A method, device and equipment for detecting the health state of a hydrogen production system using electrolysis of seawater
By constructing a multimodal fusion pre-trained neural network model and combining it with self-supervised comparative learning, the problem of health status assessment of seawater electrolysis hydrogen production system in complex marine environments was solved. This enabled accurate assessment and performance degradation early warning of the seawater electrolysis hydrogen production system, improving the system's operational reliability and maintenance efficiency.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- STATE OCEAN TECH CENT
- Filing Date
- 2026-01-14
- Publication Date
- 2026-06-23
AI Technical Summary
Existing technologies struggle to accurately assess the overall health of seawater electrolysis hydrogen production systems in complex marine environments, particularly lacking the ability to identify and comprehensively evaluate the degradation trends of key components in the early stages. Traditional diagnostic methods are ill-suited to the nonlinear and time-varying degradation processes of these systems, and the application of artificial intelligence in electrolysis hydrogen production systems is insufficient.
A multimodal fusion pre-trained neural network model is constructed and optimized by combining a self-supervised contrastive learning method. The health status assessment of the seawater electrolysis hydrogen production system is realized by using multi-source operating data. The model is trained and optimized by acquiring a multimodal dataset, and features are extracted and fused using a temporal convolutional network and self-supervised contrastive learning to output a health status score.
It enables accurate health status assessment of the entire life cycle of seawater electrolysis hydrogen production system under conditions of limited labeled data or no abnormal samples, improving operational reliability and maintenance efficiency, and supporting performance degradation early warning and online monitoring.
Smart Images
Figure CN121496492B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of equipment condition monitoring, and in particular to a method, apparatus and equipment for monitoring the health status of a seawater electrolysis hydrogen production system. Background Technology
[0002] When operating in complex marine environments for extended periods, seawater electrolysis for hydrogen production is susceptible to factors such as power fluctuations, salt spray corrosion, electrode degradation, and membrane fouling, leading to performance degradation and even system shutdown. Existing monitoring methods often rely on single operating parameters, making it difficult to accurately reflect the overall health status of the system, and in particular, lacking the ability to identify and comprehensively assess the degradation trends of key components in their early stages.
[0003] Currently, the life-cycle health assessment of seawater electrolysis hydrogen production systems is still in its early stages, and a model system capable of integrating multi-source operational data and achieving dynamic health scoring has not yet been established. Traditional diagnostic methods rely on empirical thresholds, which cannot adapt to the nonlinear and time-varying degradation processes of the system, thus hindering the implementation of predictive maintenance.
[0004] In recent years, artificial intelligence (AI) technology has demonstrated its advantages in equipment health assessment, but its application in electrolytic hydrogen production systems is still insufficient, especially lacking a health status modeling method that combines physical mechanisms and data-driven approaches for seawater electrolysis. Therefore, there is an urgent need for an AI-based full lifecycle health status assessment model to quantify system health levels in real time, achieve performance degradation early warning and lifespan prediction, and improve the operational reliability and maintenance efficiency of electrolysis systems. Summary of the Invention
[0005] The purpose of this application is to provide a method, apparatus, and equipment for detecting the health status of a seawater electrolysis hydrogen production system, which can quantify the health status of the seawater electrolysis hydrogen production system in real time and improve the operational reliability and maintenance efficiency of the seawater electrolysis hydrogen production system.
[0006] To achieve the above objectives, this application provides the following solution:
[0007] In a first aspect, this application provides a method for detecting the health status of a seawater electrolysis hydrogen production system, including:
[0008] Obtain a multimodal dataset of a seawater electrolysis hydrogen production system; the multimodal dataset includes multiple historical multimodal time-series operational data and a health status label for each historical multimodal time-series operational data;
[0009] The neural network model is trained based on the multimodal dataset to obtain a pre-trained neural network model. The neural network model includes multiple single-modal feature extraction modules, feature fusion modules, projection modules, and prediction modules. Each single-modal feature extraction module is used to extract features from the time-series running data of one modality. The feature fusion module is used to fuse the features extracted by multiple single-modal feature extraction modules to obtain a fused feature vector. The projection module is used to map the fused feature vector to a low-dimensional space. The prediction module is used to determine the health status based on the features output by the projection module.
