Double-body valve coordination control system and method for pipe network flow closed-loop regulation

Abstract

The invention discloses a pipe network flow closed-loop regulation-oriented double-body valve coordination control system and method, and relates to the technical field of data acquisition control. The double-body valve coordination control system and method for pipe network flow closed-loop regulation comprises the following steps that S1, a double-body valve data set is collected to be subjected to segmented scaling normalization processing; s2, fragmented recombination and time sequence consistency screening are carried out, and quality scoring is carried out; s3, a double-body valve adjusting instruction is generated for combined control, and a control strategy state is output; and S4, establishing an associated index for each sampling time slice, and performing integrated version archiving and playback recalculation. The synchronization, stability and traceable operation and maintenance capability of cooperative closed-loop adjustment of the double-body valve are effectively improved, and the problems of asynchronization and mutual interference in cooperative control of the double-body valve are solved.

Description

Technical Field

[0001] This invention relates to the field of data acquisition and control technology, specifically to a dual-valve coordinated control system and method for closed-loop regulation of pipeline flow. Background Technology

[0002] As pipeline network scenarios increasingly demand higher precision, stability, and maintainability in flow closed-loop regulation, existing technologies employ single-valve or dual-valve independent control to achieve target flow and differential pressure regulation. However, in engineering implementations where dual-body valves are driven independently by two motors, asynchronous valve position responses are commonly caused by actuator dynamic differences, transmission clearances, and friction variations. Online updates of compensation parameters and coupled models often lack provable data quality constraints and safe backoff logic, making them prone to mislearning in abnormal segments, resulting in uncontrollable parameter drift and difficulties in fault location.

[0003] For example, invention patent CN119045563B discloses a constant temperature regulation method and system based on heat dissipation flow control of a heat dissipation valve. The method includes: collecting temperature influence nodes of the area to be regulated and constructing a room temperature model; setting up a radiator model and conducting tests to obtain the heat dissipation performance of the radiator model; adding the radiator model to the room temperature model; calculating the total heat generation of the area to be regulated in a preset period, calculating the theoretical heat load of the radiator model in the preset period, and calculating the heat dissipation flow of the radiator model in the preset period; generating on / off control information for the heat dissipation valve in the radiator model based on the heat dissipation flow of the radiator model in the preset period; controlling the on / off state of the heat dissipation valve based on the on / off control information; and evaluating the temperature regulation of the area to be regulated by the radiator model. This invention controls the heat dissipation flow of the radiator by controlling the on / off state of the heat dissipation valve, effectively improving the efficiency and accuracy of constant temperature regulation of the radiator.

[0004] For example, invention patent CN119472858B discloses a thermostatic intelligent valve control system and its control method, including: receiving DDC analog control signals in real time through the thermostatic intelligent valve and performing standby mode vibration-free judgment and valve adjustment control analysis to generate a valve position opening adjustment control mode and a valve flow adjustment control mode; performing real-time monitoring of the supply and return water temperatures and calculating the valve temperature difference to obtain the measured temperature difference of the intelligent valve; performing valve adjustment control logic response analysis on the thermostatic intelligent valve to generate a basic intelligent valve control mode and a constant intelligent valve control mode; obtaining the set temperature difference and set return water temperature of the intelligent valve and performing basic valve control processing and constant valve control processing on the thermostatic intelligent valve to execute the corresponding intelligent valve control work. This invention can automatically adjust the control mode according to actual needs, and has high adaptability, accuracy and energy saving.

[0005] In existing technologies, current systems employ two independently driven motors (L9110Sx2), with inconsistencies in motor dynamics, transmission clearance, and friction between the two actuators. These differences lead to asynchronous responses from the two valve bodies under the same control commands, resulting in valve position control deviations. Especially under pipeline coupling conditions, when one valve position deviates, it causes a redistribution of flow and pressure differences, creating a valve-network coupling mutual interference effect. This further exacerbates system instability, manifesting as slow tracking, oscillation, or even overshoot in closed-loop control, thus affecting the flow regulation accuracy and stability of the entire pipeline network.

[0006] Therefore, in order to address the above problems, there is an urgent need for a two-body valve coordinated control system and method for closed-loop regulation of pipeline flow. Summary of the Invention

[0007] Technical problems to be solved To address the shortcomings of existing technologies, this invention provides a coordinated control system and method for dual-body valves in pipeline network flow closed-loop regulation, which solves the problems of asynchrony and mutual interference in coordinated control of dual-body valves.

[0008] Technical solution To achieve the above objectives, this invention provides the following technical solution: a coordinated control system and method for dual-body valves for closed-loop regulation of pipeline flow, comprising: S1, collecting dual actuator drive information and pipeline observation information to construct a dual-body valve dataset, introducing a unified time base for alignment and correction, and performing segmented scaling and normalization processing; S2, performing fragmented recombination and timing consistency screening on the dual-body valve data frame set to generate a usable fragment set, constructing a compensation parameter set and performing quality scoring; S3, generating dual-body valve regulation commands based on the compensation parameter set, performing joint control, performing adaptive gating updates and safety degradation rollback, and outputting the control strategy status; S4, establishing an associated index for each sampling time slice, performing integrated version archiving and playback recalculation, and performing online health assessment, alarm classification, and automatic rollback handling.

[0009] Furthermore, the specific process of collecting dual actuator drive information and pipeline network observation information to construct a dual-body valve dataset is as follows: Basic capability information of the dual-body valve control unit is collected through the equipment management interface, control board hardware configuration, firmware parameter table, and factory-written information. Dual actuator drive information and pipeline network observation information are collected through valve action calibration sequences and pipeline network steady-state and disturbance operation processes to construct a dual-body valve dataset. The dual-body valve dataset includes: a basic dataset, a drive dataset, an observation dataset, a scheduling dataset, and a timestamp dataset. Historical closed-loop regulation acquisition sequences that have been stored under the same installation scenario and configuration conditions are used as historical operating data.

[0010] Furthermore, the specific process of introducing a unified time base for alignment and correction, and performing segmented scaling and normalization is as follows: Based on the acquired dual-body valve dataset, a unified time base is introduced to align and correct the time fields of the sampling task, drive update task, and communication arrival event. A set of dual-body valve data frames is obtained using the sampling period as the basic unit. For each time-slice record, timing consistency verification, link freshness verification, and quality detection and labeling of asynchronous and mutual interference risks are performed: For each time-slice record in the dual-body valve dataset, the dual-path drive communication of the device is retrieved from the basic dataset based on the device number. The system maps the channel, communication link type, and sampling configuration version, and performs consistency verification between the target drive commands and actual drive execution quantities of valve body A and valve body B in the current time slice. It performs segmented scaling normalization and unit diameter unification processing: the flow observation value, differential pressure and pressure observation value are segmented and scaled according to the sensor calibration range and unit conversion rules, and mapped to a unified range. The normalized fields are written into the dual-body valve data frame, and the target field, measured field, context field and quality field are integrated into a dual-body valve data frame record using the unified sampling timestamp as the index, and the dual-body valve data frame set is output.

[0011] Furthermore, the specific process of fragmenting and recombining the dual-body valve data frame set and filtering for temporal consistency to generate a usable fragment set is as follows: Input the dual-body valve data frame set, perform stable sorting, deduplication, and gap detection on a unified time axis, and aggregate data frames that meet the continuity and quality conditions into a fragment-level closed-loop sequence; perform stable sorting of the dual-body valve data frame set using a unified sampling timestamp and sampling sequence number as keys; mark data frames with backward, jump, or duplicate sampling sequence numbers, or backward timestamps with temporal anomalies, and write the anomaly type, anomaly occurrence timestamp, preceding and following sequence numbers, associated valve body identifier, and associated channel number into the temporal sequence. Anomaly list; Stage identification within each segment: The time window in which the target field remains unchanged is used as a candidate steady-state window. The sliding variance, sliding slope, and peak-to-peak value are calculated for the flow observation, differential pressure observation, and pressure observation. When the sliding variance is lower than the steady-state threshold and the absolute value of the slope is lower than the slope threshold, it is marked as a steady-state segment. When the target field undergoes a step change or the valve body enters the acceleration and homing phase, it is marked as a transient segment. Availability screening of segments is performed: Data frames with missing key fields, equipment fault flags set, continuous link CRC failures leading to outdated observations, and continuous jitter exceeding limits in the driver update task are removed. Output a set of available segments organized by segment number.

[0012] Further, the specific process of constructing the compensation parameter set and performing quality scoring is as follows: Input the available fragment set and the basic dataset; calculate the difference features between valve body A and valve body B at the drive layer, scheduling layer, and observation layer; combine the mutual interference sensitivity features extracted from the pipeline network observation response to generate the actuator difference compensation parameter set and the valve-network coupling sensitivity parameter set; normalize and regularize the compensation parameter set: linearly normalize the dead zone, gain, time constant, and coupling matrix elements to obtain the parameter vector; take the natural logarithm of the parameter stability score and multiply it by the parameter stability weight coefficient to obtain the stability term; take the natural logarithm of the fitting accuracy score and multiply it by the fitting accuracy weight coefficient to obtain the accuracy term; take the coordination consistency... The natural logarithm of the performance score is multiplied by a weighting coefficient to obtain the synergy term. The three weighted terms—stability, accuracy, and synergy—are added together and substituted into an exponential function to obtain the quality score of the compensation parameter set. The quality score is compared with the quality threshold in real time. When the quality score is lower than the quality threshold, re-identification is triggered: the next available window is prioritized to execute the calibration sequence of valve body A and B and the steady-state perturbation test segment to supplement the sample. When the quality score is greater than or equal to the quality threshold, the current compensation parameter set is determined to be available. The actuator difference compensation parameters and coupling sensitivity parameters obtained in this identification are written into the parameter lookup table as the new version's effective parameters. The compensation parameter set, parameter quality score, and risk label set are output.

[0013] Furthermore, the specific process of generating dual-valve regulation commands based on the compensation parameter set and performing joint control is as follows: Input the compensation parameter set and the target flow rate and target pressure difference of the current time slice to construct a two-layer coordinated control structure of the flow rate and pressure difference main loop and the dual-valve synchronous slave loop. Calculate the target drive increment of valve body A and valve body B and generate the target duty cycle, direction, and stage commands. The main loop takes the target flow rate or target pressure difference as the control target and calculates the observation error in each sampling period. The observation error includes flow rate error and pressure difference error. The main loop outputs the desired observation increment vector. Using the local coupling matrix in the coupling sensitivity parameter set, the desired observation increment is mapped to the dual-valve drive increment using constrained weighted least squares solution. After generating the dual-valve drive increment, the synchronous slave loop compensation term is superimposed. The synchronous slave loop outputs the synchronous suppression increment. Perform difference compensation and nonlinear correction on the dual-valve drive increment: perform dead zone compensation and friction compensation on the sub-valves according to the actuator difference compensation parameters, and output the target drive command sequence of valve body A and valve body B.

