Hydrogen energy storage equipment leakage fault early warning method and system based on deep learning

Abstract

The invention provides a hydrogen energy storage equipment leakage fault early warning method and system based on deep learning, and the method comprises the steps: collecting the multi-source sensing data of hydrogen energy storage equipment in real time, constructing a multi-dimensional feature matrix, building the nonlinear mapping of pressure fluctuation and gas concentration through a bidirectional interaction network, and carrying out the early warning of the leakage fault of the hydrogen energy storage equipment. And performing spatial modeling on the temperature gradient and the vibration spectrum to generate a fusion feature tensor. And then, inputting the fused feature tensor into a spatio-temporal reasoning model, predicting equipment leakage and energy loss, and outputting a leakage probability cloud picture and an energy loss vector. And an entropy sudden change area is identified based on an energy loss vector, when the high-density leakage area and the entropy sudden change area are overlapped in a continuous detection period, a multi-stage early warning signal is triggered, a leakage source thermodynamic diagram and a pressure balance parameter set are generated, a self-adjusting system is activated to dynamically correct the valve opening degree, network weight is updated, and a closed-loop control link is formed. The early warning accuracy and the response speed of the leakage fault of the hydrogen energy storage equipment are improved.

Description

Technical Field

[0001] The embodiments of the present application relate to the technical field of deep learning, and in particular, to a leakage fault warning method and system for hydrogen energy storage equipment based on deep learning. Background Art

[0002] Hydrogen energy storage equipment has important applications in the new energy field, but during its operation, energy loss or even safety accidents may occur due to leakage faults.

[0003] Currently, there are leakage detection schemes based on single-sensor data analysis and machine learning algorithms. For example, support vector machines or random forest algorithms are used to classify pressure or gas concentration data to determine whether there is a leakage risk. Such schemes can identify leakage characteristics to a certain extent and trigger warning signals by training historical data models.

[0004] Existing schemes mainly rely on single-type sensor data and fail to make full use of the complementarity of multi-source heterogeneous data, resulting in insufficient accuracy and robustness of leakage detection. In addition, existing schemes lack refined modeling of leakage diffusion paths and energy losses, cannot achieve leakage source localization and dynamic adjustment, and are difficult to meet the real-time warning and control requirements under complex working conditions. Summary of the Invention

[0005] The embodiments of the present application provide a leakage fault warning method and system for hydrogen energy storage equipment based on deep learning to solve the problems of poor warning accuracy and slow response speed of leakage faults in hydrogen energy storage equipment in the prior art.

[0006] In a first aspect, the embodiments of the present application provide a leakage fault warning method for hydrogen energy storage equipment based on deep learning, including: Collecting multi-source heterogeneous sensing data streams of hydrogen energy storage equipment in real time to construct a multi-dimensional feature matrix, where the sensing data streams include pressure fluctuation sequences, temperature gradient distribution fields, vibration spectrum waveforms, and gas concentration spectrograms; Inputting the multi-dimensional feature matrix into a bidirectional interaction network to establish a non-linear mapping between the pressure fluctuation sequence and the gas concentration spectrogram through a cross-attention module in the bidirectional interaction network, and using a gradient field reconstruction algorithm to perform spatial propagation modeling on the temperature gradient distribution field and the vibration spectrum waveform to generate a fusion feature tensor; Inputting the fusion feature tensor into a spatio-temporal inference model to predict hydrogen energy storage equipment leakage and energy loss using the spatio-temporal inference model, and outputting a leakage probability cloud map and an energy loss vector; Determine the entropy value mutation region based on the energy loss vector. When the high-density region in the leakage probability cloud map overlaps with the entropy value mutation region in consecutive detection periods, trigger a multi-level warning signal and generate a leakage source thermal map and a pressure balance parameter set; Activate the self-regulating system according to the leakage source thermal map and the pressure balance parameter set, so that the self-regulating system dynamically corrects the valve opening gradient of adjacent hydrogen energy storage devices and updates the node weights of the bidirectional interaction network to generate a closed-loop link.

[0007] Optionally, the spatio-temporal inference model includes a multi-layer perception constraint module, an implicit state transition model, a gas diffusion physical constraint module, and a probability density clustering module connected in sequence; Inputting the fusion feature tensor into the spatio-temporal inference model to predict the leakage and energy loss of the hydrogen energy storage device by using the spatio-temporal inference model, and outputting a leakage probability cloud map and an energy loss vector, includes: Analyze the internal structure connection relationship of the hydrogen energy storage device through the multi-layer perception constraint module to generate a topological graph of the energy transfer path; Based on the topological graph of the energy transfer path, track the energy flow field feature matrix through the dynamic recursive operator in the implicit state transition model, and the dynamic recursive operator adaptively adjusts the state transition step according to the coupling effect of the temperature gradient distribution field and the vibration frequency spectrum waveform; Input the energy flow field feature matrix into the gas diffusion physical constraint module, and construct a three-dimensional space grid coordinate system in combination with real-time environmental parameters; Simulate the diffusion path of the leaked gas in the three-dimensional space grid coordinate system through the mass conservation equation and the momentum transfer theorem to generate the diffusion path probability distribution of the leaked gas; Generate a fusion leakage situation map according to the diffusion path probability distribution; Based on the fusion leakage situation map, generate a leakage probability cloud map through the probability density clustering module, and construct an energy loss vector according to the energy flow field feature matrix.

[0008] Optionally, based on the topological graph of the energy transfer path, track the energy flow field feature matrix through the dynamic recursive operator in the implicit state transition model, and the dynamic recursive operator adaptively adjusts the state transition step according to the coupling effect of the temperature gradient distribution field and the vibration frequency spectrum waveform, includes: Decompose the temperature gradient distribution field into a heat conduction component and a thermal stress component through a preset thermo-elastic coupling equation, and generate a mechanical vibration influence factor matrix based on the frequency domain energy distribution of the vibration frequency spectrum waveform; Perform a two-channel convolution fusion operation on the heat conduction component and the mechanical vibration influence factor matrix to generate a temperature-vibration coupling coefficient matrix, where the dimension of the coupling coefficient matrix is consistent with the number of nodes in the energy transfer path topology graph; Perform state transition calculation with physical constraints on the energy transfer path topology graph, and iteratively update the energy flow field feature matrix. During each iteration, the dynamic recursion operator dynamically adjusts the state transition step size according to the spectral radius of the temperature-vibration coupling coefficient matrix.

[0009] Optionally, before performing the state transition calculation with physical constraints on the energy transfer path topology graph, the method further includes: Perform frequency-domain energy spectrum analysis on the pressure fluctuation sequence to obtain an analysis result; Extract energy mutation indices in different frequency bands from the analysis result through a band-pass filtering window to construct a pressure frequency-domain mutation feature vector; And calculate the spatial gradient direction weight coefficient based on a preset leakage diffusion direction constraint condition; Input the pressure frequency-domain mutation feature vector and the spatial gradient direction weight coefficient into a two-channel attention gating module to generate an initial value of the time-varying gain factor through the non-linear mapping function in the two-channel attention gating module; Perform dynamic normalization processing on the initial value of the time-varying gain factor using a leakage risk index threshold curve to generate a normalized time-varying gain factor; Perform a tensor multiplication operation on the normalized time-varying gain factor and the temperature-vibration coupling coefficient matrix to generate a dynamic weight parameter matrix; Embed the dynamic weight parameter matrix as a physical constraint into the initial state transition equation to obtain a state transition equation with physical constraints.

[0010] Optionally, the inputting the pressure frequency-domain mutation feature vector and the spatial gradient direction weight coefficient into a two-channel attention gating module to generate an initial value of the time-varying gain factor through the non-linear mapping function in the two-channel attention gating module includes: Perform a multi-scale decomposition operation on the pressure frequency-domain mutation feature vector to obtain multi-scale decomposition information; Perform wavelet packet transform on the multi-scale decomposition information to extract energy mutation intensity coefficients in different frequency bands; Construct a pressure frequency-domain energy distribution tensor according to the energy mutation intensity coefficients in different frequency bands; Perform directional enhancement processing on the spatial gradient direction weight coefficient to obtain a processed spatial gradient direction weight coefficient; Based on the processed spatial gradient direction weight coefficients, an anisotropic diffusion filtering algorithm is used to generate a gradient direction consistency matrix; Based on the gradient direction consistency matrix and a preset leakage diffusion direction constraint condition, a direction weight correction factor is calculated; The pressure frequency-domain energy distribution tensor and the direction weight correction factor are input into a dual-channel attention gating module, so as to establish a non-linear correlation mapping between the pressure frequency domain and the spatial gradient through a cross-attention module in the dual-channel attention gating module, and an initial attention weight matrix is generated; A non-linear activation operation is performed on the initial attention weight matrix to generate an initial value of the time-varying gain factor.

[0011] Optionally, inputting the multi-dimensional feature matrix into a bidirectional interaction network to establish a non-linear mapping between the pressure fluctuation sequence and the gas concentration spectrogram through a cross-attention module in the bidirectional interaction network, and using a gradient field reconstruction algorithm to perform spatial propagation modeling on the temperature gradient distribution field and the vibration frequency spectrum waveform to generate a fused feature tensor, including: Performing time-frequency joint analysis on the pressure fluctuation sequence to generate a pressure fluctuation time-frequency feature matrix; Performing spatial multi-resolution analysis on the gas concentration spectrogram to generate a gas concentration multi-scale feature map; The pressure fluctuation time-frequency feature matrix and the gas concentration multi-scale feature map are input into a cross-attention module to establish a correlation mapping between the pressure frequency-domain feature and the gas concentration spatial feature through a multi-head attention module, and a pressure concentration interaction feature matrix is generated; Generating a temperature gradient propagation path map according to the temperature gradient distribution field; Extracting the intrinsic mode function components of the vibration frequency spectrum waveform through an empirical mode decomposition algorithm, and generating a vibration frequency spectrum energy feature vector based on the intrinsic mode function components; Generating a temperature vibration interaction feature matrix according to the temperature gradient propagation path map and the vibration frequency spectrum energy feature vector; Performing a tensor splicing operation on the pressure concentration interaction feature matrix and the temperature vibration interaction feature matrix to generate a fused feature tensor.

