Safety monitoring system for hydrogen supply

Abstract

The invention relates to the field of hydrogen supply and safety monitoring, and discloses a safety monitoring system for hydrogen supply, which comprises an acquisition processing module for acquiring multi-source data of a hydrogen supply system in real time and preprocessing the multi-source data to obtain processed data; the prediction model module is used for constructing a prediction model by selecting a time sequence data processing algorithm based on the processing data, and training, verifying and updating the prediction model; the early warning processing module is used for predicting the safety of the hydrogen supply system according to the prediction model to obtain a prediction result and adopting emergency processing in combination with real-time multi-source data; and the optimization module is used for performing performance evaluation, algorithm improvement and upgrading and knowledge accumulation and sharing. Multi-source data are collected in real time, processed data are obtained through preprocessing, a proper algorithm is selected to construct a model according to data characteristics, decision suggestions are extracted according to a prediction result, multi-channel early warning is performed, emergency measures are taken, and safety risks are responded and handled in time.

Description

Technical Field

[0001] This invention relates to the field of hydrogen supply and safety monitoring technology, specifically a safety monitoring system for hydrogen supply. Background Technology

[0002] With the ever-increasing global demand for clean energy, hydrogen, as an efficient, clean energy carrier with broad application prospects, is finding increasingly widespread use in numerous fields. In the transportation sector, hydrogen fuel cell vehicles, with their advantages of zero emissions and high energy density, are gradually becoming an important direction for the development of new energy vehicles. In the energy sector, hydrogen can be used for distributed power generation and energy storage. However, hydrogen's flammability, explosiveness, and strong diffusion make safety during its production, storage, transportation, and use crucial. Hydrogen leaks, fires, and other safety accidents can lead to explosions, causing significant losses to life and property. Therefore, establishing a reliable hydrogen supply safety monitoring system is of great significance for ensuring the safe and stable operation of the hydrogen supply system and promoting the healthy development of the hydrogen energy industry.

[0003] Traditional safety monitoring systems for hydrogen supply lack the ability to deeply analyze and mine historical data, cannot use artificial intelligence technology to predict potential safety risks, and cannot take preventive maintenance measures in advance, making it difficult to meet the high requirements for safety monitoring in large-scale commercial hydrogen supply scenarios. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a safety monitoring system for hydrogen supply, which solves the problem that traditional safety monitoring systems for hydrogen supply cannot take preventative maintenance measures in advance and are difficult to meet the high safety monitoring requirements of large-scale commercial hydrogen supply scenarios.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a safety monitoring system for hydrogen supply, comprising: The data acquisition and processing module is used to acquire multi-source data from the hydrogen supply system in real time and preprocess it to obtain processed data. The prediction model module is used to build a prediction model based on the processed data by selecting time series data processing algorithms and to train, validate and update it. Early warning and processing module: It is used to predict the safety of the hydrogen supply system based on the prediction model, obtain the prediction results, and take emergency measures in combination with real-time multi-source data. Optimization module: Used to optimize system performance through performance evaluation, algorithm improvement and upgrade, and knowledge accumulation and sharing, and to accumulate and share knowledge.

[0006] Preferably, the data acquisition and processing module includes a data acquisition unit, a data cleaning unit, a data labeling unit, and a data storage unit. The data acquisition unit is used to acquire multi-source data from the hydrogen supply system, including hydrogen concentration, pressure, temperature, flow rate, equipment vibration, rotational speed, and current data. The data cleaning unit... Noise removal, among which To calculate the average value of the data within the window, arrive To find the range of summation, x j For a data point at position j in the original time series data, outliers are removed and duplicates are deduplicated according to the rule x < μ - 3σ or x > μ + 3σ, where μ is the mean and σ is the standard deviation. The data labeling unit is used to label the data as normal or abnormal, fault or leakage type, location coordinates and severity. The data storage unit is used to select a relational cloud database, design data tables to store different types of data, and establish inter-table relationships to obtain processed data.

