Analyzer application method based on analysis of process requirements for a spatially divided apparatus

Abstract

The application discloses an analyzer application method based on process requirements of air separation devices, relates to the technical field of analyzer application, and obtains pressure fluctuation data and precision data of different gas sampling, determines the installation positions of different types of analyzers through fuzzy reasoning, establishes a correlation model for the analyzers with the determined installation positions, acquires acoustic characteristics in the air separation device by using an acoustic sensor array when a neighboring analyzer triggers a wake-up requirement, constructs a prediction model in combination with the acoustic characteristics and the correlation model, judges whether the current analyzer needs to be woken up through the prediction model, identifies the change trend of the gas characteristics collected by the woken-up analyzer by using time series, judges whether the sampling parameters of the woken-up analyzer need to be adjusted through the change trend, and then outputs the sampling parameters of the woken-up analyzer based on a preset parameter adjustment mechanism. The method realizes data-driven accurate positioning, intelligent linkage and dynamic adjustment, and improves the detection efficiency and operation stability of the air separation device.

Description

Technical Field

[0001] This invention relates to the field of analyzer application technology, specifically to analyzer application methods based on the process requirements of air separation units. Background Technology

[0002] Air separation units are industrial equipment used to separate various components of gas in the air to produce industrial gases such as oxygen and nitrogen. They are widely used in industries such as metallurgy, petrochemicals, coal chemicals, glass, semiconductors, aerospace, and nuclear applications. Due to the continuous deterioration of the overall environment, the number of harmful components in the air that endanger the safety of air separation equipment is increasing, such as hydrocarbons and nitrogen oxides. Among these, acetylene and nitrous oxide are currently recognized as the most dangerous harmful components in the industry. These harmful components, once they enter the air separation unit and accumulate over a period of time, can cause an explosion after reaching a certain critical value, resulting in huge personal and property losses. To ensure the safe operation of the unit, various online analyzers are required, such as online carbon dioxide analyzers and online total hydrocarbon analyzers, to analyze trace amounts of hydrocarbons, carbon dioxide, and nitrous oxide in the air separation unit, monitor changes in the content of explosive components, and thus provide important reference data for the safe operation of the air separation unit.

[0003] Existing technologies rely on experience to select analyzer locations without combining pressure fluctuations with sampling accuracy for quantitative analysis; furthermore, data from neighboring analyzers lack effective correlation, making it impossible to predict detection needs in advance, resulting in wasted resources or detection delays; the sampling frequency and measurement range are fixed and cannot be dynamically adjusted according to changes in gas characteristics, affecting detection accuracy, and there is a lack of non-invasive monitoring methods based on acoustic characteristics.

[0004] Therefore, in order to address the above issues, there is an urgent need for analyzer application methods based on the process requirements of air separation units. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an analyzer application method based on the process requirements of air separation units, solving the problems of blind installation, poor coordination, rigid parameters, and insufficient adaptability to operating conditions of existing air separation unit analyzers.

[0006] To achieve the above objectives, the present invention provides the following technical solution: an analyzer application method based on the process requirements of an air separation unit, comprising the following steps: Step S1, acquiring pressure fluctuation data at different parts of the air separation unit during the sampling process, as well as accuracy data for sampling different gases, and outputting installation scores for different types of analyzers at different parts of the air separation unit through fuzzy inference to determine the installation positions of different types of analyzers; Step S2, for analyzers with determined installation positions, identifying the correlation between the gas characteristics to be collected and the gas characteristics collected by neighboring analyzers, and establishing a correlation model; Step S3, when a neighboring analyzer triggers a wake-up request, acquiring acoustic characteristics within the air separation unit using an acoustic sensor array, constructing a prediction model by combining the acoustic characteristics and the correlation model, and determining whether the current analyzer needs to be woken up using the prediction model; Step S4, identifying the changing trend of the gas characteristics collected by the woken-up analyzer using time series analysis, determining whether the sampling parameters of the woken-up analyzer need to be adjusted based on the changing trend, and then outputting the sampling parameters of the woken-up analyzer based on a preset parameter adjustment mechanism.

[0007] Further, step S1 is specifically analyzed as follows: pressure fluctuation data of different parts of the air separation unit during the sampling process is obtained through a pressure sensor array, and the accuracy data of different gas sampling is obtained in combination with the process requirement database. Then, the pressure fluctuation data and the accuracy data of different gas sampling are divided into different fuzzy sets, and fuzzy rules are established. The fuzzy rules take the pressure fluctuation data and the accuracy data of different gas sampling as inputs and output the fuzzy set of installation scores of different types of analyzers at each part. The fuzzy set of installation scores of different types of analyzers at each part is defuzzified to obtain the installation score of different types of analyzers at each part. The air separation unit part with the highest installation score is recorded as the installation position of different types of analyzers.

[0008] Furthermore, the correlation model is specifically established and analyzed as follows: the direction and intensity of the influence between the gas characteristics collected by each analyzer are determined by Granger causality, a preliminary correlation network is established, and a condition triggering rule base is established. The parameter weights of the correlation model are dynamically adjusted according to the air separation unit load rate and product purity.

[0009] Furthermore, the specific analysis of constructing a prediction model by combining acoustic features and an association model is as follows: training samples are extracted based on historical data, and the features of the training samples include trigger parameters based on the association model and acoustic features; random forest is used as the basis for constructing the prediction model, random forest parameters are set, the performance of the prediction model is optimized through cross-validation, and the wake-up probability output by the prediction model is probabilistically calibrated so that the error between the wake-up probability and the actual wake-up situation is lower than the error threshold; the prediction model is deployed, and the wake-up probability of the current analyzer is output; when the wake-up probability exceeds the wake-up probability threshold, the current analyzer is woken up.