[0012] Optionally, determining an entropy value mutation region based on the energy loss vector. When the high-density region in the leakage probability cloud map and the entropy value mutation region spatially overlap within consecutive detection periods, a multi-level warning signal is triggered and a leakage source thermal map and a pressure balance parameter set are generated, including: Performing a multi-dimensional decomposition operation on the energy loss vector to generate an energy loss feature matrix; Identifying abnormal energy entropy values according to the energy loss feature matrix to generate an entropy value mutation region; Perform spatial clustering analysis on the leakage probability cloud map to generate high-density regions; When the high-density region and the entropy value mutation region spatially overlap within at least two consecutive detection periods, trigger a multi-level warning signal; Generate a leakage source heat map corresponding to the high-density region, and calculate the pressure compensation parameters of multiple adjacent hydrogen energy storage devices according to the leakage source heat map and a preset pressure balance equation to generate a pressure balance parameter set.

[0013] In a second aspect, an embodiment of the present application provides a hydrogen energy storage device leakage fault warning system based on deep learning, including: A collection module, configured to collect multi-source heterogeneous sensing data streams of hydrogen energy storage devices in real time to construct a multi-dimensional feature matrix, where the sensing data streams include pressure fluctuation sequences, temperature gradient distribution fields, vibration spectrum waveforms, and gas concentration spectrograms; An input module, configured to input the multi-dimensional feature matrix into a bidirectional interaction network, establish a non-linear mapping between the pressure fluctuation sequence and the gas concentration spectrogram through a cross-attention module in the bidirectional interaction network, and perform spatial propagation modeling on the temperature gradient distribution field and the vibration spectrum waveform by using a gradient field reconstruction algorithm to generate a fused feature tensor; An output module, configured to input the fused feature tensor into a spatio-temporal inference model to predict the leakage and energy loss of hydrogen energy storage devices by using the spatio-temporal inference model, and output a leakage probability cloud map and an energy loss vector; A trigger module, configured to determine an entropy value mutation region based on the energy loss vector, and when the high-density region in the leakage probability cloud map and the entropy value mutation region spatially overlap within consecutive detection periods, trigger a multi-level warning signal and generate a leakage source heat map and a pressure balance parameter set; An activation module, configured to activate a self-regulating system according to the leakage source heat map and the pressure balance parameter set, so that the self-regulating system dynamically corrects the valve opening gradient of adjacent hydrogen energy storage devices and updates the node weights of the bidirectional interaction network to generate a closed-loop link.

[0014] In a third aspect, an embodiment of the present application provides a computing device, including a processor and a memory, where a computer program is stored in the memory, and the processor is configured to run the computer program to execute any one of the methods for warning of hydrogen energy storage device leakage faults based on deep learning in the first aspect.

[0015] In a fourth aspect, an embodiment of the present application provides a computer storage medium, on which computer program instructions are stored, and when the computer program instructions are executed by a processor, the method for warning of hydrogen energy storage device leakage faults based on deep learning according to any one of the first aspects is implemented.

[0016] In an embodiment of the present application, a method for early warning of leakage faults of hydrogen energy storage devices based on deep learning is provided. The method includes: collecting multi-source heterogeneous sensing data streams of hydrogen energy storage devices in real time to construct a multi-dimensional feature matrix, where the sensing data streams include pressure fluctuation sequences, temperature gradient distribution fields, vibration spectrum waveforms, and gas concentration spectrograms; inputting the multi-dimensional feature matrix into a bidirectional interaction network to establish a non-linear mapping between the pressure fluctuation sequence and the gas concentration spectrogram through a cross-attention module in the bidirectional interaction network, and using a gradient field reconstruction algorithm to perform spatial propagation modeling on the temperature gradient distribution field and the vibration spectrum waveform to generate a fused feature tensor; inputting the fused feature tensor into a spatio-temporal inference model to predict leakage and energy loss of hydrogen energy storage devices using the spatio-temporal inference model, and outputting a leakage probability cloud map and an energy loss vector; determining an entropy value mutation region based on the energy loss vector, and when the high-density region in the leakage probability cloud map spatially overlaps with the entropy value mutation region within consecutive detection periods, triggering a multi-level warning signal and generating a leakage source heat map and a pressure balance parameter set; activating a self-regulating system according to the leakage source heat map and the pressure balance parameter set, so that the self-regulating system dynamically corrects the valve opening gradient of adjacent hydrogen energy storage devices and updates the node weights of the bidirectional interaction network to generate a closed-loop link.

[0017] The technical solution provided by the present application has the following beneficial effects: By integrating multi-source data such as pressure fluctuation sequences, temperature gradient distribution fields, vibration spectrum waveforms, and gas concentration spectrograms, the present application constructs a multi-dimensional feature matrix, providing a comprehensive and high-dimensional data basis for subsequent analysis, and improving the comprehensiveness and accuracy of leakage detection. Using a cross-attention module to establish a non-linear mapping between pressure fluctuations and gas concentrations, and combining a gradient field reconstruction algorithm to perform spatial modeling on temperature gradients and vibration spectra to generate a fused feature tensor, realizing the deep fusion of multi-source data and enhancing the feature expression ability. Accurately predicting leakage and energy loss through a spatio-temporal inference model, generating a leakage probability cloud map and an energy loss vector, providing a basis for the quantitative analysis and positioning of leakage risks. By identifying the entropy value mutation region and performing spatial overlap analysis with the high-density region of the leakage probability cloud map, early warning of leakage risks is realized, triggering a multi-level warning signal, and improving the timeliness and accuracy of the warning. According to the leakage source heat map and the pressure balance parameter set, dynamically correcting the valve opening gradient of adjacent devices and updating the node weights of the bidirectional interaction network to form a closed-loop control link, realizing the dynamic regulation of leakage risks and system optimization.

[0018] Furthermore, the embodiment of the present application also analyzes the internal structure connection relationship of the hydrogen energy storage device through a multi-layer perception constraint module to generate a topological graph of the energy transfer path; based on the topological graph, uses the dynamic recursion operator in the implicit state transfer model to track the energy flow field feature matrix, and adaptively adjusts the state transfer step size in combination with the coupling effect of the temperature gradient and the vibration spectrum; inputs the energy flow field feature matrix into the gas diffusion physical constraint module, constructs a three-dimensional space grid coordinate system, and simulates the diffusion path of the leaked gas through the mass conservation equation and the momentum transfer theorem to generate a probability distribution of the diffusion path; generates a fusion leakage situation map based on the probability distribution of the diffusion path, and finally generates a leakage probability cloud map through the probability density clustering module and constructs an energy loss vector.

[0019] Moreover, this step accurately tracks the energy flow field features through the multi-layer perception constraint and the implicit state transfer model, and combines the gas diffusion physical constraint module to realize the three-dimensional simulation of the diffusion path of the leaked gas, generating a high-precision leakage probability cloud map and an energy loss vector. This method improves the accuracy of leakage positioning and the refinement degree of energy loss analysis, providing a reliable basis for the quantitative assessment and dynamic regulation of leakage risks.

[0020] These aspects or other aspects of the present application will be more clearly understood in the following description of the embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0022] Figure 1 It is a flowchart of a method for warning of leakage faults of a hydrogen energy storage device based on deep learning provided by an embodiment of the present application; Figure 2 It is a schematic structural diagram of a system for warning of leakage faults of a hydrogen energy storage device based on deep learning provided by an embodiment of the present application; Figure 3 It is a schematic structural diagram of a computing device provided by an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0023] In order to enable those skilled in the art to better understand the solution of the present application, the following will clearly and completely describe the technical solutions in the embodiments of the present application in conjunction with the drawings in the embodiments of the present application.

[0024] In some processes described in the specification, claims, and the above-mentioned drawings of this application, multiple operations appear in a specific order. However, it should be clearly understood that these operations can be executed not in the order in which they appear herein or in parallel. The serial numbers of the operations, such as 101, 102, etc., are only used to distinguish different operations, and the serial numbers themselves do not represent any execution order. Additionally, these processes may include more or fewer operations, and these operations can be executed sequentially or in parallel. It should be noted that the descriptions such as "first", "second", etc. in this text are used to distinguish different messages, devices, modules, etc., and do not represent a sequence, nor do they limit that "first" and "second" are of different types.

[0025] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without making creative efforts belong to the scope of protection of the present application.

[0026] The research and development idea of this solution is to collect multi-source heterogeneous sensing data of hydrogen energy storage devices in real time, construct a multi-dimensional feature matrix, and use the cross-attention module in the bidirectional interaction network to establish a non-linear mapping between pressure fluctuations and gas concentrations. At the same time, a gradient field reconstruction algorithm is used to model the spatial propagation of temperature gradients and vibration spectra to generate a fused feature tensor. Then, the fused feature tensor is input into a spatio-temporal inference model to predict device leakage and energy loss, and output a leakage probability cloud map and an energy loss vector. Based on the energy loss vector, the entropy mutation region is identified. When the high-density region in the leakage probability cloud map spatially overlaps with the entropy mutation region in consecutive detection periods, a multi-level warning signal is triggered, and a leakage source thermal map and a pressure balance parameter set are generated. Finally, according to the leakage source thermal map and the pressure balance parameter set, the self-regulating system is activated to dynamically correct the valve opening gradient of adjacent devices and update the node weights of the bidirectional interaction network, forming a closed-loop control link to achieve precise early warning and dynamic regulation of leakage risks.