[0007] Preferably, the data acquisition unit includes a combustible gas detection subunit, a flame detection subunit, a hydrogen detection subunit, an equipment detection subunit, a historical data collection subunit, and a data transmission subunit. The combustible gas detection subunit is used to detect combustible gases in the working environment of the hydrogen supply system. The flame detection subunit is used to detect flames in the working environment of the hydrogen supply system, including open flames and smoldering flames. The hydrogen detection subunit is used to detect hydrogen data of the hydrogen supply system, including concentration, pressure, temperature, and flow rate. The equipment detection subunit is used to detect equipment data of the hydrogen supply system, including vibration, rotational speed, and current. The historical data collection subunit is used to collect detailed records of past faults and leaks in the hydrogen supply system, including the time, location, type, cause, handling measures, and consequences of the events. The data transmission subunit is used to transmit the various data.

[0008] Preferably, the prediction model module includes a model selection unit, a model training unit, and a validation and update unit. The model selection unit is used to select an algorithm for time series data processing based on the characteristics of multi-source data. The model training unit is used to update parameters using the mean squared error loss function and the stochastic gradient descent algorithm to obtain the prediction model. The validation and update unit is used to divide the dataset, adjust the hyperparameters using the validation set, and periodically update the training set with new data and adjust the hyperparameters.

[0009] Preferably, the features include the time-series characteristics of the data, the diversity of data types, the trend and fluctuation range of the data, and the algorithms for processing the time-series data include Long Short-Term Memory (LSTM) network algorithms, gated recurrent unit (ROU) algorithms, and autoregressive integral moving average (ARM) models. The mean squared error loss function includes... in For the predicted value, y (i) The true label is N, and the number of samples is N. The stochastic gradient descent algorithm includes... Where θ is a parameter and α is the learning rate.

[0010] Preferably, the early warning processing module includes a prediction unit, a decision generation unit, and an alarm processing unit. The prediction unit is used to input the processed data collected and preprocessed in real time into the prediction model, and output the prediction result of the safety status of the hydrogen supply system through model calculation. The decision generation unit is used to extract decision suggestions corresponding to different fault types and severity from the decision database according to the prediction results and based on the pre-set fault type and severity classification rules. The alarm processing unit is used to issue early warnings through appropriate channels and take emergency measures according to the decision suggestions.

[0011] Preferably, the corresponding channels include audible and visual alarms, SMS notifications, and email reminders; the prediction results include determining whether the hydrogen supply system is in a normal state, whether there is a risk of failure, or whether there is a risk of leakage; the decision recommendations include countermeasures for different types and severity of failures; and the emergency response includes closing relevant valves, evacuating surrounding personnel, and activating emergency rescue plans.

[0012] Preferably, the optimization module includes a performance evaluation unit, an improvement and upgrade unit, and an accumulation and sharing unit. The performance evaluation unit statistically analyzes and visualizes early warning indicators. The improvement and upgrade unit is used to integrate long short-term memory network models using ensemble learning algorithms to explore new algorithms. The accumulation and sharing unit is used to establish an online knowledge base, use knowledge graph technology to store knowledge in the form of triples, and update and share the knowledge.

[0013] Preferably, the early warning indicators include early warning accuracy. Fault diagnosis accuracy Among them TP warn To ensure the correct number of warnings, FP warn TP is the number of false alarms. diag To accurately diagnose the number of failures, FP diag To erroneously diagnose the number of faults, the ensemble learning algorithm integrates a long short-term memory network model, including... Where w k The weights are determined through cross-validation.

[0014] A safety monitoring method for hydrogen supply includes the following steps: S1. Data Acquisition and Processing: The data acquisition and processing module collects multi-source data from the hydrogen supply system in real time and preprocesses it to obtain processed data. S2. Predictive Model: Based on the processed data, the predictive model module constructs a predictive model by selecting time series data processing algorithms and then trains, validates, and updates it. S3. Early Warning and Handling: The early warning and handling module predicts the safety of the hydrogen supply system based on the prediction model, obtains the prediction results, and takes emergency measures in combination with real-time multi-source data. S4. Optimization: Through the optimization module, performance is evaluated, algorithms are improved and upgraded, and knowledge is accumulated and shared to optimize system performance and accumulate and share knowledge.

