Iot monitoring platform

Abstract

This invention provides an IoT monitoring platform, relating to the field of IoT technology, comprising: real-time acquisition of fuel cell stack parameters, hydrogen system parameters, and environmental parameters of hydrogen energy equipment via IoT terminals, and processing of data feature points to obtain corresponding feature point data; constructing a multi-dimensional topological directed structure for the hydrogen energy equipment based on the feature point data; obtaining the parameter fluctuation coefficient and parameter output efficiency of the corresponding node based on the feature point data of each node in the multi-dimensional topological directed structure; coupling and correlating the parameter fluctuation coefficient and parameter output efficiency to obtain the actual state risk coefficient of the corresponding node; constructing an intelligent early warning model for equipment status to obtain the state early warning risk coefficient of the corresponding hydrogen energy equipment; comprehensively judging the actual state risk coefficient and state early warning risk coefficient of the same hydrogen energy equipment to obtain the state risk fluctuation coefficient of the corresponding hydrogen energy equipment, and then optimizing the intelligent early warning model for equipment status.

Description

Technical Field

[0001] This invention relates to the field of Internet of Things (IoT) technology, specifically to an IoT monitoring platform. Background Technology

[0002] Existing IoT monitoring technologies applied to hydrogen fuel cell power supply equipment for drones mostly only achieve basic parameter collection and simple alarms, lacking in-depth data processing and feature mining. They cannot establish correlations between equipment operating parameters, making it difficult to accurately analyze equipment energy efficiency and potential risks. Furthermore, existing early warning models are mostly based on fixed thresholds, failing to dynamically optimize based on real-time equipment operating data. This results in insufficient accuracy and specificity of warnings, and difficulty in accurately diagnosing equipment faults. Consequently, they cannot meet the actual needs of efficient, safe, and stable operation of hydrogen fuel cell equipment, exhibiting problems such as incomplete monitoring, superficial analysis, inaccurate early warnings, and a lack of self-optimization capabilities in the models. These issues hinder the improvement of intelligent operation and maintenance levels for hydrogen fuel cell power supply equipment for drones. Therefore, this paper proposes an IoT monitoring platform. Summary of the Invention

[0003] To address the problem of how to accurately analyze the energy efficiency status and potential risks of equipment, the present invention aims to provide an intelligent Internet of Things (IoT) monitoring platform.

[0004] To achieve the above objectives, the present invention provides the following technical solution: an Internet of Things (IoT) monitoring platform, which includes: an IoT access module, a device management module, a device operation and maintenance module, an alarm center module, and a rule engine module; The IoT access module is used to collect real-time parameters of the fuel cell stack, hydrogen system, and environmental parameters of hydrogen energy equipment via IoT terminals. The equipment management module is used to process data feature points of fuel cell stack parameters, hydrogen system parameters, and environmental parameters to obtain feature point data of corresponding data types; and to construct multi-dimensional topological directed structures of hydrogen energy equipment based on the feature point data. The equipment operation and maintenance module obtains the parameter fluctuation coefficient and parameter output efficiency of the corresponding node based on the feature point data of each node in the multidimensional topological directed structure; and couples and correlates the parameter fluctuation coefficient and parameter output efficiency to obtain the actual state risk coefficient of the corresponding node. The alarm center module is used to construct an intelligent early warning model for equipment status; and to obtain the status early warning risk coefficient of the corresponding hydrogen energy equipment based on the intelligent early warning model for equipment status. The rules engine module comprehensively judges the actual state risk coefficient and state warning risk coefficient of several nodes of the same hydrogen energy device to obtain the state risk fluctuation coefficient of the corresponding hydrogen energy device, and optimizes the intelligent early warning model of the device state based on the state risk fluctuation coefficient.

[0005] Furthermore, the process of processing data feature points of the fuel cell stack parameters, hydrogen system parameters, and environmental parameters to obtain feature point data of corresponding data types includes: Outliers for corresponding fuel cell stack parameters are identified based on statistical criteria, and parameter anomalies are determined by the fuel cell stack's rated parameters and the duration of the anomaly. The fuel cell stack parameters at the corresponding acquisition time for each outlier are obtained and correlated with the hydrogen system parameters and environmental parameters at the same acquisition time. Furthermore, the fuel cell stack parameters, hydrogen system parameters, and environmental parameters for several acquisition cycles adjacent to the acquisition time of the outlier are obtained. The peak value, valley value, mean value, and variance of each parameter within these acquisition cycles are calculated and recorded as secondary feature parameters. The fuel cell stack parameters, hydrogen system parameters, and environmental parameters corresponding to the outlier are used as primary feature parameters. The primary and secondary feature parameters are then classified according to data type to form feature point data for the corresponding data type.

[0006] Furthermore, the process of constructing the multidimensional topological directed structure of the hydrogen energy device based on the aforementioned feature point data includes: The multidimensional topological directed structure includes a fuel cell stack topological directed structure and a hydrogen system topological directed structure. The fuel cell stack parameters corresponding to outliers of different hydrogen energy devices are used as the reference nodes of the fuel cell stack topological directed structure, and the fuel cell stack parameters at different acquisition times within several acquisition cycles adjacent to the outlier are used as the topological nodes of the fuel cell stack topological directed structure. The hydrogen system parameters corresponding to the outlier are used as the reference nodes of the hydrogen system topological directed structure, and the hydrogen system parameters at different acquisition times within several acquisition cycles adjacent to the outlier are used as the topological nodes of the hydrogen system topological directed structure. Based on the working logic and operating time sequence of the hydrogen energy devices, directed association relationships between nodes are defined. Node associations are established between the fuel cell stack topological directed structure and the hydrogen system topological directed structure to create parameter interaction links between fuel cell stack parameters and hydrogen system parameters. The data coupling relationships of the feature points corresponding to the parameter interaction links are labeled to form the multidimensional topological directed structure of the hydrogen energy devices.

[0007] Furthermore, the process of obtaining the parameter fluctuation coefficient and parameter output efficiency of the corresponding node based on the feature point data of each node in the multidimensional topological directed structure includes: Based on the feature point data of each node in the multidimensional topological directed structure, the difference of feature point data of the same node at different acquisition times is calculated, and the parameter change rate is obtained by combining the time interval. The change rate is then converted into parameter fluctuation coefficient through normalization. Based on the functions of the components of the hydrogen energy equipment, a parameter output efficiency calculation model is established. Based on the characteristic point data of each node in the multidimensional topological directed structure and the parameter output efficiency calculation model, the corresponding input parameters and output parameters are calculated. The parameter output efficiency of the corresponding node is calculated by the ratio of the output parameters to the input parameters.

[0008] Furthermore, the process of coupling and correlating the parameter fluctuation coefficient and parameter output efficiency to obtain the actual state risk coefficient of the corresponding node includes: A two-dimensional coupling correlation matrix is ​​constructed based on the parameter fluctuation coefficient and parameter output efficiency of the corresponding node. The correlation strength of the coupling elements in the two-dimensional coupling correlation matrix is ​​analyzed based on the Pearson correlation analysis algorithm to obtain the corresponding dynamic weight adjustment factor. A dual-feature coupling algorithm is constructed based on the dynamic weight adjustment factor. The parameter fluctuation coefficient and parameter output efficiency of the corresponding node are mapped nonlinearly based on the dual-feature coupling algorithm to obtain the initial coupling feature value of the corresponding node. Based on the historical fault data and risk status parameters of the corresponding nodes, a risk calibration model is constructed. The initial coupling feature value of the corresponding node is input into the risk calibration model, and a calibration coefficient is generated by combining the node fault frequency and fault severity. The risk calibration model is then corrected to obtain the coupling feature correction value. Through risk status threshold division, the coupling feature correction value is mapped to the actual state risk coefficient of the corresponding node.

[0009] Furthermore, the process of constructing an intelligent early warning model for equipment status and obtaining the corresponding status early warning risk coefficient for hydrogen energy equipment includes: Using historical feature data and historical actual state risk coefficients of each node in a multidimensional topological directed structure as training samples, a training set and a test set are divided. An intelligent early warning model for equipment status is constructed using an LSTM neural network. The input layer is the historical feature data of each node, the hidden layer is used to explore the deep correlation between the historical feature data and the historical actual state risk coefficients and the state risk of hydrogen energy equipment, and the output layer is the state early warning risk coefficient of hydrogen energy equipment. The intelligent early warning model for equipment status is iteratively trained and tested using the training set and the test set. The real-time acquired feature point data of each node and the actual state risk coefficient are input into the corresponding intelligent early warning model of equipment status to calculate the state early warning risk coefficient of the corresponding hydrogen energy equipment.

