A flow meter online calibration system and calibration method

Abstract

The present invention provides a flowmeter online calibration system and method. First, a real-time operating data set of a target flowmeter in a pipeline operating environment is collected, encompassing dynamic monitoring data and environmental status data. The data set is then preprocessed to obtain an operating data sequence that can describe the changing trends of the flowmeter's operating status at multiple consecutive time points. This data sequence is then input into a pre-trained calibration parameter generation model to generate a calibration parameter set containing a flow correction coefficient, an environmental compensation factor, and an error adjustment threshold. Based on this calibration parameter set, the measurement output data is adjusted in real time, and a calibrated flow measurement result is generated and compared with a standard flow reference value for verification. If the deviation is greater than the error adjustment threshold, the calibration parameter set is regenerated and iteratively adjusted until the deviation is within the threshold range, thereby achieving high-precision, adaptive online flowmeter calibration.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial production, and in particular to an online calibration system and a calibration method for a flow meter. Background Art

[0002] In numerous fields, including industrial production, energy transmission, and environmental monitoring, accurate fluid flow measurement is crucial for precise control of production processes, rational resource utilization, and stable system operation. As a key device for measuring fluid flow, the accuracy of flowmeters directly impacts the performance and efficiency of the entire system.

[0003] Currently, the traditional method for flow meter calibration is mostly offline. This method requires removing the flow meter from the actual pipeline system and transporting it to a specialized calibration laboratory for calibration using standard flow equipment. Offline calibration is not only cumbersome, time-consuming, and labor-intensive, but also causes pipeline system interruptions, impacting production continuity and increasing operating costs.

[0004] Even some existing technologies attempt online calibration, but they often focus solely on the flowmeter's output flow data, ignoring the impact of the pipeline's operating environment on flowmeter measurements. Furthermore, existing calibration techniques lack adaptability to dynamic changes in the flowmeter's operating state. Flowmeter performance changes over time during long-term operation, and fixed calibration methods are unable to track these changes in real time, resulting in a gradual decrease in calibration accuracy, making it difficult to meet the increasingly demanding flow measurement precision.

[0005] Furthermore, existing calibration processes often involve a relatively simple and crude approach to verifying and adjusting calibration results. These methods often involve simply comparing measured values ​​with standard values, lacking an effective feedback mechanism to ensure the accuracy of calibration results. Once deviations occur, it is difficult to quickly and accurately adjust calibration parameters to achieve the desired calibration results. Summary of the Invention

[0006] In view of the above-mentioned problems, in combination with the first aspect of the present invention, an embodiment of the present invention provides a flow meter online calibration method, the method comprising:

[0007] Collecting a real-time operating data set of a target flow meter in a pipeline operating environment, the real-time operating data set including dynamic monitoring data generated by the target flow meter during fluid transportation and environmental status data corresponding to the pipeline operating environment;

[0008] Performing data preprocessing on the real-time operation data set to obtain a preprocessed operation data sequence, wherein the preprocessed operation data sequence is used to describe an operation state change trend of the target flow meter at multiple consecutive time nodes;

[0009] Inputting the preprocessed operating data sequence into a pre-trained calibration parameter generation model to generate a calibration parameter set that matches the current operating state of the target flow meter, the calibration parameter set including a flow correction coefficient, an environmental compensation factor, and an error adjustment threshold;

[0010] Adjusting the measurement output data of the target flow meter in real time based on the calibration parameter set to generate a calibrated flow measurement result, and comparing and verifying the calibrated flow measurement result with a preset standard flow reference value;

[0011] When the deviation between the calibrated flow measurement result and the standard flow reference value is greater than the error adjustment threshold, an updated calibration parameter set is regenerated and the real-time adjustment is iteratively performed until the deviation is within the error adjustment threshold.

[0012] On the other hand, an embodiment of the present invention also provides a flow meter online calibration system, including a processor and a machine-readable storage medium, wherein the machine-readable storage medium is connected to the processor, the machine-readable storage medium is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the machine-readable storage medium to implement the above method.

[0013] Based on the above aspects, the embodiment of the present application collects a real-time operating data set containing dynamic monitoring data and environmental status data, and then pre-processes the real-time operating data set to obtain an operating data sequence. The operating data sequence can clearly describe the operating status change trend of the target flow meter at multiple consecutive time nodes, thereby integrating the time series characteristics into the calibration process. Furthermore, a pre-trained calibration parameter generation model is used to generate a calibration parameter set, in which the comprehensive application of the flow correction coefficient, the environmental compensation factor and the error adjustment threshold realizes the multi-dimensional precise correction of the flow meter measurement error. The flow correction coefficient adjusts the deviation of the flow measurement itself, the environmental compensation factor effectively compensates for the interference of different environmental conditions on the measurement results, and the error adjustment threshold provides a quantitative control standard for the calibration accuracy. This multi-parameter collaborative calibration method is a major breakthrough in the traditional single parameter calibration, significantly improving the flexibility and adaptability of the calibration, and can better cope with various complex and changeable actual working conditions.

[0014] Then, based on the calibration parameter set, the measured output data is adjusted in real time and compared with the standard flow reference value for verification, realizing dynamic feedback and real-time optimization of the calibration process, ensuring that the flow meter always maintains high measurement accuracy during operation, avoiding the accumulation of measurement deviations caused by long-term operation or environmental changes, and greatly improving the timeliness and accuracy of flow measurement.

[0015] Finally, when the deviation between the calibrated flow measurement result and the standard flow reference value exceeds the error adjustment threshold, iterative calibration is performed until the deviation falls within the threshold range. This further ensures the accuracy of the calibration result, allowing the flowmeter to automatically adapt to changing operating conditions and continuously provide high-precision flow measurement results. This adaptive calibration method effectively reduces the frequency and workload of manual calibration and improves the stability and automation level of the entire flow measurement system. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a schematic diagram of the execution flow of the flow meter online calibration method provided by an embodiment of the present invention.

[0017] Figure 2 Schematic diagram of the hardware architecture of the flow meter online calibration system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0018] The present invention will be described in detail below with reference to the accompanying drawings. Figure 1 FIG1 is a flow chart of an online calibration method for a flow meter provided by an embodiment of the present invention. The online calibration method for a flow meter is introduced in detail below.

[0019] Step S110 , collecting a real-time operating data set of the target flow meter in the pipeline operating environment, wherein the real-time operating data set includes dynamic monitoring data generated by the target flow meter during fluid transportation and environmental status data corresponding to the pipeline operating environment.

[0020] In this embodiment, in a liquid flow measurement scenario, consider a liquid delivery pipeline system within a chemical enterprise. A target flowmeter is installed on the pipeline to measure the flow rate of a chemical solution. This pipeline system is located in a specific environment, and the surrounding environmental data has a direct impact on the flowmeter's measurement accuracy. The dynamic monitoring data generated by the target flowmeter during the fluid delivery process includes multiple aspects. For example, as the chemical solution flows through the pipeline, the sensor within the flowmeter continuously monitors the solution's flow rate. This can be understood as a precise velocity monitor within the pipeline, constantly recording the solution's flow rate. Furthermore, regarding pressure, since the liquid exerts pressure on the pipe wall as it flows through the pipeline, the flowmeter also monitors changes in this pressure, helping to monitor information such as the stability of the liquid's flow state within the pipeline. Regarding temperature, the chemical solution itself has a certain temperature, and during its flow, heat exchange with the pipe wall may occur, resulting in temperature changes. The flowmeter also monitors the solution's temperature.

[0021] In addition to these dynamic monitoring data directly related to liquid flow, environmental status data corresponding to the pipeline operating environment can also be loaded into the real-time operation dataset. For example, the ambient temperature has a significant impact on the entire pipeline system. If the ambient temperature is low, it may cause the pipe wall to dissipate heat faster, thereby affecting the temperature distribution of the solution within the pipe and, in turn, the flowmeter's measurement accuracy. The humidity of the surrounding environment is also not to be ignored. High humidity environments may cause condensation on the outer wall of the pipe, which may have a subtle effect on some physical properties of the pipe and indirectly affect the flowmeter's measurement. In addition, the space where the pipeline is located may contain some vibration sources, such as large mechanical equipment operating nearby, which generates vibrations that are transmitted to the pipeline. This vibration may interfere with the normal operation of the flowmeter's internal components and affect the measurement accuracy. Therefore, this vibration-related data is also collected in the real-time operation dataset. Therefore, by comprehensively collecting this data, a rich raw data foundation is provided for subsequent flow calibration.

[0022] Step S120 , performing data preprocessing on the real-time operation data set to obtain a preprocessed operation data sequence, wherein the preprocessed operation data sequence is used to describe the operation state change trend of the target flow meter at multiple consecutive time nodes.

[0023] Let's continue with the example of liquid flow measurement in the above-mentioned chemical enterprise. For the temperature monitoring sub-data, the outer wall temperature measurement value and the internal temperature measurement value of the fluid in the pipeline where the target flowmeter is located are extracted, and the difference between the outer wall temperature measurement value and the internal temperature measurement value of the fluid is calculated to obtain a temperature gradient change sequence. Assuming that when a chemical solution flows in a pipeline, the temperature sensor on the outer wall of the pipeline measures the outer wall temperature of 25 degrees Celsius at a certain moment, while the temperature sensor inside the fluid measures the temperature of the solution as 30 degrees Celsius, then the difference between the two is -5 degrees Celsius. As time goes by, this difference at different moments constitutes a temperature gradient change sequence. This temperature gradient change sequence can reflect the dynamic situation of heat exchange between the liquid and the pipeline wall at different time nodes, which helps to discover temperature factors that may affect flow measurement.

