Intelligent flow distribution adjustment method and system for parallel operation of multi-column system

Abstract

The application provides a multi-column system parallel operation flow distribution intelligent adjustment method and system, relates to the flow regulation technical field, and the method comprises the following steps: collecting the current opening degree dynamic parameters of the valve, and constructing a time displacement trajectory equation to predict the inertia lag position of the valve based on the initial opening degree instruction of the electric regulating valve, so as to obtain a feedforward correction opening degree instruction; collecting the outlet header pressure fluctuation data and calculating the pressure disturbance compensation amount, integrating the pressure disturbance compensation amount and the feedforward correction opening degree instruction, and obtaining a final execution opening degree instruction. The application realizes the regulation of flow distribution, and improves the system flow balance and operation stability.

Description

Technical Field

[0001] This invention relates to the field of flow regulation technology, and in particular to a method and system for intelligent flow distribution regulation when a multi-column system is running in parallel. Background Technology

[0002] In the field of fluid control involving the parallel operation of multiple columns, such as large-scale chromatographic separation and multi-stage filtration, maintaining a balanced distribution of flow in each branch is usually an important consideration for the stable operation of the system. Currently, most such systems tend to adopt a feedback control strategy based on the PID (proportional-integral-derivative) algorithm, that is, by monitoring the actual flow at the outlet of each column and comparing it with the set target, the generated deviation signal is used to drive the electric control valve to adjust the opening.

[0003] However, under certain complex actual operating conditions, due to differences in the state of the filling medium inside each parallel column, subtle differences in pipeline layout, or changes in local flow resistance, the fluid flowing through each branch may exhibit certain dynamic time delay characteristics. Existing conventional control methods, when dealing with such problems, sometimes treat each branch as a relatively independent unit for adjustment, or largely rely on static decoupling logic. Taking the multi-column parallel operation of a certain type of industrial adsorption separation device as an example, when the system load fluctuates and the flow needs to be redistributed, if there are differences in the mass transfer resistance inside each adsorption column, the lag time of the fluid response will... They may not be entirely the same; in this case, if a traditional PID controller mainly outputs commands based on instantaneous flow error, the prediction mechanism for dynamic time delay deviation may not be sufficient. Sometimes, the adjustment of the slower-responding branch may not be fully matched with the actual demand. This may cause the valve of that branch to be adjusted too large in the initial stage, and when the fluid finally overcomes the lag and arrives, it may cause a certain degree of overshoot, which in turn prompts the system to make reverse correction. In some cases, this adjustment process may prolong the time required for the flow between each column to reach equilibrium, and may even cause slight fluctuations in the outlet main pipe pressure. Summary of the Invention

[0004] This invention provides a method and system for intelligent adjustment of flow distribution when a multi-column system is running in parallel, which enables rapid adjustment of flow distribution and improves the system's flow balance and operational stability.

[0005] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows:

[0006] Firstly, a method for intelligent adjustment of flow distribution during parallel operation of a multi-column system, the method comprising:

[0007] Step 1: Calculate the dynamic time delay deviation of each column based on the state dataset and construct the synchronized state vector; map the synchronized state vector to the actual pose of the virtual rigid body in multi-dimensional space, compare it with the target pose of the virtual rigid body, calculate the attitude correction amount, and use the attitude correction amount to correct the synchronized state vector to obtain the corrected synchronized state vector.

[0008] Step 2: Input the corrected synchronized state vector into the pre-trained adaptive control model to obtain the theoretical flow correction amount for each branch; Based on the real-time pressure and flow correlation characteristics of each column in the corrected synchronized state vector, construct the instantaneous flow opening mapping relationship of each electric regulating valve under the current working condition, match the theoretical flow correction amount of each branch with the instantaneous flow opening mapping relationship to obtain the valve opening change amplitude and the initial opening command of each electric regulating valve.

[0009] Step 3: Collect the current dynamic parameters of the valve opening, and based on the initial opening command of the electric regulating valve, construct the time displacement trajectory equation to predict the valve's inertial lag position and obtain the feedforward correction opening command; collect the outlet main pipe pressure fluctuation data and calculate the pressure interference compensation amount, integrate the pressure interference compensation amount and the feedforward correction opening command to obtain the final execution opening command;

[0010] Step 4: Based on the final execution opening command, collect the actual instantaneous flow data of each column and compare it with the preset target flow distribution standard to calculate the flow distribution deviation, update the system operating information, and achieve balanced flow distribution.

[0011] Secondly, the intelligent flow distribution adjustment system for multi-column systems operating in parallel includes:

[0012] The correction module is used to calculate the dynamic time delay deviation of each column based on the state dataset and construct the synchronized state vector; map the synchronized state vector to the actual pose of the virtual rigid body in multi-dimensional space, compare it with the target pose of the virtual rigid body, calculate the attitude correction amount, and use the attitude correction amount to correct the synchronized state vector to obtain the corrected synchronized state vector.

[0013] The control module is used to input the corrected synchronized state vector into the pre-trained adaptive control model to obtain the theoretical flow correction amount of each branch; based on the real-time pressure and flow correlation characteristics of each column in the corrected synchronized state vector, the instantaneous flow opening mapping relationship of each electric regulating valve under the current working condition is constructed, and the theoretical flow correction amount of each branch is matched with the instantaneous flow opening mapping relationship to obtain the valve opening change amplitude and the initial opening command of each electric regulating valve.

[0014] The correction module is used to collect the dynamic parameters of the valve's current opening degree, and based on the initial opening degree command of the electric regulating valve, construct a time displacement trajectory equation to predict the valve's inertial lag position and obtain the feedforward correction opening degree command; collect the outlet main pipe pressure fluctuation data and calculate the pressure interference compensation amount, integrate the pressure interference compensation amount and the feedforward correction opening degree command to obtain the final execution opening degree command;

[0015] The verification module is used to collect the actual instantaneous flow data of each column based on the final execution opening command, compare it with the preset target flow allocation standard to calculate the flow allocation deviation, update the system operating information, and achieve balanced flow allocation.

[0016] The above-described solution of the present invention has at least the following beneficial effects:

[0017] By constructing a virtual rigid body pose, the complex fluid coupling interference of multiple cylinders is quantified into attitude correction quantities, improving the synchronization level of the system state. A feedforward correction mechanism based on the time displacement trajectory equation is introduced to predict and compensate for the inertial lag of the electric regulating valve. Combined with an adaptive control model, the flow regulation becomes faster and more accurate. A dual control strategy of feedforward prediction and outlet pressure negative feedback fine-tuning is adopted to quickly respond to changes in the setpoint, suppress the interference caused by the pressure fluctuation of the outlet main pipe, and establish a complete closed-loop mechanism. When the flow deviation exceeds the range, the state dataset is automatically updated and the correction quantity is recalculated until the target is reached, ensuring the long-term balance of flow distribution in each branch. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating the intelligent flow distribution adjustment method for a multi-column system operating in parallel, provided by an embodiment of the present invention.

[0019] Figure 2 This is a schematic diagram of an intelligent flow distribution adjustment system for a multi-column system operating in parallel, provided by an embodiment of the present invention. Detailed Implementation

[0020] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0021] like Figure 1 As shown, embodiments of the present invention propose an intelligent flow distribution adjustment method for multi-column systems operating in parallel, the method comprising the following steps:

[0022] Step 1: Calculate the dynamic time delay deviation of each column based on the state dataset and construct the synchronized state vector; map the synchronized state vector to the actual pose of the virtual rigid body in multi-dimensional space, compare it with the target pose of the virtual rigid body, calculate the attitude correction amount, and use the attitude correction amount to correct the synchronized state vector to obtain the corrected synchronized state vector.

[0023] Step 2: Input the corrected synchronized state vector into the pre-trained adaptive control model to obtain the theoretical flow correction amount for each branch; Based on the real-time pressure and flow correlation characteristics of each column in the corrected synchronized state vector, construct the instantaneous flow opening mapping relationship of each electric regulating valve under the current working condition, match the theoretical flow correction amount of each branch with the instantaneous flow opening mapping relationship to obtain the valve opening change amplitude and the initial opening command of each electric regulating valve.

[0024] Step 3: Collect the current dynamic parameters of the valve opening, and based on the initial opening command of the electric regulating valve, construct the time displacement trajectory equation to predict the valve's inertial lag position and obtain the feedforward correction opening command; collect the outlet main pipe pressure fluctuation data and calculate the pressure interference compensation amount, integrate the pressure interference compensation amount and the feedforward correction opening command to obtain the final execution opening command;

[0025] Step 4: Based on the final execution opening command, collect the actual instantaneous flow data of each column and compare it with the preset target flow distribution standard to calculate the flow distribution deviation, update the system operating information, and achieve balanced flow distribution.

[0026] In this embodiment of the invention, by constructing a virtual rigid body pose, the complex fluid coupling interference of multiple cylinders is quantified into an attitude correction quantity, thereby improving the synchronization level of the system state. A feedforward correction mechanism based on the time displacement trajectory equation is introduced to predict and compensate for the inertial lag of the electric regulating valve. Combined with an adaptive control model, the flow regulation becomes faster and more accurate. A dual control strategy of feedforward prediction and outlet pressure negative feedback fine-tuning is adopted to quickly respond to changes in the set value, suppress the interference caused by the pressure fluctuation of the outlet main pipe, and establish a complete closed-loop mechanism. When the flow deviation exceeds the range, the state dataset is automatically updated and the correction quantity is recalculated until the target is reached, ensuring the long-term balance of the flow distribution of each branch.

[0027] In a preferred embodiment of the present invention, collecting inlet and outlet pressure, temperature, and instantaneous flow data of each column's electrical sensing device to generate a status dataset may include:

[0028] In this embodiment of the invention, before data acquisition, all electrical sensing devices (pressure transmitters, temperature sensors, and instantaneous flow meters) of all column branches are uniformly calibrated. The pressure transmitters are calibrated for zero point and range according to the system process pressure range to eliminate the inherent offset of different transmitters. Temperature sensors, such as PT100 RTDs and K-type thermocouples, are calibrated using a standard temperature source to correct the linearity error of temperature measurement. Instantaneous flow meters, such as electromagnetic / vortex flow meters, are calibrated using a standard flow calibration device to ensure that the deviation between the measured flow value and the actual value is within the allowable range of the process. The acquisition clocks of all electrical sensing devices are synchronized with the unified reference clock of the system, and a unified sampling trigger rule is set to avoid misalignment of the data time dimension due to the asynchronous clocks of various sensors. The acquisition points of each column are determined: the pressure data acquisition point is the straight pipe section after the column inlet valve and before the outlet valve; the temperature data acquisition point is the main flow area of ​​the inlet and outlet pipes; and the instantaneous flow acquisition point is the full-flow section of the branch pipe, ensuring that the acquired data can truly reflect the actual operating status of the column.

[0029] For each column, the raw pressure values ​​are collected in real time via electrically sensing pressure transmitters installed at its inlet and outlet ends. During the collection process, the system reference clock stamp corresponding to each pressure value is recorded synchronously, and the column number to which the data belongs is marked, such as column 1 inlet pressure P1-in, column 1 outlet pressure P1-out, column 2 inlet pressure P2-in, and column 2 outlet pressure P2-out, to avoid data confusion between different columns and different measuring points. Simultaneously, the raw inlet and outlet temperatures are collected synchronously using electrically sensing temperature sensors on the inlet and outlet pipelines of each column, and are also linked to the system reference clock stamp. The column numbers are used, such as column 1 inlet temperature T1-in and column 1 outlet temperature T2-out. For minor temperature fluctuations caused by pipeline heat dissipation and local eddies, no correction is made for now, and the original collected values ​​are retained. They will be adjusted in the subsequent preprocessing stage in combination with fluid transmission characteristics. The instantaneous flow rate at each sampling moment is collected through the electrical sensing instantaneous flow meter of each column branch, and the clock stamp and column number are bound, such as column 1 instantaneous flow rate Q1 and column 2 instantaneous flow rate Q2. During the collection, it is necessary to avoid instantaneous flow rate jumps caused by pipeline vibration and fluid cavitation, and retain the complete trend of the original collected data.