[0010] Furthermore, the process of comprehensively judging the actual state risk coefficient and state warning risk coefficient of several nodes of the same hydrogen energy equipment to obtain the state risk fluctuation coefficient of the corresponding hydrogen energy equipment includes: Extract the actual state risk coefficients of all nodes of the same hydrogen energy equipment, and combine them with the state warning risk coefficient of the hydrogen energy equipment to calculate the relative deviation rate between the actual state risk coefficient of each node and the state warning risk coefficient of the equipment. A deviation judgment threshold is set; when the relative deviation rate of a node is less than or equal to the deviation judgment threshold, the node is determined to be stable; when the relative deviation rate is greater than the deviation judgment threshold, the node is determined to be fluctuating; the number of fluctuating nodes is counted, and the proportion of fluctuating nodes to the total number of nodes is calculated as the node fluctuation weight; the relative deviation rates of all nodes of the hydrogen energy equipment are weighted and summed to obtain the initial state risk fluctuation coefficient; combined with the overall operating conditions of the hydrogen energy equipment, an operating condition correction coefficient is introduced to adjust the initial state risk fluctuation coefficient, thereby obtaining the state risk fluctuation coefficient of the corresponding hydrogen energy equipment.

[0011] Furthermore, the process of optimizing the intelligent early warning model for equipment status based on the aforementioned status risk fluctuation coefficient includes: Set an optimization trigger threshold for the state risk fluctuation coefficient; when the state risk fluctuation coefficient is greater than or equal to the optimization trigger threshold, initiate the equipment state intelligent early warning model optimization process; extract feature point data, actual state risk coefficient, equipment state early warning risk coefficient, and state risk fluctuation coefficient of each node of the hydrogen energy equipment during the period when the state risk fluctuation coefficient exceeds the threshold, as model optimization samples; based on the model optimization samples, adjust the number of neurons and activation function of the hidden layer of the equipment state intelligent early warning model, and optimize the weight allocation of feature data; replace the original equipment state intelligent early warning model, and establish a model optimization log to record the optimization data for each optimization.

[0012] The present invention also provides a computer-readable storage medium storing a computer program, the computer program being executed by a processor of the Internet of Things monitoring platform.

[0013] Compared with existing technologies, the beneficial effects of this invention are as follows: This IoT monitoring platform, through the collaborative work of multiple modules, achieves full-process control and in-depth analysis of hydrogen energy equipment operation data, effectively solving the shortcomings of existing monitoring technologies. A dedicated data acquisition module comprehensively captures various key parameters of hydrogen energy equipment, ensuring the real-time and comprehensiveness of data acquisition; the data processing module, through feature point processing and multi-dimensional topological directed structure construction, clearly presents the correlation between various parameters, providing a reliable data foundation for subsequent energy efficiency analysis and risk assessment; the energy efficiency analysis module, through multi-feature coupling and correlation, can accurately identify the actual operating status risks of each node of the equipment; the intelligent early warning module, with the early warning model built using an LSTM neural network, improves the accuracy and timeliness of equipment status early warning; the intelligent diagnostic module not only achieves comprehensive judgment of equipment status but also dynamically optimizes the early warning model based on the state risk fluctuation coefficient, realizing closed-loop management of monitoring, analysis, early warning, diagnosis, and optimization, significantly improving the intelligent level of hydrogen energy equipment operation and maintenance, ensuring stable and efficient equipment operation, reducing operation and maintenance costs, and providing technical support for the large-scale application of hydrogen energy equipment. Attached Figure Description

[0014] 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 recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0015] Figure 1 This is a schematic diagram of the modules of an IoT monitoring platform.

[0016] Figure 2 A schematic diagram for constructing a multidimensional topological directed structure.

[0017] Figure 3 A flowchart for mapping the risk coefficient of the actual state.

[0018] Figure 4 This is a complete flowchart of the model. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be described in detail below. Obviously, the described embodiments are merely some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other implementation methods obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0020] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0021] like Figure 1 As shown, the IoT monitoring platform includes: an IoT access module, a device management module, a device operation and maintenance module, an alarm center module, and a rule engine module. The IoT access module is used to collect real-time parameters of the fuel cell stack, hydrogen system, and environment of hydrogen energy equipment via IoT terminals.

[0022] In this embodiment, taking the 602 Institute fuel cell stack as an example, a fuel cell stack parameter acquisition unit is deployed at the fuel cell stack output end; the fuel cell stack parameter acquisition unit includes, but is not limited to, a current acquisition unit, a voltage acquisition unit, etc. A hydrogen system parameter acquisition unit is deployed in the hydrogen pipeline; the hydrogen system parameter acquisition unit includes, but is not limited to, a hydrogen purity acquisition unit, a hydrogen supply pressure acquisition unit, a hydrogen supply flow acquisition unit, etc. An environmental parameter acquisition unit is arranged in the equipment compartment; the environmental parameter acquisition unit includes, but is not limited to, a temperature acquisition unit, a relative humidity acquisition unit, etc. The acquisition period is set to an hourly time granularity; the acquisition time in the acquisition period is in minute time granularity.

[0023] All parameters carry a unified timestamp and, after validity verification, are uploaded to the device management module in the form of a structured data packet.

[0024] The equipment management module is used to process data feature points of fuel cell stack parameters, hydrogen system parameters, and environmental parameters to obtain feature point data of corresponding data types; based on the feature point data, a multi-dimensional topological directed structure of hydrogen energy equipment is constructed.

[0025] In this embodiment, the specific process of processing data feature points of the fuel cell stack parameters, hydrogen system parameters, and environmental parameters to obtain feature point data of corresponding data types includes: Step S10: Data preprocessing; In this embodiment, considering the rated operating parameters of the 602 Institute's fuel cell stack, including but not limited to the stack's rated output current of 80A, rated single-cell voltage of 0.65-0.85V, rated hydrogen supply pressure of the hydrogen system of 0.4-0.6MPa, and equipment compartment ambient temperature of 15-35℃, the raw data collected by the IoT terminal is first preprocessed to remove null values ​​and garbled data caused by sensor malfunctions. Linear interpolation is then used to supplement the missing data. Step S11: Based on Statistical criteria were used to screen outliers in fuel cell stack parameters; It should be noted that in this embodiment, the output current in the fuel cell stack parameters is used as a processing example; the processing of other parameters is similar to the output current screening process.

[0026] In this embodiment, the core parameters of the 602 Institute's fuel cell stack, such as output current and single-cell voltage, are selected as outlier screening targets. Statistical criteria are used to calculate the mean and standard deviation of the fuel cell stack parameters; the range for identifying outliers is determined, i.e. When the collected values ​​of the fuel cell stack parameters at a certain moment exceed the range, it is initially identified as an outlier. For example, the average output current of the fuel cell stack of Institute 602 over the past 24 hours is 78A with a standard deviation of 2A. The outlier identification range is 72A-84A. If the current collected at a certain moment is 85A, it exceeds the range and is initially identified as an outlier.

[0027] Step S12: Combining the rated parameters of the fuel cell stack with the duration of continuous abnormality, the abnormal point of the fuel cell stack parameters is finally determined; It should be noted that outlier screening alone cannot eliminate misjudgments caused by instantaneous fluctuations. It is necessary to combine the rated parameters of the 602 Institute's fuel cell stack and the duration of the anomaly to further determine whether the data is abnormal.

[0028] In this embodiment, an anomaly judgment threshold is set: the rated range of the fuel cell stack output current is 70-90A, and the rated range of the single-chip voltage is 0.65-0.85V. When the initially determined outlier exceeds this rated range, the duration judgment stage is entered. An anomaly duration threshold is set: based on the fuel cell stack operation and maintenance experience of Institute 602, the parameter anomaly duration is set to ≥3 minutes (i.e., 3 sampling times, 1 sampling every 1 minute), which can be finally judged as an abnormal point of the fuel cell stack parameters. If the initial outlier (such as current 85A) exceeds the rated range (70-90A is the normal rated range of fuel cell stack output current, 85A does not exceed it and is not judged as abnormal); if a current of 68A is collected, exceeding the rated lower limit, and lasts for 3 sampling times (i.e., 3 minutes), then it is finally judged as an abnormal output current, and the sampling time corresponding to the anomaly is recorded.