[0024] For the pressure monitoring sub-data, the pressure fluctuation peak in the direction of fluid flow in the pipeline operating environment is identified, and a pressure stability evaluation index is generated based on the frequency of occurrence of the pressure fluctuation peak. During the liquid transportation process, pressure fluctuations may occur due to reasons such as the possible bends in the pipeline, the opening and closing of the valve, or the flow characteristics of the liquid itself. For example, when the liquid flows through a half-open valve, a local pressure change will be formed near the valve. If the pressure fluctuation peak occurs frequently within a certain period of time, it means that the flow state of the liquid is not very stable. Assuming that the pressure fluctuation peak occurs 5 times within 10 minutes, this frequency of occurrence is converted into a pressure stability evaluation index through a specific algorithm. This index can quantify the pressure stability during the liquid flow process, thereby providing a basis for subsequent flow calibration.

[0025] For the flow rate monitoring sub-data, the rotor speed measurement value and the blade vibration amplitude measurement value of the target flow meter are obtained, and the rotor speed measurement value and the blade vibration amplitude measurement value are time-series aligned to generate a flow rate-related feature vector. For the rotor inside the flow meter, its rotation speed is closely related to the flow rate of the liquid. When the rotor speed is faster, it usually means that the liquid flow rate is higher. At the same time, the vibration amplitude of the blade can also reflect some characteristics of the liquid flow. Assuming that at a certain moment, the rotor speed measurement value is 1000 revolutions per minute and the blade vibration amplitude is 0.5 mm, these data at the same time node are time-series aligned to generate a flow rate-related feature vector. As time goes by, the flow rate-related feature vectors at different moments can comprehensively describe the changing trend of the flow rate-related data.

[0026] Finally, the temperature gradient change sequence, the pressure stability evaluation index, and the flow rate associated characteristic vector are spliced ​​according to the time nodes to generate the pre-processed operation data sequence, wherein the data unit corresponding to each time node in the pre-processed operation data sequence includes the temperature gradient value, the pressure stability level, and the flow rate associated value. In this way, a comprehensive data unit is generated at each time node. For example, the data unit at a certain moment may be a temperature gradient value of -3 degrees Celsius, a pressure stability level of medium (derived from the previous pressure stability evaluation index), and a flow rate associated value of [1200 rpm, 0.4 mm]. This data unit fully reflects the operating status information of the target flowmeter in liquid flow measurement at that moment. The entire pre-processed operation data sequence shows the operating status change trend at multiple consecutive time nodes.

[0027] Step S130: input the preprocessed operating data sequence into a pre-trained calibration parameter generation model to generate a calibration parameter set that matches the current operating state of the target flow meter, wherein the calibration parameter set includes a flow correction coefficient, an environmental compensation factor, and an error adjustment threshold.

[0028] It is still based on the liquid flow measurement scenario of chemical enterprises. The calibration parameter generation model includes a multi-level parameter generation unit. First, the first-level parameter generation unit is called to perform nonlinear mapping processing on the temperature gradient value in the preprocessed operating data sequence to generate a primary temperature compensation factor, and the primary temperature compensation factor is weighted and fused with the internal temperature measurement value of the fluid to obtain the final temperature compensation coefficient. Assuming that a certain temperature gradient value in the preprocessed operating data sequence is -2 degrees Celsius, after the nonlinear mapping processing in the first-level parameter generation unit, the primary temperature compensation factor is 0.8. If the internal temperature measurement value of the fluid is 32 degrees Celsius at this time, through a specific weighted fusion algorithm (for example, weighted calculation according to a certain ratio), the final temperature compensation coefficient is 0.9 (this is just a hypothetical calculation result). This coefficient will be used to compensate for the temperature factor in flow measurement.

[0029] Next, the second-level parameter generation unit is called to discretize and classify the pressure stability levels in the preprocessed operating data sequence, determine the pressure condition category of the current pipeline operating environment, and match the corresponding pressure compensation weight from a preset pressure-flow relationship table based on the pressure condition category. For example, if the pressure stability level is discretized and classified to determine that the current pipeline operating environment belongs to a condition with large pressure fluctuations, the corresponding pressure compensation weight is found in the preset pressure-flow relationship table to be 1.2. This weight is derived from extensive previous experiments and statistical data, and is used to compensate flow measurements under this pressure condition to improve measurement accuracy.

[0030] Next, the third-level parameter generation unit is called to perform a sliding window mean filter on the velocity correlation values ​​in the preprocessed operational data sequence to eliminate instantaneous velocity fluctuation noise. A velocity smoothing coefficient is then calculated based on the filtered velocity correlation values. Assuming that the rotor speed measurements in the velocity correlation values ​​exhibit some transient fluctuations, a relatively smoothed velocity value is obtained after sliding window mean filtering. Based on this filtered velocity correlation value, a velocity smoothing coefficient of 0.95 is calculated. This coefficient can suppress short-term fluctuations in the velocity data, making flow measurement more stable and accurate.

[0031] Finally, the final temperature compensation coefficient, the pressure compensation weight, and the flow rate smoothing coefficient are normalized to generate the flow correction coefficient in the calibration parameter set. The product of the pressure compensation weight and the flow rate smoothing coefficient is then determined as the environmental compensation factor. After normalization, assuming a flow correction coefficient of 1.1, a pressure compensation weight of 1.2, and a flow rate smoothing coefficient of 0.95, the resulting environmental compensation factor is 1.14. Simultaneously, based on system requirements and statistical analysis of the data, an error adjustment threshold of 0.05 is determined. This error adjustment threshold will be used to subsequently determine whether the calibrated flow measurement results meet the requirements.

[0032] Step S140 , adjusting the measurement output data of the target flow meter in real time based on the calibration parameter set, generating a calibrated flow measurement result, and comparing and verifying the calibrated flow measurement result with a preset standard flow reference value.

[0033] In the case of liquid flow measurement in a chemical company, first obtain the raw flow measurement value output by the target flow meter in an uncalibrated state and extract the acquisition timestamp corresponding to the raw flow measurement value. Assume that the raw flow measurement value output by the target flow meter in an uncalibrated state at a certain moment is 100 cubic meters per hour, and the corresponding acquisition timestamp is 10:0:00 on October 1, 2023.

[0034] The corresponding temperature gradient value, pressure stability level, and flow rate correlation value are matched from the preprocessed operational data sequence based on the acquisition timestamp. Based on the aforementioned timestamp, the corresponding temperature gradient value is -1 degrees Celsius, the pressure stability level is relatively stable, and the flow rate correlation value is [1100 rpm, 0.3 mm].

[0035] The temperature gradient value is input into the compensation function corresponding to the final temperature compensation coefficient to obtain a temperature correction, and the pressure stability level is input into the adjustment function corresponding to the pressure compensation weight to obtain a pressure correction. Assuming the compensation function corresponding to the final temperature compensation coefficient is a linear function, the temperature correction calculated based on a temperature gradient value of -1°C is 5 cubic meters per hour. The adjustment function corresponding to the pressure compensation weight, based on a relatively stable pressure stability level, calculates a pressure correction of 3 cubic meters per hour.

[0036] Substituting the raw flow measurement value, the temperature correction value, and the pressure correction value into a preset flow superposition formula yields a preliminary calibrated flow value. This preliminary calibrated flow value is then multiplied by the flow rate smoothing coefficient to generate the calibrated flow measurement result. According to the flow superposition formula (e.g., raw flow measurement value + temperature correction value + pressure correction value), the preliminary calibrated flow value is 100 + 5 + 3 = 108 cubic meters per hour. Multiplying this preliminary calibrated flow value by the flow rate smoothing coefficient of 0.95 yields a calibrated flow measurement result of 102.6 cubic meters per hour.

[0037] The calibrated flow measurement result is then compared and verified against a preset standard flow reference value. A historical standard flow dataset matching the current pipeline operating environment is obtained from a remote calibration database. This historical standard flow dataset contains laboratory-calibrated benchmark flow values ​​under the same environmental conditions. Assume that the benchmark flow value obtained from the database is 103 cubic meters per hour. The absolute deviation between the calibrated flow measurement result and the benchmark flow value is calculated as |102.6 - 103| = 0.4 cubic meters per hour. A moving average of all absolute deviations within a preset time window (e.g., 1 hour) is then calculated.

[0038] Step S150 , when the deviation between the calibrated flow measurement result and the standard flow reference value is greater than the error adjustment threshold, regenerate an updated calibration parameter set and iterate the real-time adjustment until the deviation is within the error adjustment threshold.

