A Method and System for Full-Process Data Monitoring and Visualization Management of Air Separation Process

Abstract

The present invention relates to the technical field of data processing, and specifically provides a method and system for full-process data monitoring and visualization management of the air separation process, which collects key parameter data, input data vectors and output data vectors between devices in the air separation process; constructs a parameter correlation model to analyze the input data vectors and output data vectors to obtain data deviations; if the data deviation exceeds the dynamic threshold, triggers a multi-level alarm mechanism and generates a blockchain log record; generates a hash value for the key parameter data and marks the abnormal link; constructs a dual-modal anomaly detection model to analyze the abnormal link to obtain the equipment performance attenuation coefficient and the abnormal influence relationship coefficient between devices, and generates a comprehensive anomaly score through a fusion technology; when the comprehensive anomaly score is greater than the preset threshold, constructs a fault tree model to trace back the abnormal data link in reverse for the abnormal link to locate the fault source; and realizes full-process data monitoring and management through a visualization interface.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and specifically to a method and system for full-process data monitoring and visualization management of an air separation process. Background Art

[0002] In the field of industrial system data governance, the traditional monitoring system of air separation plants faces multiple technical bottlenecks. The existing architecture is limited by the structural contradictions of heterogeneous data sources, specifically manifested as follows: 1) The capture dimension of multi-modal data is missing. Measured data shows that the existing system can only obtain the basic physical quantity parameters of equipment (about 35% of the actual process feature dimensions), and it is impossible to construct a complete digital twin covering deep process parameters such as phase change processes and energy efficiency characteristics; 2) There are defects in the time series alignment of the distributed data storage architecture. When processing asynchronous data streams from sensor networks, equipment controllers, and process simulation systems, the time stamp synchronization error rate of cross-source data is as high as 17%; 3) There are performance bottlenecks in the visualization rendering pipeline. When the traditional graphics interface presents more than 32 process parameters simultaneously, the dynamic refresh rate drops below 12fps, resulting in the loss of important operating condition characteristics; 4) The scalability of the real-time computing framework is insufficient. When the batch processing mechanism based on a relational database deals with more than 150,000 data points written per second, the system throughput drops sharply by 63%; 5) The ability of data correlation analysis is weak. Traditional statistical methods are difficult to analyze the non-linear relationship of the 48-dimensional feature space hidden in the dynamic balance of the fractionating tower. During the peak production load period of holidays, these problems directly lead to a delay in process fault tracing of more than 45 minutes.

[0003] Current technological improvements mainly focus on optimizing the physical layer communication protocol, and there are three key limitations at the data governance level: First, significant information entropy attenuation occurs during the multi-source heterogeneous data fusion process, and the use of traditional ETL tools results in a 28% loss of process feature fidelity; Second, the visualization engine is limited by the two-dimensional plane mapping paradigm and cannot implement the dynamic thermodynamic trajectory projection of process parameters in the three-dimensional phase space; Third, the time series data processing adopts an offline batch calculation mode, and the response delay to millisecond-level process events exceeds 800ms. The essence of these problems lies in the lack of a full-life cycle computing system for high-dimensional industrial data.

[0004] Therefore, a method and system for full-process data monitoring and visualization management of an air separation process are proposed. Summary of the Invention

[0005] The object of the present invention is to provide a method and system for full - process data monitoring and visualization management of an air separation process. By collecting key parameter data in the air separation process, input data vectors and output data vectors between devices; constructing a parameter association model to analyze the input data vectors and output data vectors to obtain data deviations; if the data deviation exceeds the dynamic threshold, triggering a multi - level alarm mechanism and generating a blockchain log record; generating a hash value for the key parameter data and marking abnormal links; constructing a dual - mode anomaly detection model to analyze the abnormal links to obtain the equipment performance attenuation coefficient and the abnormal influence relationship coefficient between devices, and generating a comprehensive anomaly score through a fusion technology; when the comprehensive anomaly score is greater than the preset threshold, constructing a fault tree model to trace back the abnormal data link in reverse for the abnormal link to locate the fault source; and realizing full - process data monitoring and management through a visualization interface.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] A method for full - process data monitoring and visualization management of an air separation process, including:

[0008] S1. Collect key parameter data in the air separation process; deploy sensors through edge - computing nodes, analyze and pre - process the key parameter data to obtain input data vectors and output data vectors between upstream and downstream devices of the air separation process;

[0009] S2. Construct a parameter association model to analyze the input data vectors and output data vectors to obtain the logical relationship of key parameters, and combine with the dynamic threshold to obtain data deviations; if the data deviation exceeds the dynamic threshold, trigger a multi - level alarm mechanism and generate a blockchain log record;

[0010] S3. Generate a hash value for the key parameter data to form an encrypted data packet; after decrypting the data packet, recalculate the hash value and compare it with the original hash value; when they are inconsistent, automatically trigger a data re - transmission request and mark the abnormal link;

[0011] S4. Construct a dual - mode anomaly detection model to analyze the abnormal link to obtain the equipment performance attenuation coefficient and the abnormal influence relationship coefficient between devices, and generate a comprehensive anomaly score through a fusion technology; when the comprehensive anomaly score is greater than the preset threshold, construct a fault tree model to trace back the abnormal data link in reverse for the abnormal link to locate the fault source; and realize full - process data monitoring and management through a visualization interface.