[0029] Step S13: Correlate multi-parameter data and obtain parameters from adjacent periods to ensure data correlation; obtain the fuel cell stack parameters (e.g., current 68A, single-chip voltage 0.62V, etc.) at the acquisition time corresponding to the outlier point in step S12, and simultaneously correlate the hydrogen system parameters (e.g., hydrogen purity 99.95%, hydrogen supply pressure 0.38MPa, hydrogen supply flow rate 22L / min, etc.) and environmental parameters (e.g., equipment compartment temperature 36℃, relative humidity 65%, etc.) at that time to ensure the corresponding correlation of multi-source parameters at the same time, providing complete data support for subsequent feature analysis; Obtaining parameters from adjacent periods: Set the number of adjacent acquisition periods to 2 (i.e., 1 period per hour for the first 2 hours), obtain all fuel cell stack parameters, hydrogen system parameters, and environmental parameters within the two acquisition periods (12:00, 13:00) before the time corresponding to the outlier point (e.g., 14:00-01min), each period contains 60 sets of minute-level acquisition data to ensure the temporal continuity of the data, which is used for subsequent sub-feature parameter calculation.

[0030] Step S14: Obtain the primary and secondary feature parameters; For the parameters of each type obtained in the two adjacent collection cycles (a total of 2×60=120 sets of data) in step S13, the four types of secondary characteristic parameters, namely peak value, valley value, mean value and variance, are calculated based on statistical calculation formulas to quantify the fluctuation pattern of the parameters. An example: taking the output current of the fuel cell stack in two adjacent cycles as an example, the peak value of 60 sets of data is 79A, the valley value is 68A, the mean value is 75A, and the variance is 3.2. This set of values ​​is the secondary characteristic parameter of the fuel cell stack current.

[0031] The stack parameters, hydrogen system parameters, and environmental parameters corresponding to the outlier point are taken as the main feature parameters; the main feature parameters and secondary feature parameters are classified according to data type to form feature point data of corresponding data types.

[0032] It should be noted that the distinction between primary and secondary characteristic parameters is clearly defined. The stack parameters, hydrogen system parameters, and environmental parameters at the time corresponding to the outlier points in step S12 are taken as primary characteristic parameters, reflecting the instantaneous state of the parameters at the time of the anomaly. The peak value, valley value, mean, and variance calculated in step S14 are taken as secondary characteristic parameters, reflecting the fluctuation state of the parameters before and after the time of the anomaly. Classified by data type: Primary and secondary feature parameters are classified into three main types: fuel cell stack, hydrogen system, and environment, to avoid confusion between different types of parameters. Fuel cell stack feature point data includes: primary features (i.e., fuel cell parameters at the time corresponding to outliers) and secondary features (i.e., peak, valley, mean, and variance of fuel cell parameters in adjacent acquisition periods). Hydrogen system feature point data includes: primary features (i.e., hydrogen system parameters at the time corresponding to outliers) and secondary features (i.e., peak, valley, mean, and variance of hydrogen system parameters in adjacent acquisition periods). Environment feature point data includes: primary features (i.e., environmental parameters at the time corresponding to outliers) and secondary features (i.e., peak, valley, mean, and variance of environmental parameters in adjacent acquisition periods). The classified primary and secondary feature parameters are then structured and encapsulated, labeled with their corresponding data type, acquisition time, and correlation relationships, forming complete feature point data. This data is then pushed to the topological directed structure construction unit, providing core data support for the subsequent construction of multidimensional topological directed structures.

[0033] In this embodiment, the specific process of constructing the multidimensional topological directed structure of the hydrogen energy device based on the feature point data includes: It should be noted that the multidimensional topological directed structure includes the electric stack topological directed structure and the hydrogen system topological directed structure. The construction of the multidimensional topological directed structure is based on outliers or data from adjacent acquisition cycles to build the topological relationship between data, clarify the implicit correlation between data, and thus achieve real-time monitoring of the equipment by analyzing the correlation between different data.

[0034] The stack parameters corresponding to outliers of different hydrogen energy devices are used as the reference nodes of the stack topology directed structure, and the stack parameters at different acquisition times within several acquisition cycles adjacent to the outlier are used as the topology nodes of the stack topology directed structure. The hydrogen system parameters corresponding to the outlier are used as the reference nodes of the hydrogen system topology directed structure, and the hydrogen system parameters at different acquisition times within several acquisition cycles adjacent to the outlier are used as the topology nodes of the hydrogen system topology directed structure. According to the working logic and operating time sequence of the hydrogen energy devices, the directed association relationship between nodes is defined. The stack topology directed structure and the hydrogen system topology directed structure are associated with nodes to establish parameter interaction links between stack parameters and hydrogen system parameters. The data coupling relationship of the feature points corresponding to the parameter interaction links is marked to form a multidimensional topology directed structure of the hydrogen energy devices.

[0035] It should be noted that the hydrogen energy equipment is based on the hydrogen fuel cell stack. Hydrogen gas is filtered, depressurized, and stabilized by the hydrogen system before being stably delivered to the anode of the stack. Under the action of a catalyst, it decomposes into hydrogen ions and electrons. Hydrogen ions pass through the proton exchange membrane to reach the cathode, while electrons form a direct current through an external circuit. The hydrogen ions and electrons on the cathode side combine with oxygen in the air to generate water and release a small amount of heat, completing a non-combustion electrochemical reaction that directly converts the chemical energy of hydrogen into electrical energy. The output electrical energy is then supplied to external loads after voltage conversion and stabilization. At the same time, the system maintains the operating temperature, humidity, and cleanliness of the stack through hydrothermal management and environmental control to ensure that the electrochemical reaction proceeds continuously, efficiently, and stably.

[0036] In this embodiment, the 602 Institute's fuel cell stack is used as a specific implementation case to explain in detail the construction steps, node definitions, association logic, and examples of the multidimensional topological directed structure of hydrogen energy equipment. The multidimensional topological directed structure mainly includes the fuel cell stack topological directed structure and the hydrogen system topological directed structure. The two are associated through parameter interaction links to form a complete multidimensional topological system. The core purpose is to clarify the implicit associations between various parameters and provide support for real-time equipment monitoring and anomaly tracing.

[0037] Step S20: Set the reference plotting points for the directed topology of the fuel cell stack and the directed topology of the hydrogen system; It should be noted that the outlier points corresponding to parameter anomalies are used as the reference anchor points to associate parameter data from adjacent acquisition cycles to construct topology nodes; the topology range is limited to the time of the anomaly and the two adjacent acquisition cycles to ensure the continuity and correlation of data time sequence.

[0038] Taking the 602 Institute's fuel cell stack as an example, the baseline anchor point is the main characteristic parameters (such as the main characteristics of the fuel cell stack, the main characteristics of the hydrogen system, and the main characteristics of the environment) corresponding to the abnormal time (such as 14:00-01min mentioned above), which serves as the root node of the topology, i.e., the baseline node; the topology coverage is the abnormal time (14:00-01min) + two adjacent acquisition cycles (12:00 cycle and 13:00 cycle), each cycle contains 60 minutes of acquisition time, and each time corresponds to a complete set of fuel cell stack and hydrogen system parameters; auxiliary attributes are environmental feature point data, which are used as additional attributes of all nodes. They are not used to construct nodes separately, but only to label the environmental state corresponding to the nodes to assist in anomaly analysis. For example, excessively high temperatures may cause abnormal fuel cell stack parameters.

[0039] Step S21: Construct the directed structure of the fuel cell stack topology; It should be noted that the core of the directed structure of the fuel cell stack topology is the temporal correlation and anomaly correlation of the fuel cell stack parameters. The nodes are based on the acquisition time and fuel cell stack parameters, and the directed edges reflect the temporal order and parameter fluctuation transmission relationship. In this embodiment, fuel cell topology nodes are defined, categorized into three types, each carrying feature point data. The node naming convention is: acquisition period - acquisition time - corresponding fuel cell parameter type (e.g., 12:00-01min-fuel cell current). Each node is bound to the primary feature (i.e., instantaneous parameter) or secondary feature (i.e., statistical parameter of the corresponding acquisition period) of the fuel cell at the corresponding acquisition time or acquisition period. For example, a reference node is defined as either 14:00-01min-reference node or 14:00-05min-reference node, bound to corresponding data such as the primary feature: 14:00-01min-fuel cell parameter at the corresponding time or 14:00-05min-fuel cell parameter at the corresponding time, such as output current. 68A, single-chip voltage 0.62V; Adjacent cycle nodes (i.e., topology nodes): Each acquisition cycle corresponds to 1 node, bound to the corresponding data, such as the following feature: associated with the statistical parameters of the fuel cell stack parameters corresponding to 13:00-corresponding acquisition cycle or the statistical parameters of the fuel cell stack parameters corresponding to 12:00-corresponding acquisition cycle; Example: Topology node at 12:00: 12:00-peak value, valley value, mean value and variance of fuel cell stack output current or 12:00-peak value, valley value, mean value and variance of single-chip voltage of fuel cell stack, etc.; Topology node at 13:00: 13:00-peak value, valley value, mean value and variance of fuel cell stack output current or 13:00-peak value, valley value, mean value and variance of single-chip voltage of fuel cell stack, etc.