[0015] This invention provides a safety monitoring system for hydrogen supply. It has the following beneficial effects: 1. This invention collects multi-source data in real time, preprocesses the data to obtain processed data, selects appropriate algorithms to build models based on data characteristics, extracts decision suggestions based on prediction results, provides early warnings through multiple channels and takes emergency measures to respond to and handle safety risks in a timely manner, statistically analyzes and visualizes early warning indicators, integrates models, explores new algorithms, establishes a knowledge base and updates shared knowledge, and continuously improves system performance and diagnostic capabilities. This enables the early detection of potential risks and the implementation of preventive maintenance measures, meeting the high requirements for safety monitoring in large-scale commercial hydrogen supply scenarios and ensuring the safe and stable operation of the hydrogen supply system.

[0016] 2. This invention collects multi-source information such as hydrogen concentration, pressure, equipment operation data, combustible gas and flame data, and historical fault records. After filtering, outlier removal, and deduplication by the data cleaning unit, the data is further annotated in detail by the data annotation unit. Finally, the data is stored in a relational cloud database and associated with it, resulting in accurate, standardized, and orderly processed data. This ensures a comprehensive understanding of the operating status of the hydrogen supply system, providing a reliable basis for subsequent analysis and decision-making, and enabling the monitoring system to operate based on real and effective data.

[0017] 3. This invention constructs a model by selecting suitable time series data processing algorithms from a variety of algorithms based on the characteristics of multi-source data. During the training process, the mean squared error loss function and stochastic gradient descent algorithm are used to optimize the parameters, and the hyperparameters are adjusted with the help of the validation set. The training set is updated regularly with new data, enabling the model to accurately learn data patterns and accurately predict potential safety risks of the hydrogen supply system. This allows for early prevention, effectively reducing the probability of accidents and providing strong protection for the safe operation of large-scale commercial hydrogen supply scenarios.

[0018] 4. This invention uses the results of a predictive model combined with real-time multi-source data to quickly determine the safety status of the system. Once a fault or leakage risk is detected, the invention extracts the corresponding strategy from the decision library according to preset rules, issues early warnings in a timely manner through multiple channels, and quickly takes emergency measures such as closing valves, evacuating personnel, and activating emergency rescue plans. This can effectively control the spread of danger in the early stages of an accident, minimize casualties and property losses, and ensure the safety of the hydrogen supply system and the surrounding environment.

[0019] 5. This invention provides intuitive evidence for system improvement by statistically analyzing key early warning indicators and presenting them visually. It employs ensemble learning algorithms to fuse models and explore new algorithms, thereby enhancing the system's diagnostic and predictive capabilities. It establishes an online knowledge base, uses knowledge graph technology to store and update knowledge, and enables sharing. This allows the system to continuously optimize its performance, adapt to complex and ever-changing hydrogen supply scenarios, and promote the inheritance and application of knowledge, thus driving the continuous development of hydrogen supply safety monitoring technology. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the system architecture of a safety monitoring system for hydrogen supply proposed in this invention; Figure 2 This is a schematic diagram of the data acquisition and processing module architecture of a safety monitoring system for hydrogen supply proposed in this invention. Figure 3 This is a schematic diagram of the process flow for a safety monitoring method for hydrogen supply proposed in this invention. Detailed Implementation

[0021] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0022] Please see the appendix Figure 1 This invention provides a safety monitoring system for hydrogen supply, comprising: The data acquisition and processing module is used to acquire multi-source data from the hydrogen supply system in real time and preprocess it to obtain processed data. The module includes a data acquisition unit, a data cleaning unit, a data labeling unit, and a data storage unit. The data acquisition unit collects multi-source data from the hydrogen supply system, including hydrogen concentration, pressure, temperature, flow rate, equipment vibration, rotational speed, and current data. The data cleaning unit... Noise removal, among which To calculate the average value of the data within the window, arrive To find the range of summation, xj For the data point at position j in the original time series data, outliers are removed and deduplication is performed according to the rule x < μ - 3σ or x > μ + 3σ, where μ is the mean and σ is the standard deviation. The data labeling unit is used to label the data as normal or abnormal, fault or leakage type, location coordinates and severity. The data storage unit is used to select a relational cloud database, design data tables to store different types of data, and establish inter-table relationships to obtain the processed data.