[0012] Preferably, the key parameter data includes pressure, temperature, flow rate, and oxygen content;

[0013] The pre - processing includes data cleaning, data synchronization, and data normalization, and filters noise using a sliding - window mean algorithm;

[0014] The blockchain log record includes storing the abnormal timestamp, device ID, and operator information into the private chain nodes of the steel plant;

[0015] Visual management includes real-time display of device status, data trend analysis, abnormal alarm prompts, and fault tracing results.

[0016] Preferably, the parameter correlation model includes a feature extraction module, a correlation analysis module, a threshold calculation module, and a data consistency verification module;

[0017] The feature extraction module extracts features from the input data vector and output data vector to obtain the key data feature vector; the correlation analysis module analyzes the key data feature vector through linear regression to generate the logical relationship of the key parameters and predict the input data of the downstream device; the threshold calculation module calculates the dynamic threshold through the normal fluctuations and errors of the historical data of the air separation process; the data consistency verification module obtains the deviation by comparing the input data of the downstream device with the actual input data.

[0018] Preferably, the dual-modal anomaly detection model includes a data partitioning module, a time-series modal analysis module, a spatial modal analysis module, and a modal fusion module;

[0019] The data partitioning module analyzes the abnormal link to obtain abnormal time-series data and abnormal spatial topology data; and performs noise removal and normalization processing on the abnormal time-series data and abnormal spatial topology data;

[0020] The time-series modal analysis module analyzes the change trend of device parameters over time by using a long short-term memory network to obtain the long-term operation law of the device and generate the device performance attenuation coefficient;

[0021] The spatial modal analysis module analyzes the propagation path and intensity of the anomaly in the device network through a graph convolutional network to generate the anomaly influence relationship coefficient between devices;

[0022] The modal fusion module generates a comprehensive anomaly score by using a weighted fusion technique to synthesize the time-series and spatial analysis results.

[0023] Preferably, the specific calculation formula for the comprehensive anomaly score is:

[0024] ;

[0025] Where, is the comprehensive anomaly score of device , is the weight of the device performance attenuation coefficient, is the device performance attenuation coefficient, is the weight of the anomaly influence relationship coefficient between devices, For the device For the device Abnormal influence relationship coefficient For the device Set of devices directly connected

[0026] Preferably, the fault tree model includes an event module, a traceability analysis module, a probability calculation module, and a positioning module;

[0027] The event module decomposes the abnormal link event into basic events, intermediate events, and top events; the traceability analysis module starts from the top event and traces back along the logical structure in the reverse direction to identify all possible paths leading to the abnormality; the probability calculation module calculates the occurrence probability of each event in the fault tree to evaluate the possibility of each fault path; the positioning module determines the fault source by comprehensively analyzing the logical relationship, feature matching degree, and probability distribution of the fault path, and generates the positioning result of the fault source.

[0028] An air separation process full-process data monitoring and visualization management system, comprising:

[0029] A data acquisition and processing unit, configured to acquire key parameter data in the air separation process; deploy sensors through edge computing nodes, analyze and preprocess the key parameter data, and obtain the input data vector and output data vector between the upstream and downstream devices of the air separation process;

[0030] A parameter correlation model construction unit, configured to construct a parameter correlation model to analyze the input data vector and output data vector, obtain the logical relationship of the key parameters, and combine with a dynamic threshold to obtain a data deviation; if the data deviation exceeds the dynamic threshold, trigger a multi-level alarm mechanism and generate a blockchain log record;

[0031] An abnormal link acquisition unit, configured to generate a hash value for the key parameter data to form an encrypted data packet; decrypt the data packet and then recalculate the hash value and compare it with the original hash value; when they are inconsistent, automatically trigger a data retransmission request and mark the abnormal link;

[0032] A fault analysis and traceability unit, configured to construct a bimodal anomaly detection model to analyze the abnormal link, obtain the device performance attenuation coefficient and the abnormal influence relationship coefficient between devices, generate a comprehensive anomaly score through a fusion technology; when the comprehensive anomaly score is greater than a preset threshold, construct a fault tree model to trace back the abnormal data link in the reverse direction for the abnormal link, locate the fault source; and realize the monitoring and management of the full-process data through a visualization interface.

[0033] Compared with the prior art, the beneficial effects of the present invention are:

[0034] 1. The present invention deploys sensors through edge computing nodes to achieve comprehensive collection of input and output data between upstream and downstream equipment in the air separation process. Based on this, a parameter correlation model is constructed to analyze the logical relationship of key parameters, and dynamic thresholds are combined to monitor data deviation in real time. If the deviation exceeds the standard, multi-level alarms are triggered and recorded in the blockchain log, significantly improving the comprehensiveness of data collection and the accuracy of anomaly detection. The blockchain log ensures the transparency and immutability of anomaly event records, further enhancing the system reliability.

[0035] 2. The present invention generates hash values and forms encrypted data packets during data transmission to ensure data integrity and security. If there is inconsistency, data retransmission is triggered and the abnormal link is marked, significantly improving transmission reliability. In addition, the abnormal link identification function can quickly locate the problem link, avoid the spread of anomalies, improve system stability and data processing efficiency, and provide stronger data guarantee for the air separation process.