[0040] In this embodiment, directed edges are defined in the stack topology to reflect the correlation between timing and fluctuations. The direction of the directed edges is chronological, and the correlation type of each edge is labeled. Examples include: Normal fluctuation edges: such as 12:00-01min→12:00-02min→…→12:00-60min→13:00-01min→…→13:00-60min, indicating normal timing fluctuations. The edge attribute is the parameter difference between the corresponding two nodes, such as 12:00-10min-current 76A→12:00-11min-current 78A, difference +2A; Abnormal precursor edges: such as 13:00-01min→…→13 ... From 13:00 to 13:00, the edge attribute is that the current continuously decreases, causing the variance to exceed the variance threshold. For example, 13:00-00min-current 73A→…→13:00-20min-current 71A→…→13:00-40min-current 72A→…→13:00-60min-current 71A, the variance increases from 2.8 to 3.5, which is marked as an abnormal precursor. Abnormal outbreak edge: 13:00-60min→14:00-01min (baseline node), the edge attribute is that the current exceeds the rated range (71A→68A) or the voltage exceeds the rated range (0.69V→0.62V), which is marked as an abnormal outbreak.

[0041] A more complete example: 12:00-01min- (76A / 0.77V) →…→ 12:00-60min (75A / 0.76V) → 13:00-00min (74A / 0.74V) →…→ 13:00-20min (73A / 0.75V) →…→ 13:00-40min (73A / 0.75V) →…→ 13:00-60min (71A / 0.69V) → 14:00-0 1 min (68A / 0.62V, reference node); if the current variance or voltage variance between 13:00-00 min (74A / 0.74V) → … → 13:00-20 min (73A / 0.75V) exceeds the corresponding variance threshold, it is recorded as an abnormal precursor; if the current variance or voltage variance between 13:00-60 min (71A / 0.69V) → 14:00-01 min (68A / 0.62V, reference node) exceeds the corresponding rated range, it is recorded as an abnormal outbreak.

[0042] Step S22: Construct the topological directed structure of the hydrogen system; It should be noted that the construction logic of the directed structure of the hydrogen system topology is consistent with that of the fuel cell stack topology. The nodes are based on the acquisition time and hydrogen system parameters, and the directed edges reflect the temporal order and parameter fluctuation transmission relationship.

[0043] In this embodiment, hydrogen system topology nodes are defined, with the node naming rule as follows: acquisition period - acquisition time - corresponding hydrogen system parameter type (e.g., 12:00-01min - hydrogen supply pressure). Each node is bound to the primary feature (i.e., instantaneous parameter) or secondary feature (i.e., statistical parameter of the corresponding acquisition period) of the hydrogen system parameter at the corresponding time or acquisition period. For example, a baseline node is defined as either 14:00-01min - baseline node or 14:00-05min - baseline node. Corresponding data is bound to the node, such as the primary feature: 14:00-01min - hydrogen system parameter at the corresponding time or 14:00-05min - hydrogen system parameter at the corresponding time, such as hydrogen purity 99.9%. 5%, hydrogen supply pressure 0.38MPa, hydrogen supply flow rate 22L / min; Topology node: one node corresponds to each acquisition cycle, bound to the corresponding data, such as the following feature: associated with the statistical parameters of the hydrogen system parameters corresponding to 13:00-corresponding acquisition cycle or the statistical parameters of the hydrogen system parameters corresponding to 12:00-corresponding acquisition cycle; Example: Topology node for 12:00: 12:00-peak value, valley value, mean and variance of hydrogen supply pressure or 12:00-peak value, valley value, mean and variance of hydrogen purity, etc.; Topology node for 13:00: 13:00-peak value, valley value, mean and variance of hydrogen supply pressure or 13:00-peak value, valley value, mean and variance of hydrogen purity, etc.

[0044] In this embodiment, directed edges are defined in the hydrogen system topology. The direction of the directed edges is chronological, and the association type of the edges is labeled. Examples: Normal fluctuation edges: such as 12:00-01min→12:00-02min→…→12:00-60min→13:00-01min→…→13:00-60min, labeled as normal temporal fluctuation, the edge attribute is the parameter difference between the corresponding two nodes, such as 12:00-10min-hydrogen supply pressure 0.43MPa→12:00-11min-hydrogen supply pressure 0.44MPa, difference +0.01MPa; Abnormal precursor edges: such as 13:00-01min→…→13:00-02 ... From 0 to 60 minutes, the edge attribute indicates that the hydrogen supply pressure continuously decreases, causing the variance of the hydrogen supply pressure to exceed the variance threshold. For example, 13:00-00min -0.41MPa → ... → 13:00-20min -0.43MPa → ... → 13:00-40min -0.43MPa → ... → 13:00-60min -0.45MPa, the variance increases from 0.02 to 0.20, marking an abnormal precursor. Abnormal outbreak edge: 13:00-60min → 14:00-01min (baseline node, 0.38MPa), the edge attribute indicates that the hydrogen supply pressure exceeds the rated range (0.45MPa → 0.38MPa), marking an abnormal outbreak.

[0045] A more complete example: 12:00-01min- (0.43MPa) →…→12:00-60min (0.44MPa) →13:00-00min (0.42MPa) →…→13:00-20min (0.43MPa) →…→13:00-40min (0.44MPa) →…→13:00-60min (0.45MPa) →14:00-0 1 min (0.38 MPa, baseline node); if the variance of hydrogen supply pressure between 13:00-00 min (0.42 MPa) → … → 13:00-20 min (0.43 MPa) exceeds the corresponding variance threshold, it is recorded as an abnormal precursor; if the variance between 13:00-60 min (0.42 MPa) → 14:00-01 min (0.38 MPa, baseline node) exceeds the corresponding rated range, it is recorded as an abnormal outbreak.

[0046] Step S23: Establish the association between the directed topology of the fuel cell stack and the directed topology of the hydrogen system; The nodes in the directed structure of the fuel cell stack topology and the directed structure of the hydrogen system topology at the same acquisition time or the same acquisition period are bound to form associated node pairs, and the associated attributes are labeled. For example, associated node pair 1: 13:00-50min-fuel cell stack reference node (71A / 0.69V) is associated with 13:00-50min-hydrogen system reference node (0.40MPa); associated node pair 2: 14:00-00min-fuel cell stack reference node is associated with 14:00-00min-hydrogen system reference node. Construct parameter interaction links, define directed interaction links between hydrogen system parameters and fuel cell stack parameters, and label the coupling relationships of characteristic point data corresponding to the links. Examples: Interaction link 1: such as the coupling relationship between hydrogen system supply pressure and fuel cell stack output current in the corresponding acquisition period. The coupling coefficient is calculated using the Pearson product moment correlation coefficient algorithm, and the closer the coupling coefficient is to 1, the higher the coupling degree. Analyze the implicit relationships between characteristic point data. Interaction link 2: the coupling relationship between hydrogen system supply flow rate and fuel cell stack single-piece voltage in the corresponding acquisition period: a decrease in hydrogen supply flow rate will cause a synchronous decrease in fuel cell stack single-piece voltage. The corresponding coupling coefficient is calculated to be 0.88 using the Pearson product moment correlation coefficient algorithm, which indicates that the hydrogen supply flow rate has a high impact on the fuel cell stack single-piece voltage. The directed topological structures of the fuel cell stack and the hydrogen system are integrated through associated node pairs and parameter interaction links to form a complete multidimensional directed topological structure.

[0047] Step S24: Topology verification; By comparing the coupling relationship of parameter interaction links in the topology with the actual working pattern of the 602 Institute's fuel cell stack, the accuracy of the coupling coefficient was verified. For example, when the hydrogen supply pressure drops from 0.43MPa to 0.38MPa (a decrease of 0.05MPa), the stack current drops from 76A to 68A (a decrease of 8A), and the calculated coupling coefficient is 0.92, which conforms to the actual pattern of 0.01MPa→1.6A. Anomaly threshold calibration was performed, and the judgment criteria of anomaly precursor nodes in the topology were calibrated based on the 602 Institute's operation and maintenance experience to improve the accuracy of anomaly warning. The verified multidimensional topology directed structure was pushed to the equipment operation and maintenance module and the alarm center module for anomaly tracing and real-time monitoring.

[0048] The equipment operation and maintenance module obtains the parameter fluctuation coefficient and parameter output efficiency of the corresponding node based on the feature point data of each node in the multidimensional topological directed structure; and couples and correlates the parameter fluctuation coefficient and parameter output efficiency to obtain the actual state risk coefficient of the corresponding node.