[0027] Figure 1 The flowchart of a leakage fault early warning method for hydrogen energy storage devices based on deep learning provided by the embodiments of the present application is as Figure 1 shown, and the method includes: Step 101: Collect multi-source heterogeneous sensing data streams of hydrogen energy storage devices in real time to construct a multi-dimensional feature matrix.

[0028] In this step, the sensed data stream includes a pressure fluctuation sequence, a temperature gradient distribution field, a vibration spectrum waveform, and a gas concentration spectrogram. The pressure fluctuation sequence refers to the continuous data of the internal pressure of the device changing with time, which is used to reflect the stability of the device operation state. The temperature gradient distribution field refers to the spatial distribution of the internal temperature of the device, which is used to analyze heat conduction and thermal stress. The vibration spectrum waveform refers to the distribution of the device vibration signal in the frequency domain, which is used to identify mechanical vibration characteristics. The gas concentration spectrogram refers to the spatial distribution of the gas concentration around the device, which is used to detect the diffusion of leaked gas. The multi-dimensional feature matrix refers to integrating multi-source heterogeneous data into a high-dimensional matrix for subsequent analysis and modeling.

[0029] In the embodiment of the present application, through a variety of sensors installed on the hydrogen energy storage device (such as pressure sensors, temperature sensors, vibration sensors, gas concentration sensors), data such as the pressure fluctuation sequence, the temperature gradient distribution field, the vibration spectrum waveform, and the gas concentration spectrogram are collected in real time; these data are aligned according to the time stamp and spatial position to construct a multi-dimensional feature matrix, where each row of the matrix represents the multi-source data at a time point, and each column represents the feature dimension of a certain type of sensor data.

[0030] For example, in a large hydrogen storage station, engineers installed a variety of sensors at key parts of the hydrogen storage tank to monitor the state changes of hydrogen in real time. Through the use of a dedicated data acquisition system, these raw data are transmitted to the central server and converted into a multi-dimensional feature matrix, providing a basis for further analysis.

[0031] Step 102: Input the multi-dimensional feature matrix into a bidirectional interaction network to establish a non-linear mapping between the pressure fluctuation sequence and the gas concentration spectrogram through the cross-attention module in the bidirectional interaction network, and use the gradient field reconstruction algorithm to perform spatial propagation modeling on the temperature gradient distribution field and the vibration spectrum waveform to generate a fused feature tensor.

[0032] In this step, the bidirectional interaction network refers to a deep learning model that can simultaneously process the bidirectional interaction relationship between multi-source data. The cross-attention module is used to establish a non-linear mapping relationship between different data features, such as the correlation between pressure fluctuation and gas concentration. The fused feature tensor refers to a high-dimensional tensor generated after deeply fusing the features of multi-source data for subsequent prediction tasks.

[0033] In the embodiment of the present application, the multi-dimensional feature matrix is input into the bidirectional interaction network, and the non-linear mapping relationship between the pressure fluctuation sequence and the gas concentration spectrogram is calculated through the cross-attention module; at the same time, the gradient field reconstruction algorithm is used to perform spatial propagation modeling on the temperature gradient distribution field and the vibration spectrum waveform to analyze their coupling effect; finally, the pressure-gas concentration interaction feature and the temperature-vibration interaction feature are tensor spliced to generate a fused feature tensor.

[0034] For example, based on the multi-dimensional feature matrix obtained in the previous step, the researchers used a two-way interaction network to conduct an in-depth analysis of this data. In particular, through the cross-attention module, the relationship between the pressure fluctuations and gas concentration anomalies at potential leakage risk points was identified, and the possible fault areas were determined through the gradient field reconstruction algorithm, generating a fused feature tensor.

[0035] Step 103: Input the fused feature tensor into a spatio-temporal inference model to use the spatio-temporal inference model to predict hydrogen energy storage device leakage and energy loss, and output a leakage probability cloud map and an energy loss vector.

[0036] In this step, the spatio-temporal inference model is a deep learning model that combines temporal and spatial features for predicting leakage and energy loss. The leakage probability cloud map refers to a distribution map of the leakage probability in the space inside or around the device, used to locate the leakage risk area. The energy loss vector refers to a vector representing the energy loss situation of the device, used to quantify the impact of leakage on the device performance.

[0037] In the embodiment of the present application, the fused feature tensor is input into the spatio-temporal inference model. The model analyzes temporal and spatial features to predict the device leakage probability and energy loss; the leakage probability cloud map is generated through probability density clustering, reflecting the distribution of leakage risks; the energy loss vector is generated by analyzing the energy flow field feature matrix to quantify the energy loss caused by leakage.

[0038] For example, after analyzing the fused feature tensor using the spatio-temporal inference model, the system successfully predicted several high-risk leakage areas and calculated the corresponding energy loss vectors, providing an important basis for subsequent risk assessment.

[0039] Step 104: Determine the entropy mutation region based on the energy loss vector. When the high-density region in the leakage probability cloud map spatially overlaps with the entropy mutation region within consecutive detection periods, trigger a multi-level warning signal and generate a leakage source heat map and a pressure balance parameter set.

[0040] In this step, the entropy mutation region refers to the region where the entropy value changes in the energy loss vector, indicating abnormal energy fluctuations. The high-density region refers to the region in the leakage probability cloud map where the data points are densely distributed and exceed the preset density threshold. The basis for distinguishing high and low densities is to calculate the local density of the data points through a spatial clustering algorithm and set a density threshold. The region above the threshold is the high-density region, indicating a higher leakage probability; the region below the threshold is the low-density region, indicating a lower leakage probability. Specific numerical examples are as follows: set the minimum number of points to 5 and the neighborhood radius to 0.5. If the number of data points per unit area in a certain region exceeds 10, it is determined as a high-density region; if it is less than 5, it is a low-density region. For example, if there are 12 data points per square meter in a certain region, exceeding the threshold of 10, it is marked as a high-density region. The multi-level warning signal is a warning mechanism triggered hierarchically according to the severity of the leakage risk. Different levels of warning signals correspond to different response measures to ensure timely and effective handling of the leakage risk. Different levels of warning correspond to different prompting methods. Specifically, the prompting methods corresponding to the first-level warning (low risk) are: voice prompt, prompt box, etc. Example: The system issues a voice prompt "Potential leakage risk detected, please check the equipment" and pops up a prompt box on the monitoring interface. The prompting methods corresponding to the second-level warning (medium risk) are: SMS notification, light warning, etc. Example: The system sends an SMS to relevant personnel "Medium leakage risk detected, please check the equipment immediately" and activates a yellow warning light at the equipment site. The prompting methods corresponding to the third-level warning (high risk) are: emergency alarm, automatic system shutdown. Example: The system triggers an emergency alarm sound, sends an SMS "High leakage risk detected, the system will be automatically shut down", and automatically shuts down relevant equipment to prevent further leakage.

[0041] In the embodiment of the present application, the entropy mutation region is calculated based on the energy loss vector to identify abnormal energy fluctuations; spatial clustering analysis is performed on the leakage probability cloud map to determine the high-density region; when the high-density region and the entropy mutation region spatially overlap in consecutive detection cycles, a multi-level warning signal is triggered, and a leakage source heat map and a pressure balance parameter set are generated.

[0042] For example, during the detection process, the system found that the energy loss in some areas increased, indicating that there might be a leakage risk. Further analysis confirmed that these areas coincided with the high-density areas on the leakage probability cloud map. Immediately, the warning program was activated, and a detailed leakage source heat map and adjustment suggestions were generated.

[0043] Step 105: Activate the self-regulating system according to the leakage source heat map and the pressure balance parameter set, so that the self-regulating system dynamically corrects the valve opening gradient of adjacent hydrogen energy storage devices and updates the node weights of the bidirectional interaction network to generate a closed-loop link.

[0044] In this step, the self-regulating system refers to a system that can dynamically adjust the operating parameters of equipment according to the leakage risk. The valve opening gradient refers to the opening adjustment strategy of adjacent equipment valves, which is used to balance the pressure distribution. The closed-loop link is a control link that realizes the dynamic adjustment and optimization of the system through a feedback mechanism.

[0045] In the embodiments of the present application, the leakage source location is determined according to the leakage source heat map, and the valve opening gradient of adjacent equipment is calculated in combination with the pressure balance parameter set; the self-regulating system dynamically adjusts the valve opening to balance the pressure distribution; at the same time, the node weights of the two-way interaction network are updated to optimize the model performance, forming a closed-loop control link.

[0046] For example, after receiving the warning signal, the self-regulating system immediately takes action to adjust the valve opening of the relevant hydrogen storage tank, alleviating the local overpressure condition. At the same time, by continuously learning and adjusting the network weights, the safety and efficiency of the entire system are improved.

[0047] This solution accurately predicts leakage and energy loss by collecting multi-source heterogeneous data in real time, constructing a multi-dimensional feature matrix, and using a two-way interaction network and a spatio-temporal reasoning model; based on the overlap analysis of the entropy mutation region and the leakage probability cloud map, multi-level warning signals are triggered to generate a leakage source heat map and a pressure balance parameter set; finally, the self-regulating system dynamically adjusts the operating parameters of the equipment to form a closed-loop control link, improving the warning accuracy and control efficiency of the leakage fault of the hydrogen energy storage equipment.