[0030] Based on the reasonable range of the process and the rules of data continuity, invalid data is screened and removed. For example, when the inlet pressure of a column exceeds the allowable range of 0.1~2.0MPa, the instantaneous flow rate is negative (flowmeter malfunction), or the temperature deviates from the normal process range by more than ±20℃, it is judged as an anomaly. At the same time, abrupt data jumps in continuous sampling, such as the column outlet pressure jumping from 0.8MPa to 5.0MPa instantly, and there is no transition trend between three adjacent sampling points, are judged to be an anomaly caused by electromagnetic interference and are also removed to avoid errors. Erroneous data affects the validity of the dataset. To address data gaps that appear after removing outliers, the missing data is filled in by combining the continuous trend of similar data in the same column. For example, if the instantaneous flow rate of a column is missing at 10:00:00.500, it is filled in with the flow rate values ​​at 10:00:00.400 (0.8 m³ / h) and 10:00:00.600 (0.82 m³ / h) according to the linear trend, thus ensuring the continuity of the dataset in the time dimension and avoiding breakpoints in the calculation of dynamic time delay deviation.

[0031] All collected data were standardized to process standard units: pressure was standardized to MPa, temperature to ℃, and instantaneous flow rate to m³ / h. This eliminated dataset confusion caused by different sensor calibration units. Using the system's reference clock stamp as the core index, the preprocessed data from all columns were integrated into a three-dimensional structure of time-column-parameter. The row dimension represents the sampling time, and the column dimension is divided into two layers: the first layer is the column number (column 1, column 2... column n), and the second layer contains the corresponding inlet pressure, outlet pressure, inlet temperature, outlet temperature, and instantaneous flow rate parameters for each column, forming a structured raw state data matrix. Metadata identifiers were added to the integrated dataset, including the collection period, the current system operating condition label, the sensor calibration batch, and outlier removal / completion records, ensuring dataset traceability. Simultaneously, the dataset was divided into blocks for storage according to process periods. 10% of the sampling points within the dataset were randomly selected to verify the consistency between the original sensor collection records and the integrated data, confirming the absence of parameter mismatches, clock stamp offsets, unit errors, etc. After verification, the final state dataset was formed.

[0032] In a preferred embodiment of the present invention, step 1 above, which involves calculating the dynamic time delay deviation of each column based on the state dataset and constructing a synchronized state vector; mapping the synchronized state vector to the actual pose of a virtual rigid body in multidimensional space and comparing it with the target pose of the virtual rigid body, calculating the attitude correction amount, and using the attitude correction amount to correct the synchronized state vector to obtain the corrected synchronized state vector, may include:

[0033] In this embodiment of the invention, step 110 involves extracting the original time-series data recorded by the electrical sensing devices of each column from the state dataset, and identifying the acquisition time tag corresponding to each pressure, temperature, and flow rate data; comparing the acquisition time tag with the system's unified reference clock, calculating the initial time offset of each column's data relative to the reference time, and using the initial time offset as the preliminary dynamic time delay deviation; specifically, this includes: extracting the original time-series data of all state parameters of each column one by one according to the column number from the preprocessed, structured, integrated, and verified state dataset, specifically including the inlet pressure of each column, Complete data collection of five parameters—outlet pressure, inlet temperature, outlet temperature, and instantaneous flow rate—ensuring no raw data point is missed. The acquisition time tag for each raw data point is identified; this tag represents the local clock time synchronously recorded by each electrical sensing device (pressure transmitter, temperature sensor, instantaneous flow meter) during data acquisition. A unified system reference clock is used as the sole time reference standard. This unified reference clock is calibrated using an NTP network clock to ensure millisecond-level time accuracy. All electrical sensing devices in the column have been synchronized with this reference clock, as recorded in the status dataset. During the data collection phase, clock synchronization is completed. The acquisition time tag of each piece of raw data for each column is compared one by one with the corresponding time of the system's unified reference clock. The time difference between the two is calculated, and this time difference is the initial time offset of the data relative to the system reference time. Since all electrical sensing devices in the same column collect data synchronously, their acquisition time tag time offsets are basically consistent. Therefore, the average of the initial time offsets of all data in that column is taken as the initial dynamic time delay deviation of that column, which is used to characterize the initial lag of the column's data relative to the system reference time. The specific calculation method is as follows: the initial dynamic time delay deviation of a certain column... The delay deviation is the average of the (collection time label - corresponding time of the system's unified reference clock) of all data in the column. For example, column 1 has 10 traffic data, and the difference between the collection time label of each data and the corresponding time of the system reference clock is 0.048 seconds, 0.050 seconds, 0.052 seconds...0.051 seconds, respectively. After adding these 10 differences and dividing by 10, we get the average value of 0.050 seconds. This 0.050 seconds is the initial dynamic delay deviation of column 1. Similarly, the initial dynamic delay deviation of all columns is calculated in the same way to ensure that each column has a unique initial dynamic delay deviation value.

[0034] Step 111: Resample the original time series data according to the preliminary dynamic time delay deviation. Map the state parameters collected at different physical times for each cylinder to the same virtual time axis through interpolation operations to obtain an intermediate state data group after time alignment. Specifically, it includes: setting a unified virtual time axis for the system. The sampling interval of this virtual time axis is the same as that of the original data. For example, if the original data is collected every 100 milliseconds, the sampling interval of the virtual time axis is also 100 milliseconds. The starting time of the virtual time axis is the same as the starting collection time of the system's unified reference clock, and the ending time is the same as the ending collection time of the original data, ensuring that the virtual time axis can completely cover the collection period of all original data.

[0035] For each cylinder, determine the correspondence between the original data of the cylinder and the virtual time axis according to its preliminary dynamic time delay deviation. If the preliminary dynamic time delay deviation of a certain cylinder is positive, it means that the data of this cylinder lags behind the system reference time, and its original data needs to be shifted backward by the corresponding duration on the time axis; if the preliminary dynamic time delay deviation is negative, it means that the data of this cylinder is ahead of the system reference time, and its original data needs to be shifted forward by the corresponding duration on the time axis. After the shift, through linear interpolation operations, map the state parameters collected at different physical times for this cylinder to each sampling moment of the virtual time axis, ensuring that there is a corresponding state parameter value for this cylinder at each virtual sampling moment. The specific method of linear interpolation operation is as follows: for a certain sampling moment t on the virtual time axis, if there is no original collection data of the corresponding cylinder at this moment, find the two adjacent original collection times t1 and t2 (t1 < t < t2) before and after this moment, and the corresponding original parameter values x1 (at t1 moment) and x2 (at t2 moment), then the interpolation parameter value x at t moment is x = x1 + (t - t1) × (x2 - x1) ÷ (t2 - t1). For example, the preliminary dynamic time delay deviation of cylinder 2 is 0.030 seconds, the sampling moments of the virtual time axis are 10:00:00.000, 10:00:00.100, 10:00:00.200, the original data collection times of this cylinder are 10:00:00.030 (inlet pressure 0.78 MPa), 10:00:00.130 (inlet pressure 0.80 MPa), 10:00:00.230 (inlet pressure 0.82 MPa). Calculate the inlet pressure value x at virtual moment 10:00:00.000 through linear interpolation: x = 0.78 + (0.000 - 0.030) × (0.80 - 0.78) ÷ (0.130 - 0.030) = 0.774 MPa; similarly, calculate the parameter values of this cylinder at all sampling moments of the virtual time axis to complete the data alignment of this cylinder. Repeat the above operations to complete the resampling and time alignment of all cylinders, and integrate the state parameters of all cylinders at each sampling moment of the virtual time axis to form an intermediate state data group after time alignment.

[0036] Step 112: Based on the intermediate state data set, calculate the rate of change of values ​​between adjacent sampling points to obtain the actual rate of change of each type of state parameter for each column; based on the actual rate of change and the theoretical rate of change range determined by the fluid transmission characteristics in the pipeline, identify the residual error caused by signal transmission fluctuations; specifically, this includes: splitting the intermediate state data set by column number to obtain an independent data subset for each column; then, for each column's data subset, further splitting it by parameter type (inlet pressure, outlet pressure, inlet temperature, outlet temperature, instantaneous flow rate) to obtain the time series data for each type of parameter for that column; for the time series data of each type of parameter, calculate the rate of change of values ​​between two adjacent sampling points, where the actual rate of change of a parameter = parameter value of the next sampling point - parameter value of the previous sampling point. The parameter values ​​are then calculated by dividing the difference by the time interval between the two sampling points, i.e., the sampling interval of the virtual time axis, such as 100 milliseconds = 0.1 seconds. This gives the actual rate of change of the parameter within that time period, with the unit determined according to the parameter type. For example, in the instantaneous flow time series data of column 3, two adjacent sampling points are 10:00:00.100 (0.8 m³ / h) and 10:00:00.200 (0.82 m³ / h), and the time interval between the two sampling points is 0.1 seconds. Therefore, the actual rate of change of the instantaneous flow rate within that time period is (0.82 - 0.8) ÷ 0.1 = 0.2 m³ / h·second. Similarly, the actual rates of change of the inlet pressure, outlet pressure, inlet temperature, and outlet temperature of the column, as well as the actual rates of change of all other parameters of the column, are calculated.

[0037] Based on the fluid transport characteristics in the pipeline, the theoretical rate of change range for each type of state parameter is predetermined. This range is determined by combining pipeline and fluid characteristic parameters such as pipeline diameter, pipeline length, fluid viscosity, and fluid density, and is calculated using fluid mechanics formulas. This ensures that the theoretical rate of change range covers the reasonable variation range of each parameter during normal system operation, avoiding misjudgment of normal fluctuations as errors. For example, the theoretical rate of change range for instantaneous flow rate is determined to be -0.05 m³ / h·s to 0.05 m³ / h·s, the theoretical rate of change range for inlet pressure is determined to be -0.02 MPa / s to 0.02 MPa / s, and the theoretical rate of change range for temperature is determined to be -0.1℃ / s to 0.1℃ / s. The actual rate of change for each parameter of each column is compared one by one with the corresponding theoretical rate of change range. If the actual rate of change is within the theoretical rate of change range, then it is judged as... The rate of change is assumed to be normal fluctuation with no residual error. If the actual rate of change exceeds the theoretical rate of change range, the difference between the actual rate of change and the theoretical rate of change boundary value is calculated. This difference is the residual error caused by signal transmission fluctuation. The unit of the residual error is consistent with the unit of the rate of change of the corresponding parameter. For example, if the actual rate of change of instantaneous flow in column 3 is 0.06 m³ / h·s in a certain time period, which exceeds the upper limit of the theoretical rate of change of 0.05 m³ / h·s, then the residual error = 0.06 - 0.05 = 0.01 m³ / h·s. If the actual rate of change is -0.07 m³ / h·s, which exceeds the lower limit of the theoretical rate of change of -0.05 m³ / h·s, then the residual error = -0.07 - (-0.05) = -0.02 m³ / h·s. For each type of parameter of each column, the residual error is identified in the above manner, and finally the residual error values ​​of each type of parameter of each column are obtained.

[0038] Step 113: Based on the residual error, perform a correction calculation on the initial dynamic delay deviation to obtain the final dynamic delay deviation reflecting the degree of lag in the response of each cylinder; based on the final dynamic delay deviation, perform phase fine-tuning on the intermediate state data group after time alignment to obtain the fine-tuned state parameters of each cylinder at the same logical moment; specifically, the correction calculation adopts a linear correction method, and the final dynamic delay deviation = initial dynamic delay deviation + (residual error × correction coefficient). Since the residual error of different parameters of the same cylinder may be different, the average value of the residual error of all parameters of the cylinder is taken and substituted into the above formula to calculate the final dynamic delay deviation; for example, the initial dynamic delay deviation of cylinder 1 is 0.050 seconds, and the various types of... The residual errors of the parameters are 0.01 m³ / h·second (flow rate), 0.002 MPa / second (inlet pressure), 0.003 MPa / second (outlet pressure), 0.001℃ / second (inlet temperature), and 0.001℃ / second (outlet temperature). The average residual error is (0.01+0.002+0.003+0.001+0.001)÷5=0.0034 m³ / h·second (converted to flow rate unit; residual errors of other parameters are converted proportionally). The correction factor is 0.005. Therefore, the final dynamic time delay deviation is 0.050+(0.0034×0.005)=0.050017 seconds, rounded to four decimal places as 0.0500 seconds.

