An adaptive control method and system for the opening degree of a casting valve

By establishing an equivalent opening hidden state and an online correction method, the problems of valve hysteresis and dead zone effects in the casting valve control system were solved, and the stability and consistency of the casting process were achieved.

CN123067154APending Publication Date: 2026-11-10SHANGHAI TARGET IND CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611375675.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-09-07
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

The existing pouring valve control system fails to effectively characterize changes in valve hysteresis, dead zone, and flow gain per unit opening, resulting in a mismatch between control commands and actual flow capacity, causing repeated adjustments to the opening at the end of the pouring process and deviations in the cumulative pouring volume.

Method used

By establishing an equivalent implicit opening state that includes valve hysteresis, dead zone, and unit opening flow gain, the casting output is predicted, and the equivalent implicit opening state is corrected online based on the cumulative material balance residual, generating mechanical opening control commands.

Benefits of technology

This achieves the matching of mechanical valve position with actual flow capacity, reduces redundant compensation of control commands and deviation of casting volume, and improves the stability and consistency of the casting process.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123067154A_ABST
    Figure CN123067154A_ABST
Patent Text Reader

Abstract

This invention relates to the field of automatic control technology, specifically to an adaptive control method and system for the opening of a casting valve. The method acquires the control command sequence, valve position feedback sequence, and casting output data of the casting valve. Based on the relationship between the control command and the valve position response, it establishes an equivalent opening latent state, including hysteresis state, dead zone state, and unit opening flow gain. The casting output is predicted based on the latent state, and the latent state is corrected online using the cumulative material balance residual aligned with the delay interval. Based on the corrected latent state, the target casting trajectory is inverted into the target equivalent opening. A mechanical opening control command is generated by combining hysteresis branches, dead zone boundaries, and parameter reliability. This invention can distinguish between the mechanical valve position and the actual flow capacity, reduce redundant compensation caused by response delay, suppress opening repetition and cumulative casting volume overshoot in the final stage of casting, and maintain control consistency under different movement directions and casting batches.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of automatic control technology, specifically to an adaptive control method and system for the opening degree of a casting valve. Background Technology

[0002] The opening control of the pouring valve is a crucial aspect of continuous process automatic control. Common schemes use the target pouring volume, target flow rate, or liquid level of the object being poured as setpoints, and the feedback from flow meters, weighing devices, or liquid level detectors. The controller calculates the current deviation and outputs an opening adjustment command to the pouring valve drive mechanism accordingly. The controller typically employs a combination of proportional, integral, and derivative control, or a staged opening table, maintaining a larger opening at the beginning of pouring and gradually decreasing it as it approaches the target value. Valve position feedback is generally used directly as the actual opening in the control calculation, with the controller assuming a stable correspondence between the mechanical valve position and the flow capacity. When the actual pouring output is detected to be lower than the target trajectory, the controller increases the valve opening; when the actual pouring output is higher than the target trajectory, the controller decreases the valve opening. This type of scheme has a clear structure and is easy to tune under fixed operating conditions, but its control is primarily based on instantaneous deviations and mechanical valve positions, lacking a unified description of valve drive hysteresis, dead zones, and changes in flow capacity per unit opening.

[0003] To reduce the impact of the nonlinear characteristics of the casting valve on the control process, existing solutions also include opening compensation tables, dead zone compensation amounts, or hysteresis correction coefficients in addition to the basic closed-loop control. Opening compensation tables are typically obtained through offline testing, pairing multiple mechanical opening points with corresponding flow rates, and then using piecewise interpolation to establish a mapping relationship between opening and flow. Dead zone compensation usually maintains the valve position unchanged when the change in control command is less than a set threshold, and increases a fixed compensation amount when the threshold is exceeded. Hysteresis correction uses different transition curves for the opening and closing directions. Some solutions also switch parameter groups based on the casting batch, material temperature, or cumulative usage time. These measures can cover several pre-calibrated operating states, but most compensation parameters are determined before control operation. After mechanical wear, changes in transmission clearance, accumulation of deposits in the flow channel, and changes in material flow state, the original mapping relationship will gradually deviate from the actual state. Fixed compensation amounts cannot reflect the continuous changes in hysteresis and flow gain over time.

[0004] Some existing technologies update the control parameters of the casting valve using online identification or model prediction. They estimate flow gain, response time, or control model parameters based on the error between valve position feedback and actual casting output, and then use the updated model to calculate subsequent opening commands. A common implementation involves collecting control commands, valve position feedback, and flow data within a fixed time window, continuously correcting the model according to the minimum error criterion, or directly adding the cumulative casting volume deviation to the opening control value. These solutions often use the valve position feedback value as the true valve state, and the model parameters are used to correct the mapping from valve position to flow, without separating the mechanical opening from the actual flow capacity. When there is a transmission delay in the control action of the casting valve, the output change detected in the current sampling period may originate from previous control commands. If this is directly used to update the current model, it can easily lead to time mismatch. Instantaneous flow fluctuations and continuous changes in flow capacity may also simultaneously enter the parameter update process, causing the model parameters to repeatedly change around multiple interpretations, resulting in continuous additions or reversals of the opening command.

[0005] Control systems typically equate the mechanical valve position with the valve's actual flow capacity, lacking an equivalent opening state that can simultaneously characterize hysteresis, dead zone, and changes in flow gain per unit opening. They also lack a mechanism for time-consistent correction of this state using cumulative material relationships. During the casting process, there is a sequential delay between the control command, valve position arrival, and actual casting output. The same mechanical opening may correspond to different actual outputs in different directions of movement and at different stages of use. When the controller continues to increase or decrease the opening based on instantaneous deviations, control actions already issued but not yet reflected in the detection results are repeatedly compensated for; when delayed outputs subsequently appear in concentrated bursts, the deviation between the mechanical opening and the actual flow capacity is amplified. Therefore, the control system struggles to obtain a control quantity that matches the current valve's true flow state, directly causing repeated opening adjustments at the end of casting, cumulative casting exceeding the target value, and decreased control consistency across different batches. Summary of the Invention

[0006] The purpose of this invention is to provide an adaptive control method and system for the opening of a casting valve, which can solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the technical solution adopted by the present invention is as follows: An adaptive control method for the opening of a pouring valve includes: acquiring a control command sequence, a valve position feedback sequence, and process data characterizing the actual pouring output of the pouring valve; Based on the response relationship between the control command sequence and the valve position feedback sequence, an equivalent implicit opening state is established, which includes valve hysteresis state, dead zone state, and unit opening flow gain. Based on the predicted casting output from the equivalent opening hidden state, and according to the cumulative material balance residual between the predicted casting output and the process data, the equivalent opening hidden state is corrected online. Based on the corrected equivalent opening hidden state, the target pouring trajectory is inverted into the target equivalent opening. The current hysteresis branch is determined based on the valve hysteresis state, the dead zone boundary is determined based on the dead zone state, and the target equivalent opening is converted into a mechanical opening control command.

[0008] Preferably, establishing the equivalent opening hidden state includes: dividing the valve motion branches according to the increasing direction, decreasing direction and direction switching time of the control command sequence; Calculate the command change, valve position arrival amount, initial response time of casting output, and stable time of casting output in each valve motion branch; The dead zone boundary is determined based on the change in the command and the arrival of the valve position; the hysteresis state is determined based on the first response time of the pouring output and the stable time of the pouring output; and the flow gain per unit opening is determined based on the arrival of the valve position and the corresponding pouring output increment. The equivalent opening hidden state is composed of the valve motion branch, dead zone boundary, hysteresis state, and unit opening flow gain.