[0048] To solve the problems of accuracy and refinement in predicting the leakage fault of hydrogen energy storage equipment, in some embodiments, step 103: The spatio-temporal reasoning model includes a multi-layer perception constraint module, an implicit state transition model, a gas diffusion physical constraint module, and a probability density clustering module connected in sequence. The multi-layer perception constraint module refers to a component that uses the multi-layer perceptron (MLP) architecture in deep learning to process input data and identify internal structural features. It is mainly used to analyze the internal structural connection relationship of hydrogen energy storage equipment and generate a topological map of the energy transfer path. This module simulates how energy flows in the system by learning the complex relationships between different sensor data. The gas diffusion physical constraint module refers to a module designed based on physical principles, which is used to simulate and predict the diffusion behavior of gas (hydrogen in this case) under specific environmental conditions. This module combines real-time environmental parameters (such as temperature, wind speed, etc.), constructs a three-dimensional space grid coordinate system, and applies the mass conservation equation and the momentum transfer theorem to calculate the gas diffusion path. The probability density clustering module is a data analysis tool specifically used to identify dense regions or clusters in high-dimensional datasets. In the application scenario of hydrogen energy storage equipment, it is used to analyze the fusion leakage situation map, identify the regions with the highest leakage risk, and generate a leakage probability cloud map. In addition, it can also construct an energy loss vector based on the energy flow field feature matrix to help quantify the degree of energy loss.

[0049] Inputting the fusion feature tensor into the spatio-temporal inference model to predict the leakage and energy loss of the hydrogen energy storage device by using the spatio-temporal inference model, and outputting a leakage probability cloud map and an energy loss vector, including: Step 201: Analyze the internal structure connection relationship of the hydrogen energy storage device through a multi-layer perception constraint module to generate a topological graph of the energy transfer path.

[0050] In step 201, the topological graph of the energy transfer path refers to a network graph representing the internal energy transfer path of the device, where nodes represent device components and edges represent energy transfer relationships.

[0051] In the embodiment of the present application, the connection relationship of the internal structure of the hydrogen energy storage device is analyzed through a multi-layer perception constraint module, and the graph neural network is used to model the energy transfer relationship between device components to generate a topological graph of the energy transfer path. The node weights of this topological graph are determined by the energy transfer efficiency of the device components, and the edge weights are determined by the connection strength between components.

[0052] Step 202: Based on the topological graph of the energy transfer path, trace the energy flow field feature matrix through the dynamic recurrence operator in the implicit state transition model, and the dynamic recurrence operator adaptively adjusts the state transition step size according to the coupling effect of the temperature gradient distribution field and the vibration frequency spectrum waveform.

[0053] In step 202, the dynamic recurrence operator is an operator used for state transition calculation on the topological graph of the energy transfer path and can adaptively adjust the step size.

[0054] In the embodiment of the present application, the temperature gradient is decomposed into a heat conduction component and a thermal stress component through a thermo-elastic coupling equation, and a mechanical vibration influence factor matrix is generated by combining the frequency domain energy distribution of the vibration frequency spectrum, and finally a temperature-vibration coupling coefficient matrix is generated to adjust the state transition step size.

[0055] Step 203: Input the energy flow field feature matrix into the gas diffusion physical constraint module and construct a three-dimensional space grid coordinate system in combination with real-time environmental parameters.

[0056] In step 203, the three-dimensional space grid coordinate system refers to dividing the space around the device into three-dimensional grids for simulating gas diffusion.

[0057] In the embodiment of the present application, the energy flow field feature matrix is input into the gas diffusion physical constraint module, and a three-dimensional space grid coordinate system is constructed in combination with real-time environmental parameters (such as wind speed, air pressure, temperature). This coordinate system divides the space around the device into several grid cells, and each cell contains parameters such as gas concentration, temperature, and pressure.

[0058] Step 204: Simulate the diffusion path of the leaked gas in the three-dimensional space grid coordinate system through the mass conservation equation and the momentum transfer theorem, and generate the diffusion path probability distribution of the leaked gas.

[0059] In step 204, the diffusion path probability distribution is used to represent the distribution of the diffusion probability of the gas in different grid cells.

[0060] In the embodiment of the present application, the mass change and momentum transfer of the gas in each grid cell are calculated to generate the diffusion path probability distribution of the leaked gas.

[0061] Step 205: Generate a fused leakage situation map according to the diffusion path probability distribution.

[0062] In step 205, the fused leakage situation map is the result obtained after comprehensive analysis of all the collected information, which shows the possible locations of leakage and its severity.

[0063] In the embodiment of the present application, a fused leakage situation map is generated according to the diffusion path probability distribution. This map combines the gas diffusion path and the energy flow field characteristics, and can intuitively reflect the distribution and diffusion trend of the leakage risk.

[0064] Step 206: Based on the fused leakage situation map, generate a leakage probability cloud map through the probability density clustering module, and construct an energy loss vector according to the energy flow field characteristic matrix.

[0065] In the embodiment of the present application, the Gaussian mixture model is used to perform clustering analysis on the leakage situation map to generate a leakage probability cloud map; at the same time, an energy loss vector is constructed according to the energy flow field characteristic matrix to quantify the energy loss caused by leakage.

[0066] The following is a specific example: In a hydrogen energy storage device, the multi-layer perception constraint module analyzes the internal structure of the device to generate a topological map of the energy transfer path; the implicit state transition model tracks the energy flow field characteristic matrix and discovers an abnormal coupling between the temperature gradient and the vibration frequency spectrum in a certain area; the gas diffusion physical constraint module constructs a three-dimensional space grid coordinate system to simulate the diffusion path of the leaked gas and generates a diffusion path probability distribution; a fused leakage situation map is generated based on the diffusion path probability distribution, showing that the leaked gas diffuses towards the southeast direction of the device; the probability density clustering module generates a leakage probability cloud map and discovers that the leakage probability in the southeast area is as high as 80%, and at the same time constructs an energy loss vector, showing an increase in energy loss in this area. The system triggers a secondary warning signal, generates a thermal map of the leakage source, and dynamically adjusts the valve openings of adjacent devices to balance the pressure distribution.

[0067] Through the collaborative work of a multi-layer perceptron constraint module, an implicit state transition model, a gas diffusion physical constraint module, and a probability density clustering module, this solution achieves refined prediction and dynamic regulation of leakage faults in hydrogen energy storage devices. Specifically, the generated energy transfer path topology map and energy flow field feature matrix provide a data basis for leakage prediction. The leakage probability cloud map and energy loss vector generated by the gas diffusion physical constraint module and the probability density clustering module improve the accuracy of leakage location and energy loss analysis. Finally, through the self-regulating system, the dynamic optimization of the device operation parameters is realized, forming a closed-loop control link, effectively improving the safety and operation efficiency of the hydrogen energy storage device.

[0068] In order to further improve the accuracy and dynamic adjustment ability of the energy flow field feature tracking of the hydrogen energy storage device, in some embodiments, step 202: Based on the energy transfer path topology map, the energy flow field feature matrix is tracked through the dynamic recurrence operator in the implicit state transition model, and the dynamic recurrence operator adaptively adjusts the state transition step size according to the coupling effect of the temperature gradient distribution field and the vibration frequency spectrum waveform, including: Step 301: Decompose the temperature gradient distribution field into a heat conduction component and a thermal stress component through a preset thermoelastic coupling equation, and generate a mechanical vibration influence factor matrix based on the frequency domain energy distribution of the vibration frequency spectrum waveform.

[0069] In step 301, the preset thermoelastic coupling equation is a mathematical model used to describe the influence of temperature changes (heat) on the internal stress distribution (elasticity) of materials and the interaction between the two. The heat conduction component refers to the part of the temperature change caused by heat conduction in the temperature gradient distribution field. The thermal stress component refers to the part of the temperature change caused by thermal stress in the temperature gradient distribution field. The mechanical vibration influence factor matrix is a matrix representing the influence of the vibration frequency spectrum waveform on the mechanical vibration of the device.

[0070] In the embodiments of this application, the temperature gradient distribution field is decomposed into a heat conduction component and a thermal stress component through a preset thermoelastic coupling equation, respectively representing the temperature changes caused by heat conduction and thermal stress; at the same time, based on the frequency domain energy distribution of the vibration frequency spectrum waveform, the fast Fourier transform is used to extract the energy values of the main frequency bands, and a mechanical vibration influence factor matrix is generated to quantify the influence of vibration on the device.

[0071] Step 302: Perform a two-channel convolution fusion operation on the heat conduction component and the mechanical vibration influence factor matrix to generate a temperature-vibration coupling coefficient matrix.

[0072] In step 302, the dual-channel convolution fusion operation is a technique for fusing two feature matrices through convolution operations. The temperature-vibration coupling coefficient matrix refers to a matrix representing the coupling effect between the temperature gradient and the vibration spectrum, and its dimension is consistent with the number of nodes in the energy transfer path topology diagram.

[0073] In the embodiment of the present application, a dual-channel convolution fusion operation is performed on the heat conduction component and the mechanical vibration influence factor matrix. The convolutional neural network is used to perform convolution processing on the two feature matrices respectively to extract their spatial features; then, a temperature-vibration coupling coefficient matrix is generated through feature splicing and a fully connected layer, and each element of this matrix represents the coupling strength between the temperature and vibration of the corresponding node.

[0074] Step 303: Perform state transition calculation with physical constraints on the energy transfer path topology diagram, and iteratively update the energy flow field feature matrix. In each iteration process, the dynamic recurrence operator dynamically adjusts the state transition step size according to the spectral radius of the temperature-vibration coupling coefficient matrix.

[0075] In step 303, the state transition calculation with physical constraints means introducing physical constraints (such as energy conservation and momentum conservation) in the state transition calculation to ensure the rationality of the calculation results. The spectral radius refers to the maximum absolute value of the matrix eigenvalues and is used to measure the stability of the matrix. The state transition step size refers to the step size of each iteration in the state transition calculation, which affects the calculation efficiency and accuracy.

[0076] In the embodiment of the present application, a state transition calculation with physical constraints is performed on the energy transfer path topology diagram to iteratively update the energy flow field feature matrix. Specifically, in each iteration process, the dynamic recurrence operator dynamically adjusts the state transition step size according to the spectral radius of the temperature-vibration coupling coefficient matrix: if the spectral radius is large, the step size is reduced to improve stability; if the spectral radius is small, the step size is increased to improve the calculation efficiency. Finally, an updated energy flow field feature matrix is generated.