[0039] After obtaining the final dynamic delay deviation of each column, the intermediate state data group obtained in step 111 is phase-fine-tuned based on this to further eliminate the delay deviation and ensure that the state parameters of all columns correspond to the same logical time. The fine-tuning rule is as follows: if the final dynamic delay deviation of a column is positive, it means that the response of the column is still lagging, and all parameters of the column in the intermediate state data group need to be fine-tuned backward by the corresponding time on the virtual time axis (fine-tuning time = final dynamic delay deviation); if the final dynamic delay deviation is negative, it means that the response of the column is ahead, and all parameters of the column need to be fine-tuned forward by the corresponding time on the virtual time axis. During fine-tuning, the linear interpolation operation in step 111 is still used to ensure that the phase deviation is correct. To ensure the continuity of parameter values ​​after adjustment and avoid data breakpoints, for example, if the final dynamic delay deviation of column 1 is 0.002 seconds (positive), then all parameters of this column in the intermediate state data group are finely adjusted backward by 0.002 seconds, that is, the parameters at time 10:00:00.100 on the virtual time axis are finely adjusted to time 10:00:00.102. The parameter values ​​at each sampling time after fine adjustment are calculated by linear interpolation. If the final dynamic delay deviation of column 4 is -0.001 seconds (negative), then all its parameters are finely adjusted forward by 0.001 seconds to ensure that the parameters of this column are aligned with the parameters of other columns at the same logical time. After completing the phase fine adjustment of all columns, the finely adjusted state parameters of all columns at the same logical time are obtained.

[0040] Step 114: Based on the fine-tuned state parameters, encapsulate the inlet and outlet pressure, temperature, and instantaneous flow rate values ​​of each column at the same logical moment in a preset order to construct a synchronized state vector. Specifically, this includes: determining a preset encapsulation order, which is a fixed and uniform rule. First, arrange the columns in ascending order of their column numbers. For each column, encapsulate its fine-tuned state parameters in the order of inlet pressure, outlet pressure, inlet temperature, outlet temperature, and instantaneous flow rate. That is, first encapsulate the inlet pressure, outlet pressure, inlet temperature, outlet temperature, and instantaneous flow rate of column 1, then encapsulate the above five types of parameters for column 2, and so on, until all column parameters are encapsulated. During encapsulation, arrange the five types of parameter values ​​of each column in sequence to form a one-dimensional array. This one-dimensional array is the synchronized state vector, and the dimension of the vector is equal to the number of columns multiplied by 5 (5 types per column). (Parameters), for example, the system contains 3 columns. At the same logical moment, the fine-tuned state parameters of each column are as follows: Column 1: inlet pressure 0.8MPa, outlet pressure 0.6MPa, inlet temperature 45℃, outlet temperature 43℃, instantaneous flow rate 0.85m³ / h; Column 2: inlet pressure 0.82MPa, outlet pressure 0.61MPa, inlet temperature 44℃, outlet temperature 42℃, instantaneous flow rate 0.84m³ / h; Column 3: inlet pressure 0.79MPa, outlet pressure 0.59MPa, inlet temperature 46℃, outlet temperature 44℃, instantaneous flow rate 0.86m³ / h; then the synchronized state vector encapsulated in a preset order is [0.8, 0.6, 45, 43, 0.85, 0.82, 0.61, 44, 42, 0.84, 0.79, 0.59, 46, 44, 0.86]. In this vector, the position of each value corresponds to a fixed column and parameter type.

[0041] For each logical moment, a synchronized state vector must be constructed in the manner described above to ensure that there is a corresponding synchronized state vector at each moment during system operation, which is used to reflect the overall operating status of the system in real time. At the same time, during the encapsulation process, the value and position of each parameter must be checked to avoid parameter mismatch or position reversal, and to ensure the accuracy of the synchronized state vector.

[0042] Step 115: Based on the synchronized state vector, map the instantaneous flow rate values ​​of each column to the coordinate origin position of the virtual rigid body in multi-dimensional space; map the inlet and outlet pressure difference data to the pitch and yaw angles of the rigid body; and map the temperature data to the roll angle of the rigid body, thus constructing the actual pose of the virtual rigid body representing the current system operating state. Specifically, this includes: determining the coordinate origin position of the virtual rigid body; extracting the instantaneous flow rate values ​​of each column in the synchronized state vector; multiplying the instantaneous flow rate value of each column by a spatial scaling factor to complete the spatial coordinate transformation; and transforming the data... The value serves as the coordinate origin of the virtual rigid body in multidimensional space. The spatial scaling factor is determined based on the total system flow range, ranging from 0.1 to 1. Its function is to convert the flow rate value into coordinate values ​​suitable for multidimensional spatial representation, ensuring that the coordinate values ​​are within a reasonable range. For example, if the total system flow rate ranges from 0 to 10 m³ / h, and the spatial scaling factor is 0.1, then 1 m³ / h of flow rate corresponds to a coordinate value of 0.1. For instance, if the instantaneous flow rate of column 1 is 0.85 m³ / h, and the spatial scaling factor is 0.1, the transformed value is 0.85 × 0.1 = 0.085. The instantaneous flow rate of column 2 is 0.84 m³ / h, which is transformed to 0.084. The instantaneous flow rate of column 3 is 0.86 m³ / h, which is transformed to 0.086. These three values ​​are used as the coordinates of the origin of the virtual rigid body along the X, Y, and Z axes in three-dimensional space, respectively, i.e., the origin position is (0.085, 0.084, 0.086), thus determining the basic spatial positioning of the virtual rigid body. The pitch and yaw angles of the virtual rigid body are determined by extracting the inlet and outlet pressure data of each column from the synchronized state vector and calculating the inlet and outlet pressures of each column. The pressure difference is calculated as: Inlet / Outlet Pressure = Inlet Pressure - Outlet Pressure, in MPa. The pressure difference between the inlet and outlet of each column is multiplied by an angle conversion factor and mapped to a rotation angle relative to the origin of the virtual rigid body coordinate system. This yields the pressure rotation angle data. The angle conversion factor is determined based on the system's pressure regulation sensitivity and is set to 5 degrees / MPa. Its function is to convert the pressure difference into a rotation angle of the virtual rigid body, ensuring that changes in angle accurately reflect changes in pressure difference. For example, for every 1 MPa change in pressure difference, the rotation angle changes by 5 degrees.

[0043] Based on the obtained pressure rotation angle data, the horizontal rotation angle component (XY plane) is used as the pitch angle of the virtual rigid body, and the vertical rotation angle component (Z-axis direction) is used as the yaw angle of the virtual rigid body. This completes the attitude definition of the virtual rigid body in two dimensions. For example, the pressure difference between the inlet and outlet of column 1 is 0.2 MPa, and the mapped rotation angle is 0.2 × 5 = 1 degree, where the horizontal component is 0.6 degrees (pitch angle) and the vertical component is 0.4 degrees (yaw angle); the pressure difference between the inlet and outlet of column 2 is 0.21 MPa. a. The mapped rotation angle is 1.05 degrees, with a horizontal component of 0.63 degrees and a vertical component of 0.42 degrees. Take the average of the pitch and yaw angles of all columns as the final pitch and yaw angles of the virtual rigid body, i.e., pitch angle = (0.6 + 0.63 + 0.57) ÷ 3 = 0.6 degrees (pressure difference between inlet and outlet of column 3 is 0.18 MPa, horizontal component is 0.57 degrees after mapping), yaw angle = (0.4 + 0.42 + 0.38) ÷ 3 = 0.4 degrees (vertical component is 0.38 degrees after mapping of column 3).

[0044] The roll angle of the virtual rigid body is determined by extracting the inlet and outlet temperature data of each column from the synchronized state vector. The temperature fluctuation of each column is calculated as: Temperature fluctuation = Current temperature - System rated temperature (in °C). The system rated temperature is determined according to the system process requirements, such as 45 °C. The temperature fluctuation of each column is multiplied by an angle conversion factor (1 degree / °C, unitless) to convert it into a rotational component about the virtual rigid body's axis (Z-axis), generating the roll angle and completing the virtual rigid body's attitude information in the third dimension. For example, the inlet temperature of column 1 is 4... 5℃, temperature fluctuation = 45 - 45 = 0℃, the mapped rotation component is 0 × 1 = 0 degrees; the outlet temperature of column 1 is 43℃, temperature fluctuation = 43 - 45 = -2℃, the mapped rotation component is -2 × 1 = -2 degrees; take the average value of the mapped rotation components of the temperature fluctuations of all columns as the final roll angle of the virtual rigid body, that is, roll angle = (0 - 2 + 1) ÷ 3 = -0.33 degrees (the inlet and outlet temperature fluctuations of column 2 are -1℃ and -2℃ respectively, and those of column 3 are 1℃ and -1℃ respectively); set the origin position (X, Y, Z) and pitch angle of the virtual rigid body coordinates determined above. ), yaw angle ( ), roll angle ( The parameters are fused to form a six-dimensional parameter set, which represents the actual pose of the virtual rigid body, characterizing the current operating state of the multi-column system. This set can be represented as (X, Y, Z, ...). , , Each parameter corresponds to a certain operating characteristic of the system: the position of the origin corresponds to the total flow state, the pitch and yaw angles correspond to the pressure distribution state, and the roll angle corresponds to the temperature fluctuation state.

[0045] Step 116: Based on the system's preset total flow target value and the ideal distribution ratio of each branch, construct the virtual rigid body target pose representing the ideal equilibrium state in a multi-dimensional space identical to the actual pose of the constructed virtual rigid body. Specifically, this includes: determining the system's preset total flow target value, set according to system process requirements (e.g., 10 m³ / h), and the ideal flow distribution ratio of each parallel branch, determined according to system design requirements (e.g., equal distribution or distribution according to column performance, such as equal distribution for 3 columns with an ideal distribution ratio of 1:1:1); splitting the total flow target value according to the ideal distribution ratio to obtain the ideal instantaneous flow value of each column; the calculation method is: ideal instantaneous flow of a column = system total flow target value × ideal instantaneous flow of that column. The allocation ratio is as follows: for example, the total system flow target value is 10 m³ / h, and it is equally distributed among the three columns. The ideal instantaneous flow of each column is 10 × (1 / 3) ≈ 3.333 m³ / h. Using the same spatial scaling factor as in step 115, the ideal instantaneous flow values ​​of each column are transformed into spatial coordinates. The transformed values ​​are used as the coordinate origin of the virtual rigid body target pose. For example, if the spatial scaling factor is 0.1, the ideal instantaneous flow of each column is 3.333 m³ / h, and the transformed value is 3.333 × 0.1 ≈ 0.333. Therefore, the coordinate origin of the virtual rigid body target is (0.333, 0.333, 0.333), which indicates that the total system flow has reached the target value and the flow of each branch is equally distributed.