[0009] Preferably, online correction of the equivalent opening hidden state includes: determining the delay action interval of each control command on the predicted casting output based on the hysteresis state, and aligning the process data with the corresponding control command according to the delay action interval; The sign retention and direction of change of the cumulative material balance residual are calculated over multiple consecutive sampling periods. The residuals that meet the preset continuity conditions are defined as continuous residuals, and the residuals that do not meet the preset continuity conditions are defined as fluctuating residuals. The unit aperture flow gain and the dead zone boundary are updated only using the continuous residual, while keeping the equivalent aperture hidden state parameter within the sampling period corresponding to the fluctuation residual unchanged.

[0010] Preferably, converting the target equivalent opening into the mechanical opening control command includes: determining the candidate opening change amount based on the deviation between the target equivalent opening and the current equivalent opening latent state; Based on the current valve movement direction, select the corresponding positive or negative mapping relationship from the valve hysteresis state, and combine it with the dead zone boundary to map the candidate opening change amount into the candidate mechanical opening increment; The allowable range of variation of the candidate mechanical opening increment is limited based on the parameter confidence of the equivalent opening hidden state; The candidate mechanical opening increments are superimposed with the current mechanical opening to generate a mechanical opening control command.

[0011] Preferably, determining the hysteresis state includes: taking the moment when each control command exceeds the dead zone boundary as the start of the response event, taking the moment when the process data first deviates from the pre-event reference interval as the start of the output response, and taking the moment when the slope of the change in the process data continuously enters the stable interval as the end of the output response. Based on the response event start point, output response start point, and output response end point, construct the response waiting segment and the effective action segment; When the effective action segments of adjacent response events overlap, the pouring output increment within the overlapping interval is split according to the time decay weight corresponding to the change in each control command, and the hysteresis state corresponding to each response event is updated using the split pouring output increment.

[0012] Preferably, updating the unit opening flow gain and the dead zone boundary using the persistent residual includes: determining whether the source of the residual belongs to flow gain offset or dead zone boundary offset based on the distribution difference of the persistent residual during the process of increasing and decreasing the opening. When the continuous residual under the same opening change direction accumulates proportionally with the opening change, the unit opening flow gain is updated according to the corresponding control command within the delay action interval. When the persistent residuals after the direction switch occur in the interval where the mechanical opening changes but the casting output does not respond, update the dead zone boundary of the corresponding opening change direction. Apply directional consistency constraints to the updated parameters to ensure that the dead zone boundaries in the direction of increasing opening and decreasing opening are recorded independently, while sharing the flow gain change trend within the continuous opening interval.

[0013] Preferably, each change in control command and its corresponding casting output response are determined as response events. Based on the degree of matching between the predicted casting output and the actual casting output in the most recent response events, hysteresis state confidence, dead zone boundary confidence and flow gain confidence are generated respectively. The upper limit of the opening increment is determined according to the minimum value among the three confidence levels, and the sign of the opening increment is determined according to the deviation direction between the target equivalent opening and the current equivalent opening. When any confidence level is lower than the preset confidence threshold, the candidate mechanical opening increment is decomposed into a continuous sequence of trial increments, and the next trial increment is generated after the casting output response corresponding to the previous trial increment ends.

[0014] Preferably, it further includes: establishing an event state vector for each response event, consisting of the valve motion branch, mechanical opening range, response waiting time, effective duration, and unit opening output increment; The state transition direction is determined based on the difference between the state vectors of adjacent events, and the hysteresis state and the unit opening flow gain are synchronously corrected only when the state transition directions of multiple consecutive response events are consistent. When the state transition directions are inconsistent, the corresponding response event is marked as a transient event and excluded from the parameter update sequence; When a subsequent response event has the same migration direction as the transient event, the transient event is reintroduced into the parameter update sequence.

[0015] Preferably, update weights for the fast correction channel and the slow correction channel are assigned based on the reliability of the parameters; The fast calibration channel adjusts the candidate mechanical opening increment based on the valve position arrival amount and response waiting time of a single response event, thereby correcting the current mechanical opening control command. The slow correction channel corrects the unit opening flow gain and the dead zone boundary based on the cumulative material balance residual corresponding to the continuous response events; When any confidence level is lower than a preset confidence threshold, only the fast correction channel is enabled and the slow correction channel is frozen; When all confidence levels reach the preset confidence threshold and the direction of change of the continuous residuals is consistent, the update weights of the two correction channels are configured according to the proportion of each confidence level.

[0016] An adaptive control system for the opening degree of a casting valve includes a data acquisition interface, a memory, and a processor; The data acquisition interface is used to acquire the control command sequence, valve position feedback sequence, and process data characterizing the actual pouring output of the pouring valve. The memory stores instructions that can be executed by the processor; The processor is used to: establish an equivalent openness hidden state, including valve hysteresis state, dead zone state, and unit opening flow gain, based on the response relationship between the control command sequence and the valve position feedback sequence. Based on the predicted casting output from the equivalent opening hidden state, and according to the cumulative material balance residual between the predicted casting output and the process data, the equivalent opening hidden state is corrected online. Based on the corrected equivalent opening hidden state, the target pouring trajectory is inverted into the target equivalent opening, and the target equivalent opening is converted into a mechanical opening control command based on the valve hysteresis state and the dead zone state.

[0017] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. Based on the control command sequence, valve position feedback sequence, and actual pouring output process data, an equivalent implicit opening state is established, including valve hysteresis state, dead zone state, and unit opening flow gain, so that the mechanical valve position and actual flow capacity are represented separately in the control calculation. After the predicted pouring output and actual process data form a cumulative material balance residual, the timing is aligned according to the delay interval of the control action, and continuous residuals and fluctuating residuals are distinguished based on the continuity of the residuals. Only the continuous residuals are used to correct the equivalent implicit opening state. The corrected equivalent opening state is used to invert the target pouring trajectory into the target equivalent opening, and then combined with the hysteresis branch and dead zone boundary corresponding to the current movement direction to generate the mechanical opening control command. The above processing ensures that historical control actions, current mechanical valve position, and actual pouring output participate in the calculation under the same time relationship, reducing the situation where control actions that have not yet completed the response are compensated again, and making the opening command adjust with the actual flow capacity of the valve, suppressing repeated increases and decreases in the final stage of pouring and preventing the cumulative pouring amount from exceeding the target value.