[0077] The following is a specific example: In an application scenario of a hydrogen energy storage station, first, the monitored temperature gradient distribution field is divided into a heat conduction part and a thermal stress part through the thermoelastic coupling equation, and the mechanical vibration influence factor matrix is extracted from the vibration sensor data. Then, these data are processed using the dual-channel convolution fusion algorithm to generate a temperature-vibration coupling coefficient matrix. Finally, a state transition calculation with physical constraints is applied on the energy transfer path topology diagram to dynamically adjust the state transition step size, accurately tracking the changes in the energy flow field, and improving the accuracy and timeliness of leakage prediction.

[0078] Through the thermo-elastic coupling equation and the dual-channel convolution fusion operation, this solution realizes the refined coupling analysis of the temperature gradient and the vibration spectrum, generating a temperature-vibration coupling coefficient matrix; combined with the state transition calculation with physical constraints and the dynamic recursive operator, it realizes the efficient iterative update of the energy flow field characteristic matrix, improves the accuracy and dynamic adjustment ability of the energy flow field characteristic tracking, and provides a reliable data basis for subsequent leakage prediction and energy loss analysis.

[0079] In order to further improve the accuracy and rationality of the state transition calculation, in some embodiments, step 303: before performing the state transition calculation with physical constraints on the energy transfer path topology diagram, the method further includes: Step 401: Perform frequency-domain energy spectrum analysis on the pressure fluctuation sequence to obtain an analysis result.

[0080] In step 401, the analysis result refers to the energy distribution of the pressure fluctuation sequence in the frequency domain.

[0081] In the embodiments of the present application, when performing frequency-domain energy spectrum analysis on the pressure fluctuation sequence, the Fourier transform is used to convert the time-domain pressure signal into a frequency-domain energy spectrum, obtaining the energy distribution at different frequencies, providing a data basis for subsequent extraction of the energy mutation index.

[0082] Step 402: Extract the energy mutation indices in different frequency bands from the analysis result through a band-pass filter window to construct a pressure frequency-domain mutation feature vector.

[0083] In step 402, the band-pass filter window is a signal processing term, referring to a filter that can select the signal components within a specific frequency range (i.e., frequency band). The energy mutation index refers to the intensity representing the energy mutation within a certain frequency band. The pressure frequency-domain mutation feature vector refers to a vector composed of the energy mutation indices in different frequency bands.

[0084] In the embodiments of the present application, the energy mutation indices in different frequency bands are extracted from the frequency-domain energy spectrum through a band-pass filter window. The sliding window technique is used to segment the frequency-domain energy spectrum, calculate the energy mutation index for each frequency band, and finally construct a pressure frequency-domain mutation feature vector.

[0085] Step 403: Calculate the spatial gradient direction weight coefficient based on the preset leakage diffusion direction constraint condition.

[0086] In step 403, the leakage diffusion direction constraint condition refers to the direction limit on how the leakage spreads set according to the physical model and historical data. The spatial gradient direction weight coefficient refers to the weight coefficient representing the influence of the gas diffusion direction on the pressure fluctuation.

[0087] In the embodiments of the present application, according to the physical laws of gas diffusion (such as diffusion from high pressure to low pressure), combined with the internal pressure distribution of the equipment, the gradient direction weight coefficient at each spatial position is calculated.

[0088] Step 404: Input the pressure frequency-domain mutation feature vector and the spatial gradient direction weight coefficient into the dual-channel attention gating module to generate an initial value of the time-varying gain factor through the non-linear mapping function in the dual-channel attention gating module.

[0089] In step 404, the dual-channel attention gating module refers to a module that combines the attention mechanism and is used to establish the correlation between the pressure frequency-domain features and the spatial gradient features. The initial value of the time-varying gain factor refers to the initial value representing the correlation strength between the pressure frequency-domain features and the spatial gradient features.

[0090] In the embodiments of the present application, the pressure frequency-domain mutation feature vector and the spatial gradient direction weight coefficient are input into the dual-channel attention gating module, and the non-linear mapping function is used to establish the correlation mapping between the two to generate the initial value of the time-varying gain factor.

[0091] Step 405: Perform dynamic normalization processing on the initial value of the time-varying gain factor using the leakage risk index threshold surface to generate the normalized time-varying gain factor.

[0092] In step 405, the leakage risk index threshold surface is the threshold surface used for dynamically normalizing the time-varying gain factor. The normalized time-varying gain factor refers to the time-varying gain factor after dynamic normalization processing, and its value range is [0, 1].

[0093] In the embodiments of the present application, the initial value of the time-varying gain factor is dynamically normalized using the leakage risk index threshold surface, and the initial value of the time-varying gain factor is mapped to the range of [0, 1] to generate the normalized time-varying gain factor.

[0094] Step 406: Perform a tensor multiplication operation on the normalized time-varying gain factor and the temperature vibration coupling coefficient matrix to generate a dynamic weight parameter matrix.

[0095] In step 406, the dynamic weight parameter matrix refers to the weight matrix combined by the time-varying gain factor and the temperature vibration coupling coefficient matrix.

[0096] In the embodiments of the present application, a tensor multiplication operation is performed on the normalized time-varying gain factor and the temperature vibration coupling coefficient matrix to generate a dynamic weight parameter matrix. This matrix is used to adjust the weight distribution in the state transition calculation.

[0097] Step 407: Embed the dynamic weight parameter matrix into the initial state transition equation as a physical constraint to obtain a state transition equation with physical constraints.

[0098] In step 407, the state transition equation with physical constraints is a modified traditional state transition equation considering physical reality, which contains all the information obtained from the previous steps.

[0099] In the embodiments of the present application, the dynamic weight parameter matrix is embedded into the initial state transition equation as a physical constraint to obtain the state transition equation with physical constraints. This equation considers the coupling effects of pressure fluctuations, temperature gradients, and vibration spectra in the state transition calculation, improving the accuracy and rationality of the calculation.

[0100] The following is a specific example: In an application scenario of a hydrogen energy storage station, first, perform frequency-domain energy spectrum analysis on the pressure fluctuation sequence, extract the energy mutation index in different frequency bands, and construct the pressure frequency-domain mutation feature vector. Then, calculate the spatial gradient direction weight coefficient based on the leakage diffusion direction constraint condition, and generate the initial value of the time-varying gain factor through the dual-channel attention gating module. Next, normalize this initial value using the leakage risk index threshold curve, and finally combine it with the temperature-vibration coupling coefficient matrix to generate the dynamic weight parameter matrix and apply it to the state transition equation, achieving high-precision tracking of the energy flow pattern and greatly improving the effectiveness of the leakage warning system.

[0101] This solution realizes the refined coupling analysis of pressure fluctuations, temperature gradients, and vibration spectra through frequency-domain energy spectrum analysis, band-pass filtering, spatial gradient direction weight calculation, and the dual-channel attention gating module; generates the dynamic weight parameter matrix through dynamic normalization and tensor multiplication operations and embeds it into the state transition equation, improving the accuracy and rationality of the state transition calculation, and providing reliable technical support for the leakage prediction and energy loss analysis of hydrogen energy storage devices.

[0102] In order to further improve the correlation mapping accuracy between the pressure frequency-domain features and the spatial gradient features, in some embodiments, step 404: inputting the pressure frequency-domain mutation feature vector and the spatial gradient direction weight coefficient into the dual-channel attention gating module to generate the initial value of the time-varying gain factor through the non-linear mapping function in the dual-channel attention gating module includes: Step 501: Perform a multi-scale decomposition operation on the pressure frequency-domain mutation feature vector to obtain multi-scale decomposition information.

[0103] In step 501, the multi-scale decomposition information includes a low-frequency scale, a medium-frequency scale, and a high-frequency scale. The low-frequency scale captures the global trend and slow-changing features in the pressure fluctuation sequence, the medium-frequency scale reflects the energy distribution and fluctuation features in the medium-frequency range, and the high-frequency scale extracts the fast-changing detail information and local mutation features.

[0104] In the embodiments of the present application, in order to perform multi-scale decomposition operations on the pressure frequency-domain mutation feature vectors, for example, empirical mode decomposition or wavelet transform is used to decompose the pressure frequency-domain mutation feature vectors into sub-signals of multiple scales, obtaining multi-scale decomposition information, which provides a data basis for extracting the energy mutation intensity coefficients subsequently.

[0105] Step 502: Perform wavelet packet transform on the multi-scale decomposition information to extract the energy mutation intensity coefficients of different frequency bands.

[0106] In step 502, wavelet packet transform refers to a more refined frequency-domain analysis method than wavelet transform, which can extract more detailed frequency band characteristics. The energy mutation intensity coefficient represents the intensity of energy mutation within a certain frequency band.

[0107] In the embodiments of the present application, wavelet packet transform is used to perform frequency-domain analysis on the sub-signals of each scale, and calculate the energy mutation intensity coefficients of each frequency band.

[0108] Step 503: Construct a pressure frequency-domain energy distribution tensor according to the energy mutation intensity coefficients of different frequency bands.

[0109] In step 503, the pressure frequency-domain energy distribution tensor refers to a high-dimensional tensor composed of the energy mutation intensity coefficients of different frequency bands, representing the distribution of pressure frequency-domain energy.

[0110] In the embodiments of the present application, a pressure frequency-domain energy distribution tensor is constructed according to the energy mutation intensity coefficients of different frequency bands. Each dimension of this tensor corresponds to the energy mutation intensity coefficient of a frequency band, and is used to represent the multi-scale distribution characteristics of pressure frequency-domain energy.

[0111] Step 504: Perform directional enhancement processing on the spatial gradient direction weight coefficients to obtain the processed spatial gradient direction weight coefficients.

[0112] In step 504, the processed spatial gradient direction weight coefficients refer to the weight coefficients after directional enhancement processing.