[0010] Furthermore, the triggering parameter based on the association model is specifically represented as follows: the association model outputs the influence intensity between the gas characteristics collected by the neighboring analyzer and the current analyzer, and the influence intensity threshold is stored in the conditional triggering rule base. When the influence intensity between the gas characteristics collected by the neighboring analyzer and the current analyzer exceeds the influence intensity threshold, the difference between the influence intensity and the influence intensity threshold is obtained as the triggering parameter of the association model.

[0011] Further, step S4 is specifically analyzed as follows: the changing trend of the gas characteristics collected by the awakened analyzer is identified using time series analysis; the extreme point detection and slope within the sliding window are used to determine whether the sampling parameters need to be adjusted; if the sampling parameters need to be adjusted, the sampling parameters of the awakened analyzer are output using a parameter adjustment mechanism, wherein the sampling parameters include the sampling frequency and the sampling range; the parameter adjustment mechanism is specifically: the influence correlation between the changing trend of the collected gas characteristics identified based on historical data and the process requirements of the air separation unit and the adjustment ratio of the sampling frequency and the sampling range, respectively.

[0012] Furthermore, the specific analysis of whether the sampling parameters need to be adjusted by detecting extreme points within the sliding window and determining the slope is as follows: set the size of the sliding window, perform first-order difference processing on the gas characteristics within the sliding window to obtain a rate of change sequence; detect the local maxima and minima of the rate of change sequence; if the number of extreme points is greater than or equal to 2 and the difference between the local maxima and minima exceeds a preset difference threshold, determine that the changing trend of the collected gas characteristics is a significant changing trend; and output the changing trend type using slope analysis, wherein the changing trend type includes rising, falling, and fluctuating.

[0013] Furthermore, it also includes parameter optimization verification using a digital twin model. Specific steps include: constructing a digital twin model of the air separation unit, integrating real-time analyzer data and equipment status data. The real-time analyzer data includes collected gas characteristics and sampling parameters, while the equipment status data includes equipment temperature and pressure. The parameter adjustment effect is simulated in the digital twin, and the deviation between the actual air separation unit and the digital twin model is compared. If the deviation exceeds a deviation threshold, parameter correction is triggered. The parameter correction includes adjusting the gas diffusion coefficient and equipment response delay. The effect of the corrected parameter adjustment is verified through the digital twin model. The critical value of the corresponding parameter when the deviation between the actual air separation unit and the digital twin model is lower than the deviation threshold is used as the parameter adjustment scheme and sent to the actual analyzer for execution, forming a closed-loop parameter optimization.

[0014] The present invention has the following beneficial effects:

[0015] This analyzer application method based on the process requirements of air separation units overcomes the limitations of traditional single-index selection through a multi-dimensional fusion-based installation location optimization method. By combining fuzzy reasoning and integer programming, it achieves scientific decision-making for installation locations. Dynamic correlation modeling technology solves the problem of poor adaptability of fixed correlation models, and improves the collaborative efficiency of the analyzer through process status identification and anomaly propagation prediction. The acoustic feature-assisted wake-up mechanism creatively transforms equipment operating noise into useful monitoring signals, which has higher sensitivity and reliability compared with traditional timed wake-up or single-parameter triggering. The adaptive parameter adjustment method achieves dynamic optimization of sampling parameters through reinforcement learning, reducing system energy consumption while ensuring measurement accuracy. Digital twin-assisted decision-making deeply integrates virtual simulation with actual monitoring, enabling early fault detection and accurate prediction, and providing technical support for preventive maintenance. The self-optimization capability enables it to continuously adapt to process changes and equipment aging, extending the effective service life of the system and reducing operation and maintenance costs.

[0016] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0017] Figure 1 This is a flowchart of the analyzer application method based on the process requirements of an air separation unit according to the present invention. Detailed Implementation

[0018] This application embodiment achieves data-driven precise positioning, intelligent linkage, and dynamic adjustment through an analyzer application method based on the process requirements of the air separation unit, thereby improving the detection efficiency and operational stability of the air separation unit.

[0019] The overall approach of this application is as follows: quantify the installation location of the analyzer based on fuzzy inference; mine the value of nearby detection data using an association model; achieve intelligent wake-up of the analyzer through acoustic features and prediction models; and combine time series to dynamically adjust sampling parameters, forming a closed loop of "positioning-linkage-wake-up-adjustment".

[0020] Please see Figure 1This invention provides a technical solution: an analyzer application method based on the process requirements of an air separation unit, comprising the following steps: Step S1, acquiring pressure fluctuation data of different parts of the air separation unit during the sampling process, and accuracy data for sampling different gases, and outputting installation scores of different types of analyzers at different parts of the air separation unit through fuzzy inference to determine the installation location of different types of analyzers; Step S2, for the analyzer with the determined installation location, identifying the correlation between the gas characteristics to be collected by it and the gas characteristics collected by neighboring analyzers, and establishing a correlation model; Step S3, when a neighboring analyzer triggers a wake-up request, using an acoustic sensor array to acquire acoustic characteristics within the air separation unit, combining the acoustic characteristics with the correlation model to construct a prediction model, and using the prediction model to determine whether the current analyzer needs to be woken up; Step S4, using time series to identify the changing trend of the gas characteristics collected by the woken-up analyzer, determining whether the sampling parameters of the woken-up analyzer need to be adjusted based on the changing trend, and then outputting the sampling parameters of the woken-up analyzer based on a preset parameter adjustment mechanism.

