Intelligent early warning and safety interlocking control method for abnormal leakage of liquefied gas based on pattern recognition

Abstract

The invention relates to the technical field of industrial automation control and chemical safety monitoring, in particular to a liquefied gas abnormal leakage intelligent early warning and safety interlocking control method based on mode recognition. An ideal fluid dynamic reference model is constructed, and a theoretical leakage-free flow reference is calculated; a synthetic analysis step: injecting sampling bypass impedance parameters to generate theoretical sampling expected flow, and calculating a theoretical residual vector; a double-track difference calculation step: calculating a real-time flow and a real residual vector; a similarity judgment and control step: calculating morphological similarity; if yes, normal sampling is determined and interlocking is shielded; if the amplitude is smaller than the threshold value and exceeds the limit, abnormal leakage is judged and turn-off is triggered; according to the invention, the problem of false positive alarm caused by online sampling is effectively solved.

Description

Technical Field

[0001] This invention relates to the fields of industrial automation control and chemical safety monitoring technology, specifically to a method for intelligent early warning and safety interlock control of abnormal liquefied gas leaks based on pattern recognition. Background Technology

[0002] In the field of liquefied gas transmission process control, monitoring the sealing integrity of regulating valves and pipeline systems is crucial to ensuring production safety. In order to monitor the quality of the medium in real time, the system is usually equipped with online sampling equipment, and the branch line needs to be opened from time to time to sample the fluid. Existing monitoring solutions mainly rely on single flow or pressure threshold judgment logic, lacking the ability to deeply analyze the fluid dynamics under complex operating conditions. When performing online sampling operations, the sudden changes in flow and pressure fluctuations caused by bypass diversion are highly similar to real media leakage in terms of physical signal characteristics. Since traditional methods cannot effectively identify the essential differences between these two types of operating conditions, normal sampling disturbances are easily misjudged as abnormal leaks, resulting in false positive alarms and erroneously triggering the safety interlock system, leading to accidental valve closure and production interruption. In addition, if the alarm threshold is relaxed simply to avoid sampling false alarms, it will reduce the system's detection sensitivity to small, gradual leaks, masking potential safety hazards. Therefore, how to construct an intelligent identification mechanism based on fluid dynamics principles, effectively eliminate false alarm interference, and ensure the timeliness and accuracy of safety interlock control while accurately distinguishing between normal sampling operations and actual leakage accidents in real time, has become an urgent technical problem to be solved. Summary of the Invention

[0003] To address the aforementioned technical problems, this invention provides a method for intelligent early warning and safety interlock control of abnormal liquefied gas leaks based on pattern recognition. Specifically, the technical solution of this invention includes: The system collects real-time operating data of the control valve and the operational status signals of the online sampling device; the real-time operating data includes the pressure before the valve, the pressure after the valve, the valve opening degree, and the fluid temperature. An ideal fluid dynamics benchmark model is constructed based on the inherent flow characteristics of the control valve. The theoretical leak-free flow benchmark under the current valve opening and current pressure difference is calculated using the ideal fluid dynamics benchmark model. In response to the action status signal indicating that sampling is started, the preset sampling bypass impedance parameter is retrieved, the sampling bypass impedance parameter is injected into the ideal fluid dynamics reference model, the theoretical sampling expected flow rate is synthesized, and the theoretical residual vector of the theoretical sampling expected flow rate relative to the theoretical leak-free flow rate reference is calculated. The real-time flow is calculated based on the real-time operating data, and the actual residual vector of the real-time flow relative to the theoretical leak-free flow benchmark is calculated. The similarity between the actual residual vector and the theoretical residual vector is calculated to obtain the morphological similarity between the two. Configured to perform the following logical judgment: If the morphological similarity is greater than the preset safety similarity threshold, the current state is determined to be a normal sampling operation, and a shielding command is generated to shield the trigger signal of the safety interlocking system. If the morphological similarity is not greater than the preset safety similarity threshold, and the amplitude of the actual residual vector exceeds the preset leakage dead zone, the current state is determined to be an abnormal leakage, and a control signal is generated to trigger the safety interlock system to perform a valve shut-off operation. If the morphological similarity is not greater than the preset security similarity threshold, and the magnitude of the actual residual vector does not exceed the preset leakage dead zone, the current state is determined to be background disturbance, and the current operating state is maintained.

[0004] Preferably, the real-time operating data of the control valve and the action status signals of the online sampling device are collected, including: The upstream pressure and downstream pressure of the control valve are obtained by a pressure transmitter. The actual valve opening feedback value of the control valve is obtained through the valve positioner; The temperature of the liquefied gas flowing through the regulating valve is obtained by a temperature sensor; The status of the switch contacts of the online sampling device is obtained through the digital input module as the action status signal.

[0005] Preferably, constructing an ideal fluid dynamics benchmark model based on the inherent flow characteristics of the control valve includes: Obtain the original flow coefficient curve provided by the control valve manufacturer and establish a mapping function between valve opening and flow coefficient; The fluid temperature is used to query a preset table of liquefied gas density temperature change characteristics to determine the current fluid density compensation coefficient. Based on the fluid dynamics flow equation, the flow coefficient output by the mapping function, the square root of the pressure difference between the inlet and outlet pressures, and the fluid density compensation coefficient are correlated to construct the ideal fluid dynamics benchmark model.

[0006] Preferably, the theoretical leak-free flow rate benchmark under the current valve opening and current pressure difference is calculated using the ideal fluid dynamics benchmark model, and its calculation logic is expressed as follows: The theoretical leak-free flow rate benchmark is equal to the ideal flow coefficient corresponding to the current valve opening multiplied by the square root of the pressure difference to density ratio. The ideal fluid dynamics benchmark model defines the theoretical flow state of the control valve under absolutely leak-free and unsampled flow conditions.