[0113] In the embodiments of the present application, directional enhancement processing is performed on the spatial gradient direction weight coefficients. The Gaussian filtering algorithm is used to smooth the weight coefficients, and at the same time enhance the weight values in the main diffusion directions, obtaining the processed spatial gradient direction weight coefficients.

[0114] Step 505: Based on the processed spatial gradient direction weight coefficients, use the anisotropic diffusion filtering algorithm to generate a gradient direction consistency matrix.

[0115] In step 505, the gradient direction consistency matrix represents a matrix of spatial gradient direction consistency.

[0116] In the embodiments of the present application, based on the processed spatial gradient direction weight coefficients, an anisotropic diffusion filtering algorithm is used to generate a gradient direction consistency matrix. This algorithm adjusts the filtering intensity according to the gradient direction to generate a consistency matrix for quantifying the consistency of the gradient direction.

[0117] Step 506: Calculate a direction weight correction factor based on the gradient direction consistency matrix and a preset leakage diffusion direction constraint condition.

[0118] In step 506, the direction weight correction factor is a factor used to correct the spatial gradient direction weight to improve the rationality of weight allocation.

[0119] In the embodiments of the present application, based on the gradient direction consistency matrix and a preset leakage diffusion direction constraint condition, a direction weight correction factor is calculated. Specifically, according to the consistency matrix and the leakage diffusion direction constraint condition, the weight allocation is adjusted to generate a direction weight correction factor.

[0120] Step 507: Input the pressure frequency-domain energy distribution tensor and the direction weight correction factor into a dual-channel attention gating module to establish a non-linear correlation mapping between the pressure frequency domain and the spatial gradient through the cross-attention module in the dual-channel attention gating module, and generate an initial attention weight matrix.

[0121] In step 507, the initial attention weight matrix is an initial matrix representing the correlation strength between the pressure frequency domain and the spatial gradient features.

[0122] In the embodiments of the present application, the pressure frequency-domain energy distribution tensor and the direction weight correction factor are input into the dual-channel attention gating module, and a non-linear correlation mapping between the pressure frequency domain and the spatial gradient is established through the cross-attention module to generate an initial attention weight matrix. This matrix represents the correlation strength between the two.

[0123] Step 508: Perform a non-linear activation operation on the initial attention weight matrix to generate an initial value of the time-varying gain factor.

[0124] In the embodiments of the present application, a non-linear activation operation is performed on the initial attention weight matrix to perform a non-linear transformation on the matrix and generate an initial value of the time-varying gain factor. This value is used for subsequent dynamic normalization processing.

[0125] The following is a specific example: In a certain hydrogen energy storage device, a multi-scale decomposition operation is performed on the pressure frequency domain mutation feature vector to obtain multi-scale decomposition information; the energy mutation intensity coefficients of different frequency bands are extracted through wavelet packet transform to construct a pressure frequency domain energy distribution tensor; the directional enhancement processing is performed on the spatial gradient direction weight coefficient to generate a gradient direction consistency matrix; the direction weight correction factor is calculated in combination with the leakage diffusion direction constraint condition; the pressure frequency domain energy distribution tensor and the direction weight correction factor are input into a dual-channel attention gating module to generate an initial attention weight matrix; the initial value of the time-varying gain factor is generated through a non-linear activation operation for subsequent state transition calculation.

[0126] Through multi-scale decomposition, wavelet packet transform, directional enhancement processing, and a cross-attention module, this solution realizes the refined correlation mapping between the pressure frequency domain features and the spatial gradient features; the generated pressure frequency domain energy distribution tensor and direction weight correction factor improve the accuracy of feature expression, and the finally generated initial value of the time-varying gain factor provides a basis for high-precision weight allocation for subsequent state transition calculation, further improving the accuracy and reliability of hydrogen energy storage device leakage prediction.

[0127] In order to further improve the fusion effect of the multi-dimensional feature matrix and the feature expression ability, in some embodiments, step 102: inputting the multi-dimensional feature matrix into a bidirectional interaction network to establish a non-linear mapping between the pressure fluctuation sequence and the gas concentration spectrogram through the cross-attention module in the bidirectional interaction network, and using the gradient field reconstruction algorithm to perform spatial propagation modeling on the temperature gradient distribution field and the vibration frequency spectrum waveform to generate a fusion feature tensor, including: Step 601: Perform time-frequency joint analysis on the pressure fluctuation sequence to generate a pressure fluctuation time-frequency feature matrix.

[0128] In step 601, time-frequency joint analysis refers to simultaneously analyzing the characteristics of a signal in the time domain and the frequency domain to capture the dynamic changes and frequency distribution of the signal. The pressure fluctuation time-frequency feature matrix refers to a high-dimensional matrix representing the joint characteristics of the pressure fluctuation sequence in the time domain and the frequency domain.

[0129] In the embodiments of the present application, time-frequency joint analysis is performed on the pressure fluctuation sequence to convert the pressure fluctuation sequence from the time domain to the time-frequency domain to generate a pressure fluctuation time-frequency feature matrix. This matrix contains information in three dimensions: time, frequency, and energy, and is used to comprehensively describe the dynamic characteristics of pressure fluctuations.

[0130] Step 602: Perform spatial multi-resolution analysis on the gas concentration spectrogram to generate a gas concentration multi-scale feature map.

[0131] In step 602, spatial multi-resolution analysis refers to decomposing the gas concentration spectrogram at different spatial scales to capture multi-level concentration distribution characteristics. The gas concentration multi-scale feature map refers to the feature distribution map representing the gas concentration spectrogram at different spatial scales. The multi-scale of the gas concentration multi-scale feature map includes large scale, medium scale, and small scale. The large scale captures the distribution trend and global characteristics of the gas concentration within the overall space of the device, the medium scale reflects the change characteristics and transition regions of the gas concentration within the local area, and the small scale extracts the detailed information and local mutation characteristics of the gas concentration within the fine spatial range.

[0132] In the embodiment of the present application, spatial multi-resolution analysis is performed on the gas concentration spectrogram, and the gas concentration spectrogram is decomposed into feature maps of multiple spatial scales using the Gaussian pyramid algorithm to generate a gas concentration multi-scale feature map. This map contains concentration distribution information at different resolutions and is used to describe the spatial variation characteristics of the gas concentration.

[0133] Step 603: Input the pressure fluctuation time-frequency feature matrix and the gas concentration multi-scale feature map into the cross-attention module to establish an association mapping between the pressure frequency domain features and the gas concentration spatial features through the multi-head attention module, and generate a pressure-concentration interaction feature matrix.

[0134] In step 603, the pressure-concentration interaction feature matrix represents a high-dimensional matrix of the association mapping between the pressure frequency domain features and the gas concentration spatial features.

[0135] In the embodiment of the present application, the pressure fluctuation time-frequency feature matrix and the gas concentration multi-scale feature map are input into the cross-attention module, and an association mapping between the pressure frequency domain features and the gas concentration spatial features is established through the multi-head attention module. Specifically, the multi-head attention mechanism is used to calculate the association weights between the two to generate a pressure-concentration interaction feature matrix.

[0136] Step 604: Generate a temperature gradient propagation path map according to the temperature gradient distribution field.

[0137] In step 604, the temperature gradient propagation path map represents a map of the propagation path of the temperature gradient inside the device.

[0138] In the embodiment of the present application, a temperature gradient propagation path map is generated according to the temperature gradient distribution field, and the heat conduction equation is used to simulate the propagation path of the temperature gradient inside the device to generate a temperature gradient propagation path map. This map is used to describe the spatial propagation characteristics of the temperature gradient.

[0139] Step 605: Extract the intrinsic mode function components of the vibration frequency spectrum waveform through the empirical mode decomposition algorithm, and generate a vibration frequency spectrum energy feature vector based on the intrinsic mode function components.

[0140] In step 605, the intrinsic mode function components represent the components of different frequency components in the signal. The vibration spectrum energy feature vector represents the feature vector of the energy distribution of the vibration spectrum waveform.

[0141] In the embodiments of the present application, the intrinsic mode function components of the vibration spectrum waveform are extracted by the empirical mode decomposition algorithm. The vibration spectrum waveform is decomposed into multiple intrinsic mode function components using empirical mode decomposition, and a vibration spectrum energy feature vector is generated based on these components.

[0142] Step 606: Generate a temperature-vibration interaction feature matrix according to the temperature gradient propagation path map and the vibration spectrum energy feature vector.

[0143] In step 606, the temperature-vibration interaction feature matrix represents a high-dimensional matrix of the correlation mapping between the temperature gradient and the vibration spectrum features.

[0144] In the embodiments of the present application, a temperature-vibration interaction feature matrix is generated according to the temperature gradient propagation path map and the vibration spectrum energy feature vector. Specifically, a convolutional neural network is used to perform feature fusion on the two to generate a temperature-vibration interaction feature matrix.

[0145] Step 607: Perform a tensor splicing operation on the pressure-concentration interaction feature matrix and the temperature-vibration interaction feature matrix to generate a fused feature tensor.

[0146] In the embodiments of the present application, a tensor splicing operation is performed on the pressure-concentration interaction feature matrix and the temperature-vibration interaction feature matrix to generate a fused feature tensor. This tensor contains multi-source feature information of pressure, gas concentration, temperature, and vibration, and is used for subsequent leakage prediction and energy loss analysis.

[0147] The following is a specific example: In a certain hydrogen energy storage device, a time-frequency joint analysis is performed on the pressure fluctuation sequence to generate a pressure fluctuation time-frequency feature matrix; a spatial multi-resolution analysis is performed on the gas concentration spectrogram to generate a gas concentration multi-scale feature map; the two are input into a cross-attention module to generate a pressure-concentration interaction feature matrix; a temperature gradient propagation path map is generated according to the temperature gradient distribution field; the intrinsic mode function components of the vibration spectrum waveform are extracted by the empirical mode decomposition algorithm to generate a vibration spectrum energy feature vector; a temperature-vibration interaction feature matrix is generated according to the temperature gradient propagation path map and the vibration spectrum energy feature vector; finally, the pressure-concentration interaction feature matrix and the temperature-vibration interaction feature matrix are spliced to generate a fused feature tensor, providing high-dimensional feature input for subsequent leakage prediction.