[0049] In this embodiment, the process of obtaining the parameter fluctuation coefficient and parameter output efficiency of the corresponding node based on the feature point data of each node in the multidimensional topological directed structure includes: Based on the feature point data of each node in the multidimensional topological directed structure, the difference of feature point data of the same node at different acquisition times is calculated, and the parameter change rate is obtained by combining the time interval. The change rate is then converted into parameter fluctuation coefficient through normalization. It should be noted that the parameter fluctuation coefficient is used to quantify the degree of fluctuation of the parameters of each node in a multidimensional topological directed structure. The larger the fluctuation coefficient, the more unstable the operation of the equipment component corresponding to the node is, and the more prone it is to failure.

[0050] Based on the relationships between nodes in a multidimensional topological directed structure and the temporal sequence of directed edges, feature point data of the same associated node at different acquisition times are extracted, such as the output current and single-chip voltage of the reference node in the electric stack topological directed structure; and the hydrogen supply pressure and hydrogen supply flow rate in the hydrogen system topological directed structure, to ensure the temporal continuity of the data.

[0051] In this embodiment, taking the output current in the topology node from 13:00-40min to 13:00-60min as an example, i.e., 13:00-40min-73A→13:00-50min-71A→13:00-60min-71A, the time interval is 10min, and the corresponding sampling values ​​at different times are 73A, 71A, and 71A respectively. Alternatively, taking 13:00-39min-72A→13:00-49min-72A→13:00-59min-73A as an example, the time interval is 10min, and the corresponding sampling values ​​at different times are 72A, 72A, and 73A respectively.

[0052] Based on the extracted time-series output current, the difference in feature point data between two adjacent acquisition times with a time interval of 10 minutes is calculated. This difference is then divided by the time interval between the two acquisition times. In this embodiment, the acquisition times are in the minute range, and the time interval is 10 minutes, to obtain the parameter change rate. The unit is adjusted according to the parameter type, such as the current change rate: A / min. Calculate the current change rate from 13:00-40min-73A to 13:00-50min-71A: (71A-73A) / 10min=-0.2A / min, where the negative sign indicates a current decrease; calculate the current change rate from 13:00-50min-71A to 13:00-60min-71A: (71A-71A) / 10min=0A / min. Calculate the current change rate from 13:00-39min-72A to 13:00-49min-72A: (72A-72A) / 10min=0A / min; the current change rate from 13:00-49min-72A to 13:00-59min-73A: (72A-73A) / 10min=-0.1A / min.

[0053] The min-max normalization method is used to convert the calculated parameter change rate into a parameter fluctuation coefficient between 0 and 1, eliminating the influence of different parameter units and facilitating subsequent coupled calculations. The closer the parameter fluctuation coefficient is to 1, the more drastic the parameter fluctuation and the worse the equipment operation stability; the closer it is to 0, the more stable the parameter operation.

[0054] Based on the rated current range of 70A-90A for the 602 fuel cell stack, the maximum threshold for the current change rate is determined to be 0.5A / min; the current change rate from 13:00-40min to 13:00-50min is -0.2A / min, with a corresponding parameter fluctuation coefficient of |-0.2| / 0.5=0.4; the current change rate from 13:00-50min to 13:00-60min is 0A / min, with a corresponding parameter fluctuation coefficient of 0 / 0.5=0; the current change rate from 13:00-39min-72A to 13:00-49min-72A is 0A / min, with a corresponding parameter fluctuation coefficient of 0 / 0.5=0; the current change rate from 13:00-49min-72A to 13:00-59min-73A is -0.1A / min, with a corresponding parameter fluctuation coefficient of |-0.1| / 0.5=0.2.

[0055] Based on the functions of hydrogen energy equipment components, a parameter output efficiency calculation model is established. Based on the characteristic point data of each node in the multidimensional topological directed structure and the parameter output efficiency calculation model, the corresponding input parameters (not limited to the input power of nodes in the fuel cell stack topological directed structure, the input hydrogen flow rate of nodes in the hydrogen system topological directed structure) and output parameters (not limited to the output power of nodes in the fuel cell stack topological directed structure, the output hydrogen flow rate of nodes in the hydrogen system topological directed structure) are calculated. The parameter output efficiency of the corresponding node is calculated by the ratio of the output parameters to the input parameters.

[0056] It should be noted that the parameter output efficiency is used to quantify the energy conversion efficiency of the corresponding equipment components at each node. The lower the efficiency, the higher the energy consumption and the more degraded the performance of the equipment components.

[0057] Based on the functions of different components of hydrogen energy equipment, parameter output efficiency calculation models for directed topology structures of fuel cell stacks and hydrogen systems are established respectively. The selection criteria for input and output parameters are clarified to ensure that the model calculation results conform to actual operation and maintenance needs.

[0058] It should be noted that the node parameter output efficiency model is: Parameter output efficiency = Input parameters / Output parameters; Based on the topological node feature point data, the input parameters of the corresponding nodes of the directed structure of the fuel cell stack topology are extracted, such as the input power, which is the product of the input current and the input voltage; the input parameters of the corresponding nodes of the directed structure of the hydrogen system fuel cell stack topology, such as the input hydrogen energy = hydrogen supply flow rate × input hydrogen supply pressure × time, other parameters are not further explained.

[0059] Based on the main feature data of the topology nodes, the output parameters of the corresponding nodes of the directed structure of the fuel cell stack topology are calculated. For example, the output power = output current × single-cell voltage × number of single cells in the fuel cell stack. The number of single cells in the fuel cell stack of Institute 602 is calculated as 100 cells. The output parameters of the corresponding nodes of the directed structure of the hydrogen system topology are calculated, such as the output hydrogen energy of the hydrogen system = output hydrogen flow rate × output hydrogen supply pressure × time. Other parameters are not further explained.

[0060] From the number of feature points of each node in the multidimensional topological directed structure, extract the corresponding node parameters and output parameters required for the efficiency model to ensure the accuracy and correlation of the parameters; for associated node pairs.

[0061] For example, in the directed structure of the hydrogen system topology, the input parameters of the reference node from 14:00 to 00:00 are: input hydrogen supply pressure 0.40 MPa, input hydrogen supply flow rate 23 L / min × 1 min = 23 L, and input hydrogen energy = 0.40 MPa × 23 L = 9.2 MPa·L; the output parameters are: output hydrogen supply pressure 0.38 MPa, output hydrogen supply flow rate 22 L / min × 1 min = 22 L, and output hydrogen energy = 0.38 MPa × 22 L = 8.36 MPa·L; the corresponding parameter output efficiency of the reference node is 8.36 MPa·L / 9.2 MPa·L × 100% ≈ 90.87%.

[0062] The extracted input and output parameters are substituted into the corresponding node parameter output efficiency model to calculate the parameter output efficiency of each node. Based on the rated efficiency range of the equipment of Institute 602, the validity of the calculated parameter output efficiency is verified.

[0063] In this embodiment, the process of coupling and correlating the parameter fluctuation coefficient and the parameter output efficiency to obtain the actual state risk coefficient of the corresponding node includes: A two-dimensional coupling correlation matrix is ​​constructed based on the parameter fluctuation coefficient and parameter output efficiency of the corresponding node. The correlation strength of the coupling elements in the two-dimensional coupling correlation matrix is ​​analyzed based on the Pearson correlation analysis algorithm to obtain the corresponding dynamic weight adjustment factor. A dual-feature coupling algorithm is constructed based on the dynamic weight adjustment factor. The parameter fluctuation coefficient and parameter output efficiency of the corresponding node are mapped nonlinearly based on the dual-feature coupling algorithm to obtain the initial coupling feature value of the corresponding node. Based on the historical fault data and risk status parameters of the corresponding nodes, a risk calibration model is constructed. The initial coupling feature value of the corresponding node is input into the risk calibration model, and a calibration coefficient is generated by combining the node fault frequency and fault severity. The risk calibration model is then corrected to obtain the coupling feature correction value. Through risk status threshold division, the coupling feature correction value is mapped to the actual state risk coefficient of the corresponding node.

[0064] In this embodiment, the energy efficiency fluctuation coefficient and parameter output efficiency of the corresponding node are used as coupling elements in the matrix. For example, the reference node of the directed structure of the fuel cell topology from 14:00-00min has an energy efficiency fluctuation coefficient of 0.8 for the current and a normalized parameter output efficiency of 0.585 for the current. The energy efficiency fluctuation coefficient of the voltage is 0.2 and the normalized parameter output efficiency of the voltage is 0.885. The constructed two-dimensional coupling correlation matrix is ​​as follows: .