[0018] 2. Valve motion branches are divided according to opening increase, opening decrease, and direction switching. Dead zone boundaries, hysteresis states, and flow gain per unit opening are determined using control command changes, valve position arrival, initial output response, and output stabilization times, enabling separate recording of valve characteristics under different motion directions. For cases where adjacent control action response intervals overlap, the output increment corresponding to each control command is split based on time decay weights to prevent subsequent control actions from absorbing response data from previous actions. Transient events, continuous changes, and model uncertainties are distinguished using response event state vectors, state transition directions, and parameter reliability, thereby limiting mechanical opening increments or splitting candidate increments into continuous trial increments. A fast correction channel corrects the current mechanical opening command, while a slow correction channel updates flow gain and dead zone boundaries based on continuously accumulated material balance residuals. Both types of corrections configure update ranges according to reliability, reducing the interference of instantaneous fluctuations on long-term parameters and maintaining consistency between model updates and the actual valve response process. Attached Figure Description

[0019] Figure 1 A flowchart of an adaptive control method for the opening degree of a casting valve; Figure 2 This is a comparison diagram of equivalent opening state transition and mechanical opening; Figure 3 A comparison chart showing the predicted casting output versus the actual output; Figure 4 This is a graph showing the cumulative material balance residuals and local fluctuations. Figure 5 For target equivalent opening inversion and target trajectory tracking diagram; Figure 6 Flowchart for establishing and online correcting the equivalent openness hidden state; Figure 7 Output a contribution attribution graph for overlapping response events; Figure 8 A graph showing the residual persistence index and the decision threshold; Figure 9 Flowchart for parameter reliability assessment and fast / slow correction channel coordinated control; Figure 10 A graph showing parameter reliability and control increment limiting; Figure 11 This is a composite diagram of the fast and slow correction channels and the control increment. Detailed Implementation

[0020] In one embodiment, please refer to Figure 1 An adaptive control method for the opening of a casting valve includes acquiring a control command sequence, a valve position feedback sequence, and process data characterizing the actual casting output of the casting valve; establishing an equivalent opening latent state that includes valve hysteresis state, dead zone state, and unit opening flow gain based on the response relationship between the control command sequence and the valve position feedback sequence; predicting the casting output based on the equivalent opening latent state; correcting the equivalent opening latent state online based on the cumulative material balance residual between the predicted casting output and the process data; inverting the target casting trajectory into a target equivalent opening based on the corrected equivalent opening latent state; determining the current hysteresis branch based on the valve hysteresis state; determining the dead zone boundary based on the dead zone state; and converting the target equivalent opening into a mechanical opening control command.

[0021] In this embodiment, the control command sequence consists of the mechanical opening target value output by the controller at continuous sampling time, and the valve position feedback sequence consists of the actual mechanical position returned by the pouring valve actuator. The process data is selected from at least one of the instantaneous pouring flow rate, cumulative pouring volume, mass change of the pouring container, or state quantity that can equivalently characterize the material output per unit time. Before the data enters the control calculation, a sampling index is established according to a unified time reference, and the time when each control command is issued, the time when the valve position starts to change, the time when the valve position arrives, and the time when the corresponding change in the process data occurs are recorded, so that data from different sources can establish a corresponding relationship based on the same control event.

[0022] Specifically, the controller does not directly recognize the valve position feedback value as the actual flow opening of the valve. Instead, it uses the mechanical opening, hysteresis branch, dead zone boundary, response lag degree, and unit opening flow gain to represent the equivalent opening latent state. The equivalent opening latent state is used to describe the actual material throughput capacity that the current mechanical position can form under a given direction of movement and flow conditions. Among them, the mechanical opening reflects the position state of the valve actuator, the hysteresis branch reflects the difference between the opening increase path and the opening decrease path, the dead zone boundary reflects the opening interval where the control command changes but the process output has not yet undergone an attributable change, and the unit opening flow gain reflects the amount of change in casting output corresponding to each unit change in the equivalent opening.

[0023] In a preferred embodiment, the equivalent open-degree hidden state is calculated according to the following state transition relationship: ; in, This represents the equivalent aperture at the k-th sampling time. This represents the mechanical opening command at the k-th sampling time. This represents the hysteresis branch determined by the direction of the opening change. This represents the opening propagation coefficient of the corresponding hysteresis branch. Indicates the corresponding hysteresis branch at the th The dead zone boundary at each sampling time. Represents the hysteresis regression coefficient. Represents a sign function, when The value is 1 for positive, -1 for negative, and 0 for zero. This means taking the larger of the two values. When the change in mechanical opening does not exceed the dead zone boundary, the equivalent opening remains near the previous state. When the change in mechanical opening exceeds the dead zone boundary, only the excess part is included in the change in equivalent opening.

[0024] Please refer to Figure 2 , Figure 2 The horizontal axis represents the unified sampling sequence number, covering the continuous control range of valve opening from low to high, from high to low, and direction switching during batch playback; the vertical axis represents the opening percentage, with the curves corresponding to the mechanical opening command, the equivalent opening estimate, and the effective change state triggered by dead zone crossing, respectively. A legend is uniformly placed above the image to distinguish the meaning of each curve.

[0025] from Figure 2 As can be seen, after the mechanical opening command makes a step change, the equivalent opening does not immediately follow with the same amplitude. Instead, it remains approximately stable until the dead zone is crossed, and then enters a slow upward or downward process after crossing the boundary. Near the direction switch, the equivalent opening curve exhibits regression and lag phenomena relative to the mechanical opening curve, indicating that the branch coefficient, dead zone boundary, and hysteresis regression term in the state transition jointly determine the current flow capacity.

[0026] Figure 2 The smoothing delay and directional correlation difference of the equivalent opening indicate that the control system cannot directly replace the actual flow state with mechanical commands, but needs to continuously update the implicit state of the equivalent opening in each sampling period. The subsequent casting output prediction and target opening inversion are both based on this state as input.

[0027] In this embodiment, the state transition relationship stores parameters according to the opening increase branch and the opening decrease branch respectively. When the mechanical opening command changes from small to large, the parameters are... The decision is to increase the branch size and call the corresponding opening transfer coefficient, dead zone boundary, and hysteresis regression coefficient. When the mechanical opening command changes from large to small, the following will be implemented: The method is to reduce the branch and call another set of parameters. When the mechanical opening command remains unchanged, the previous hysteresis branch is maintained and only the hysteresis regression term is calculated to reflect the process that the valve position has been reached while the actual flow capacity is still approaching the stable state. After the direction switch occurs, the historical state of the previous branch is not cleared immediately. Instead, the equivalent opening before the switch is recorded as the initial state of the new branch, so that the opening state is kept continuous during the reversal.

[0028] The casting output prediction is based on the equivalent opening implicit state and the unit opening flow gain. The prediction object can be the material output increment of the current sampling period or the cumulative casting volume from the current time to the preset prediction endpoint. If the process data is the instantaneous casting flow rate, the equivalent opening is multiplied by the unit opening flow gain after the response delay and the reference output term is added. If the process data is the cumulative casting volume, the predicted output increment of each sampling period is calculated first and then the accumulation is performed. The reference output term is determined by the stable data of the valve being near the current equivalent opening and the control command not changing. When the direction changes or the opening interval shifts, the corresponding reference interval is reselected.

[0029] Preferably, the predicted casting output at the k-th sampling time is determined according to the following relationship: ; in, This represents the predicted casting output at the k-th sampling time. This represents the unit aperture flow gain at the k-th sampling time. This represents the response delay expressed in terms of the number of sampling periods. This represents the equivalent openness at a corresponding historical moment. This represents the baseline output term formed by the current static pressure state and steady flow state of the material. When the value is 3, it means that the output at the k-th sampling time mainly corresponds to the equivalent opening at the (k-3)-th sampling time. When the flow rate decreases, the predicted output corresponding to the same equivalent opening also decreases synchronously, thereby preventing the controller from misjudging the decrease in flow capacity as insufficient mechanical opening.