[0021] Specifically, step S1 is analyzed as follows: pressure fluctuation data of different parts of the air separation unit during the sampling process is obtained through the pressure sensor array, and the accuracy data of different gas sampling is obtained in combination with the process requirement database. Then, the pressure fluctuation data and the accuracy data of different gas sampling are divided into different fuzzy sets, and fuzzy rules are established. The fuzzy rules take the pressure fluctuation data and the accuracy data of different gas sampling as inputs and output the fuzzy set of installation scores of different types of analyzers at each part. The fuzzy set of installation scores of different types of analyzers at each part is defuzzified to obtain the installation score of different types of analyzers at each part. The air separation unit part with the highest installation score is recorded as the installation position of different types of analyzers.

[0022] In this implementation plan, pressure fluctuation data includes pressure fluctuation amplitude, pressure change frequency, and pressure gradient distribution. Specifically: pressure fluctuation amplitude represents the maximum deviation of the pressure value from the average pressure per unit time, reflecting airflow stability (e.g., a sudden pressure surge in a distillation column may lead to component turbulence); pressure change frequency represents the number of pressure fluctuations per unit time, reflecting the dynamic characteristics of the process (e.g., the pressure change frequency is high during compressor start-up and shutdown); and pressure gradient distribution represents the pressure difference at different locations in the air separation unit, used to determine the gas flow direction and diffusion characteristics (e.g., the pressure gradient at the inlet and outlet of the heat exchanger affects the representativeness of the sampling). The pressure fluctuation data is acquired as follows: pressure sensor arrays are deployed at key locations in the air separation unit (e.g., compressor outlet, distillation column trays, heat exchanger inlet and outlet). Pressure signals are collected in real time using analog signal conditioning circuits (amplification and filtering), with a sampling frequency of no less than 10Hz. Analog-to-digital converters (ADCs) convert the analog pressure signals into digital signals, which are then transmitted to the data processing module via industrial Ethernet or fieldbus (e.g., PROFINET, Modbus). A Kalman filter algorithm is used to denoise the raw pressure data, eliminating the influence of noise such as pipeline vibration and electromagnetic interference on the calculation of fluctuation parameters.

[0023] Accuracy data for different gas sampling includes target gas concentration deviation, component response time, and cross-interference coefficient. Among them: target gas concentration deviation represents the absolute value of the difference between the actual sampled concentration and the concentration required by the process, reflecting the sampling's ability to capture the target gas (e.g., the detection deviation of impurity content in high-purity oxygen sampling); component response time represents the analyzer's reaction speed to changes in gas composition, reflecting the efficiency of sampling pipelines and pretreatment (e.g., excessive lag time leading to control lag in trace water detection); cross-interference coefficient represents the degree of interference of non-target gases on the detection results, used to evaluate the sampling's relevance (e.g., the interference of nitrogen mixed in argon on thermal conductivity analyzers). The methods for obtaining accurate sampling data for different gases are as follows: extract the theoretical concentration range, allowable deviation threshold, and interfering gas type of the target gas in each part from the process design document of the air separation unit, establish a standardized parameter table, collect the detection data of the analyzer under typical operating conditions, calculate the target gas concentration deviation and component response time, obtain the sampling accuracy characteristic curve of each part by fitting with the least squares method, or simulate the operating conditions of the air separation unit in a laboratory environment, inject different proportions of interfering gas, measure the cross-interference coefficient, form an interference matrix and store it in the process requirement database.

[0024] The specific analysis of the fuzzy sets of installation scores for different types of analyzers at various locations is as follows: Pressure fluctuation data and the accuracy data of different gas sampling are defined as input variables and divided into different fuzzy sets, for example: "Low", "Medium", and "High" for pressure fluctuation data, and "Low", "Medium", and "High" for the accuracy data of different gas sampling. The installation scores of different types of analyzers at various locations are defined as output variables and divided into different fuzzy sets, for example: "Low", "Medium", and "High" for the installation scores of different types of analyzers at various locations. Based on the direct proportionality between the accuracy data of different gas sampling and the installation scores of different types of analyzers at various locations, and the inverse proportionality between the pressure fluctuation data and the installation scores of different types of analyzers at various locations, fuzzy rules are established. An example of the fuzzy rules is as follows:

[0025] If we label the accuracy data of different gas sampling as Q, the pressure fluctuation data as G, and the installation scores of different types of analyzers at various locations as R, then we can define:

[0026] Rule 1: IF (Q is High) AND (G is Short) THEN (R is High)

[0027] Rule 2: IF (Q is Short) AND (G is High) THEN (R is Low) ...

[0028] It should be noted that the division of fuzzy sets can be adjusted according to the actual situation. For example, although this embodiment uses three fuzzy sets as an example, in reality, pressure fluctuation data, accuracy data of different gas sampling, and installation scores of different types of analyzers in various parts can be divided into more than three sets to facilitate better installation effects of different types of analyzers in various parts.

[0029] The judgment of high, medium, and low pressure fluctuation data, and high, medium, and low accuracy data of different gas sampling data, can be made by setting thresholds according to the actual situation, which will not be elaborated here.

[0030] By quantifying pressure fluctuations and sampling accuracy into an installation score through fuzzy reasoning, the subjectivity of traditional empirical methods is avoided, ensuring that the analyzer installation location matches the process requirements and reducing detection errors caused by improper location. Combining pressure fluctuation and sampling accuracy data covers the fluid characteristics and detection requirements of different parts of the air separation unit, enhancing the adaptability of the analysis system to complex operating conditions such as high pressure, low pressure, and impurity fluctuations. Through explicit fuzzy set partitioning and defuzzification steps, a reusable algorithm framework is formed, facilitating the rapid deployment of analyzer networks in different air separation units and improving system integration efficiency.

