A method for monitoring and visualizing management of full flow data of air separation process

Abstract

The present application belongs to the technical field of industrial process monitoring and data visualization management, and particularly relates to a kind of air separation process whole process data monitoring and visualization management method, comprising: collecting each process section monitoring signal is obtained by pre-processing uniform time series dataset;Segment calibration obtains segmented state anchor point set and operation envelope library;According to the inter-segment coupling relationship, the time lag calibration is carried out in the load band to obtain the process propagation time lag matrix;The upstream disturbance is deduced to the expected response and compared with the measured response to calculate the consistency, to obtain the abnormal identification result;Combined with real-time margin and change rate, the touch edge time is calculated to obtain the hierarchical hidden state risk window;The above results are superimposed to the process topology map to generate the verification priority jointly weighted, to obtain the hierarchical visualization management interface.The present application can realize the active identification of cross-section disturbance propagation, the effective distinction of flexible fluctuation and internal anomaly, and the early perception and linkage visualization output of hidden state risk.

Description

Technical Field

[0001] This invention belongs to the field of industrial process monitoring and data visualization management technology, specifically relating to a method for full-process data monitoring and visualization management of air separation process. Background Technology

[0002] In recent years, with the rapid increase in the proportion of new energy power generation and the deepening of power market reform, the operating environment of liquid air separation units has undergone fundamental changes. Frequent grid peak-shaving commands, time-of-use pricing mechanisms, and random fluctuations in wind and solar power output have jointly forced air separation units to repeatedly switch between multiple load levels in a short period of time. Unlike the traditional steady-state operation mode, the coupling relationship between the various stages of air compression, pre-cooling and purification, main heat exchange, distillation separation, and liquid storage and delivery is more complex in frequent load change scenarios. The same disturbance signal exhibits dispersed characteristics with different arrival times and response amplitudes in different stages. Coupled with factors such as equipment performance degradation and actuator response lag, the whole process monitoring faces three challenges: cross-stage time mismatch, difficulty in identifying the nature of the status, and inability to detect potential risks in advance.

[0003] Currently, the industrial sector commonly uses DCS or SCADA systems as data aggregation platforms. These systems integrate temperature, pressure, flow, level, and analyzer signals, presenting the plant status to operators through real-time process displays, historical trend curves, and over-limit alarm lists. When parameters exceed limits, personnel use their experience to assess and handle the situation. While this type of solution is relatively mature in terms of data aggregation and routine display, the various processing stages operate independently, and a holistic management framework that integrates the propagation patterns of disturbances between work sections with the overall risk evolution trend of the entire plant has not yet been established.

[0004] However, the above-mentioned solutions have significant limitations in scenarios with frequent load changes: On the one hand, the existing system is based on single-point numerical display and fixed threshold triggering, lacking the ability to perform cross-section correlation analysis. It cannot quantify the time difference and degree of deviation of upstream disturbances affecting downstream sections, easily misjudging normal process elastic responses as abnormalities, and may also mask true internal degradation signals with large fluctuations. On the other hand, existing visualization methods can only reflect the over-limit status of each point at the current moment, and cannot integrate the disturbance origin, propagation path, expected over-limit time, and operation response sequence into a coherent management framework. This results in potential risks of key units such as cold boxes often being passively perceived only after actual over-limits, missing the opportunity for preventive intervention. To address this, this invention proposes a full-process data monitoring and visualization management method for liquid air separation units with frequent load changes. Summary of the Invention

[0005] In view of the aforementioned existing problems, the present invention is proposed.

[0006] This invention provides a method for full-process data monitoring and visualization management of air separation processes. The aim is to address the shortcomings of existing monitoring systems in frequent load-changing liquid air separation unit operation scenarios. These systems lack the ability to identify cross-process disturbance propagation, making it impossible to jointly quantify the arrival time lag and deviation degree of upstream load disturbances in each downstream process section. This results in difficulty distinguishing between normal flexible fluctuations and internal anomalies. Furthermore, existing visualization methods remain at the level of static process diagrams and fixed threshold alarms, failing to integrate disturbance sources, propagation paths, future limit exceedance time windows, and job inspection sequences into a coherent full-process management view. Consequently, latent risks in key units such as cold boxes are often only passively perceived after actual limit exceedances, missing opportunities for preventative intervention.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0008] A method for full-process data monitoring and visualization management of air separation technology, comprising:

[0009] The monitoring signals of each process segment of the entire air separation process are collected and preprocessed to obtain a unified time series dataset; the unified time series dataset is segmented and calibrated according to each process segment to obtain a segmented state anchor point set and a running envelope library.

[0010] Based on the segmented state anchor point set and the running envelope library, the load band execution time delay calibration is performed according to the inter-segment coupling relationship to obtain the process propagation time delay matrix; using the process propagation time delay matrix as the path basis, the upstream disturbance is extrapolated to the expected response of each downstream process segment, compared with the measured response and the consistency is calculated to obtain the anomaly identification result;

[0011] Using the margin parameters of each process segment in the aforementioned operating envelope library as boundaries, and combining real-time margins and change rates to calculate the edge contact time, a hierarchical latent risk window is obtained. The anomaly identification results, the process propagation time delay matrix, and the hierarchical latent risk window are superimposed on the process topology diagram, and a verification priority order is generated through joint weighting to obtain a hierarchical visual management interface.

[0012] As a preferred implementation, the specific steps for acquiring and preprocessing the monitoring signals are as follows: each signal is connected to the DCS via a fieldbus and uniformly collected and written into the real-time database; the sampling period is set according to the signal type; the preprocessing eliminates time deviation based on the control system master station clock; if the short-term missing value does not exceed the preset number of periods, the previous value is kept and filled; if it exceeds the preset number of periods, the reliability is downgraded.

[0013] As a preferred implementation, the specific steps for segmented calibration according to each process segment are as follows: based on the correspondence between the instrument tag number and the process segment, each monitoring signal is assigned to the corresponding process segment to obtain a subset of historical time-series signals; steady-state operation segments are extracted using the signal change rate and duration as steady-state judgment conditions; steady-state anchor points are obtained by weighting and fusing the design operating parameters and the steady-state mean within a preset number of historical days according to preset weights; the load band is divided into low-load, medium-load, and high-load operating intervals; the upper and lower boundaries of normal offset are determined using preset percentiles to obtain the operating envelope; when the sample is insufficient, it is supplemented by linear interpolation of adjacent load bands.

[0014] As a preferred embodiment, the specific steps of performing time delay calibration based on inter-segment coupling relationship in load bands are as follows: identify effective coupling topology based on whether there is actual medium flow or energy transfer; extract disturbance response samples using the topology as path constraints, identify the upstream disturbance start time and downstream response arrival time to obtain inter-segment propagation time delay, and the disturbance start time determination must meet the dual conditions of offset amplitude and duration; statistically analyze the time delay value and propagation weight of effective samples in the same load band, and summarize them to form a process propagation time delay matrix.

[0015] As a preferred embodiment, the specific steps of summarizing to form the process propagation time delay matrix further include: during the load band switching transition, using the relative position of the current load value between the center values ​​of adjacent load bands as the interpolation coefficient, linearly interpolating the calibration time delay value and the propagation weight respectively; only when a certain process segment stably shows a propagation response in no less than a preset load band number threshold, it is solidified and written into the process propagation time delay matrix.

[0016] As a preferred implementation, the specific steps for extrapolating the upstream disturbance to the desired response and calculating the consistency degree are as follows: using the calibration time delay value as the time offset and the propagation weight as the amplitude attenuation basis, the measured offset of the upstream disturbance is extrapolated downstream step by step along the effective coupling topology. When multiple propagation edges converge, the weighted sum is used to obtain the combined desired response offset value. When crossing intermediate process sections, the combined value is used as the starting offset value of the next level to continue the extrapolation. The measured response offset value is calculated using the steady-state anchor point of the corresponding load band as the zero bias reference. After obtaining the two, the consistency degree is calculated according to the formula for the same source disturbance.

[0017] As a preferred implementation, the criteria for determining the anomaly identification results are as follows: when the consistency of the same source disturbance is not lower than the first threshold, it is determined to be a flexible operation same source fluctuation; when it is between the second threshold and the first threshold, it is determined to be a mixed state to be observed; when it is lower than the second threshold and no less than a preset monitoring quantity threshold number of monitoring quantities in the same process section deviates from the same direction of the operating envelope boundary within a continuous sampling period of no less than a preset period threshold number of sampling periods, it is determined to be an internal anomaly with priority.

[0018] As a preferred implementation, the specific steps for calculating the contact time by combining real-time margin and rate of change are as follows: the boundaries of each margin parameter are determined based on the corresponding load band operating envelope limit offset and each load operating interval is independently valued; the original difference is obtained by subtracting the margin parameter boundary from the current collected value. When the original difference is positive, it is taken as the real-time margin; when it is zero or negative, it is marked as the risk boundary has been reached; when the rate of change is negative, the expected contact time is obtained by dividing the real-time margin by the absolute value of the rate of change.

[0019] As a preferred implementation, the specific steps for the graded labeling of the expected contact time are as follows: the rate of change is continuously updated in a sliding window manner, and the expected contact time is recalculated for each sampling period; when the absolute value of the rate of change is less than the preset lower limit for preventing zero, the lower limit is used instead for calculation; based on the expected contact time, it is divided into four levels: red urgent window, orange warning window, yellow attention window, and normal monitoring status, which are summarized to form a graded hidden risk window.

[0020] As a preferred implementation, the specific steps for generating a hierarchical visual management interface by jointly weighting the verification priority order are as follows: a process topology map is constructed with each process segment as a node and the effective propagation edge as a directed connection; anomaly identification results are marked with node color codes, and hierarchical latent risk windows are marked with overlaid warning rings; in the case of crossover, those with higher risk levels are given priority; the verification priority order score is calculated by jointly weighting three normalized indicators: risk urgency, propagation coverage, and data credibility; the score is sorted and output to the job task card; and the suggested verification operation is automatically generated based on the pre-registered verification rule table, using the process segment name and anomaly type combination as the search key.