[0065] In this embodiment, a dynamic weight adjustment factor is obtained based on the Pearson correlation analysis algorithm. Correlation strength analysis is performed on all coupling elements in the two-dimensional coupling correlation matrix. The correlation coefficient r between the energy efficiency fluctuation coefficient and the parameter output efficiency is calculated using the Pearson correlation analysis algorithm, with r ranging from [-1, 1]. The dynamic weight adjustment factor is determined based on the magnitude of the correlation coefficient. The larger the absolute value of the correlation coefficient, the stronger the correlation between the two indicators, and the more the dynamic weight adjustment factor favors the indicator with the higher correlation.

[0066] Based on the operation and maintenance experience of Institute 602, the following dynamic weight adjustment rules are set: When the correlation coefficient r ≥ 0.7, the corresponding dynamic weight adjustment factor is 1.1, indicating a strong positive correlation, meaning that the greater the energy efficiency fluctuation and the worse the parameter output efficiency, the more the dynamic weight adjustment factor tilts towards the energy efficiency fluctuation coefficient, increasing its weight ratio; when 0.3 ≤ |r| < 0.7, the corresponding dynamic weight adjustment factor is 1, indicating a moderate correlation, and the weight adjustment factor maintains a basic balance; when |r| < 0.3, the corresponding dynamic weight adjustment factor is 0.9, indicating a weak correlation, and the dynamic weight adjustment factor tilts towards the parameter output efficiency, focusing on the energy efficiency degradation of the equipment.

[0067] A dual-feature coupling algorithm is constructed based on a dynamic weight adjustment factor. The algorithm is based on a two-dimensional coupling correlation matrix and the dynamic weight adjustment factor. The core of the algorithm is to achieve precise coupling between the energy efficiency fluctuation coefficient and the parameter output efficiency through dynamic weight allocation, so as to avoid the problem that fixed weights cannot adapt to the operating states of different nodes.

[0068] In this embodiment, the formula for the dual-feature coupling algorithm is: Initial coupling feature value = [Energy efficiency fluctuation coefficient × (0.6 × dynamic weight adjustment factor)] + [(1 - normalized value of parameter output efficiency) × (0.4 ÷ dynamic weight adjustment factor)], where 0.6 and 0.4 are the basic weights, and the dynamic weight adjustment factor dynamically corrects the basic weights to ensure that the coupling result matches the actual operating state of the node.

[0069] For example, at the 14:00-00min fuel cell reference node, the dynamic weight adjustment factor is 1.1. Substituting into the formula, we get the corresponding initial coupling characteristic value = [0.8×(0.6×1.1)]+[(1-0.585)×(0.4÷1.1)]≈(0.8×0.66)+(0.415×0.364)≈0.528+0.151≈0.679.

[0070] Based on historical fault data and risk status parameters of hydrogen energy equipment at Institute 602, a risk calibration model is constructed. The model input is the initial coupling characteristic value of the node, and the output is the calibration coefficient. The initial coupling characteristic value of the corresponding node is input into the risk calibration model. Combined with the node's historical fault frequency (e.g., the number of faults in the past 6 months) and fault severity (e.g., minor fault, general fault, severe fault), values ​​of 1, 3, and 5 are assigned respectively to generate corresponding calibration coefficients. These coefficients correct the initial coupling characteristic value, resulting in a corrected coupling characteristic value, thus improving the accuracy of risk assessment. The calibration formula is: Corrected coupling characteristic value = Initial coupling characteristic value × (1 + Calibration coefficient), where the calibration coefficient is set according to the fault situation: no historical faults, calibration coefficient is 0; minor faults, calibration coefficient is 0.1; general faults, calibration coefficient is 0.2; severe faults, calibration coefficient is 0.3. For example, the baseline node at 14:00-00min: this node has had one general fault in the past 6 months, with a severity value of 3, generating a calibration coefficient of 0.2. For example, the corrected coupling characteristic value = 0.679 × (1 + 0.2) ≈ 0.815. Risk state thresholds are defined, and actual state risk coefficients are obtained by mapping. The coupling feature correction value is positively correlated with the risk coefficient, thus clarifying the risk classification standard. Among them, the actual state risk coefficient is [0,0.2), which represents no risk; [0.2,0.4), which represents low risk; [0.4,0.6), which represents medium risk; [0.6,0.8), which represents slightly high risk; and [0.8,1], which represents extremely high risk.

[0071] The alarm center module is used to construct an intelligent early warning model for equipment status; based on the intelligent early warning model for equipment status, the corresponding status early warning risk coefficient of the hydrogen energy equipment is obtained.

[0072] In this embodiment, the process of constructing an intelligent early warning model for equipment status and obtaining the corresponding status early warning risk coefficient for hydrogen energy equipment includes: Using historical feature data and historical actual state risk coefficients of each node in a multidimensional topological directed structure as training samples, a training set and a test set are divided. An intelligent early warning model for equipment status is constructed using an LSTM neural network. The input layer is the historical feature data of each node, the hidden layer is used to explore the deep correlation between the historical feature data and the historical actual state risk coefficients and the state risk of hydrogen energy equipment, and the output layer is the state early warning risk coefficient of hydrogen energy equipment. The intelligent early warning model for equipment status is iteratively trained and tested using the training set and the test set. The real-time acquired feature point data of each node and the actual state risk coefficient are input into the corresponding intelligent early warning model of equipment status to calculate the state early warning risk coefficient of the corresponding hydrogen energy equipment.

[0073] In this embodiment, the historical feature point data and historical actual state risk coefficients of each node in the multidimensional topological directed structure are used as the core training samples, as follows: Multidimensional topological node data of hydrogen energy equipment from Institute 602 over the past 12 months were extracted, covering all scenarios including normal operation, abnormal precursors, and abnormal outbreaks, to ensure the comprehensiveness and representativeness of the samples. It should be noted that each sample includes historical feature point data of the node and the risk coefficient of the node's historical actual state. The samples were divided into training set and test set in an 8:2 ratio. The training set was used for model parameter training, and the test set was used for model performance verification to avoid overfitting.

[0074] In this embodiment, an intelligent early warning model for equipment status is constructed using an LSTM neural network. Leveraging its expertise in processing time-series data, the model uncovers the deep correlation between node features and equipment risks. The model structure is designed to meet the equipment monitoring needs of Institute 602, as detailed below: The model's input layer receives historical feature data from each node in a multidimensional topological directed structure, including but not limited to time-series sequences of core parameters such as stack output current, single-chip voltage, hydrogen supply pressure, and hydrogen flow rate. The input dimension matches the number of node feature parameters. The model's hidden layer consists of three LSTM hidden layers with 64 neurons per layer. The ReLU function is used to mine the deep correlation between historical feature data, historical actual state risk coefficients, and the state risk of hydrogen energy equipment, capturing the intrinsic link between parameter time-series fluctuations and risk evolution. A Dropout layer with a dropout rate of 0.2 is also added to prevent overfitting. The model's output layer outputs the state warning risk coefficient of the hydrogen energy equipment, with a value range of 0-10, consistent with the actual state risk coefficient (0 for no risk, 10 for extremely high risk), facilitating comprehensive judgment by the subsequent rule engine module.

[0075] In this embodiment, the intelligent early warning model for equipment status is iteratively trained and tested using training and testing sets to ensure that the model's accuracy meets the operational and maintenance requirements of Institute 602. The specific process is as follows: The model was trained iteratively using mean squared error (MSE) as the loss function and the Adam optimizer (learning rate set to 0.001) for 100 iterations. The accuracy of the training set was calculated after each training iteration. Training was stopped when the accuracy reached 90% or higher and the loss function tended to stabilize. The test set data is input into the trained model, and the deviation between the model's predicted state warning risk coefficient and the actual risk coefficient is calculated. When the deviation rate is ≤5%, the model is deemed qualified. If the deviation rate is >5%, the number of hidden layer neurons or the activation function is adjusted, and the training is iterated again until the model is qualified. The qualified intelligent equipment status warning model is then embedded into the alarm center module for real-time warning calculation.

[0076] The real-time acquired feature point data of each node (real-time collected parameters of the fuel cell stack and hydrogen system in the multi-dimensional topological directed structure) and the actual state risk coefficient are synchronously input into the solidified equipment state intelligent early warning model. The model outputs the corresponding state early warning risk coefficient of the hydrogen energy equipment through the calculation of the nonlinear mapping of the LSTM neural network. ,in, Indicates the activation function; This represents the output layer weight matrix; Indicates the output layer bias term; This represents the core computational function of a neural network, used to uncover the deep correlation between time-series features and risks; This represents the real-time input of node feature point data; This represents the risk coefficient of the actual state of the node as input in real time. Represents the input layer weight matrix; Indicates the input layer bias term; Represents the hidden layer weight matrix; Represents the hidden layer bias term; state warning risk coefficient. This is used to provide real-time early warnings of equipment status. For example, if the actual risk coefficient is 8.2, after inputting it into the model, the calculated risk coefficient for the status warning is 8.3, which deviates from the actual risk coefficient by 0.1, meeting the deviation threshold requirement.