[0031] Specifically, the correlation model is established and analyzed as follows: the direction and intensity of the influence between the gas characteristics collected by each analyzer are determined by Granger causality, a preliminary correlation network is established, and a condition triggering rule base is established. The parameter weights of the correlation model are dynamically adjusted according to the load rate of the air separation unit and the purity of the product.

[0032] In this implementation plan, the specific steps of Granger causality analysis are as follows: The gas characteristic time series collected by each analyzer are denoised and normalized to eliminate the influence of dimensional differences and random noise; the ADF test or KPSS test is used to determine whether the time series is stationary, and non-stationary series are differencing until stationary; information criteria (such as AIC, BIC) are used to determine the optimal lag order for Granger causality testing to ensure a balance between model complexity and fitting effect; for each pair of gas characteristic time series, a bidirectional Granger causality test is performed to determine whether a unidirectional or bidirectional causal relationship exists; the strength level of the causal relationship is determined based on the magnitude of the F statistic, and a correlation network topology containing the direction and strength of influence is generated; the significance of the causal relationship is verified by Monte Carlo simulation or Bootstrap method, and spurious causal relationships are eliminated.

[0033] The specific steps for constructing the condition trigger rule base are as follows: Based on historical operating data and process safety standards, set initial impact intensity thresholds for each correlation; classify and store the thresholds according to process stage (such as startup, steady state, shutdown) and operating condition type (such as high load, low purity); assign priority to each trigger rule to ensure that it is executed in order of importance when multiple rules are satisfied simultaneously; regularly collect actual operating data and optimize the threshold parameters through machine learning algorithms to improve the accuracy of the trigger rules; set up a fault tolerance mechanism to automatically switch to safe mode and record anomaly logs when the trigger parameters exceed the preset range; and establish a graphical interface for the rule base to support manual intervention and rule editing, realizing human-machine collaborative optimization.

[0034] The specific steps for dynamically adjusting parameter weights are as follows: Real-time acquisition of air separation unit load rate and product purity data; determination of the current operating condition using a pattern recognition algorithm; pre-definition of an operating condition-weight mapping table; matching the corresponding parameter weight combination from the mapping table based on the current operating condition; smooth transition of parameter weights using a weighted average method during operating condition switching to avoid drastic model fluctuations; comparison of the correlation model's prediction results with actual detection data; dynamic fine-tuning of weight parameters based on the error magnitude; and the ability to set a manual intervention interface, allowing process experts to manually adjust weight parameters based on experience. An example of a conditional trigger rule library is: when a neighboring analyzer detects that the argon content exceeds a preset argon content threshold, the oxygen analyzer is triggered to enter a standby state.

[0035] Granger causality analysis is used to determine the causal relationships between gas characteristics, avoiding the spurious correlation problem caused by traditional methods that rely solely on correlation analysis, thus improving the reliability of model predictions. A condition trigger rule base is established to enable intelligent wake-up of the analyzer, reducing detection and lowering system energy consumption and equipment wear. Parameter weights are dynamically adjusted according to load rate and product purity, enabling the correlation model to adapt to different operating conditions of the air separation unit and improving system robustness.

[0036] Specifically, the prediction model constructed by combining acoustic features and correlation models is analyzed as follows: training samples are extracted based on historical data. The features of the training samples include trigger parameters based on the correlation model and acoustic features; random forest is used as the basis for constructing the prediction model. Random forest parameters are set, and the performance of the prediction model is optimized through cross-validation. The wake-up probability output by the prediction model is calibrated so that the error between the wake-up probability and the actual wake-up situation is lower than the error threshold; the prediction model is deployed, and the wake-up probability of the current analyzer is obtained. When the wake-up probability exceeds the wake-up probability threshold, the current analyzer is woken up.

[0037] In this implementation scheme, the triggering parameters based on the association model are specifically represented as follows: the association model outputs the influence intensity between the gas characteristics collected by the neighboring analyzer and the current analyzer, and the influence intensity threshold is stored in the conditional triggering rule base. When the influence intensity between the gas characteristics collected by the neighboring analyzer and the current analyzer exceeds the influence intensity threshold, the difference between the influence intensity and the influence intensity threshold is obtained as the triggering parameter of the association model.

[0038] Acoustic characteristics include frequency distribution characteristics, time-domain characteristics, spatial characteristics, and statistical characteristics. Among them, frequency distribution characteristics include dominant frequency components, harmonic content, and frequency band energy distribution, reflecting the gas flow state and equipment operating state; time-domain characteristics represent sound pressure level, sound intensity, sound signal duration, rise / fall time, etc., reflecting the intensity and rate of change of acoustic events; spatial characteristics represent the phase difference and sound pressure gradient between channels of the acoustic sensor array, used to locate the sound source and determine the sound wave propagation path; statistical characteristics represent the mean, variance, kurtosis, skewness, etc. of the acoustic signal, describing the statistical characteristics and non-stationarity of the signal. The acoustic feature acquisition steps are as follows: At least four high-sensitivity acoustic sensors are arranged in a ring around key parts of the air separation unit (such as pipe bends, valves, and tower connections) to form a spatial array; acoustic signals from each sensor are synchronously acquired at a sampling frequency of at least 20 kHz for at least 10 seconds per acquisition; bandpass filtering is applied to the raw acoustic signals to remove environmental noise and interference from the equipment's inherent frequencies; time-frequency analysis is used to extract the time-domain, frequency-domain, and spatial characteristic parameters of the acoustic signals; the subset of acoustic features that contribute most to the wake-up decision is selected through correlation analysis or feature importance ranking; and the extracted acoustic features are normalized to eliminate the influence of dimensions.