[0007] Preferably, the sampling bypass impedance parameters are injected into the ideal fluid dynamics benchmark model to synthesize the theoretical sampling expected flow rate, including: A fluid impedance network model of the sampling pipeline is established, which describes the relationship between the sampling flow rate and the pressure difference across the regulating valve. Based on the ideal fluid dynamics benchmark model, the sampling flow rate calculated by the fluid impedance network model is superimposed to generate the theoretical expected sampling flow rate; The theoretical sampling expected flow rate characterizes the total flow rate data characteristics that the system should present under normal sampling operation.

[0008] Preferably, the similarity calculation of the actual residual vector and the theoretical residual vector to obtain the morphological similarity between the two includes: Construct real residual sequences and theoretical residual sequences that include the time dimension; The degree of morphological matching between the actual residual sequence and the theoretical residual sequence is calculated using a cosine similarity algorithm or a dynamic time warping algorithm, and is used as the morphological similarity. The morphological similarity is used to quantify whether the current actual flow deviation matches the expected sampling fluid dynamics characteristics.

[0009] Preferably, determining the current state as an abnormal leak includes: When the morphological similarity indicates that the two morphologies do not match, identify whether the real residual vector exhibits a linear growth characteristic or an independent fluctuation characteristic that does not have a following characteristic related to pressure changes; If the magnitude of the actual residual vector is continuously greater than the leakage dead zone and does not have the flow resistance compliance characteristic of the sampling operation, then it is confirmed that the regulating valve or pipeline has a physical sealing failure, and an abnormal leakage alarm signal is generated and output.

[0010] Preferably, the preset sampling bypass impedance parameter is determined in the following way: During the system commissioning phase, once it has been confirmed that there are no leaks, controlled sampling operations are performed. Record the actual pressure difference changes and flow deviation data during the sampling process; The flow resistance coefficient of the sampling pipeline is fitted by the least squares method, and the flow resistance coefficient is stored in the controller's knowledge base as the preset sampling bypass impedance parameter.

[0011] Compared with the prior art, the present invention has the following beneficial effects: 1. This method achieves accurate identification of online sampling operations by constructing an ideal fluid dynamics benchmark model and introducing synthetic analysis concepts. It achieves the effect of synthesizing theoretical sampling expected flow rates using preset sampling bypass impedance parameters and determining the operating conditions by calculating the morphological similarity between the actual residual vector and the theoretical residual vector. Compared with existing technologies that rely solely on a single flow rate or pressure threshold for monitoring, this invention can distinguish between normal fluid fluctuations caused by sampling operations and actual physical leaks based on waveform morphology and physical mechanisms. This effectively solves the problem of false positive alarms and accidental triggering of safety interlock shutdowns caused by sampling operations, ensuring the continuity of production operations. 2. This method optimizes the physical fidelity of the baseline model by using first-principles fluid density compensation and valve inherent characteristic modeling; it achieves the effect of calculating the operating density using real-time acquired temperature and pressure data combined with regression coefficients from a standard physical property database, and constructing an absolutely leak-free baseline by combining the valve opening mapping function; compared with existing monitoring schemes that are easily affected by changes in medium physical properties and historical statistical deviations, this invention eliminates model input errors caused by the variability of liquefied gas operating conditions, ensures the accuracy of the flow baseline in terms of physical dimensions and thermodynamic state, and significantly improves the system's ability to detect minute leaks; 3. This method enhances the robustness of the early warning logic by introducing a composite similarity calculation algorithm that includes time alignment and amplitude constraints. It achieves the effect of eliminating sensor response lag using cross-correlation functions before similarity calculation and strictly constraining flow amplitude differences through energy deviation attenuation rate. Compared with traditional simple waveform comparison technology, this invention not only focuses on the consistency of trends but also forces the verification of the magnitude matching degree of flow energy, avoiding the misjudgment of serious leakage with similar waveforms but huge amplitudes as normal sampling. It solves the logical loopholes that may exist in single-form matching and ensures the security of the issuance of safety interlock shielding commands. 4. This method achieves adaptive configuration and fault classification of the system through in-situ parameter identification and multi-dimensional feature verification mechanism; it achieves the effect of fitting the actual pipeline flow resistance coefficient by least squares method during the commissioning stage and identifying abnormal types by combining linear growth characteristics and pressure following characteristics during the operation stage; compared with the existing technology that relies on theoretical design parameters or cannot distinguish the type of leakage, this invention calibrates the model parameters by field measured data, eliminates the error caused by differences in engineering installation, and can further distinguish progressive leakage, random disturbance and systematic failure, providing detailed diagnostic basis for subsequent maintenance decisions. Attached Figure Description