[0014] Furthermore, the specific process of adaptive gating update and safety degradation rollback, and outputting the control strategy state, is as follows: Input the quality score value generated for each time slice; construct an adaptive gating update and safety degradation rollback mechanism; adaptively update the coupling sensitivity parameter and actuator difference compensation parameter; trigger control strategy degradation and parameter rollback; to avoid mislearning link obsolescence, scheduling jitter, and strong transient noise as coupling drift or actuator difference, an update threshold is introduced to determine whether to update; a sliding statistic is obtained by gating with the moving average and moving minimum and then normalizing to obtain a quality score value for K consecutive sampling periods; the observation freshness factor is obtained by comparing the freshness score with the hysteresis upper limit and normalizing it; and the observation freshness factor is obtained by comparing the CRC failure rate, retransmission count, timeout count, arbitration conflict count, or lost data. The link health factor is obtained by comparing the packet count with its respective upper limit and normalizing it; the scheduling stability factor is obtained by comparing the period deviation, jitter peak, and critical section blocking duration with the upper limit and normalizing them; the device and steady-state window factors are obtained by gating and normalizing the device fault and reset flags, the allowable range of power supply voltage and onboard temperature, and the steady-state update window indication; the minimum value of the sliding statistic and the four factors is calculated and substituted into the truncation function to obtain the update admission value; the update admission value is compared with the admission threshold in real time. The parameter update of the coupling matrix and compensation parameters is allowed only when the update admission value is greater than or equal to the admission threshold. When the update admission value is less than the admission threshold, the parameter update of the current time slice is stopped, and the previous version of the coupling matrix and compensation parameter set continues to be effective. The control strategy state with gating update capability is output.

[0015] Furthermore, the specific process of establishing an associated index for each sampling time slice and performing integrated version archiving and replay recalculation is as follows: For each sampling time slice, an associated index is established for data, instructions, parameters, and firmware. Version archiving and replay recalculation are performed, outputting a traceable closed-loop operation evidence chain. Using the device number, segment number, and sampling sequence number as primary keys, target fields, measured fields, link statistics fields, scheduling jitter fields, and quality marker fields for the same time slice are aggregated to form a time-slice-level operation record. On the onboard side, immutable append writing is performed on the original closed-loop acquisition records, control instruction records, and parameter change records, and an evidence chain digest is generated using a three-layer digest mechanism: time-slice hash, segment hash tree root value, and cross-link witness. On the platform side, idempotent deduplication, hash tree proof verification, cross-link witness consistency verification, and order reordering are performed on the segments. Replay recalculation uses an operator-based replay and a comparative stripping attribution mechanism. A closed-loop operation evidence chain is output.

[0016] Furthermore, the specific process for online health assessment, alarm classification, and automatic rollback is as follows: Input the quality score value for each time slice, update the admission value, link health factor, and scheduling stability factor; conduct online health assessment and handling rules; generate alarm events, degradation and rollback actions, and handling result records; online health assessment introduces a risk score value, integrates risk evidence into a single risk quantity for alarm classification; multiply the indicator's out-of-limit value by the amplitude sensitivity coefficient to obtain the out-of-limit item; multiply the out-of-limit item by the product of the proportion of the duration of continuous out-of-limit indicators and the persistence amplification coefficient to obtain the persistence item; substitute the persistence item into the arctangent function to obtain the contribution item; add the contribution items corresponding to the indicators participating in the fusion in sequence and then add them to the numerical stability item. The numerator is obtained by summing the weight coefficients of all indicators and dividing by the numerator. The denominator is obtained by summing the weight coefficients and dividing by the denominator. The risk score is obtained by summing the weight coefficients and dividing by the denominator. When the risk score is consistently higher than the drift risk threshold, it is determined to be an unlearnable drift risk, triggering a freeze update and recording the reason code. When the risk score is higher than the parameter unavailability threshold, it is determined to be a parameter unavailability risk, triggering parameter rollback and writing the rollback version number into the control strategy status. Automatic rollback handling includes: writing a rollback pointer for each parameter update and each effective version. The rollback pointer records the previous version number, the previous version quality score, the previous version applicable working condition bucket, and the effective time. The alarm event table, the degradation and rollback action table, and the handling result table are output.

[0017] Furthermore, the second aspect of this invention provides a coordinated control system for dual-body valves for closed-loop regulation of pipeline flow, applied to a coordinated control method for dual-body valves for closed-loop regulation of pipeline flow, comprising: a data acquisition and transmission module for acquiring dual actuator drive information and pipeline observation information to construct a dual-body valve dataset, introducing a unified time base for alignment and correction, and performing segmented scaling and normalization processing; a data processing and analysis module for fragmenting and recombining the dual-body valve data frame set and filtering for time sequence consistency, generating a set of usable fragments, constructing a compensation parameter set, and performing quality scoring; a valve coordinated control and compensation module for generating dual-body valve regulation commands based on the compensation parameter set, performing joint control, performing adaptive updates of gating and safety degradation rollback, and outputting the control strategy status; and an optimization and fault diagnosis module for establishing an associated index for each sampling time slice, performing integrated version archiving and playback recalculation, and performing online health assessment, alarm classification, and automatic rollback handling.

[0018] Beneficial effects The present invention has the following beneficial effects: (1) This invention calculates the differential features of valve body A and valve body B in the drive layer, scheduling layer and observation layer, and extracts mutual interference sensitivity features and asynchronous and mutual interference risk labels to realize online observability, quantification and location of "actuator asynchronous causes valve and network coupling mutual interference", and can identify risk segments of slow tracking, oscillation and overshoot in advance.

[0019] (2) This invention introduces scheduling jitter indicators such as the periodic deviation, preemption times and blocking duration of sampling tasks and driving update tasks, establishes a scheduling stability factor and participates in gating judgment, which can suppress the decrease in closed-loop effective bandwidth and low-frequency oscillation caused by uneven driving updates under MCU multi-task load fluctuation conditions, thereby improving control stability and engineering adaptability.

[0020] (3) In this invention, by constructing a quality score for compensation parameters and updating the admission value, and by adopting a multi-factor gating mechanism, the coupling matrix and the actuator difference compensation parameters are allowed to be updated in small steps only when the data is fresh, the link is healthy, the scheduling is stable and in a steady-state low-disturbance segment. This avoids the noise being mistakenly learned as drift in abnormal segments, and significantly improves the safety and convergence reliability of online adaptive updates.

[0021] (4) In this invention, by coordinating the synchronous compensation of dual valves and the differential compensation of actuators, the differences such as low duty cycle start-up lag, direction switching return difference and equivalent gain difference are offset, so that the actual execution of dual valves is compressed from obvious asynchrony to a small deviation range, reducing the flow and pressure redistribution disturbance caused by valve position deviation, thereby reducing overshoot rate and mutual interference risk and improving closed-loop tracking accuracy.

[0022] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0023] Figure 1 This is a flowchart of the dual-body valve coordinated control method for closed-loop regulation of pipeline flow according to the present invention. Figure 2 This is a framework diagram of the dual-body valve coordinated control system for closed-loop regulation of pipeline flow according to the present invention. Figure 3 This is a visualization diagram of the timing of the dual-body valve actuation and pipeline network observation in this invention; Figure 4 This is a comparison chart of the quality scores of the compensation parameters of this invention; Figure 5 This is a comparison diagram of the effects of the dual-valve synchronous control of the present invention. Detailed Implementation

[0024] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] Please see Figures 1-5This invention provides a technical solution: a coordinated control system and method for dual-body valves for closed-loop regulation of pipeline flow, comprising: S1, collecting dual actuator drive information and pipeline observation information to construct a dual-body valve dataset, introducing a unified time base for alignment and correction, and performing segmented scaling and normalization processing; S2, performing fragmented recombination and timing consistency screening on the dual-body valve data frame set to generate a usable fragment set, constructing a compensation parameter set and performing quality scoring; S3, generating dual-body valve regulation commands based on the compensation parameter set, performing joint control, performing adaptive gating updates and safety degradation rollback, and outputting the control strategy status; S4, establishing an associated index for each sampling time slice, performing integrated version archiving and playback recalculation, and performing online health assessment, alarm classification, and automatic rollback handling.

[0026] Specifically, the process of collecting dual actuator drive information and pipeline network observation information to construct a dual-body valve dataset is as follows: Basic capability information of the dual-body valve control unit is collected through the equipment management interface, control board hardware configuration, firmware parameter table, and factory-written information. Dual actuator drive information and pipeline network observation information are collected through valve action calibration sequences and pipeline network steady-state and disturbance operation processes to construct the dual-body valve dataset. The dual-body valve dataset includes: a basic dataset, a drive dataset, an observation dataset, a scheduling dataset, and a timestamp dataset. Historical closed-loop regulation acquisition sequences that have been stored under the same installation scenario and configuration conditions are used as historical operating data.

[0027] By reading the hardware resources and communication interface configuration of the control board, the system collects the MCU model and clock resources, storage resources and peripheral interface mapping, driver model and channel number of the dual motor drive channels, drive pin mapping and forward / reverse control logic of valve body A and valve body B, link type and configuration parameters of wired bus interface and wireless interface, and activation status of onboard temperature acquisition channel and power rail monitoring channel. The above capability information is written into the basic capability data record of the control board according to the device number to form a basic dataset. The basic capability information of the control board comes from at least the control board circuit and device configuration: STM32F103C8T6 control unit, two L9110S motor drivers, MBUS bus transceiver and isolation device, LoRa wireless module RA-01H, EEPROMAT24C02, temperature acquisition device DS18B20 and NTC sampling port, as well as the hardware configuration and interface connection information of power conversion and voltage regulation circuit. When the equipment is in a non-production adjustment period, a valve action calibration sequence is issued. For valve body A and valve body B, segmented duty cycle ramp-up, direction switching, holding, and homing actions are performed respectively. The drive PWM duty cycle, PWM frequency (or timer period), forward and reverse control bits, drive enable bits, action stage identifier, cumulative action pulse count, and cumulative power-on duration are collected for each sampling period. Simultaneously, the onboard power rail voltage (collected via MCUADC or power monitoring point), onboard temperature (collected via DS18B20 or NTC channel), and drive output logic state are collected. The starting threshold, equivalent dead zone width, direction switching backlash difference, stage response delay (from command issuance to drive state stabilization), and jitter amplitude during the holding phase are statistically analyzed for each valve body in different duty cycle ranges. These statistics are then compared with the equipment number, valve body number, and calibration sequence number. The data is written into the dual-executor drive process records to construct the drive dataset. The scheduling dataset includes: the planned cycle of the sampling task, the actual trigger time of the sampling task, the execution duration of the sampling task, the planned cycle of the drive update task, the actual trigger time of the drive update task, the execution duration of the drive update task, the length of the task ready queue, the number of task preemptions, the longest critical section blocking duration, the number of highest priority inversions, and the watchdog feeding interval statistics. It is obtained by embedding points in the MCU firmware for the sampling task, drive update task, communication receiving and parsing task, and storage writing task. When the task is switched in and out and timer interrupt is triggered, the DWT_CYCCNT or SysTick counter, RTOS task status register and ready queue length are read, and the cycle deviation and execution jitter are accumulated and statistically analyzed online to characterize the scheduling jitter and blocking risk.The timestamp and sequence identifier dataset includes: a unified timestamp for the acquisition node (milliseconds or microseconds), a local timestamp for the device, a timestamp received by the gateway (if forwarded by the gateway), a sampling sequence number, a control round number, a valve body identifier (A or B), a data source channel number (flow, differential pressure, temperature, drive, link, or scheduling), a sampling configuration version number, a control parameter version number, and a firmware version number. The timestamp dataset is uniformly assigned by the acquisition control unit after each completion of sensor readings, drive status readbacks, link statistics updates, and scheduling data collection. The unified timestamp for the acquisition node is generated by the MCU system time base (RTC or SysTick) after NTP time synchronization. The local timestamp for the device is generated by the fields reported by each peripheral and slave device or by the local RTC. The timestamp received by the gateway is written by the gateway when the message arrives. The sampling sequence number and control round number are generated incrementally by the control task in each sampling cycle and written together with the configuration and firmware version number for out-of-order detection, frame loss location, cross-version caliber tracing, and closed-loop reproduction experiments.