[0021] Beneficial effects

[0022] 1. This invention constructs a process propagation time delay matrix to explicitly quantify the arrival time delay and expected response of upstream load disturbances in each downstream process section, and compares and judges the measured response with the expected response based on the consistency of the same source disturbance. This helps to distinguish between flexible operation fluctuations and internal process anomalies under frequent load changes, and to a certain extent improves the problem of false alarms of normal fluctuations and missed alarms of real anomalies caused by the lack of cross-section disturbance propagation identification capabilities in the prior art.

[0023] 2. This invention calculates the estimated contact time by combining the remaining margin of each process section with the rate of change of the margin. Compared with the single-parameter fixed threshold triggering alarm method in the prior art, this invention helps to present key safety margins such as cold box temperature difference margin and tower pressure difference margin in a graded and quantified manner before they exceed the limit. It can maintain a relatively continuous perception of hidden risks under frequent load changes and provide operators with a certain time window for preventive intervention.

[0024] 3. This invention overlays the anomaly identification results, process propagation time delay matrix, and hierarchical hidden risk windows onto the process topology diagram, and generates a job inspection priority order by weighting risk urgency, propagation coverage, and data credibility. This helps to transform scattered local alarm information into a full-process management view that covers disturbance sources, propagation paths, risk windows, and job task cards, enabling operators to obtain causal basis for handling within a single interface. Attached Figure Description

[0025] Figure 1 This is an exemplary flowchart of the air separation process full-process data monitoring and visualization management method provided in the embodiments of the present invention.

[0026] Figure 2 This diagram compares the effectiveness of the air separation process full-process data monitoring and visualization management method provided in this embodiment of the invention with existing technologies. Detailed Implementation

[0027] To make the technical means, creative features, and achieved objectives and effects of this invention easier to understand, the invention is further described below with reference to specific embodiments. However, the following embodiments are merely preferred embodiments of this invention and not all of them. Other embodiments obtained by those skilled in the art based on the embodiments described herein without creative effort are all within the protection scope of this invention. Unless otherwise specified, the experimental methods in the following embodiments are conventional methods, and the materials and reagents used in the following embodiments are commercially available unless otherwise specified.

[0028] Example 1 combined Figure 1 The following is an exemplary flowchart of a method for monitoring and visualizing data throughout the entire air separation process. The specific implementation steps are as follows:

[0029] The monitoring signals of each process segment of the entire air separation process are collected and preprocessed to obtain a unified time series dataset.

[0030] By collecting and preprocessing the monitoring signals of each process segment in the entire air separation process, a data input basis with time consistency is provided for subsequent parameter calibration, disturbance inference and risk calculation of each process segment;

[0031] The entire air separation process includes an air compression section, a purification section, a main heat exchange section, a distillation cold box section, and a liquid storage and delivery section. The air compression section pressurizes the feed air to the required process pressure; the purification section removes moisture and carbon dioxide from the air; the main heat exchange section facilitates heat exchange between the feed air and product air; the distillation cold box section separates the air into components such as oxygen and nitrogen under low-temperature conditions; and the liquid storage and delivery section manages the storage and delivery of liquid products. The monitoring signals cover temperature, pressure, flow rate, liquid level, and component analysis signals for each of the above process sections, as well as key execution parameters. The system includes status signals of the actuator and external load command signals. Temperature signals are measured locally by thermocouples or resistance temperature detectors, pressure signals are sampled locally by pressure transmitters, flow signals are measured online by orifice plates or vortex flow meters, liquid level signals are continuously detected by differential pressure level gauges or radar level gauges, component analysis signals are periodically sampled and analyzed by online analyzers, actuator status signals are reported in real time by valve position feedback devices, and external load command signals are sent and accessed by the dispatch system. After each of the above signals is measured by field instruments, it is connected to the Distributed Control System (DCS) via fieldbus. The DCS collects the data and writes it into the real-time database, forming a raw signal storage layer that can be read by subsequent steps. Considering the differences in the rate of change of different signals, the sampling period is set differently according to the signal type. The sampling period for fast-changing signals is preferably 1 to 5 seconds, and the sampling period for slow-changing signals is preferably 30 to 60 seconds, so as to ensure that the time resolution of various signals meets the needs of subsequent disturbance propagation calculations.

[0032] After signal acquisition, the monitoring signals undergo further preprocessing. This preprocessing prioritizes time alignment and uses the control system master station clock as a unified reference to eliminate time deviations introduced by communication delays at each acquisition node. Furthermore, for short-term missing data during acquisition, if consecutive missing data does not exceed three similar sampling periods, the previous value is retained to fill the gaps. If the missing data exceeds this range, the reliability of the signal is downgraded to avoid interference from low-quality signals in subsequent calculations. After the above acquisition and preprocessing, the monitoring signals from each process segment are uniformly aligned on the time axis, forming the unified time-series dataset. This unified time-series dataset refers to the entire process multi-source time-series signal set, after time alignment and data quality processing, uniformly aligned on the time axis. It includes continuous historical and real-time records of all temperature, pressure, flow rate, liquid level, and component analysis signals from all process segments of the air separation process, serving as a unified data source for subsequent segmental calibration, disturbance inference, and risk calculation for each process segment.

[0033] The unified time series dataset is segmented and labeled according to each process segment to obtain a segmented state anchor point set and a running envelope library.

[0034] Based on the unified time series dataset, the normal and abnormal offset ranges of each process segment under different load conditions are quantitatively distinguished by segmented aggregation and load band parameter calibration, providing benchmark parameters for subsequent disturbance inference and risk boundary calculation.

[0035] The segmented calibration includes segmented aggregation and parameter calibration;

[0036] The segmented aggregation is based on the correspondence table between the instrument tag number of each monitoring signal and the process segment to which it belongs. Each monitoring signal in the unified time series data is assigned to its corresponding process segment, resulting in an independent subset of historical time series signals for each process segment.

[0037] The aforementioned historical time series signal subset refers to the independent time series signal set corresponding to each process segment after the monitoring signals in the unified time series dataset are classified and organized according to the correspondence table between instrument tag number and process segment. It includes the historical records of all temperature, pressure, flow rate, liquid level and component analysis signals of the process segment, and is used for subsequent independent execution parameter calibration of each process segment.

[0038] After obtaining a subset of historical time-series signals, parameter calibration is performed. This calibration includes two steps: steady-state anchor point extraction and load band operation envelope division. For steady-state anchor point extraction, the steady-state judgment condition is that the signal change rate between adjacent sampling points is lower than a preset threshold and the duration is not less than a set window. For example, the steady-state judgment condition is that the signal change rate between adjacent sampling points is lower than 0.1% of the full scale of the signal per sampling cycle and the duration is not less than 20 minutes. Steady-state operation segments that meet the judgment conditions are extracted from the historical time-series signal subset. Using the design operating condition parameters as the theoretical benchmark value and the average of the steady-state operation segments from the most recent 15-30 days as the field measured benchmark value, a weighted fusion is performed with a weight of 0.3 for the design operating condition parameters and 0.7 for the field measured benchmark value to obtain the steady-state anchor point for each process section under the corresponding load level. The reason for selecting the above weight ratio is that after long-term operation, the actual conditions such as equipment performance and pipeline resistance have deviated from the ideal design state. The field measured benchmark has a better representational ability of the current operating state than the design operating condition parameters. The design operating parameters still serve to suppress short-term data fluctuations and prevent anchor point drift, therefore they are assigned a low but not zero weight. The steady-state anchor point refers to the typical stable operating parameter benchmark value of each process section under a specific load level, obtained by weighting and fusing the design operating parameters as the theoretical benchmark and the average of the most recent 15-30 day steady-state operation segments as the field measurement benchmark. This value is used to characterize the normal operation reference center of each process section under the corresponding load level. For example, the steady-state anchor point of the main cooling liquid oxygen level of the distillation cold box section under the 70%-100% high load operation range is the weighted result of the design value and the recent steady-state average value. This value serves as the normal operation reference center of the liquid level parameter of the distillation cold box section under the high load operation range. Subsequent measured values ​​and expected responses are calculated based on this as the zero bias benchmark. The design operating parameters are the rated operating parameter values ​​of each process section determined by process calculations and loaded into the design documents during the unit design stage. They reflect the theoretical operating benchmark of the unit under ideal conditions and are used to obtain the steady-state anchor point by weighted fusion with the field measurement benchmark.Regarding the division of the load band operating envelope, the load band is divided into a 20%–40% low-load operating range, a 40%–70% medium-load operating range, and a 70%–100% high-load operating range. These three segments are determined based on the process characteristics of the liquid air separation unit. The 40% load level is near the lower limit for maintaining normal liquid-to-gas ratio balance in the distillation column system; below this level, the hydrodynamic characteristics within the column change significantly. The 70% load level is near the transition boundary for the air compressor unit from partial load operation to high-efficiency operation; above this level, the coupling response characteristics of each process segment tend to be consistent. The 5th percentile is taken from all valid historical samples within each load band interval as the statistical population. Using the lower bound of normal offset and the 95th percentile as the upper bound of normal offset, the operating envelopes of each process section under low load, medium load, and high load operating ranges are obtained. Low load operation represents the load reduction condition when the power grid is shaving off or downstream gas consumption is low, used for independent calibration of the normal offset range of parameters for each process section under low load conditions. The medium load operating range represents the daily transitional operating condition of the unit, used for independent calibration of the normal offset range of parameters for each process section under medium load conditions. The high load operating range represents the normal production condition of the corresponding unit at full load or near full load, used for independent calibration of the normal offset range of parameters for each process section under high load conditions. When a certain load... When the number of valid samples within an interval is less than 30, linear interpolation is performed using the upper and lower boundaries of the operating envelope of adjacent load zone intervals as endpoints, based on the load percentage. Specifically, the interpolation ratio is used to calculate the interpolation results for the upper and lower boundaries of the operating envelope, based on the relative position of the target load zone's center load value between the center load values ​​of two adjacent calibrated load zones, thus obtaining the complete operating envelope for each load zone. After the above segmented calibration, the steady-state anchor points of each process segment are summarized to form the segmented state anchor point set, and the corresponding operating envelopes of each load zone are summarized to form the operating envelope library for subsequent steps. The segmented state anchor point set is generated by each process segment under the corresponding load level using design operating parameters and recent steady-state operating samples. The weighted fusion yields a summary of steady-state anchor points, which serve as a reference center for calculating the zero-bias benchmark of the measured response offset values ​​of each subsequent process segment and for comparing and determining the consistency of similar disturbances. Subsequent steps involving the use of steady-state anchor points are all read from this segmented state anchor point set. The operating envelope library only stores the upper and lower bounds of the operating envelope for each load band and does not include the steady-state anchor points themselves. The operating envelope library is formed by summarizing the operating envelopes independently calibrated for each process segment under low, medium, and high load operating ranges. It provides a normal offset range benchmark for load band adaptation for determining the margin parameter boundaries, calculating real-time margins, and determining the risk touch-edge time for each subsequent process segment.