[0012] The present invention will be further explained below with reference to the accompanying drawings and embodiments: Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0013] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0014] Example 1: Please see Figure 1 A method for intelligent early warning and safety interlock control of abnormal liquefied gas leaks based on pattern recognition, comprising: Collect real-time operating data of the control valve and the action status signals of the online sampling equipment; real-time operating data includes inlet pressure, outlet pressure, valve opening degree and fluid temperature; An ideal fluid dynamics benchmark model is constructed based on the inherent flow characteristics of the control valve. The theoretical leak-free flow benchmark under the current valve opening and current pressure difference is calculated using the ideal fluid dynamics benchmark model. In response to the action status signal indicating that sampling is started, the preset sampling bypass impedance parameters are retrieved, the sampling bypass impedance parameters are injected into the ideal fluid dynamics reference model, the theoretical sampling expected flow rate is synthesized, and the theoretical residual vector of the theoretical sampling expected flow rate relative to the theoretical leak-free flow rate reference is calculated. Real-time flow is calculated based on real-time operational data, and the actual residual vector of real-time flow relative to the theoretical leak-free flow benchmark is calculated. The similarity between the actual residual vector and the theoretical residual vector is calculated to obtain the morphological similarity between the two. Configured to perform the following logical judgment: If the morphological similarity is greater than the preset safety similarity threshold, the current state is determined to be a normal sampling operation, and a shielding command is generated to shield the trigger signal of the safety interlocking system. If the morphological similarity is not greater than the preset safety similarity threshold, and the magnitude of the actual residual vector exceeds the preset leakage dead zone, the current state is determined to be an abnormal leakage, and a control signal is generated to trigger the safety interlock system to perform valve shut-off operation. If the morphological similarity is not greater than the preset safety similarity threshold, and the magnitude of the actual residual vector does not exceed the preset leakage dead zone, the current state is determined to be a background disturbance, and the current operating state is maintained.

[0015] This embodiment details an intelligent early warning logic based on synthetic analysis, aiming to solve the problem of false positive alarms caused by online sampling operations in liquefied gas delivery systems. The system performs data acquisition and status sensing steps, obtaining real-time operating data of the regulating valve through a sensor network, including the inlet pressure. Pressure after valve Valve opening and fluid temperature Simultaneously monitor the operational status signals of the online sampling equipment. The signal A high level indicates that sampling is enabled; to fully disclose the specific implementation method of calculating real-time flow based on real-time operating data, this embodiment provides a detailed algorithm flow that can be implemented programmatically: the system collects the raw mass flow rate in the pipeline through a physical flow metering unit, such as a Coriolis mass flow meter. The unit is kg / s; Using real-time collected fluid temperature and average pressure Computational fluid real-time operating density The calculation model is

[0016] in, For standard density, its reference condition is explicitly defined as 20. C, 101.325 kPa, to eliminate the uncertainty of the calculation benchmark. These are the preset coefficients of thermal expansion and compressibility, respectively; to ensure full disclosure in the instruction manual, they are clearly defined here: The physical unit is , The physical unit is Furthermore, the specific values ​​of these two coefficients are not arbitrarily selected, but rather obtained by calling standard physical property databases such as NISTREFPROP to acquire density data of specific media, such as propane and butane, within the operating temperature and pressure range, and then performing multiple linear regression. For example, for liquid propane under normal operating conditions... , This ensures the physical accuracy of density compensation calculations; through division operations. Calculate the real-time volumetric flow rate The unit is This step rigorously defines the process from the raw sensor signal to the model input variables. The conversion logic ensures that it matches the theoretical leak-free flow rate benchmark of the embodiment. Complete consistency in physical dimensions and thermodynamic states, thus meeting the pseudocode-level disclosure requirements; Based on the inherent flow characteristics of the control valve, a flow model is constructed to describe the flow under ideal conditions of absolute zero leakage and no sampling. This model is then used to calculate the theoretical zero-leakage flow benchmark under the current operating conditions. This step, through first-principles modeling, eliminates potential subtle leakage biases that may be hidden in historical statistical data; in response to When sampling is activated, the system does not directly disable the alarm; instead, it retrieves the preset sampling bypass impedance parameters. This parameter is injected into the aforementioned ideal model to synthesize the theoretically sampled expected flow rate. And calculate the theoretical residual vector of the expected flow rate relative to the baseline flow rate. , when the signal When sampling is indicated to be off, the theoretical sampling expected flow rate equal ,at this time Automatically zeroed out to a zero vector, which serves as the zero baseline for subsequent similarity comparisons. This vector is obtained through a difference formula. It is defined to characterize the theoretical flow increment caused purely by the sampling operation; The system processes real-time operational data and calculates real-time traffic. And further calculate its value relative to the theoretical value. The real residual vector The calculation logic is as follows: This establishes the foundation for the dual analysis of physical observation and model prediction; based on this, [the following is discussed]. and Similarity calculation is performed to obtain morphological similarity. The system executes a three-state logic decision: in response to... Greater than the preset security similarity threshold It is determined to be a normal sampling and the shielding interlock is activated; in response to and The amplitude is explicitly defined in this embodiment as the absolute value of the instantaneous residual sample. To eliminate ambiguity in the statistical definition and maintain consistency with the implementation, exceeding the preset leakage dead zone. If the leakage is detected, the valve will be shut off; otherwise, it will be considered a background disturbance. In addition, for the preset security similarity threshold and leak dead zone The system performs adaptive configuration during the initialization phase: It is usually set between 0.85 and 0.95, with the specific value depending on the background level of electromagnetic interference at the site, in order to strike a balance between noise tolerance and waveform differentiation; The setup involves collecting background noise data during a period of no leakage and no operation, and then calculating its standard deviation. ,Pick to This ensures a statistically reliable confidence level of over 99.7%, avoiding misjudgments caused by random noise from the sensor.

[0017] Example 2: The real-time operating data of the control valve and the operational status signals of the online sampling equipment are collected, including: The upstream pressure and downstream pressure of the control valve are obtained by a pressure transmitter. The actual valve opening feedback value of the control valve is obtained through the valve positioner; The temperature of the liquefied gas flowing through the regulating valve is obtained by a temperature sensor; The status of the switch contacts of the online sampling device is obtained through the digital input module as an action status signal.