[0039] In the aforementioned scenario of liquid flow measurement at a chemical enterprise, if the previously calculated deviation of 0.4 cubic meters per hour is greater than the error adjustment threshold of 0.05 cubic meters per hour, then the operation of regenerating an updated calibration parameter set will be triggered. This may be because certain factors were not fully considered during the previous calibration process or the environment has changed. When regenerating the updated calibration parameter set, the pre-processed operating data sequence will be analyzed and processed again according to the method in step S130, and the calculation method or weight of each parameter may be adjusted. For example, it may be found that the previous nonlinear mapping of the temperature gradient value is not accurate enough, so the mapping function in the first-level parameter generation unit is adjusted and parameters such as the temperature compensation factor are recalculated. Then, according to step S140, the measurement output data of the target flow meter is adjusted in real time to obtain a new calibrated flow measurement result. Continue to compare and verify with the preset standard flow reference value. If the deviation is still greater than the error adjustment threshold, the process will be iterated until the deviation is within the error adjustment threshold range. For example, after multiple iterations, the calibrated flow measurement result is 102.9 cubic meters per hour, which deviates from the baseline flow value of 103 cubic meters per hour by 0.1 cubic meters per hour, which is less than the error adjustment threshold of 0.05 cubic meters per hour. At this time, it is considered that the flow measurement has achieved the required accuracy.

[0040] Based on the above steps, the embodiment of the present application collects a real-time operating data set containing dynamic monitoring data and environmental status data, and then pre-processes the real-time operating data set to obtain an operating data sequence. The operating data sequence can clearly describe the operating status change trend of the target flow meter at multiple continuous time nodes, thereby integrating the time series characteristics into the calibration process. Furthermore, a pre-trained calibration parameter generation model is used to generate a calibration parameter set, in which the comprehensive application of the flow correction coefficient, the environmental compensation factor and the error adjustment threshold realizes the multi-dimensional precise correction of the flow meter measurement error. The flow correction coefficient adjusts the deviation of the flow measurement itself, the environmental compensation factor effectively compensates for the interference of different environmental conditions on the measurement results, and the error adjustment threshold provides a quantitative control standard for the calibration accuracy. This multi-parameter collaborative calibration method is a major breakthrough in the traditional single parameter calibration, significantly improving the flexibility and adaptability of the calibration, and can better cope with various complex and changeable actual working conditions.

[0041] Then, based on the calibration parameter set, the measured output data is adjusted in real time and compared with the standard flow reference value for verification, realizing dynamic feedback and real-time optimization of the calibration process, ensuring that the flow meter always maintains high measurement accuracy during operation, avoiding the accumulation of measurement deviations caused by long-term operation or environmental changes, and greatly improving the timeliness and accuracy of flow measurement.

[0042] Finally, when the deviation between the calibrated flow measurement result and the standard flow reference value exceeds the error adjustment threshold, iterative calibration is performed until the deviation falls within the threshold range. This further ensures the accuracy of the calibration result, allowing the flowmeter to automatically adapt to changing operating conditions and continuously provide high-precision flow measurement results. This adaptive calibration method effectively reduces the frequency and workload of manual calibration and improves the stability and automation level of the entire flow measurement system.

[0043] In one possible implementation, the real-time operation data set includes a plurality of operation sub-data of the flow meter within a preset time period, the operation sub-data including temperature monitoring sub-data, pressure monitoring sub-data, and flow rate monitoring sub-data. Step S120 includes:

[0044] Step S121 , extracting the outer wall temperature measurement value and the fluid internal temperature measurement value of the pipe where the target flow meter is located from the temperature monitoring sub-data, and performing difference calculation on the outer wall temperature measurement value and the fluid internal temperature measurement value to obtain a temperature gradient change sequence.

[0045] Step S122 : identifying the pressure fluctuation peak value in the fluid flow direction in the pipeline operation environment based on the pressure monitoring sub-data, and generating a pressure stability evaluation index based on the occurrence frequency of the pressure fluctuation peak value.

[0046] Step S123 , obtaining the rotor speed measurement value and the blade vibration amplitude measurement value of the target flow meter for the flow velocity monitoring sub-data, and performing time-series alignment on the rotor speed measurement value and the blade vibration amplitude measurement value to generate a flow velocity associated feature vector.

[0047] Step S124, splicing the temperature gradient change sequence, the pressure stability evaluation index and the flow rate association characteristic vector according to time nodes to generate the preprocessed operation data sequence, wherein the data unit corresponding to each time node in the preprocessed operation data sequence contains the temperature gradient value, the pressure stability level and the flow rate association value.

[0048] In this embodiment, a target flowmeter is used to measure the flow rate of a chemical solution in a pipeline system. The real-time operational data set contains multiple operational sub-data, of which the temperature monitoring sub-data is particularly important. The pipeline in which the target flowmeter is located has a temperature sensor installed on its outer wall to measure the outer wall temperature. A sensor is also installed inside the pipeline to measure the internal temperature of the fluid. Temperature data is continuously collected over a specific time period, such as the one-hour preset time period from 9:00 AM to 10:00 AM on October 1, 2023. At 9:00 AM, the outer wall temperature of the pipeline is measured to be 23 degrees Celsius, while the internal temperature of the fluid is 28 degrees Celsius. The difference between the two is calculated to be -5 degrees Celsius. Over time, at 9:10 AM, the outer wall temperature reaches 24 degrees Celsius, while the internal temperature of the fluid is 29 degrees Celsius, a difference of -5 degrees Celsius. At 9:20 AM, the outer wall temperature reaches 25 degrees Celsius, while the internal temperature reaches 30 degrees Celsius, a difference of -5 degrees Celsius. And so on. The differences calculated at different time points throughout the preset time period constitute a temperature gradient change sequence. This temperature gradient change sequence can reflect the dynamics of heat exchange between the liquid and the pipe wall during this time period. For example, if the temperature gradient remains stable, it indicates that the heat exchange is in a relatively stable state. If the temperature gradient fluctuates significantly, it may mean that the surrounding environment of the pipe or the flow state of the liquid has changed. This is of great significance for subsequent analysis of the flow meter's operating status and calibration of flow measurement.

[0049] Regarding pressure monitoring data, as chemical solutions flow through pipelines, pressure fluctuations occur in the direction of the liquid's flow due to factors such as the pipeline's layout, valve status, and the physical properties of the liquid itself. Multiple pressure sensors are installed along the pipeline to monitor pressure fluctuations. During the preset time period, the sensors continuously collect pressure data to identify pressure fluctuation peaks. For example, at 9:05:00, the pressure suddenly rises from the normal 1.5 MPa to 2.0 MPa. This is a pressure fluctuation peak. Over time, multiple such pressure fluctuation peaks are detected. The frequency of these pressure fluctuation peaks is calculated. For example, suppose eight pressure fluctuation peaks occur within an hour. Based on this frequency, a specific algorithm is used to generate a pressure stability assessment index. If the frequency of pressure fluctuation peaks is high, for example, exceeding a certain threshold (this threshold is determined based on extensive experiments and historical data), it indicates that the liquid flow is unstable, and the pressure stability assessment index will reflect this instability. If the frequency of pressure fluctuation peaks is low and within a reasonable range, it indicates that the pressure is relatively stable, and the pressure stability assessment index will also reflect this stability. This pressure stability evaluation index plays a key role in analyzing the flow characteristics of liquids in pipelines and evaluating the factors affecting the measurement accuracy of flow meters.

[0050] For flow rate monitoring data, the rotor speed within the target flowmeter is closely related to the liquid flow rate, while the blade vibration amplitude can also reflect the characteristics of the liquid flow. Over a preset time period, the rotor speed and blade vibration amplitude measurements of the target flowmeter are continuously acquired. For example, at 9:00:00, the rotor speed is measured at 900 rpm and the blade vibration amplitude is 0.3 mm. Time-aligning these two data points forms part of a flow rate correlation feature vector at that moment. At 9:10:00, the rotor speed is measured at 950 rpm and the blade vibration amplitude is 0.4 mm, and these two data points are also time-aligned. Over time, these data points at different time points throughout the preset time period are combined to generate a flow rate correlation feature vector. This flow rate correlation feature vector comprehensively records flow rate-related information at different time points, including the correlation between rotor speed and blade vibration amplitude. It can reflect the actual liquid flow rate from multiple perspectives, providing rich flow rate-related information for subsequent flow calibration.

[0051] Finally, the obtained temperature gradient change sequence, pressure stability evaluation index, and flow velocity correlation feature vector are concatenated at each time node to generate a preprocessed operational data sequence. At each time node, a data unit is formed, containing the temperature gradient value, pressure stability level, and flow velocity correlation value. For example, at 9:00:00, the temperature gradient value is -5 degrees Celsius, the pressure stability evaluation index is calculated and classified as medium (the medium level is determined based on the previously established pressure stability evaluation index classification criteria), and the flow velocity correlation value is [900 rpm, 0.3 mm]. This constitutes a complete data unit. Over time, these data units at different time nodes form the preprocessed operational data sequence. This preprocessed operational data sequence comprehensively describes the operating state trends of the target flowmeter at multiple consecutive time nodes. It covers information on temperature, pressure, and flow velocity, providing a comprehensive and accurate data foundation for subsequent input into the calibration parameter generation model, thereby facilitating more precise calibration of the target flowmeter's measurement results and improving measurement accuracy in liquid flow measurement scenarios in chemical enterprises.

[0052] In a possible implementation, the calibration parameter generation model includes a multi-level parameter generation unit, and step S130 includes:

[0053] Step S131, calling the first-level parameter generation unit, performing nonlinear mapping processing on the temperature gradient value in the preprocessed operation data sequence, generating a primary temperature compensation factor, and weightedly fusing the primary temperature compensation factor with the internal temperature measurement value of the fluid to obtain a final temperature compensation coefficient.

[0054] Step S132: Call the second-level parameter generation unit to discretize and classify the pressure stability levels in the preprocessed operating data sequence, determine the pressure condition category to which the current pipeline operating environment belongs, and match the corresponding pressure compensation weight from the preset pressure-flow relationship table based on the pressure condition category.