[0028] Historical operational data includes: target flow rate time series and target differential pressure time series for different time periods; target valve position allocation strategy identifier sequence and control round number and control parameter version number sequence; pipeline network observation time series (flow rate observation, differential pressure observation or upstream and downstream pressure observation, and pipeline network temperature and medium temperature observation); dual actuator drive process time series (target duty cycle or direction or stage sequence of valve body A and valve body B, actual duty cycle or direction or enable position sequence, cumulative pulse count sequence and energization duration sequence, stage switchover arrival time stamp sequence, target and actual drive deviation sequence, and response asynchrony sequence); link and freshness statistics sequence (CRC check results, retransmission). The sequence includes: number of times, timeouts, bus arbitration and collision counts, wireless packet loss and retransmission counts, receive queue backlog length, arrival timestamps, and freshness scores; scheduling jitter and operational health sequences (planned period, actual trigger time, execution duration, task preemption count, longest critical section blocking duration, priority inversion count, watchdog feed interval statistics, reset flag and fault flag sequences); and segment metadata and quality tags bound to the above sequences (segment number, segment start and end time, sampling configuration version number, control parameter version number, firmware version number, timing anomaly flag, link risk flag, suspected mutual interference flag, and high-risk closed-loop segment flag). In this implementation scheme, through the above steps, multi-source data from control board hardware configuration, calibration sequence, closed-loop operation process, wired and wireless link statistics, and task scheduling statistics are transformed into a dual-body valve dataset with clear sources and physical meanings, providing a unified data foundation for asynchronous identification of dual actuators, suppression of valve-network coupling interference, and online correction of adaptive servo parameters.

[0029] Specifically, the process of introducing a unified time base for alignment and correction, and performing segmented scaling and normalization, is as follows: Based on the acquired dual-body valve dataset, a unified time base is introduced to align and correct the time fields of the sampling task, drive update task, and communication arrival event. A set of dual-body valve data frames is obtained using the sampling period as the basic unit. For each time-slice record, timing consistency verification, link freshness verification, and quality detection and labeling of asynchronous and mutual interference risks are performed. For each time-slice record in the dual-body valve dataset, the dual-drive channel mapping, communication link type, and other parameters of the device are retrieved from the basic dataset based on the device number. The sampling configuration version performs consistency verification on the target drive commands and actual drive execution quantities of valve body A and valve body B in the current time slice; it writes the difference between the target duty cycle and the actual duty cycle, the difference between the target direction and the actual direction, and the difference between the target stage and the actual stage into the drive deviation field, and calculates the response asynchrony between valve body A and valve body B (the difference between the drive deviations of the two valves, the difference between the cumulative pulse counts and the energization duration of the two valves, and the time difference between the stage switching arrival of the two valves); in terms of time axis alignment, it sorts and checks the collected records using the sampling sequence number and the local sampling timestamp as keys, and marks records with abrupt changes, repetitions, or reversals in the sampling sequence number as timing anomalies. The system employs several techniques: marking and statistical analysis of the periodic deviation between sampling and driver update tasks. When driver update jitter exceeds a threshold, the corresponding time slice is marked as uneven, explaining the risk of decreased closed-loop effective bandwidth and low-frequency oscillations. Regarding link freshness, a data freshness score is calculated using arrival timestamps and queue backlog lengths for both MBUS and LoRa links. Traffic and differential pressure observations that are "not the latest" or "not from the same time" are marked as insufficiently fresh to prevent the controller from introducing reverse regulation based on outdated observations when calculating valve positions. Link statistics are analyzed for CRC failures, sudden increases in retransmissions / timeouts, or elevated arbitration conflicts. When time slices are marked as link risk segments, a mutual interference criterion is constructed based on the linkage between response asynchronous quantities and pipeline network observation quantities: when the asynchronous quantities of valve body A and valve body B increase, and the flow and differential pressure observations show opposite directions or overshoot amplification (e.g., the target remains unchanged while the flow or differential pressure fluctuation amplitude increases significantly), the time slice is marked as suspected valve-network coupling mutual interference; when suspected mutual interference occurs simultaneously with uneven drive updates and insufficient link freshness, the time slice is marked as a high-risk closed-loop segment, providing a basis for elimination and weight reduction for adaptive servo parameter updates and coupling compensation model training.

[0030] The process involves segmented scaling and normalization, along with unified unit caliber processing: Flow, differential pressure, and pressure observations are segmented and scaled according to sensor calibration ranges and unit conversion rules, mapping them to a unified range; the target duty cycle, actual duty cycle, and drive deviation of valve bodies A and B are normalized according to the upper limit of PWM resolution, unifying the duty cycle representation under different timer frequency divisions and resolutions to a dimensionless range; the cumulative pulse count and power-on duration are scaled according to the sampling period and action stage, converting them into drive intensity per unit time and relative progress characteristics within a stage, eliminating scale differences caused by different sampling periods and action durations; task jitter fields (period deviation, execution duration, longest blocking duration) are normalized according to the upper bound of the quantiles within their respective statistical windows, forming horizontally comparable scheduling pressure characteristics; link statistics fields (retransmission count, timeout count, collision count, packet loss count, queue backlog length) are normalized according to the upper limit of the link type's capacity and the sliding window peak value, forming link congestion intensity characteristics; and the normalized fields are then labeled with " The "_norm" suffix is ​​written into the dual-body valve data frame for feature input coupling compensation model training and asynchronous mutual interference identification. Using a unified sampling timestamp as an index, target fields (target flow rate or target differential pressure, target duty cycle or direction or stage of valve bodies A and B), measured fields (flow observation, differential pressure or pressure observation, actual duty cycle or direction or stage of valve bodies A and B, cumulative pulse count or power-on duration), context fields (onboard temperature, supply voltage, task jitter, link statistics), and quality fields (timing anomaly marker, link risk marker, suspected mutual interference marker, high-risk closed-loop segment marker) are integrated into a single dual-body valve data frame record, outputting a set of dual-body valve data frames. Fields in the dual-body valve data frame related to "drive channel, communication link, onboard storage, and temperature acquisition" can all be traced back to the control board's dual L9110S drive channels, wired MBUS and LoRa interfaces, EEPROM, and temperature acquisition channel hardware and interface configuration.

[0031] like Figure 3The visualization of the timing of dual-valve actuation and pipeline network observation provides interpretable timing evidence for coupling sensitivity identification, gating update access, and degradation rollback triggering. The red and blue lines in the diagram represent the actual duty cycle (or equivalent opening) of valve body A and valve body B over time, respectively. The two curves exhibit a periodic rise and fall, indicating that the controller continuously issues adjustment actions to the dual valves within different time slices. Simultaneously, the curves of the two valves show a certain amplitude and phase difference near the peaks and troughs, reflecting that differences in motor dynamics, friction, and transmission clearance between the two actuators project as asynchronous valve position responses. The right vertical axis corresponds to the green curve, representing the observed response on the pipeline network side. It can be seen that the pipeline network response is highly correlated with the changes in the dual valve opening in time: when the dual valve duty cycle increases, the pipeline network response rises synchronously; when the dual valve duty cycle decreases, the pipeline network response falls accordingly, demonstrating the direct impact of the actuator actions on the pipeline network state under valve-network coupling. The green curve exhibits slight lag and local spikes at certain rising and falling inflection points. Common sources include link arrival lag, sampling and driver update jitter, and dynamic delays caused by the inertia of the network itself. These phenomena directly affect the effective closed-loop bandwidth and overshoot risk.

[0032] This implementation plan significantly reduces the risks of reverse regulation, low-frequency oscillations, and overshoot amplification caused by disordered, outdated observations, and non-uniform updates. On the other hand, it transforms the asynchrony of dual valves and valve-grid coupling interference from "phenomena" into quantifiable response asynchrony quantities, suspected interference, and risk segment labels. This provides a direct and interpretable chain of evidence for coupling sensitivity identification, gating update admission determination, and degradation backoff triggering, ultimately improving the stability, synchronization, and engineering maintainability of closed-loop regulation.

[0033] Specifically, the process of fragmenting and recombining the dual-body valve data frame set and filtering for temporal consistency to generate a usable fragment set is as follows: Input the dual-body valve data frame set, perform stable sorting, deduplication, and gap detection on a unified time axis, and aggregate data frames that meet continuity and quality conditions into a fragment-level closed-loop sequence: Use a unified sampling timestamp and sampling sequence number as keys to stably sort the dual-body valve data frame set; mark data frames with backward, jump, or duplicate sampling sequence numbers, or backward timestamps, with temporal anomaly markers, and write the anomaly type, anomaly occurrence timestamp, preceding and following sequence numbers, associated valve body identifier, and associated channel number into the temporal anomaly list; deduplication uses the "primary key consistency and tolerance consistency" criterion: using (equipment number, sampling sequence number, ... The sampling configuration version number is used as the strong primary key. If the strong primary keys are completely identical, it is determined to be a duplicate frame. When the strong primary keys are identical but there is a slight difference in the sampling timestamp, if the timestamp difference does not exceed the deduplication tolerance, it is still determined to be a duplicate report in the same time slice and only one record is retained. The deduplication tolerance is determined by the statistical distribution of the timestamp difference of duplicate reports with the same sequence number in the historical running data. The upper quantile value is taken as the upper limit of the tolerance to cover the small time offset introduced by normal link retransmission and retransmission without accidentally deleting real new samples. When the strong primary keys are identical but the difference of the key observation fields (flow and pressure difference) exceeds the observation tolerance, it is determined to be a conflicting duplicate frame. Both records are retained and registered as data conflict anomalies in the anomaly list to trace possible cache overwriting, out-of-order recalculation, or inconsistency of multiple sources.

[0034] When the time interval between adjacent data frames exceeds the upper limit threshold, it is marked as a time gap event, and the missing duration and start and end times of the missing data are recorded. Gap detection is based on two criteria: sequence gap and time gap. Using the sampling sequence number as the key, if the difference between the sequence numbers of two adjacent frames is greater than 1, it is determined to be a sequence gap and the range of missing sequences is recorded. Using the sampling timestamp as the key, if the time difference between two adjacent frames is greater than the upper limit threshold of the interval, it is determined to be a time gap. The upper limit threshold of the interval is obtained by statistically analyzing the distribution of normal sampling intervals in historical running data, taking the upper quantile value of the normal interval and multiplying it by the interval amplification factor. This is used to prevent excessive triggering of gaps when there is slight jitter, and at the same time, it can sensitively identify real lost sampling and long blockages. When sequence gaps and time gaps occur simultaneously, the gap is marked as a gap, which is used to increase the weight in the segment availability screening.