[0148] Through time-frequency joint analysis, spatial multi-resolution analysis, cross-attention module and empirical mode decomposition algorithm, this solution realizes the deep fusion of multi-source features such as pressure, gas concentration, temperature and vibration; the generated fusion feature tensor improves the comprehensiveness and accuracy of feature expression, provides high-dimensional and high-precision feature input for hydrogen energy storage equipment leakage prediction and energy loss analysis, and further improves the accuracy and reliability of the prediction model.

[0149] In order to further improve the accuracy and dynamic regulation ability of hydrogen energy storage equipment leakage warning, in some embodiments, step 104: determining an entropy value mutation region based on the energy loss vector, and when the high-density region in the leakage probability cloud map overlaps with the entropy value mutation region in consecutive detection periods in space, triggering a multi-level warning signal and generating a leakage source thermal map and a pressure balance parameter set, including: Step 701: Perform a multi-dimensional decomposition operation on the energy loss vector to generate an energy loss feature matrix.

[0150] In step 701, the energy loss feature matrix represents the feature distribution matrix of energy loss in multiple dimensions. The multiple dimensions in the multi-dimensional decomposition operation include the time dimension, the space dimension and the energy dimension. The time dimension analyzes the change trend of energy loss over time, the space dimension describes the distribution characteristics of energy loss at different spatial positions of the equipment, and the energy dimension reflects the distribution of energy loss at different energy levels. Through multi-dimensional decomposition, the multi-faceted characteristics of energy loss can be comprehensively captured, providing multi-dimensional feature support for subsequent identification of entropy value mutation regions.

[0151] In the embodiments of the present application, a multi-dimensional decomposition operation is performed on the energy loss vector, and independent component analysis is used to decompose the energy loss vector into features in multiple dimensions to generate an energy loss feature matrix. This matrix is used to describe the distribution characteristics of energy loss in different dimensions.

[0152] Step 702: Identify abnormal energy entropy values according to the energy loss feature matrix to generate an entropy value mutation region.

[0153] In step 702, the abnormal energy entropy value represents the region with abnormal entropy value in the energy loss feature matrix.

[0154] In the embodiments of the present application, abnormal energy entropy values are identified according to the energy loss feature matrix, and an entropy value calculation algorithm is used to analyze the energy loss feature matrix to identify the region with abnormal entropy value and generate an entropy value mutation region.

[0155] Step 703: Perform spatial clustering analysis on the leakage probability cloud map to generate a high-density region.

[0156] In the embodiments of the present application, spatial clustering analysis is performed on the leakage probability cloud map to identify high-density regions. This region indicates that the leakage probability is higher than other regions.

[0157] Step 704: When the high-density region and the entropy value mutation region spatially overlap within at least two consecutive detection periods, trigger a multi-level warning signal.

[0158] In the embodiments of the present application, when the high-density region and the entropy value mutation region spatially overlap within at least two consecutive detection periods, a multi-level warning signal is triggered. Specifically, according to the area of the overlapping region and the leakage probability value, the risk level is judged and the corresponding level of warning signal is triggered.

[0159] Step 705: Generate a leakage source heat map corresponding to the high-density region, and calculate pressure compensation parameters for multiple adjacent hydrogen energy storage devices according to the leakage source heat map and a preset pressure balance equation, so as to generate a pressure balance parameter set.

[0160] In the embodiments of the present application, a leakage source heat map corresponding to the high-density region is generated, and pressure compensation parameters for multiple adjacent hydrogen energy storage devices are calculated according to the leakage source heat map and a preset pressure balance equation, and a pressure balance parameter set is generated. This parameter set is used to dynamically adjust the valve opening degrees of adjacent devices to balance the pressure distribution.

[0161] The following is a specific example: In a certain hydrogen energy storage device, a multi-dimensional decomposition operation is performed on the energy loss vector to generate an energy loss feature matrix; abnormal energy entropy values are identified according to the energy loss feature matrix to generate an entropy value mutation region; spatial clustering analysis is performed on the leakage probability cloud map to generate a high-density region; when the high-density region and the entropy value mutation region overlap within two consecutive detection periods, a secondary warning signal is triggered, and a leakage source heat map is generated; according to the leakage source heat map and the pressure balance equation, the pressure compensation parameters of adjacent devices are calculated to generate a pressure balance parameter set for dynamically adjusting the valve opening degree.

[0162] This solution realizes accurate early warning and dynamic regulation of the leakage risk of hydrogen energy storage devices through multi-dimensional decomposition, entropy value analysis, spatial clustering and a multi-level early warning mechanism; the generated leakage source heat map and pressure balance parameter set improve the accuracy of leakage positioning and pressure balance, providing reliable technical support for the safe operation of the device.

[0163] Figure 2 It is a schematic structural diagram of a leakage fault early warning system for hydrogen energy storage devices based on deep learning provided by the embodiments of the present application. As Figure 2 shown, the system includes: A collection module 21 for collecting in real time multi-source heterogeneous sensing data streams of a hydrogen energy storage device to construct a multi-dimensional feature matrix, where the sensing data streams include a pressure fluctuation sequence, a temperature gradient distribution field, a vibration spectrum waveform, and a gas concentration spectrogram.

[0164] An input module 22 for inputting the multi-dimensional feature matrix into a bidirectional interaction network to establish a non-linear mapping between the pressure fluctuation sequence and the gas concentration spectrogram through a cross-attention module in the bidirectional interaction network, and using a gradient field reconstruction algorithm to perform spatial propagation modeling on the temperature gradient distribution field and the vibration spectrum waveform to generate a fused feature tensor.

[0165] An output module 23 for inputting the fused feature tensor into a spatio-temporal inference model to predict leakage and energy loss of the hydrogen energy storage device using the spatio-temporal inference model, and outputting a leakage probability cloud map and an energy loss vector.

[0166] A trigger module 24 for determining an entropy value mutation region based on the energy loss vector, and when the high-density region in the leakage probability cloud map spatially overlaps with the entropy value mutation region in consecutive detection periods, triggering a multi-level warning signal and generating a leakage source heat map and a set of pressure balance parameters.

[0167] An activation module 25 for activating a self-regulating system according to the leakage source heat map and the set of pressure balance parameters, so that the self-regulating system dynamically corrects the valve opening gradient of adjacent hydrogen energy storage devices and updates the node weights of the bidirectional interaction network to generate a closed-loop link.

[0168] Figure 2 The described leakage fault warning system for hydrogen energy storage devices based on deep learning can execute Figure 1 The described leakage fault warning method for hydrogen energy storage devices based on deep learning in the illustrated embodiment, the implementation principle and technical effects will not be elaborated further. For the leakage fault warning system for hydrogen energy storage devices based on deep learning in the above embodiment, the specific ways in which each module and unit perform operations have been described in detail in the embodiments related to the method, and will not be elaborated here.

[0169] In a possible design, Figure 2 The leakage fault warning system for hydrogen energy storage devices based on deep learning in the illustrated embodiment can be implemented as a computing device, such as Figure 3 shown, and this computing device can include a storage component 31 and a processing component 32; The storage component 31 stores one or more computer instructions, where the one or more computer instructions are called and executed by the processing component 32.

[0170] The processing component 32 aboveFigure 1 A method for early warning of leakage faults of hydrogen energy storage devices based on deep learning in the described embodiment.

[0171] Among them, the processing component 32 may include one or more processors to execute computer instructions to complete all or part of the steps in the above method. Of course, the processing component may also be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors or other electronic components for executing the above method.

[0172] The storage component 31 is configured to store various types of data to support the operation of the terminal. The storage component can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disks or optical discs.

[0173] Of course, the computing device may necessarily also include other components, such as input / output interfaces, display components, communication components, etc.

[0174] The input / output interface provides an interface between the processing component and the peripheral interface module, and the above peripheral interface module may be an output device, an input device, etc.

[0175] The communication component is configured to facilitate communication between the computing device and other devices in a wired or wireless manner, etc.

[0176] Among them, the computing device can be a physical device or an elastic computing host provided by a cloud computing platform. In this case, the computing device can refer to a cloud server. The above-mentioned processing components, storage components, etc. can be basic server resources rented or purchased from a cloud computing platform.

[0177] An embodiment of the present application also provides a computer storage medium storing a computer program, which when executed by a computer can implement the above-mentioned Figure 1 A hydrogen energy storage device leakage fault warning method based on deep learning shown in the embodiment.

[0178] Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working processes of the above-described systems, devices, and units can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0179] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art can understand and implement it without creative labor.

[0180] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the essence of the above technical solution, or the part that contributes to the prior art, can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0181] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application and are not intended to limit them. Although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments or perform equivalent replacements for some of the technical features. These modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A hydrogen energy storage equipment leakage fault warning method based on deep learning, characterized in that: include: Collect multi-source heterogeneous sensor data streams of hydrogen energy storage equipment in real time to construct a multi-dimensional feature matrix, wherein the sensor data streams include pressure fluctuation sequences, temperature gradient distribution fields, vibration spectrum waveforms, and gas concentration spectra; Inputting the multidimensional feature matrix into a bidirectional interactive network, so as to establish a nonlinear mapping between the pressure fluctuation sequence and the gas concentration spectrum through a cross-attention module in the bidirectional interactive network, and using a gradient field reconstruction algorithm to perform spatial propagation modeling on the temperature gradient distribution field and the vibration spectrum waveform to generate a fusion feature tensor; Input the fused feature tensor into the spatiotemporal reasoning model, so as to use the spatiotemporal reasoning model to predict the leakage and energy loss of the hydrogen energy storage device, and output a leakage probability cloud map and an energy loss vector; Determine the entropy mutation area based on the energy loss vector, and when the high-density area in the leakage probability cloud map overlaps with the entropy mutation area in a continuous detection cycle, trigger a multi-level warning signal and generate a leakage source thermal map and a pressure balance parameter set; The self-regulating system is activated according to the leakage source thermodynamic map and the pressure balance parameter set, so that the self-regulating system dynamically corrects the valve opening gradient of the adjacent hydrogen energy storage device and updates the node weights of the bidirectional interactive network to generate a closed-loop link.