[0018] This embodiment specifies the hardware implementation for data acquisition, ensuring that the physical meaning of the input data is clear and that it is synchronized in time; the system acquires the upstream pressure of the regulating valve through a high-precision pressure transmitter. and downstream valve pressure These two parameters constitute the core power source driving fluid flow; simultaneously, the actual valve opening feedback value of the regulating valve is obtained through the intelligent valve positioner. Here, actual position feedback is collected instead of control commands to eliminate errors caused by mechanical lag in the actuator; the temperature of the liquefied gas flowing through the regulating valve is obtained through an RTD temperature sensor installed on the pipeline. This is used for subsequent density compensation; The controller obtains the switching contact status of the online sampling device, i.e., the dry contact signal of the solenoid valve, through the digital input module (DI), as the operation status signal. To ensure the feasibility of the step in Example 1, which calculates real-time flow based on real-time operational data, at the hardware and algorithm levels, this example further clarifies that: the system is equipped with a high-precision mass flow meter to acquire the raw mass flow signal. ; The controller has a built-in density correction algorithm based on the formula:

[0019] Perform real-time calculations, among which, It combines temperature With pressure The comprehensive density compensation coefficient, i.e., fluid specific gravity, is used to convert mass flow rate into a value comparable to the baseline model. Operating volume flow rate in the same physical dimension To clarify the calculation logic of this coefficient, this embodiment specifies... The specific functional form is defined as the ratio of the current operating density to the standard density, that is:

[0020] Its specific calculation is based on the linear state equations of the embodiment. Alternatively, the higher-order state equations in the following embodiments are solved to ensure the model closure under different accuracy requirements; This explicit logic for converting physical dimensions eliminates the ambiguity that the term "flow rate" may cause, namely the difference between mass and volume, and ensures the physical validity of residual calculations. This embodiment constructs a high-fidelity physical sensing layer by clearly defining the sensor type and acquisition location. Especially in scenarios with variable operating conditions such as liquefied gas, the use of valve position feedback values ​​instead of command values ​​can accurately reflect the actual flow area of ​​the valve, effectively avoiding model input errors caused by valve stem jamming or positioning deviation, thereby significantly improving the calculation accuracy of the benchmark model and the robustness of the system.

[0021] Example 3: The construction of an ideal fluid dynamics benchmark model based on the inherent flow characteristics of a control valve includes: Obtain the original flow coefficient curve provided by the control valve manufacturer and establish a mapping function between valve opening and flow coefficient; The fluid temperature is used to look up the preset liquefied gas density temperature change characteristic table to determine the current fluid density compensation coefficient. Based on the fluid dynamics flow equation, the flow coefficient output by the mapping function, the square root of the pressure difference between the inlet and outlet pressures, and the fluid density compensation coefficient are correlated to construct an ideal fluid dynamics benchmark model.

[0022] This embodiment details the construction process of an ideal fluid dynamics benchmark model, which serves as the reference zero point for the entire early warning system; it also obtains the original flow coefficient curve provided by the control valve manufacturer, i.e. The relationship between valve opening and valve degree is established by fitting the discrete data into a cubic polynomial using the least squares method. The unit is percentage, and it is related to the ideal flow coefficient. Mapping function:

[0023] in, To obtain the standards provided by the control valve manufacturer -The discrete data points of the valve opening are fitted with specific polynomial coefficients by regression fitting using the least squares method to ensure that the model curve is strictly consistent with the physical characteristics of the valve. Using the collected fluid temperature Determine the current fluid density compensation coefficient. To ensure the reproducibility of the algorithm at the code level and to eliminate quantization noise caused by the lookup table method, this embodiment preferably uses a second-order analytic function based on the physical properties of the medium to replace the lookup table: setting a reference temperature. ,calculate:

[0024] in, This is a preset thermal expansion correction coefficient. Note: To differentiate the different accuracy requirements of the measurement layer and the model layer, a second-order fitting coefficient is used here. , here adopt As physical property parameters of the theoretical model layer, they are clearly distinguished from those used for linear correction of the sensor hardware layer in Example 1. To prevent confusion regarding physical meaning; for example, for propane media, take , For other liquefied gas media in the embodiments, such as butane and LNG, the coefficients are... The method for determining the density is as follows: query standard physical property databases such as NISTREFPROP, obtain the theoretical density data of the medium at at least 3 temperature points within the working temperature range, calculate its ratio to the density at the reference temperature, and obtain the continuous and differentiable density correction term by regression fitting the above second-order polynomial using the least squares method. Based on the fluid dynamics flow equation, the above parameters are correlated to construct a model, which outputs a theoretical leak-free flow rate benchmark. The calculation is as follows:

[0025] Introduced in the formula To prevent errors in calculations under the square root due to negative pressure difference, the model constructs a pure zero-state benchmark through explicit analytical functions.

[0026] Example 4: The theoretical leak-free flow rate baseline is calculated using an ideal fluid dynamics baseline model under the current valve opening and current pressure difference. The calculation logic is as follows: The theoretical leak-free flow rate benchmark is equal to the ideal flow coefficient corresponding to the current valve opening multiplied by the square root of the pressure difference to density ratio. Among them, the ideal fluid dynamics benchmark model defines the theoretical flow state of the control valve under absolutely leak-free and unsampled flow conditions.