[0010] Based on the unlabeled multimodal operation data of the seawater electrolysis hydrogen production system, the pre-trained neural network model was optimized using a self-supervised comparative learning method to obtain the optimized pre-trained neural network model.
[0011] Based on the real-time multimodal operation data of the seawater electrolysis hydrogen production system, the real-time health status of the system is determined using the optimized pre-trained neural network model.
[0012] Secondly, this application provides a health status detection device for a seawater electrolysis hydrogen production system, comprising:
[0013] The data acquisition unit is used to acquire a multimodal dataset of the seawater electrolysis hydrogen production system; the multimodal dataset includes multiple historical multimodal time-series operation data and a health status label for each historical multimodal time-series operation data.
[0014] A network pre-training module is used to train a neural network model based on the multimodal dataset to obtain a pre-trained neural network model. The neural network model includes multiple single-modal feature extraction modules, a feature fusion module, a projection module, and a prediction module. Each single-modal feature extraction module is used to extract features from the time-series running data of one modality. The feature fusion module is used to fuse the features extracted by multiple single-modal feature extraction modules to obtain a fused feature vector. The projection module is used to map the fused feature vector to a low-dimensional space. The prediction module is used to determine the health status based on the features output by the projection module.
[0015] The network optimization module is used to optimize the pre-trained neural network model based on the unlabeled multimodal operation data of the seawater electrolysis hydrogen production system, using a self-supervised comparative learning method, to obtain the optimized pre-trained neural network model.
[0016] The health monitoring module determines the real-time health status of the seawater electrolysis hydrogen production system based on the real-time multimodal operation data of the system and using the optimized pre-trained neural network model.
[0017] Thirdly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-described method for detecting the health status of a seawater electrolysis hydrogen production system.
[0018] According to the specific embodiments provided in this application, this application has the following technical effects: by constructing a multimodal fusion pre-trained neural network model and optimizing the pre-trained neural network model by combining a self-supervised contrastive learning method, this application achieves accurate assessment of the health status of the seawater electrolysis hydrogen production system throughout its entire life cycle under conditions of only a small amount of labeled data or even no abnormal samples. Attached Figure Description
[0019] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0020] Figure 1 This is an application environment diagram of a health status detection method for a seawater electrolysis hydrogen production system according to an embodiment of this application.
[0021] Figure 2 This is a flowchart illustrating a health status detection method for a seawater electrolysis hydrogen production system according to an embodiment of this application.
[0022] Figure 3 This is a schematic diagram of the structure of a temporal convolutional network in one embodiment of this application.
[0023] Figure 4 This is a partial structural diagram of a pre-trained neural network model in one embodiment of this application.
[0024] Figure 5 This is a schematic diagram of a fine-tuning framework based on self-supervised contrastive learning in one embodiment of this application.
[0025] Figure 6 This is a schematic diagram of the functional modules of a health status detection device for a seawater electrolysis hydrogen production system provided in an embodiment of this application.
[0026] Figure 7 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation
[0027] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0028] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.
[0029] The health status detection method for seawater electrolysis hydrogen production systems provided in this application embodiment can be applied to, for example... Figure 1 In the application environment shown, terminal 101 communicates with server 102 via a network. A data storage system can store the data that server 102 needs to process. The data storage system can be set up independently, integrated into server 102, or placed in the cloud or on another server. Terminal 101 can send operating data of the seawater electrolysis hydrogen production system to server 102. Server 102 detects the health status of the seawater electrolysis hydrogen production system based on the received operating data and determines the real-time health status of the system. Server 102 can provide feedback on the real-time health status of the seawater electrolysis hydrogen production system to terminal 101. Furthermore, in some embodiments, the health status detection method for the seawater electrolysis hydrogen production system can also be implemented independently by server 102 or terminal 101.
[0030] The terminal 101 can be, but is not limited to, various desktop computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, and smart in-vehicle devices. Portable wearable devices can include smartwatches, smart bracelets, and head-mounted devices. The server 102 can be implemented using a standalone server or a server cluster composed of multiple servers, or it can be a cloud server.