[0036] 3. The present invention uses a dual-modal anomaly detection model, combines time series and spatial analysis, deeply analyzes abnormal links, generates equipment performance decay coefficients and abnormal influence relationship coefficients between equipment, and calculates a comprehensive anomaly score through fusion technology. When the score exceeds the threshold, a fault tree model is automatically constructed for backward tracing to accurately locate the fault source, improving the accuracy of troubleshooting. The visualization interface intuitively presents the full-process operation status, helping operators quickly respond to anomalies. Compared with the prior art, the present invention greatly improves the efficiency of fault management and decision support capabilities. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 It is a schematic flow chart of a full-process data monitoring and visualization management method for the air separation process provided by the present invention;

[0038] Figure 2 It is a schematic structural diagram of a full-process data monitoring and visualization management system for the air separation process provided by the present invention;

[0039] Figure 3 It is a schematic structural diagram of the parameter correlation model provided by the embodiment of the present invention;

[0040] Figure 4 It is a schematic diagram of full-process data monitoring for the air separation process provided by the embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0041] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0042] The air separation process is an industrial process that separates air into important gases such as nitrogen, oxygen, and argon, and is widely used in multiple fields such as chemical industry, metallurgy, and energy. During the operation of an air separation unit, it involves multiple key links and complex process controls, and usually relies on precise data monitoring and management systems to ensure its efficient and safe operation. However, there are still some obvious deficiencies in the existing technology in terms of data flow and monitoring management of air separation units. First of all, the current air separation process data acquisition system mainly relies on scattered sensor devices, and the data acquisition in each link is not comprehensive and accurate enough to reflect the operating status of the entire air separation unit in real time and accurately. Secondly, the data processing and storage methods in each link of the air separation unit are different, lacking a unified standard and intelligent data flow interface, resulting in low data flow efficiency between different links and making it difficult to achieve seamless docking and integration of data. Moreover, most of the existing air separation process data monitoring systems only monitor local links, especially in terms of equipment operating status and fault diagnosis, lacking a global monitoring and visualization management platform for the entire air separation process flow, and it is difficult to support more efficient decision-making analysis and process optimization. In addition, the real-time monitoring and early warning mechanism of traditional systems is also relatively weak, often relying on manual records or regular sampling, with poor real-time performance and unable to quickly respond to abnormal changes in the process. Although some modern data monitoring technologies have begun to be applied, due to the low data update frequency or imperfect feedback mechanism, it is still difficult to achieve rapid early warning of system failures or fine control of the process.

[0043] Embodiment 1

[0044] Based on this, the present invention provides a method and system for full-process data monitoring and visualization management of an air separation process. Please refer to Figures 1 to 2 , and the technical solution is as follows:

[0045] As an implementation manner of the present invention, referring to Figure 1 S1 in, S1 is applied to the data acquisition and processing unit of a full-process data monitoring and visualization management system for an air separation process. The data acquisition and processing unit is used to acquire key parameter data in the air separation process; sensors are deployed through edge computing nodes to analyze and preprocess the key parameter data to obtain input data vectors and output data vectors between upstream and downstream devices in the air separation process.

[0046] S1. Acquire key parameter data in the air separation process; deploy sensors through edge computing nodes to acquire output data and input data between upstream and downstream devices in the air separation process, and perform preprocessing to generate input data vectors and output data vectors between devices;

[0047] Furthermore, the key parameter data includes pressure, temperature, flow rate, and oxygen content; the preprocessing includes data cleaning, data synchronization, and data normalization, and uses the sliding window mean algorithm to filter noise;

[0048] In the air separation process, the acquisition and monitoring of key parameter data are crucial for ensuring the stable operation of each system part. The following is a detailed description of the main system parts and their key parameters:

[0049] Air compression system: Monitor the inlet and outlet pressures of the compressor to ensure that the compression process operates within a safe and efficient range; monitor the inlet and outlet temperatures of the compressor to prevent equipment damage caused by overheating;

[0050] Precooling and purification system: Monitor the temperature of the air in the precooler to ensure effective reduction of the air temperature; monitor the air flow rate to ensure the precooling effect;

[0051] Heat exchange system: Monitor the inlet and outlet temperatures of the heat exchanger to ensure the heat exchange efficiency; monitor the flow rate of the cooling medium to ensure the heat exchange effect;

[0052] Rectification separation system: Monitor the temperatures of each tray of the rectification column to ensure the separation effect; monitor the top and bottom pressures of the rectification column to maintain appropriate separation conditions; monitor the flow rates of the product and waste gas to ensure the product output and purity;

[0053] Expansion refrigeration system: Monitor the inlet and outlet temperatures of the expander to ensure the refrigeration effect; monitor the inlet and outlet pressures of the expander to maintain appropriate expansion conditions;

[0054] Product transportation and storage system: Monitor the transportation pressure of the product gas to ensure the safety of the transportation process; monitor the temperature of the product gas to prevent high temperature from affecting the product quality.

[0055] By real-time monitoring and management of the above key parameters, the stable operation of each system part of the air separation process can be effectively guaranteed. Some key parameter data is shown in Table 1.

[0056] Table 1 is used to record the key parameter data (such as pressure, temperature, flow rate, etc.) of each system part in the air separation process. It is collected through edge computing nodes and sensors to ensure the comprehensiveness and accuracy of data collection, providing a basis for subsequent analysis.

