pooling) operation is used: the average value of each feature channel is calculated along the entire time dimension. This operation compresses the variable-length time sequence feature sequence (length depends on the input segment length) into a fixed-length vector (128 dimensions), which contains statistical information of the features in the entire time segment, while significantly reducing the data dimension.

[0058] Step S4: Classification and recognition model training and testing, input the fixed-length feature vector of the training set obtained into the XGBoost classifier for training, and obtain the trained XGBoost leakage recognition model; input the fixed-length feature vector of the test set into the trained XGBoost leakage recognition model, output the corresponding leakage level prediction label and leakage position prediction label, and complete the hydrogen leakage recognition.

[0059] Step S4 uses the XGBoost classifier to complete the final leakage recognition task, as follows:

[0060] Training of the XGBoost classifier: input the 128-dimensional feature vector of the training set sample obtained in S3 and its corresponding leakage level label and leakage position label into the XGBoost model for training. For the leakage position recognition XGBoost classifier, the main parameter settings are: learning rate = 0.1, maximum depth of tree = 8, and number of decision trees = 161. For the leakage level recognition XGBoost classifier, the main parameter settings are: learning rate = 0.05, maximum depth of tree = 8, and number of decision trees = 321. To improve the performance of the model, a feature enhancement strategy is used during training: the preliminary diagnosis prediction result (probability distribution) generated by the first layer of decision trees is spliced with the original 128-dimensional feature vector to form an enhanced feature vector; the enhanced feature vector is input into an autoencoder (Autoencoder) for feature transformation; the enhanced features output by the autoencoder are input into the next layer of decision trees for more detailed diagnosis. This process is iterated until the accuracy of the model on the validation set no longer improves significantly, and finally a trained XGBoost leakage recognition model is obtained.

[0061] Test the trained XGBoost leakage identification model: input the 128-dimensional feature vector of the test set sample obtained in S3 into the trained XGBoost model. The model will output two prediction results for each test sample: one is the prediction label of the leakage level (Level 1-5), and the other is the prediction label of the leakage position (leakage point 1-4).

[0062] Step S5: Performance evaluation, based on the leakage level prediction label and the leakage position prediction label obtained in step S4, compare with the corresponding true label, calculate the quantitative evaluation index, wherein:

[0063] The quantitative evaluation index includes overall accuracy (Accuracy), recall rate (Recall) and F1 score (F1-score) calculated by category. The position recognition accuracy of this embodiment is 93.33%, and the level recognition accuracy is 94.17%. The recall rate and F1 score can be calculated according to the leakage level or position category, which can more accurately evaluate the recognition ability of the model for each category (especially when the data is unbalanced).

[0064] Visualize the classification results and intuitively display the distribution of the classification results by drawing a confusion matrix. As shown in Figure 4 Part a of the confusion matrix for leakage position recognition is shown in Figure 4 Part b of the confusion matrix for leakage level recognition is shown in. The higher the value on the diagonal line of the confusion matrix (the number of correct classifications), the higher the recognition accuracy of the corresponding category; the non-diagonal cells reveal the main error classification patterns of the model (such as which categories are easily confused).

[0065] The advantages of the present application are:

[0066] This embodiment effectively solves the problems of insufficient recognition accuracy and poor real-time performance of the prior art in complex sealed spaces and multi-factor leakage scenarios by combining the advantages of TCN, MHSA and XGBoost. The rich data set constructed by CFD simulation ensures the coverage of the model for multiple working conditions. Experimental results show that this method achieves a high accuracy of more than 93% in leakage position and level recognition, significantly better than traditional methods, and the model has good robustness and interpretability, meeting the urgent needs of fuel cell vehicles for real-time and accurate monitoring of hydrogen safety.

[0067] The above is only the preferred embodiment of the present application, it should be pointed out that, for those skilled in the art, without departing from the principles of the present application, can also make a number of improvements and refinements, such as adjusting the TCN layer or expansion rate, change the number of MHSA head, optimization XGBoost parameters, increase other sensor data source (such as temperature, pressure) fusion, etc., these improvements and refinements should also be considered as the protection scope of the present application.

[0068] A leakage identification electronic device, comprising a processor and a memory, the memory stores computer readable instructions, when the computer readable instructions are executed by the processor, the steps of the data-driven fuel cell vehicle hydrogen storage tank hydrogen leakage identification method are run.

[0069] A storage medium, having a computer program stored thereon, the computer program is executed by a processor to run the steps of the data-driven fuel cell vehicle hydrogen storage tank hydrogen leakage identification method.