[0039] Acoustic characteristics can reflect the fluid dynamics (such as airflow disturbance and eddy formation) and mechanical conditions (such as valve vibration and pipe friction) within the air separation unit. These microscopic changes often appear earlier than changes in gas composition and can serve as early warning indicators, enabling proactive decision-making to wake up the analyzer.

[0040] The steps for predicting the wake-up probability output by the model are as follows: Extract the associated model trigger parameters and corresponding acoustic features from historical data, label whether an actual wake-up occurs, and form a training dataset; construct the prediction model using the random forest algorithm, setting initial parameters such as the number of decision trees, maximum depth, and minimum number of sample splits; divide the training data into multiple subsets, use K-fold cross-validation to evaluate model performance, and adjust parameters to optimize metrics such as accuracy and recall; calibrate the original probability output by the model using Platt scaling or Isotonic regression to ensure that the calibrated probability matches the actual wake-up frequency; verify the error rate of the calibrated model using an independent test dataset to ensure that the error is below a preset threshold; input the associated model trigger parameters and acoustic features at the current moment into the calibrated prediction model to output the wake-up probability.

[0041] The wake-up probability threshold is a critical value used to determine whether to wake up the analyzer. Values ​​above this threshold indicate the analyzer needs to be woken up for detection; values ​​below this threshold maintain standby mode to conserve resources. The threshold is obtained by analyzing the false alarm and false negative rates under different wake-up probabilities in historical operating data and selecting the probability value that minimizes their sum as the initial threshold. Alternatively, the energy cost of waking up the analyzer, the losses caused by detection delays, and the maintenance costs due to false alarms can be considered to establish a cost function to solve for the optimal threshold. Furthermore, the threshold can be dynamically adjusted based on operating parameters such as the air separation unit's load rate and product purity requirements through a preset mapping relationship. Process experts can also manually correct the threshold based on experience, forming a human-machine collaborative threshold optimization mechanism.

[0042] The impact intensity threshold can be obtained in the following ways: based on the design parameters and operation manual of the air separation unit, determine the theoretical correlation intensity threshold between various gas characteristics; or analyze the correlation intensity distribution under normal and abnormal operating conditions in historical operating data, and take the upper limit of the 95% confidence interval under normal operating conditions as the initial threshold; or observe the changes in system performance by changing the threshold parameters, and select the threshold point with the highest sensitivity to wake-up decision; or use online learning algorithms to continuously optimize the impact intensity threshold based on the latest operating data to adapt to changes such as unit aging and process adjustment.

[0043] By combining acoustic features with associated model trigger parameters, the limitations of single gas detection are overcome, and the early detection capability of potential changes in air separation units is improved. The prediction model is optimized through cross-validation and probabilistic calibration to ensure the reliability of wake-up decisions and reduce false alarm and false negative rates. The wake-up threshold is dynamically adjusted based on real-time acoustic features, so that the analyzer can maintain the best response state under different operating conditions and reduce energy waste.

[0044] Specifically, step S4 is analyzed as follows: the changing trend of the gas characteristics collected by the awakened analyzer is identified using time series analysis; the extreme point detection and slope within the sliding window are used to determine whether the sampling parameters need to be adjusted; if the sampling parameters need to be adjusted, the sampling parameters of the awakened analyzer are output using the parameter adjustment mechanism. The sampling parameters include the sampling frequency and the sampling range. The parameter adjustment mechanism is specifically: the influence correlation between the changing trend of the collected gas characteristics identified based on historical data and the process requirements of the air separation unit and the adjustment ratio of the sampling frequency and the sampling range, respectively.

[0045] In this implementation scheme, the specific analysis of whether the sampling parameters need to be adjusted by detecting extreme points within the sliding window and determining the slope is as follows: Set the size of the sliding window, perform first-order difference processing on the gas characteristics within the sliding window to obtain a rate of change sequence; detect the local maxima and minima of the rate of change sequence; if the number of extreme points is greater than or equal to 2 and the difference between the local maxima and minima exceeds a preset difference threshold, determine that the changing trend of the collected gas characteristics is a significant changing trend; use slope analysis to output the changing trend type, which includes rising, falling, and fluctuating trends.

[0046] The steps for identifying trends using time series analysis are as follows: 1. Receive gas characteristic time series data collected by the awakened analyzer. Clean the data, removing outliers and missing values, and normalize the data dimensions. 2. Based on the process characteristics of the air separation unit and historical data fluctuations, set an appropriate sliding window size to divide the time series data into multiple overlapping or non-overlapping window segments. 3. Perform first-order difference calculations on the gas characteristic data within each sliding window to obtain a rate of change sequence reflecting the rate of data change. 4. Traverse the rate of change sequence, detecting local maxima and minima, and recording the positions and values ​​of the extreme points. 5. Count the number of detected extreme points and calculate the difference between adjacent extreme points. When the number of extreme points is greater than or equal to 2 and the difference between local maxima and minima exceeds a preset difference threshold, the trend of gas characteristic changes within that window is determined to be a significant trend; otherwise, the trend is considered insignificant. 6. Summarize and analyze the judgment results of all sliding windows. If multiple windows are determined to have significant trends, the current gas characteristic is determined to show a significant change; otherwise, the trend is determined to be stable.