[0031] In one exemplary embodiment, such as Figure 2 As shown, a method for detecting the health status of a seawater electrolysis hydrogen production system is provided. This method is executed by a computer device, specifically by a terminal or server alone, or by both a terminal and a server. In this embodiment, the method is applied to... Figure 1 Taking server 102 as an example, the explanation includes the following steps 201 to 204.
[0032] Step 201: Obtain the multimodal dataset of the seawater electrolysis hydrogen production system.
[0033] The multimodal dataset is derived from the enterprise's historical operation records, including multiple historical multimodal time-series operation data and the health status label of each historical multimodal time-series operation data.
[0034] Historical multimodal time-series operational data includes key monitoring parameters for the hydrogen separator, oxygen separator, electrolyte circulation pump, waste heat recovery device, and core components of the electrolyzer. Key monitoring parameters include, but are not limited to: electrolysis voltage, electrolysis current, electrolyte temperature, electrolyte pH, gas purity, gas pressure, circulation flow rate, electrode impedance change rate, and membrane resistance.
[0035] Each time-series operational data point is organized according to a time sequence, including the specific values and units of each parameter, the health status level of the seawater electrolysis hydrogen production system at the corresponding time, and the health threshold range of each parameter under normal operating conditions. The health status is divided into four levels: "Excellent," "Good," "Average," and "Poor." Data labeling is based on equipment operating parameters, health thresholds, and engineering experience, combined with system-level comprehensive evaluation results to generate labels. The labeling process comprehensively considers the degree of deviation from parameter thresholds, the duration of abnormalities, the coupling effect between components, and historical failure modes, and uses weighted voting and the worst-case priority principle to determine the overall system status. The classification of health status levels in this application includes: the degree of deviation of key performance indicators from benchmark values, the magnitude of energy efficiency decline, the frequency of fault alarms, and manual operation and maintenance records.
[0036] In practical applications, the original multimodal operation data is first cleaned, aligned, normalized, and segmented in time series. Then, an optimized pre-trained neural network model is used to detect the health status and output a health score. At the same time, the health level is mapped to four levels: "excellent", "good", "medium" and "poor" by setting a threshold or classifier.
[0037] Step 202: Train the neural network model using the multimodal dataset to obtain a pre-trained neural network model. The neural network model undergoes supervised pre-training using the multimodal dataset.
[0038] The neural network model includes multiple single-modal feature extraction modules, a feature fusion module, a projection module, and a prediction module. Each single-modal feature extraction module is used to extract features from the time-series running data of one modality. The feature fusion module is used to fuse the features extracted by multiple single-modal feature extraction modules to obtain a fused feature vector. The projection module is used to map the fused feature vector to a low-dimensional space. The prediction module is used to determine the health status based on the features output by the projection module.
[0039] The single-modal feature extraction module includes a temporal convolutional network and a vector segmentation submodule. The temporal convolutional network is used to extract features from the temporal data of a single modality, obtaining a high-dimensional vector. The vector segmentation submodule is used to segment the high-dimensional vector along the channel dimension into the principal feature vector and guiding coefficient of the corresponding modality. The features of each modality include the principal feature vector and the guiding coefficient.
[0040] like Figure 3 As shown, the single-modality feature extraction module uses a temporal convolutional network to process the temporal data of a single modality. The backbone is a 1×1 convolutional layer, and the remaining parts consist of residual blocks. Each residual block contains two dilated causal convolutional layers, followed by layer normalization, ReLU activation function, and Dropout layer. Compared with batch normalization, layer normalization can more effectively address the differences in feature distributions of different parameters, thereby maximizing feature discovery. Layer normalization is applied after the dilated causal convolutional layer within each residual block to eliminate the impact of differences in the dimensions and distributions of different sensor parameters on feature extraction.