[0027] This embodiment further mathematically defines the computational logic, clarifying the physical constraint relationship between flow rate and pressure difference; in the above formula, the subscripts... Strictly defined as the discrete-time index of the current sampling period, it should be noted that in the formulas throughout the text, the function form... With subscript form Both represent the variables at discrete sampling times. The numerical values ​​of the two are mathematically equivalent; that is... , , They represent the time intervals. Valve opening degree, pressure difference, and density ratio are collected or calculated; theoretical leak-free flow rate benchmark is calculated using an ideal fluid dynamics benchmark model. The computational logic is specifically expressed as follows:

[0028] Regarding the dimensional equilibrium constant Its definition must be based on strict unit system constraints: among which, the dimensional balance constant The general calculation formula is defined as follows:

[0029] in, For reference water density, the value is taken as 1000. Or relative specific gravity 1, This is a table of unit conversion factors determined according to the selected unit system; under this general definition, the conversion factor is valid if and only if the flow coefficient... use (Right now (value), pressure difference use And target traffic The unit is When, constant The following conversion logic applies only to values ​​of that type; if the unit system changes, it must be based on... The following corrections were made: Unit conversion factor; in, Current sampling time Dimensional equilibrium constant: Used to convert the dimension of the square root of the pressure difference on the right side of the formula to the dimension of the volumetric flow rate on the left side; its specific value depends on... The system of units used, such as , and pressure units, such as , For example, when Units are , Units are ,and for When the value is, If conversion to the International System of Units (SI) is required... and Then, a corresponding physical conversion factor needs to be introduced; to ensure the sufficiency of the disclosure of this technical solution, the specific method for determining this conversion factor is clearly given here: based on the standard definition of the flow coefficient, when for The unit is defined as At a pressure difference of 1 bar, the factor calculation formula for conversion to SI units is as follows: ;when for When, the corresponding SI conversion factor The calculation is as follows:

[0030] By clarifying this conversion relationship, the accuracy of the formula in different unit systems is ensured; Current opening degree Corresponding ideal flow coefficient Current pressure difference, i.e. The ideal fluid dynamics benchmark model defines the theoretical flow state of the control valve under absolutely leak-free and unsampled flow conditions, based on the fluid specific gravity or density ratio at the current moment. This embodiment transforms physical laws into calculable mathematical constraints, clarifying the core characteristic that the baseline flow rate is proportional to the square root of the pressure difference. In actual operation scenarios, any flow fluctuation that does not conform to this square root law will be regarded as a potential anomaly. This provides a solid theoretical basis for distinguishing between random noise and physical leakage, enabling the system to identify risks based on physical essence rather than simple threshold exceedance.

[0031] Example 5: By injecting the sampling bypass impedance parameters into an ideal fluid dynamics baseline model, the theoretically expected sampling flow rate is synthesized, including: A fluid impedance network model of the sampling pipeline is established, which describes the relationship between the sampling flow rate and the pressure difference across the control valve. Based on the ideal fluid dynamics benchmark model, the sampling flow rate calculated by the fluid impedance network model is superimposed to generate the theoretical expected sampling flow rate; The theoretical sampling expected flow rate characterizes the total flow rate data characteristics that the system should present under normal sampling operation.

[0032] This embodiment details the process of synthesizing sampling signals through parameter injection, a crucial step in distinguishing between sampling and leakage. The sampling pipeline is treated as a fluid impedance connected in parallel to the main pipeline, and a fluid impedance network model is established to describe the sampling flow rate. Pressure difference across the regulating valve The dependency relationship, that is

[0033] in, To sample bypass impedance parameters, This is the fluid density compensation coefficient; to conform to the general industrial fluid dynamics notation standard, the sampling bypass flow rate coefficient is defined here. This coefficient has the same physical meaning as the valve. The value represents the flow capacity of the bypass pipeline; To clarify the parameters To understand the physical nature of the equation and eliminate dimensional ambiguity, this embodiment specifically points out that the formula is essentially the standard flow equation in industrial fluid mechanics, such as... The engineering representation of ; among which, This is a dimensionless fluid density ratio, i.e., specific gravity. Defined as the equivalent flow coefficient of the sampling bypass; to overcome the ambiguity of the technical solution that may result from simply relying on engineering conventions, this embodiment clearly defines... Physical dimensions and constraints: According to the above flow equation, where, The equivalent flow resistance coefficient of the sampling bypass is defined according to physical relationships. In practical industrial applications, for ease of engineering understanding, it is usually expressed as a dimensionless equivalent flow coefficient. The coefficient Commonly used with valves The values ​​have the same physical meaning and unit, that is This enables a standardized measurement of bypass impedance parameters; This clear physical definition makes It is no longer merely a black-box parameter dependent on field regression, but an impedance metric with clear physical meaning. While its value can be obtained through regression from field measurement data in engineering implementation, the aforementioned dimensional definition provides a theoretical verification boundary for the data's validity. For example, if engineering units are used… and If the values ​​need to be converted accordingly, it ensures that the expected flow rate of the theoretical sampling is implemented on the basis of a clear physical relationship, thus eliminating the risk of calculation errors caused by confusion in the unit system. Based on the ideal fluid dynamics benchmark model, the calculated sampling split flow rate is superimposed to generate the theoretical expected sampling flow rate. The formula is This traffic This characterizes the total flow data characteristics that the system should present under normal sampling operations; it should be noted that this linear superposition formula... Based on the following physical assumptions: the flow rate of the sampling loop The flow rate is much smaller than that of the main pipeline, or the upstream pressure source has strong rigidity, ensuring that the sampling operation will not cause a pressure difference in the main pipeline at the moment of initiation. Significant pressure drop; if the actual operating conditions do not meet this assumption, a pressure difference coupling correction factor needs to be introduced; This embodiment greatly improves the accuracy of identifying sampling conditions by transforming unknown interference into known features. In complex industrial sites, if the actual data matches the theoretical expected flow rate, it indicates that the current deviation is entirely caused by sampling; otherwise, it indicates the presence of an additional leakage source. This method effectively incorporates the sampling operation into the expected model of the system, avoiding misjudgment of leakage accidents due to sudden changes in flow rate caused by sampling.