[0047] Sampling frequency refers to the number of times an analyzer collects gas characteristics per unit time. When gas characteristics change slowly, reducing the sampling frequency can reduce the equipment's operating load and data processing volume, saving energy. When gas characteristics change drastically, such as during the start-up of an air separation unit or the load adjustment phase, increasing the sampling frequency can capture data changes more intensively, ensuring that key information is not missed and providing timely and accurate data support for process control.

[0048] The sampling range represents the range of gas characteristic parameters that the analyzer can detect. Adjusting the sampling range allows for the reasonable setting of the detection range based on the actual changes in gas characteristics. When gas characteristic values ​​are close to or exceed the original sampling range, expanding the sampling range can prevent data overflow or inaccurate measurements. When gas characteristic values ​​are relatively stable and concentrated in a small range, narrowing the sampling range can improve detection accuracy and make the measurement results more precise.

[0049] The steps for constructing the parameter adjustment mechanism are as follows: First, collect gas characteristic change trend data from the awakened analyzer under different operating conditions of the air separation unit, along with adjustment records of sampling frequency and sampling range under corresponding operating conditions, and establish a historical database. Second, combine the air separation unit's design documents, operating procedures, and process expert experience to clarify the accuracy, frequency, and other requirements for gas characteristic detection at different process stages and operating conditions, and compile a process requirement parameter table. Third, use data mining algorithms to analyze the potential correlation between gas characteristic change trends (increasing, decreasing, fluctuating) and the sampling frequency adjustment ratio and sampling range adjustment ratio in historical data, and determine the adjustment rules and range of sampling parameters under different change trends. Fourth, use historical data to simulate and verify the mined correlations and adjustment rules, and evaluate the rationality of the rules by comparing the accuracy and effectiveness of the detection data before and after adjustment. Fifth, optimize the correlations and rules based on the verification results until they meet the process requirements and detection accuracy requirements. Sixth, integrate the optimized correlations and adjustment rules into the parameter adjustment mechanism, enabling it to automatically calculate and output the corresponding sampling frequency adjustment ratio and sampling range adjustment ratio based on the real-time identified gas characteristic change trends.

[0050] The sliding window size is set as follows: Based on the process characteristics of the air separation unit and past operating experience, process experts or technicians set an initial sliding window size as a reference value for the initial stage of system operation. During system operation, the fluctuation and frequency of gas characteristic data are monitored in real time. When the data fluctuation is large and the frequency of change is high, the sliding window size is appropriately reduced to capture data changes more precisely. When the data fluctuation is small and the frequency of change is low, the sliding window size is increased to reduce the computational load and data processing pressure. Historical data can also be analyzed periodically to statistically analyze the accuracy and effectiveness indicators of time series analysis under different sliding window sizes (such as detection error rate and trend identification accuracy). Based on the analysis results, the optimal sliding window size is searched using optimization algorithms (such as genetic algorithms and particle swarm optimization algorithms), and the window size is continuously adjusted and optimized.

[0051] The steps for using slope analysis to output the trend type are as follows: Linearly fit the gas characteristic data within the sliding window and calculate the slope of the fitted line. This slope reflects the average trend of the gas characteristic data within the window. Based on the calculated slope value, the trend type is determined as follows: when the slope is greater than 0, the trend type is upward; when the slope is less than 0, the trend type is downward; when the slope is close to 0 and fluctuates within a certain error range, combined with the extreme point detection results, if there are multiple extreme points with small differences, the trend type is determined as fluctuating. A slope threshold range is set. When the slope exceeds the threshold range for a normal upward or downward trend, it is considered an abnormal situation. Further comprehensive judgment is made by combining other data analysis methods (such as data mutation detection and historical data comparison) to ensure the accuracy of the trend type judgment.

[0052] By employing time series analysis and sliding window detection, the system can keenly capture subtle changes in gas characteristics, determine the need for sampling parameter adjustments in real time, and ensure that the analyzer's detection data accurately reflects the operating status of the air separation unit, avoiding detection lag or data distortion caused by fixed parameters. The sampling frequency and range are dynamically adjusted based on the changing trends of gas characteristics. When gas characteristics are stable, the sampling frequency is reduced to minimize equipment wear and data redundancy; when changes are significant, the sampling frequency is increased and the sampling range is expanded to ensure complete acquisition of key data and achieve efficient resource utilization. The parameter adjustment mechanism establishes an impact correlation between historical data and process requirements, enabling the system to adapt to different operating conditions and process fluctuations in the air separation unit, improving the stability and reliability of the analyzer in complex environments, and ensuring the safe and efficient operation of the air separation unit.

[0053] Specifically, this also includes parameter optimization verification using a digital twin model. The specific steps include: constructing a digital twin model of the air separation unit, integrating real-time analyzer data and equipment status data. The real-time analyzer data includes collected gas characteristics and sampling parameters, while the equipment status data includes equipment temperature and pressure. The parameter adjustment effect is simulated in the digital twin, and the deviation between the actual air separation unit and the digital twin model is compared. If the deviation exceeds a deviation threshold, parameter correction is triggered. Parameter correction includes adjusting the gas diffusion coefficient and equipment response delay. The effect of the corrected parameter adjustment is verified through the digital twin model. The critical value of the corresponding parameter when the deviation between the actual air separation unit and the digital twin model is lower than the deviation threshold is used as the parameter adjustment scheme and sent to the actual analyzer for execution, forming a closed-loop parameter optimization.