[0030] Please refer to Figure 3 , Figure 3 The horizontal axis represents the unified sampling sequence number, and the vertical axis represents the instantaneous casting output intensity, with units expressed according to the flow rate units in the system sampling data. The curves represent the actual casting output, the predicted output after delay alignment, and the trend line after the reference output term is included. The legend is uniformly placed above the image to facilitate direct comparison of the phase and amplitude differences of different curves.

[0031] from Figure 3 As can be seen, when the mechanical opening command changes, the actual casting output exhibits an observable lag relative to the equivalent opening; the peak and trough positions of the predicted curve are basically aligned with the actual curve after introducing the response delay. If this delay is not considered, the predicted output will show changes in rise and fall prematurely, causing the residual to be incorrectly interpreted as long-term parameter drift.

[0032] Figure 3 The degree of fit between the predicted curve and the actual curve indicates that the unit opening flow gain, delay step size, and baseline output term jointly determine the timing estimate of the casting output; this result provides a data basis for subsequent cumulative residual calculation, persistence determination, and deduction of in-transit response from the end-of-pipe remaining quantity.

[0033] In this embodiment, the cumulative material balance residual is not directly based on the prediction error of a single sampling point. Instead, it is calculated by integrating or accumulating the actual casting output and the predicted casting output within an observation interval consistent with the response delay. The starting point of the observation interval is set at the moment after the control command exceeds the dead zone boundary and passes through the estimated response delay, and the ending point is set at the moment before the output enters the stable interval or before the next control event begins to affect the output. If the responses of adjacent control events overlap, the actual output is first split according to the time contribution of each control event, and then corresponding residuals are formed respectively, so as to avoid the delayed output of a control action being incorrectly assigned to subsequent control actions.

[0034] The cumulative material balance residual is calculated according to the following relationship: ; in, This represents the cumulative material balance residual at the k-th sampling time. This represents the actual casting output at the j-th sampling time. This indicates the corresponding predicted casting output. This indicates the number of sampling periods contained in the current observation interval. This represents the duration of the j-th sampling period. This means that the output deviation of each period within the observation interval is converted into material quantity and then summed. When the actual output is higher than the predicted output in consecutive periods, Maintaining positive growth, when positive and negative deviations alternate and cancel each other out. Keep it within a small range.

[0035] Please refer to Figure 4 , Figure 4 The horizontal axis represents the unified sampling sequence number corresponding to the end of the residual window, and the vertical axis represents the cumulative material balance residual, with the unit consistent with the cumulative casting volume. The curve represents the result of accumulating the difference between the actual output and the predicted output within the current window according to the sampling interval, supplemented by a zero baseline and local fluctuation range, to observe the direction, amplitude, and duration of the residual.

[0036] from Figure 4 As can be seen, the residuals caused by single-point disturbances typically exhibit short-term spikes, followed by a rapid decline during window sliding; however, conditions with continuous gain decreases or dead zones widening cause the residuals to maintain the same direction over a longer period and gradually accumulate. This difference indicates that material balance residuals can reflect both the deviation from the actual pouring volume and whether the deviation has persistent characteristics that can be used for long-term parameter correction.

[0037] Figure 4 The cumulative direction and duration of the residual curve are used to distinguish between continuous residuals and fluctuating residuals. Only residuals that continuously maintain the same direction and whose amplitude meets the judgment requirements should enter the slow correction channel. Short fluctuations are only used for the current period to avoid writing occasional noise into long-term parameters.

[0038] Furthermore, the controller determines whether the residual is a continuous residual or a fluctuating residual based on the sign retention, direction consistency, and normalized accumulation degree of the cumulative material balance residual. The sign retention is obtained by statistically analyzing the proportion of sampling periods with the same sign of the residual increment within the observation interval. The direction consistency is obtained by comparing the directions of adjacent residual increments. The normalized accumulation degree is obtained by the ratio of the absolute value of the cumulative residual to the sum of the absolute errors of each sampling period. Only when the above quantities simultaneously meet the preset continuity condition will the corresponding residual be used to update the equivalent opening degree hidden state parameter. Residuals that do not meet the condition are retained as process fluctuation records and are not included in long-term parameter correction.

[0039] In this embodiment, online calibration is performed separately according to the parameter source. When the persistent residual accumulates approximately proportionally with the change in equivalent opening within the same motion branch, the residual is attributed to the flow gain offset per unit opening, and correction is made based on the ratio of the persistent residual to the corresponding equivalent opening integral. When persistent residuals occur during periods when the machine opening has changed but the process output has not yet left the reference range, the residuals are attributed to dead zone boundary offset and corrected based on the range of machine opening changes that do not produce attributable output. Correction is applied when the output first response time shifts continuously in the same direction relative to the historical response delay. Each parameter is updated only based on data that can be clearly attributed, thus avoiding the simultaneous unconstrained changes of multiple parameters driven by the same residual.

[0040] The target pouring trajectory is represented by the relationship between the cumulative pouring volume of the target and time or the instantaneous pouring output envelope of the target. At each sampling moment, the controller obtains the current target value, the current actual value, and the corrected equivalent opening hidden state. Based on the unit opening flow gain, it inversely calculates the target equivalent opening required to satisfy the target trajectory. During the inversion process, the current hysteresis branch and its dead zone boundary are retained. If the output change direction required by the target trajectory is consistent with the current motion branch, the equivalent opening increment is calculated along the current branch. If the output change direction required by the target trajectory is opposite to the current motion branch, the mechanical opening change required to cross the current dead zone boundary is calculated first, and then the effective equivalent opening increment after entering the opposite branch is calculated.

[0041] Preferably, the target equivalent opening is calculated according to the following relationship: ; in, This represents the target equivalent aperture at the k-th sampling time. This represents the target output corresponding to the target casting trajectory at the k-th sampling time. This indicates that the inversion denominator should remain a positive number and not zero. The fractional expression represents converting the difference between the target output and the predicted output into the required equivalent change in opening. Greater than When the equivalent opening of the target increases, Less than When the equivalent opening of the target decreases, When the difference is small, the same output difference corresponds to a larger equivalent opening change requirement.

[0042] Please refer to Figure 5 , Figure 5 The horizontal axis represents the unified sampling sequence number, and the vertical axis represents the opening percentage or equivalent opening dimension. The curves represent the current equivalent opening, the target equivalent opening obtained by inversion from the target casting trajectory, and the candidate control results after confidence limiting, respectively. The legend is uniformly placed at the top of the image to facilitate observation of the difference between the target trajectory and the actual state.

[0043] from Figure 5As can be seen, when the target casting output is higher than the current predicted output, the target equivalent opening curve rises; when the final stage approaches the target cumulative amount, the target equivalent opening gradually falls back, instead of using a fixed valve closing threshold that deviates from the model. The magnitude of the change in the target curve is affected by the unit opening flow gain and the stability term; when the gain is low, the same output error will result in a larger equivalent opening correction requirement.

[0044] Figure 5 The difference between the equivalent opening of the target and the current equivalent opening constitutes the main basis for generating mechanical commands. The system then combines hysteresis branches, dead zone boundaries and parameter reliability to limit the amplitude, converting the target trajectory into executable, traceable mechanical opening commands with a safety margin.