[0034] Example 6: The similarity between the actual residual vector and the theoretical residual vector is calculated to obtain the morphological similarity between the two, including: Construct real residual sequences and theoretical residual sequences that include the time dimension; The cosine similarity algorithm or dynamic time warping algorithm is used to calculate the degree of morphological matching between the actual residual sequence and the theoretical residual sequence, which is used as the morphological similarity. Morphological similarity is used to quantify whether the current actual flow deviation matches the expected sampling fluid dynamics characteristics.

[0035] This embodiment specifically illustrates how to quantify the degree of matching between actual deviations and theoretical expectations through morphological similarity, and specifically addresses the risk of misjudgment caused by scale invariance; the system initializes two lengths of A first-in-first-out circular buffer, wherein, For example, a time window, such as 30 seconds. The sampling period is set to, for example, 100ms; at each sampling time, the latest actual residual is recorded. and theoretical residual Push them into the buffer respectively and And remove the oldest data; for the fluid transmission lag and sensor response time differences that are common in industrial sites, such as pressure transmitters responding faster than flow meters, direct point-to-point similarity calculation may fail due to phase misalignment. Therefore, before calculating the cosine similarity, this embodiment introduces a temporal alignment preprocessing step: defining the maximum allowable hysteresis window. For example, 5 sampling periods, in the interval Internal sliding theory residual sequence Calculate its relationship with the actual residual sequence. The cross-correlation function is used to select the optimal time delay corresponding to the cross-correlation peak. If the theoretical residual sequence The variance is lower than the calculation precision threshold. This threshold is based on the machine precision of the processor's floating-point operations. And the sensor resolution lower limit setting is intended to prevent calculation divergence caused by the denominator approaching zero, so it is directly set To avoid meaningless alignment searches on zero vectors, Perform translation correction to generate the aligned sequence. ; To prevent situations where waveforms are similar in shape but differ greatly in amplitude, such as large leaks exhibiting waveforms similar to those of sampling, and are therefore misjudged as normal sampling, this embodiment employs a composite similarity algorithm:

[0036] in, Cosine similarity is used for calculation, and the aligned sequences are used. :

[0037] Symbols in the formula Explicitly defined as a vector Norm, Euclidean norm, that is, for a vector , To characterize the energy features of the signal; among which, The amplitude similarity factor is defined as:

[0038] in, To prevent extremely small positive numbers with a denominator of zero, for example, take This parameter The value of is not set arbitrarily, but should be selected as the full-scale resolution of the system's flow sensor. This ensures that while numerical calculations are stable, minute physical fluctuations in flow are not masked. parameter The physical meaning of is the attenuation rate of energy deviation, and its value is determined according to the formula. To ensure the accuracy and reproducibility of this key parameter configuration, this embodiment discloses a rigorous determination method based on field statistics, rather than relying on general empirical estimates: 1. The determination of the maximum relative amplitude deviation tolerance is as follows: Under stable reference conditions with no system leakage and no sampling operations, the quantification standard for this stable reference condition is: within the acquisition period, the variance of fluid pressure fluctuation is less than the range. And the temperature drift rate is less than To ensure the purity of statistical data; Continuously collect operational data for at least 24 hours and calculate the relative amplitude residual sample at each moment. ; Calculate the mean of this sample set with standard deviation ,set up ; use The criteria ensure that 99.9996% of normal background noise fluctuations are contained within the tolerance, thus precisely defining the maximum permissible deviation boundary under normal operating conditions; 2. The determination of the lower limit of similarity at the deviation boundary: Under the same benchmark dataset mentioned above, the statistical morphological similarity... The distribution of the vector is used, and the 0.1 percentile value of this distribution is selected, which is the lowest similarity performance after removing extreme outliers, denoted as . ;set up A 20% engineering margin is introduced; through the above statistical calibration method, those skilled in the art can calculate a unique and reliable accuracy based on the specific field sensor accuracy environment. These parameters are used to quantitatively set the parameters to achieve the goal of strictly constraining the deviation of energy amplitude. This composite index It considers both the consistency of waveform trends and strictly constrains the deviation of energy amplitude, ensuring that high similarity is only output when both waveform and flow rate conform to the sampling characteristics; furthermore, this product form effectively avoids logical loopholes, i.e., when... High but Low, such as a large leak, or Low but High, such as sensor noise, when, the product result All will be reduced to a safe threshold. This ensures that the system can correctly distinguish between normal sampling and abnormal operating conditions, avoiding misjudgments caused by product traps.

[0039] Example 7: Determining the current state as an abnormal leak includes: When the morphological similarity indicates that the two morphologies do not match, identify whether the actual residual vector exhibits a linear growth characteristic or an independent fluctuation characteristic that does not have a following characteristic related to pressure changes. If the magnitude of the actual residual vector is consistently greater than the leakage dead zone and does not have the flow resistance compliance characteristic of the sampling operation, then it is confirmed that the control valve or pipeline has a physical sealing failure, and an abnormal leakage alarm signal is generated and output.