[0041] In a specific application example, such as Figure 4 As shown, identity embedding is performed on n modal temporal execution data, and then each data point is input into n single-modal feature extraction modules, where n is the number of modalities. Assume the original single-modal temporal execution data is... L is the length of the single-modal time-series running data, which is expanded after identity embedding to: .in, For identity embedding extended single-modal time-series runtime data, This is the Lth data point in the single-modal timing data. For modal identification, For character identification, and All are learnable embedding parameter vectors.
[0042] Each unimodal feature extraction module processes the received time-series runtime data, outputting a high-dimensional vector. This high-dimensional vector is then divided into two parts along the channel dimension: the first part serves as the principal feature vector for that modality, and the second part, after processing with an activation function, serves as the guiding coefficient for that modality. During the feature extraction process for each modality, its corresponding unimodal feature extraction module acts as the main branch, while the unimodal feature extraction modules for the other modalities act as auxiliary branches.
[0043] The feature fusion module fuses features extracted by multiple single-modality feature extraction modules. This process includes: for the i-th modality, aggregating the mean of the guidance coefficients of the remaining modalities to obtain the aggregated guidance coefficient of the i-th modality. , i ∈[1, nThe principal eigenvector of the i-th mode. The aggregation guiding coefficient of the i-th mode Element-wise multiplication yields the modulation feature vector of the i-th mode. The fused feature vector is obtained by summing the modulation feature vectors of all modes. : .
[0044] Step 203: Based on the unlabeled multimodal operation data of the seawater electrolysis hydrogen production system, the pre-trained neural network model is optimized using a self-supervised comparative learning method to obtain the optimized pre-trained neural network model.
[0045] Pre-trained neural network models can learn general operating condition characteristics; however, in practical applications, seawater electrolysis hydrogen production systems exhibit significant differences under various operating conditions. To accurately assess the health status of seawater electrolysis hydrogen production systems, it is necessary to fine-tune the model using actual operating condition data. Typically, in actual operating conditions, only data under healthy conditions (i.e., normal data when the health status is excellent or good) can be collected for seawater electrolysis hydrogen production systems. Data under abnormal health conditions is relatively scarce.
[0046] Therefore, this application constructs a fine-tuning framework based on self-supervised contrastive learning, which utilizes unlabeled multimodal operating data under actual working conditions containing only health statuses of "excellent" or "good" to optimize a pre-trained neural network model. For example... Figure 5 As shown, the fine-tuning framework includes a structurally similar network to be fine-tuned (i.e., a pre-trained neural network model) and a target network. The target network includes multiple unimodal feature extraction modules, feature fusion modules, and projection modules from the pre-trained neural network model; that is, the target network does not contain a prediction module. The parameters of the pre-trained neural network model are... The parameters of the target network are .
[0047] In a specific application example, step 203 includes steps 31 to 38.
[0048] Step 31: Enhance the unlabeled multimodal operating data of the seawater electrolysis hydrogen production system to obtain the first multimodal enhancement signal. Second multimodal enhancement signal .
[0049] Specifically, at least one of amplitude scaling and Gaussian noise is randomly applied to the unlabeled multimodal running data to improve feature representation capability and generate a first multimodal enhancement signal. Second multimodal enhancement signal The pre-trained neural network model and the target network are input separately and follow certain regularization relationships to ensure the balance between the two enhancement methods.
[0050] Step 32: Using the multiple single-modal feature extraction modules and feature fusion modules of the pre-trained neural network model, the first multimodal enhanced signal is sequentially subjected to feature extraction and fusion to obtain the first fused feature vector. The second multimodal enhancement signal is sequentially subjected to feature extraction and fusion using multiple single-modal feature extraction modules and feature fusion modules of the target network to obtain a second fused feature vector. .
[0051] Step 33: The projection module of the pre-trained neural network model is used to map the first fused feature vector to a low-dimensional space to obtain the first transformed feature vector. The projection module of the target network maps the second fused feature vector to a low-dimensional space to obtain the second transformed feature vector. .
[0052] Specifically, the projection module is a two-layer fully connected neural network. The middle layer uses the ReLU activation function, while the output layer has no activation function. Using ReLU as the activation function can improve sparsity, reduce information loss, speed up fine-tuning, and simplify the model. The fully connected neural network is a multilayer perceptron.