[0055] Step S133, calling the third-level parameter generation unit, performing sliding window mean filtering on the velocity correlation values ​​in the preprocessed operation data sequence to eliminate instantaneous velocity fluctuation noise, and calculating the velocity smoothing coefficient based on the filtered velocity correlation values.

[0056] Step S134, normalizing the final temperature compensation coefficient, the pressure compensation weight, and the flow rate smoothing coefficient to generate a flow correction coefficient in the calibration parameter set, and determining the product of the pressure compensation weight and the flow rate smoothing coefficient as the environmental compensation factor.

[0057] In this embodiment, when the preprocessed operational data sequence is input into the pretrained calibration parameter generation model, the multi-level parameter generation unit within the calibration parameter generation model begins executing its corresponding tasks. First, the first-level parameter generation unit is invoked to perform nonlinear mapping on the temperature gradient values ​​in the preprocessed operational data sequence. For example, during the flow of a chemical solution through a pipeline, the previously obtained temperature gradient value sequence contains values ​​at different time points. For example, at 9:00:00 AM on October 1, 2023, the temperature gradient value is -5 degrees Celsius. The first-level parameter generation unit processes this -5 degree Celsius temperature gradient value using its internally defined nonlinear mapping function (which is constructed based on extensive experimental data and theoretical analysis) to obtain a primary temperature compensation factor. Assume this primary temperature compensation factor is 0.8. This primary temperature compensation factor is then weighted and fused with the measured internal fluid temperature. At 9:00:00 AM, the measured internal fluid temperature is 28 degrees Celsius. Using a specific weighted fusion algorithm (determined based on the pipeline system and liquid characteristics), the final temperature compensation coefficient is calculated. Assume that the weighted calculation process is: final temperature compensation coefficient = 0.8 × 0.1 + 28 × 0.9 = 25.28 (here 0.1 and 0.9 are weighted coefficients determined according to system characteristics). This final temperature compensation coefficient reflects the comprehensive impact of temperature factors on flow measurement and will play an important role in the subsequent flow calibration process.

[0058] Then the second-level parameter generation unit is called to discretize and classify the pressure stability levels in the pre-processed operating data sequence. In the previous pressure monitoring process, the pressure stability evaluation index was obtained based on the frequency of pressure fluctuation peaks, and the pressure stability level was determined. For example, during the entire measurement period, if the frequency of pressure fluctuation peaks is high, after analysis and comparison with the set classification standards, the pressure stability level is determined to be an unstable category. Then, based on this pressure stability level, the pressure condition category to which the current pipeline operating environment belongs is determined, and then the corresponding pressure compensation weight is matched from the preset pressure-flow relationship table. Assuming that the currently determined pressure condition category is an unstable category, the corresponding pressure compensation weight of 1.2 is found in the pressure-flow relationship table. This pressure compensation weight is a compensation coefficient set based on a large amount of experimental and historical data for accurate flow measurement under such unstable pressure conditions. It reflects the need for correction of flow measurement under such pressure conditions.

[0059] The third-level parameter generation unit is then called to perform a sliding window mean filter on the flow velocity correlation values ​​in the preprocessed operating data sequence. The previously acquired flow velocity correlation values ​​include the rotor speed and blade vibration amplitude measurements of the target flowmeter, such as rotor speed and blade vibration amplitude data at different time points. Due to transient interference factors during liquid flow, small fluctuations in rotor speed and blade vibration amplitude may occur, which can affect the accuracy of flow measurement. For example, at 9:00:00, the rotor speed is measured to be 900 rpm and the blade vibration amplitude is 0.3 mm; at 9:01:00, the rotor speed is measured to be 910 rpm and the blade vibration amplitude is 0.32 mm, and so on. Sliding window mean filtering (the sliding window size is determined based on the system's dynamic characteristics and the requirements for flow velocity stability, for example, a window size of 5 minutes) is used to eliminate these transient flow velocity fluctuations. Assuming that relatively smooth flow velocity correlation values ​​are obtained after filtering, the flow velocity smoothing coefficient is calculated based on these filtered flow velocity correlation values. Assume that according to a specific calculation method (this method is determined based on the physical properties of the liquid in the pipeline and the working principle of the flow meter), the flow rate smoothing coefficient is calculated to be 0.95. This flow rate smoothing coefficient can effectively suppress short-term fluctuations in the flow rate data, making the flow measurement more stable and accurate.

[0060] The final temperature compensation coefficient, pressure compensation weight, and flow rate smoothing coefficient obtained above are normalized to generate the flow correction coefficient in the calibration parameter set. Assuming the normalization formula is: Flow Correction Coefficient = (Final Temperature Compensation Coefficient + Pressure Compensation Weight + Flow Rate Smoothing Coefficient) / (Final Temperature Compensation Coefficient × Pressure Compensation Weight × Flow Rate Smoothing Coefficient), substitute the previously obtained values ​​into the calculation to obtain the flow correction coefficient. Simultaneously, the product of the pressure compensation weight and the flow rate smoothing coefficient is determined as the environmental compensation factor, for example, 1.2 × 0.95 = 1.14. This environmental compensation factor accounts for the combined effects of pressure and flow rate on the environment and will be used to correct for flow measurement deviations caused by environmental factors during subsequent flow calibration.

[0061] In a possible implementation, step S140 includes:

[0062] Step S141 : obtaining an original flow measurement value output by the target flow meter in an uncalibrated state, and extracting a collection timestamp corresponding to the original flow measurement value.

[0063] Step S142 : matching corresponding temperature gradient values, pressure stability levels, and flow rate correlation values ​​from the pre-processed operation data sequence according to the acquisition timestamp.

[0064] Step S143 , inputting the temperature gradient value into the compensation function corresponding to the final temperature compensation coefficient to obtain a temperature correction value, and inputting the pressure stability level into the adjustment function corresponding to the pressure compensation weight to obtain a pressure correction value.

[0065] In step S144, the original flow measurement value, the temperature correction value, and the pressure correction value are substituted into a preset flow superposition formula to obtain a preliminary calibrated flow value, and the preliminary calibrated flow value is multiplied by the flow rate smoothing coefficient to generate the calibrated flow measurement result.

[0066] In this embodiment, when real-time adjustments are made to the target flow meter's output data based on the calibration parameter set to generate a calibrated flow measurement result, the raw flow measurement value output by the target flow meter in an uncalibrated state is first obtained, and the acquisition timestamp corresponding to the raw flow measurement value is extracted. For example, at 9:0:00 AM on October 1, 2023, the raw flow measurement value output by the uncalibrated target flow meter is 100 cubic meters per hour, and the acquisition timestamp is 9:0:00 AM.

[0067] Based on this acquisition timestamp, the corresponding temperature gradient, pressure stability level, and flow rate correlation value are matched from the preprocessed operational data sequence. At 9:00:00, the previously calculated temperature gradient value is -5 degrees Celsius, the pressure stability level is unstable, and the flow rate correlation value is [900 rpm, 0.3 mm].

[0068] The temperature gradient value is input into the compensation function corresponding to the final temperature compensation coefficient to obtain the temperature correction. Assuming the compensation function corresponding to the final temperature compensation coefficient is based on a physical model, substituting -5 degrees Celsius into this function yields a temperature correction of 5 cubic meters per hour. Simultaneously, the pressure stability level is input into the adjustment function corresponding to the pressure compensation weight to obtain the pressure correction. Assuming the adjustment function for the pressure compensation weight is based on an unstable pressure stability level, the resulting pressure correction is 3 cubic meters per hour.

[0069] Substituting the raw flow measurement value, temperature correction, and pressure correction into the preset flow superposition formula yields a preliminary calibrated flow value. Assuming the flow superposition formula is: Preliminary Calibration Flow Value = Raw Flow Measurement Value + Temperature Correction + Pressure Correction, the preliminary calibration flow value is 100 + 5 + 3 = 108 cubic meters per hour. The preliminary calibration flow value is then multiplied by the flow velocity smoothing coefficient to generate the calibrated flow measurement result. The calibrated flow measurement result is 108 × 0.95 = 102.6 cubic meters per hour. This calibrated flow measurement result comprehensively accounts for the effects of temperature, pressure, and flow velocity on the raw flow measurement value. By applying the calibration parameter set, the raw flow measurement value is corrected, resulting in a more accurate flow measurement result that meets the flow measurement accuracy requirements of chemical companies in liquid flow measurement scenarios.

[0070] In a possible implementation, step S140 may further include:

[0071] Step S145 , obtaining a historical standard flow data set that matches the current pipeline operating environment from a remote calibration database, wherein the historical standard flow data set includes reference flow values ​​calibrated in a laboratory under the same environmental conditions.

[0072] In this embodiment, this remote calibration database stores a large amount of historical flow data under different pipeline operating environments in chemical companies. In the current chemical solution pipeline flow measurement scenario, the historical standard flow data set in the database contains benchmark flow values ​​calibrated in the laboratory under the same environmental conditions. These benchmark flow values ​​are accurate flow values ​​measured for specific pipelines, liquid types, and environmental conditions using high-precision measuring equipment under strictly controlled laboratory conditions. For example, for the current pipeline, the composition, temperature, pressure range of the chemical solution inside it, as well as the material, diameter, and other environmental factors of the pipeline all match certain historical data stored in the database, so that the corresponding benchmark flow value can be obtained from it.