[0023] The data acquisition unit includes a combustible gas detection subunit, a flame detection subunit, a hydrogen detection subunit, an equipment detection subunit, a historical data collection subunit, and a data transmission subunit. The combustible gas detection subunit is used to detect combustible gases in the working environment of the hydrogen supply system. The flame detection subunit is used to detect flames in the working environment of the hydrogen supply system, including open flames and smoldering flames. The hydrogen detection subunit is used to detect hydrogen data in the hydrogen supply system, including concentration, pressure, temperature, and flow rate. The equipment detection subunit is used to detect equipment data in the hydrogen supply system, including vibration, rotational speed, and current. The historical data collection subunit is used to collect detailed records of past faults and leaks in the hydrogen supply system, including the time, location, type, cause, handling measures, and consequences of the events. The data transmission subunit is used to transmit all the data.

[0024] Specifically, by acquiring multi-source data from the hydrogen supply system in real time and preprocessing it, processed data is obtained. This processed data is then used to collect data on combustible gases, flames, hydrogen (concentration, pressure, temperature, flow rate), equipment (vibration, speed, current), and records of past faults and leaks in the hydrogen supply system's operating environment, utilizing principles such as catalytic combustion, ultraviolet sensing, and electrochemical sensing. For example, combustible gas detection modules consisting of 8-channel catalytic combustion detectors are deployed in the unloading area, compressor area, hydrogen storage cylinder group area, and hydrogen refueling area. The detector ends are fixed to the process pipelines using 316L stainless steel compression fittings. These detectors are ExdⅡCT4 explosion-proof rated, with a range of 0-100% LEL and an output signal of 4-20mA. This allows for real-time and accurate detection of hydrogen concentration in each area, and stable transmission of the concentration signal. Furthermore, in the same area as the combustible gas detection modules, a flame detection module consisting of 8-channel infrared-ultraviolet composite detectors is installed. These flame detectors are mounted 3m above the ground using explosion-proof brackets, are ExdⅡCT6 explosion-proof rated, and have a detection distance of up to 25m. This allows for the timely detection of flame conditions in various areas, especially for the effective identification of potential fire sources.

[0025] These data are transmitted to the data cleaning unit via a data transmission subunit, and each I / O module is connected via a PROFINET bus. The I / O modules within the PLC control cabinet are connected to field devices via shielded twisted-pair cables. The cabinet grounding resistance is ≤4Ω, and the PLC system has a 15% margin in the number of I / O points. The control cabinet is equipped with dual 24VDC power supplies. This establishes a stable and efficient data transmission and control network, ensuring rapid acquisition and processing of data from field devices, and providing good scalability and redundancy.

[0026] The data cleaning unit uses the mean filtering formula Noise removal is performed by eliminating outliers according to the rule x < μ - 3σ or x > μ + 3σ, and deduplication is also carried out. Next, the data labeling unit labels the cleaned data with normal or abnormal status, fault or leak type, location coordinates, and severity. Then, the data storage unit uses a relational cloud database, designs data tables to store different types of data, and establishes relationships between tables. This obtains accurate, standardized, and orderly processed data, enabling effective organization and preliminary analysis of multi-source raw data from the hydrogen supply system, providing reliable data support for the construction of subsequent prediction models.

[0027] The prediction model module is used to build, train, validate, and update a prediction model based on the processed data by selecting time series data processing algorithms. The prediction model module includes a model selection unit, a model training unit, and a validation and update unit. The model selection unit is used to select time series data processing algorithms according to the characteristics of multi-source data. The model training unit is used to update parameters using the mean squared error loss function and the stochastic gradient descent algorithm to obtain the prediction model. The validation and update unit is used to divide the dataset, adjust hyperparameters using the validation set, and periodically update the training set with new data and adjust hyperparameters.

[0028] Features include the time-series characteristics of the data, the diversity of data types, and the trends and fluctuation ranges of the data. Algorithms for time-series data processing include Long Short-Term Memory (LSTM) networks, gated recurrent unit (GRU) algorithms, and autoregressive integral moving average (ARM) models. Mean squared error loss functions include... in For the predicted value, y (i) Where N is the true label and N is the number of samples, the stochastic gradient descent algorithm includes... Where θ is a parameter and α is the learning rate.