[0057] Table 1 Key Parameter Data Acquisition Table for Air Separation Process

[0058]

[0059] As an implementation manner of the present invention, refer to Figure 1S2 in it is applied to the parameter correlation model construction unit of a full-process data monitoring and visualization management system for air separation process. The parameter correlation model construction unit is used to construct a parameter correlation model to analyze the input data vector and output data vector, obtain the logical relationship of key parameters, and combine with the dynamic threshold to obtain data deviation. If the data deviation exceeds the dynamic threshold, a multi-level alarm mechanism is triggered, and a blockchain log record is generated.

[0060] S2. Construct a parameter correlation model to analyze the input data vector and output data vector, obtain the logical relationship of key parameters, combine with the dynamic threshold to obtain data deviation. If the data deviation exceeds the dynamic threshold, a multi-level alarm mechanism is triggered, and a blockchain log record is generated. The blockchain log record includes storing the abnormal timestamp, device ID, and operator information to the private chain node of the steel plant.

[0061] Further, the parameter correlation model includes a feature extraction module, a correlation analysis module, a threshold calculation module, and a data consistency verification module. Refer to Figure 3 ;

[0062] The feature extraction module extracts features from the input data vector and output data vector to obtain the key data feature vector. The correlation analysis module analyzes the key data feature vector through linear regression to generate the logical relationship of key parameters and predict the input data of downstream devices. The threshold calculation module calculates the dynamic threshold through the normal fluctuations and errors of the historical data of the air separation process. The data consistency verification module obtains the deviation by comparing the input data of the downstream device with the actual input data.

[0063] The parameter correlation model is trained on the historical air separation process data through the random forest algorithm to learn the logical relationship between key parameters. The model parameters are optimized through methods such as cross-validation to improve performance. Through the trained correlation analysis model, the input data of downstream devices can be predicted based on the input data. The specific data is shown in Table 2.

[0064] Table 2 Training data table of parameter correlation model

[0065]

[0066] Table 2 is used to display the historical data samples used during the training of the parameter correlation model, including input features (such as pressure, temperature) and output features (such as flow rate), as well as the label of whether it is abnormal, for revealing the logical relationship of key parameters.

[0067] In this embodiment, a parameter association model is established to analyze the input and output data vectors, revealing the logical relationships between key parameters. Combining with a dynamic threshold, data deviations are identified. When the data deviation exceeds the set threshold, a multi-level alarm mechanism is triggered, and a blockchain log record is generated to ensure the security and traceability of the data.

[0068] As an implementation manner of the present invention, referring to Figure 1 S3 in [reference], S3 is applied to an abnormal link acquisition unit of a full-process data monitoring and visualization management system for air separation process. The abnormal link acquisition unit is used to generate a hash value for the key parameter data to form an encrypted data packet; recalculate the hash value after decrypting the data packet and compare it with the original hash value; when they are inconsistent, automatically trigger a data retransmission request and mark the abnormal link.

[0069] S3. Generate a hash value for the key parameter data to form an encrypted data packet; recalculate the hash value after decrypting the data packet and compare it with the original hash value; when they are inconsistent, automatically trigger a data retransmission request and mark the abnormal link;

[0070] In this embodiment, a hash value is generated for the key parameter data to form an encrypted data packet. After decryption, the hash value is recalculated and compared with the original hash value. When they are inconsistent, a data retransmission request is automatically triggered, and the abnormal link is marked to ensure the accuracy and integrity of data transmission.

[0071] As an implementation manner of the present invention, referring to Figure 1 S4 in [reference], S4 is applied to a fault analysis and traceability unit of a full-process data monitoring and visualization management system for air separation process. The fault analysis and traceability unit is used to construct a dual-modal anomaly detection model to analyze the abnormal link, obtain the equipment performance attenuation coefficient and the abnormal influence relationship coefficient between devices, and generate a comprehensive anomaly score through a fusion technology; when the comprehensive anomaly score is greater than a preset threshold, construct a fault tree model to trace back the abnormal data link of the abnormal link in reverse to locate the fault source; and realize the visualization of the full-process data monitoring and management through a visualization interface, including real-time display of equipment status, data trend analysis, abnormal alarm prompt, and fault traceability result.

[0072] S4. Construct a dual-modal anomaly detection model to analyze the abnormal link, obtain the equipment performance attenuation coefficient and the abnormal influence relationship coefficient between devices, and generate a comprehensive anomaly score through a fusion technology; when the comprehensive anomaly score is greater than a preset threshold, construct a fault tree model to trace back the abnormal data link of the abnormal link in reverse to locate the fault source; and realize the monitoring and management of the full-process data through a visualization interface.

[0073] Furthermore, the dual-modal anomaly detection model includes a data division module, a time-series modal analysis module, a spatial modal analysis module, and a modal fusion module;

[0074] The data partitioning module analyzes the abnormal link to obtain abnormal time-series data and abnormal spatial topology data; and performs noise removal and normalization processing on the abnormal time-series data and abnormal spatial topology data.

[0075] The time-series modal analysis module analyzes the changing trend of device parameters over time by using a long short-term memory network, obtains the long-term operation law of the device, and generates a device performance attenuation coefficient.

[0076] The spatial modal analysis module analyzes the propagation path and intensity of abnormalities in the device network through a graph convolutional network, and generates an abnormal influence relationship coefficient between devices.