[0073] Step S146 , calculating the absolute deviation between the calibrated flow measurement result and the reference flow value, and calculating a moving average of all absolute deviations within a preset time window.

[0074] For example, if the calibrated flow measurement result is 102.6 cubic meters per hour, and the baseline flow value obtained from the database is 103 cubic meters per hour, then the absolute deviation is |102.6 - 103| = 0.4 cubic meters per hour. Furthermore, a moving average of all absolute deviations within a preset time window is required. Assuming the preset time window is one hour, multiple calibrated flow measurements are obtained at regular intervals (e.g., every 10 minutes) within that hour. The absolute deviations between these measurements and the corresponding baseline flow values ​​are calculated. Then, a moving average of these absolute deviations is calculated using a moving average algorithm. This moving average more comprehensively reflects the deviations between the calibrated flow measurement results and the baseline flow values ​​over time, minimizing the impact of random deviations at a single moment in time.

[0075] Step S147 : When the moving average value is greater than the error adjustment threshold, a calibration parameter update instruction is triggered, where the calibration parameter update instruction is used to instruct to regenerate the updated calibration parameter set.

[0076] Step S148 , dynamically weighting the flow correction coefficient and the environmental compensation factor in the updated calibration parameter set to generate a new error adjustment threshold, and feeding the new error adjustment threshold back to the calibration parameter generation model for parameter self-optimization.

[0077] For example, if the error adjustment threshold is set to 0.05 cubic meters per hour, and the calculated moving average is 0.1 cubic meters per hour, which is greater than the error adjustment threshold, a calibration parameter update instruction will be triggered. This instruction is used to instruct the regeneration of an updated calibration parameter set to further adjust the flow measurement results. The flow correction coefficient and environmental compensation factor in the updated calibration parameter set are dynamically weighted to generate a new error adjustment threshold. Assuming the updated flow correction coefficient is 1.15 and the environmental compensation factor is 1.14, the new error adjustment threshold is calculated to be 0.06 cubic meters per hour according to a specific dynamic weighting algorithm (this algorithm is determined based on the system's historical data and optimization strategy). This new error adjustment threshold is then fed back to the calibration parameter generation model for parameter self-optimization, allowing the calibration parameter generation model to adjust its internal parameter generation mechanism according to the new error adjustment threshold, thereby improving the accuracy of flow measurement.

[0078] In one possible implementation, the pre-trained calibration parameter generation model is trained by the following steps:

[0079] Step S210: Acquire a historical pipeline operation data set, where the historical pipeline operation data set includes uncalibrated operation data collected by multiple flow meters under different working conditions and their corresponding laboratory calibration data.

[0080] Step S220 , performing noise filtering and outlier removal on the uncalibrated operating data to generate a training operating data sequence, and converting the laboratory calibration data into a standard flow reference sequence.

[0081] For example, a historical pipeline operation dataset includes uncalibrated operating data collected by multiple flowmeters in a chemical plant under different operating conditions, along with their corresponding laboratory calibration data. These different operating conditions include varying chemical solution compositions, pipeline pressure ranges, temperature environments, and flow rate ranges. For example, some flowmeters measure the flow of a high-concentration chemical solution under high temperature and high pressure, while others measure the flow of a low-concentration solution under low temperature and low pressure. Uncalibrated operating data under these different operating conditions requires noise filtering and outlier removal to generate training data sequences. Uncalibrated operating data may contain noise and outliers due to sensor failure, electromagnetic interference, or other external factors. For example, electromagnetic interference generated by the startup of a nearby large motor may cause a flowmeter to collect a flow rate value significantly outside the normal range at a specific moment. This value is considered an outlier. Outliers are detected by identifying sudden changes in the uncalibrated operating data at consecutive time points. A sudden change in the data is characterized by a difference from the previous data point exceeding a preset sudden change threshold. Assuming a preset mutation threshold of 100 rpm (for rotor speed measurements), if the rotor speed measurement at a given moment differs from the previous moment by more than 100 rpm, this data point is flagged as a suspected outlier. This mutation data point is then subjected to contextual correlation analysis. The variance of a preset number of data points before and after the mutation point (e.g., five before and five after) is calculated and compared with the historical mean variance. If the deviation exceeds a preset threshold (derived from historical data statistics), the mutation data point is identified as an outlier. For outliers, linear interpolation of the preceding and following data points is used to replace them. For example, for an abnormal rotor speed value, a reasonable replacement value is calculated by linear interpolation of the preceding and following normal rotor speed values. The replaced uncalibrated operating data is then subjected to wavelet transform denoising. This wavelet transform decomposes the data into different frequency bands, extracts the noise components in these frequency bands, and filters them out of the original signal to generate the operating data series for training. Simultaneously, the laboratory calibration data is converted into a standard flow reference series, which serves as the target value for model training.

[0082] Step S230 : constructing an initial parameter generation model, wherein the initial parameter generation model includes a temperature compensation branch, a pressure compensation branch, and a flow rate smoothing branch connected in series.

[0083] In step S240, the training operation data sequence is input into the initial parameter generation model, and a predicted temperature compensation coefficient is generated through the temperature compensation branch, a predicted pressure compensation weight is generated through the pressure compensation branch, and a predicted flow rate smoothing coefficient is generated through the flow rate smoothing branch.

[0084] Step S250, calculating the mean square error between the predicted temperature compensation coefficient and the true value of the temperature compensation in the standard flow reference sequence to obtain a temperature loss value, and calculating the cross entropy loss value between the predicted pressure compensation weight and the true value of the pressure compensation to obtain a pressure loss value.

[0085] Step S260, performing weighted summation on the temperature loss value, the pressure loss value, and the cosine similarity between the predicted flow rate smoothing coefficient and the true flow rate smoothing value to obtain a total training loss value, and performing backpropagation optimization on the initial parameter generation model based on the total training loss value until convergence.

[0086] In this embodiment, in the temperature compensation branch, for example, the temperature gradient values ​​in the training operating data sequence are processed using a multilayer perceptron network. The temperature gradient values ​​are input into the first hidden layer of the multilayer perceptron network for nonlinear feature transformation, resulting in a first hidden layer feature vector. This first hidden layer contains multiple neurons, each of which performs a nonlinear transformation on the input temperature gradient values ​​using a specific activation function (e.g., ReLU function). The first hidden layer feature vector is then input into the second hidden layer of the multilayer perceptron network for feature dimensionality reduction, resulting in a second hidden layer feature vector. This second hidden layer has fewer neurons than the first hidden layer, and the first hidden layer feature vector is processed using a specific weight matrix and activation function to obtain a more compact feature representation. Finally, the second hidden layer feature vector is concatenated with the external ambient temperature measurement to generate a temperature fusion feature. This temperature fusion feature is then input into the output layer of the multilayer perceptron network to generate a predicted temperature compensation coefficient.

[0087] In the pressure compensation branch, the pressure stability levels in the training data sequence are converted into a one-hot encoded vector. For example, if the pressure stability levels are classified into stable, medium, and unstable, the corresponding one-hot encoded vector for medium pressure stability is [0, 1, 0]. This one-hot encoded vector is then input into the one-dimensional convolutional layer of the convolutional neural network for local feature extraction. The convolution kernel in the convolutional layer slides over the one-hot encoded vector with a set step size, performing a convolution operation to extract key local features. The convolution output features are then max-pooled, for example, by taking the maximum value within a small local region, to extract key pressure pattern features. These key pressure pattern features are then input into the fully connected layer for probability distribution mapping. The fully connected layer uses multiple neurons and a weight matrix to map the input key pressure pattern features into a probability distribution vector. Finally, the output of the fully connected layer is processed using a softmax activation function to generate a class probability distribution corresponding to the predicted pressure compensation weight. The class with the highest probability in the class probability distribution is used as the predicted pressure compensation weight.

[0088] The flow rate smoothing branch includes a temporal attention mechanism module. The flow rate association values ​​in the training data sequence are input into the query vector generator of the temporal attention mechanism module to generate a query vector for the current time node. This query vector reflects the importance weight of the current flow rate association value within the entire time series. An attention weight is then calculated between the query vector and the key vectors of the historical time nodes. This attention weight represents the degree of correlation between the current flow rate association value and the historical flow rate association values. Based on this attention weight, the value vectors of the historical time nodes are weighted and summed to obtain a context-aware flow rate feature representation. This context-aware flow rate feature representation integrates information from the current and historical flow rate association values. Finally, this context-aware flow rate feature representation is input into a sigmoid activation function to generate a predicted flow rate smoothing coefficient between 0 and 1. This predicted flow rate smoothing coefficient is used to suppress short-term fluctuations in the flow rate data.

[0089] The temperature loss value is calculated by calculating the mean squared error (MSE) between the predicted temperature compensation coefficient and the true temperature compensation value in the standard flow reference sequence. For example, if the predicted temperature compensation coefficient is 0.9 and the true temperature compensation value is 0.95, the temperature loss value is calculated according to the MSE formula. Simultaneously, the cross-entropy loss between the predicted pressure compensation weight and the true pressure compensation value is calculated to obtain the pressure loss value. The total training loss value is obtained by weightedly summing the cosine similarity of the predicted flow rate smoothing coefficient and the true flow rate smoothing value. This total training loss value combines the prediction errors for temperature, pressure, and flow rate. Based on the total training loss value, the initial parameter generation model is back-propagated and optimized. By adjusting the weight parameters in the model, the total training loss value is gradually reduced until convergence. Convergence means that the error between the model's prediction results and the standard flow reference sequence has reached an acceptable minimum. At this point, model training is complete, and a calibration parameter set can be accurately generated for flow calibration in liquid flow measurement scenarios in chemical enterprises.