[0029] Specifically, based on the processed data, a predictive model is constructed and trained, validated, and updated by selecting time series data processing algorithms. This is achieved by considering the characteristics of multi-source data—time series features, diverse types, varying trends, and fluctuating ranges—and selecting appropriate time series data processing algorithms from sources such as Long Short-Term Memory (LSTM) networks, gated recurrent unit (GRU) algorithms, and autoregressive integral moving average (ARM) models. The model training unit employs the mean squared error loss function. Using stochastic gradient descent algorithm The model parameters are continuously updated to build a preliminary prediction model. The validation update unit divides the dataset into training, validation, and test sets. It uses the validation set to adjust hyperparameters, such as the learning rate and the number of hidden layer neurons, and periodically updates the training set with new data and readjusts the hyperparameters. This achieves the optimization of the prediction model's ability to accurately learn and predict hydrogen supply system data, enabling it to more accurately predict potential safety risks of the system.

[0030] The early warning processing module is used to predict the safety of the hydrogen supply system based on the prediction model, obtain the prediction results, and take emergency measures in combination with real-time multi-source data. The early warning processing module includes a prediction unit, a decision generation unit, and an alarm processing unit. The prediction unit is used to input the real-time collected and pre-processed data into the prediction model, and output the prediction results of the safety status of the hydrogen supply system through model calculation. The decision generation unit is used to extract decision suggestions corresponding to different fault types and severity levels from the decision database based on the prediction results and according to the pre-set fault type and severity classification rules. The alarm processing unit is used to issue early warnings through appropriate channels and take emergency measures based on the decision suggestions.

[0031] The corresponding channels include audible and visual alarms, SMS notifications, and email reminders. The prediction results include whether the hydrogen supply system is in normal condition, whether there is a risk of failure, or whether there is a risk of leakage. The decision-making recommendations include response measures for different types and severity of failures. Emergency handling includes shutting down relevant valves, evacuating surrounding personnel, and activating emergency rescue plans.

[0032] Specifically, the safety of the hydrogen supply system is predicted using a predictive model. The predicted results are then combined with real-time multi-source data to implement emergency response measures. The pre-processed, real-time collected data is input into the trained predictive model, which calculates based on learned patterns and rules, outputting a prediction of the hydrogen supply system's safety status. The decision generation unit, based on the prediction results and according to pre-defined fault type and severity classification rules, extracts corresponding decision suggestions from the decision database. For example, if the prediction results indicate an abnormally high hydrogen concentration reaching the moderate leakage standard, the decision generation unit will extract response strategies for moderate hydrogen leakage from the decision database.

[0033] The alarm processing unit issues warnings through various channels, including audible and visual alarms (such as flashing alarm lights and loudspeaker audible alarms), SMS notifications (sending warning SMS messages to relevant personnel's mobile phones), and email alerts (sending detailed warning emails to maintenance personnel's email addresses). It then takes emergency measures based on decision-making recommendations, such as closing relevant valves to prevent further hydrogen leakage, evacuating surrounding personnel to ensure their safety, and activating emergency rescue plans to prepare for potential dangerous situations. Other measures include shut-off devices driven by pneumatic ball valves and solenoid valves. The emergency shut-off valve uses a Swagelok pneumatic ball valve with a 316L valve body and a response time ≤2s. Its drive gas path is connected to the nitrogen manifold via a 1 / 2" compression fitting. Emergency shut-off buttons are available at the station level (shutting down the entire station) and the equipment level (isolator isolation). The knob reset normally closed contact has a capacity of 3A@24VDC. When the hydrogen concentration is >0.4%, the detector signal is converted by the PLC analog input module, triggering an audible and visual alarm on the host computer; when the concentration is >1%, the PLC outputs a DO signal to drive the shut-off valve to close the corresponding pipeline and start the forced ventilation system. After the flame detector identifies the fire source, it directly triggers the entire station's ESD through the dry contact signal, and simultaneously links the fire sprinkler system. Thus, when hydrogen leakage exceeds the standard or a flame is detected, it can quickly and accurately execute the shut-off action, link other safety equipment, and effectively control the spread of the dangerous situation. Therefore, it can respond promptly when there is a safety risk in the hydrogen supply system, realizing effective early warning and rapid handling of safety risks in the hydrogen supply system, and minimizing the harm caused by safety accidents.