[0046] Based on the preset inlet and outlet pressure difference under ideal system operating conditions, and determined according to system design requirements (e.g., 0.2 MPa), the same angle conversion coefficient (5 degrees / MPa) as in step 115 is used to map the preset inlet and outlet pressure difference to ideal rotation angles. The horizontal component is the ideal pitch angle, and the vertical component is the ideal yaw angle. Since the inlet and outlet pressure differences of each column are consistent during ideal system operation, the ideal pitch and yaw angles are fixed values. For example, with a preset inlet and outlet pressure difference of 0.2 MPa, the mapped rotation angle is 1 degree, and the ideal pitch angle is... 0.6 degrees, the ideal yaw angle is 0.4 degrees, characterizing the balanced pressure distribution of the system; under ideal operating conditions, the temperature of each column should be maintained at the rated temperature value without temperature fluctuation, therefore the temperature fluctuation is 0℃. Using the same angle conversion coefficient as in step 115 (1 degree / ℃), the mapped rotation component is 0 degrees, that is, the virtual rigid body target roll angle is 0 degrees, characterizing the system temperature stability without abnormal fluctuation; the virtual rigid body target coordinate origin position, ideal pitch angle, ideal yaw angle, and 0-degree roll angle are fused to form a six-dimensional parameter set (X target, Y target, Z target, ... Target, Target, The set of virtual rigid body target poses represents the ideal equilibrium operating state of the system. It is in the same multi-dimensional space as the actual virtual rigid body pose and can be directly superimposed and compared in space.

[0047] Step 117: Spatially superimpose and compare the actual pose of the virtual rigid body with the target pose of the virtual rigid body, calculate the spatial difference between the two, and decompose the spatial difference into a rotation component representing the degree of imbalance in the flow ratio of each branch and a translation component representing the degree of deviation in the total flow of the system. Specifically, this includes: aligning the origins of the actual pose and the target pose of the virtual rigid body to the same spatial coordinate system, comparing the corresponding pose parameters one by one, that is, comparing the coordinate differences of the three axes X, Y, and Z of the origin, as well as the pitch angle. Yaw angle Roll angle The angle difference is used to calculate the overall spatial difference between the actual pose of the virtual rigid body and the target pose using a spatial distance algorithm. This difference comprehensively reflects the degree of deviation between the system's operating state and the ideal state. The specific formula is as follows: ,in These are the parameters corresponding to the actual pose of the virtual rigid body; These are the corresponding parameters for the virtual rigid body target pose. For example, if the actual pose parameters of the virtual rigid body are (0.085, 0.084, 0.086, 0.6°, 0.4°, -0.33°), and the target pose parameters are (0.333, 0.333, 0.333, 0.6°, 0.4°, 0°), then the overall spatial difference... ≈ The value ≈0.541 indicates a deviation between the actual operating state and the ideal state of the system, mainly due to flow distribution and temperature fluctuations. The calculated overall spatial difference is decomposed into two independent components based on its source, corresponding to different types of imbalance. The rotation component is composed of the pitch, yaw, and roll angle differences between the actual and target poses of the virtual rigid body. It primarily characterizes the degree of imbalance in the flow ratio of each parallel branch. The larger the angle difference, the larger the rotation component, indicating a greater deviation between the flow distribution ratio of each branch and the ideal ratio, and a more uneven pressure distribution and temperature fluctuation. The translation component is composed of the coordinate differences along the X, Y, and Z axes between the actual and target poses of the virtual rigid body. It primarily characterizes the degree of deviation between the total system flow and the target total flow. The larger the coordinate difference, the larger the translation component, indicating a greater deviation of the total system flow from the target value, and that the overall flow supply does not meet process requirements. This decomposition clearly distinguishes the two core causes of flow distribution imbalance (proportional imbalance and total volume deviation).

[0048] Step 118: Based on the rotation component, calculate the rotation matrix describing the uneven flow distribution; simultaneously, based on the translation component, calculate the translation vector describing the total deviation; fuse the rotation matrix and translation vector to obtain the attitude correction amount that simultaneously reflects the proportional misalignment and the total deviation; specifically, the rotation matrix, a 3×3 matrix used to quantify the deviation corresponding to the rotation component, describes the rotation correction amount of the virtual rigid body around each axis, with the specific formula as follows:

[0049] ,

[0050] For example =0.6°−0.6°=0° (0rad), =0.4°−0.4°=0° (0rad), =−0.33°−0°=−0.33° (≈-0.00576rad), substituting into the formula, we obtain the rotation matrix. , This matrix indicates that the current rotational deviation mainly stems from minor deviations in the roll angle, requiring targeted correction. The translation vector, a 3×1 vector used to quantify the deviation corresponding to the translation components, describes the translation correction amount of the virtual rigid body's coordinate origin. The specific formula is as follows: ,For example, =0.085−0.333=−0.248mm, =0.084−0.333=−0.249mm, =0.086−0.333=−0.247mm, then the translation vector This vector indicates that the current virtual rigid body coordinate origin has a negative deviation in all three axes, meaning the total system flow is lower than the target value, requiring positive translation correction. The calculated rotation matrix will then be used. With translation vector A fusion calculation is performed, incorporating weighting coefficients, to generate an attitude correction quantity that simultaneously reflects both the flow ratio imbalance and the total flow deviation. The specific formula is as follows: ,in This is the weighting coefficient for the rotational component, ranging from 0 to 1. It is used to adjust the correction weight for flow ratio misalignment; the higher the priority of flow ratio control, the better. The larger the value; This is the weighting coefficient for the translation component, ranging from 0 to 1. It is used to adjust the correction weight for the total flow deviation; the higher the priority of total flow control, the better. The larger the value, the stronger the constraint. + =1, ensuring the reasonableness of the correction amount and avoiding over-correction or under-correction. For example, setting the flow ratio control priority higher than the total flow control priority, taking... =0.6, =0.4, substitute into the rotation matrix Translation vector The attitude correction amount is obtained. After calculation, the attitude correction amount is approximately [[0.501, 0.00346, 0], [0.00346, 0.5004, 0], [0, 0, 0.5012]]. This correction amount includes both rotational correction for flow ratio imbalance and translational correction for total flow deviation.

[0051] Step 119: Based on the attitude correction amount, perform correction calculations on each parameter value in the synchronized state vector to obtain the corrected parameter values; based on the corrected parameter values, repackage them in a preset order to obtain the corrected synchronized state vector; specifically, this includes: setting corresponding parameter correction coefficients according to different parameter types. These coefficients are determined based on the system's adjustment sensitivity and process requirements. Their core function is to convert the attitude correction amount into the corresponding parameter correction value, ensuring that the corrected parameter values ​​meet the process's allowable range. The specific parameter correction coefficients are set as follows: the pressure parameter (inlet pressure, outlet pressure) correction coefficient is 0.01 MPa / unit correction amount, that is, every 1 unit of attitude correction amount corresponds to a 0.01 MPa pressure parameter correction; the temperature parameter (inlet... The correction factor for temperature (outlet temperature) is 0.1℃ / unit correction, meaning that for every unit of attitude correction, there is a 0.1℃ correction for the temperature parameter; the correction factor for flow rate parameter (instantaneous flow rate) is 0.005m³ / h / unit correction, meaning that for every unit of attitude correction, there is a 0.005m³ / h correction for the flow rate parameter; the correction rule is: corrected parameter value = original parameter value + (value of the dimension corresponding to the attitude correction × parameter correction factor), where the value of the dimension corresponding to the attitude correction is determined according to the parameter type, with pressure parameter corresponding to the element in the first row and first column of the rotation matrix, temperature parameter corresponding to the element in the second row and second column of the rotation matrix, and flow rate parameter corresponding to the element in the third row and third column of the rotation matrix. Combined with the corresponding components of the translation vector, the specificity of the correction is ensured.

[0052] For example, in step 114, the original parameter value of the inlet pressure of column 1 in the synchronized state vector is 0.8 MPa, the value of the dimension corresponding to the attitude correction is 0.501, and the pressure parameter correction coefficient is 0.01 MPa / unit correction. Therefore, the corrected inlet pressure of column 1 = 0.8 + (0.501 × 0.01) = 0.8 + 0.00501 = 0.80501 MPa, rounded to four decimal places, is 0.8050 MPa; the original parameter value of the instantaneous flow rate of column 1 is 0.85 m³ / h, the value of the dimension corresponding to the attitude correction is 0.5012, and the flow parameter correction coefficient is... With a unit correction of 0.005 m³ / h, the corrected instantaneous flow rate of column 1 is 0.85 + (0.5012 × 0.005) = 0.85 + 0.002506 = 0.852506 m³ / h, rounded to four decimal places as 0.8525 m³ / h. Similarly, the corrected values ​​of all parameters in the synchronization state vector are calculated in the same way. The corrected synchronization state vector is then repackaged and generated. All the corrected parameter values ​​are repackaged in the same preset order as in step 114 to form a one-dimensional array, which is the corrected synchronization state vector.

[0053] By extracting raw data, calculating and correcting dynamic time delay deviations, it ensures that all column parameters are aligned at the same logical moment. By mapping multi-dimensional system operating parameters to virtual rigid body poses, abstract problems such as flow distribution imbalance, uneven pressure distribution, and temperature fluctuations are transformed into intuitive spatial pose deviations, thereby improving the stability of multi-column system operation.

[0054] In a preferred embodiment of the present invention, step 2 above involves inputting the corrected synchronized state vector into a pre-trained adaptive control model to obtain the theoretical flow correction for each branch; based on the real-time pressure and flow correlation characteristics of each column in the corrected synchronized state vector, constructing the instantaneous flow opening mapping relationship of each electric regulating valve under the current operating condition; matching the theoretical flow correction for each branch with the instantaneous flow opening mapping relationship to obtain the valve opening change amplitude and the initial opening command of each electric regulating valve, which may include:

[0055] In this embodiment of the invention, step 220 involves loading the corrected synchronized state vector as feature input data for the current operating condition into a pre-trained adaptive control model, and using the decoupled flow response law learned in the adaptive control model to extract features from the feature input data, thereby obtaining the extracted feature data. Specifically, this includes collecting historical operating data of the multi-column parallel system under all operating conditions, which mainly includes three types of data: first, the synchronized state vector of each column after dynamic time delay correction under different fluid viscosities and load conditions; second, the target value of the total system flow and the ideal flow distribution ratio of each column corresponding to each operating condition; and third, the flow change and pressure fluctuation data of the electric regulating valve under different opening degrees. Outliers are removed, and missing data is filled in by aligning it by time, ultimately forming a complete training dataset containing feature input data and flow correction labels, covering low / medium / high load and load fluctuation scenarios. The model adopts an end-to-end structure of feature encoder + fusion matching mode + policy inference mode. The feature encoder consists of 1D-CNN + LSTM, and its core function is to extract deep features of the synchronized state vector, separate the flow coupling interference between each column, and learn the decoupled flow response law. The fusion matching mode consists of fully connected layers to realize the dynamic fusion of extracted features and total flow target value. The policy inference mode includes an adaptive parameter adjustment layer and an output layer, which are responsible for adjusting the control policy parameters. The model is pre-trained using a publicly available multi-column fluid control dataset. This allows the model to learn the basic correlations between flow rate and pressure / temperature, as well as the coupling interference characteristics between multiple columns, and to grasp the basic characteristics of decoupled flow response. The pre-trained model is then fine-tuned using a pre-constructed full-condition training dataset. The focus is on training the model's ability to extract features from the synchronized state vector after dynamic delay compensation, its ability to fuse and match with the total flow target value, and its adaptive adjustment ability of control strategy parameters under different operating conditions. During fine-tuning, batch training is used to progressively optimize model parameters, ensuring the model can adapt to the present invention. Specific operating conditions require a composite loss function consisting of three parts to constrain the model training process and ensure the accuracy of the model output. The loss function specifically includes feature extraction loss, fusion matching loss, and flow correction loss. The feature extraction loss measures the deviation between the features extracted by the model and the actual decoupled response pattern. The fusion matching loss measures the deviation between the fusion result and the target load requirement. The flow correction loss measures the deviation between the theoretical flow correction amount output by the model and the actual required correction amount. The parameters of each layer of the model are continuously optimized until the loss function converges to the preset threshold. At this point, the model has fully mastered the decoupled flow response pattern and the operating condition adaptation logic, and the pre-training is completed.