[0077] The rules engine module comprehensively judges the actual state risk coefficient and state warning risk coefficient of several nodes of the same hydrogen energy device to obtain the state risk fluctuation coefficient of the corresponding hydrogen energy device, and optimizes the intelligent early warning model of the device state based on the state risk fluctuation coefficient.

[0078] In this embodiment, the process of comprehensively judging the actual state risk coefficient and state warning risk coefficient of several nodes of the same hydrogen energy device to obtain the state risk fluctuation coefficient of the corresponding hydrogen energy device includes: Extract the actual state risk coefficients of all nodes of the same hydrogen energy equipment, and combine them with the state warning risk coefficient of the hydrogen energy equipment to calculate the relative deviation rate between the actual state risk coefficient of each node and the state warning risk coefficient of the equipment. In this embodiment, the actual state risk coefficients of all nodes in the directed topology of the 602 reactor stack and hydrogen system are extracted. Combined with the state warning risk coefficient of the hydrogen energy equipment, the relative deviation rate of each node is calculated by the following formula to quantify the degree of deviation between the actual risk and the warning risk of the node, where the relative deviation rate = |actual state risk coefficient of the corresponding node - state warning risk coefficient of the corresponding node| / state warning risk coefficient of the corresponding node.

[0079] A deviation judgment threshold is set; when the relative deviation rate of a node is less than or equal to the deviation judgment threshold, the node is determined to be stable; when the relative deviation rate is greater than the deviation judgment threshold, the node is determined to be fluctuating; the number of fluctuating nodes is counted, and the proportion of fluctuating nodes to the total number of nodes is calculated as the node fluctuation weight; the relative deviation rates of all nodes of the hydrogen energy equipment are weighted and summed to obtain the initial state risk fluctuation coefficient; combined with the overall operating conditions of the hydrogen energy equipment, an operating condition correction coefficient is introduced to adjust the initial state risk fluctuation coefficient, thereby obtaining the state risk fluctuation coefficient of the corresponding hydrogen energy equipment.

[0080] In this embodiment, based on the operation and maintenance experience of Institute 602, a deviation judgment threshold of 10% is set, which can be dynamically adjusted according to the aging degree of the equipment. The status of each node is judged based on the relative deviation rate: when the relative deviation rate of a node is ≤10%, the node is judged to be stable, indicating that the actual risk of the node matches the warning risk, and the prediction accuracy of the warning model for the node meets the standard; when the relative deviation rate of a node is >10%, the node is judged to be fluctuating, indicating that the actual risk of the node deviates significantly from the warning risk, and the prediction accuracy of the warning model for the node is insufficient, requiring close attention.

[0081] The number of nodes exhibiting fluctuations within the same hydrogen energy equipment is counted. The proportion of fluctuating nodes to the total number of nodes is calculated as the node fluctuation weight, with a value ranging from 0 to 1. A higher weight indicates more significant overall node fluctuations in the equipment: Node fluctuation weight = Number of fluctuating nodes / Total number of nodes in the equipment. For example, if a hydrogen energy equipment has 12 total nodes, and 4 of them exhibit fluctuations, the node fluctuation weight would be approximately 4 / 12 ≈ 0.33.

[0082] The initial state risk fluctuation coefficient is calculated by weighting and summing the relative deviation rates of all nodes of the hydrogen energy equipment, with the weights being the node fluctuation weights. The initial state risk fluctuation coefficient is calculated as follows: Initial state risk fluctuation coefficient = Σ (relative deviation rate of a single node × node fluctuation weight); For example, if the relative deviation rates of four fluctuating nodes of a certain hydrogen energy equipment are 52.5%, 48.2%, 35.7%, and 29.6% respectively, and the node fluctuation weight is 0.33, the initial state risk fluctuation coefficient = (52.5% + 48.2% + 35.7% + 29.6%) × 0.33 ≈ 51.1%.

[0083] Taking into account the load rate of the 602 Institute's fuel cell stack, the stability of the hydrogen supply system, and the ambient temperature, an operating condition correction coefficient is introduced to adjust the initial state risk fluctuation coefficient, ensuring that the coefficient closely matches the actual operating state of the equipment. The rules for setting the operating condition correction coefficient are as follows: 1. Under normal operating conditions, i.e., load rate is [50%, 80%]: operating condition correction factor = 1.0; 2. High load conditions for the equipment, i.e., load rate > 80%: operating condition correction factor = 1.1; 3. Low load condition of equipment, i.e., load rate < 50%: operating condition correction factor = 0.9.

[0084] The final state risk fluctuation coefficient = initial state risk fluctuation coefficient × operating condition correction coefficient, with a value range of 0-1. The higher the coefficient, the more severe the equipment state fluctuation, and the more the accuracy of the early warning model needs to be optimized.

[0085] In this embodiment, the process of optimizing the intelligent early warning model of equipment status based on the state risk fluctuation coefficient includes: Set an optimization trigger threshold for the state risk fluctuation coefficient; when the state risk fluctuation coefficient is greater than or equal to the optimization trigger threshold, initiate the equipment state intelligent early warning model optimization process; extract feature point data, actual state risk coefficient, equipment state early warning risk coefficient, and state risk fluctuation coefficient of each node of the hydrogen energy equipment during the period when the state risk fluctuation coefficient exceeds the threshold, as model optimization samples; based on the model optimization samples, adjust the number of neurons and activation function of the hidden layer of the equipment state intelligent early warning model, and optimize the weight allocation of feature data; replace the original equipment state intelligent early warning model, and establish a model optimization log to record the optimization data for each optimization.

[0086] In this embodiment, based on the operational experience of Institute 602, the optimization trigger threshold for the state risk fluctuation coefficient is set to 50%, which can be dynamically adjusted, to determine the equipment state risk fluctuation coefficient: When the state risk fluctuation coefficient is ≥50%, it indicates that the equipment state fluctuates drastically and the prediction accuracy of the early warning model is insufficient. The equipment state intelligent early warning model optimization process should be initiated. When the state risk fluctuation coefficient is less than 50%, it indicates that the equipment state fluctuation is mild and the prediction accuracy of the early warning model meets the standard. Therefore, the optimization process is not initiated, and only the current state risk fluctuation coefficient is recorded for reference in subsequent model iterations.

[0087] Extract relevant data from each node of the hydrogen energy equipment during the period when the state risk fluctuation coefficient exceeds the threshold, and use it as a sample set for model optimization. The sample set should include characteristic point data of each node, actual state risk coefficient, state warning risk coefficient, state risk fluctuation coefficient, relative deviation rate, and number of fluctuating nodes.

[0088] Based on the optimized sample set, the core parameters of the intelligent early warning model for equipment status were adjusted, with a focus on optimizing the weight allocation of feature data to improve the model's prediction accuracy for fluctuating nodes. Specific adjustments included: adjusting the number of neurons in the hidden layer: based on the complexity of the optimized sample, the number of neurons in the original 3-layer hidden layer was increased from 64 to 80, enhancing the model's ability to fit complex time-series data; adjusting the activation function: replacing the ReLU activation function in the hidden layer with the LeakyReLU function to solve the gradient vanishing problem during model training and improve the model's ability to mine fluctuating node data; optimizing the weight allocation of feature data: increasing the weight ratio of core feature parameters corresponding to fluctuating nodes (such as fuel cell current and hydrogen supply pressure in the hydrogen system) in the model input layer (e.g., from 0.2 to 0.3), and reducing the weight of non-fluctuating feature parameters, so that the model focuses on the core parameters of fluctuating nodes; adjusting the model training parameters: adjusting the learning rate from 0.001 to 0.0008 and increasing the number of iterations to 120 epochs to improve the stability and accuracy of the optimized model. Detailed records are kept of complete data for each optimization, including optimization trigger time, state risk fluctuation coefficient, optimization sample set information, parameter adjustment content, post-optimization model accuracy, optimization person in charge, etc., forming a model optimization archive for subsequent model iteration, fault tracing and operation and maintenance experience summary, in line with the equipment operation and maintenance standardization requirements of Institute 602.

[0089] Optionally, in this embodiment, those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing the hardware related to the terminal device. The program can be stored in a computer-readable storage medium, which may include: flash drive, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0090] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0091] If the integrated units in the above embodiments are implemented as software functional units and sold or used as independent products, they can be stored in the aforementioned computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause one or more electronic devices to execute all or part of the steps of the methods described in the various embodiments of this application.

[0092] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0093] In the several embodiments provided in this application, it should be understood that the disclosed application can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection of units or modules may be electrical or other forms.