[0045] Specifically, when converting the target equivalent opening into a mechanical opening control command, the controller calculates the deviation between the target equivalent opening and the current equivalent opening, determines the candidate motion branch based on the deviation direction, and then reads the opening transfer coefficient, dead zone boundary, and hysteresis regression coefficient from the mapping relationship corresponding to the candidate motion branch. If the candidate motion branch is the same as the current hysteresis branch, the equivalent opening deviation is divided by the corresponding opening transfer coefficient and the dead zone amount that has not yet been crossed is added. If the candidate motion branch is different from the current hysteresis branch, the backlash crossing amount required for direction switching, the dead zone amount of the opposite branch, and the effective opening change amount are sequentially synthesized into the mechanical opening change amount, and a continuity constraint is applied to the calculation result.

[0046] In this embodiment, the allowable range of candidate mechanical opening increments is limited according to the parameter confidence of the equivalent opening latent state. The parameter confidence is generated for hysteresis state, dead zone boundary and unit opening flow gain respectively. The controller calculates the matching degree between the predicted output and the actual output, the consistency of the parameter update direction and the repeatability of the stable interval from the control events to which the responses have been completed most recently. The smaller value among the three confidence levels is taken as the current control confidence. When the current control confidence is high, the complete candidate mechanical opening increment is allowed to be output at once. When the current control confidence is low, the candidate mechanical opening increment is split into multiple trial increments with consistent directions. The next trial increment is determined after the output response of the previous trial increment is completed.

[0047] In this embodiment, the mechanical opening control command is formed by superimposing the current mechanical opening and the limited mechanical opening increment. After the command is issued, the controller establishes a new response event record and freezes the initial parameter snapshot corresponding to the event. Subsequent sampled data enters the event record according to the response delay. The event loop is only completed when the valve position reaches the process, the process output first response, and the process output stability conditions are met. The event loop data is used to update the equivalent opening hidden state and parameter reliability. If a new control requirement occurs before the event is completed, the new control command is recorded as an overlapping event and the response contribution split is initiated, instead of directly overwriting the historical attribution relationship of the previous event.

[0048] In one embodiment, please refer to Figure 6 Establishing the equivalent opening hidden state also includes dividing the valve motion branches according to the increasing direction, decreasing direction and direction switching time of the control command sequence, calculating the command change, valve position arrival amount, casting output first response time and casting output stabilization time in each valve motion branch, determining the dead zone boundary based on the command change and valve position arrival amount, determining the hysteresis state based on the casting output first response time and casting output stabilization time, determining the unit opening flow gain based on the valve position arrival amount and the corresponding casting output increment, and combining the valve motion branches, dead zone boundary, hysteresis state and unit opening flow gain to form the equivalent opening hidden state.

[0049] Specifically, the direction of increase, the direction of decrease, and the timing of the direction switch are determined by comparing adjacent control commands and adjacent valve position feedback. When the control command increases and the valve position feedback subsequently increases, an increase branch event is formed. When the control command decreases and the valve position feedback subsequently decreases, a decrease branch event is formed. When the control command direction has already switched but the valve position feedback is still changing along the original direction, this period is recorded as the direction switch transition zone and belongs to the previous branch. The process continues until the valve position feedback change direction is consistent with the new control command before entering the new branch. This process avoids the inertia or hysteresis of the actuator causing the direction switch timing to be determined in advance.

[0050] The change in command is taken as the algebraic difference between the mechanical opening command between the start and end of the control event. The valve position arrival is taken as the algebraic difference after the valve position feedback reaches stability within the same event. The initial response time of the casting output is determined by comparing the process data with the reference interval before the event. The stable time of the casting output is determined by the slope and fluctuation range of the output change within the continuous sampling period. If the process data does not leave the reference interval after the event occurs, the corresponding event is recorded as an event that has not crossed the dead zone, and the maximum mechanical opening change that has occurred within the event is added to the candidate set of the dead zone boundary. If the process data changes continuously, the cumulative mechanical opening change at the time of the initial response is used as the basis for the correction of the dead zone boundary.

[0051] In a preferred embodiment, the dead zone boundary is not directly replaced by the maximum value of a single unresponsive event. Instead, multiple candidate values ​​are stored within the same motion branch and adjacent mechanical opening intervals. Weights are assigned according to the completeness of the event corresponding to the candidate value, the stability of the process data, and the clarity of the response attribution. The current dead zone boundary is then updated using the weighted median position. Event completeness is used to exclude records where a new command occurs before the valve position is reached. Process data stability is used to exclude records where the baseline interval continuously drifts. Clarity of response attribution is used to exclude records where multiple events overlap and contribution cannot be split.

[0052] In one embodiment, determining the hysteresis state includes taking the moment when each control command exceeds the dead zone boundary as the response event start point, the moment when the process data first deviates from the pre-event baseline interval as the output response start point, and the moment when the slope of the process data continuously enters the stable interval as the output response end point. Based on the response event start point, the output response start point, and the output response end point, a response waiting segment and an effective action segment are constructed. When the effective action segments of adjacent response events overlap, the casting output increment within the overlapping interval is split according to the time decay weight corresponding to the change amount of each control command, and the hysteresis state corresponding to each response event is updated using the split casting output increment.

[0053] Furthermore, the response waiting segment extends from the start of the response event to the start of the output response. This time period is used to estimate the delay before the control command begins to act on the actual casting output after passing through the mechanical transmission and flow process. The effective action segment extends from the start of the output response to the end of the output response. This time period is used to estimate the continuous impact of the control action on the change in the casting output. When a new control event in the opposite direction occurs within the effective action segment, the previous event is not immediately truncated. Instead, the actual output increment within the overlapping time period is allocated according to the remaining response weight of the previous event at each sampling time and the initial response weight of the subsequent event.

[0054] Preferably, the output attribution value obtained by the i-th response event at the j-th sampling time is determined according to the following relationship: ; in, This represents the increment of the casting output obtained by the i-th response event at the j-th sampling time. This represents the set of response events that are still in the effective period at the j-th sampling time. This represents the time contribution weight of the i-th response event. This represents the increase in the actual pouring volume at the j-th sampling time relative to the previous sampling time. This represents the effective mechanical opening change after the i-th response event exceeds the dead zone boundary. This represents the response attenuation coefficient of the corresponding hysteresis branch. Indicates the current sampling time. This indicates the starting point of the output response for the i-th response event.

[0055] Please refer to Figure 7 , Figure 7 The horizontal axis represents the sampling sequence number or event range where responses overlap, and the vertical axis represents the pouring output increment attributed to each response event within the current sampling period. The stacked areas correspond to the contribution share of different response events, and the legend is uniformly placed at the top of the image to show the output attribution obtained by each event at the same sampling time.

[0056] from Figure 7 As can be seen, when the response intervals of multiple mechanical opening commands overlap, the more recently occurring events with higher confidence weights receive a higher output share, while the contribution of earlier events gradually decreases as the time interval increases. The sum of the heights of each stacked region at the same sampling point remains consistent with the mass increment of that sampling point, indicating that the contribution splitting process satisfies the material conservation requirement.

[0057] Figure 7 The change in the contribution area indicates that event closure and parameter updates cannot simply attribute all the outputs of the overlapping interval to the last instruction. Instead, they should be separated according to the event confidence and the first response time of the output to avoid the long-term gain, dead zone and delay estimates being contaminated by overlapping events.