[0039] Based on the segmented state anchor point set and the running envelope library, the process propagation time delay matrix is ​​obtained by performing load band execution time delay calibration according to the inter-segment coupling relationship.

[0040] Based on the segmented state anchor point set and the running envelope library, by identifying the physical coupling relationship between each process segment and calibrating the load band time delay of historical disturbance response samples, the propagation time and propagation weight of upstream disturbances to each downstream process segment are explicitly quantified to obtain the process propagation time delay matrix, providing a path basis for the expected response deduction.

[0041] The premise of time delay calibration is to clarify the effective coupling relationship between each process segment. To this end, the connection relationship between each process segment in the process flow diagram is retrieved. The existence of actual medium flow or energy transfer is used as the criterion. Only process segment pairs with direct pipe connection, heat exchange coupling or gas-liquid phase change transfer relationship are established. No propagation edge is established for process segment pairs without the above physical coupling relationship, so as to avoid introducing false associations without actual process basis. The effective coupling topology between each process segment is obtained. The effective coupling topology refers to the directed topology graph formed by each process segment as node and the inter-segment connection with actual medium flow or energy transfer as edge after the inter-segment coupling relationship is identified. It is used to constrain the subsequent time delay calibration to be performed only between process segment pairs with physical basis.

[0042] The inter-segment coupling relationship refers to the actual medium flow or energy transfer relationship between each process segment in the entire air separation process due to direct pipeline connection, heat exchange coupling, or gas-liquid phase change transmission. The physical connection between each process segment in the process flow diagram is used as the identification basis. Effective propagation edges are established only between process segment pairs with the above physical coupling relationship to constrain the calibration range of the process propagation time delay matrix and ensure that each propagation edge in the matrix has actual process basis rather than statistical correlation inference.

[0043] Using the effective coupled topology as a path constraint, time-delay calibration is performed on the load band for each coupled process section. Specifically, three types of disturbance response samples are extracted from the historical database: historical step disturbance samples, load switching samples, and start-up / shutdown samples. This ensures that the time-delay calibration results cover the disturbance propagation characteristics of the device under different operating conditions. Among them, historical step disturbance samples refer to historical time-series segments in the historical operation records where the parameters of the upstream process section change abruptly due to sudden changes in external load commands or rapid action of key valves. Load switching samples refer to historical time-series segments where the parameters of each process section change continuously during the switching process between low load, medium load, and high load operating ranges. Start-up / shutdown samples refer to historical time-series segments where the parameters of each process section transition from the initial state to the steady state during the start-up or shutdown process of the device.

[0044] For various types of samples, the moment when a significant change occurs in the upstream process section monitoring signal is identified as the disturbance initiation moment. The determination of a significant change requires simultaneously meeting the following two conditions: First, the signal offset amplitude exceeds the operating envelope boundary of the corresponding load band in the operating envelope library; second, this offset state is continuous for at least 10 times the sampling period of the signal in the time dimension and the offset direction is consistent, excluding situations where the signal briefly exceeds the envelope boundary due to statistical tail fluctuations in normal operation data. Only when both conditions are met simultaneously is it considered a valid disturbance initiation event. For example, if the air compression section outlet pressure signal shows a unidirectional deviation exceeding the operating envelope boundary of the corresponding load band at a certain moment and the duration is at least 10 times the sampling period, the moment of the first occurrence of this continuous deviation is taken as the disturbance initiation moment. The moment when the corresponding downstream process section monitoring signal response deviates from the operating envelope boundary of the corresponding load band in the operating envelope library is identified as the response arrival moment. For example, if the main heat exchange section outlet temperature signal shows a continuous deviation exceeding the operating envelope boundary of the corresponding load band at a certain moment after the disturbance initiation moment, that moment is taken as the response arrival moment. The difference between the two is calculated to obtain the inter-segment propagation delay corresponding to the sample. The propagation delay refers to the time difference between the start time of the disturbance where the upstream process segment monitoring signal changes significantly and the arrival time of the response of the corresponding downstream process segment monitoring signal deviating from the operating envelope boundary. It reflects the actual delay time required for the disturbance signal to be transmitted from the upstream process segment to the downstream process segment through the medium flow or energy transfer path. It is used to assign values ​​to the delay elements of the corresponding process segment pairs in the process propagation delay matrix. For multiple valid samples within the same load band interval, two independent calculations are performed: delay value statistics and propagation weight statistics. In terms of delay value statistics, the median of the inter-segment propagation delay obtained from each valid sample is taken to obtain the calibration delay value of the corresponding process segment pair under the load band. In terms of propagation weight statistics, while obtaining the inter-segment propagation delay, the ratio of the downstream process segment monitoring signal response offset amplitude to the upstream process segment monitoring signal disturbance offset amplitude in the sample is extracted as the single propagation attenuation ratio corresponding to the sample. The arithmetic mean of the single propagation attenuation ratios of multiple valid samples within the same load band interval is taken to obtain the propagation weight of the corresponding process segment pair under the load band.

[0045] The calibration time delay value refers to the disturbance response sample that meets the valid sample condition extracted from historical step disturbance samples, load switching samples and start-up and shutdown samples for each coupled process segment pair within the corresponding load band interval. The result obtained by taking the statistical median of the inter-segment propagation time delay corresponding to each sample reflects the typical value of the disturbance propagation time of the coupled process segment pair under the corresponding load band. It is used for assigning the corresponding element in the process propagation time delay matrix and determining the time offset in the expected response deduction.

[0046] The propagation weight is a positive number between 0 and 1. The closer it is to 1, the smaller the amplitude attenuation when the upstream disturbance propagates to the downstream process section. The closer it is to 0, the greater the amplitude attenuation. When there are less than 3 effective samples in the corresponding load band for a certain process section, the propagation weight of the adjacent load band is used as an approximation. After sufficient sample accumulation, it is recalibrated. The effective sample refers to the disturbance response sample in the corresponding load band interval where the disturbance start time and response arrival time can be clearly identified and the inter-segment propagation path is consistent with the effective coupling topology. It is used to participate in the time delay statistical calculation of the corresponding process section pair under the load band.

[0047] Considering the statistical reliability of calibration results when the sample size is insufficient, the calibration time delay value is only fixed and written into the process propagation time delay matrix when a process segment pair consistently exhibits a propagation response in at least three load band samples. Process segment pairs that do not meet the above requirements are marked as pending accumulation and are not included in the matrix for the time being. Pending accumulation indicates that a process segment pair has fewer than three load bands that meet the valid sample conditions in the current historical samples, and its time delay calibration result does not yet have statistical reliability. Therefore, it is not written into the process propagation time delay matrix for the time being and will be recalibrated after sufficient samples have been accumulated. This process involves inter-segment coupling identification and analysis. After load band time delay calibration, when the device is in the load band switching transition process, i.e., when the current load value is near the boundary between two adjacent load band intervals, the calibration time delay value and propagation weight of the two adjacent load bands are read simultaneously. Using the relative position of the current load value between the center load values ​​of the two load bands as the interpolation coefficient, linear interpolation is performed on the time delay value and propagation weight respectively. The difference between the current load value and the center value of the low load band is divided by the difference between the center values ​​of the two adjacent load bands to obtain the interpolation coefficient α. Subtracting α from 1 yields the corresponding complementary coefficient 1-α. The calibration time delay value of the low load band is multiplied by the complementary coefficient 1-α, and the high load band... The load band calibration time delay value is multiplied by the interpolation coefficient α, and the two are added together to obtain the transition time delay value under the current load level. The low load band calibration propagation weight is multiplied by the complementary coefficient 1-α, and the high load band calibration propagation weight is multiplied by the interpolation coefficient α, and the two are added together to obtain the transition propagation weight under the current load level. The obtained transition time delay value and transition propagation weight replace the single load band calibration value as the expected response extrapolation parameters applicable to the transition period, and participate in the subsequent calculation of the consistency of the same source disturbance to ensure the continuity of the extrapolation calculation during the load band switching process. Each process section is used as the matrix row and column index, and the calibration time delay value under the corresponding load band is used as the index. The propagation weights are matrix elements, which are summarized to form a process propagation delay matrix. The process propagation delay matrix refers to a two-dimensional parameter matrix with each process segment of the entire space separation process as the row and column index, and the propagation delay values ​​and propagation weights obtained by calibration of historical disturbance response samples under low load, medium load and high load operating ranges for each coupled process segment as matrix elements. Each non-zero element in the matrix corresponds to an effective propagation edge verified by physical coupling relationship, reflecting the actual delay time required for the disturbance of the upstream process segment to reach the corresponding downstream process segment through the medium flow or energy transfer path, and is used for the deduction and calculation of the expected response offset value.

[0048] The propagation weight refers to the amplitude attenuation coefficient determined by the arithmetic mean of the ratio of the downstream process section response offset amplitude to the upstream process section disturbance offset amplitude in the historical disturbance response sample within the corresponding load band of the same coupled process section. The value range is 0 to 1. It is used to calculate the amplitude attenuation of the upstream disturbance offset during the expected response offset value extrapolation process, as well as the weighted summation calculation when multiple upstream propagation edges converge.