[0056] After the data is loaded, the decoupled flow response rules learned in advance in the adaptive control model are used to perform deep feature extraction on the feature input data. The specific extraction process is as follows: The model's feature encoder uses a one-dimensional convolutional neural network to extract local features from the corrected synchronized state vector, capturing the instantaneous change trends of inlet pressure, outlet pressure, and instantaneous flow rate of each column, as well as the pressure-flow correlation trend between columns. The extracted local features are then integrated temporally through a long short-term memory network to eliminate the temporal interference remaining due to dynamic delays in the data of each column, separate the flow coupling interference between columns, and extract decoupled temporal features that can reflect the independent flow response characteristics of a single column. Finally, the extracted feature data is obtained, which contains three types of core information: the independent flow response characteristics of each column, the pressure-flow coupling elimination characteristics between columns, and the flow state deviation characteristics under the current operating conditions, providing core support for subsequent fusion matching and strategy adjustment.

[0057] Step 221: The feature data is fused and matched with the system's preset total flow target value within the adaptive control model to obtain a fusion matching result reflecting the current load demand and fluid state; based on the fusion matching result, the control strategy parameters of the adaptive control model are dynamically adjusted to obtain the adjusted control strategy parameters; specifically, the fusion formula is as follows: ,in The fusion matching result is a one-dimensional feature vector that can comprehensively reflect the degree of matching between the current fluid state and the system load requirements. The closer the fusion result is to the preset standard value, the smaller the deviation between the current working condition and the target working condition. The core feature vector extracted in step 220 contains key information such as the independent flow response and coupling elimination of each column; The feature vector of the total traffic target value is the system's preset total traffic target value, which is divided according to the ideal distribution ratio of each column and mapped into a feature vector with the same dimension as the feature data. It is used to characterize the system's load demand. The feature data weighting coefficient ranges from 0.4 to 0.7 and is dynamically adjusted according to the current operating conditions. When the operating conditions are in a load fluctuation state, the value is 0.7 to prioritize the current fluid state characteristics; when the operating conditions are in a stable load state, the value is 0.4 to prioritize the system load demand. The total flow target value weight coefficient ranges from 0.3 to 0.6. It satisfies the constraint that the weight coefficients of the feature data add up to 1. The dynamic adjustment logic is the opposite of that of the feature data weight coefficients to ensure the rationality of the fusion result.

[0058] After obtaining the fusion matching result, the adaptive control model dynamically adjusts its internal control strategy parameters based on this result to ensure that the control strategy can adapt to the load requirements and fluid state of the current operating condition. The control strategy parameters include three core parameters, and the specific adjustment logic is as follows: First, the coupling suppression parameter is adjusted according to the degree of flow coupling residue between each column in the fusion matching result. If the fusion result shows a high degree of flow coupling residue between each column, it indicates that the coupling interference has not been completely eliminated. Therefore, the coupling suppression parameter is increased to strengthen the model's weight in extracting decoupling features and further suppress coupling interference. If the fusion result shows a low degree of coupling residue, the coupling suppression parameter is decreased to avoid over-suppression leading to feature loss and to ensure the accuracy of the features extracted by the model. Second, the strategy learning rate is adjusted according to the deviation between the fusion matching result and the ideal state. The control strategy parameters are adjusted as follows: First, the larger the deviation, the greater the difference between the current and target operating conditions. A larger learning rate is needed to accelerate the dynamic adaptation of the model's control strategy and quickly reduce the deviation. Second, a smaller deviation results in a smaller learning rate, preventing excessive policy adjustments that could lead to flow overshoot and ensuring the stability of policy adjustments. Third, the flow correction weight is adjusted based on the deviation between the target and current total flow. A larger deviation indicates a greater deviation from the target value, requiring a larger flow correction weight to enhance the model's response to flow correction and accelerate the overall flow's approach to the target value. A smaller deviation results in a smaller flow correction weight, ensuring the accuracy of flow correction and preventing over-correction that could cause flow fluctuations. After dynamically adjusting these three types of parameters, the adjusted control strategy parameters are obtained.

[0059] Step 222: Based on the adjusted control strategy parameters and the corrected synchronized state vector, perform inference calculations to calculate the theoretical flow correction amount required for each parallel branch to eliminate the remaining state deviation and achieve the target allocation ratio, thus obtaining the theoretical flow correction amount for each branch. Specifically, this includes: extracting the current actual instantaneous flow of each column from the corrected synchronized state vector; and calculating the target allocation instantaneous flow of each column based on the system's preset total flow target value and the ideal allocation ratio of each column. The calculation method is: target allocation instantaneous flow of a column = system total flow target value × ideal allocation ratio of that column. For example, if the system total flow target value is 10 m³ / h and the ideal allocation ratio of the three columns is 1:1:1, then the target allocation instantaneous flow of each column = 10 × The flow rate is approximately 3.333 m³ / h. The initial flow deviation for each column is calculated as follows: Initial flow deviation for a column = Target allocated instantaneous flow rate for that column - Current actual instantaneous flow rate for that column. A positive initial flow deviation indicates insufficient flow for that column, requiring an increase in flow; a negative initial flow deviation indicates excessive flow for that column, requiring a decrease in flow. Combining the flow correction weight and coupling suppression parameters in the adjusted control strategy parameters, the initial flow deviation for each column is corrected to eliminate coupling interference and residual state deviation between columns, yielding the theoretical flow correction for each branch. The specific correction formula is as follows: ,in For the first The theoretical flow correction for each column is the core result of the final output, reflecting the flow adjustment range required for that column to achieve the target allocation ratio. The flow correction weight is the flow correction weight adjusted in step 221, with a value range of 0.8 to 1.2. It is used to amplify or reduce the correction magnitude of the initial flow deviation to ensure that the flow correction strength is adapted to the current operating conditions. For the first The initial flow deviation of each column is the difference between the target instantaneous flow rate of the column calculated in the previous stage and the current actual instantaneous flow rate; This is the coupling suppression parameter, which is the adjusted coupling suppression parameter in step 221. Its value ranges from 0.1 to 0.3, and it is used to suppress the effect of other column flow deviations on the first... The coupling interference of individual columns ensures the accuracy of the theoretical flow correction. It is the sum of the initial flow deviations of the other columns, excluding the first column. The sum of the initial flow deviations of all other columns outside the first column is used to characterize the effect of other columns on the first column. The degree of coupling interference of each column.

[0060] Step 223: Based on the corrected synchronized state vector, extract the real-time pressure and flow correlation features of each column, and construct the instantaneous flow opening mapping relationship of each electric regulating valve under the current operating condition based on the real-time pressure and flow correlation features. Specifically, this includes: extracting the real-time pressure and flow correlation features of each column; extracting the real-time inlet pressure, real-time outlet pressure, and real-time instantaneous flow from the corrected synchronized state vector for each column; and then calculating the pressure-flow correlation coefficient for each column. This coefficient characterizes the degree of influence of pressure changes on flow changes. The calculation method is: pressure-flow correlation coefficient = change in real-time instantaneous flow ÷ change in real-time inlet-outlet pressure difference. The specific calculation logic is: select two adjacent sampling times... For real-time data, first calculate the instantaneous flow rate difference between the previous and subsequent moments, then calculate the inlet and outlet pressure difference between the previous and subsequent moments, where the inlet and outlet pressure difference = inlet pressure - outlet pressure. Finally, divide the instantaneous flow rate difference by the inlet and outlet pressure difference to obtain the pressure-flow rate correlation coefficient for the column. For example, if the real-time instantaneous flow rates of column 1 at two adjacent sampling moments are 0.85 m³ / h and 0.87 m³ / h, and the inlet and outlet pressure differences are 0.2 MPa and 0.22 MPa, respectively, then the instantaneous flow rate change = 0.87 - 0.85 = 0.02 m³ / h; the inlet and outlet pressure difference change = 0.22 - 0.2 = 0.02 MPa; and the pressure-flow rate correlation coefficient = 0.02 ÷ 0.02 = 1 m³ / h·MPa. -1 This indicates that for every 0.1 MPa change in the pressure difference between the inlet and outlet of the column, the instantaneous flow rate changes by 0.1 m³ / h.

[0061] A mapping relationship between the instantaneous flow rate and opening degree of each electric control valve under the current operating condition is constructed. Based on the extracted pressure-flow correlation features, and combined with the current actual opening degree and historical opening degree-flow response data of each electric control valve, a quantitative mapping relationship is established. This mapping relationship can calculate the corresponding instantaneous flow rate of the valve body based on the opening degree of the electric control valve, and conversely, it can determine the corresponding opening degree change based on the required flow rate change. The specific formula is as follows: ,in The instantaneous flow rate of the column corresponding to the electric regulating valve is the actual instantaneous flow rate of the column at a certain opening degree of the electric regulating valve. This represents the opening degree of the electric control valve, ranging from 0 to 100%. It characterizes the degree to which the electric control valve is open; the larger the opening degree, the greater the fluid flow rate. The flow rate coefficient is determined by the pressure-flow rate correlation coefficient under the current operating conditions. The larger the pressure-flow rate correlation coefficient, the better. The larger the value, the more significant the impact of the change in opening degree on the change in flow rate; This is a linear correction coefficient, determined by the electric control valve's own performance parameters and the current fluid viscosity, used to correct for deviations in the linear correlation between the valve opening and flow rate. The base flow offset is the basic leakage flow rate when the electric control valve opening is 0. It is determined by the sealing performance of the electric control valve and is usually taken to be close to 0. During the construction process, it is determined by fitting the pressure-flow correlation coefficient under the current operating conditions using historical opening-flow response data. , , The specific values ​​of the three coefficients ensure that the mapping relationship can accurately match the current working conditions. After the fitting is completed, the instantaneous flow opening mapping relationship of each electric regulating valve under the current working conditions is obtained.

[0062] Step 224: Match the theoretical flow correction of each branch with the instantaneous flow opening mapping relationship to determine the opening change required for each electric control valve to achieve the theoretical flow correction, and obtain the valve opening change amplitude of each electric control valve; superimpose the valve opening change amplitude of each electric control valve onto the current actual opening reference of the corresponding electric control valve to obtain the initial opening command of the electric control valve; specifically, this includes: determining the valve opening change amplitude of each electric control valve; for each column, substituting the theoretical flow correction of that column into the corresponding instantaneous flow opening mapping relationship, and solving in reverse to obtain the required electric control valve opening change, i.e., the valve opening change amplitude. The specific solution logic is as follows: if the theoretical flow correction of each branch is matched with the instantaneous flow opening mapping relationship, the valve opening change amplitude is obtained. If the theoretical flow correction is positive, it indicates that the column flow needs to be increased, and the corresponding electric control valve opening needs to be increased. The required opening value to reach the target flow is calculated using a mapping relationship. The difference between this opening value and the current actual opening of the electric control valve is the valve opening change range (positive value). If the theoretical flow correction is negative, it indicates that the column flow needs to be reduced, and the corresponding electric control valve opening needs to be decreased. The required opening value to reach the target flow is calculated using a mapping relationship. The difference between this opening value and the current actual opening of the electric control valve is the valve opening change range (negative value). For example, if the theoretical flow correction for column 1 is 0.1398 m³ / h, the corresponding instantaneous flow opening mapping relationship is as follows: =0.0001× ²+0.008× +0.01, the current actual opening of the electric control valve is 60%, corresponding to a current flow rate of 0.85 m³ / h. The target flow rate is 0.85 plus 0.1398, which equals 0.9898 m³ / h. Substituting the target flow rate into the mapping relationship, the required opening is approximately 68%. Therefore, the valve opening change is equal to 68 minus 60, which equals 8%, meaning the electric control valve opening needs to be increased by 8%. The valve opening change of each electric control valve is then superimposed onto the current actual opening reference of the corresponding electric control valve. The calculation method is as follows: the initial opening of the electric control valve... Initial opening command = current actual opening of electric control valve + valve opening change range. For example, if the current actual opening of the electric control valve in column 1 is 60% and the valve opening change range is 8%, then the initial opening command of the electric control valve in column 1 is 60 plus 8, which equals 68%. The theoretical flow correction of column 2 is -0.1002 m³ / h, the valve opening change range is -5%, and the current actual opening is 65%. Then the initial opening command of the electric control valve in column 2 is 65 plus (-5), which equals 60%. Through the above calculation, the initial opening command of each electric control valve is obtained.