[0058] In this embodiment, when there is only one response event in the effective range at the same sampling time, the response event obtains the full actual casting output increment. When there are two or more overlapping response events, each event receives the output increment according to the weight of its time contribution. The weight of an earlier event decreases as time increases, while the weight of a later event increases as its effective opening changes. If the effective mechanical opening change of an event is 0, its time contribution weight is 0 and it does not participate in the allocation. The split output increments are accumulated in each event record to calculate the unit opening flow gain and hysteresis duration.

[0059] In one embodiment, online correction of the equivalent opening latent state further includes determining the delay action interval of each control command on the predicted casting output based on the hysteresis state, aligning the process data with the corresponding control commands according to the delay action interval, calculating the sign retention and change direction consistency of the cumulative material balance residuals in multiple consecutive sampling periods, determining the residuals that meet the preset continuity conditions as continuous residuals, determining the residuals that do not meet the preset continuity conditions as fluctuating residuals, updating the unit opening flow gain and dead zone boundary only using continuous residuals, and keeping the equivalent opening latent state parameters unchanged in the sampling period corresponding to the fluctuating residuals.

[0060] Specifically, the delay action interval is jointly determined by the response waiting period and the effective action period. Changes in process data within the response waiting period are not directly attributed to the current control command. Changes in process data within the effective action period are attributed to the current control command according to the event contribution relationship. The continuity conditions include at least the residual sign remaining consistent across multiple sampling periods, the cumulative residual value changing in the same direction, and no irrelevant abrupt changes occurring in the reference interval before and after the event. If only a single-point deviation exists or the deviation directions of adjacent periods alternate, the corresponding deviation will be included in the fluctuation residual record without changing the unit opening flow gain, dead zone boundary, and response delay.

[0061] In a preferred embodiment, the persistence determination quantity is calculated according to the following relationship: ; in, This represents the residual persistence determination value at the k-th sampling time. This indicates the prevention of positive numbers with a denominator of 0. The first product factor represents the consistency of the error sign, and the second product factor represents the degree to which accumulated errors are not canceled out by errors in the opposite direction. When all errors have the same sign and accumulate continuously, both product factors are close to 1. When errors frequently reverse direction, at least one product factor decreases, and the controller will... Residuals that reach the persistence threshold and whose direction is consistent across multiple consecutive observation intervals are defined as persistent residuals.

[0062] Please refer to Figure 8 , Figure 8 The horizontal axis represents the uniform sampling sequence number at the end of the sliding decision window, and the vertical axis represents the residual persistence index. The solid line represents the persistence level calculated based on the consistency of the residual direction and the cumulative residual amplitude within the window, and the threshold line represents the decision boundary required to enter the slow correction channel. The legend is uniformly placed above the image.

[0063] from Figure 8 As can be seen, although random fluctuations or single-point anomalies may cause local residual spikes, their persistence index is usually below the threshold due to the alternation of positive and negative directions. When multiple consecutive response events show deviations in the same direction, the persistence index will rise significantly and exceed the threshold. This phenomenon indicates that persistence determination not only focuses on the magnitude of the residuals, but also on directional stability and effective accumulation within the window.

[0064] Figure 8 Intervals exceeding the threshold can be identified as persistent residuals, which can be used as the basis for updating unit opening flow gain, dead zone boundary, and response delay; fluctuation residuals that do not exceed the threshold only participate in the current cycle processing and do not enter the long-term parameter sequence.

[0065] In a preferred embodiment, updating the unit opening flow gain and dead zone boundary using persistent residuals includes determining whether the source of the residual is flow gain offset or dead zone boundary offset based on the distribution difference of the persistent residuals during the process of increasing and decreasing the opening. When the persistent residuals under the same opening change direction accumulate proportionally with the amount of opening change, the unit opening flow gain is updated according to the corresponding control command within the delay action interval. When the persistent residuals after the direction switch are concentrated in the interval where the mechanical opening changes but the casting output does not respond, the dead zone boundary of the corresponding opening change direction is updated. The updated parameters are subject to directional consistency constraints so that the dead zone boundary in the direction of increasing opening and the dead zone boundary in the direction of decreasing opening are recorded independently and share the flow gain change trend within the continuous opening interval.

[0066] In this embodiment, the residual source is determined by comparing the correlation between the persistent residual and the effective mechanical opening change. If the persistent residuals of multiple events are in the same direction and have similar values ​​after being divided by their respective effective mechanical opening changes, the unit opening flow gain is used as the main update object. If the persistent residual mainly appears in the unresponsive interval after the direction switch and disappears after crossing a certain mechanical opening range, the dead zone boundary is used as the main update object. If both types of features exist at the same time, the dead zone boundary is updated first according to the unresponsive interval, and then the unit opening flow gain is updated with the effective mechanical opening change after deducting the dead zone, so as to avoid the dead zone estimation bias being absorbed into the flow gain.

[0067] Directional consistency constraints include storing the dead zone boundaries for increasing and decreasing opening directions separately, and restricting the dead zone boundary after a single update from crossing confirmed valid response points in adjacent opening intervals. The trend of flow gain change remains continuous between adjacent equivalent opening intervals. If a complete response event is missing in a certain opening interval, the trend extrapolation value of the adjacent interval is used as a temporary value instead of a confirmation parameter, until the interval obtains complete event data before replacement. During the parameter update process, the parameters before the update, the parameters after the update, and the event index that triggered the update are saved simultaneously for subsequent state transition judgment.

[0068] In one embodiment, converting the target equivalent opening degree into a mechanical opening degree control command further includes determining the candidate opening degree change amount based on the deviation between the target equivalent opening degree and the current equivalent opening degree, selecting the corresponding forward mapping relationship or reverse mapping relationship from the valve hysteresis state according to the current valve movement direction, mapping the candidate opening degree change amount to the candidate mechanical opening degree increment in combination with the dead zone boundary, limiting the allowable variation range of the candidate mechanical opening degree increment according to the parameter reliability of the equivalent opening degree hidden state, and superimposing the limited candidate mechanical opening degree increment with the current mechanical opening degree to generate a mechanical opening degree control command.

[0069] Specifically, the forward mapping relationship corresponds to the equivalent opening conversion path when the mechanical opening increases, and the reverse mapping relationship corresponds to the equivalent opening conversion path when the mechanical opening decreases. When the candidate opening change is consistent with the current direction of motion, the current path parameters are directly called. When the candidate opening change is opposite to the current direction of motion, the mechanical opening amount required to return from the current hysteresis branch to the direction switching boundary is first determined, and then the mechanical opening amount required to cross the dead zone in the opposite direction is determined. After that, the remaining equivalent opening change is converted into the mechanical opening amount according to the opening transmission coefficient of the opposite branch. The three parts are algebraically superimposed to form the candidate mechanical opening increment.

[0070] In one embodiment, please refer to Figure 9 Each change in control command and its corresponding casting output response is defined as a response event. Based on the matching degree between the predicted casting output and the actual casting output in the most recent response events, hysteresis state confidence, dead zone boundary confidence, and flow gain confidence are generated respectively. The upper limit of the opening increment is determined according to the minimum value of the three confidence levels, and the sign of the opening increment is determined according to the deviation direction between the target equivalent opening and the current equivalent opening. When any confidence level is lower than the preset confidence threshold, the candidate mechanical opening increment is decomposed into a continuous sequence of trial increments, and the next trial increment is generated after the casting output response corresponding to the previous trial increment ends.