[0090] In a possible implementation, step S220 includes:

[0091] Step S221 : identifying mutation data points at consecutive time nodes in the uncalibrated operating data, wherein the mutation data point is characterized in that the difference between the mutation data point and the previous data point is greater than a preset mutation threshold.

[0092] When measuring the flow rate of chemical solutions, uncalibrated operating data includes various measurements, such as flow rate, pressure, and temperature, which change over time. For example, consider temperature measurement data. Suppose, when measuring the flow rate of a chemical solution in a pipeline, temperature data is continuously collected over a period of time. For example, from 9:00 AM to 10:00 AM on October 1, 2023, temperature values ​​are collected every 10 seconds. Under normal circumstances, temperature changes should be relatively smooth, but sudden changes can occur due to factors such as temporary sensor failures or external interference. Assuming a preset sudden change threshold of 2 degrees Celsius (this threshold is determined based on extensive historical data and the normal range of chemical solution temperature fluctuations), if the temperature measured at 9:10:00 is 30 degrees Celsius and at 9:10:10 is 33 degrees Celsius, the difference of 3 degrees Celsius exceeds the preset sudden change threshold. Therefore, the temperature data point at 9:10:10 is identified as a sudden change.

[0093] Step S222 : performing context correlation analysis on the mutation data point, calculating the variance of a preset number of data points before and after the mutation data point, and comparing the degree of deviation of the variance with the historical variance mean.

[0094] Assume the preset number of data points is five before and after the sudden change at 9:10:10. Specifically, the variance of the five data points before and after the sudden change at 9:10:10 (i.e., the nine data points between 9:10:00 and 9:10:50, excluding the sudden change) is calculated. A specific variance calculation method is used to obtain the variance of these nine data points. Simultaneously, the mean historical variance under similar normal operating conditions is calculated based on historical data. Assume the mean historical variance is 0.5, while the currently calculated variance is 1.5. A deviation threshold is set, for example, 0.8 (this threshold is determined based on data stability requirements and historical data statistics). Since 1.5 - 0.5 = 1, and 1 is greater than 0.8, the deviation exceeds the threshold and the sudden change is marked as an outlier. This outlier is replaced by linear interpolation of the preceding and following data points. This linear interpolation calculation is performed using the temperature values ​​at 9:10:00 and 9:10:20 (the first normal data point after the sudden change). Assume that the temperature at 9:10:00 is 30 degrees Celsius and the temperature at 9:10:20 is 31 degrees Celsius. Through linear interpolation, the reasonable temperature value at the outlier position of 9:10:10 is 30.5 degrees Celsius, and this value is used to replace the original outlier.

[0095] Step S223: When the deviation degree is greater than a set threshold, the mutation data point is marked as an outlier, and the outlier is replaced by linear interpolation of the previous and next data points.

[0096] Step S224 , performing wavelet transform denoising processing on the replaced uncalibrated operating data, extracting noise components of different frequency bands and filtering them out from the original signal, and generating the training operating data sequence.

[0097] In chemical flow measurement, noise in uncalibrated operating data can come from a variety of sources, such as sensor electronic noise and subtle interference from the surrounding pipeline environment. Wavelet transform is a mathematical method that decomposes signals into distinct frequency bands. For replaced measurement data such as temperature, pressure, and flow rate, wavelet transform is used to decompose these data into different frequency bands. For example, temperature data may be decomposed into a low-frequency component (reflecting slowly changing temperature trends) and a high-frequency component (which may contain noise and some rapid temperature fluctuations). The noise components in these frequency bands are then extracted and filtered out of the original signal. If significant high-frequency noise is detected in a certain frequency band, the noise component in this band can be removed by setting an appropriate threshold or filtering algorithm, resulting in cleaner temperature data. The same process is performed for pressure and flow rate data. After this process, all uncalibrated operating data is cleaned, generating a training data sequence that more accurately reflects the actual operating conditions of chemical solution flow measurement, providing a high-quality data foundation for subsequent model training.

[0098] In a possible implementation, the temperature compensation branch includes a multilayer perceptron network, and step S240 includes:

[0099] Step S241: Input the temperature gradient value in the training operation data sequence into the first hidden layer of the multi-layer perceptron network for nonlinear feature transformation to obtain a first hidden layer feature vector.

[0100] Step S242: Input the first hidden layer feature vector into the second hidden layer of the multilayer perceptron network to perform feature dimensionality reduction processing to obtain a second hidden layer feature vector.

[0101] Step S243: Concatenate the second hidden layer feature vector with the external environment temperature measurement value to generate a temperature fusion feature, and input the temperature fusion feature into the output layer of the multi-layer perceptron network to generate the predicted temperature compensation coefficient.

[0102] The temperature compensation branch includes a multilayer perceptron network, which generates predicted temperature compensation coefficients. The temperature gradient values ​​in the training data sequence are input into the first hidden layer of the multilayer perceptron network for nonlinear feature transformation, resulting in the first hidden layer eigenvector. In the chemical solution flow measurement scenario, the temperature gradient values ​​in the training data sequence reflect the time-varying temperature difference between the chemical solution and the pipe wall. The first hidden layer of the multilayer perceptron network contains multiple neurons, each with its own weight and bias. When the temperature gradient value is input into the first hidden layer, each neuron performs calculations based on a specific activation function. For example, if the activation function is ReLU (RectifiedLinearUnit), for the input temperature gradient value, the neuron will calculate the ReLU (temperature gradient value * weight + bias). Through the calculations of multiple neurons, the input temperature gradient value undergoes a nonlinear transformation, resulting in the first hidden layer eigenvector. This first hidden layer eigenvector is no longer a simple temperature gradient value; after the nonlinear transformation, it contains more complex features representing the relationship between the temperature gradient and other factors.

[0103] The first hidden layer feature vector is input into the second hidden layer of the multilayer perceptron network for feature dimensionality reduction, resulting in a second hidden layer feature vector. The second hidden layer has fewer neurons than the first. Each element in the first hidden layer feature vector is multiplied by the weight of a second hidden layer neuron, then added with the bias. The result is an activation function (such as the ReLU function) to generate the second hidden layer feature vector. This process, due to the reduced number of neurons in the second hidden layer, achieves feature dimensionality reduction, compressing and refining the information in the first hidden layer feature vector to extract more representative and critical features. For example, if the first hidden layer feature vector has 10 elements, after calculation in the second hidden layer (assuming it has 5 neurons), the resulting second hidden layer feature vector will have only 5 elements. These 5 elements can more effectively represent the relationship between the temperature gradient value and the flow measurement.

[0104] The second hidden layer feature vector is concatenated with the measured external ambient temperature to generate a fused temperature feature. This fused temperature feature is then fed into the output layer of the multilayer perceptron network to generate the predicted temperature compensation coefficient. In chemical solution flow measurement scenarios, the measured external ambient temperature is also a significant factor. Assume that an ambient temperature sensor is installed around the pipeline and measures an external ambient temperature of 25 degrees Celsius. This external ambient temperature value is concatenated with the second hidden layer feature vector. For example, if the second hidden layer feature vector is [0.1, 0.2, 0.3, 0.4, 0.5], the resulting concatenated temperature fused feature might be [25, 0.1, 0.2, 0.3, 0.4, 0.5] (this assumes a simple concatenation; actual concatenation may vary depending on the specific network structure and data processing requirements). This fused temperature feature is then fed into the output layer of the multilayer perceptron network. The neurons in the output layer perform calculations based on previously learned weights and biases, as well as a specific activation function (such as a linear function or other suitable function), ultimately generating the predicted temperature compensation coefficient. This predicted temperature compensation coefficient reflects the degree of temperature compensation required to accurately measure the flow rate of the chemical solution under the current temperature gradient and external ambient temperature. It will play an important role in the subsequent calibration parameter generation process, and is used to perform temperature-related compensation on the flow measurement results to improve the measurement accuracy.

[0105] In one possible implementation, the pressure compensation branch includes a combined structure of a convolutional neural network and a fully connected layer, and step S240 further includes:

[0106] Step S244: convert the pressure stability level in the training operation data sequence into a one-hot encoding vector, and input the one-hot encoding vector into the one-dimensional convolution layer of the convolutional neural network for local feature extraction.

[0107] In step S245 , the convolution output features are subjected to maximum pooling processing to extract key pressure pattern features, and the key pressure pattern features are input into the fully connected layer for probability distribution mapping.

[0108] Step S246: Process the output of the fully connected layer through a softmax activation function to generate a category probability distribution corresponding to the predicted pressure compensation weight, and use the category corresponding to the highest probability in the category probability distribution as the predicted pressure compensation weight.

[0109] The flow rate smoothing branch includes a temporal attention mechanism module, and step S240 further includes:

[0110] Step S247: Input the flow rate correlation value in the training running data sequence into the query vector generator of the temporal attention mechanism module to generate a query vector for the current time node.

[0111] Step S248: Calculate the attention weight between the query vector and the key vector of the historical time node, and perform weighted summation on the value vector of the historical time node based on the attention weight to obtain a context-aware flow rate feature representation.