[0034] The optimization module is used to optimize system performance through performance evaluation, algorithm improvement and upgrading, and knowledge accumulation and sharing. The optimization module includes a performance evaluation unit, an improvement and upgrading unit, and an accumulation and sharing unit. The performance evaluation unit collects and visualizes early warning indicators. The improvement and upgrading unit is used to explore new algorithms by using ensemble learning algorithms to fuse long short-term memory network models. The accumulation and sharing unit is used to establish an online knowledge base, use knowledge graph technology to store knowledge in the form of triples, and update and share the knowledge.

[0035] Early warning indicators include early warning accuracy. Fault diagnosis accuracy Among them TP warn To ensure the correct number of warnings, FP warn TP is the number of false alarms. diag To accurately diagnose the number of failures, FP diag To erroneously diagnose the number of faults, an ensemble learning algorithm integrates a long short-term memory network model, including... Where w k The weights are determined through cross-validation.

[0036] Specifically, through performance evaluation, algorithm improvement and upgrades, and knowledge accumulation and sharing, system performance is optimized, and knowledge is accumulated and shared to statistically analyze the accuracy of early warnings. Fault diagnosis accuracy The system includes early warning indicators, which are then visualized using charts (such as line graphs showing the change in early warning accuracy over time, and bar charts comparing the diagnostic accuracy of different fault types). Various monitoring data are presented to operators through an intuitive visual interface, allowing them to easily understand the system's operating status in real time. This enables real-time monitoring of the hydrogen supply system's operation, facilitating timely problem detection and decision-making by operators. The upgraded unit employs ensemble learning algorithms, such as... By integrating a Long Short-Term Memory (LSTM) network model and actively exploring new algorithms, an online knowledge base is established using shared units. Knowledge is stored in triples (e.g., hydrogen leak, cause, pipeline corrosion) using knowledge graph technology. This knowledge is updated based on newly emerging fault cases and solutions, and knowledge sharing is achieved through the system's internal network or a dedicated sharing platform. Through these operations, the system's performance and diagnostic capabilities are continuously improved, enabling continuous optimization and knowledge accumulation of the hydrogen supply safety monitoring system. This allows the system to better adapt to the complex and ever-changing operating environment of the hydrogen supply system, ensuring its long-term stable and safe operation.

[0037] The data acquisition and processing module collects multi-source data from the hydrogen supply system in real time, covering data such as hydrogen concentration, pressure, temperature, flow rate, equipment vibration, rotation speed, and current, as well as environmental data such as combustible gases and flames, and historical data on past faults and leaks. The data cleaning unit removes noise, eliminates outliers, and deduplicates the data, while the data labeling unit labels the data with status, type, location, and severity. Finally, the processed data is stored in a relational cloud database, thereby obtaining accurate, standardized, and orderly processed data, providing reliable data support for subsequent analysis.

[0038] Based on this processed data, the prediction model module selects appropriate time series data processing algorithms from algorithms such as Long Short-Term Memory Network (LSTM), Gated Recurrent Unit (GRU), and Autoregressive Integral Moving Average (AMI) to construct the prediction model, taking into account the time series characteristics, type diversity, trend, and fluctuation range of the data. The model is trained using the mean squared error loss function and stochastic gradient descent algorithm, and the hyperparameters are adjusted using the validation set. The training set is periodically updated with new data, and the hyperparameters are readjusted to construct a model that can accurately learn and optimize its prediction capabilities, thus achieving accurate prediction of potential safety risks in the hydrogen supply system.