[0053] Step 34: Use the prediction module of the pre-trained neural network model. The first transformed feature vector is subjected to a nonlinear transformation to obtain the health status prediction result. .
[0054] Step 35: Perform L2 norm normalization on the first transformed feature vector and the second transformed feature vector respectively to obtain the first normalized feature and the second normalized feature.
[0055] Step 36: Calculate the mean squared error loss based on the health status prediction result and the second normalized feature: ;in, For mean square error loss, The results of health status prediction after L2 norm normalization are as follows. This is the second normalization feature.
[0056] Step 37: Calculate the symmetric loss based on the first normalized feature and the stopping gradient of the target network. ;in, For symmetrical loss, The stopping gradient of the target network. L2 norm normalized , This is the first normalized feature.
[0057] Step 38: Adjust the parameters of the pre-trained neural network model and the target network according to the mean squared error loss and the symmetric loss until the mean squared error loss and the symmetric loss converge to obtain the optimized pre-trained neural network model.
[0058] The parameters of the pre-trained neural network model and the target network are adjusted using the following formula:
[0059] ;
[0060] ;
[0061] in, For the parameters of the pre-trained neural network model, For the parameters of the target network, For learning rate, To control the hyperparameters of the update speed, , For the total loss function, , For mean square error loss, For symmetrical loss, For the parameters of the model optimizer, For model optimizers.
[0062] The optimized pre-trained neural network model is used to evaluate the health status of the seawater electrolysis hydrogen production system in real time and output a health score or health status level.
[0063] Step 204: Based on the real-time multimodal operation data of the seawater electrolysis hydrogen production system, the optimized pre-trained neural network model is used to determine the real-time health status of the seawater electrolysis hydrogen production system.
[0064] The optimized pre-trained neural network model is deployed on edge computing devices or cloud servers to enable online monitoring of the health status of the seawater electrolysis hydrogen production system and early warning of degradation.
[0065] This application further generates a health trend curve based on the real-time health status of the seawater electrolysis hydrogen production system and issues performance degradation early warning information, supporting remote access and diagnosis.
[0066] This application constructs a multimodal fusion pre-trained neural network model and combines it with a fine-tuning strategy based on self-supervised contrastive learning to achieve accurate health status assessment of seawater electrolysis hydrogen production systems under conditions with only limited labeled data or even no outliers. This not only improves the model's generalization ability under complex operating conditions but also provides a feasible technical path for future intelligent operation and maintenance. The optimized pre-trained neural network model can accurately assess the health status of seawater electrolysis hydrogen production systems throughout their entire lifecycle.
[0067] Based on the same inventive concept, this application also provides an apparatus for implementing the method described above. The solution provided by this apparatus is similar to the solution described in the above method; therefore, specific limitations in one or more apparatus embodiments provided below can be found in the limitations of the method described above, and will not be repeated here.
[0068] In one exemplary embodiment, such as Figure 6 As shown, a health status detection device for a seawater electrolysis hydrogen production system is provided, comprising the following functional modules.
[0069] The data acquisition unit 601 is used to acquire a multimodal dataset of the seawater electrolysis hydrogen production system. The multimodal dataset includes multiple historical multimodal time-series operational data and a health status label for each historical multimodal time-series operational data.
[0070] The network pre-training module 602 is used to train the neural network model based on the multimodal dataset to obtain a pre-trained neural network model. The neural network model includes multiple single-modal feature extraction modules, a feature fusion module, a projection module, and a prediction module. Each single-modal feature extraction module is used to extract features from the time-series data of one modality. The feature fusion module is used to fuse the features extracted by multiple single-modal feature extraction modules to obtain a fused feature vector. The projection module is used to map the fused feature vector to a low-dimensional space. The prediction module is used to determine the health status based on the features output by the projection module.
[0071] The network optimization module 603 is used to optimize the pre-trained neural network model based on the unlabeled multimodal operation data of the seawater electrolysis hydrogen production system using a self-supervised comparative learning method, so as to obtain the optimized pre-trained neural network model.
[0072] The health monitoring module 604 determines the real-time health status of the seawater electrolysis hydrogen production system based on the real-time multimodal operation data of the system and using the optimized pre-trained neural network model.