[0063] By inferring from a pre-trained adaptive control model and constructing an opening mapping relationship based on real-time pressure-flow correlation features, the theoretical flow correction of each branch and the opening of the electric regulating valve are matched, enabling the flow of each parallel branch to reach the target distribution ratio and improving the stability of flow distribution in the multi-column parallel system.

[0064] In a preferred embodiment of the present invention, step 3 above, which involves collecting dynamic parameters of the current valve opening and constructing a time-displacement trajectory equation to predict the valve's inertial lag position based on the initial opening command of the electric regulating valve, and obtaining a feedforward correction opening command; collecting outlet main pipe pressure fluctuation data and calculating pressure interference compensation, and integrating the pressure interference compensation and the feedforward correction opening command to obtain the final execution opening command, may include:

[0065] In this embodiment of the invention, step 330 involves synchronously collecting the current real-time opening position, rated operating speed, and maximum acceleration parameters of each electric regulating valve according to the initial opening command of the electric regulating valve, thereby obtaining a set of valve dynamic parameters. Specifically, this includes: determining the collection range and triggering logic; performing dynamic parameter collection only on each electric regulating valve for which an initial opening command has been generated, ensuring that the collection object and the target of the control command are completely consistent, avoiding invalid collection and resource waste; synchronously collecting three types of core dynamic parameters: collecting the current real-time opening position of the valve through the position sensor built into the electric regulating valve (this parameter characterizes the current actual opening degree of the valve, with a value range of 0 to 100% and a collection accuracy to 0.1%); collecting the rated operating speed of the valve through a valve operating speed detection device (this parameter is the standard operating rate calibrated at the valve's factory, characterizing the maximum opening change capability that the valve can achieve per unit time); and collecting the maximum acceleration parameter of the valve through an acceleration detection module. The maximum allowable change in valve opening acceleration during operation reflects the speed and smoothness of valve action. During data acquisition, millisecond-level timestamps are used to synchronously record the acquisition time of the three types of parameters, ensuring that the acquisition sequence of the three parameters for the same electric control valve is completely consistent, eliminating the impact of timing deviations on subsequent calculation results. The validity of the acquired parameters is verified by checking whether the current real-time opening position is within the process allowable range of 0 to 100%. If it exceeds this range, it is judged as abnormal data and immediately re-acquired. The rated operating speed and maximum acceleration parameters are checked to ensure they are consistent with the factory calibration values ​​of the electric control valve. If there is a deviation and it exceeds the allowable error range, the detection device needs to be checked for malfunction and re-acquired to ensure the authenticity and accuracy of the parameters. The valve dynamic parameter set is integrated and constructed by combining the verified current real-time opening position, rated operating speed, and maximum acceleration parameters in a fixed order to form the dynamic parameter set of a single electric control valve.

[0066] Step 331: Based on the current real-time opening position in the valve dynamic parameter set as the starting point of motion, and using the rated operating speed and maximum acceleration parameters as motion constraints, combined with the target position pointed to by the initial opening command of the electric control valve, a time-displacement trajectory equation describing the entire process of the valve's movement from the starting point to the end point is constructed. Specifically, this includes: extracting core motion parameters; extracting the current real-time opening position of a single electric control valve from the valve dynamic parameter set, using it as the starting opening of the valve movement; extracting the opening value pointed to by the initial opening command of the electric control valve, using it as the target opening of the valve movement; extracting the rated operating speed as the velocity constraint condition for the valve movement; and extracting the maximum acceleration parameter as... The acceleration constraint condition for valve movement ensures that the valve movement process conforms to its own mechanical performance limitations, avoiding mechanical shock and overshoot. The valve movement is divided into stages, and the direction of valve movement is determined according to the relationship between the initial opening and the target opening. If the target opening is greater than the initial opening, the valve performs an increasing opening movement; if the target opening is less than the initial opening, the valve performs a decreasing opening movement. Regardless of the direction of movement, the valve follows a three-stage movement law of acceleration, constant speed, and deceleration. First, it accelerates to the rated operating speed with maximum acceleration, then moves at a constant speed with the rated operating speed, and finally decelerates uniformly to the target opening with maximum acceleration. This ensures smooth valve movement and avoids mechanical wear and flow fluctuations caused by excessively fast movement.

[0067] Calculate the key parameters for each motion phase: Acceleration phase (acceleration time = rated operating speed ÷ maximum acceleration), valve opening change during acceleration phase = 0.5 × maximum acceleration × acceleration time²; Deceleration phase (deceleration time = acceleration time), valve opening change during deceleration phase = acceleration phase opening change; Uniform speed phase (uniform speed phase opening change = target opening - starting opening - acceleration phase opening change - deceleration phase opening change). If the calculation result is negative, it indicates that the valve does not need to pass through the uniform speed phase and directly enters the deceleration phase from the acceleration phase; Uniform speed time equals the uniform speed phase opening change divided by the rated operating speed, i.e., uniform speed time = uniform speed phase opening change ÷ rated operating speed. Construct the time-displacement trajectory equation, separately for the three phases, uniformly using time as the independent variable and valve opening as the dependent variable. The equation is as follows: Acceleration phase (0 ≤ time ≤ acceleration time) is... The constant speed segment (acceleration time < time ≤ acceleration time + constant speed time) is... The deceleration phase (acceleration time + constant speed time < time ≤ total motion time) is... ,in For any time The corresponding theoretical valve opening represents the theoretical command position of the valve at that moment; This represents the initial valve opening, i.e., the current real-time valve opening position. This refers to the valve's maximum acceleration parameter. This refers to the change in the opening of the acceleration section; The rated operating speed of the valve; To speed up the process; The target valve opening is the opening value indicated by the initial opening command of the electric control valve. The total valve movement time is equal to the acceleration time + constant speed time + deceleration time. The above equation can describe the valve's opening change pattern from the initial opening degree to the target opening degree.

[0068] Step 332: Calculate the lag deviation between the actual position and the theoretical commanded position of the valve due to mechanical inertia and fluid resistance when the valve moves along the trajectory described by the time-displacement trajectory equation, obtaining the inertial lag position prediction result. Specifically, this includes: determining the cause of the lag deviation; during the movement of the electric regulating valve, it possesses mechanical inertia, and the mass of the valve components prevents its movement speed from instantly following the theoretical commanded speed; simultaneously, fluid resistance is generated when the fluid flows inside the valve, the magnitude of which is related to fluid viscosity and flow velocity, further hindering the valve's action response. The combined effect of these two factors causes the actual valve position to lag behind the theoretical commanded position described by the time-displacement trajectory equation, resulting in a lag deviation; calculating the lag deviation value, which is the sum of the lag deviation caused by mechanical inertia and the lag deviation caused by fluid resistance, using the following formula: ,in This is the valve inertial hysteresis deviation value, that is, the amount of hysteresis between the actual position and the theoretical commanded position; The hysteresis deviation caused by mechanical inertia is calculated by multiplying the mechanical inertia coefficient by the valve's acceleration. ,in The mechanical inertia coefficient is determined by the valve's own mass and transmission structure, and its value ranges from 0.01 to 0.05. The hysteresis deviation caused by fluid resistance is calculated by multiplying the fluid resistance coefficient by the fluid velocity. ,in This is the fluid resistance coefficient, determined by the fluid viscosity and valve diameter, with a value ranging from 0.02 to 0.06. The actual flow velocity of the fluid inside the valve can be calculated from the instantaneous flow rate of the cylinder in the corrected synchronized state vector, combined with the valve diameter (flow velocity = instantaneous flow rate ÷ valve cross-sectional area). Using the time-displacement trajectory equation, the inertial lag deviation value at any given moment is calculated. During the acceleration, constant velocity, and deceleration phases of the valve's movement, the acceleration and fluid velocity at the corresponding moments are substituted to calculate the lag deviation value at each moment, ensuring that the lag deviation prediction covers the entire valve movement process. The inertial lag position prediction result is then integrated, and the average value of the lag deviation values ​​at each moment during the entire valve movement process is taken as the final inertial lag position prediction result.

[0069] Step 333: Based on the inertial lag position prediction result, perform advance compensation on the initial opening command of the electric control valve to obtain the feedforward corrected opening command. After executing the feedforward corrected opening command, calculate the instantaneous pressure fluctuation difference between the measured pressure and the set stable pressure based on the real-time monitored pressure data of the parallel system outlet main pipe, and obtain the instantaneous pressure fluctuation difference. Specifically, this includes: generating the feedforward corrected opening command. The core purpose of feedforward correction is to offset the opening deviation caused by valve inertial lag, ensuring that the actual valve position can accurately reach the target opening. The specific calculation method is: feedforward corrected opening command = initial opening command of electric control valve + inertial lag position prediction result. For example, if the initial opening command of the electric control valve is 68% and the inertial lag position prediction result is 0.5%, then the feedforward corrected opening command = 68% + 0.5% = 68.5%. After generating the feedforward correction opening command, the command is immediately sent to each electric regulating valve to execute the feedforward correction operation. Simultaneously, the pressure data of the parallel system's outlet manifold is monitored in real time by a pressure sensor installed on the outlet manifold. The sampling frequency is consistent with the valve movement sampling frequency to ensure that the pressure data reflects the pressure change after valve opening adjustment in real time. The instantaneous pressure fluctuation difference is calculated as follows: Instantaneous pressure fluctuation difference = Measured pressure - Set stable pressure. If the instantaneous pressure fluctuation difference is positive, it indicates that the outlet manifold pressure is higher than the set value, indicating excessive pressure interference; if it is negative, it indicates that the outlet manifold pressure is lower than the set value, indicating excessive pressure interference; if it is 0, it indicates that the pressure is stable and there is no fluctuation interference. After calculation, the instantaneous pressure fluctuation difference at each sampling moment is recorded.

[0070] Step 334: Based on the instantaneous pressure fluctuation difference, convert it using a preset negative feedback gain coefficient to obtain the pressure disturbance compensation amount; integrate the pressure disturbance compensation amount with the feedforward correction opening command to obtain the final execution opening command; specifically, this includes: calculating the pressure disturbance compensation amount, based on the instantaneous pressure fluctuation difference obtained in step 333, converting the pressure fluctuation difference into the corresponding valve opening compensation amount using a preset negative feedback gain coefficient, the specific calculation formula is as follows: ,in This is the pressure disturbance compensation amount, used to compensate for the impact of pressure fluctuations on flow distribution; The negative feedback gain coefficient is determined by the system pressure regulation sensitivity and ranges from 0.1 to 0.3. Its function is to convert the pressure fluctuation difference into a reasonable opening compensation amount to avoid over-compensation or under-compensation. This is the instantaneous pressure fluctuation difference, i.e., the difference between the measured pressure calculated in step 333 and the set stable pressure. For example, if the instantaneous pressure fluctuation difference is 0.02 MPa and the negative feedback gain coefficient is 0.2, then the pressure disturbance compensation amount = 0.2 × 0.02 = 0.004%, indicating that the 0.02 MPa pressure fluctuation needs to be compensated by fine-tuning the valve opening. If the instantaneous pressure fluctuation difference is -0.01 MPa, then the pressure disturbance compensation amount = 0.2 × (-0.01) = -0.002%, corresponding to fine-tuning the valve opening in the opposite direction. The calculated pressure disturbance compensation amount is integrated with the feedforward correction opening command obtained in step 333. The specific calculation method is as follows: the final execution opening command = feedforward correction opening command + pressure disturbance compensation amount. During the integration process, it is necessary to ensure that the opening command is within the process allowable range of 0 to 100%. If the integrated opening command exceeds the range, it needs to be corrected to the corresponding boundary value (100% if it exceeds the upper limit, and 0% if it exceeds the lower limit) to avoid the valve action exceeding the mechanical limit. For example, if the feedforward correction opening command is 68.5% and the pressure interference compensation is 0.004%, then the final executed opening command = 68.5% + 0.004% = 68.504%; if the pressure interference compensation is -0.002%, then the final executed opening command = 68.5% - 0.002% = 68.498%. The final executed opening command is sent to each electric regulating valve to guide the valve to complete the action, ensuring balanced flow distribution and stable outlet main pressure in the multi-column parallel system.