2. The method according to claim 1, characterized in that The spatiotemporal reasoning model includes a multi-layer perception constraint module, an implicit state transfer model, a gas diffusion physical constraint module and a probability density clustering module connected in sequence; The step of inputting the fused feature tensor into the spatiotemporal reasoning model to use the spatiotemporal reasoning model to predict leakage and energy loss of hydrogen energy storage equipment, and outputting a leakage probability cloud map and an energy loss vector includes: The internal structure connection relationship of the hydrogen energy storage device is analyzed through the multi-layer perception constraint module to generate the energy transfer path topology diagram; Based on the energy transfer path topology diagram, the energy flow field characteristic matrix is ​​tracked by a dynamic recursive operator in an implicit state transfer model, and the dynamic recursive operator adaptively adjusts the state transfer step size according to the coupling effect between the temperature gradient distribution field and the vibration spectrum waveform; The energy flow field characteristic matrix is ​​input into the gas diffusion physical constraint module, and a three-dimensional space grid coordinate system is constructed in combination with real-time environmental parameters; The diffusion path of the leaked gas in the three-dimensional space grid coordinate system is simulated by the mass conservation equation and momentum transfer theorem, and the diffusion path probability distribution of the leaked gas is generated; Generate a fusion leakage situation map according to the diffusion path probability distribution; Based on the fused leakage situation map, a leakage probability cloud map is generated through a probability density clustering module, and an energy loss vector is constructed according to the energy flow field characteristic matrix.

3. The method according to claim 2, characterized in that The method of tracking the energy flow field characteristic matrix based on the energy transfer path topology diagram through a dynamic recursive operator in an implicit state transfer model, wherein the dynamic recursive operator adaptively adjusts the state transfer step size according to the coupling effect between the temperature gradient distribution field and the vibration spectrum waveform, comprises: Decomposing the temperature gradient distribution field into a heat conduction component and a thermal stress component by presetting a thermoelastic coupling equation, and generating a mechanical vibration influence factor matrix based on the frequency domain energy distribution of the vibration spectrum waveform; Performing a dual-channel convolution fusion operation on the heat conduction component and the mechanical vibration influence factor matrix to generate a temperature vibration coupling coefficient matrix, wherein the dimension of the coupling coefficient matrix is ​​consistent with the number of nodes of the energy transfer path topology diagram; A state transfer calculation with physical constraints is performed on the energy transfer path topology diagram, and the energy flow field characteristic matrix is ​​iteratively updated. During each iteration, the dynamic recursive operator dynamically adjusts the state transfer step size according to the spectral radius of the temperature vibration coupling coefficient matrix.

4. The method according to claim 3, characterized in that Before performing the state transfer calculation with physical constraints on the energy transfer path topology graph, the method further includes: Performing frequency domain energy spectrum analysis on the pressure fluctuation sequence to obtain analysis results; Extracting energy mutation indexes in different frequency bands from the analysis results through a bandpass filter window, and constructing a pressure frequency domain mutation feature vector; And calculate the spatial gradient direction weight coefficient based on the preset leakage diffusion direction constraint condition; Inputting the pressure frequency domain mutation feature vector and the spatial gradient direction weight coefficient into a dual-channel attention gating module to generate an initial value of a time-varying gain factor through a nonlinear mapping function in the dual-channel attention gating module; Dynamically normalizing the initial value of the time-varying gain factor using a leakage risk index threshold surface to generate a normalized time-varying gain factor; Performing tensor multiplication operation on the normalized time-varying gain factor and the temperature vibration coupling coefficient matrix to generate a dynamic weight parameter matrix; The dynamic weight parameter matrix is ​​embedded into the initial state transfer equation as a physical constraint to obtain a state transfer equation with physical constraints.

5. The method according to claim 4, characterized in that The step of inputting the pressure frequency domain mutation feature vector and the spatial gradient direction weight coefficient into a dual-channel attention gating module to generate an initial value of a time-varying gain factor through a nonlinear mapping function in the dual-channel attention gating module includes: Performing a multi-scale decomposition operation on the pressure frequency domain mutation feature vector to obtain multi-scale decomposition information; Performing wavelet packet transformation on the multi-scale decomposition information to extract energy mutation intensity coefficients of different frequency bands; According to the energy mutation intensity coefficients of different frequency bands, the pressure frequency domain energy distribution tensor is constructed; Performing directional enhancement processing on the spatial gradient directional weight coefficient to obtain a processed spatial gradient directional weight coefficient; Based on the processed spatial gradient direction weight coefficients, an anisotropic diffusion filter algorithm is used to generate a gradient direction consistency matrix; The directional weight correction factor is calculated based on the gradient directional consistency matrix and the preset leakage diffusion directional constraint condition; Inputting the pressure frequency domain energy distribution tensor and the directional weight correction factor into a dual-channel attention gating module, so as to establish a nonlinear correlation mapping between the pressure frequency domain and the spatial gradient through a cross attention module in the dual-channel attention gating module, and generate an initial attention weight matrix; A nonlinear activation operation is performed on the initial attention weight matrix to generate an initial value of a time-varying gain factor.

6. The method according to claim 1, characterized in that The multidimensional feature matrix is ​​input into a bidirectional interactive network to establish a nonlinear mapping between the pressure fluctuation sequence and the gas concentration spectrum through a cross attention module in the bidirectional interactive network, and a gradient field reconstruction algorithm is used to perform spatial propagation modeling on the temperature gradient distribution field and the vibration spectrum waveform to generate a fusion feature tensor, including: Performing a time-frequency joint analysis on the pressure fluctuation sequence to generate a pressure fluctuation time-frequency feature matrix; Performing spatial multi-resolution analysis on the gas concentration spectrum to generate a multi-scale characteristic map of gas concentration; Input the pressure fluctuation time-frequency feature matrix and the gas concentration multi-scale feature map into the cross attention module, so as to establish the correlation mapping between the pressure frequency domain feature and the gas concentration space feature through the multi-head attention module, and generate the pressure concentration interaction feature matrix; Generate a temperature gradient propagation path diagram according to the temperature gradient distribution field; Extracting the intrinsic mode function components of the vibration spectrum waveform by an empirical mode decomposition algorithm, and generating a vibration spectrum energy feature vector based on the intrinsic mode function components; Generate a temperature-vibration interaction feature matrix according to the temperature gradient propagation path diagram and the vibration spectrum energy feature vector; A tensor concatenation operation is performed on the pressure-concentration interaction feature matrix and the temperature-vibration interaction feature matrix to generate a fused feature tensor.

7. The method according to claim 1, characterized in that The entropy mutation region is determined based on the energy loss vector, and when the high-density region in the leakage probability cloud map overlaps with the entropy mutation region in a continuous detection cycle, a multi-level warning signal is triggered and a leakage source thermal map and a pressure balance parameter set are generated, including: Performing a multi-dimensional decomposition operation on the energy loss vector to generate an energy loss feature matrix; Identifying abnormal energy entropy values ​​according to the energy loss characteristic matrix to generate an entropy value mutation region; Performing spatial cluster analysis on the leakage probability cloud map to generate high-density areas; When the high-density area and the entropy value mutation area overlap in space for at least two consecutive detection cycles, a multi-level warning signal is triggered; A leakage source thermodynamic map corresponding to the high-density area is generated, and pressure compensation parameters of a plurality of adjacent hydrogen energy storage devices are calculated according to the leakage source thermodynamic map and a preset pressure balance equation to generate a pressure balance parameter set.

8. A hydrogen energy storage equipment leakage fault warning system based on deep learning, characterized in that: include: A collection module is used to collect multi-source heterogeneous sensor data streams of hydrogen energy storage equipment in real time to construct a multi-dimensional feature matrix, wherein the sensor data stream includes a pressure fluctuation sequence, a temperature gradient distribution field, a vibration spectrum waveform, and a gas concentration spectrum; An input module, used for inputting the multidimensional feature matrix into a bidirectional interactive network, so as to establish a nonlinear mapping between the pressure fluctuation sequence and the gas concentration spectrum through a cross attention module in the bidirectional interactive network, and to perform spatial propagation modeling on the temperature gradient distribution field and the vibration spectrum waveform using a gradient field reconstruction algorithm, so as to generate a fusion feature tensor; An output module, used to input the fused feature tensor into a spatiotemporal reasoning model, so as to use the spatiotemporal reasoning model to predict leakage and energy loss of hydrogen energy storage equipment, and output a leakage probability cloud map and an energy loss vector; A trigger module, used to determine the entropy mutation area based on the energy loss vector, and when the high-density area in the leakage probability cloud map and the entropy mutation area overlap in space within a continuous detection cycle, trigger a multi-level warning signal and generate a leakage source thermal map and a pressure balance parameter set; An activation module is used to activate the self-regulating system according to the leakage source thermal map and the pressure balance parameter set, so that the self-regulating system dynamically corrects the valve opening gradient of the adjacent hydrogen energy storage device and updates the node weights of the bidirectional interactive network to generate a closed-loop link.

9. A computing device, characterized in that It comprises a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement a deep learning-based hydrogen energy storage device leakage fault warning method as described in any one of claims 1 to 7.

10. A computer storage medium, characterized in that: A computer program is stored, and when the computer program is executed by a computer, a leakage fault warning method for hydrogen energy storage equipment based on deep learning is implemented as described in any one of claims 1 to 7.

Images