[0035] The segment boundary conditions include at least: changes in control round number, changes in sampling configuration version number, changes in control parameter version number, changes in link freshness flag from normal to insufficient, setting of link risk flag, setting of driver update unevenness flag, and setting of reset flag; when a step change in target field is detected (change in target flow or target differential pressure), switching between valve body A and B action stages (stage identifier change point), or a sudden increase in the backlog length of the link arrival queue exceeding the queue threshold, the current segment is forcibly terminated and a new segment is started to avoid mixing different control contexts into the same closed-loop segment; Within each segment, stage identification is performed: the time window in which the target field remains unchanged is used as the candidate steady-state window; the sliding variance, sliding slope, and peak-to-peak value are calculated for the flow rate observation, differential pressure observation, and pressure observation; when the sliding variance is lower than the steady-state threshold and the absolute value of the slope is lower than the slope threshold, it is marked as a steady-state segment; when the target field undergoes a step change or the valve body enters the acceleration and homing stage, it is marked as a transient segment; the steady-state threshold and slope threshold are obtained through quantile statistics and maintained in separate buckets according to the medium temperature range, differential pressure range, and power supply voltage range.

[0036] Segment availability screening is performed: data frames with missing key fields, device fault flags, continuous link CRC failures leading to outdated observations, and continuous excessive jitter in driver update tasks are removed. Anomaly density metrics are calculated at the segment level, including the proportion of time-series anomalies, link risk, suspected mutual interference, and uneven driver updates. When the anomaly density exceeds a density threshold, the segment is marked as a low-confidence segment and written to the segment quality summary field. The anomaly types in the time-series anomaly list are defined using an enumeration method, including at least: E01 - Sequence number backwards (the sampling sequence number of two adjacent frames decreases), E02 - Sequence number jump (the difference between the sampling sequence numbers of two adjacent frames is greater than 1), E03 - Sequence number duplication (duplicate reporting due to strong primary key consistency), E04 - Timestamp backwards (the sampling timestamp of two adjacent frames decreases and the decrease exceeds the time backwards tolerance), E05 - Timestamp drift (the deviation of the timestamp difference between two adjacent frames from the expected sampling period exceeds the drift threshold), E06 - Time gap (the time difference between two adjacent frames exceeds the upper limit of the interval threshold), E... 07-Conflicting duplicate frames (strong primary key consistency but key observation difference exceeds observation tolerance), E08-Out-of-order arrivals (arrival order differs from the order after stable timestamp sorting exceeding the out-of-order threshold); where, the time backoff tolerance is determined by the statistical distribution of timestamp backoff amplitude of correctable out-of-order segments in historical running data, used to distinguish between slight out-of-order that can be corrected by stable sorting and unacceptable true time backoff; the drift threshold is jointly determined by the nominal value of the sampling period and the sampling period jitter distribution in historical running data, used to identify periodic deviations caused by clock anomalies or task blocking, and outputs a set of available segments organized by segment number.

[0037] This implementation plan avoids distorting the coupling sensitivity fitting and compensation parameter identification by piecing together incomparable samples. On the other hand, it enables modeling training, gating updates, and effect evaluation to perform consistency statistics based on available fragments, remove or reduce the weight of low-confidence fragments, and quickly trace back to the responsibility chain corresponding to the specific anomaly type when closed-loop oscillations or overshoot events occur, thereby significantly improving the quality of compensation parameters, update security, and engineering maintainability.

[0038] Specifically, the process of constructing the compensation parameter set and performing quality scoring is as follows: Input the available fragment set and the basic dataset, calculate the differential features between valve body A and valve body B at the drive layer, scheduling layer, and observation layer, and combine this with the mutual interference sensitivity features extracted from the pipeline network observation response. Calculate the asynchronous differential features frame-by-frame based on the available fragment set: construct paired features for valve body A and B based on the target duty cycle, actual duty cycle, direction position, stage identifier, cumulative pulse count, and energization duration to obtain the duty cycle difference, the actual deviation difference of the target projection, the stage arrival time difference, and the drive intensity difference per unit time; the drive intensity per unit time is determined by the "energization duration increment or sampling period" and the "pulse count increment". The sampling period is calculated to characterize the equivalent propulsion difference in a valve-position encoder-free scenario. Regarding mutual disturbance response feature extraction: within the steady-state segment where the target field remains unchanged, the fluctuation amplitude (peak-to-peak value or variance) of the observed flow rate and differential pressure is statistically analyzed and correlated with the response asynchrony to obtain mutual disturbance amplification characteristics; within the steady-state micro-perturbation segment, local least squares fitting is performed on (drive strength, drive intensity) and (flow rate, differential pressure) to obtain a local coupling matrix. The matrix elements characterize the sensitivity of the valve body drive difference to the redistribution of observed quantities, forming coupling sensitivity characteristics; the phase lag characteristic is obtained by the peak lag of the cross-correlation between the response asynchrony and the flow rate and differential pressure residuals. An actuator difference compensation parameter set and a valve-network coupling sensitivity parameter set are generated.

[0039] The actuator differential compensation parameter set includes at least the equivalent dead zone parameter, equivalent gain parameter, equivalent first-order hysteresis time constant, and hysteresis error parameter. The equivalent dead zone parameter is obtained by locating the minimum effective duty cycle corresponding to the "first time the observed quantity crosses the response threshold" in the step segment, and a robust estimate is formed by statistically analyzing quantiles across multiple segments. The equivalent gain parameter is obtained by estimating the slope of the "drive intensity - observed quantity" in the steady-state segment, and is maintained separately according to temperature, pressure difference, and power supply voltage. The equivalent time constant is obtained by fitting the first-order inertia of the rising segment of the step response. The hysteresis error parameter is estimated by estimating the offset of the response starting point under the same drive intensity in the direction position reversal segment. The valve-network coupling sensitivity parameter set consists of a local coupling matrix and a coupling strength score. The coupling strength score is obtained by combining the matrix condition number, fitting residual, and mutual disturbance amplification characteristics, and is used to measure mutual disturbance sensitivity.

[0040] The compensation parameter set is normalized and regularized: the dead zone, gain, time constant, and coupling matrix elements are linearly normalized to obtain the parameter vector; the upper and lower bounds of the linear normalization are determined according to the physical feasible range and historical steady-state distribution: a lower and upper bound are given for each dimension parameter, the lower and upper bounds are given by the hardware specifications and structural constraints to provide an initial feasible range, and the upper and lower quantile values ​​of the parameter are statistically analyzed in the steady-state segment of historical running data to shrink and correct the feasible range; smoothing constraints and regularization terms are added during the fitting process, and parameter jumps in adjacent update cycles are limited by penalizing the "parameter norm" and "parameter increment" to avoid numerical divergence caused by limited samples or noise; the regularization constraint includes at least two terms: a magnitude penalty term for the normalized parameter vector (limiting the overall scale of the parameters) and an increment penalty term for the difference of the normalized parameters between two adjacent updates (limiting the abrupt jump amplitude), the weights of which are determined by the statistical correlation of "parameter jump amplitude, closed-loop oscillation and overshoot event" in historical running data. For parameters with continuous update cycles, an exponentially weighted moving average is used for time smoothing to track the slow drift of pipeline operating conditions and actuator status and suppress the impact of single abnormal segments on parameters.

[0041] The parameter change magnitude characterization value is synchronously calculated by the parameter manager in the control task each time a new version of the compensation parameter set (including actuator difference compensation parameters and valve-network coupling sensitivity parameters) is generated and written into the effective parameter table. Specifically, it is obtained by reading the current effective parameter vector and the previous version parameter vector, taking the norm of the difference vector between the two, and writing the parameter version number, update timestamp, and component differences of the parameter vector into the parameter change field of the data frame to characterize the change magnitude brought about by parameter version iteration; the fitting residual characterization value is output by the coupling prediction and compensation calculation module in the control task. After generating the predicted flow rate and pressure difference for the next time slice, when the measured flow rate and measured pressure difference are collected in the next time slice, the flow rate residual is calculated as the measured flow rate minus the predicted flow rate, and the pressure difference residual is calculated as the measured pressure difference minus the predicted pressure difference. Within a sliding window, the root mean square error, mean absolute error, and maximum absolute residual of the residual sequence are statistically analyzed to obtain the fitting residual characterization value. The above residual statistics and the corresponding window length are written into the residual evaluation field of the data frame to quantify the model fitting error and support the parameter quality scoring and the determination of triggering re-identification or parameter backoff.

[0042] Both the parameter change magnitude characterization value and the fitting residual characterization value are derived from online statistics and records during the closed-loop regulation process. The parameter change magnitude index is generated by the parameter manager of the control task each time the compensation parameter set is updated. The parameter change magnitude index includes at least the compensation parameter version number switching flag, the difference between the old and new versions of key compensation coefficients (such as dead zone compensation, gain compensation, coupling compensation coefficients, etc.), and the equivalent gain drift calculated by "unit time drive intensity and flow and pressure difference response" within the steady-state window. The above changes are written into the parameter change field of the data frame according to the sampling period to characterize the compensation. The compensation parameters change and drift slowly over time; the fitting residual index is predicted by the control task or edge prediction module based on the currently active local coupling model and compensation model for the flow rate and pressure difference (or pressure difference) of the next time slice. Then, the flow rate residual and pressure difference residual are obtained by subtracting the measured observation value from the predicted value. The root mean square error (RMSE), mean absolute error (MAE), maximum absolute residual, and residual variance are statistically analyzed within the sliding window. The above residual statistics are written into the residual evaluation field of the data frame to quantify the model interpretation error and provide a basis for parameter quality scoring, re-identification, and backoff determination.

[0043] The stability term is obtained by multiplying the natural logarithm of the parameter stability score by the parameter stability weighting coefficient; the accuracy term is obtained by multiplying the natural logarithm of the fitting accuracy score by the fitting accuracy weighting coefficient; and the synergy term is obtained by multiplying the natural logarithm of the consistency score by the tradeoff weighting coefficient. The three weighted terms—stability, accuracy, and synergy—are summed and substituted into an exponential function to obtain the quality score of the compensation parameter set, which quantifies the usability of the compensation parameter set. The specific formula for calculating the quality score is as follows: ; In the formula, This represents the quality score of the compensation parameter set, used to quantitatively evaluate the usability of the parameter set; The parameter stability weighting coefficient is determined and normalized by statistically analyzing the sensitivity of the closed-loop stability index under parameter changes in historical operating data. Its value ranges from 0 to 1. It is used to emphasize the requirement of "parameters not changing excessively" in different scenarios and to prevent frequent parameter updates from causing jitter in the controller compensation amount. The parameter stability score is obtained by subtracting the parameter change magnitude from the stability threshold and standardizing it. The stability deviation is then obtained by performing a Logistic mapping on the stability deviation. This score is used to suppress parameter mutations caused by insufficient samples or noise when determining the effectiveness of a parameter. The fitting accuracy weighting coefficient is determined by statistically analyzing the sensitivity between the fitting residual index and the closed-loop tracking performance index (such as steady-state tracking error, transient recovery time, steady-state fluctuation, and mutual disturbance amplification) in historical operating data. The value range is 0-1. It is used to emphasize the requirement that "the parameters should be able to explain the observation and fit the coupling relationship" in different scenarios, and to avoid sacrificing the compensation effectiveness in pursuit of stability. The fitting accuracy score is obtained by subtracting the fitting residual value from the accuracy threshold and standardizing it. The accuracy deviation is then obtained by performing a Logistic mapping on the accuracy deviation. This score is used to determine whether the current parameter has actual compensation value and whether additional calibration segments need to be re-identified. The collaborative consistency weight coefficient is determined by statistically analyzing the impact of "stability and accuracy mismatch" on closed-loop risk in historical operating data. Its value ranges from 0 to 1 and is used to further suppress the mismatch between stability and accuracy. The collaborative consistency score is obtained by first linearly normalizing the parameter change magnitude index and the fitting residual index, and then calculating the absolute difference between the two and applying exponential decay. It is used to improve the robustness of parameter version selection to the real closed-loop risk.