[0049] The aforementioned time delay calibration refers to the process of identifying the start time of a disturbance where the upstream process section monitoring signal changes significantly and the arrival time of the corresponding downstream process section monitoring signal response deviating from the operating envelope boundary for each effective propagation edge in the inter-segment coupling relationship, using historical step disturbance samples, load switching samples, and start-up / shutdown samples as data sources. The difference between these two times is used to calculate the propagation time delay corresponding to the sample. The statistical median of multiple effective samples within the same load band is then taken to obtain the calibration time delay value of the corresponding process section pair under that load band. This process is used to provide time delay parameter values ​​supported by measured data for each propagation edge element in the process propagation time delay matrix.

[0050] Based on the process propagation time delay matrix, the upstream disturbance is extrapolated to the expected response of each downstream process segment. The response is compared with the measured response and the consistency is calculated to obtain the anomaly identification result.

[0051] Using the process propagation time delay matrix as the path basis and the real-time monitoring signal in the unified time series dataset as the comparison benchmark, the expected response of each downstream process section is extrapolated from the upstream load disturbance along the propagation path, and the expected response is quantitatively compared with the measured response of the corresponding process section. The consistency of the same source disturbance is calculated to obtain the anomaly identification result, thereby achieving effective differentiation between flexible load fluctuations and internal process anomalies.

[0052] Specifically, when a significant change in the upstream process segment monitoring signal is detected, the time offset is calculated using the calibration time delay value of the corresponding process segment pair in the process propagation time delay matrix. This time offset refers to the propagation time delay value of the corresponding coupled process segment pair in the process propagation time delay matrix after load band time delay calibration, reflecting the actual delay time required for the upstream disturbance to propagate to the downstream process segment via the physical coupling path. This time offset is used to determine the starting point for the expected response extrapolation. The upstream disturbance signal is then extrapolated sequentially to each downstream process segment along the propagation edges in the effective coupling topology. The time offset is based on the calibration time delay value under the load band corresponding to each effective propagation edge in the process propagation time delay matrix, and the amplitude attenuation is based on the propagation weight corresponding to each effective propagation edge. The upstream process segment monitoring signal is then read from the unified time series dataset at the time of the disturbance. The measured offset after the start time is multiplied by the corresponding propagation weight to obtain the attenuated propagation amplitude, and then shifted backward to the corresponding time according to the calibrated time delay value. This is then propagated sequentially to each downstream process segment along the propagation edge in the effective coupling topology. When a downstream process segment simultaneously receives propagation amplitudes from multiple upstream propagation edges, the multiple propagation amplitudes are weighted and summed using the corresponding weights of each propagation edge as coefficients to obtain the combined expected response offset value of the process segment at the corresponding time. When the propagation path crosses an intermediate process segment, the combined expected response offset value of the intermediate process segment is used as the starting offset of the next level of propagation, and this process is sequentially extrapolated to the next downstream process segment until all downstream process segments in the effective coupling topology are covered, thus obtaining the expected response offset value of each downstream process segment at the corresponding time.

[0053] The upstream disturbance refers to a significant change event in the monitoring signal of the upstream process section in the entire space separation process, where the offset amplitude exceeds the operating envelope boundary of the load band corresponding to the operating envelope library at a certain sampling time and the duration meets the preset conditions. This time is taken as the disturbance start time, and the measured response offset value of the corresponding process section is taken as the disturbance amplitude. It is used to perform the expected response deduction to each downstream process section along the corresponding effective propagation edge in the process propagation time delay matrix.

[0054] The downstream process segment expected response refers to the parameter offset that the corresponding process segment should theoretically exhibit at the corresponding time after the measured response offset value of the upstream disturbance is deduced step by step along the effective coupling topology to the corresponding downstream process segment based on the time offset value of the calibration time delay value of the corresponding process segment pair in the process propagation time delay matrix and the propagation weight is used as the amplitude attenuation basis. It is used to compare with the measured response offset value to calculate the consistency of the same source disturbance.

[0055] Simultaneously, the measured response offset value of the corresponding downstream process segment at the same moment is read from the unified time series dataset. Taking the steady-state anchor point of the load band corresponding to the segmented state anchor point set as the zero-bias reference benchmark, the measured acquisition value of the monitoring signal of the corresponding downstream process segment at the current moment is read from the unified time series dataset. The measured acquisition value is subtracted from the steady-state anchor point of the load band corresponding to the segmented state anchor point set to obtain the deviation of the monitoring signal from the normal operation reference center. When the obtained difference is positive, it indicates that the measured acquisition value is higher than the steady-state anchor point. When the obtained difference is negative, it indicates that the measured acquisition value is lower than the steady-state anchor point. The obtained difference is taken as the actual deviation of the monitoring signal at the current moment, which is the value of the actual deviation. It is used to compare and calculate with the expected response offset value in the same source disturbance consistency formula, and serves as the basis for the value of the measured response offset value. The actual offset is the difference between the measured value of the monitoring signal of the corresponding downstream process segment and the steady-state anchor point of the corresponding load band in the segmented state anchor point set, reflecting the degree to which the current operating state of the process segment deviates from the normal reference. The expected response offset refers to the parameter offset that the process segment should theoretically exhibit at the corresponding time after extrapolating the upstream disturbance signal to the corresponding downstream process segment along the effective coupling topology based on the time offset of the corresponding process segment pair in the process propagation time delay matrix. It is used to compare with the measured response offset value to calculate the consistency of the same source disturbance. The measured response offset refers to the actual value of the monitoring signal of the corresponding downstream process segment at the same time read from the unified time series dataset, and the actual deviation calculated with the steady-state anchor point of the corresponding load band in the segmented state anchor point set as the zero-bias reference. It is used to compare with the expected response offset value to calculate the consistency of the same source disturbance.

[0056] After completing the expected response derivation and measured response extraction, the consistency degree of the homogeneous disturbance at time t is calculated using the following formula:

[0057] ,

[0058] in, This represents the consistency of the same source perturbation at time t, and its value ranges from [0,1]. This represents the measured response offset value of the j-th monitored quantity at time t; This represents the expected response offset value of the j-th monitored quantity at time t, derived from the process propagation time delay matrix. This represents the confidence weight of the j-th monitoring quantity, which is assigned based on the data quality label of the corresponding signal. The weight of the signal with a confidence downgrade is reduced accordingly. To prevent tiny positive numbers with a denominator of zero, the preferred value is... t is the time series index of the monitoring signal sampling, representing the sampling time, and j is the number index of the monitoring quantity of each downstream process section, representing the monitoring quantity participating in the consistency calculation of the same source disturbance;

[0059] The consistency of the same source disturbance The closer it is to 1, the better the measured response of each downstream process section matches the expected response derived from the time delay matrix. The current offset can be attributed to the normal propagation of the upstream load disturbance. The measured response refers to the actual collected value of the monitoring signal of the corresponding downstream process section read from the unified time series dataset at the current moment, that is, the original engineering quantity value directly reported by the instrument without any benchmark deduction processing. The closer the value is to 0, the more significant the deviation between the measured and expected responses, indicating that the current offset contains anomalous components that cannot be explained by upstream load disturbances; based on the consistency of the aforementioned co-source disturbances... Output anomaly identification results according to the following criteria: when When the offset is greater than or equal to the first threshold, the current offset is determined to be a flexible operation fluctuation of the same source, which is a recoverable offset driven by external load and does not require triggering abnormal handling; when the second threshold is less than or equal to the first threshold, the current offset is determined to be a flexible operation fluctuation of the same source, which is a recoverable offset driven by external load and does not require triggering abnormal handling. When the current state is less than the first threshold, it is determined to be a mixed state to be observed, and the relevant process segment is marked for continuous monitoring; when If the measured response offset values ​​of at least two monitored quantities within the same process segment deviate from the corresponding operating envelope boundary in the same direction for at least five consecutive sampling periods, and the result is deemed an internal anomaly, triggering an anomaly flag for the corresponding process segment, the first threshold refers to the consistency of the same source disturbance. The upper decision boundary is preferably set to 0.75. If the current offset is not lower than the first threshold, it is determined to be a flexible operation homogeneous fluctuation. This is used to distinguish between recoverable external load-driven offsets and mixed states requiring further observation. External load-driven offsets refer to parameter offsets caused by changes in load commands from upstream process sections or external disturbance inputs, propagated to downstream process sections via the effective coupling topology. The difference between the measured response offset value and the expected response offset value is within the first threshold range, falling within the normal elastic response range of the device and requiring no abnormal handling. Flexible operation homogeneous fluctuations are... The device operating state corresponding to a value not lower than the first threshold indicates that the measured response offset values ​​of each downstream process section are highly consistent with the expected response offset values ​​derived from the process propagation time delay matrix. The current offset can be completely attributed to the normal propagation of the offset driven by external load, and belongs to a self-recoverable process elastic state; the mixed state refers to... The device operating status corresponding to the position between the second and first thresholds indicates that the current measured response offset value contains both homologous components that can be explained by external load-driven offsets and deviation components whose causes cannot yet be clearly attributed. Continuous monitoring of the relevant process sections is necessary, pending further investigation at subsequent time points. After further evolution, a new judgment is made; the second threshold refers to the consistency of the same source perturbation. The lower decision boundary is preferably 0.40. When the deviation falls below the second threshold and the local process segment offset remains concentrated, it is identified as an internal anomaly and prioritized to trigger the anomaly marker output of the corresponding process segment. After the above deduction, comparison, and consistency calculation, the anomaly identification results at each time point are obtained and can be called by the hierarchical visualization management interface. The anomaly identification results refer to the consistency of each process segment in the entire space separation process at each sampling time based on the same source disturbance. The calculated state determination conclusions include three determination types: flexible operation homogeneous fluctuation, mixed state to be observed, and internal anomaly priority. These are used as the input basis for generating the state labeling of each process section and the priority order of job verification in the hierarchical visualization management interface.