[0071] In this embodiment, the reliability of the hysteresis state is determined by the dispersion between the actual response waiting time and the predicted response waiting time of the most recent complete response events. The reliability of the dead zone boundary is determined by the repetition of the non-response interval after the direction switch and the concentration of the dead zone candidate values. The reliability of the flow gain is determined by the deviation between the actual output increment and the predicted output increment after splitting. All reliability values ​​are normalized to between 0 and 1. If the event data is incomplete, the event overlap cannot be fully split, or the process data baseline interval is unstable, the corresponding reliability is reduced. The controller uses the minimum value among the three to limit the mechanical opening increment to prevent a large control quantity from being output according to the higher reliability of other parameters when a certain key parameter lacks support.

[0072] In a preferred embodiment, the current control confidence level and the mechanical opening increment limit relationship are determined according to the following formula: ; in, This represents the current control confidence level at the k-th sampling time. Indicates the reliability of the hysteresis state. Indicates the confidence level of the dead zone boundary. Indicates the confidence level of the flow gain. This indicates an undefined increment in the candidate mechanical opening. This represents the incremental benchmark determined based on the current mechanical opening range and the remaining change in the target trajectory. This means restricting the input value to within a given upper and lower boundary. When the value is 0.4, the allowable increment range is ±0.4 times the increment base. When the value is 0.9, the allowable range expands accordingly.

[0073] Please refer to Figure 10 , Figure 10 The horizontal axis represents the unified sampling sequence number or parameter snapshot update sequence number, while the vertical axis simultaneously expresses the confidence level and the control increment limiting ratio. The bars or steps correspond to the hysteresis state confidence, dead zone boundary confidence, flow gain confidence, and the minimum confidence after the combination of the three, respectively. The control increment curve represents the executable range of the candidate instruction after limiting.

[0074] from Figure 10 As can be seen, whenever any confidence component decreases, the overall confidence level decreases accordingly, and the candidate mechanical opening increment is compressed into a narrower safety range. Within the low confidence range, the system tends to break down larger candidate increments into consecutive trial increments, waiting for the previous response to close before issuing the next one, in order to reduce the impact of parameter uncertainty on the casting output.

[0075] Figure 10 The synchronous relationship between the decline in confidence level and the tightening of the limit indicates that the control command is not only determined by the target error, but also constrained by the confidence level of the model parameters. When the data quality is restored, the response event is closed and the verification is completed, the limit range is gradually released as the confidence level recovers.

[0076] In a preferred embodiment, the trial increment sequence maintains the same direction as the candidate mechanical opening increment, and the single trial increment is allocated according to the current control confidence. After the previous trial increment is issued, the controller does not generate additional increments in the same direction during the corresponding response waiting period. After entering the effective action period, the current equivalent opening is corrected according to the actual output change. After the output response ends or the accumulated material balance residual reaches the stability judgment condition, the remaining target equivalent opening deviation is recalculated, and then it is decided whether to output the next trial increment. If the actual output change direction is inconsistent with the predicted direction, the original trial increment sequence is terminated and the hysteresis branch and dead zone boundary are re-determined.

[0077] In one embodiment, an event state vector is established for each response event, consisting of valve motion branch, mechanical opening range, response waiting time, effective duration, and unit opening output increment. The state transition direction is determined based on the difference between adjacent event state vectors. Only when the state transition directions of multiple consecutive response events are consistent, the hysteresis state and unit opening flow gain are synchronously corrected using the cumulative material balance residual of the corresponding response event. When the state transition directions are inconsistent, the corresponding response event is marked as a transient event and excluded from the parameter update sequence. When subsequent response events have the same transition direction as transient events, the transient events are reintroduced into the parameter update sequence.

[0078] Specifically, the mechanical opening interval in the event state vector is divided according to the continuity of the equivalent opening. The response waiting time and effective action time are converted according to the unified sampling time. The unit opening output increment is calculated by dividing the cumulative casting output obtained by the event by the effective mechanical opening change after exceeding the dead zone boundary. The difference between adjacent event state vectors is compared with the change scale of the historical stable interval. If at least two components of the response waiting time, effective action time and unit opening output increment change continuously in the same direction, the state transition direction is determined. If only a single component changes in isolation or the adjacent events change in opposite directions, the corresponding change is treated as a transient event.

[0079] Transient event exclusion only restricts its immediate participation in long-term parameter updates. The control commands, valve position feedback, output response, and cumulative material balance residuals of the transient event are still fully preserved. When subsequent events exhibit state migration in the same direction within the same motion branch and adjacent mechanical opening intervals, the previous transient event and subsequent events are included in the migration sequence and the parameter update amount is recalculated. When the subsequent event recovers to the state before migration, the transient event exclusion status is maintained. This process ensures that a single flow fluctuation does not directly change the long-term model, while retaining early change data to confirm slow migration.

[0080] In one embodiment, update weights for the fast and slow correction channels are assigned based on parameter confidence levels. The fast correction channel corrects the current mechanical opening control command based on the valve position arrival amount, response waiting time, and candidate mechanical opening increment of a single response event. The slow correction channel corrects the unit opening flow gain and dead zone boundary based on the cumulative material balance residuals corresponding to continuous response events. When any confidence level is lower than a preset confidence threshold, only the fast correction channel is enabled and the slow correction channel is frozen. When all confidence levels reach the preset confidence thresholds and the direction of change of continuous residuals is consistent, update weights for the two correction channels are configured according to the proportion of each confidence level.

[0081] In this embodiment, the fast correction channel does not rewrite the long-term flow gain and dead zone boundary. Instead, it temporarily corrects the mechanical opening increment for the next control cycle based on the difference between the actual valve position arrival amount and the candidate mechanical opening increment in the current response event. The temporary correction is released after the current response event ends. The slow correction channel updates the long-term parameters only when multiple consecutive events form a state migration in the same direction and the cumulative material balance residual meets the persistence condition. If any confidence level is lower than the threshold, the update amount of the slow correction channel is kept at 0, and event data that can restore the parameter confidence level is collected by probing the incremental sequence.

[0082] In a preferred embodiment, the synthesis relationship between the fast correction channel and the slow correction channel is determined according to the following formula: ; in, This represents the fast correction weight at the k-th sampling time. Indicates slow correction weights. This represents the temporary mechanical opening correction amount obtained based on the current response event. This indicates the correction amount calculated based on long-term parameters obtained from continuous response events. This represents the mechanical opening correction amount after the two correction channels are combined. When the current control confidence decreases, the fast correction weight increases; when the current control confidence and the residual persistence determination amount increase simultaneously, the slow correction weight increases.

[0083] Please refer to Figure 11 , Figure 11 The horizontal axis represents the unified sampling sequence number, and the vertical axis represents the control correction amount after mechanical opening increment or channel weight conversion. The curves correspond to the fast correction increment, slow correction increment, and the final increment after their combination, respectively. The legend is uniformly placed at the top of the image to observe the contribution relationship of different channels within the same control cycle.