[0044] The system compares the quality score with the quality threshold in real time. When the quality score is lower than the quality threshold, a re-identification is triggered: on the one hand, the next available window is prioritized to execute the calibration sequences of valve bodies A and B and the steady-state perturbation test segment to supplement the samples; on the other hand, the compensation parameter set with a normal quality score from the previous version is used. When the quality score is greater than or equal to the quality threshold, the current compensation parameter set is determined to be available. The actuator difference compensation parameters and coupling sensitivity parameters obtained in this identification are written into the parameter lookup table as the parameters effective in the new version, and the corresponding quality score, segment number set, and timestamp are recorded. At the same time, the asynchronous level label, mutual interference level label, and high-risk closed-loop label set are output. The compensation parameter set, parameter quality score, and risk label set are output.

[0045] Table 1, showing the quality score data for compensation parameters, is used to evaluate the overall usability of the compensation parameter set. For parameter version PARAM-001, the compensation parameter set has the following scores: Valve body A quality score of 0.92, Valve body B quality score of 0.89, average response time of 125, flow control error of 3.2, synchronization error of 8.5, and system stability of 91, indicating good overall usability. For parameter version PARAM-002, the compensation parameter set has the following scores: Valve body A quality score of 0.94, Valve body B quality score of 0.91, average response time of 118, flow control error of 2.8, synchronization error of 6.2, and system stability of 93, with a quality score of 0.91, indicating even better overall usability. The compensation parameter set corresponding to parameter version PARAM-003 has the following characteristics: Valve body A quality score of 0.88, Valve body B quality score of 0.85, average response time of 142 seconds, flow control error of 4.5 seconds, synchronization error of 12.3 seconds, system stability of 85, and a quality score of 0.82, indicating that the overall usability of this compensation parameter set is average. The compensation parameter set corresponding to parameter version PARAM-004 has the following characteristics: Valve body A quality score of 0.91, Valve body B quality score of 0.88, average response time of 130 seconds, flow control error of 3 seconds, synchronization error of 7.8 seconds, system stability of 90, and a quality score of 0.87, indicating that the overall usability of this compensation parameter set is good. The compensation parameter set corresponding to parameter version PARAM-005 has the following characteristics: Valve body A quality score of 0.96, Valve body B quality score of 0.94, average response time of 108, flow control error of 2.1, synchronization error of 4.5, system stability of 96, and a quality score of 0.95, demonstrating excellent overall usability. The compensation parameter set corresponding to parameter version PARAM-006 has the following characteristics: Valve body A quality score of 0.93, Valve body B quality score of 0.92, average response time of 115, flow control error of 2.5, synchronization error of 5.2, system stability of 94, and a quality score of 0.93, demonstrating moderate overall usability.

[0046] Table 1 Compensation Parameter Quality Score Data Table like Figure 4The comparison chart of compensation parameter quality scores visually illustrates the differences in usability and selection criteria for different compensation parameter versions. The top left subplot compares the quality scores of valve body A and valve body B, with a dashed line indicating a quality threshold of 0.85. It can be seen that all versions are generally not lower than the threshold, but the degree of balance between the two actuators varies. The scores of both valve bodies rise simultaneously with a smaller difference, indicating more effective suppression of asynchronous operation of the two actuators and less likelihood of triggering valve-network coupling interference. The top right subplot compares the average response time and flow control error (dual-axis). There are significant differences between the different versions in terms of "fast response" and "low error": PARAM-003 has the longest response time and the highest error, representing a typical version where "compensation mismatch leads to closed-loop blunting / error amplification"; PARAM-005 is superior in both response time and error, demonstrating that the compensation parameters not only improve steady-state error but also enhance the effective closed-loop bandwidth and tracking capability. The bottom left subplot shows a comparison of overshoot rates. The overshoot rate of PARAM-005 is approximately 4.5%, making it the only version significantly below the threshold, indicating stronger suppression of chain oscillations under network coupling conditions. The lower right subplot summarizes the overall score. The overall score of PARAM-005, highlighted, shows a score of approximately 0.95, indicating it is the best overall and can be prioritized.

[0047] In this implementation scheme, by utilizing duty cycle difference, stage arrival time difference, and unit time drive intensity difference, the asynchronous response and effective propulsion difference of the two actuators can still be stably characterized under the condition of no valve position encoder, thereby forming identifiable dead zone, gain, hysteresis, and backlash compensation; providing a unified criterion for gating update, re-identification trigger, and parameter rollback, reducing the risk of mislearning and mutual interference amplification, and improving closed-loop tracking accuracy, synchronous control effect, and overall stability.

[0048] Specifically, the process of generating dual-body valve regulation commands based on the compensation parameter set and performing joint control is as follows: Input the compensation parameter set and the target flow rate and target pressure difference of the current time slice; construct a two-layer coordinated control structure consisting of a main loop for flow rate and pressure difference and a synchronous slave loop for dual valves; calculate the target drive increments for valve body A and valve body B and generate target duty cycle, direction, and stage commands; the main loop uses the target flow rate or target pressure difference as the control target, and calculates the observation error in each sampling period. The observation error includes flow rate error and pressure difference error; the flow rate error is obtained by subtracting the target flow rate from the measured flow rate, and the pressure difference error is obtained by subtracting the target pressure difference from the measured pressure difference; select the main control variable according to the control mode: when using a flow rate closed loop, the flow rate error is the main variable, and the pressure difference error is used as a constraint or auxiliary variable; when using a pressure difference closed loop, the pressure difference error is the main variable, and the flow rate error is used as a constraint or auxiliary variable; the main loop outputs the desired observation increment vector, which includes at least the desired flow rate increment and the desired pressure difference increment, used to describe the direction and magnitude of the observed change expected to be achieved in the next time slice.

[0049] The desired observation increment is mapped to a dual-valve drive increment using a constrained weighted least squares solution based on the local coupling matrix in the coupling sensitivity parameter set. The dual-valve drive increment includes at least the unit-time drive intensity increment of valve body A and the unit-time drive intensity increment of valve body B. The local coupling matrix is ​​derived from the local fitting results of the drive intensity increments of valve body A and valve body B and the observation increments of flow and pressure difference in the steady-state perturbation segment, and is used to characterize the sensitivity of valve body drive to the redistribution of observations. Constraints and regularization terms are introduced when solving for the drive increment: the constraints include at least the upper and lower limits of duty cycle, the upper limit of duty cycle change rate, the minimum hold time for direction switching, and the mutual exclusion rule for stage switching. The regularization term is used to suppress drastic changes in the solution and avoid the divergence of drive increment caused by ill-conditioned coupling matrix. This is achieved by adding a drive increment norm penalty and a drive increment difference penalty with the previous time slice to the solution objective, so that the output drive increment changes smoothly within the physically executable range. The constraint types of weighted least squares with constraints include at least three categories: First, saturation constraints, used to limit the upper and lower bounds of the duty cycle increment of valve body A and valve body B, ensuring that the updated duty cycle falls within the allowable range; second, rate constraints, used to limit the duty cycle change rate per unit time from not exceeding the upper limit of the change rate, avoiding overshoot and mechanical shock caused by sudden drive changes; and third, smoothing constraints, used to limit the difference between the drive increment of the current time slice and the drive increment of the previous time slice from not exceeding the smoothing upper limit, so that the drive output changes continuously. At the same time, the minimum hold time for direction switching and the mutual exclusion rule for stage switching are used as logical constraints: when the hold time is not met, the drive increment is prohibited from crossing the zero point, causing the direction to flip; when the stage mutual exclusion is triggered, the combination of the two valve increments is limited to falling within the allowable set. The dimension of the local coupling matrix is ​​consistent with the observed and driving quantities: when the flow rate increment and pressure difference increment are used as observations and the driving increments of valve body A and valve body B are used as control quantities, the local coupling matrix is ​​a 2x2 matrix, which is used to map the dual valve driving increment to the observed increments of flow rate and pressure difference. When the matrix invertibility is insufficient or close to singular, a weighted least squares solution is obtained by adding a regularization term, so that the solution is equivalent to finding a stable solution in the pseudo-inverse sense. The condition number or the minimum singular value is used as the criterion. When the ill-conditioning degree exceeds the ill-conditioning degree threshold, the coupling compensation weight is reduced or the mode is degenerated into master-slave valve or single valve mode to avoid the divergence of driving increment.

[0050] After generating the dual-valve drive increment, a synchronous slave loop compensation term is superimposed. The synchronous slave loop takes the response asynchrony as the core feedback quantity. The response asynchrony is obtained by combining the difference in actual duty cycle between valve bodies A and B, the difference in drive intensity per unit time, the difference in stage arrival time, and the difference in deviation between the target and the actual value. The synchronous slave loop outputs a synchronous suppression increment, which is to reduce the difference between the two valves: when valve body A advances too fast, it suppresses the drive increment of valve body A and compensates for the drive increment of valve body B; when valve body B advances too fast, it applies the opposite. The gain of the synchronous slave loop is given by the actuator difference compensation parameter set or called according to the working condition bucket. It is at least related to the dead zone difference, equivalent gain difference, and backlash difference. It is used to convert the dynamic difference of the motor, transmission clearance, and friction difference into a synchronous servo adjustment quantity that can be compensated online.

[0051] For incremental dual-valve drives, differential compensation and nonlinear correction are performed: dead zone compensation and friction compensation are performed on sub-valves according to the actuator differential compensation parameters, and a start-up threshold compensation is introduced for the duty cycle range to cross the static friction dead zone; backlash compensation and stage holding protection are introduced for direction switching scenarios to avoid gap swing and mutual interference amplification caused by frequent reversals; anti-integral saturation and amplitude and speed limiting processing are performed on drive commands. When the duty cycle reaches the upper or lower threshold, the integral term is frozen or rolled back to the proportional term to prevent error accumulation and overshoot.

[0052] The main loop drive increment, synchronization suppression increment, and differential compensation and correction are integrated into target duty cycle, direction, and stage commands for valve body A and valve body B. The coupling matrix version number, compensation parameter version number, and synchronization slave loop gain bucket number used to generate the commands are written into the control command record for closed-loop reproduction and parameter traceability. The target drive command sequence for valve body A and valve body B is output, enabling the active suppression of oscillations and overshoot caused by asynchronous operation of the dual actuators and valve-network coupling interference while tracking the target flow rate and target pressure difference.