[0060] In one embodiment, a liquid air separation unit receives a grid peak-shaving command, rapidly switching its load from a 70% high-load operating range to a 40% medium-load operating range. The outlet pressure of the air compression section subsequently drops significantly. Based on the process propagation time delay matrix, this disturbance is expected to propagate to the main heat exchange section after approximately 180 seconds, and further to the distillation cold box section after approximately 320 seconds. Within this time window, the outlet temperature of the main heat exchange section and the top pressure of the distillation cold box section successively deviate. The measured response offset values ​​of each downstream process section are compared with the expected response offset values ​​derived from the process propagation time delay matrix to calculate the consistency of the common source disturbance. If the value remains above the first threshold, it is determined that the above deviations are all fluctuations of the same source in flexible operation driven by external load, and there is no need to trigger abnormal handling. Based on this, the operator can determine that the current device is in the normal load switching response process, avoiding unnecessary intervention operations due to misjudgment under the background of large fluctuations.

[0061] Meanwhile, the liquid oxygen level in the distillation cold box section continued to drop after the load switch, and its measured response offset value significantly exceeded the expected response offset value derived from the process propagation time delay matrix. The calculated consistency of the same source disturbance... If the liquid level drops below the second threshold and the deviation remains concentrated in the distillation cold box section, it is determined to be an internal anomaly with priority, triggering an anomaly flag for that process section. Subsequent maintenance confirmed that the anomaly was caused by a local decrease in heat exchange efficiency on the main cooling liquid oxygen side, and was unrelated to external load switching. The above process shows that the introduction of the consistency of the same source disturbance effectively solves the problem that operators cannot distinguish between flexible fluctuations and internal anomalies in the context of large load fluctuations, avoiding the dilemma of false alarms of normal fluctuations and missed alarms of real anomalies in the existing technology.

[0062] Using the margin parameters of each process segment in the aforementioned operational envelope library as boundaries, and combining real-time margin and rate of change to calculate the contact time, a hierarchical hidden risk window is obtained.

[0063] Using the margin parameters of each process segment in the operating envelope library as the source of risk boundaries, the margin parameters include, but are not limited to, the remaining space of key temperature, pressure, liquid level, and flow parameters of each process segment from the upper or lower boundary of the corresponding load band operating envelope; the current collected values ​​of the margin parameters of each process segment are read in real time from the unified time series dataset, and the difference between the current collected value and the operating envelope boundary of the corresponding load band in the operating envelope library is calculated to obtain the real-time margin of each process segment; within the continuous sampling window, the rate of change is obtained by dividing the difference of the real-time margin at adjacent sampling times by the sampling time interval. When the rate of change is negative... When the margin continuously narrows, the estimated contact time is obtained by dividing the real-time margin by the absolute value of the rate of change. Based on the estimated contact time, risk levels are assigned: a red "urgent window" indicates a contact time of no more than 15 minutes; an orange "warning window" indicates a contact time between 15 and 45 minutes; a yellow "concern window" indicates a contact time between 45 and 120 minutes; and no risk flag is triggered if the contact time exceeds 120 minutes. The estimated contact time and corresponding risk level for each process segment are summarized to form the hierarchical implicit risk window, which is then used in the generation of the hierarchical visual management interface.

[0064] The estimated time to reach the boundary refers to the quotient obtained by dividing the current real-time margin of each process segment by the larger of the absolute value of the corresponding rate of change and the preset zero lower limit, under the assumption that the current real-time margin and the rate of change remain unchanged. It reflects the estimated remaining time required for the margin parameter of each process segment to reach the boundary of the margin parameter from the current position under the current narrowing trend. It is used as the basis for the classification and labeling of the risk level of each process segment in the hierarchical hidden risk window and the calculation of the risk urgency score in the priority score of the inspection.

[0065] The anomaly identification results, process propagation time delay matrix and hierarchical hidden risk window are superimposed on the process topology diagram, and the verification priority order is generated by joint weighting to obtain a hierarchical visual management interface;

[0066] Based on the anomaly identification results, the process propagation time delay matrix, and the hierarchical latent risk windows, a process topology map is first constructed using each process segment of the entire space separation process as nodes and each effective propagation edge in the process propagation time delay matrix as a directed connection. On this basis, the anomaly identification results are marked to the corresponding process segment nodes using node color codes: green for flexible operation with homogeneous fluctuations, yellow for the mixed state to be observed, and red for priority internal anomalies. The hierarchical latent risk windows are superimposed and marked in the form of warning rings around the nodes, with the color of the warning rings corresponding to the risk level, and the outer side of the warning rings is marked. The estimated contact time is calculated; then, the priority order of each process segment is calculated by weighting three indicators: risk urgency, propagation coverage, and data credibility. The preferred weights for the three indicators are 0.5, 0.3, and 0.2, respectively. The process segments with the highest scores are sorted from highest to lowest and the top 3 to 5 are output to the job task card. Each task card item includes the process segment name, anomaly type, estimated contact time, and suggested verification operations. This results in a hierarchical and visual management interface covering the source of disturbance, propagation path, risk window, and job task card, which is used by operators to perform actions.

[0067] Based on the processing results corresponding to the hierarchical visual management interface, write back and update various parameters to obtain a continuously corrected parameter library;

[0068] Based on the operation confirmation results, maintenance conclusions, and final disposal results corresponding to the hierarchical visualization management interface, the disposal results are written back to the sample library to update the process propagation time delay matrix, operating envelope library, segmented state anchor point set, and visualization priority parameters, thus obtaining a continuously corrected parameter library. In the next monitoring cycle, each step of segmented calibration, time delay calibration, expected response deduction, margin calculation, and verification priority generation reads the updated parameters from the continuously corrected parameter library to replace the initial calibration parameters for calculation, so that the entire process monitoring forms a continuous closed loop of acquisition, calibration, deduction, identification, risk calculation, visualization output, disposal write-back, parameter update, and re-acquisition.

[0069] Combination Figure 2 The chart shows a comparison of the effectiveness of the air separation process data monitoring and visualization management method with existing technologies. The black bars represent the technical effectiveness of the present invention, while the gray bars represent the technical effectiveness of existing technologies. Figure 2 It can be seen that the technical effect of the present invention is superior to that of the prior art.

[0070] Example 2, based on Example 1, provides a method for full-process data monitoring and visualization management of air separation technology. It focuses on explaining the construction details of the hierarchical latent risk window in claim 1, the generation rules of the layered visualization management interface, and the write-back update mechanism of the continuously correcting parameter library. Building upon the closed-loop system for cross-section disturbance propagation identification and anomaly judgment established in Example 1, each process section has formed a unified set of anomaly identification results and hierarchical latent risk windows. This example further addresses how to use these results as a driving force to achieve full-process visualization management for operators under the linkage constraints of three types of information: disturbance source, propagation path, and future time window exceeding limits. It also ensures real-time traceability and continuous self-correction capabilities for the risk status and job verification sequence of each process section. The specific solution is as follows:

[0071] In one embodiment, the construction of the hierarchical latent risk window further includes:

[0072] The margin parameters of each process segment in the operating envelope library are used as boundaries. These margin parameters include the temperature difference margin, tower pressure difference margin, liquid level reserve margin, reflux margin, and critical valve opening margin of the distillation cold box segment. The boundaries of the margin parameters of each process segment are determined based on the limit offset of the corresponding load band operating envelope in the operating envelope library. They dynamically correspond to the load band intervals, and the low-load, medium-load, and high-load operating intervals each have independent margin boundary values. These are used as the upper limit of the benchmark for subsequent real-time margin calculations. For example, the margin boundary of the liquid level reserve margin of the distillation cold box segment in the high-load operating interval is determined based on the limit offset of the 70% to 100% load band operating envelope, while in the low-load operating interval, it is re-determined based on the 20% to 40% load band operating envelope. The two are independent of each other and are not universally applicable, so as to ensure that the margin boundary matches the current actual operating load level of the unit.

[0073] The aforementioned hierarchical latent risk window refers to the set of full-process risk status formed by summing the estimated contact time of each process segment based on the estimated contact time calculated by combining the current real-time margin and change rate of the margin parameters of each process segment, and dividing each process segment into four risk levels: red urgent window, orange warning window, yellow attention window, and normal monitoring status according to the first time threshold, the second time threshold, and the third time threshold. This set is used to provide a quantitative representation of the current latent risk urgency of each process segment to the hierarchical visualization management interface.

[0074] The aforementioned temperature margin refers to the remaining space of the minimum allowable temperature difference between the hot and cold ends of the main condenser-evaporator in the distillation cold box section, reflecting the safety reserve of the main condenser's heat exchange efficiency. The minimum allowable temperature difference is derived from the heat exchange design parameters of the main condenser-evaporator determined by process calculations and loaded into the design documents during the device design phase. The specific value is based on the minimum allowable operating value of the main condenser temperature difference in the design documents.

[0075] The aforementioned tower pressure differential margin refers to the remaining space between the currently measured pressure difference between the upper or lower column of the distillation cold box section and the flooding critical pressure differential, reflecting the degree to which the gas-liquid load in the tower approaches the flooding boundary; the flooding critical pressure differential is derived from the flooding pressure differential limit value of the corresponding tower type under rated operating conditions in the equipment performance data table provided by the tower equipment manufacturer. When the performance of the equipment deteriorates after long-term operation, the manufacturer's rated value is corrected based on the results of the most recent tower performance calibration test.

[0076] The liquid level reserve margin refers to the remaining space between the current liquid level of the liquid storage tank and the minimum safe liquid level, reflecting the sustainable capacity of liquid product delivery scheduling. The minimum safe liquid level is derived from the lower limit liquid level value for safe operation registered for each storage tank in the unit's operating procedures. This value is determined by the process designer based on the storage tank's geometric parameters, the pump inlet net positive suction head requirement, and the minimum storage requirement for delivery scheduling.