[0112] Step S249: Input the context-aware flow rate feature representation into a sigmoid activation function to generate the predicted flow rate smoothing coefficient between 0 and 1, wherein the predicted flow rate smoothing coefficient is used to suppress short-term fluctuations in flow rate data.

[0113] In this embodiment, the pressure compensation branch includes a combined structure of a convolutional neural network and a fully connected layer, which is used to generate predicted pressure compensation weights. First, the pressure stability level in the training data sequence is converted into a one-hot encoded vector, and this one-hot encoded vector is input into the one-dimensional convolutional layer of the convolutional neural network for local feature extraction. In the process of measuring the flow rate of chemical solutions in pipelines, the pressure stability level is a quantitative description of the pressure stability of the liquid flow in the pipeline. For example, based on the previous analysis of the frequency of pressure fluctuation peaks and pressure changes, the pressure stability level is divided into three levels: stable, moderately stable, and unstable. When the pressure stability level is moderately stable, it is converted into a one-hot encoded vector. If stable, moderately stable, and unstable are assigned to three positions in the one-hot encoded vector, then the one-hot encoded vector corresponding to moderately stable is [0, 1, 0]. This one-hot encoded vector is input into the one-dimensional convolutional layer of the convolutional neural network. The one-dimensional convolutional layer contains multiple convolution kernels, each with a specific size and weight value. The convolution kernel slides on the one-hot encoded vector according to a set step size to perform the convolution operation. For example, with a convolution kernel size of 3 and a stride of 1, when sliding over the one-hot encoded vector [0, 1, 0], the kernel multiplies the elements in the vector and then sums them to produce the convolution output feature. This convolution output feature extracts local features from the one-hot encoded vector. These local features reflect the characteristic information of the pressure stability level at different local locations, which helps in the subsequent identification of pressure pattern characteristics.

[0114] Max pooling is performed on the convolutional output features to extract key pressure pattern features. These features are then fed into a fully connected layer for probability distribution mapping. The convolutional output features obtained above may contain multiple values. Max pooling selects the maximum value within a small local region as the representative value. For example, the convolutional output features are divided into several small regions, and the maximum value within each region is selected. These maximum values ​​are combined to form the key pressure pattern feature. This key pressure pattern feature extracts the most significant information from the convolutional output features and better represents the key pattern in the pressure stability level. This key pressure pattern feature is then fed into a fully connected layer. The fully connected layer consists of multiple neurons, each with a connection weight to each element of the input key pressure pattern feature. Using these connection weights, the fully connected layer maps the key pressure pattern features to a probability distribution vector. This probability distribution vector represents the probability of different pressure compensation weight categories.

[0115] The output of the fully connected layer is processed using a softmax activation function to generate a class probability distribution corresponding to the predicted pressure compensation weight. The class corresponding to the highest probability in the class probability distribution is used as the predicted pressure compensation weight. The softmax activation function converts the numerical output of the fully connected layer into probability values, so that the sum of these probability values ​​is 1. For example, after processing with the softmax activation function, the resulting class probability distribution is [0.2, 0.5, 0.3], corresponding to different pressure compensation weight categories. The class with the highest probability is 0.5, which is assumed to be the medium pressure compensation weight category. This medium pressure compensation weight category is then determined as the predicted pressure compensation weight. This predicted pressure compensation weight reflects the weight required to pressure compensate the flow measurement results at the current pressure stability level, based on the relationship between pressure and flow learned from historical data and the model. It plays an important role in the subsequent flow calibration process, correcting for flow measurement deviations caused by pressure factors.

[0116] The flow rate smoothing branch includes a temporal attention mechanism module, which generates the predicted flow rate smoothing coefficient. The flow rate correlation values ​​from the training data sequence are input into the query vector generator of the temporal attention mechanism module to generate a query vector for the current time node. In the chemical solution flow measurement scenario, the flow rate correlation values ​​include information such as the rotor speed measurement and blade vibration amplitude measurement of the target flow meter. These flow rate correlation values ​​vary at different time nodes, reflecting the dynamic changes in liquid flow rate. When the flow rate correlation values ​​are input into the query vector generator of the temporal attention mechanism module, the query vector generator processes them based on its internal parameters and algorithms. For example, the query vector generator may include multiple linear transformation layers and activation function layers. After the flow rate correlation values ​​are calculated through these layers, a query vector for the current time node is generated. This query vector contains the importance weight of the current flow rate correlation value within the entire time series and reflects the relationship between the current flow rate correlation value and the flow rate correlation values ​​at other time nodes.

[0117] Attention weights are calculated between the query vector and the key vectors of each historical time node. Based on the attention weights, the value vectors of each historical time node are weighted and summed to obtain a context-aware flow velocity feature representation. In the temporal attention mechanism, each historical time node has a corresponding key vector and value vector. The key vector is a feature representation of the flow velocity associated value at the historical time node, while the value vector contains additional information related to the flow velocity associated value (such as environmental factors related to flow velocity). The similarity between the query vector and the key vector of each historical time node is calculated (for example, using a dot product operation). These similarities are then converted into attention weights using a softmax function. These attention weights represent the degree of correlation between the current flow velocity associated value and the flow velocity associated value of each historical time node. Based on these attention weights, the value vectors of the historical time nodes are weighted and summed. For example, assuming there are three historical time nodes with corresponding attention weights of 0.3, 0.5, and 0.2, and the corresponding value vectors of the historical time nodes are [1, 2, 3], [4, 5, 6], and [7, 8, 9], respectively, then the context-aware flow rate feature representation obtained by weighted summation is 0.3*[1, 2, 3]+0.5*[4, 5, 6]+0.2*[7, 8, 9]=[3.4, 4.7, 5.6]. This context-aware flow rate feature representation integrates information such as the current and historical flow rate correlation values ​​and related environmental factors, and can more comprehensively reflect the actual flow rate.

[0118] The context-aware flow velocity feature representation is input into a sigmoid activation function to generate a predicted flow velocity smoothing coefficient between 0 and 1. This coefficient is used to suppress short-term fluctuations in flow velocity data. The sigmoid activation function converts the context-aware flow velocity feature representation into a value between 0 and 1. For example, inputting [3.4, 4.7, 5.6] into the sigmoid activation function yields a value, let's say 0.8. This predicted flow velocity smoothing coefficient of 0.8 indicates the degree of smoothing applied to the flow velocity data. When flow velocity data exhibits short-term fluctuations, multiplying the flow velocity data by this predicted flow velocity smoothing coefficient effectively suppresses these fluctuations, making the flow velocity data smoother and more stable, thereby improving flow measurement accuracy. In chemical solution flow measurement scenarios, the flow of liquid in a pipeline can be affected by various factors (such as slight opening and closing of valves and slight vibrations of pumps), resulting in short-term fluctuations in flow velocity. This predicted flow velocity smoothing coefficient can compensate for these fluctuations, making the final flow measurement more reliable.

[0119] Figure 2 FIG. 1 shows the hardware structure of a flow meter online calibration system 100 for implementing the above-mentioned flow meter online calibration method provided by an embodiment of the present invention. Figure 2 As shown, the flow meter online calibration system 100 may include a processor 110 , a machine-readable storage medium 120 , a bus 130 , and a communication unit 140 .

[0120] The machine-readable storage medium 120 can store data and / or instructions. In some embodiments, the machine-readable storage medium 120 can store data obtained from an external terminal. In some embodiments, the machine-readable storage medium 120 can store data and / or instructions used by the flow meter online calibration system 100 to execute or use to complete the exemplary methods described herein.

[0121] During the specific implementation process, one or more processors 110 execute computer executable instructions stored in the machine-readable storage medium 120, so that the processor 110 can execute the flow meter online calibration method of the above method embodiment. The processor 110, the machine-readable storage medium 120 and the communication unit 140 are connected through the bus 130, and the processor 110 can be used to control the sending and receiving actions of the communication unit 140.

[0122] The specific implementation process of the processor 110 can refer to the various method embodiments executed by the above-mentioned flow meter online calibration system 100. The implementation principles and technical effects are similar and will not be repeated here in this embodiment.

[0123] In addition, an embodiment of the present invention further provides a readable storage medium, in which computer-executable instructions are preset. When a processor executes the computer-executable instructions, the above-mentioned flow meter online calibration method is implemented.

[0124] It should be noted that in order to simplify the description of the present invention and thus help understand one or more embodiments of the invention, in the foregoing description of the embodiments of the present invention, multiple features are sometimes combined into one embodiment, figure or description thereof.