[0040] This embodiment further refines the abnormal leakage determination logic, aiming to distinguish between physical leakage and sensor failure through feature extraction; the determination logic only considers morphological similarity. This is triggered under the premise that normal sampling has been excluded; the system performs the following parallel feature verifications: 1. Linear growth feature recognition: for features within the buffer... Perform univariate linear regression on the sequence and calculate the slope. With the coefficient of determination To address the issue of unclear parameter sources, a positive growth threshold is defined here. ,coefficient The physical basis comes from the signal-to-noise ratio threshold theory in signal processing: considering that industrial noise usually follows a normal distribution, we take... noise boundary Double can be established This provides a safety margin, thereby statistically reducing the probability of false positives due to noise-induced random drift to a minimum. It should be understood that the coefficient 1.5 is an empirical value recommended based on typical operating conditions. In practical applications, those skilled in the art can adjust the coefficient within the range of 1.2 to 1.8 according to the background noise level of the actual system and the sensitivity requirements for detecting minute leaks, without departing from the technical concept of the present invention. That is, the drift amount per unit time exceeds the leakage dead zone. In the time window The coefficient is 1.5 times the mean distribution value. This is to balance detection sensitivity and anti-interference capability: if the coefficient is set too low, such as The system is susceptible to false alarms due to long-period baseline drift caused by daily variations in ambient temperature; if the coefficient is set too high, such as... This will significantly prolong the detection time for minute, gradual leaks; if and 1. Ensure the statistical significance of the trend rather than random noise, and determine it as a gradual leakage failure; 2. Independent fluctuation feature identification: calculate With the square root sequence of pressure difference Pearson correlation coefficient ; To avoid steady-state operating conditions Constantness leads to a division-by-zero error with zero variance, if Standard deviation This indicates that the system is in a stable voltage state, and the Pearson correlation coefficient is mathematically undefined. Under this condition, the system no longer performs correlation calculations but instead executes the steady-state deviation determination logic: as long as the magnitude of the actual residual vector is within a certain range... Defined as This represents an instantaneous physical flow deviation that consistently exceeds the preset leakage dead zone. That is, it satisfies the condition for continuous counting:

[0041] The logical execution that satisfies the continuous counting condition is strictly defined as: if the current time satisfies The counter will be reset immediately. Only when The cumulative value exceeds Time-triggered judgment is used to filter out occasional spike interference; Larger than the leakage dead zone And morphological similarity The indication of a non-sampling state, i.e., the existence of independent abnormal fluctuations that are not pressure-following, is considered to satisfy condition b, thus confirming a leak; if Then calculate ; The comprehensive judgment logic is as follows: if the magnitude of the actual residual vector is continuous, that is, if it is constant... Each sampling period remains above the limit, for example The duration corresponds to 5 seconds to filter transient interference, which is greater than the leakage dead zone. If any of the following conditions are met, a physical seal failure of the control valve or pipeline is confirmed, and an abnormal leakage alarm signal is generated: (a) exhibiting a linear growth characteristic ( and (b) exhibits independent fluctuation characteristics. (); Here, it is necessary to specifically explain the physical meaning of this characteristic: In fluid mechanics, the flow rate of conventional physical leakage, such as the porosity formed by valve seat wear, usually varies with the pressure difference. Changes, i.e., high correlation; if detected This means that flow fluctuations are decoupled from pressure difference changes. This usually indicates a non-porous systematic failure, such as zero drift of flow sensors, random noise interference of electronic systems, or forced oscillations caused by upstream pumping systems that are unrelated to the pipeline pressure difference. Although such anomalies are not typical valve leaks, in the safety interlock control strategy, to prevent uncontrollable deterioration of operating conditions, the system classifies them as generalized physical seal failures or loss of system reliability, thus triggering safety interlocks to ensure absolute safety; (c) Although it has pressure following characteristics ( ), but because If the flow resistance is below the threshold, meaning that the sampling operation does not have the flow resistance compliance characteristic, it indicates that the pressure following behavior is not caused by the preset sampling bypass, but by the unknown orifice plate effect, i.e., physical leakage orifice. This logic fully covers three physical forms: gradual leakage, random disturbance leakage, and orifice plate leakage that conforms to the laws of fluid mechanics.

[0042] Example 8: The preset sampling bypass impedance parameters are determined in the following way: During the system commissioning phase, once it has been confirmed that there are no leaks, controlled sampling operations are performed. Record the actual pressure difference changes and flow deviation data during the sampling process; The flow resistance coefficient of the sampling pipeline is fitted by the least squares method, and the flow resistance coefficient is stored in the controller's knowledge base as a preset sampling bypass impedance parameter.

[0043] This embodiment introduces the method for obtaining the core parameter sampling bypass impedance parameter, i.e., the system's self-learning process; during the system commissioning phase after confirming no leakage, i.e., the golden time, a controlled sampling operation is performed; during this period, the actual differential pressure change during the sampling process is recorded. Fluid temperature is used to determine the density compensation coefficient during commissioning by referring to a table. , and flow deviation data The flow resistance coefficient of the sampling pipeline was fitted using the least squares method. The specific fitting process is as follows: As a regression model, let the intermediate observed variables Target variable Based on minimizing the sum of squared residuals The principle is to directly use analytical formulas for calculation. :

[0044] The calculated The preset sampling bypass impedance parameters are stored in the controller's knowledge base; This embodiment simplifies the complex physical characteristics of the sampling pipeline into an equivalent lumped parameter. By regression fitting of field measured data, it ensures that the model parameters can truly reflect the characteristics of the current physical system. This in-situ parameter identification method avoids the errors caused by relying solely on theoretical design parameters, making the generated theoretical sampling expected flow rate highly accurate, thereby ensuring the effectiveness of subsequent synthetic analysis.