[0053] like Figure 5The diagram shows a comparison of the effects of dual-valve synchronous control, contrasting the coordinated control effects with and without synchronous compensation. The top diagram shows the condition without synchronous compensation: the actual duty cycle (or equivalent opening) curves of valve A and valve B exhibit continuous phase shifts and amplitude deviations across multiple cycles. This manifests as inconsistent response speeds and non-coinciding peak and valley positions under the same control command. The corresponding "response asynchrony" (green dashed line) fluctuates significantly throughout the process and maintains a high average value, approximately 8.36% as shown in the diagram. This indicates that the dynamic differences in the motors, transmission clearances, and friction differences of the two actuators are not suppressed. Valve position deviations can cause flow / pressure differential redistribution under pipeline coupling, easily inducing valve-network interference, leading to slower closed-loop tracking and increased risks of oscillation or overshoot. The bottom diagram shows the condition with synchronous compensation: the curves of valve A and valve B essentially overlap, peak-valley synchronicity is significantly improved, and the inter-valve deviation band (shaded area) is significantly narrowed. Simultaneously, the "response asynchrony" is suppressed within a low amplitude range, with the average value decreasing to approximately 1.46%. This indicates that the compensation strategy can offset the differences between the two valves in low duty cycle start-up, direction switching return difference, and equivalent gain difference online, enabling the two valves to participate in closed-loop regulation with more consistent drive execution, thereby reducing the flow and pressure redistribution disturbances caused by asynchrony and improving the stability and controllability of the pipeline closed loop.

[0054] In this implementation scheme, the control target not only tracks the set value, but also explicitly constrains the differential pressure and redistribution effect, reducing overshoot and mutual interference amplification caused by a single index drive. The "desired flow rate and differential pressure change" is decomposed into the drive intensity increment of valve body A and valve body B through a two-by-two local coupling matrix, and the solution results are restricted to the physical executable domain by saturation, rate, smoothing and logic constraints. When the coupling matrix is ​​ill-conditioned, it automatically reduces the weight or degenerates to the master-slave valve or single valve mode to avoid drive divergence.

[0055] Specifically, the process of adaptive gating update and safety degradation rollback, and outputting the control strategy state, is as follows: Input the quality score value generated for each time slice; construct an adaptive gating update and safety degradation rollback mechanism; adaptively update the coupling sensitivity parameter and actuator difference compensation parameter; trigger control strategy degradation and parameter rollback; when the fitting residual continuously increases and the mutual interference level label rises, it is determined that the current coupling relationship has drifted or mutual interference has increased, and the system enters the candidate state for allowing update; the gating condition is used to determine whether to perform parameter update: the moving average of the compensation parameter set quality score value over the most recent K consecutive sampling periods is not less than the quality threshold and within the window... The minimum value is not lower than the lower threshold, the freshness score of flow and differential pressure observation is not less than the freshness threshold and the observation arrival lag does not exceed the lag upper limit, and the CRC failure ratio, the average number of retransmissions, the average number of timeouts, and the bus arbitration conflict count or wireless packet loss count within the K window do not exceed their respective upper limits, the absolute value of the period deviation of the driving update task does not exceed the jitter average upper limit and the maximum deviation does not exceed the jitter peak upper limit and the longest critical section blocking time does not exceed the blocking upper limit, the equipment fault flag and reset flag are both zero and the power supply voltage and onboard temperature are within the allowable range and the voltage change rate does not exceed the limit, and the current segment is in a steady-state update window.

[0056] To avoid mislearning link obsolescence, scheduling jitter, and strong transient noise as coupling drift or actuator differences, an update admission value is introduced to determine whether to update. A sliding statistic, taking quality scores over K consecutive sampling periods, is obtained by gating and normalizing using the sliding mean and sliding minimum. An observation freshness factor is obtained by comparing the freshness score with the hysteresis upper limit and then normalizing it. A link health factor is obtained by comparing the CRC failure rate, retransmission count, timeout count, arbitration conflict count, or packet loss count with their respective upper limits and then normalizing it. A scheduling stability factor is obtained by comparing period deviation, jitter peak, and critical section blocking duration with their upper limits and then normalizing it. Device and steady-state window factors are obtained by gating and normalizing using equipment fault and reset flags, power supply voltage and onboard temperature allowable ranges, and steady-state update window indications. The minimum of the sliding statistic and the four factors is calculated and substituted into a truncation function to obtain the update admission value. The formula for calculating the update admission value is as follows: ; In the formula, Indicates the first The update threshold for time slices ranges from 0 to 1 and is used to quantify whether a time slice meets the comprehensive conditions of being "learnable and updatable". Indicates the most recent consecutive The sliding statistic of the quality score value for each sampling period, with a value range of 0-1, is used to characterize the stability and availability of the parameter set in the near term; The observation freshness factor, which represents the freshness of observations and the freshness factor after arrival lag gating, has a value range of 0-1 and is used to suppress erroneous adjustment caused by stale observations. This represents the link health factor, with a value ranging from 0 to 1, and is used to suppress mislearning of interfering link segments; This represents the scheduling stability factor, with a value ranging from 0 to 1, used to suppress the risk of closed-loop oscillations caused by scheduling jitter; This represents the device and steady-state window factor, with a value range of 0-1, used to ensure that updates are only made in steady-state segments where the device is healthy and the target is not switching.

[0057] The system compares and updates the admission value and the admission threshold in real time. Only when the updated admission value is greater than or equal to the admission threshold is the parameter update of the coupling matrix and compensation parameters allowed. Otherwise, the update is frozen to avoid mislearning noise as coupling drift or actuator difference in the case of outdated links, scheduling jitter or strong transient phases. When the updated admission value is less than the admission threshold, the parameter update of the current time slice is stopped and the previous version of the coupling matrix and compensation parameter set continues to be effective. At the same time, the time slice is marked as an unupdateable segment and the freeze count is accumulated. When the number of consecutive freezes reaches the preset number or the fitting residual continues to rise, the rollback or re-identification process is triggered.

[0058] The parameter adaptive update adopts a "local, bucketed, and limited" strategy: For coupling sensitivity parameters, recursive least squares updates are performed within the current operating condition bucket (temperature range, differential pressure range, power supply voltage range), outputting a new local coupling matrix and calculating the fitting residual and condition number as availability criteria; for actuator difference compensation parameters, only subsets strongly correlated with current observation evidence are updated. For example, when low duty cycle response hysteresis is significant, the start-up threshold or dead zone parameter is updated; when the fallback hysteresis occurs after direction switching, the backlash parameter is updated; when the difference in response slope between the two valves is stable under the same drive intensity, the equivalent gain difference parameter is updated; each update applies an upper limit to the parameter increment and performs time smoothing to ensure slow parameter evolution; safety degradation and rollback mechanisms include: freezing parameter updates and triggering control strategy degradation when the quality score value is below the quality threshold or the high-risk closed-loop label is set.

[0059] The control strategy degradation includes at least two modes: one is the "master-slave valve synchronization mode", which selects the valve body with higher historical stability, healthier link, or more stable actuator response as the master valve, which bears the main adjustment, while the other valve body acts as the slave valve and only executes the synchronization suppression term to follow the master valve, reducing the coupling degree of freedom to suppress mutual interference; the other is the "single valve steady-state holding mode", which fixes one valve near the most recent steady-state opening when mutual interference is amplified or the link is outdated, allowing only the other valve to make small adjustments, avoiding redistribution oscillations caused by the simultaneous action of the two valves; Parameter rollback includes: when the coupling residual increases instead of decreasing within several consecutive windows, or the number of conditions in the coupling matrix exceeds the condition number threshold, rollback uses the compensation parameter set and coupling matrix version with normal quality scores from the previous version, and records the rollback reason code, rollback timestamp, and associated segment number; after entering the rollback state, the next available window is prioritized for executing steady-state perturbation segments or calibration segments to supplement quality samples that can be used for re-identification; outputting the control strategy state with gating update capability. This ensures stable closed-loop operation even in engineering scenarios with strong pipeline coupling, significant actuator asynchrony, and fluctuations in links and scheduling, and provides traceable degradation and rollback closed-loop functionality.

[0060] In this implementation, outdated links, task jitter, and noise samples in strong transient phases are automatically excluded from the learning domain to avoid "mislearning noise as drift". The local coupling matrix and actuator difference compensation parameters are only updated in small steps within the current operating condition bucket, and the residuals and condition numbers are calculated simultaneously as availability criteria to ensure that the parameters evolve slowly and do not cause the drive to diverge due to ill-conditioned coupling.

[0061] Specifically, the process of establishing an associated index for each sampling time slice and performing integrated version archiving and playback recalculation is as follows: For each sampling time slice, an associated index is established for data, instructions, parameters, and firmware. Versioned archiving and playback recalculation are performed, outputting a traceable closed-loop operational evidence chain. Using the device number, segment number, and sampling sequence number as primary keys, the target field, measured field, link statistics field, scheduling jitter field, and quality marker field of the same time slice are aggregated to form a time-slice-level operational record. For each time-slice-level operational record, the duty cycle, direction, and stage of valve body A or B target, as well as the corresponding instruction generation time, are bound to the control instruction record. The version number of the compensation parameter, coupling matrix, admission threshold, and firmware that were effective at that time are also bound to form a version context. Simultaneously, a gating criterion vector is written to bind the input and output of a control decision to an updatable context, avoiding the inability to reproduce the judgment result of the gating conditions at that time by only recording the parameter version.

[0062] On the onboard side, immutable append writing is performed on the original closed-loop acquisition records, control command records, and parameter change records. A three-layer digest mechanism—time-slice hashing, fragment hash tree root value, and cross-link witnessing—is used to generate an evidence chain digest. The first layer is time-slice hashing, where key fields are concatenated in a fixed serialization order for each time slice and then hashed. Key fields include at least the sampling sequence number, sampling timestamp, target flow or target pressure difference, measured flow and measured pressure difference, target drive commands for valve A and valve B, actual drive execution amounts for valve A and valve B, link statistics, scheduling jitter, and gating criterion vector. The second layer is fragment-level fragment hash tree root. The first layer constructs a hash tree by sequentially organizing all time-slice-level hashes within the fragment to obtain the root hash. The root hash value, along with the fragment's start and end timestamps, sampling sequence range, and parameter version range, is written to the fragment header. The second layer is cross-link witnessing. The fragment-level hash tree root value in the fragment header, along with the key link event digest (such as a sudden increase in verification failures, a sudden increase in timeouts, reset events, and parameter version switching events), is written to the independent witness logs of the wired bus link and the wireless link to form dual-channel witness consistency. The platform side only determines the fragment to be valid when the fragment-level hash tree root values ​​in the witness logs of the two links are consistent or can be aligned within the allowed time window. This is used to suppress single-path tampering or retransmission insertion.

[0063] The platform performs idempotent deduplication, hash tree proof verification, cross-link witness consistency verification, and order reordering on the segments. The segments that pass the verification are written into the time series database, and foreign key associations are established in the segment directory table, time slice index table, parameter version table, gating criterion table, and alarm event table. The gating criterion table records the gating criterion vector and threshold version, supporting offline review of "falsely released updates" and "falsely frozen updates".