Claims

1. A flow meter online calibration method, characterized in that: The method comprises: Collecting a real-time operating data set of a target flow meter in a pipeline operating environment, the real-time operating data set including dynamic monitoring data generated by the target flow meter during fluid transportation and environmental status data corresponding to the pipeline operating environment; Performing data preprocessing on the real-time operation data set to obtain a preprocessed operation data sequence, wherein the preprocessed operation data sequence is used to describe an operation state change trend of the target flow meter at multiple consecutive time nodes; Inputting the preprocessed operating data sequence into a pre-trained calibration parameter generation model to generate a calibration parameter set that matches the current operating state of the target flow meter, the calibration parameter set including a flow correction coefficient, an environmental compensation factor, and an error adjustment threshold; Adjusting the measurement output data of the target flow meter in real time based on the calibration parameter set to generate a calibrated flow measurement result, and comparing and verifying the calibrated flow measurement result with a preset standard flow reference value; When the deviation between the calibrated flow measurement result and the standard flow reference value is greater than the error adjustment threshold, regenerating an updated calibration parameter set and iteratively performing the real-time adjustment until the deviation is within the error adjustment threshold; The calibration parameter generation model includes a multi-level parameter generation unit, and the pre-processed operation data sequence is input into the pre-trained calibration parameter generation model to generate a calibration parameter set that matches the current operation state of the target flow meter, including: calling a first-level parameter generation unit to perform nonlinear mapping processing on the temperature gradient value in the preprocessed operation data sequence to generate a primary temperature compensation factor, and weightedly fusing the primary temperature compensation factor with the fluid internal temperature measurement value to obtain a final temperature compensation coefficient; calling the second-level parameter generation unit to discretize and classify the pressure stability levels in the preprocessed operation data sequence, determine the pressure operating condition category to which the current pipeline operation environment belongs, and match the corresponding pressure compensation weight from a preset pressure-flow relationship table based on the pressure operating condition category; calling the third-level parameter generation unit to perform sliding window mean filtering on the flow velocity correlation values ​​in the preprocessed operation data sequence to eliminate instantaneous flow velocity fluctuation noise, and calculating a flow velocity smoothing coefficient based on the filtered flow velocity correlation values; The final temperature compensation coefficient, the pressure compensation weight, and the flow rate smoothing coefficient are normalized to generate a flow correction coefficient in the calibration parameter set, and the product of the pressure compensation weight and the flow rate smoothing coefficient is determined as the environmental compensation factor.

2. The flowmeter online calibration method according to claim 1, characterized in that: The real-time operation data set includes a plurality of operation sub-data of the flow meter within a preset time period, the operation sub-data including temperature monitoring sub-data, pressure monitoring sub-data, and flow rate monitoring sub-data. The data preprocessing of the real-time operation data set to obtain a preprocessed operation data sequence includes: Extracting the outer wall temperature measurement value and the fluid internal temperature measurement value of the pipe where the target flow meter is located from the temperature monitoring sub-data, and performing a difference calculation between the outer wall temperature measurement value and the fluid internal temperature measurement value to obtain a temperature gradient change sequence; identifying, for the pressure monitoring sub-data, a pressure fluctuation peak in the fluid flow direction in the pipeline operating environment, and generating a pressure stability evaluation index based on the occurrence frequency of the pressure fluctuation peak; For the flow rate monitoring sub-data, obtaining a rotor speed measurement value and a blade vibration amplitude measurement value of the target flow meter, and performing time-series alignment on the rotor speed measurement value and the blade vibration amplitude measurement value to generate a flow rate correlation feature vector; The temperature gradient change sequence, the pressure stability evaluation index and the flow rate associated characteristic vector are spliced ​​according to time nodes to generate the preprocessed operation data sequence, wherein the data unit corresponding to each time node in the preprocessed operation data sequence includes the temperature gradient value, the pressure stability level and the flow rate associated value.

3. The flow meter online calibration method according to claim 1, characterized in that: The step of adjusting the measurement output data of the target flow meter in real time based on the calibration parameter set to generate a calibrated flow measurement result includes: Obtaining a raw flow measurement value output by the target flow meter in an uncalibrated state, and extracting an acquisition timestamp corresponding to the raw flow measurement value; matching corresponding temperature gradient values, pressure stability levels, and flow rate correlation values ​​from the preprocessed operation data sequence according to the acquisition timestamp; Inputting the temperature gradient value into the compensation function corresponding to the final temperature compensation coefficient to obtain a temperature correction value, and inputting the pressure stability level into the adjustment function corresponding to the pressure compensation weight to obtain a pressure correction value; The original flow measurement value, the temperature correction value, and the pressure correction value are substituted into a preset flow superposition formula to obtain a preliminary calibrated flow value, and the preliminary calibrated flow value is multiplied by the flow velocity smoothing coefficient to generate the calibrated flow measurement result.

4. The flow meter online calibration method according to claim 3, characterized in that: The comparing and verifying the calibrated flow measurement result with a preset standard flow reference value includes: Acquire a historical standard flow data set that matches the current pipeline operating environment from a remote calibration database, wherein the historical standard flow data set includes reference flow values ​​calibrated in a laboratory under the same environmental conditions; Calculating the absolute deviation between the calibrated flow measurement result and the reference flow value, and calculating a moving average of all absolute deviations within a preset time window; When the moving average value is greater than the error adjustment threshold, triggering a calibration parameter update instruction, wherein the calibration parameter update instruction is used to instruct to regenerate the updated calibration parameter set; The flow correction coefficient and the environmental compensation factor in the updated calibration parameter set are dynamically weighted to generate a new error adjustment threshold, and the new error adjustment threshold is fed back to the calibration parameter generation model for parameter self-optimization.

5. The flow meter online calibration method according to claim 4, characterized in that: The pre-trained calibration parameter generation model is trained by the following steps: Acquire a historical pipeline operation data set, wherein the historical pipeline operation data set includes uncalibrated operation data collected by multiple flow meters under different operating conditions and corresponding laboratory calibration data; performing noise filtering and outlier removal on the uncalibrated operating data to generate a training operating data sequence, and converting the laboratory calibration data into a standard flow reference sequence; Constructing an initial parameter generation model, wherein the initial parameter generation model includes a temperature compensation branch, a pressure compensation branch, and a flow rate smoothing branch connected in series; Inputting the training operation data sequence into the initial parameter generation model, generating a predicted temperature compensation coefficient through the temperature compensation branch, generating a predicted pressure compensation weight through the pressure compensation branch, and generating a predicted flow rate smoothing coefficient through the flow rate smoothing branch; Calculating the mean square error between the predicted temperature compensation coefficient and the true value of the temperature compensation in the standard flow reference sequence to obtain a temperature loss value, and calculating the cross entropy loss value between the predicted pressure compensation weight and the true value of the pressure compensation to obtain a pressure loss value; The temperature loss value, the pressure loss value, and the cosine similarity of the predicted flow rate smoothing coefficient and the true flow rate smoothing value are weightedly summed to obtain a total training loss value, and the initial parameter generation model is back-propagated and optimized based on the total training loss value until convergence.

6. The flow meter online calibration method according to claim 5, characterized in that: The performing noise filtering and outlier removal on the uncalibrated operating data to generate an operating data sequence for training includes: Identifying mutation data points at consecutive time nodes in the uncalibrated running data, wherein the mutation data point is characterized by a difference between the mutation data point and a previous data point being greater than a preset mutation threshold; Performing contextual relevance analysis on the mutation data point, calculating the variance of a preset number of data points before and after the mutation data point, and comparing the degree of deviation of the variance value from the historical variance mean; When the deviation is greater than a set threshold, the mutation data point is marked as an outlier, and the outlier is replaced by linear interpolation of the previous and next data points; The replaced uncalibrated operating data is subjected to wavelet transform denoising processing, noise components of different frequency bands are extracted and filtered out from the original signal, and the training operating data sequence is generated.

7. The flow meter online calibration method according to claim 6, characterized in that: The temperature compensation branch includes a multi-layer perceptron network, and generating a predicted temperature compensation coefficient through the temperature compensation branch includes: Inputting the temperature gradient value in the training operation data sequence into the first hidden layer of the multilayer perceptron network for nonlinear feature transformation to obtain a first hidden layer feature vector; Inputting the first hidden layer feature vector into the second hidden layer of the multilayer perceptron network for feature dimensionality reduction processing to obtain a second hidden layer feature vector; The second hidden layer feature vector is concatenated with the external ambient temperature measurement value to generate a temperature fusion feature, and the temperature fusion feature is input into the output layer of the multi-layer perceptron network to generate the predicted temperature compensation coefficient.

8. The flow meter online calibration method according to claim 7, characterized in that: The pressure compensation branch includes a combined structure of a convolutional neural network and a fully connected layer, and generating a predicted pressure compensation weight through the pressure compensation branch includes: Converting the pressure stability level in the training operation data sequence into a one-hot encoded vector, and inputting the one-hot encoded vector into a one-dimensional convolutional layer of the convolutional neural network for local feature extraction; Performing maximum pooling processing on the convolution output features to extract key pressure pattern features, and inputting the key pressure pattern features into the fully connected layer for probability distribution mapping; Processing the output of the fully connected layer through a softmax activation function to generate a category probability distribution corresponding to the predicted pressure compensation weight, and taking the category corresponding to the highest probability in the category probability distribution as the predicted pressure compensation weight; The flow rate smoothing branch includes a temporal attention mechanism module, and the generation of a predicted flow rate smoothing coefficient through the flow rate smoothing branch includes: Inputting the flow rate correlation value in the training running data sequence into the query vector generator of the temporal attention mechanism module to generate a query vector for the current time node; Calculating an attention weight between the query vector and the key vector of the historical time node, and performing a weighted summation on the value vector of the historical time node based on the attention weight to obtain a context-aware flow rate feature representation; The context-aware flow rate feature representation is input into a sigmoid activation function to generate the predicted flow rate smoothing coefficient between 0 and 1, wherein the predicted flow rate smoothing coefficient is used to suppress short-term fluctuations in flow rate data.

9. A flow meter online calibration system, characterized in that: The flow meter online calibration system includes a processor and a memory, the memory is connected to the processor, the memory is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the memory to implement the flow meter online calibration method described in any one of claims 1 to 8.

Images