[0071] By synchronously collecting dynamic parameters of the valve to construct trajectory equations, predicting inertial lag positions and performing advance compensation, and simultaneously collecting pressure fluctuation data and calculating compensation amounts, the generated final execution opening command combines feedforward compensation and negative feedback regulation functions, thereby improving the response speed of electric regulating valve opening control.

[0072] In a preferred embodiment of the present invention, step 4 above, based on the final execution opening command, involves collecting the actual instantaneous flow data of each column and comparing it with a preset target flow distribution standard to calculate the flow distribution deviation, updating the system operating information, and achieving balanced flow distribution. This step may include:

[0073] In this embodiment of the invention, step 440 involves controlling the electric regulating valve to execute the final execution opening command, and synchronously collecting the actual instantaneous flow data of each column after the command execution is completed to obtain the feedback flow data for the current cycle. Specifically, this includes: synchronously sending the final execution opening command generated in step 334 to all electric regulating valves in the system. The command sending process uses millisecond-level synchronous triggering to ensure that all valves start operating simultaneously, avoiding flow distribution imbalance caused by command sending timing deviations. The operating status of each electric regulating valve is monitored in real time, and the valve opening changes are fed back in real time through the valve's built-in position sensor until the actual opening of all electric regulating valves reaches the opening value pointed to by the final execution opening command, at which point the command execution is determined to be complete. If some valves do not reach the target opening, the command is resent until all valves complete the resentment. The valve opening is adjusted to ensure the integrity and consistency of the command execution. After the command is executed, the synchronous acquisition of the actual instantaneous flow data of each column is immediately initiated. The actual instantaneous flow data of each column is collected through the flow sensors installed at the outlet of each column. The acquisition frequency is consistent with the valve action monitoring frequency to ensure that the flow data can reflect the actual flow state after the valve opening is adjusted. The validity of the collected actual instantaneous flow data is verified to check whether the flow data is within the reasonable range allowed by the process. If it exceeds the range, it is judged as abnormal data and is immediately re-acquired. The timing of the acquisition of flow data of each column is checked to eliminate the impact of timing deviation on subsequent comparison calculations. After verification, all valid data are integrated to obtain the feedback flow data of the current cycle. This data includes the actual instantaneous flow of each column in the system.

[0074] Step 441: Compare the feedback flow data of the current round with the preset target flow allocation standard in real time to calculate the flow allocation deviation value at the current moment. Specifically, this includes: determining the core comparison benchmark. The preset target flow allocation standard is determined based on the total system flow target value and the ideal allocation ratio of each column. That is, the target instantaneous flow of each column, the sum of the target instantaneous flows of all columns equals the total system flow target value, and the target values ​​conform to the preset allocation ratio to form a uniform and balanced flow allocation benchmark. First, based on the total system flow target value, determine an circumscribed circle. The circumference of the circumscribed circle corresponds to the total system flow, and the radius of the circumscribed circle corresponds to the average allocation coefficient of the total flow, ensuring that the size of the circumscribed circle matches the total system flow. Then, based on the number of columns in the system, generate circumscribed regular polygons with corresponding number of sides. For example, if the system contains 3 columns, generate a circumscribed equilateral triangle; if it contains 4 columns, generate a circumscribed square, and so on, so that each vertex of the regular polygon corresponds to the target instantaneous flow position of a column.

[0075] The generation process of the vertices of the circumscribed regular polygon must ensure that all vertices are evenly distributed on the outside of the circle, and that the arc length between any two adjacent vertices is equal. In flow distribution, this corresponds to the equal difference in the target instantaneous flow rate between any two adjacent cylinders, perfectly meeting the requirement of balanced flow distribution. When generating vertices, the circumference is evenly divided according to the number of cylinders, using the center of the circumscribed circle as a reference. Each division point corresponds to a vertex of the regular polygon, and each vertex is equidistant from the center, corresponding to the balanced correlation between the target flow rate of each cylinder and the total flow rate reference. After vertex generation, the feedback flow data of the current round is mapped to the planar coordinate system of the circumscribed regular polygon. The actual instantaneous flow rate of each cylinder corresponds to an actual point, and each vertex corresponds to the target instantaneous flow rate point of that cylinder. By comparing the positional deviations of the actual points with the corresponding vertices, the direction and degree of flow deviation for each column are determined. If the actual point is located inside the corresponding vertex, it indicates that the actual flow of the column is less than the target flow; if it is located outside, it indicates that the actual flow is greater than the target flow. The distance between the point and the vertex directly corresponds to the magnitude of the flow deviation. Combining the deviation direction and degree determined by the above algorithm, the flow deviation of each column is quantitatively calculated. The deviation value of a single column = actual instantaneous flow - target instantaneous flow. Then, the overall flow distribution deviation value of the system is calculated. The overall deviation value is equal to the sum of the absolute values ​​of the individual deviation values ​​of all columns, divided by the number of columns, to obtain the average deviation value, which is used as the flow distribution deviation value at the current moment, quantitatively reflecting the overall balance of flow distribution in the system.

[0076] Step 442: Determine whether the flow distribution deviation value exceeds the preset allowable error range, and obtain the deviation judgment result. According to the deviation judgment result, if it is determined that it does not exceed the allowable error range, then it is determined that the flow distribution has reached a balanced state and the current adjustment process ends; if it is determined that it exceeds the allowable error range, then the feedback flow data of the current round collected this time is used as new flow data and added to the past state dataset to obtain the updated state dataset. Specifically, this includes: determining the preset allowable error range, which is determined according to the system process requirements and is the critical value of flow deviation to ensure the stable operation of the system. The range value needs to be combined with the actual needs of the industrial site to ensure that when the deviation is within the range, the system flow distribution is balanced, the outlet pressure is stable, and it will not affect the subsequent process operation. The flow distribution deviation value calculated in step 441 at the current moment is compared with the preset allowable error range to obtain the deviation judgment result. If the current flow distribution deviation is less than or equal to the upper limit of the allowable error range and greater than or equal to the lower limit of the allowable error range, it is determined that it is within the allowable error range, indicating that the flow distribution of each column in the current system has reached a balanced state and no further adjustment is needed. The current adjustment process ends and enters the stable operation monitoring stage. If the current flow distribution deviation exceeds the preset allowable error range, it indicates that there is still an imbalance in the current flow distribution and further iterative correction is needed. In this case, the feedback flow data of the current round collected this time is added to the past status dataset as new flow data. The past status dataset contains feedback flow data, deviation data, and operating condition data collected in all previous rounds of adjustment. When adding new data, it is necessary to sort them according to the order of adjustment rounds, and mark the adjustment time and deviation value corresponding to the data to ensure the integrity and time sequence of the dataset. After the addition is completed, the updated status dataset is obtained.

[0077] Step 443: Based on the updated state dataset, trigger a new round of iterative loop and re-execute the entire process from calculating dynamic delay deviation, constructing synchronized state vectors, to calculating attitude correction quantities using the updated state dataset, to obtain optimized control commands based on the new deviation characteristics. Specifically, this includes: automatically triggering a new round of iterative adjustment loop based on the updated state dataset. The triggering logic of the iterative loop is as follows: when the state dataset is updated, the system automatically identifies that there is still a deviation in traffic allocation and starts the iterative mechanism to ensure that traffic allocation continuously approaches an equilibrium state. After the iterative loop starts, using the updated state dataset, re-execute the entire process from calculating dynamic delay deviation, constructing synchronized state vectors, to calculating attitude correction quantities. Each step uses the updated state data as input to ensure the optimization and targeting of the control commands. Recalculate the dynamic delay deviation and combine it with the feedback flow in the updated state dataset. The system collects data, analyzes the transmission delay of flow signals in each column, calculates the dynamic delay deviation of each column, corrects the delay deviation caused by flow changes, and ensures the accuracy of delay compensation. It then reconstructs the synchronized state vector, collecting real-time pressure, flow, and temperature parameters of each column based on the corrected dynamic delay deviation, and constructs a new synchronized state vector to eliminate the impact of delay and flow deviations. Next, it recalculates the attitude correction, analyzes the flow imbalance attitude of each column using the newly constructed synchronized state vector, calculates the corresponding attitude correction, and specifically corrects the flow deviation of each column. Finally, it generates optimized control commands, based on the recalculated attitude correction and the deviation trend reflected in the updated state dataset, re-executes the relevant processes of steps 2 and 3, and generates optimized control commands based on the new deviation characteristics. These commands are more targeted than the previous round's control commands and can compensate for the current flow distribution deviation.

[0078] Step 444: Based on the optimized control command, repeat the iterative process of controlling the valves, collecting feedback, comparing deviations, and judging results to continuously correct the control command and achieve balanced flow distribution in the multi-column system. Specifically, this includes: sending the optimized control command generated in step 443 to each electric regulating valve, controlling the valves to execute the optimized command, adjusting the valve opening, and correcting the current flow distribution deviation; during the valve execution process, monitoring the valve opening and flow changes of each column in real time to ensure that the command is executed properly; after the command is executed, repeat the operation of step 440, synchronously collecting the actual instantaneous flow data of each column to obtain a new round of feedback flow data; then repeat the operation of step 441 again, sending the new round of feedback flow data... The new flow distribution deviation value is calculated by comparing it with the target flow distribution standard and using the algorithm for generating vertices of circumscribed regular polygons. Then, the operation of step 442 is repeated to determine whether the new deviation value exceeds the allowable error range. If the new deviation value still exceeds the allowable error range, the process of steps 442 to 444 is repeated to continuously update the state dataset, trigger the iterative loop, generate optimized control commands, and adjust the valve opening until the flow distribution deviation value calculated in a certain round is within the preset allowable error range. The entire iterative process continues, and the control commands in each round are optimized based on the feedback data of the previous round to gradually reduce the flow distribution deviation and finally achieve balanced flow distribution in each column of the multi-column system.

[0079] Achieving closed-loop control of flow regulation not only ensures the balance and stability of flow distribution in each column, but also enhances the system's adaptive adjustment capability.

[0080] like Figure 2 As shown, embodiments of the present invention also provide an intelligent flow distribution adjustment system for multi-column systems operating in parallel, comprising:

[0081] The correction module is used to calculate the dynamic time delay deviation of each column based on the state dataset and construct the synchronized state vector; map the synchronized state vector to the actual pose of the virtual rigid body in multi-dimensional space, compare it with the target pose of the virtual rigid body, calculate the attitude correction amount, and use the attitude correction amount to correct the synchronized state vector to obtain the corrected synchronized state vector.

[0082] The control module is used to input the corrected synchronized state vector into the pre-trained adaptive control model to obtain the theoretical flow correction amount of each branch; based on the real-time pressure and flow correlation characteristics of each column in the corrected synchronized state vector, the instantaneous flow opening mapping relationship of each electric regulating valve under the current working condition is constructed, and the theoretical flow correction amount of each branch is matched with the instantaneous flow opening mapping relationship to obtain the valve opening change amplitude and the initial opening command of each electric regulating valve.

[0083] The correction module is used to collect the dynamic parameters of the valve's current opening degree, and based on the initial opening degree command of the electric regulating valve, construct a time displacement trajectory equation to predict the valve's inertial lag position and obtain the feedforward correction opening degree command; collect the outlet main pipe pressure fluctuation data and calculate the pressure interference compensation amount, integrate the pressure interference compensation amount and the feedforward correction opening degree command to obtain the final execution opening degree command;

[0084] The verification module is used to collect the actual instantaneous flow data of each column based on the final execution opening command, compare it with the preset target flow allocation standard to calculate the flow allocation deviation, update the system operating information, and achieve balanced flow allocation.