[0045] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for intelligent early warning and safety interlock control of abnormal liquefied gas leaks based on pattern recognition, characterized in that, include: The system collects real-time operating data of the control valve and the operational status signals of the online sampling device; the real-time operating data includes the pressure before the valve, the pressure after the valve, the valve opening degree, and the fluid temperature. An ideal fluid dynamics benchmark model is constructed based on the inherent flow characteristics of the control valve. The theoretical leak-free flow benchmark under the current valve opening and current pressure difference is calculated using the ideal fluid dynamics benchmark model. In response to the action status signal indicating that sampling is started, the preset sampling bypass impedance parameter is retrieved, the sampling bypass impedance parameter is injected into the ideal fluid dynamics reference model, the theoretical sampling expected flow rate is synthesized, and the theoretical residual vector of the theoretical sampling expected flow rate relative to the theoretical leak-free flow rate reference is calculated. The real-time flow is calculated based on the real-time operating data, and the actual residual vector of the real-time flow relative to the theoretical leak-free flow benchmark is calculated. The similarity between the actual residual vector and the theoretical residual vector is calculated to obtain the morphological similarity between the two. Configured to perform the following logical judgment: If the morphological similarity is greater than the preset safety similarity threshold, the current state is determined to be a normal sampling operation, and a shielding command is generated to shield the trigger signal of the safety interlocking system. If the morphological similarity is not greater than the preset safety similarity threshold, and the amplitude of the actual residual vector exceeds the preset leakage dead zone, the current state is determined to be an abnormal leakage, and a control signal is generated to trigger the safety interlock system to perform a valve shut-off operation. If the morphological similarity is not greater than the preset security similarity threshold, and the magnitude of the actual residual vector does not exceed the preset leakage dead zone, the current state is determined to be background disturbance, and the current operating state is maintained.

2. The intelligent early warning and safety interlock control method for abnormal liquefied gas leakage based on pattern recognition as described in claim 1, characterized in that, The real-time operating data of the control valve and the operational status signals of the online sampling equipment are collected, including: The upstream pressure and downstream pressure of the control valve are obtained by a pressure transmitter. The actual valve opening feedback value of the control valve is obtained through the valve positioner; The temperature of the liquefied gas flowing through the regulating valve is obtained by a temperature sensor; The status of the switch contacts of the online sampling device is obtained through the digital input module as the action status signal.

3. The intelligent early warning and safety interlock control method for abnormal liquefied gas leakage based on pattern recognition as described in claim 2, characterized in that, The construction of an ideal fluid dynamics benchmark model based on the inherent flow characteristics of a control valve includes: Obtain the original flow coefficient curve provided by the control valve manufacturer and establish a mapping function between valve opening and flow coefficient; The fluid temperature is used to query a preset table of liquefied gas density temperature change characteristics to determine the current fluid density compensation coefficient. Based on the fluid dynamics flow equation, the flow coefficient output by the mapping function, the square root of the pressure difference between the inlet and outlet pressures, and the fluid density compensation coefficient are correlated to construct the ideal fluid dynamics benchmark model.

4. The intelligent early warning and safety interlock control method for abnormal liquefied gas leakage based on pattern recognition as described in claim 3, characterized in that, The theoretical leak-free flow rate benchmark is calculated using the ideal fluid dynamics benchmark model under the current valve opening and current pressure difference. The calculation logic is as follows: The theoretical leak-free flow rate benchmark is equal to the ideal flow coefficient corresponding to the current valve opening multiplied by the square root of the pressure difference to density ratio. The ideal fluid dynamics benchmark model defines the theoretical flow state of the control valve under absolutely leak-free and unsampled flow conditions.

5. The intelligent early warning and safety interlock control method for abnormal liquefied gas leakage based on pattern recognition as described in claim 1, characterized in that, The sampling bypass impedance parameters are injected into the ideal fluid dynamics baseline model to synthesize the theoretical sampling expected flow rate, including: A fluid impedance network model of the sampling pipeline is established, which describes the relationship between the sampling flow rate and the pressure difference across the regulating valve. Based on the ideal fluid dynamics benchmark model, the sampling flow rate calculated by the fluid impedance network model is superimposed to generate the theoretical expected sampling flow rate; The theoretical sampling expected flow rate characterizes the total flow rate data characteristics that the system should present under normal sampling operation.

6. The intelligent early warning and safety interlock control method for abnormal liquefied gas leakage based on pattern recognition as described in claim 1, characterized in that, The similarity between the actual residual vector and the theoretical residual vector is calculated to obtain the morphological similarity between them, including: Construct real residual sequences and theoretical residual sequences that include the time dimension; The degree of morphological matching between the actual residual sequence and the theoretical residual sequence is calculated using a cosine similarity algorithm or a dynamic time warping algorithm, and is used as the morphological similarity. The morphological similarity is used to quantify whether the current actual flow deviation matches the expected sampling fluid dynamics characteristics.

7. The intelligent early warning and safety interlock control method for abnormal liquefied gas leakage based on pattern recognition as described in claim 6, characterized in that, Determining the current state as an abnormal leak includes: When the morphological similarity indicates that the two morphologies do not match, identify whether the real residual vector exhibits a linear growth characteristic or an independent fluctuation characteristic that does not have a following characteristic related to pressure changes; If the magnitude of the actual residual vector is continuously greater than the leakage dead zone and does not have the flow resistance compliance characteristic of the sampling operation, then it is confirmed that the regulating valve or pipeline has a physical sealing failure, and an abnormal leakage alarm signal is generated and output.

8. The intelligent early warning and safety interlock control method for abnormal liquefied gas leakage based on pattern recognition as described in claim 1, characterized in that, The preset sampling bypass impedance parameters are determined in the following way: During the system commissioning phase, once it has been confirmed that there are no leaks, controlled sampling operations are performed. Record the actual pressure difference changes and flow deviation data during the sampling process; The flow resistance coefficient of the sampling pipeline is fitted by the least squares method, and the flow resistance coefficient is stored in the controller's knowledge base as the preset sampling bypass impedance parameter.

Images