[0077] The modal fusion module generates a comprehensive anomaly score by comprehensively analyzing the time-series and spatial analysis results through a weighted fusion technique.

[0078] Construct a dual-modal anomaly detection model to analyze the abnormal link, obtain the device performance attenuation coefficient and the abnormal influence relationship coefficient between devices. Through the fusion technique, generate a comprehensive anomaly score. When the comprehensive anomaly score exceeds the preset threshold, construct a fault tree model, trace back the abnormal data link in reverse, and locate the fault source. Through the visualization interface, realize the monitoring and management of the whole-process data, and improve the intelligent level of the system.

[0079] Furthermore, the specific calculation formula for the comprehensive anomaly score is:

[0080] ;

[0081] Where is the comprehensive anomaly score of device , is the weight of the device performance attenuation coefficient, is the device performance attenuation coefficient, is the weight of the abnormal influence relationship coefficient between devices, is the abnormal influence relationship coefficient of device on device , is the set of devices directly connected to device .

[0082] In this embodiment, the calculation method of the comprehensive anomaly score optimizes the accuracy of the anomaly score by weighted fusion of the device performance attenuation coefficient and the abnormal influence relationship coefficient, making the prediction of device failures more scientific and reliable.

[0083] Furthermore, the fault tree model includes an event module, a traceback analysis module, a probability calculation module, and a positioning module.

[0084] The event module decomposes the abnormal link event into basic events, intermediate events, and top events; the traceability analysis module starts from the top event and traces back along the logical structure in the reverse direction to identify all possible paths leading to the abnormality; the probability calculation module evaluates the possibility of each fault path by calculating the occurrence probability of each event in the fault tree; the positioning module determines the fault source by comprehensively analyzing the logical relationship, feature matching degree, and probability distribution of the fault paths, and generates the positioning result of the fault source.

[0085] The introduction of the fault tree model can effectively trace various events in the abnormal link and evaluate the possibility of the fault path through probability calculation. By comprehensively analyzing the fault path, the fault source can be quickly located, improving the maintenance efficiency and response speed of the system.

[0086] The present invention realizes the data acquisition, analysis, and monitoring of the entire process of the air separation process by combining advanced technologies such as edge computing, blockchain, and deep learning. First, key parameter data in the air separation process are collected in real time through edge computing nodes and sensors, and the data are preprocessed to ensure the accuracy and timeliness of the data. By constructing a parameter correlation model, the relationship between the input and output data of devices can be effectively analyzed, potential data deviations can be discovered in a timely manner, and an alarm mechanism is triggered in combination with dynamic thresholds to effectively ensure the stability of the system operation. Using blockchain technology for data recording ensures the immutability and traceability of the data, enhancing the security and reliability of the system. In addition, the dual-modal anomaly detection model accurately identifies device anomalies and their mutual influences through deep learning technology, comprehensively analyzing time-series and spatial data, effectively improving the accuracy of device fault warning. At the same time, the fault tree model can accurately trace the abnormal link, quickly locate the fault source, and optimize the fault handling process. The visual management interface displays the device status, data trends, and abnormal alarms in real time, enhancing the decision-making support ability of managers and comprehensively improving the monitoring efficiency and operation and maintenance level of the air separation process. These innovative points jointly promote the intelligent management of data in the entire process of the air separation process.

[0087] Embodiment 2

[0088] Please refer to Figures 1 to 2 , the present invention provides a method for monitoring and visual management of the entire process data of an air separation process, which is applied to a system for monitoring and visual management of the entire process data of an air separation process. The technical solution is as follows:

[0089] As an implementation manner of the present invention, referring to Figure 1S1 in the system is applied to the data acquisition and processing unit of a full-process data monitoring and visualization management system for air separation process. The data acquisition and processing unit is used to collect key parameter data in the air separation process. Sensors are deployed through edge computing nodes to analyze and preprocess the key parameter data, and input data vectors and output data vectors between upstream and downstream equipment in the air separation process are obtained.

[0090] S1. Collect key parameter data in the air separation process; deploy sensors through edge computing nodes to analyze and preprocess the key parameter data, and obtain input data vectors and output data vectors between upstream and downstream equipment in the air separation process;

[0091] Furthermore, the key parameter data includes pressure, temperature, flow rate, and oxygen content; the preprocessing includes data cleaning, data synchronization, and data normalization, and the sliding window mean algorithm is used to filter noise;

[0092] As an implementation manner of the present invention, referring to Figure 1 S2 in the system is applied to the parameter correlation model construction unit of a full-process data monitoring and visualization management system for air separation process. The parameter correlation model construction unit is used to construct a parameter correlation model to analyze the input data vectors and output data vectors, obtain the logical relationship of key parameters, and combine with dynamic thresholds to obtain data deviations. If the data deviation exceeds the dynamic threshold, a multi-level alarm mechanism is triggered, and a blockchain log record is generated.

[0093] S2. Construct a parameter correlation model to analyze the input data vectors and output data vectors, obtain the logical relationship of key parameters, combine with dynamic thresholds to obtain data deviations. If the data deviation exceeds the dynamic threshold, a multi-level alarm mechanism is triggered, and a blockchain log record is generated; the blockchain log record includes storing the abnormal timestamp, device ID, and operator information to the private chain node of the steel plant, referring to Table 3;

[0094] Table 3 Abnormal Link Marking Table

[0095]

[0096] Table 3 is used to record the abnormal link information detected during the data transmission process, including abnormal time, device ID, operator, and abnormal type, to ensure data security and traceability.