[0077] The reflux margin refers to the remaining space between the liquid nitrogen reflux flow rate in the distillation cold box section and the minimum reflux flow rate required to maintain the distillation balance of the column system, reflecting the stable reserve of the column system's separation effect; the minimum reflux flow rate is derived from the minimum reflux ratio of the distillation column determined by process calculations during the unit design stage. The specific value of the minimum reflux flow rate is obtained by multiplying the minimum reflux ratio by the feed rate under the corresponding load level, and then loaded into the design file as the operating benchmark.

[0078] The critical valve opening margin refers to the remaining stroke ratio of the current opening of the critical control valve in each process section from the fully open or fully closed limit position, reflecting the adjustable space of the actuator; the fully open and fully closed limit positions are derived from the preset soft limit values ​​of stroke for each critical control valve in the DCS configuration parameters. These soft limit values ​​are determined by the instrument engineer during the commissioning phase based on the rated stroke parameters provided by the valve manufacturer and combined with the on-site pipeline conditions, and are written into the control system. When the valve is repaired or replaced, it is recalibrated and updated by the instrument engineer.

[0079] The aforementioned limit offset refers to the maximum allowable offset of the margin parameter of each process segment within the operating envelope of the corresponding load band in the operating envelope library. It is determined by the difference between the boundary value of the side with the largest deviation from the corresponding steady-state anchor point in the upper and lower boundaries of the corresponding load band operating envelope and the steady-state anchor point, and is used as the risk boundary value of the corresponding margin parameter under the load band.

[0080] After determining the margin parameter boundaries for each process segment, the current acquired values ​​of the margin parameters for each process segment are read in real time from the unified time-series dataset. Using the steady-state anchor point of the corresponding load band in the segmented state anchor point set as the zero-bias reference, the difference between the current acquired value and the margin parameter boundary is calculated to obtain the real-time margin for each process segment. The original difference is obtained by subtracting the corresponding load band margin parameter boundary value from the current acquired value of the margin parameter for each process segment. When the original difference is positive, it indicates that the current acquired value has not yet touched the margin parameter boundary, and the original difference is taken as the real-time margin for that process segment, proceeding to the subsequent change rate calculation and expected contact time calculation stage. When the original difference is negative or zero... When the current collected value has reached or exceeded the margin parameter boundary, the process segment is directly marked as having reached the risk boundary, the real-time margin is recorded as zero, the estimated contact time is recorded as zero, and the red urgent window is directly triggered, without proceeding to the estimated contact time calculation stage. Within the continuous sampling window, the difference between the real-time margin value at the next sampling time and the real-time margin value at the previous sampling time is divided by the corresponding sampling time interval to obtain the rate of change of the real-time margin for each process segment. A negative rate of change indicates that the real-time margin is continuously narrowing, while a positive rate of change indicates that the real-time margin is recovering. Only when the rate of change is negative will the subsequent estimated contact time calculation stage be initiated.

[0081] The real-time margin refers to the value of the original difference obtained by subtracting the corresponding load zone margin parameter boundary value from the current collected value of the margin parameter of each process section under positive conditions. When the original difference is positive, the real-time margin is equal to the positive value. When the original difference is zero or negative, the real-time margin is recorded as zero. It reflects the remaining space between the current operating state of the process section and the risk boundary, and is used for subsequent change rate calculation and expected contact time calculation.

[0082] The "risk boundary reached" state refers to the difference between the current collected value of the margin parameter of each process segment and the value of the margin parameter boundary of the corresponding load band being less than or equal to zero. That is, the current collected value has reached or exceeded the margin parameter boundary determined by the limit offset of the corresponding load band in the operating envelope library. This indicates that the current operating state of the process segment has no remaining safety margin. At this time, the expected contact time is recorded as zero, directly triggering the red urgent window mark and outputting to the corresponding process segment node in the hierarchical visualization management interface, prompting the operator to check and handle it immediately.

[0083] The rate of change of the real-time margin refers to the quotient obtained by dividing the difference between the real-time margin value at the next sampling moment and the real-time margin value at the previous sampling moment within the continuous sampling window by the corresponding sampling time interval. It reflects the degree to which the real-time margin narrows or recovers over time and is used in conjunction with the real-time margin to calculate the expected touch-off time.

[0084] The aforementioned estimated contact time calculation step refers to a calculation process that is triggered only when the real-time margin change rate is negative. The real-time margin is divided by the larger of the absolute value of the change rate and the lower limit of zero protection to obtain the estimated remaining time for each process segment to reach the margin parameter boundary under the current change trend.

[0085] The estimated contact time for each process segment is obtained by dividing the real-time margin by the absolute value of the rate of change. This estimated contact time is based on the approximate assumption that the current rate of change remains relatively stable over a short period. However, in scenarios with frequent load changes, the rate of change itself may change rapidly. To reduce the resulting estimation bias, the rate of change value is continuously updated using a sliding window. The window length is preferably the weighted average of the most recent 3 to 5 consecutive sampling periods, with higher weights for sampling points closer to the current time, to make the estimated rate of change closer to the current actual trend. Simultaneously, the estimated contact time is recalculated and refreshed in each sampling period, rather than being fixed after a single calculation. This allows for timely adjustments in the next sampling period when the rate of change reverses direction or undergoes a sudden change in amplitude. When the absolute value of the rate of change is less than the estimated contact time, the estimated contact time is adjusted accordingly. When the defense has a zero lower limit, the zero lower limit is used instead of the absolute value of the rate of change in the calculation. Based on the expected contact time, each process segment is graded and labeled according to the following criteria: when the expected contact time does not exceed the first time threshold, it is marked as a red urgent window, requiring immediate verification and handling; when the expected contact time is between the first time threshold and the second time threshold, it is marked as an orange warning window, requiring early intervention and attention; when the expected contact time is between the second time threshold and the third time threshold, it is marked as a yellow attention window, requiring continuous tracking; when the expected contact time exceeds the third time threshold, no risk label is triggered, and the corresponding process segment maintains normal monitoring status; after the above real-time margin calculation, rate of change determination, and graded labeling, the expected contact time and corresponding risk level of each process segment are summarized to form the graded hidden risk window.

[0086] The estimated contact time refers to the estimated remaining time required for the margin parameter of each process segment to narrow from its current position to the margin parameter boundary, under the assumption that the current real-time margin and the rate of change remain unchanged. It is used for the classification and determination of the hidden risk window level.

[0087] The preset zero-prevention lower limit refers to the smallest positive number used in the division operation when the absolute value of the rate of change approaches zero during the calculation of the expected contact time. The preferred value is... This is used to prevent the result of a division operation from diverging to infinity when the absolute value of the rate of change is zero, thus causing the estimated contact time to lose its meaning.

[0088] The first time threshold is the urgency judgment boundary of the expected contact time, preferably 15 minutes, used to distinguish between the red urgent window that requires immediate action and the orange warning window that requires early intervention; the second time threshold is the warning judgment boundary of the expected contact time, preferably 45 minutes, used to distinguish between the orange warning window and the yellow attention window that requires continuous tracking; the third time threshold is the attention judgment boundary of the expected contact time, preferably 120 minutes, used to distinguish between the yellow attention window and the normal monitoring state.

[0089] In one embodiment, the generation of the hierarchical visual management interface further includes:

[0090] First, a process topology graph is constructed based on the effective propagation edges between process segments in the process propagation delay matrix. Each process segment in the entire space division process is used as a node, and the propagation edges between each coupled process segment pair, verified by physical coupling, are used as directed lines. The direction of the lines is from the upstream process segment to the downstream process segment. The calibrated delay values ​​and propagation weights under the corresponding load bands are labeled on the lines, consistent with the corresponding elements in the process propagation delay matrix. This yields a process topology graph base map reflecting the overall disturbance propagation topology. The generation and updating of the above process topology graph base map and subsequent dynamic overlay annotations are implemented using a web-based graphics rendering engine. Specifically, the process topology graph base map is stored in vector graphics format on the visualization server. The position coordinates of each process segment node and directed connection are pre-configured based on the physical spatial relationship of each process segment in the process flow diagram. Dynamic status information such as anomaly identification results and hierarchical hidden risk windows are calculated by the background computing service in each sampling period and pushed to the front-end visualization interface in the form of structured data messages. After receiving the data, the front end performs incremental refresh on the color mark, warning ring and numerical label of the corresponding process segment node. The refresh cycle is consistent with the shortest sampling cycle of the unified time series dataset. The job task card is embedded in the right side of the visualization interface in the form of an independent panel. Each item is scored and arranged according to the priority of verification and is updated synchronously with each calculation refresh. Operators can complete the closed loop of status viewing and handling operations in the same interface.

[0091] The aforementioned process topology map refers to a static topology map constructed with each process segment of the entire space separation process as nodes and each effective propagation edge in the process propagation time delay matrix as a directed connection. It only presents the physical coupling connection relationship and corresponding time delay label between each process segment, without any dynamic state information, and is used as the underlying graphic basis for subsequent anomaly identification results and hierarchical hidden risk window overlay labeling.

[0092] The aforementioned process topology map refers to a dynamic visualization graphic formed by overlaying and annotating the anomaly identification results and hierarchical hidden risk windows onto the corresponding process segment nodes on the basis of the process topology map base map. The node color marks and early warning ring markings are updated synchronously with the changes in the real-time monitoring signals of each process segment, and it serves as the core display carrier for the hierarchical visualization management interface.

[0093] Based on the aforementioned process topology map, the anomaly identification results and hierarchical latent risk windows are superimposed and annotated onto the corresponding process segment nodes: the anomaly identification results are marked in the form of node color codes, with green, yellow, and red node color codes corresponding to flexible operation homogeneous fluctuations, observed mixed states, and internal anomaly priority, respectively. Operators can directly determine the current state of the corresponding process segment from the node color codes; the hierarchical latent risk windows are superimposed and annotated in the form of an outer warning ring around the node, where the expected contact time is marked numerically on the outside of the warning ring. By superimposing and annotating the node color codes determined by the anomaly identification results and the warning rings determined by the hierarchical latent risk windows, operators can simultaneously obtain two types of information at the same node: the current state nature and the urgency of future risks. Furthermore, when the same When the anomaly identification results of a process segment node intersect with the hierarchical latent risk window, the higher risk level is prioritized for operational guidance. Specifically, if the anomaly identification result of a process segment is a flexible operation with the same source fluctuation but the warning loop enters the red urgent window, a red urgent label is added next to the node color mark, indicating to the operator that although the current offset can be attributed to external load driving, the margin is close to the boundary and the margin parameter status still needs to be checked immediately. Conversely, if the anomaly identification result of a process segment is an internal anomaly with priority but the expected contact time exceeds the third time threshold, the red node color mark is maintained but the warning loop is not displayed, indicating that the anomaly has not yet triggered margin risk but the cause of the anomaly needs to be investigated as soon as possible. Among them, process segment nodes that have not triggered risk marking do not display warning loops and maintain normal monitoring status markings.