[0064] The replay recalculation employs an operator-based replay mechanism and a control-based attribution mechanism. The original replay operator is defined as a control algorithm that replays the data sequentially over a specified segment. The inputs are the target value, observations, link freshness, and scheduling jitter recorded at that time. It calls the coupling matrix and compensation parameters in effect at that time, and the output is the predictive control output and predicted observations. The replay deviation is obtained by subtracting these from the actual recorded control commands and observations. To attribute the deviation to root causes such as link obsolescence, scheduling jitter, drive saturation, parameter mismatch, and gating misjudgment, a family of control replay operators is constructed, and the corresponding replay deviation is calculated for each operator. The control replay operators include at least: a link de-obsolescence operator, which replaces the observation input with fresh observations compensated for arrival lag or aligned observations obtained through nearest neighbor interpolation; a scheduling de-jitter operator, which resamples the sampling and drive updates according to an ideal period and removes non-uniform updates caused by task blocking; and a drive de-saturation operator, which replaces the drive execution model with an ideal actuator model with no dead zone, no backlash, and infinite amplitude, or replaces the recorded actual... The actual execution quantity is inversely solved into the equivalent control quantity and then replayed; the parameter rollback operator replaces the coupling matrix and compensation parameters with the previous version or a candidate version with a higher quality score in the same working condition bucket before replaying; the gating counterfactual operator forces the update access judgment to be allowed or disallowed and enables or freezes the online parameter update logic accordingly before replaying, which is used to verify the impact of gate mis-release or mis-freezing on closed-loop deviation; for each time slice, the deviation reduction of each control replay operator relative to the original replay operator is compared, the root cause category with the largest deviation reduction is taken as the master attribution code, and the deviation reduction corresponding to each category is written as the attribution contribution into the attribution vector; when the replay deviation exceeds the deviation threshold and the deviation reduction corresponding to the master attribution code exceeds the attribution threshold, the time slice is marked as an attributable deviation event, and the master attribution code, attribution contribution, associated parameter version and gating criterion vector are recorded, which is used to upgrade the execution link deviation from a single alarm to an explainable, locatable and reproducible root cause event; output the closed-loop operation evidence chain.

[0065] This implementation scheme enables provable detection of missing, inserted, tampered, and version mismatched segments; it allows any closed-loop deviation to be reproduced and its explainable root cause located within the context of the input and version at the time, thereby significantly improving traceability, reproducibility, and fault attribution efficiency, and reducing the risk of false alarms and blind rollbacks.

[0066] Specifically, the process of online health assessment, alarm classification, and automatic rollback is as follows: Input the quality score value for each time slice, update the admission value, link health factor, and scheduling stability factor; conduct online health assessment and handling rules; and generate alarm events, degradation and rollback actions, and handling result records. The online health assessment is based on sliding window statistics: calculate the moving average, moving upper quantile, and duration of continuous over-limit for the quality score value, fitted residual characterization value, response asynchrony, mutual interference sensitivity characteristics, observation freshness score, and scheduling jitter index, respectively, to form a health vector; introduce a risk score value on the basis of the health vector, and fuse risk evidence into a single risk quantity for alarm classification.

[0067] Multiplying the out-of-limit value of the indicator by the amplitude sensitivity coefficient yields the out-of-limit term; multiplying the out-of-limit term by the product of the proportion of consecutive out-of-limit duration and the persistence amplification coefficient yields the persistence term; substituting the persistence term into the arctangent function yields the contribution term; summing the contribution terms corresponding to the indicators participating in the fusion sequentially, and then adding them to the numerical stability term yields the numerator; summing the weight coefficients of all indicators and dividing by the numerator yields the denominator; summing the weight coefficients and dividing by the denominator yields the risk score; the specific calculation formula for the risk score is as follows: ; In the formula, This represents the risk score value for time slice t, with a value ranging from 0 to 1. It is used to quantify the current closed-loop operation risk and drive alarm classification and handling actions. The number of indicators participating in the fusion is determined by the number of entries in the risk evidence indicator set configured by the system, and is used to determine the dimension of risk fusion calculation; Indicates the first The weight coefficients of the indicators are obtained by statistically analyzing the correlation and contribution of "indicator over-limit and alarm or rollback events" or "indicator over-limit and closed-loop oscillation or mutual interference labels" in historical operating data. The values ​​range from 0 to 1 and are used to adjust the contribution ratio of different indicators to the overall risk score. This is a numerical stability term, obtained by setting a constant at the implementation end to avoid the denominator being zero. Its value range is a positive number greater than 0 and less than 0.01, used to ensure the stability of harmonic fusion calculation and avoid numerical anomalies. Indicates the first The out-of-limit values ​​of a class of indicators are obtained by subtracting the observed value of the indicator from the trigger threshold and then standardizing it according to the scale parameter. The value range is real numbers and negative values ​​are allowed. It is used to unify indicators of different dimensions to a comparable out-of-limit scale. This indicates the percentage of duration or percentage of consecutive exceedances in the most recent K sampling periods. It is obtained by statistical analysis of exceedance markers within a sliding window and has a value range of 0 to 1. It is used to make continuous exceedances more sensitive to risk. The amplitude sensitivity coefficient is obtained by fitting and calibrating the "probability of occurrence of over-limit values ​​and alarm and rollback events or mutual interference level labels" in historical operating data. The value range is a real number greater than 0, which is used to adjust the response slope of the risk evidence strength to the "instantaneous over-limit amplitude". The continuous amplification factor is determined by the statistical relationship between the "proportion of continuous over-limit and closed-loop oscillation, mutual interference amplification and false update events" in historical operating data. The value range is a real number greater than 0, which is used to adjust the over-limit amplitude and the amplification intensity of continuous over-limit on risk evidence.

[0068] When the update admission value remains below the admission threshold for the most recent K consecutive sampling periods and the fitted residual characterization value shows a continuous upward trend, or the risk score remains above the drift risk threshold, it is determined to be an unlearnable drift risk, triggering a freeze update and recording the reason code; when the quality score falls below the quality threshold, or the risk score is above the parameter unavailability threshold, it is determined to be a parameter unavailability risk, triggering parameter rollback and writing the rollback version number to the control strategy status; when the link health factor or observation freshness score remains below the threshold for a long period within the sliding window, or the risk evidence strength corresponding to the link health gap remains high and pushes the risk score value across the observation freshness threshold, it is determined to be an unlearnable drift risk. When the confidence threshold is insufficient, it is determined to be a risk of insufficient observation confidence, triggering the control strategy to downgrade to master-slave valve synchronization mode or single valve steady-state maintenance mode, and increasing the observation freshness gating strength; when the scheduling stability factor is lower than the threshold and uneven drive updates continue to occur, or when the scheduling jitter intensity corresponds to the risk evidence strength that remains high and pushes the risk score value to cross the scheduling bandwidth insufficient threshold, it is determined to be a risk of insufficient scheduling bandwidth, triggering a reduction in control update frequency, expansion of drive speed limit and freezing of parameter updates; when a reset event occurs, it is determined to be a device safety risk, triggering a safety shutdown, locking the valve position to a safe opening or switching to a minimum opening, and reporting an emergency alarm. Risk assessment employs a mutually exclusive priority mechanism: equipment safety risks have the highest priority, and once triggered, they override and terminate other risk assessments and handling; parameter unavailability risks take precedence over non-learnable drift risks, and when both are met simultaneously, parameter rollback and freeze updates are executed first to avoid freezing without rollback when parameters are unavailable; non-learnable drift risks are triggered only when the parameter quality score is still not lower than the quality threshold and the parameter version does not need to be rolled back, and are used to identify a state where the parameter is available but not allowed to be learned; insufficient observation reliability risks and insufficient scheduling bandwidth risks are operating environment risks, and when triggered, they can coexist with parameter unavailability and non-learnability, but handling actions (degradation, rate limiting, freeze) must not cancel the already effective parameter rollback and safety stop updates.

[0069] Automatic rollback handling includes: writing a rollback pointer for each parameter update and each effective version; the rollback pointer records the previous version number, the previous version quality score, the previous version's applicable operating condition bucket, and the effective time; when a rollback is triggered, the previous version's coupling matrix and compensation parameters are restored according to the rollback pointer, and the version difference before and after the rollback, the triggering condition, the associated segment number, and the effects after the rollback (residual change, oscillation number change, mutual interference label change) are written into the handling result record; for the case of frequent rollbacks of the same device in a short period of time, a locking learning strategy is triggered: automatic updates are prohibited within a preset locking window, and only steady-state perturbation segments are allowed to be collected for offline re-identification; an alarm event table, a degradation and rollback action table, and a handling result table are output.

[0070] This implementation plan enables the early identification of risk trends and automatic triggering of freeze updates, parameter rollback, and control strategy degradation under complex operating conditions such as outdated links, scheduling jitter, asynchronous dual valves, and valve-network interference. This avoids compensation drift and oscillation amplification caused by mislearning. At the same time, it maintains closed-loop controllability and stable operation when observation reliability is insufficient or bandwidth decreases. The results of the handling are recorded to form a verifiable operation and maintenance closed loop, which significantly improves the stability and robustness of pipeline network flow closed-loop regulation.

[0071] Specifically, the second aspect of the present invention provides a coordinated control system for dual-body valves for closed-loop regulation of pipeline flow, applied to a coordinated control method for dual-body valves for closed-loop regulation of pipeline flow, comprising: a data acquisition and transmission module, used to acquire dual actuator drive information and pipeline observation information to construct a dual-body valve dataset, introduce a unified time base for alignment and correction, and perform segmented scaling and normalization processing; by synchronously acquiring target commands, actual execution quantities, pipeline observations, link statistics and scheduling jitter at the granularity of sampling period as time slice, the data source is ensured to be traceable and can directly support the calculation of asynchronous and mutual interference risks. The data processing and analysis module is used to fragment and reassemble the data frame set of the dual-body valve and filter for timing consistency, generating a usable fragment set. Through stable sorting, deduplication, gap detection, and fragment boundary segmentation, it isolates different control contexts and abnormal fragments, constructs a compensation parameter set, and performs quality scoring. The valve coordination control and compensation module is used to generate dual-body valve adjustment commands based on the compensation parameter set, perform joint control, adaptive gating updates, and safety degradation rollback. It solves the dual-valve drive increment using local coupling matrices and constrained weighted least squares, and only updates parameters in small steps when the update threshold is met, outputting the control strategy state. The optimization and fault diagnosis module is used to establish an associated index for each sampling time slice, perform integrated version archiving and playback recalculation, and perform online health assessment, alarm classification, and automatic rollback handling. It binds data frames, control commands, parameter versions, and gating criteria into a version context and supports playback comparison, enabling the location of abnormal events to the link, scheduler, actuator, or parameter mismatch root cause and forming executable rollback and handling records.

[0072] In this implementation scheme, the inter-valve asynchronous operation and valve-network interference are measurable, identifiable, compensable, and traceable: target instructions, execution quantities, observations, links, and scheduling states are unified and normalized to the same time base using time-slice-level data frames, so that the interference risk and compensation quality can be directly calculated; different control contexts and abnormal segments are isolated through fragmented recombination and quality screening to ensure the reliability of compensation parameter identification and quality scoring.

[0073] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0074] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A coordinated control method for two-body valves for closed-loop regulation of pipeline flow, characterized in that, Includes the following steps: S1. Collect dual actuator drive information and pipeline observation information to construct a dual-body valve dataset, introduce a unified time base for alignment and correction, and perform segmented scaling and normalization processing. S2, perform fragmentation and timing consistency screening on the data frame set of the dual-body valve to generate a set of usable fragments, construct a set of compensation parameters and perform quality scoring; S3 generates dual-body valve adjustment commands based on the compensation parameter set, performs joint control, performs adaptive gating update and safety degradation backoff, and outputs the control strategy state; S4 establishes an associated index for each sampling time slice, performs integrated version archiving and playback recalculation, and conducts online health assessment, alarm classification, and automatic rollback processing.