[0097] Furthermore, the parameter correlation model includes a feature extraction module, an association analysis module, a threshold calculation module, and a data consistency verification module;

[0098] The feature extraction module extracts features from the input data vector and the output data vector to obtain a key data feature vector; the correlation analysis module analyzes the key data feature vector through linear regression to generate the logical relationship of key parameters and predict the input data of downstream devices; the threshold calculation module calculates a dynamic threshold through the normal fluctuations and errors of historical data of the air separation process; the data consistency verification module obtains a deviation by comparing the input data of the downstream device with the actual input data.

[0099] As an implementation manner of the present invention, referring to Figure 1 S3 in, S3 is applied to an abnormal link acquisition unit of an air separation process full-process data monitoring and visualization management system. The abnormal link acquisition unit is used to generate a hash value for the key parameter data to form an encrypted data packet; recalculate the hash value after decrypting the data packet and compare it with the original hash value; when they are inconsistent, automatically trigger a data retransmission request and mark the abnormal link.

[0100] S3. Generate a hash value for the key parameter data to form an encrypted data packet; recalculate the hash value after decrypting the data packet and compare it with the original hash value; when they are inconsistent, automatically trigger a data retransmission request and mark the abnormal link;

[0101] S4. Construct a bimodal anomaly detection model to analyze the abnormal link, obtain the equipment performance attenuation coefficient and the abnormal influence relationship coefficient between devices, and generate a comprehensive anomaly score through a fusion technology; when the comprehensive anomaly score is greater than a preset threshold, construct a fault tree model to trace back the abnormal data link in reverse for the abnormal link to locate the fault source; and realize the monitoring and management of the full-process data through a visualization interface.

[0102] Further, the bimodal anomaly detection model includes a data division module, a time series modal analysis module, a spatial modal analysis module, and a modal fusion module;

[0103] The data division module analyzes the abnormal link to obtain abnormal time series data and abnormal spatial topology data; and performs noise removal and normalization processing on the abnormal time series data and abnormal spatial topology data;

[0104] The time series modal analysis module analyzes the change trend of device parameters over time by using a long short-term memory network to obtain the long-term operation law of the device and generate the equipment performance attenuation coefficient;

[0105] The equipment performance attenuation coefficient is calculated by the time series modal analysis module and is used to measure the attenuation degree of the equipment performance over time. The specific steps and formulas are as follows:

[0106] Data basis: Time series data of key device parameters collected from edge computing nodes (such as historical and real-time values of pressure, temperature, etc.);

[0107] The specific calculation method is as follows:

[0108] Feature extraction: Extract trend features from time-series data, such as the rate of change of parameters over time;

[0109] Model prediction: Use the LSTM model to predict the parameter values under normal operating conditions based on historical data;

[0110] Decay calculation: Compare the actual parameter values with the predicted values and calculate the degree of performance decay;

[0111] Equipment performance decay coefficient The calculation formula is:

[0112] ;

[0113] Where, is the length of the evaluation time window, is the equipment at time actual key parameter value, is the parameter value of the equipment predicted by the LSTM model at time indicating the expected value under normal operating conditions, is the reference parameter value of the equipment under normal operating conditions; represents the average deviation degree of equipment performance. The smaller the value, the milder the performance decay, and the larger the value, the more severe the decay;

[0114] The spatial modal analysis module analyzes the propagation path and intensity of anomalies in the equipment network through a graph convolutional network, and generates an anomaly influence relationship coefficient between equipment;

[0115] The anomaly influence relationship coefficient between equipment is calculated by the spatial modal analysis module and is used to quantify the transfer intensity of anomalies between equipment. The specific steps and formulas are as follows:

[0116] Data basis: The equipment connection relationship graph constructed based on the preprocessed spatial topology data, and the relevant data of the abnormal link.

[0117] The specific calculation method is as follows: Use the equipment in the air separation process as nodes and the input-output relationship between equipment as edges to construct an equipment network diagram; Use the GCN model to learn the abnormal propagation path and intensity between equipment, and calculate the influence relationship through the node embedding vector; The anomaly influence relationship coefficient between equipment is specifically calculated as:

[0118] ;

[0119] Where, is the activation function, is the learnable weight matrix in the GCN, used to adjust the feature importance, concatenates and into a vector to represent the features of the device pair, is the bias term.

[0120] The modal fusion module generates a comprehensive anomaly score by weighted fusion technology, integrating the results of temporal and spatial analysis.

[0121] Furthermore, the specific calculation formula for the comprehensive anomaly score is:

[0122] ;

[0123] where, is the comprehensive anomaly score of device , is the weight of the device performance decay coefficient, is the device performance decay coefficient, is the weight of the abnormal influence relationship coefficient between devices, is the abnormal influence relationship coefficient of device on device , is the set of devices directly connected to device .

[0124] Furthermore, the fault tree model includes an event module, a traceability analysis module, a probability calculation module, and a positioning module;

[0125] The event module decomposes the abnormal link event into basic events, intermediate events, and top events; the top event is the highest-level event of the abnormal data link, such as "abnormal input pressure of the distillation column"; the intermediate event is the failure of the subsystem or component that causes the top event to occur, such as "abnormal compressor"; the basic event is the direct cause of the intermediate event or top event, such as "valve leakage" or "sensor failure".