[0039] Based on the prediction results of the prediction model and combined with real-time collected multi-source data, the early warning processing module determines the existence of fault risk or leakage risk. The decision generation unit extracts decision suggestions corresponding to different fault types and severity levels from the decision database according to the pre-set fault type and severity classification rules. The alarm processing unit issues early warnings through channels such as audible and visual alarms, SMS notifications, and email reminders, and takes emergency response measures such as closing relevant valves, evacuating surrounding personnel, and activating emergency rescue plans according to the decision suggestions. This enables timely response and rapid handling when safety risks occur, effectively controlling the spread of dangerous situations.

[0040] The optimization module uses the performance evaluation unit to statistically analyze early warning indicators such as early warning accuracy and fault diagnosis accuracy, and displays them visually. The improvement and upgrade unit uses an ensemble learning algorithm to integrate a long short-term memory network model to explore new algorithms. The accumulation and sharing unit establishes an online knowledge base, uses knowledge graph technology to store knowledge in the form of triples, and updates and shares the knowledge, thereby continuously improving the system's performance and diagnostic capabilities, and realizing the continuous optimization and knowledge accumulation of the hydrogen supply safety monitoring system.

[0041] This enables comprehensive real-time monitoring, accurate prediction, and rapid response of the hydrogen supply system, thereby solving the problem that traditional hydrogen supply safety monitoring systems cannot take preventive maintenance measures in advance and are unable to meet the high requirements for safety monitoring in large-scale commercial hydrogen supply scenarios.

[0042] Please see the appendix Figure 2 A safety monitoring method for hydrogen supply includes the following steps: S1. Data Acquisition and Processing: The data acquisition and processing module collects multi-source data from the hydrogen supply system in real time and preprocesses it to obtain processed data. S2. Predictive Model: Based on the processed data, the predictive model module constructs a predictive model by selecting time series data processing algorithms and then trains, validates, and updates it. S3. Early Warning and Handling: The early warning and handling module predicts the safety of the hydrogen supply system based on the prediction model, obtains the prediction results, and takes emergency measures in combination with real-time multi-source data. S4. Optimization: Through the optimization module, performance is evaluated, algorithms are improved and upgraded, and knowledge is accumulated and shared to optimize system performance and accumulate and share knowledge.

[0043] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A safety monitoring system for hydrogen supply, characterized in that, include: Acquisition and processing module: Used to acquire multi-source data from the hydrogen supply system in real time and preprocess it to obtain processed data; Predictive Model Module: Used to build, train, validate, and update predictive models based on processed data by selecting time series data processing algorithms; Early warning and processing module: It is used to predict the safety of the hydrogen supply system based on the prediction model, obtain the prediction results, and take emergency measures in combination with real-time multi-source data. Optimization module: Used to optimize system performance through performance evaluation, algorithm improvement and upgrade, and knowledge accumulation and sharing, and to accumulate and share knowledge.

2. The safety monitoring system for hydrogen supply according to claim 1, characterized in that: The data acquisition and processing module includes a data acquisition unit, a data cleaning unit, a data labeling unit, and a data storage unit. The data acquisition unit is used to acquire multi-source data from the hydrogen supply system, including hydrogen concentration, pressure, temperature, flow rate, equipment vibration, rotational speed, and current data. The data cleaning unit... Noise removal, among which To calculate the average value of the data within the window, arrive To find the range of summation, x j For a data point at position j in the original time series data, outliers are removed and duplicates are deduplicated according to the rule x < μ - 3σ or x > μ + 3σ, where μ is the mean and σ is the standard deviation. The data labeling unit is used to label the data as normal or abnormal, fault or leakage type, location coordinates and severity. The data storage unit is used to select a relational cloud database, design data tables to store different types of data, and establish inter-table relationships to obtain processed data.

3. A safety monitoring system for hydrogen supply according to claim 2, characterized in that: The data acquisition unit includes a combustible gas detection subunit, a flame detection subunit, a hydrogen detection subunit, an equipment detection subunit, a historical data collection subunit, and a data transmission subunit. The combustible gas detection subunit is used to detect combustible gases in the working environment of the hydrogen supply system. The flame detection subunit is used to detect flames in the working environment of the hydrogen supply system, including open flames and smoldering flames. The hydrogen detection subunit is used to detect hydrogen data of the hydrogen supply system, including concentration, pressure, temperature, and flow rate. The equipment detection subunit is used to detect equipment data of the hydrogen supply system, including vibration, rotational speed, and current. The historical data collection subunit is used to collect detailed records of past faults and leaks in the hydrogen supply system, including the time, location, type, cause, handling measures, and consequences of the events. The data transmission subunit is used to transmit all the data.