2. The dual-valve coordinated control method for closed-loop regulation of pipeline flow according to claim 1, characterized in that: The specific process for constructing a dual-body valve dataset by collecting dual-actuator drive information and pipeline observation information is as follows: The basic capability information of the dual-body valve control unit is collected through the device management interface, control board hardware configuration, firmware parameter table and factory-written information. The dual actuator drive information and pipeline observation information are collected through valve action calibration sequence and pipeline steady state and disturbance operation process to construct dual-body valve dataset. The dual-body valve dataset includes: basic dataset, drive dataset, observation dataset, scheduling dataset and timestamp dataset; the closed-loop regulation historical acquisition sequence that has been stored under the same installation scenario and configuration conditions is used as historical operation data.

3. The dual-valve coordinated control method for closed-loop regulation of pipeline flow according to claim 1, characterized in that: The specific process of introducing a unified time base for alignment and correction, and performing piecewise scaling and normalization is as follows: Based on the acquired dual-body valve dataset, a unified time base is introduced to align and correct the time fields of sampling tasks, drive update tasks, and communication arrival events. A set of dual-body valve data frames is obtained with the sampling period as the basic unit. For each time slice record, time sequence consistency verification, link freshness verification, and quality detection and marking of asynchronous and mutual interference risks are performed: For each time slice record in the dual-body valve dataset, the dual-drive channel mapping, communication link type, and sampling configuration version of the device are retrieved from the basic dataset according to the device number. The consistency verification of the target drive command and actual drive execution of valve body A and valve body B in the current time slice is performed; Segmented scaling normalization and unit diameter unification processing are performed: The flow observation value, differential pressure, and pressure observation value are segmented and scaled according to the sensor calibration range and unit conversion rules, and mapped to a unified range; The normalized fields are written into the dual-body valve data frame. Using the unified sampling timestamp as the index, the target field, measured field, context field, and quality field are integrated into a single dual-body valve data frame record, and the dual-body valve data frame set is output.

4. The dual-valve coordinated control method for closed-loop regulation of pipeline flow according to claim 1, characterized in that: The specific process of fragmenting and reassembling the dual-body valve data frame set and performing time-series consistency filtering to generate a usable fragment set is as follows: Input a set of dual-body valve data frames, perform stable sorting, deduplication, and gap detection on a unified time axis, and aggregate data frames that meet continuity and quality conditions into a segment-level closed-loop sequence: The dual-body valve data frame set is stably sorted using a unified sampling timestamp and sampling sequence number as keys. Data frames with backward, jump, or duplicate sampling sequence numbers, or backward timestamps, are marked with timing anomaly tags, and the anomaly type, anomaly occurrence timestamp, preceding and following sequence numbers, associated valve body identifier, and associated channel number are written into a timing anomaly list; stage identification is performed within each segment: the target... The time window in which the field remains unchanged is used as a candidate steady-state window. The sliding variance, sliding slope, and peak-to-peak value are calculated for the flow observation, differential pressure observation, and pressure observation. When the sliding variance is lower than the steady-state threshold and the absolute value of the slope is lower than the slope threshold, it is marked as a steady-state segment. When the target field undergoes a step change or the valve body enters the acceleration and return-to-zero stage, it is marked as a transient segment. The segments are filtered for availability: data frames with missing key fields, equipment fault flags set, continuous link CRC failures causing the observations to be outdated, and continuous jitter exceeding the limits of the driver update task are removed. The set of available segments is output by segment number.

5. The dual-valve coordinated control method for closed-loop regulation of pipeline flow according to claim 1, characterized in that: The specific process of constructing the compensation parameter set and performing quality scoring is as follows: Input the available fragment set and the basic dataset, calculate the difference features between valve body A and valve body B at the drive layer, scheduling layer and observation layer, combine the mutual interference sensitivity features extracted from the pipeline network observation response, and generate the actuator difference compensation parameter set and the valve-network coupling sensitivity parameter set; normalize and regularize the compensation parameter set: linearly normalize the dead zone, gain, time constant and coupling matrix elements to obtain the parameter vector; The stability term is obtained by multiplying the natural logarithm of the parameter stability score by the parameter stability weighting coefficient; the accuracy term is obtained by multiplying the natural logarithm of the fitting accuracy score by the fitting accuracy weighting coefficient; and the coordination term is obtained by multiplying the natural logarithm of the coordination consistency score by the trade-off weighting coefficient. The three weighted terms of stability, accuracy, and coordination are added together and substituted into the exponential function to obtain the quality score of the compensation parameter set. The quality score is compared with the quality threshold in real time. When the quality score is lower than the quality threshold, re-identification is triggered: the next available window is prioritized to execute the calibration sequence of valve body A and B and the steady-state perturbation test segment to supplement the sample. When the quality score is greater than or equal to the quality threshold, the current compensation parameter set is determined to be in an available state. The actuator difference compensation parameters and coupling sensitivity parameters obtained in this identification are written into the parameter lookup table as the new version effective parameters. The compensation parameter set, parameter quality score, and risk label set are output.

6. The dual-valve coordinated control method for closed-loop regulation of pipeline flow according to claim 1, characterized in that: The specific process of generating dual-body valve adjustment commands based on the compensation parameter set and performing joint control is as follows: Input the compensation parameter set and the target flow rate and target pressure difference of the current time slice to construct a two-layer coordinated control structure of the main loop for flow rate and pressure difference and the synchronous slave loop for dual valves. Calculate the target drive increment for valve body A and valve body B and generate the target duty cycle, direction, and stage commands. The main loop uses the target flow rate or target pressure difference as the control target and calculates the observation error in each sampling period, including flow rate error and pressure difference error. The main loop outputs the desired observation increment vector. Using the local coupling matrix in the coupling sensitivity parameter set, the desired observation increment is mapped to the dual valve drive increment using constrained weighted least squares solution. After generating the dual valve drive increment, the synchronous slave loop compensation term is superimposed. The synchronous slave loop outputs the synchronization suppression increment and performs difference compensation and nonlinear correction on the dual valve drive increment: dead zone compensation and friction compensation are performed on the sub-valves according to the actuator difference compensation parameters, and the target drive command sequence for valve body A and valve body B is output.

7. The dual-valve coordinated control method for closed-loop regulation of pipeline flow according to claim 1, characterized in that: The specific process of performing adaptive gating updates and security degradation rollbacks, and outputting the control strategy state is as follows: Input the quality score generated per time slice to construct an adaptive update and safety degradation rollback mechanism for risk gating. Adaptively update the coupling sensitivity parameter and actuator difference compensation parameter to trigger control strategy degradation and parameter rollback. To avoid mislearning link staleness, scheduling jitter, and strong transient noise as coupling drift or actuator difference, an update threshold is introduced to determine whether to update. After gating with the moving average and moving minimum, a moving statistic is obtained by normalization after taking the quality score values ​​for K consecutive sampling periods. The observation freshness factor is obtained by comparing the freshness score with the hysteresis upper limit and normalizing it. The link health factor is obtained by comparing the CRC failure rate, retransmission count, timeout count, arbitration conflict count, or packet loss count with their respective upper limits and normalizing it. The scheduling stability factor is obtained by comparing the period deviation, jitter peak, and critical section blocking duration with the upper limits and normalizing it. The device and steady-state window factors are obtained by gating and normalizing the equipment fault and reset flags, the allowable range of power supply voltage and onboard temperature, and the steady-state update window indication. Calculate the minimum of the moving statistic and the four factors, and substitute it into the cutoff function to obtain the updated admission value; The system compares and updates the admission value and the admission threshold in real time. The system allows parameter updates to the coupling matrix and compensation parameters only when the updated admission value is greater than or equal to the admission threshold. When the updated admission value is less than the admission threshold, the system stops parameter updates for the current time slice and keeps the previous version of the coupling matrix and compensation parameter set in effect. The system outputs the control strategy state with gating update capability.

8. The dual-valve coordinated control method for closed-loop regulation of pipeline flow according to claim 1, characterized in that: The specific process of establishing an associated index for each sampling time slice and performing integrated version archiving and playback recalculation is as follows: For each sampling time slice, an associated index is established for data, instructions, parameters, and firmware. Versioned archiving and replay recalculation are performed to output a traceable closed-loop operation evidence chain. Using the device number, segment number, and sampling sequence number as primary keys, target fields, measured fields, link statistics fields, scheduling jitter fields, and quality marker fields for the same time slice are aggregated to form a time slice-level operation record. On the onboard side, the original closed-loop acquisition records, control instruction records, and parameter change records are immutably appended, and an evidence chain digest is generated using a three-layer digest mechanism of time slice hash, segment hash tree root value, and cross-link witness. On the platform side, idempotent deduplication, hash tree proof verification, cross-link witness consistency verification, and order reordering are performed on the segments. Replay recalculation adopts an operator-based replay and comparative stripping attribution mechanism. A closed-loop operation evidence chain is output.

9. The dual-valve coordinated control method for closed-loop regulation of pipeline flow according to claim 1, characterized in that: The specific process for conducting online health assessments, alarm classification, and automatic rollback is as follows: Input the quality score for each time slice, update the admission value, link health factor, and scheduling stability factor to perform online health assessment and handling rules, and generate alarm events, degradation and rollback actions, and handling result records: The online health assessment introduces a risk score, integrates risk evidence into a single risk quantity for alarm classification, multiplies the index exceeding the limit by the amplitude sensitivity coefficient to obtain the exceeding item; multiplies the exceeding item by the product of the proportion of the duration of continuous exceeding the limit and the persistence amplification coefficient to obtain the persistence item, substitutes the persistence item into the arctangent function to obtain the contribution item; the contribution items corresponding to the indicators participating in the fusion are added sequentially and then added to the numerical stability item to obtain the numerator; The risk score is obtained by summing the weight coefficients of all indicators and dividing by the numerator. When the risk score value is consistently higher than the drift risk threshold, it is determined to be an unlearnable drift risk, triggering a freeze update and recording the reason code. When the risk score is higher than the parameter unavailability threshold, it is determined as a parameter unavailability risk, triggering parameter rollback and writing the rollback version number into the control strategy status; automatic rollback handling includes: writing a rollback pointer for each parameter update and each effective version, and the rollback pointer records the previous version number, the previous version quality score, the previous version applicable working condition bucket and the effective time. Output alarm event table, degradation and rollback action table, and handling result table.

10. A two-body valve coordinated control system for closed-loop regulation of pipeline flow, employing the two-body valve coordinated control method for closed-loop regulation of pipeline flow as described in any one of claims 1-9, characterized in that, include: The data acquisition and transmission module is used to collect dual actuator drive information and pipeline observation information to construct a dual-body valve dataset, introduce a unified time base for alignment and correction, and perform segmented scaling and normalization processing. The data processing and analysis module is used to fragment and reassemble the data frame set of the dual-body valve and filter the timing consistency to generate a set of usable fragments, construct a set of compensation parameters and perform quality scoring. The valve coordination control and compensation module is used to generate dual-body valve adjustment commands based on the compensation parameter set, perform joint control, perform adaptive gating updates and safety degradation backoff, and output the control strategy status. The optimization and fault diagnosis module is used to establish an associated index for each sampling time slice, perform integrated version archiving and playback recalculation, and conduct online health assessment, alarm classification, and automatic rollback processing.

Images