[0054] In this implementation plan, the steps for constructing the digital twin model of the air separation unit are as follows: Based on the air separation unit design drawings and 3D scanning data, a 3D geometric model of the unit is constructed using professional modeling software, covering core equipment and connection structures such as air compressors, distillation columns, heat exchangers, and pipelines, ensuring that the model's geometric dimensions are consistent with the actual unit; material properties (such as density and thermal conductivity), thermodynamic parameters (specific heat capacity and latent heat of phase change), and fluid dynamic parameters (viscosity and diffusion coefficient) are assigned to each component in the model, with parameter values ​​referencing design documents and experimental test data; the automated control logic of the air separation unit is analyzed, and control algorithms and strategies such as temperature control, pressure regulation, and flow control are ported to the digital twin model to establish a virtual control module corresponding to the actual control system; a standardized data interface is designed to achieve data interaction with real-time analyzers, equipment sensors, and the control system, ensuring that the model can receive real-time detection data and control commands; after the model is built, historical operating data is input, and the model output results are compared with the historical data of the actual unit. By adjusting model parameters and correcting logical relationships, the model output error is brought within an acceptable range, completing model calibration.

[0055] Real-time analyzer data directly reflects the gas composition and detection status within the air separation unit, providing real-time process parameters for the digital twin model. This enables the model to simulate based on actual operating conditions, ensuring that the simulation results are closely related to the actual operating status. The real-time updated analyzer data is used to monitor process change trends. When abnormal fluctuations occur in gas characteristics, the model can promptly capture the changes and simulate the effects of different parameter adjustment schemes, providing data support for parameter optimization and achieving dynamic optimization of the air separation unit's operation.

[0056] Equipment status data reflects the operating condition of the equipment. Integrating this data allows digital twin models to monitor the health status of the equipment in real time, detect potential faults such as overheating and abnormal pressure in advance, provide early warning information for equipment maintenance, and avoid unplanned downtime. Equipment status directly affects the process performance of air separation units. For example, changes in heat exchanger temperature will affect gas heat exchange efficiency, and pressure fluctuations in compressors will change gas flow. Integrating equipment status data into the model allows the model to more realistically simulate the interaction between the equipment and the process, improving the accuracy of the simulation results.

[0057] The deviation threshold is a critical indicator that measures the difference between the actual air separation unit and its digital twin model. It is used to judge the degree of conformity between the model simulation results and the actual situation. When the deviation exceeds the threshold, it indicates that the model simulation effect is poor or the actual unit is operating abnormally, requiring parameter correction or adjustment of the operating status. The threshold can be obtained as follows: Based on the air separation unit's design standards, process requirements, and historical operating experience, domain experts set an initial deviation threshold, such as setting the gas concentration deviation threshold to ±2% of the actual value and the equipment pressure deviation threshold to ±5%. Alternatively, actual data and model simulation data under stable operating conditions of the air separation unit can be collected, and the deviation values ​​of the two in different parameter dimensions (gas composition, equipment pressure, temperature, etc.) can be calculated. A reasonable deviation threshold range can be determined through statistical analysis (such as calculating standard deviation and confidence intervals). A dynamic adjustment mechanism for the deviation threshold can also be established to monitor the deviation between the actual unit and the model in real time and the adjustment effect. When the deviation still frequently exceeds the threshold after multiple adjustments, the threshold range is automatically reduced; if the system is stable and the deviation is small after adjustment, the threshold range is appropriately expanded to balance the sensitivity of parameter adjustment and system stability.

[0058] The significance of adjusting the gas diffusion coefficient and equipment response delay is as follows: Regarding the gas diffusion coefficient, which affects the gas mixing and separation process within the air separation unit, adjusting this coefficient allows the digital twin model to more accurately simulate the diffusion behavior of gases in pipelines and towers, thereby optimizing key process indicators such as distillation efficiency and product purity, and improving the model's simulation accuracy. When the composition of the feed gas in the air separation unit changes or operating conditions are adjusted, the gas diffusion characteristics will change. By adjusting the gas diffusion coefficient, the model can quickly adapt to changes in operating conditions, providing reliable simulation results for process parameter optimization and ensuring stable operation of the unit under different operating conditions. Regarding the equipment response delay, after receiving control commands, the actual equipment exhibits a certain response delay due to factors such as mechanical inertia and signal transmission delay. Adjusting the equipment response delay parameter ensures that the equipment behavior in the digital twin model is consistent with the actual equipment, avoiding control deviations caused by differences between the model and actual responses, and improving the effectiveness of the control strategy. Furthermore, by simulating the control effects under different equipment response delays, the rationality of existing control strategies can be evaluated, providing a basis for improving control algorithms and adjusting control parameters, reducing problems such as control overshoot and oscillation, and achieving stable and efficient operation of the air separation unit.

[0059] In summary, this application has at least the following effects:

[0060] The analyzer installation location is determined by fuzzy reasoning, and dynamic wake-up and parameter adjustment are achieved by combining acoustic features with gas features, thereby improving detection accuracy and response speed. Data linkage between neighboring analyzers is achieved by using correlation models and prediction models, reducing redundant detection and optimizing the operating efficiency of the air separation unit. Sampling parameters are automatically adjusted based on time series analysis to minimize sampling errors and adapt to the complex operating conditions of the air separation process.

[0061] Those skilled in the art will understand that embodiments of the present invention can be provided as methods. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0062] This invention is described with reference to a flowchart of a method according to embodiments of the invention. It should be understood that the combination of each step in the flowchart can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing device, generate instructions for implementing the process. Figure 1 A device for a function specified in one or more processes.

[0063] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 The function specified in one or more processes.

[0064] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 Steps of a specified function in one or more processes.