[0094] The node color codes refer to the color markings applied to the corresponding process segment nodes in the process topology diagram based on the current determination type of each process segment in the anomaly identification results. Flexible operation with homogeneous fluctuations corresponds to green node color codes, indicating that the current offset can be attributed to external load driving and requires no action; mixed states to be observed correspond to yellow node color codes, indicating that there are deviation components whose cause cannot yet be clearly attributed and require continuous monitoring; internal anomalies are prioritized, corresponding to red node color codes, indicating that the current offset contains significant abnormal components and requires immediate verification. This is used by operators to visually identify the current state of each process segment in the process topology diagram.

[0095] After completing the overlay annotation of the three types of information, a joint weighted calculation of the verification priority score is performed on each process segment based on three indicators: risk urgency, propagation coverage, and data credibility, to obtain the verification priority order of each process segment. The risk urgency is determined based on the expected contact time of the corresponding process segment in the hierarchical latent risk window. Using the third time threshold as a normalization benchmark, the difference between the third time threshold and the expected contact time is divided by the third time threshold to obtain a risk urgency score between 0 and 1. The shorter the expected contact time, the closer the score is to 1. When the expected contact time is zero or has reached the risk boundary, the score is 1. When the expected contact time exceeds the third time threshold, the score is 0. The propagation coverage is based on all downstream processes reachable along the effective propagation edge from the process segment as the disturbance source in the process propagation time delay matrix. The number of process segments is determined using the total number of process segments minus 1 as a normalization benchmark. The number of reachable downstream process segments is divided by the total number of process segments minus 1 to obtain a propagation coverage score between 0 and 1. The data reliability is determined based on the proportion of signals marked as normal in all monitored signals of the corresponding process segment to the total number of monitored signals in that process segment. This proportion is the data reliability score between 0 and 1. The higher the proportion of normal signals, the higher the score; the higher the proportion of signals with degraded reliability, the lower the score. The above three scores are multiplied by their corresponding weighting coefficients and then summed to obtain the inspection priority score for each process segment. The preferred weighting coefficients are 0.5, 0.3, and 0.2, respectively. The process segments with the highest scores are sorted from high to low according to the inspection priority scores, and the top 3 to 5 process segment links are output to the job task card.

[0096] The aforementioned verification priority order refers to the priority score for each process segment calculated by weighting three indicators: risk urgency, dissemination coverage, and data credibility. The scores are arranged from highest to lowest to determine the order of handling for each process segment. A higher score indicates a higher level of risk urgency, a wider range of disturbance impact, and stronger data credibility for that process segment, and therefore should be prioritized for verification and handling by operators. This is used to arrange and output verification items in the job task card, ensuring that operators can determine the order of handling based on objective quantitative evidence rather than subjective experience when abnormal markers appear in multiple process segments simultaneously.

[0097] Based on the priority score of the verification, a hierarchical visual management interface is generated, covering the source of disturbance, propagation path, risk window and job task card. The source of disturbance layer presents the starting process segment and its abnormality type pointed to by the current anomaly identification result, so that operators can clearly understand the origin location and nature judgment of the current offset, and avoid the problem of not being able to locate the root cause when multiple process segments have state changes at the same time.

[0098] The propagation path layer presents the propagation process of disturbance from the source process segment to each downstream process segment in the form of directed paths with time delay annotations, based on the corresponding effective propagation edges in the process propagation time delay matrix. This is used by operators to understand the current impact range of the disturbance and the expected disturbance sequence of each downstream process segment, and to judge in advance which process segments need to be focused on in the future.

[0099] The risk window layer presents the hierarchical implicit risk window labels and expected contact times for each downstream process segment, which is used by operators to obtain the risk urgency ranking of each process segment before the parameters exceed the limit, providing a time window basis for the execution of subsequent job verification priority order;

[0100] The task card layer scores according to the verification priority order, outputting the 3 to 5 highest priority process links as verification items that operators can directly execute. Each verification item includes the corresponding process segment name, anomaly type label, estimated contact time, and suggested verification operation. The suggested verification operation is automatically generated by the system based on a pre-registered verification rule table. The pre-registered verification rule table uses the combination of process segment name and anomaly type as the search key, and outputs a description of the verification action that the operator should perform under that combination. The rule table content is pre-compiled and stored in the system by process technicians based on the equipment operating procedures and historical maintenance experience. When the search key matches a corresponding item in the rule table, the verification action description of that item is directly read as the suggested verification operation output. When the search key does not have an exact match in the rule table, a general verification template corresponding to the process segment name is used as a substitute output, and a prompt mark is added to the task card to remind the operator that the suggested verification operation is a general template rather than a specific item and needs to be adjusted according to the actual situation on site. This allows the operator to perform the action accordingly.

[0101] In one embodiment, the construction of the continuously corrected parameter library further includes:

[0102] Based on the operation confirmation results, maintenance conclusions, and final handling results corresponding to the hierarchical visual management interface, the above handling results are categorized by event type and written back to the sample library. The operation confirmation result refers to the record of the operator's confirmation or denial of the anomaly judgment conclusion after performing verification according to the job task card. The maintenance conclusion refers to the actual identification result of the cause of the anomaly after the equipment maintenance is completed. The final handling result refers to the complete handling record of this disturbance event after the device returns to normal operation. Among the above three types of results, the operation confirmation result and the final handling result are mandatory items. Only after both are recorded can the event be written into the sample library to form a valid review sample. The maintenance conclusion is a conditional item. When the event triggers the equipment maintenance process, the maintenance conclusion must be completed before it can be entered into the database. If the event is determined to be non-needing maintenance after operation confirmation, it can be entered into the database without waiting for the maintenance conclusion. When there is a discrepancy between the operation confirmation result and the maintenance conclusion in the identification of the cause of the abnormality, the maintenance conclusion shall prevail, and the relevant judgment mark in the operation confirmation result shall be corrected before entering into the database to ensure the uniqueness of the identification of the cause of abnormality in the sample database. When the maintenance is not yet completed, the event is marked as pending closure in the sample database and will not participate in the centralized update statistical calculation. It will be converted into a formal and valid review sample after the maintenance conclusion is completed and the consistency verification is completed.

[0103] Based on the valid review samples, the process propagation delay matrix, operational envelope library, and visualization priority parameters are updated as follows: The process propagation delay matrix is ​​updated using the measured propagation delay values ​​of each process segment in this disturbance event as new samples, and the median of the delay statistics for the corresponding process segment pairs is recalculated, replacing the original calibrated delay values ​​in the matrix; the operational envelope library is updated using the measured response offset values ​​of each process segment in this disturbance event as new samples, and the 5th and 95th percentile boundaries of the parameters for each process segment under the corresponding load band are recalculated, replacing the original operational envelope boundary values; the visualization priority parameters are updated by adjusting the weights of the three indicators—risk urgency, propagation coverage, and data reliability—based on the execution results of this job task card. When a certain indicator... When the actual guiding effect of an indicator on the priority of verification in this event deviates from the expected value by more than 10% of the current weight value of that indicator, the corresponding weight will be adjusted in steps, with each adjustment step being 0.02. Specifically, the historical average proportion of high-priority process segments confirmed as valid anomalies in all valid review samples within the most recent centralized update cycle will be used as the expected benchmark value. If the actual proportion of high-priority process segments corresponding to a certain indicator that are confirmed as valid anomalies in this update cycle is higher than the expected benchmark value, then the weight of that indicator will be increased by one step, and vice versa. During the first centralized update, since there is no historical average proportion available for reference, 0.5 will be used as the initial expected benchmark value. After the three weights are adjusted, they will be re-normalized to a sum of 1.To ensure the consistency of the weighting system, a centralized update is preferably performed every 20 or more valid review samples or monthly. To prevent coupling deviation between the updated running envelope and the time delay matrix, a step-by-step execution strategy is adopted for each centralized update: first, the process propagation time delay matrix is ​​updated based on the current running envelope; then, the running envelope is updated again using the updated time delay matrix as a constraint. The constraint means that when recalculating the 5th and 95th percentile boundaries of the running envelope, the disturbance start time and response arrival time in the historical samples are re-identified using the new calibration time delay values ​​of each process segment pair in the updated time delay matrix. Only disturbance response segments that still meet the valid sample conditions after re-identification based on the new time delay values ​​are included in the percentile statistics of the running envelope. The calculation is performed to eliminate interference from invalid samples incorrectly included due to deviations in the old time delay values ​​on the envelope boundary. The new time delay value refers to the updated result obtained by replacing the original calibration time delay value in the process propagation time delay matrix with the measured propagation time delay value of each process segment in the accumulated valid review samples during the continuous correction parameter library update process. This new sample is then used to re-execute the calculation of the time delay statistical median of the corresponding process segment pair, reflecting the latest typical value of the disturbance propagation time between each coupled process segment pair in the current operation stage of the device. It is used to re-identify the disturbance start time and response arrival time in the historical samples in the operation envelope update process to eliminate interference from invalid samples incorrectly included due to deviations in the original calibration time delay value on the statistical calculation of the operation envelope boundary.