[0085] It should be noted that this system is a system corresponding to the above method. All implementation methods in the above method embodiments are applicable to this embodiment and can achieve the same technical effect.

[0086] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for intelligent adjustment of flow distribution during parallel operation of a multi-column system, characterized in that, The method includes: Step 1: Collect inlet and outlet pressure, temperature, and instantaneous flow data of the electrical sensing devices in each column to form a state dataset. Based on the state dataset, calculate the dynamic time delay deviation of each column and construct a synchronized state vector. Specifically, this includes: extracting the original time series data recorded by the electrical sensing devices of each column from the state dataset and identifying the acquisition time label corresponding to each pressure, temperature, and flow data; comparing the acquisition time label with the system's unified reference clock to calculate the initial time offset of each column's data relative to the reference time, and using the initial time offset as the preliminary dynamic time delay deviation; resampling the original time series data based on the preliminary dynamic time delay deviation, and mapping the state parameters collected by each column at different physical times to the same virtual time axis through interpolation to obtain a time-aligned intermediate state data set; and calculating the rate of change of values ​​between adjacent sampling points based on the intermediate state data set. The actual rate of change of each state parameter of each column; based on the actual rate of change and the theoretical rate of change range determined by the fluid transmission characteristics in the pipeline, the residual error caused by signal transmission fluctuations is identified; based on the residual error, the preliminary dynamic time delay deviation is corrected and calculated to obtain the final dynamic time delay deviation reflecting the degree of response lag of each column; based on the final dynamic time delay deviation, the intermediate state data group after time alignment is phase-fine-tuned to obtain the fine-tuned state parameters of each column at the same logical moment; based on the fine-tuned state parameters, the inlet and outlet pressure, temperature and instantaneous flow rate values ​​of each column at the same logical moment are encapsulated in a preset order to construct a synchronized state vector; the synchronized state vector is mapped to the actual pose of a virtual rigid body in multi-dimensional space and compared with the target pose of the virtual rigid body to calculate the attitude correction amount; the attitude correction amount is used to correct the synchronized state vector to obtain the corrected synchronized state vector. Step 2: Input the corrected synchronized state vector into the pre-trained adaptive control model to obtain the theoretical flow correction amount for each branch; Based on the real-time pressure and flow correlation characteristics of each column in the corrected synchronized state vector, construct the instantaneous flow opening mapping relationship of each electric regulating valve under the current working condition, match the theoretical flow correction amount of each branch with the instantaneous flow opening mapping relationship to obtain the valve opening change amplitude and the initial opening command of each electric regulating valve. Step 3: Collect the current dynamic parameters of the valve opening, and based on the initial opening command of the electric regulating valve, construct the time displacement trajectory equation to predict the valve's inertial lag position and obtain the feedforward correction opening command; collect the outlet main pipe pressure fluctuation data and calculate the pressure interference compensation amount, integrate the pressure interference compensation amount and the feedforward correction opening command to obtain the final execution opening command; Step 4: Based on the final execution opening command, collect the actual instantaneous flow data of each column and compare it with the preset target flow distribution standard to calculate the flow distribution deviation, update the system operating information, and achieve balanced flow distribution.

2. The intelligent flow distribution adjustment method for a multi-column system operating in parallel as described in claim 1, characterized in that, The corrected synchronized state vector includes: Based on the synchronized state vector, the instantaneous flow values ​​of each column are mapped to the coordinate origin position of the virtual rigid body in the multi-dimensional space, the inlet and outlet pressure difference data are mapped to the pitch and yaw angles of the rigid body, the temperature data are mapped to the roll angle of the rigid body, and the actual pose of the virtual rigid body representing the current system operating state is constructed. Based on the system's preset total traffic target value and the ideal distribution ratio of each branch, a virtual rigid body target pose representing the ideal equilibrium state is constructed in the same multidimensional space as the actual pose of the constructed virtual rigid body. The actual pose of the virtual rigid body is spatially superimposed and compared with the target pose of the virtual rigid body. The spatial difference between the two is calculated and decomposed into a rotation component that characterizes the degree of imbalance in the flow ratio of each branch and a translation component that characterizes the degree of deviation in the total flow of the system. Based on the rotation component, the rotation matrix describing the uneven distribution of flow is obtained, and based on the translation component, the translation vector describing the total deviation is obtained. The rotation matrix and the translation vector are fused to obtain the attitude correction amount that reflects both the proportional misalignment and the total deviation. Based on the attitude correction amount, each parameter value in the synchronized state vector is corrected and calculated to obtain the corrected parameter values; based on the corrected parameter values, they are repackaged in a preset order to obtain the corrected synchronized state vector.

3. The intelligent flow distribution adjustment method for a multi-column system operating in parallel as described in claim 2, characterized in that, Constructing the actual pose of the virtual rigid body representing the current system operating state includes: Based on the synchronized state vector, the instantaneous flow rate values ​​of each column are extracted, and the extracted instantaneous flow rate values ​​are processed by spatial coordinate transformation to obtain the transformed instantaneous flow rate values; the transformed instantaneous flow rate values ​​are converted into the coordinate origin position of the virtual rigid body in multi-dimensional space to determine the basic spatial positioning of the virtual rigid body. Based on the position of the origin, the corresponding inlet and outlet pressure data are extracted from the synchronized state vector and the pressure difference is calculated. The pressure difference is then mapped to a rotation angle relative to the position of the origin to obtain the pressure rotation angle data. Based on the pressure rotation angle data, pitch angle and yaw angle, which characterize the fluid resistance properties, are generated respectively, thus completing the attitude definition of the virtual rigid body in two dimensions. Based on the pitch angle and the yaw angle, the temperature data is extracted from the synchronized state vector, the temperature fluctuation is converted into a rotational component around the axis of the virtual rigid body, the roll angle is generated, and the attitude information of the virtual rigid body in the third dimension is completed. By fusing the coordinate origin position, pitch angle, yaw angle, and roll angle, a virtual rigid body actual pose representing the current unbalanced operating state of the multi-column system is constructed.

4. The intelligent flow distribution adjustment method for a multi-column system operating in parallel as described in claim 3, characterized in that, Step 2 includes: The corrected synchronized state vector is loaded into the pre-trained adaptive control model as the feature input data of the current operating condition, and the decoupled flow response law learned in the adaptive control model is used to extract features from the feature input data to obtain the extracted feature data. The feature data is fused and matched with the system's preset total flow target value within the adaptive control model to obtain a fusion matching result that reflects the current load demand and fluid state; based on the fusion matching result, the control strategy parameters of the adaptive control model are dynamically adjusted to obtain the adjusted control strategy parameters. Based on the adjusted control strategy parameters and the corrected synchronized state vector, inference calculations are performed to calculate the theoretical flow correction amount of each parallel branch required to eliminate the remaining state deviation and achieve the target allocation ratio, thus obtaining the theoretical flow correction amount of each branch. Based on the corrected synchronized state vector, the real-time pressure and flow correlation features of each column are extracted, and based on the real-time pressure and flow correlation features, the instantaneous flow opening mapping relationship of each electric regulating valve under the current working condition is constructed to obtain the instantaneous flow opening mapping relationship. The theoretical flow correction of each branch is matched with the instantaneous flow opening mapping relationship to determine the opening change required for each electric control valve to achieve the theoretical flow correction, and the valve opening change range of each electric control valve is obtained. The valve opening change range of each electric control valve is superimposed on the current actual opening reference of the corresponding electric control valve to obtain the initial opening command of the electric control valve.

5. The intelligent flow distribution adjustment method for a multi-column system operating in parallel according to claim 4, characterized in that, Step 3 includes: Based on the initial opening command of the electric regulating valve, the current real-time opening position, rated operating speed and maximum acceleration parameters of each electric regulating valve are collected synchronously to obtain the valve dynamic parameter set; Based on the current real-time opening position in the valve dynamic parameter set as the starting point of motion, and the rated action speed and maximum acceleration parameters as motion constraints, combined with the target position pointed to by the initial opening command of the electric regulating valve, a time displacement trajectory equation describing the entire process of the valve moving from the starting point to the end point is constructed. The lag deviation of the actual position relative to the theoretical commanded position caused by mechanical inertia and fluid resistance is calculated when the valve moves along the motion trajectory described by the time displacement trajectory equation, and the inertial lag position prediction result is obtained. Based on the inertial lag position prediction results, the initial opening command of the electric regulating valve is compensated for in advance to obtain the feedforward correction opening command; after executing the feedforward correction opening command, the instantaneous pressure fluctuation difference between the measured pressure and the set stable pressure is calculated based on the real-time monitoring of the parallel system outlet main pipe pressure data to obtain the instantaneous pressure fluctuation difference. Based on the instantaneous pressure fluctuation difference, the pressure disturbance compensation amount is obtained by converting it through a preset negative feedback gain coefficient; the pressure disturbance compensation amount is then integrated with the feedforward correction opening command to obtain the final execution opening command.

6. The intelligent flow distribution adjustment method for a multi-column system operating in parallel according to claim 5, characterized in that, The advance compensation is to superimpose a pre-compensation amount that is equal in magnitude and opposite in direction to the lag deviation value into the initial command.

7. The intelligent flow distribution adjustment method for a multi-column system operating in parallel as described in claim 6, characterized in that, Achieving balanced traffic distribution includes: The electric regulating valve is controlled to execute the final opening command, and the actual instantaneous flow data of each column is collected synchronously after the command is executed to obtain the feedback flow data of the current cycle; The feedback traffic data of the current round is compared with the preset target traffic allocation standard in real time to calculate the traffic allocation deviation value at the current moment. Determine whether the traffic allocation deviation exceeds the preset allowable error range, and obtain the deviation judgment result. If it is determined that the deviation does not exceed the allowable error range, then the traffic allocation is determined to have reached a balanced state and the current adjustment process ends; if it is determined that the deviation exceeds the allowable error range, then the feedback traffic data of the current round collected this time is used as new traffic data and added to the past state dataset to obtain the updated state dataset. Based on the updated state dataset, a new round of iterative loop is triggered, and the entire process from calculating dynamic time delay deviation, constructing synchronized state vectors to calculating attitude correction quantities is re-executed using the updated state dataset to obtain optimized control commands generated based on new deviation characteristics. Based on the optimized control commands, the iterative process of controlling valves, collecting feedback, comparing deviations, and judging results is repeatedly executed to continuously correct the control commands and achieve balanced flow distribution in the multi-column system.

8. A smart flow distribution adjustment system for multi-column systems operating in parallel, wherein the system implements the method as described in any one of claims 1 to 7, characterized in that, include: The correction module is used to calculate the dynamic time delay deviation of each column based on the state dataset and construct the synchronized state vector; map the synchronized state vector to the actual pose of the virtual rigid body in multi-dimensional space, compare it with the target pose of the virtual rigid body, calculate the attitude correction amount, and use the attitude correction amount to correct the synchronized state vector to obtain the corrected synchronized state vector. The control module is used to input the corrected synchronized state vector into the pre-trained adaptive control model to obtain the theoretical flow correction amount of each branch; based on the real-time pressure and flow correlation characteristics of each column in the corrected synchronized state vector, the instantaneous flow opening mapping relationship of each electric regulating valve under the current working condition is constructed, and the theoretical flow correction amount of each branch is matched with the instantaneous flow opening mapping relationship to obtain the valve opening change amplitude and the initial opening command of each electric regulating valve. The correction module is used to collect the dynamic parameters of the valve's current opening degree, and based on the initial opening degree command of the electric regulating valve, construct a time displacement trajectory equation to predict the valve's inertial lag position and obtain the feedforward correction opening degree command; collect the outlet main pipe pressure fluctuation data and calculate the pressure interference compensation amount, integrate the pressure interference compensation amount and the feedforward correction opening degree command to obtain the final execution opening degree command; The verification module is used to collect the actual instantaneous flow data of each column based on the final execution opening command, compare it with the preset target flow allocation standard to calculate the flow allocation deviation, update the system operating information, and achieve balanced flow allocation.

Images