[0073] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 7As shown, the computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores multimodal operational data of the seawater electrolysis hydrogen production system. The I / O interfaces are used for information exchange between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When executed by the processor, the computer program implements a health status detection method for the seawater electrolysis hydrogen production system.
[0074] Those skilled in the art will understand that Figure 7 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.
[0075] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.
[0076] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.
[0077] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.
[0078] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).
[0079] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, etc., and are not limited to these.
[0080] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0081] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.
Claims
1. A method for detecting the health status of a seawater electrolysis hydrogen production system, characterized in that, The method includes: Obtain a multimodal dataset of a seawater electrolysis hydrogen production system; the multimodal dataset includes multiple historical multimodal time-series operational data and a health status label for each historical multimodal time-series operational data; The neural network model is trained according to the multi-modal data set to obtain a pre-trained neural network model; the neural network model comprises a plurality of single-modal feature extraction modules, a feature fusion module, a projection module and a prediction module; each single-modal feature extraction module is used for feature extraction on time-series operation data of one mode; specifically, n modal time-series operation data is input into n single-modal feature extraction modules after identity embedding, and n is the number of modes; assuming that the original single-modal time-series operation data is , L is the length of the single-modal time-series operation data, and is expanded to after identity embedding; wherein, is the single-modal time-series operation data after identity embedding expansion, is the Lth data in the single-modal time-series operation data, is a mode identifier, is a role identifier, and are all learnable embedding parameter vectors; The single-modal feature extraction module includes a temporal convolutional network and a vector segmentation submodule. The temporal convolutional network is used to extract features from the temporal data of a single modality to obtain a high-dimensional vector. The vector segmentation submodule is used to segment the high-dimensional vector along the channel dimension into the principal feature vector and guiding coefficient of the corresponding modality. The features of each modality include the principal feature vector and the guiding coefficient. The feature fusion module is used to fuse the features extracted by multiple single-modality feature extraction modules to obtain a fused feature vector. The process of fusing the features extracted by multiple single-modality feature extraction modules includes: for the i-th modality, the guiding coefficients of the other modalities are averaged and aggregated to obtain the aggregated guiding coefficient of the i-th modality; the principal feature vector of the i-th modality is multiplied element-wise with the aggregated guiding coefficient of the i-th modality to obtain the modulation feature vector of the i-th modality; the modulation feature vectors of all modalities are added together to obtain the fused feature vector. The projection module is used to map the fused feature vector to a low-dimensional space; the prediction module is used to determine the health status based on the features output by the projection module. Based on the unlabeled multimodal operation data of the seawater electrolysis hydrogen production system, the pre-trained neural network model was optimized using a self-supervised comparative learning method to obtain the optimized pre-trained neural network model. Based on the real-time multimodal operation data of the seawater electrolysis hydrogen production system, the real-time health status of the system is determined using the optimized pre-trained neural network model. In practical applications, the original multimodal operation data is first cleaned, aligned, normalized, and segmented according to time series. Then, the optimized pre-trained neural network model is used to detect the health status and output a health score. Simultaneously, the health score is mapped to four levels: "excellent," "good," "medium," and "poor" by setting a threshold or classifier.
2. The method for detecting the health status of a seawater electrolysis hydrogen production system according to claim 1, characterized in that, The historical multimodal time-series operation data, the unlabeled multimodal operation data, and the real-time multimodal operation data all include key monitoring parameters of the hydrogen separator, oxygen separator, electrolyte circulation pump, waste heat recovery device, and core components of the electrolyzer. The key monitoring parameters include: electrolysis voltage, electrolysis current, electrolyte temperature, electrolyte pH value, gas purity, gas pressure, circulation flow rate, electrode impedance change rate, and membrane resistance.