[0065] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the invention.

[0066] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. An analyzer application method based on the process requirements of an air separation unit, characterized in that, Includes the following steps: Step S1: Obtain pressure fluctuation data of different parts of the air separation unit during the sampling process, as well as accuracy data for sampling different gases. Through fuzzy inference, output the installation score of different types of analyzers at different parts of the air separation unit to determine the installation location of different types of analyzers. Step S2: For the analyzer with a determined installation location, identify the correlation between the gas characteristics it needs to collect and the gas characteristics collected by nearby analyzers, and establish a correlation model. Step S3: When a nearby analyzer triggers a wake-up request, the acoustic features inside the air separation unit are obtained using an acoustic sensor array. A prediction model is constructed by combining the acoustic features with the correlation model. The prediction model is then used to determine whether the current analyzer needs to be woken up. Step S4: Use time series to identify the changing trend of gas characteristics collected by the awakened analyzer, determine whether the sampling parameters of the awakened analyzer need to be adjusted based on the changing trend, and then output the sampling parameters of the awakened analyzer based on the preset parameter adjustment mechanism. The specific analysis and establishment of the correlation model is as follows: The influence direction and intensity between gas characteristics collected by each analyzer are determined by Granger causality, a preliminary correlation network is established, and a condition triggering rule base is established. The parameter weights of the correlation model are dynamically adjusted according to the load rate of the air separation unit and the purity of the product. The triggering parameters based on the association model are specifically represented as follows: The correlation model outputs the influence strength between the gas characteristics collected by the neighboring analyzer and the current analyzer. The influence strength threshold is stored in the conditional trigger rule base. When the influence strength between the gas characteristics collected by the neighboring analyzer and the current analyzer exceeds the influence strength threshold, the difference between the influence strength and the influence strength threshold is used as the trigger parameter of the correlation model.

2. The analyzer application method based on the process requirements of an air separation unit according to claim 1, characterized in that, Step S1 is analyzed in detail as follows: Pressure fluctuation data of different parts of the air separation unit during the sampling process is acquired by a pressure sensor array, and the accuracy data of different gas sampling is obtained by combining the process requirement database. Then, the pressure fluctuation data and the accuracy data of different gas sampling are divided into different fuzzy sets, and fuzzy rules are established. The fuzzy rules take the pressure fluctuation data and the accuracy data of different gas sampling as inputs and output the fuzzy sets of installation scores of different types of analyzers at each part. The fuzzy sets of installation scores of different types of analyzers at each part are defuzzified to obtain the installation scores of different types of analyzers at each part. The air separation unit part with the highest installation score is recorded as the installation position of different types of analyzers.

3. The analyzer application method based on the process requirements of an air separation unit according to claim 1, characterized in that, The specific analysis of constructing a prediction model by combining acoustic features and correlation models is as follows: Training samples are extracted from historical data, and the features of the training samples include trigger parameters and acoustic features based on the correlation model. Based on random forest as the foundation for predictive model construction, random forest parameters are set, and the performance of the predictive model is optimized through cross-validation. The wake-up probability output by the predictive model is calibrated so that the error between the wake-up probability and the actual wake-up situation is lower than the error threshold. Deploy a predictive model and output the wake-up probability of the current analyzer. When the wake-up probability exceeds the wake-up probability threshold, wake up the current analyzer.

4. The analyzer application method based on the process requirements of an air separation unit according to claim 1, characterized in that, Step S4 is analyzed in detail as follows: The changing trend of gas characteristics collected by the woken-up analyzer is identified by time series analysis. The extreme point detection and slope within the sliding window determine whether the sampling parameters need to be adjusted. If the sampling parameters need to be adjusted, the sampling parameters of the woken-up analyzer are output using the parameter adjustment mechanism. The sampling parameters include sampling frequency and sampling range. The parameter adjustment mechanism is specifically defined as the correlation between the changing trends of the collected gas characteristics identified based on historical data and the process requirements of the air separation unit, and the adjustment ratios of the sampling frequency and sampling range.

5. The analyzer application method based on the process requirements of an air separation unit according to claim 4, characterized in that, The specific analysis of determining whether the sampling parameters need adjustment by detecting extreme points and judging the slope within the sliding window is as follows: Set the sliding window size, perform first-order difference processing on the gas characteristics within the sliding window, and obtain the rate of change sequence; The local maxima and minima of the rate of change sequence are detected. If the number of extreme points is greater than or equal to 2 and the difference between the local maxima and minima exceeds a preset difference threshold, the trend of change of the collected gas characteristics is determined to be a significant trend. The slope analysis outputs the trend type, which includes rising, falling, and fluctuating trends.

6. The analyzer application method based on the process requirements of an air separation unit according to claim 1, characterized in that, It also includes parameter optimization and verification using digital twin models, with specific steps including: Construct a digital twin model of the air separation unit, integrating real-time analyzer data and equipment status data. The real-time analyzer data includes the characteristics of the collected gas and sampling parameters, while the equipment status data includes equipment temperature and equipment pressure. The effect of parameter adjustment is simulated in the digital twin, and the deviation between the actual air separation unit and the digital twin model is compared. If the deviation exceeds the deviation threshold, parameter correction is triggered. The parameter correction includes adjusting the gas diffusion coefficient and the equipment response delay. The effect of the corrected parameter adjustment is verified by using a digital twin model. When the deviation between the actual air separation unit and the digital twin model is lower than the deviation threshold, the critical value of the corresponding parameter is used as the parameter adjustment scheme and sent to the actual analyzer for execution, thus forming a closed-loop parameter optimization.

Images