[0094] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0095] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0096] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. An Internet of Things (IoT) monitoring platform, characterized in that: The IoT monitoring platform includes: an IoT access module, a device management module, a device operation and maintenance module, an alarm center module, and a rule engine module; The IoT access module is used to collect real-time parameters of the fuel cell stack, hydrogen system, and environmental parameters of hydrogen energy equipment via IoT terminals. The equipment management module is used to process data feature points of fuel cell stack parameters, hydrogen system parameters, and environmental parameters to obtain feature point data of corresponding data types; and to construct multi-dimensional topological directed structures of hydrogen energy equipment based on the feature point data. The equipment operation and maintenance module obtains the parameter fluctuation coefficient and parameter output efficiency of the corresponding node based on the feature point data of each node in the multidimensional topological directed structure; and couples and correlates the parameter fluctuation coefficient and parameter output efficiency to obtain the actual state risk coefficient of the corresponding node. The alarm center module is used to construct an intelligent early warning model for equipment status; and to obtain the status early warning risk coefficient of the corresponding hydrogen energy equipment based on the intelligent early warning model for equipment status. The rules engine module comprehensively judges the actual state risk coefficient and state warning risk coefficient of several nodes of the same hydrogen energy device to obtain the state risk fluctuation coefficient of the corresponding hydrogen energy device, and optimizes the intelligent early warning model of the device state based on the state risk fluctuation coefficient.

2. The IoT monitoring platform according to claim 1, characterized in that, The process of processing data feature points of fuel cell stack parameters, hydrogen system parameters, and environmental parameters to obtain feature point data of corresponding data types includes: Outliers for corresponding fuel cell stack parameters are identified based on statistical criteria, and parameter anomalies are determined by the fuel cell stack's rated parameters and the duration of the anomaly. The fuel cell stack parameters at the corresponding acquisition time for each outlier are obtained and correlated with the hydrogen system parameters and environmental parameters at the same acquisition time. Furthermore, the fuel cell stack parameters, hydrogen system parameters, and environmental parameters for several acquisition cycles adjacent to the acquisition time of the outlier are obtained. The peak value, valley value, mean value, and variance of each parameter within these acquisition cycles are calculated and recorded as secondary feature parameters. The fuel cell stack parameters, hydrogen system parameters, and environmental parameters corresponding to the outlier are used as primary feature parameters. The primary and secondary feature parameters are then classified according to data type to form feature point data for the corresponding data type.

3. The IoT monitoring platform according to claim 2, characterized in that, The process of constructing the multidimensional topological directed structure of hydrogen energy devices based on the aforementioned feature point data includes: The multidimensional topological directed structure includes the electric stack topological directed structure and the hydrogen system topological directed structure; The stack parameters corresponding to outliers of different hydrogen energy devices are used as the reference nodes of the stack topology directed structure, and the stack parameters at different acquisition times within several acquisition cycles adjacent to the outlier are used as the topology nodes of the stack topology directed structure. The hydrogen system parameters corresponding to the outlier are used as the reference nodes of the hydrogen system topology directed structure, and the hydrogen system parameters at different acquisition times within several acquisition cycles adjacent to the outlier are used as the topology nodes of the hydrogen system topology directed structure. According to the working logic and operating time sequence of the hydrogen energy devices, the directed association relationship between nodes is defined. The stack topology directed structure and the hydrogen system topology directed structure are associated with nodes to establish parameter interaction links between stack parameters and hydrogen system parameters. The data coupling relationship of the feature points corresponding to the parameter interaction links is marked to form a multidimensional topology directed structure of the hydrogen energy devices.

4. The IoT monitoring platform according to claim 3, characterized in that, The process of obtaining the parameter fluctuation coefficient and parameter output efficiency of the corresponding node based on the feature point data of each node in the multidimensional topological directed structure includes: Based on the feature point data of each node in the multidimensional topological directed structure, the difference of feature point data of the same node at different acquisition times is calculated, and the parameter change rate is obtained by combining the time interval. The change rate is then converted into parameter fluctuation coefficient through normalization. Based on the functions of the components of the hydrogen energy equipment, a parameter output efficiency calculation model is established. Based on the characteristic point data of each node in the multidimensional topological directed structure and the parameter output efficiency calculation model, the corresponding input parameters and output parameters are calculated. The parameter output efficiency of the corresponding node is calculated by the ratio of the output parameters to the input parameters.

5. The IoT monitoring platform according to claim 4, characterized in that, The process of coupling and correlating parameter fluctuation coefficients and parameter output efficiency to obtain the actual state risk coefficient of the corresponding node includes: A two-dimensional coupling correlation matrix is ​​constructed based on the parameter fluctuation coefficient and parameter output efficiency of the corresponding node. The correlation strength of the coupling elements in the two-dimensional coupling correlation matrix is ​​analyzed based on the Pearson correlation analysis algorithm to obtain the corresponding dynamic weight adjustment factor. A dual-feature coupling algorithm is constructed based on the dynamic weight adjustment factor. The parameter fluctuation coefficient and parameter output efficiency of the corresponding node are mapped nonlinearly based on the dual-feature coupling algorithm to obtain the initial coupling feature value of the corresponding node. Based on the historical fault data and risk status parameters of the corresponding nodes, a risk calibration model is constructed. The initial coupling feature value of the corresponding node is input into the risk calibration model, and a calibration coefficient is generated by combining the node fault frequency and fault severity. The risk calibration model is then corrected to obtain the coupling feature correction value. Through risk status threshold division, the coupling feature correction value is mapped to the actual state risk coefficient of the corresponding node.

6. The IoT monitoring platform according to claim 5, characterized in that, The process of constructing an intelligent early warning model for equipment status and obtaining the corresponding status early warning risk coefficient for hydrogen energy equipment includes: Using historical feature data and historical actual state risk coefficients of each node in a multidimensional topological directed structure as training samples, a training set and a test set are divided. An intelligent early warning model for equipment status is constructed using an LSTM neural network. The input layer is the historical feature data of each node, the hidden layer is used to explore the deep correlation between the historical feature data and the historical actual state risk coefficients and the state risk of hydrogen energy equipment, and the output layer is the state early warning risk coefficient of hydrogen energy equipment. The intelligent early warning model for equipment status is iteratively trained and tested using the training set and the test set. The real-time acquired feature point data of each node and the actual state risk coefficient are input into the corresponding intelligent early warning model of equipment status to calculate the state early warning risk coefficient of the corresponding hydrogen energy equipment.

7. The IoT monitoring platform according to claim 6, characterized in that, The process of comprehensively judging the actual state risk coefficient and state warning risk coefficient of several nodes of the same hydrogen energy equipment to obtain the state risk fluctuation coefficient of the corresponding hydrogen energy equipment includes: Extract the actual state risk coefficients of all nodes of the same hydrogen energy equipment, and combine them with the state warning risk coefficient of the hydrogen energy equipment to calculate the relative deviation rate between the actual state risk coefficient of each node and the state warning risk coefficient of the equipment. A deviation judgment threshold is set; when the relative deviation rate of a node is less than or equal to the deviation judgment threshold, the node is determined to be stable; when the relative deviation rate is greater than the deviation judgment threshold, the node is determined to be fluctuating; the number of fluctuating nodes is counted, and the proportion of fluctuating nodes to the total number of nodes is calculated as the node fluctuation weight; the relative deviation rates of all nodes of the hydrogen energy equipment are weighted and summed to obtain the initial state risk fluctuation coefficient; combined with the overall operating conditions of the hydrogen energy equipment, an operating condition correction coefficient is introduced to adjust the initial state risk fluctuation coefficient, thereby obtaining the state risk fluctuation coefficient of the corresponding hydrogen energy equipment.

8. The IoT monitoring platform according to claim 7, characterized in that, The process of optimizing the intelligent early warning model for equipment status based on the aforementioned status risk fluctuation coefficient includes: Set an optimization trigger threshold for the state risk fluctuation coefficient; when the state risk fluctuation coefficient is greater than or equal to the optimization trigger threshold, initiate the equipment state intelligent early warning model optimization process; extract feature point data, actual state risk coefficient, equipment state early warning risk coefficient, and state risk fluctuation coefficient of each node of the hydrogen energy equipment during the period when the state risk fluctuation coefficient exceeds the threshold, as model optimization samples; based on the model optimization samples, adjust the number of neurons and activation function of the hidden layer of the equipment state intelligent early warning model, and optimize the weight allocation of feature data; replace the original equipment state intelligent early warning model, and establish a model optimization log to record the optimization data for each optimization.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that is executed by a processor to implement the Internet of Things monitoring platform according to any one of claims 1-8.

Images