[0070] The above series of detailed description is only for the specific description of the feasible embodiment of the present application, and is not used to limit the protection scope of the present application. Any equivalent embodiment or change made without departing from the spirit of the present application shall be included in the protection scope of the present application.

Claims

1. A data-driven method for identifying hydrogen leaks in the hydrogen storage tank of a fuel cell vehicle, characterized in that, Includes the following steps: S1: Dataset Construction and Preprocessing: Based on CFD simulation, multidimensional hydrogen leakage time series data covering different ventilation conditions, different leakage levels, different leakage directions, and different leakage locations are generated to form a dataset; the time series data is subjected to sliding window segmentation and standardization, and the corresponding leakage level labels and leakage location labels are labeled; the processed dataset is divided into training set and test set; S2: Temporal Feature Extraction: The time-series data of the training set and the test set are input into the temporal convolutional network TCN for feature extraction. The temporal convolutional network TCN contains multiple one-dimensional causal convolutional layers. By setting the dilation rate of the dilation convolution, multi-scale temporal dependencies are captured, and high-dimensional temporal feature sequences are output. S3: Feature Enhancement and Compression: The high-dimensional temporal feature sequence output from S2 is input into the multi-head self-attention mechanism module. By calculating the attention weights at different time steps, key features related to the leakage sensitive area are dynamically enhanced. Then, the attention-enhanced feature sequence is pooled along the time dimension to compress it into a fixed-length feature vector. S4: Classification and Recognition Model Training and Testing: Input the fixed-length feature vector of the training set obtained in S3 into the XGBoost classifier for training to obtain the trained XGBoost leak recognition model; input the fixed-length feature vector of the test set into the trained XGBoost leak recognition model to output the corresponding leak level prediction label and leak location prediction label, thus completing the hydrogen leak recognition.

2. The data-driven hydrogen leakage identification method for hydrogen storage tanks of fuel cell vehicles according to claim 1, characterized in that, In step S1, the different ventilation conditions include enclosed space, natural ventilation, and forced ventilation with different wind speeds; the different leakage levels include multiple preset leakage severity levels; the different leakage directions include up, down, left, and right; and the different leakage locations include multiple preset points on the hydrogen storage tank.

3. The data-driven hydrogen leakage identification method for hydrogen storage tanks of fuel cell vehicles according to claim 1, characterized in that, In step S2, the temporal convolutional network TCN contains at least 4 one-dimensional causal convolutional layers with an inflation rate sequence of 1, 2, 4, 8; and the kernel size is set to 5×1.

4. The data-driven hydrogen leakage identification method for hydrogen storage tanks of fuel cell vehicles according to claim 1, characterized in that, In step S3, the multi-head self-attention mechanism uses 4 heads and the dimension of the key vector is set to 64; the pooling operation is global average pooling.

5. The data-driven hydrogen leakage identification method for hydrogen storage tanks of fuel cell vehicles according to claim 1, characterized in that, In step S4, the XGBoost classifier is trained using gradient boosting decision trees, with the learning rate, maximum depth, number of decision trees, and multi-class objective function set. Specifically, for the leak location recognition XGBoost classifier, the learning rate is set to 0.1, the maximum depth is set to 8, and the number of decision trees is set to 161; for the leak level recognition XGBoost classifier, the learning rate is set to 0.05, the maximum depth is set to 8, and the number of decision trees is set to 321.

6. The data-driven hydrogen leakage identification method for hydrogen storage tanks of fuel cell vehicles according to claim 1, characterized in that, The training process of the XGBoost classifier in step S4 includes: concatenating the preliminary diagnostic results generated by the first-layer decision tree with the original fixed-length feature vector obtained in S3, and inputting it into the autoencoder to generate enhanced features; inputting the enhanced features into the lower-layer decision tree for further diagnostic optimization; iteratively executing this process until the model accuracy converges, and outputting the final XGBoost leak detection model.

7. The data-driven hydrogen leakage identification method for hydrogen storage tanks of fuel cell vehicles according to claim 1, characterized in that, The method further includes: S5: Performance Evaluation: Based on the leak level prediction label and leak location prediction label obtained in step S4, compare them with the corresponding real labels, calculate the accuracy, recall, and F1 score, and evaluate the model's recognition performance.

8. A leakage detection electronic device, characterized in that, It includes a processor and a memory, the memory storing computer-readable instructions, which, when executed by the processor, perform the steps of the data-driven hydrogen storage tank hydrogen leakage identification method for fuel cell vehicles as described in any one of claims 1-7.

9. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it performs the steps in the data-driven hydrogen leakage identification method for hydrogen storage tanks of fuel cell vehicles as described in any one of claims 1-7.

Images