[0126] The traceability analysis module starts from the top event and traces back reversely along the logical structure to identify all possible paths leading to the anomaly; the event nodes in the fault tree are connected by logical gates (such as "AND gate", "OR gate") to reflect the causal relationship between events. For example, "abnormal input pressure of the distillation column" may be caused by "compressor failure" or "valve leakage" (connected by an "OR gate").

[0127] The traceability analysis module starts from the top event and traces down layer by layer to build a fault tree and identify all possible fault paths. Taking "abnormal input pressure of the distillation column" as an example, the possible fault paths include:

[0128] Path 1: Sensor failure → Data anomaly;

[0129] Path 2: Valve leakage → Pressure fluctuation;

[0130] Path 3: Compressor failure → Insufficient output pressure.

[0131] The probability calculation module evaluates the likelihood of each fault path by calculating the occurrence probabilities of events in the fault tree;

[0132] Probability of basic events: Based on historical fault data, equipment reliability data, and expert evaluation, determine the occurrence probabilities of basic events. For example, the probability of valve leakage , the probability of sensor failure .

[0133] Probabilities of intermediate events and top events: Calculate the probabilities of upper-level events according to the type of logic gate:

[0134] "OR gate": The probability that event or occurs is: ;

[0135] "AND gate": The probability that event and occur simultaneously is: .

[0136] Assume that the top event "Abnormal input pressure of the rectification column" in the fault tree is connected by "Compressor failure" (probability ) and "Valve leakage" (probability ) through an "OR gate", then the occurrence probability of the top event is:

[0137] ;

[0138] Through layer-by-layer calculation, the probability calculation module provides a quantitative likelihood assessment for each fault path.

[0139] The positioning module determines the fault source by comprehensively analyzing the logical relationship, feature matching degree, and probability distribution of the fault path, and generates the positioning result of the fault source.

[0140] Feature matching degree: Use the fault features in the expert knowledge base to match the manifestation form of real-time abnormal data. For example, if the abnormal data shows "high-frequency pressure fluctuation", and the knowledge base records that "valve leakage often shows high-frequency pressure fluctuation", then the matching degree of the "valve leakage" path is relatively high. Probability distribution: Based on the output of the probability calculation module, evaluate the occurrence probabilities of each path, and the example is referred to Table 4;

[0141] Table 4 Fault tree model analysis table

[0142]

[0143] Table 4 is used to record the occurrence probabilities of each basic event in the fault tree model and the likelihood assessment of the fault paths, and is used to trace back the abnormal link in reverse and locate the fault source.

[0144] The present invention aims to improve the monitoring accuracy and management efficiency of the air separation process. First, by deploying edge computing nodes and sensors in the air separation process, key parameter data such as pressure, temperature, flow rate, and oxygen content are collected in real time. The collected data is preprocessed to generate input and output data vectors between devices, providing a reliable data basis for subsequent analysis. Then, a parameter correlation model is constructed to analyze the input and output data vectors, revealing the logical relationships between key parameters. Combining with dynamic thresholds, data deviations are detected in a timely manner, triggering a multi-level alarm mechanism, and using blockchain technology to record abnormal information to ensure the security and traceability of the data. In addition, hash values are generated for the key parameter data to form encrypted data packets. After decryption, the hash values are recalculated and compared with the original hash values to ensure the accuracy and integrity of data transmission. When inconsistencies are found, a data retransmission request is automatically triggered, and the abnormal link is marked to improve the self-healing ability of the system. In terms of anomaly detection, a dual-modal anomaly detection model is constructed to comprehensively analyze time-series and spatial data to accurately identify device anomalies and their mutual influences. Through fusion technology, a comprehensive anomaly score is generated. When the score exceeds a preset threshold, a fault tree model is constructed to trace back the abnormal data link in reverse and quickly locate the fault source, specifically referring to Figure 4 ... Finally, through a visualization interface, the device status, data trends, and abnormal alarms are displayed in real time, enhancing the decision-making support ability of managers and comprehensively improving the monitoring efficiency and operation and maintenance level of the air separation process. In summary, the present invention realizes the intelligent management of the whole process data of the air separation process by integrating advanced technologies such as edge computing, blockchain, and deep learning, and improves the stability, security, and intelligent level of the system.