4. The safety monitoring system for hydrogen supply according to claim 1, characterized in that: The prediction model module includes a model selection unit, a model training unit, and a validation and update unit. The model selection unit is used to select an algorithm for time series data processing based on the characteristics of multi-source data. The model training unit is used to update parameters using the mean squared error loss function and the stochastic gradient descent algorithm to obtain the prediction model. The validation and update unit is used to divide the dataset, adjust the hyperparameters using the validation set, and periodically update the training set with new data and adjust the hyperparameters.

5. A safety monitoring system for hydrogen supply according to claim 4, characterized in that: The characteristics include the time-series properties of the data, the diversity of data types, and the trends and fluctuation ranges of the data. The time-series data processing algorithms include Long Short-Term Memory (LSTM) network algorithms, gated recurrent unit (GUN) algorithms, and autoregressive integral moving average (ARM) models. The mean squared error loss function includes... in For the predicted value, y (i) The true label is N, and the number of samples is N. The stochastic gradient descent algorithm includes... Where θ is a parameter and α is the learning rate.

6. A safety monitoring system for hydrogen supply according to claim 1, characterized in that: The early warning processing module includes a prediction unit, a decision generation unit, and an alarm processing unit. The prediction unit is used to input the processed data collected and preprocessed in real time into the prediction model, and output the prediction result of the safety status of the hydrogen supply system through model calculation. The decision generation unit is used to extract decision suggestions corresponding to different fault types and severity from the decision database based on the prediction results and according to the pre-set fault type and severity classification rules. The alarm processing unit is used to issue early warnings through appropriate channels and take emergency measures according to the decision suggestions.

7. A safety monitoring system for hydrogen supply according to claim 6, characterized in that: The corresponding channels include audible and visual alarms, SMS notifications, and email reminders. The prediction results include determining whether the hydrogen supply system is in a normal state, has a fault risk, or has a leakage risk. The decision recommendations include countermeasures for different fault types and severity. The emergency response includes closing relevant valves, evacuating surrounding personnel, and activating emergency rescue plans.

8. A safety monitoring system for hydrogen supply according to claim 1, characterized in that: The optimization module includes a performance evaluation unit, an improvement and upgrade unit, and an accumulation and sharing unit. The performance evaluation unit statistically analyzes and visualizes early warning indicators. The improvement and upgrade unit is used to explore new algorithms by using ensemble learning algorithms to fuse long short-term memory network models. The accumulation and sharing unit is used to establish an online knowledge base, use knowledge graph technology to store knowledge in the form of triples, and perform knowledge updates and sharing.

9. A safety monitoring system for hydrogen supply according to claim 8, characterized in that: The early warning indicators include early warning accuracy. Fault diagnosis accuracy Among them TP warn To ensure the correct number of warnings, FP warn TP is the number of false alarms. diag To accurately diagnose the number of failures, FP diag To erroneously diagnose the number of faults, the ensemble learning algorithm integrates a long short-term memory network model, including... Where w k The weights are determined through cross-validation.

10. A safety monitoring method for hydrogen supply, characterized in that: A safety monitoring system for hydrogen supply as described in any one of claims 1-9, comprising the following steps: S1. Data Acquisition and Processing: The data acquisition and processing module collects multi-source data from the hydrogen supply system in real time and preprocesses it to obtain processed data. S2. Predictive Model: Based on the processed data, the predictive model module constructs a predictive model by selecting time series data processing algorithms and then trains, validates, and updates it. S3. Early Warning and Handling: The early warning and handling module predicts the safety of the hydrogen supply system based on the prediction model, obtains the prediction results, and takes emergency measures in combination with real-time multi-source data. S4. Optimization: Through the optimization module, performance is evaluated, algorithms are improved and upgraded, and knowledge is accumulated and shared to optimize system performance and accumulate and share knowledge.

Images