3. The method for detecting the health status of a seawater electrolysis hydrogen production system according to claim 1, characterized in that, Based on unlabeled multimodal operation data of the seawater electrolysis hydrogen production system, a self-supervised comparative learning method was used to optimize the pre-trained neural network model, resulting in an optimized pre-trained neural network model, including: The unlabeled multimodal operating data of the seawater electrolysis hydrogen production system were enhanced to obtain the first multimodal enhancement signal and the second multimodal enhancement signal; The first multimodal enhancement signal is sequentially subjected to feature extraction and fusion using multiple single-modal feature extraction modules and feature fusion modules of the pre-trained neural network model to obtain a first fused feature vector; The second multimodal enhancement signal is sequentially subjected to feature extraction and fusion using multiple single-modal feature extraction modules and feature fusion modules of the target network to obtain a second fused feature vector; the target network includes multiple single-modal feature extraction modules, feature fusion modules and projection modules of a pre-trained neural network model; The projection module of the pre-trained neural network model maps the first fused feature vector to a low-dimensional space to obtain the first transformed feature vector; The projection module of the target network is used to map the second fused feature vector to a low-dimensional space to obtain the second transformed feature vector; The prediction module of the pre-trained neural network model performs a nonlinear transformation on the first transformed feature vector to obtain the health status prediction result; The first transformed feature vector and the second transformed feature vector are respectively normalized by L2 norm to obtain the first normalized feature and the second normalized feature; The mean squared error loss is calculated based on the health status prediction results and the second normalized feature. Calculate the symmetric loss based on the first normalized feature and the stopping gradient of the target network; The parameters of the pre-trained neural network model and the target network are adjusted according to the mean squared error loss and the symmetric loss until the mean squared error loss and the symmetric loss converge to obtain the optimized pre-trained neural network model.
4. The method for detecting the health status of a seawater electrolysis hydrogen production system according to claim 3, characterized in that, The formula for calculating the mean squared error loss is as follows: ； in, For mean square error loss, For the parameters of the pre-trained neural network model, For the parameters of the target network, This is the first transformed feature vector. This is the second transformation feature vector. This is the second normalized feature. For health status prediction results, L2 norm normalized , This is the prediction module.
5. The method for detecting the health status of a seawater electrolysis hydrogen production system according to claim 3, characterized in that, The formula for calculating the symmetry loss is: ； in, For symmetrical loss, This is the first transformed feature vector. This is the second transformation feature vector. The first normalized feature, The stopping gradient of the target network. L2 norm normalized .
6. The method for detecting the health status of a seawater electrolysis hydrogen production system according to claim 3, characterized in that, The parameters of the pre-trained neural network model and the target network are adjusted using the following formula: ； ； in, For the parameters of the pre-trained neural network model, For the parameters of the target network, For learning rate, To control the hyperparameters of the update speed, , For the total loss function, , For mean square error loss, For symmetrical loss, For the parameters of the model optimizer, For model optimizers.
7. A health status detection device for a seawater electrolysis hydrogen production system, characterized in that, The apparatus performs the health status detection method for a seawater electrolysis hydrogen production system according to any one of claims 1-6, and the apparatus comprises: The data acquisition unit is used to acquire a multimodal dataset of the seawater electrolysis hydrogen production system; the multimodal dataset includes multiple historical multimodal time-series operation data and a health status label for each historical multimodal time-series operation data. A network pre-training module is used to train a neural network model based on the multimodal dataset to obtain a pre-trained neural network model. The neural network model includes multiple single-modal feature extraction modules, a feature fusion module, a projection module, and a prediction module. Each single-modal feature extraction module is used to extract features from the time-series running data of one modality. The feature fusion module is used to fuse the features extracted by multiple single-modal feature extraction modules to obtain a fused feature vector. The projection module is used to map the fused feature vector to a low-dimensional space. The prediction module is used to determine the health status based on the features output by the projection module. The network optimization module is used to optimize the pre-trained neural network model based on the unlabeled multimodal operation data of the seawater electrolysis hydrogen production system, using a self-supervised comparative learning method, to obtain the optimized pre-trained neural network model. The health monitoring module determines the real-time health status of the seawater electrolysis hydrogen production system based on the real-time multimodal operation data of the system and using the optimized pre-trained neural network model.
8. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the health status detection method for a seawater electrolysis hydrogen production system according to any one of claims 1-6.