[0145] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for monitoring and visualizing the whole process data of an air separation process, characterized in that: include: S1. Collect key parameter data in air separation process; By deploying sensors on edge computing nodes, key parameter data are analyzed and preprocessed to obtain input data vectors and output data vectors between upstream and downstream equipment in the air separation process. S2. Construct a parameter association model to analyze the input data vector and the output data vector, obtain the logical relationship of key parameters, and combine the dynamic threshold to obtain the data deviation; if the data deviation exceeds the dynamic threshold, trigger a multi-level alarm mechanism and generate a blockchain log record; S3. Generate a hash value for the key parameter data to form an encrypted data packet; recalculate the hash value after decrypting the data packet and compare it with the original hash value; when there is inconsistency, automatically trigger a data retransmission request and mark the abnormal link; S4. Build a dual-modal anomaly detection model to analyze abnormal links, obtain the device performance attenuation coefficient and the anomaly impact relationship coefficient between devices, and generate a comprehensive anomaly score through fusion technology; The dual-modal anomaly detection model includes a data partitioning module, a temporal modal analysis module, a spatial modal analysis module and a modal fusion module; The data partitioning module obtains abnormal time series data and abnormal spatial topology data by analyzing the abnormal link; and removes noise and normalizes the abnormal time series data and abnormal spatial topology data; The time series modal analysis module uses a long short-term memory network to analyze the change trend of equipment parameters over time, obtains the long-term operation law of the equipment, and generates the equipment performance attenuation coefficient; The spatial modal analysis module analyzes the propagation path and intensity of anomalies in the device network through a graph convolutional network, and generates anomaly impact relationship coefficients between devices; The modality fusion module generates a comprehensive anomaly score by integrating the temporal and spatial analysis results through weighted fusion technology; When the comprehensive abnormality score is greater than the preset threshold, a fault tree model is constructed to trace the abnormal data link in reverse to locate the fault source; and the whole process data is monitored and managed through a visual interface.

2. The method for monitoring and visualizing the whole process data of air separation process according to claim 1, characterized in that: The key parameter data include pressure, temperature, flow rate and oxygen content; The preprocessing includes data cleaning, data synchronization and data normalization, and filtering noise using a sliding window mean algorithm; The blockchain log record includes storing the abnormal timestamp, equipment ID and operator information to the steel plant private chain node; Visual management includes real-time display of equipment status, data trend analysis, abnormal alarm prompts and fault tracing results.

3. The method for monitoring and visualizing the whole process data of an air separation process according to claim 1, characterized in that: The parameter association model includes a feature extraction module, an association analysis module, a threshold calculation module and a data consistency verification module; The feature extraction module obtains a key data feature vector by extracting features from the input data vector and the output data vector; The association analysis module analyzes the key data feature vector through linear regression to generate the logical relationship of key parameters and predict the downstream equipment input data; the threshold calculation module calculates the dynamic threshold through the normal fluctuations and errors of the air separation process historical data; the data consistency verification module obtains the deviation by comparing the downstream equipment input data with the actual input data.

4. The method for monitoring and visualizing the whole process data of air separation process according to claim 1, characterized in that: The specific calculation formula of the comprehensive abnormality score is: in, For equipment The comprehensive abnormality score, is the equipment performance attenuation coefficient weight, is the equipment performance attenuation coefficient, is the weight of the abnormal impact relationship coefficient between devices, For equipment About equipment Abnormal influence relationship coefficient, For equipment A collection of directly connected devices.

5. The method for monitoring and visualizing the whole process data of air separation process according to claim 1, characterized in that: The fault tree model includes an event module, a traceability analysis module, a probability calculation module and a positioning module; The event module decomposes the abnormal link event into basic events, intermediate events and top events; the tracing analysis module starts from the top event and traces back along the logical structure to identify all possible paths leading to the abnormality; The probability calculation module evaluates the possibility of each fault path by calculating the occurrence probability of each event in the fault tree; the positioning module determines the fault source and generates the positioning result of the fault source by comprehensively analyzing the logical relationship, feature matching degree and probability distribution of the fault path.

6. A data monitoring and visualization management system for the whole process of air separation process, characterized in that: include: Data acquisition and processing unit, used to collect key parameter data in air separation process; By deploying sensors on edge computing nodes, key parameter data are analyzed and preprocessed to obtain input data vectors and output data vectors between upstream and downstream equipment in the air separation process. A parameter association model construction unit, used to construct a parameter association model to analyze the input data vector and the output data vector, obtain the logical relationship of key parameters, and obtain the data deviation in combination with the dynamic threshold; if the data deviation exceeds the dynamic threshold, a multi-level alarm mechanism is triggered and a blockchain log record is generated; The abnormal link acquisition unit is used to generate a hash value for the key parameter data to form an encrypted data packet; after decrypting the data packet, the hash value is recalculated and compared with the original hash value; when there is inconsistency, a data retransmission request is automatically triggered and an abnormal link is marked; The fault analysis and tracing unit is used to build a dual-mode anomaly detection model to analyze abnormal links, obtain the equipment performance attenuation coefficient and the anomaly impact relationship coefficient between devices, and generate a comprehensive anomaly score through fusion technology; The dual-modal anomaly detection model includes a data partitioning module, a temporal modal analysis module, a spatial modal analysis module and a modal fusion module; The data partitioning module obtains abnormal time series data and abnormal spatial topology data by analyzing the abnormal link; and removes noise and normalizes the abnormal time series data and abnormal spatial topology data; The time series modal analysis module uses a long short-term memory network to analyze the change trend of equipment parameters over time, obtains the long-term operation law of the equipment, and generates the equipment performance attenuation coefficient; The spatial modal analysis module analyzes the propagation path and intensity of anomalies in the device network through a graph convolutional network, and generates anomaly impact relationship coefficients between devices; The modality fusion module generates a comprehensive anomaly score by integrating the temporal and spatial analysis results through weighted fusion technology; When the comprehensive abnormality score is greater than the preset threshold, a fault tree model is constructed to trace the abnormal data link in reverse to locate the fault source; and the whole process data is monitored and managed through a visual interface.

Images