[0104] During each update of the operating envelope, new samples are mixed with existing samples at a ratio not exceeding 30% to suppress the step shift in the subsequent disturbance identification benchmark caused by excessively large single update amplitudes. When the change amplitude of the upper and lower bounds of the operating envelope for the same process section does not exceed 0.5% of the full scale of the parameter after two consecutive concentrated updates, the parameter update is considered to have converged, and the subsequent update cycle can be extended to twice the original cycle until the next envelope change exceeds the convergence judgment range, at which point the original cycle is restored. In the event of seasonal changes, significant changes in raw material air conditions, or after a major overhaul, a new round of calibration sample set is established separately and is not mixed with existing samples for calculation. The new round of calibration sample set is accumulated in each load zone interval by no less than [number missing]. Only after 15 sets of valid samples can the original operating envelope and time delay parameters be officially replaced. During the period before the new sample set reaches the minimum data volume requirement mentioned above, the previous calibration parameters will still be used for operation, while new samples will be continuously collected and accumulated. After the above write-back and update, the process propagation time delay matrix, operating envelope library, segmented state anchor point set and visualization priority parameters are summarized to form the continuously corrected parameter library. The continuously corrected parameter library serves as the parameter reading source for each step of segmented calibration, time delay calibration, expected response deduction, margin calculation and verification priority generation in the next monitoring cycle, and continues to be effective in place of the initial calibration parameters until it is replaced by the new round of correction results after the next centralized update is completed.

[0105] The aforementioned layered visualization management interface uses the process topology diagram as its core display carrier. The anomaly identification results are superimposed and marked onto the corresponding process segment nodes using node color-coding and the hierarchical latent risk windows are marked as warning rings. A verification priority score is calculated based on a weighted average of three normalized indicators: risk urgency, propagation coverage, and data reliability. The scores are then sorted from highest to lowest to generate a four-layered structured visualization interface covering the disturbance source layer, propagation path layer, risk window layer, and job task card layer. This interface simultaneously presents the current status of each process segment, the disturbance propagation path, the urgency of latent risks, and the job verification and handling order to operators, enabling them to obtain a complete, causally related set of handling information within a single interface.

[0106] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for full-process data monitoring and visualization management of air separation process, characterized in that... ,include: The monitoring signals of each process segment of the entire air separation process are collected and preprocessed to obtain a unified time series dataset; the unified time series dataset is segmented and calibrated according to each process segment to obtain a segmented state anchor point set and a running envelope library. Based on the segmented state anchor point set and the running envelope library, the load band execution time delay calibration is performed according to the inter-segment coupling relationship to obtain the process propagation time delay matrix; using the process propagation time delay matrix as the path basis, the upstream disturbance is extrapolated to the expected response of each downstream process segment, compared with the measured response and the consistency is calculated to obtain the anomaly identification result; Using the margin parameters of each process segment in the aforementioned operating envelope library as boundaries, and combining real-time margins and change rates to calculate the edge contact time, a hierarchical latent risk window is obtained. The anomaly identification results, the process propagation time delay matrix, and the hierarchical latent risk window are superimposed on the process topology diagram, and a verification priority order is generated through joint weighting to obtain a hierarchical visual management interface.

2. The method for full-process data monitoring and visualization management of air separation technology according to claim 1, characterized in that: The specific steps for collecting and preprocessing the monitoring signals are as follows: The system collects monitoring signals from each process stage of the entire air separation process and connects the collected signals to the distributed control system via fieldbus for unified aggregation and writing into a real-time database. The sampling period is set according to the signal type. After the signal acquisition is completed, preprocessing is performed, that is, the time deviation of each acquisition node is eliminated based on the main station clock of the control system. For short-term missing signals, if the continuous missing signals do not exceed a preset number of similar sampling periods, the previous value is held to fill the gaps. If the missing signals exceed this number, the reliability of the signal is downgraded.

3. The method for monitoring and visualizing the entire air separation process data according to claim 1, characterized in that: The specific steps for segmented calibration according to each process section are as follows: After obtaining a unified time series dataset, each monitoring signal is assigned to the corresponding process section according to the correspondence between the instrument tag number and the process section, thus obtaining a subset of historical time series signals. On this basis, steady-state operation segments are extracted using the signal change rate and duration as steady-state judgment conditions, and the steady-state anchor points are obtained by weighting and fusing the design operating parameters and the steady-state mean within the preset number of historical days according to preset weights. The load band is divided into low load, medium load and high load operating ranges. The upper and lower bounds of the normal offset are determined by the preset percentile to obtain the operating envelope. When the sample is insufficient, it is supplemented by linear interpolation of adjacent load bands. The steady-state anchor points are summarized to form a segmented state anchor point set, and the operating envelope is summarized to form an operating envelope library. The two are stored independently.

4. The method for monitoring and visualizing the entire air separation process data according to claim 1, characterized in that: The specific steps for performing time delay calibration based on the inter-segment coupling relationship and load band are as follows: The coupling relationship between each process section is identified based on the existence of actual medium flow or energy transfer, resulting in an effective coupling topology. Using this topology as a path constraint, disturbance response samples under different operating conditions are extracted from the historical database. The start time of the upstream disturbance and the arrival time of the downstream response are identified to obtain the inter-segment propagation time delay. The determination of the disturbance start time must simultaneously meet the dual conditions of offset amplitude and duration. After the sample extraction is completed, the time delay value and propagation weight of the effective samples within the same load band are independently statistically analyzed. The process propagation time delay matrix is ​​formed by summarizing the calibrated time delay value and propagation weight as matrix elements.

5. The method for monitoring and visualizing the entire air separation process data according to claim 4, characterized in that: The specific steps for summarizing and forming the process propagation time delay matrix also include: During the load band switching transition, the process propagation time delay matrix uses the relative position of the current load value between the center values ​​of adjacent load bands as the interpolation coefficient to linearly interpolate the calibration time delay value and the propagation weight respectively; and only when a certain process segment stably shows a propagation response in no less than a preset load band number threshold is it solidified and written into the process propagation time delay matrix.

6. The method for monitoring and visualizing the entire air separation process data according to claim 1, characterized in that: The specific steps for extrapolating upstream disturbances to the desired response and calculating consistency are as follows: After obtaining the process propagation time delay matrix, the measured offset of the upstream disturbance is deduced downstream step by step along the effective coupling topology, based on the calibration time delay value as the time offset and the propagation weight as the amplitude attenuation value. When multiple propagation edges converge, a weighted sum is performed to obtain the combined expected response offset value. When crossing intermediate process sections, the combined value is used as the starting offset for the next level to continue the deduction, thereby obtaining the expected response offset value for each downstream process section; at the same time, the measured response offset value is calculated using the steady-state anchor point of the load zone corresponding to the segmented state anchor point set as the zero-bias reference; after obtaining both, the consistency of the same source disturbance at time t is calculated using the following formula: , in, This represents the consistency of the same source perturbation at time t, and its value ranges from [0,1]. This represents the measured response offset value of the j-th monitored quantity at time t; This represents the expected response offset value of the j-th monitored quantity at time t, derived from the process propagation time delay matrix. This represents the confidence weight of the j-th monitoring quantity, which is assigned based on the data quality label of the corresponding signal. The weight of the signal with a confidence downgrade is reduced accordingly. To prevent tiny positive numbers with a denominator of zero, t is the time series index of the monitoring signal sampling, representing the sampling time, and j is the index of the number of the monitoring quantity in each downstream process section, representing the monitoring quantity participating in the consistency calculation of the same source disturbance.

7. The method for monitoring and visualizing the entire air separation process data according to claim 6, characterized in that: The criteria for determining the anomaly identification results are as follows: After obtaining the consistency of the same source disturbance, it is classified and judged according to the judgment criteria to output the anomaly identification result. Specifically, when the consistency of the same source disturbance is not lower than the first threshold, it is judged as flexible operation same source fluctuation; when it is between the second threshold and the first threshold, it is judged as a mixed state to be observed. When the value is below the second threshold, and at least a preset monitoring threshold number of monitoring values ​​within the same process section deviates from the same direction of the operating envelope boundary within at least a preset period threshold number of sampling periods, it is determined that the internal anomaly takes priority.

8. The method for monitoring and visualizing data throughout the air separation process according to claim 1, characterized in that: The specific steps for calculating the contact time by combining real-time margin and rate of change are as follows: Using the margin parameters of each process segment in the running envelope library as boundaries, the boundaries of each margin parameter are determined based on the limit offset of the running envelope of the corresponding load band, and each load running interval is independently valued. The current collected values ​​of the margin parameters of each process segment are read in real time from the unified time series dataset. The original difference is obtained by subtracting the corresponding margin parameter boundary from the current collected value. When the original difference is positive, it is taken as the real-time margin. When it is zero or negative, it is marked as the risk boundary has been reached. The rate of change is calculated within the continuous sampling window. When the rate of change is negative, the expected contact time is obtained by dividing the real-time margin by the absolute value of the rate of change.

9. The method for monitoring and visualizing the entire air separation process data according to claim 8, characterized in that: The specific steps for the graded labeling of the expected contact time are as follows: After obtaining the expected contact time, the rate of change is continuously updated in a sliding window manner, so that the expected contact time is recalculated in each sampling period. When the absolute value of the rate of change is less than the preset zero-prevention lower limit, the lower limit is used instead of the calculation. Based on the expected contact time, it is divided into four levels: red urgent window, orange warning window, yellow attention window and normal monitoring status. The expected contact time of each process section and the corresponding level are summarized to form a hierarchical hidden risk window.

10. The method for monitoring and visualizing data throughout the air separation process according to claim 1, characterized in that: The specific steps for obtaining the hierarchical visual management interface by generating a joint weighted verification priority order are as follows: After obtaining the anomaly identification results and hierarchical hidden risk windows, a process topology base map is constructed with each process segment as a node and the effective propagation edge as a directed connection. Above this, the anomaly identification results are marked with node color codes, and the hierarchical hidden risk windows are marked with warning rings. In the case of cross-state, the one with the higher risk level is given priority. Then, the priority score for verification is calculated by weighting three normalized indicators: risk urgency, propagation coverage and data credibility. The score is sorted and output to the job task card. The suggested verification operation is automatically generated by the system based on the pre-registered verification rule table with the process section name and anomaly type as the search key, resulting in a hierarchical visual management interface.

Images