[0084] from Figure 11 As can be seen, when the reliability is low or the persistence is insufficient, the fast correction channel dominates, only correcting the current mechanical opening command and not writing it into the long-term flow gain sequence. When the persistent residual is significant and the reliability recovers, the weight of the slow correction channel gradually increases, and long-term parameters begin to participate in the update. If the slow correction is frozen, the final increment will approach the fast correction curve, reflecting the conservative control strategy during anomaly handling.

[0085] Figure 11The final increment lies between the fast and slow corrections, indicating that the control system determines the division of labor between the current command and the long-term parameters through reliability and persistence: short-term deviations are preferentially absorbed by fast corrections, and only persistent deviations verified by continuous events enter slow corrections and affect subsequent batches.

[0086] Furthermore, the fast correction channel and the slow correction channel use the same equivalent opening latent state as the calculation basis. The temporary mechanical opening correction output by the fast correction channel is superimposed on the inverse mapping result when generating control commands, but it does not enter the historical sequence of unit opening flow gain. The updated flow gain and dead zone boundary of the slow correction channel take effect at the beginning of the next complete control event, and are not replaced in the middle of the effective action segment of the current event, so as to keep the same response event from prediction, output attribution to residual calculation using consistent parameters, and avoid the output contribution from not being closed due to changing parameters in the middle of event processing.

[0087] In a preferred embodiment, when the target pouring trajectory is close to the end point, the controller continues to use the equivalent opening hidden state and the cumulative material balance residual for control. Instead of setting a fixed valve closing threshold to deviate from the model, the difference between the target cumulative pouring amount and the current cumulative pouring amount is subtracted from the predicted output of the control event that has been issued and is still in the response waiting segment or effective action segment, to obtain the remaining pouring amount that can be handled by the new mechanical opening command. The remaining pouring amount is then converted into the target equivalent opening. If the remaining pouring amount is less than the predicted output of the existing in-transit control event, the target equivalent opening is generated in the decreasing direction and the mechanical opening is inverted according to the current decreasing branch dead zone boundary.

[0088] In this embodiment, if the process data uses the cumulative pouring volume, the controller obtains the actual pouring output increment from the adjacent differences of the cumulative value and converts it into a unified material output volume according to the sampling period duration. If the process data uses the instantaneous flow rate, the controller accumulates the product of the instantaneous flow rate and the sampling period duration to form the cumulative material volume. If both types of process data are available, the controller calculates the cumulative material balance residuals separately and compares the directional consistency of the two within the complete response event. When the directions are consistent, the residuals are weighted and synthesized according to their respective stability levels. When the directions are inconsistent, the flow gain reliability is reduced and the slow correction channel is frozen until the subsequent complete event regains consistency.

[0089] In one embodiment, a casting valve opening adaptive control system includes a data acquisition interface, a memory, and a processor. The data acquisition interface is used to acquire the control command sequence, valve position feedback sequence, and process data characterizing the actual casting output of the casting valve. The memory stores instructions executable by the processor. The processor is used to establish an equivalent opening latent state, including valve hysteresis state, dead zone state, and unit opening flow gain, based on the response relationship between the control command sequence and the valve position feedback sequence. Based on the equivalent opening latent state, the processor predicts the casting output and performs online correction of the equivalent opening latent state according to the cumulative material balance residual between the predicted casting output and the process data. Based on the corrected equivalent opening latent state, the target casting trajectory is inverted into the target equivalent opening. Based on the valve hysteresis state and dead zone state, the target equivalent opening is converted into a mechanical opening control command.

[0090] In this embodiment, the data acquisition interface receives control commands, valve position feedback, and process data according to a unified timestamp, and writes missing samples, duplicate samples, and data with abnormal arrival order into the status flag. The processor only performs response event attribution on data whose time relationship can be recovered. The memory stores data in the original data area, event data area, status parameter area, and control result area. The original data area stores the input sequence before model correction. The event data area stores the response event start point, output response start point, output response end point, and output attribution quantity. The status parameter area stores the model parameters of each hysteresis branch and equivalent opening interval. The control result area stores the target equivalent opening, candidate mechanical opening increment, confidence limit result, and final mechanical opening control command.

[0091] Specifically, in each control cycle, the processor calls the state parameters corresponding to the current hysteresis branch and the equivalent opening interval, updates the equivalent opening according to the control instruction sequence and valve position feedback sequence, obtains the corresponding state from the historical equivalent opening sequence according to the response delay and calculates the predicted casting output, calculates the cumulative material balance residual according to the process data of the completed timing alignment, determines whether to update the flow gain, dead zone boundary and response delay according to the residual persistence judgment result, and then inverts the target equivalent opening according to the target casting trajectory and generates mechanical opening control instructions. Each processing stage uses the same control cycle index and the same parameter snapshot, and only after the control cycle ends will the confirmed parameter update result be submitted.

[0092] In a specific application process, after the pouring begins, the controller reads the target pouring trajectory and obtains the current mechanical opening and equivalent opening. If the target output is higher than the predicted output, the target equivalent opening and candidate mechanical opening increment are calculated in the increasing branch. After the control command is issued, a response event is established. During the response waiting period, the current event is observed without continuously adding the opening in the same direction due to the instantaneous output not changing. After the process output leaves the reference range, it enters the effective action period and accumulates the event-assigned output. After the output change slope enters the stable range, the event is closed. The implicit state of the equivalent opening is corrected according to the accumulated material balance residual. Then, the mechanical opening command for the next control cycle is calculated based on the corrected state.

[0093] In the same specific application process, when the target output begins to decrease or the cumulative pouring amount approaches the target value, the controller switches the candidate motion branch to the reduction branch, calculates the mechanical opening change required to cross the hysteresis of the original hysteresis branch and the dead zone of the reduction branch, and includes the predicted output of the historical increase opening event that is still in the effective action section into the subsequent cumulative material amount. After the reduction opening command is issued, the impact of the historical increase event and the current reduction event on the process output is distinguished according to the event contribution relationship. If the current reduction event is still in the response waiting section, the existing pouring output error is not attributed to the reduction opening being invalid. Instead, the hysteresis state of the reduction branch is updated after the corresponding output response arrives.

[0094] In this embodiment, when the actual output of multiple consecutive increasing branch events is lower than the predicted output and the residual persistence determination quantity meets the condition, the controller determines that the unit opening flow gain migrates along the decreasing direction and updates the flow gain of the corresponding opening interval in the slow correction channel. When only a single event has an output deviation and subsequent events recover, the event is marked as a transient event and the original flow gain is maintained. When the interval where the mechanical opening changes continuously but the process output does not respond after the direction switch expands, the controller updates the dead zone boundary of the corresponding decreasing branch without directly modifying the dead zone boundary of the increasing branch, ensuring that the asymmetric characteristics of the two motion directions are recorded separately.

Claims

1. A method for adaptive control of the opening degree of a casting valve, characterized in that, include: Acquire the control command sequence, valve position feedback sequence, and process data characterizing the actual pouring output of the pouring valve; Based on the response relationship between the control command sequence and the valve position feedback sequence, an equivalent implicit opening state is established, which includes valve hysteresis state, dead zone state, and unit opening flow gain. Based on the predicted casting output from the equivalent opening hidden state, and according to the cumulative material balance residual between the predicted casting output and the process data, the equivalent opening hidden state is corrected online. Based on the corrected equivalent opening hidden state, the target pouring trajectory is inverted into the target equivalent opening. The current hysteresis branch is determined based on the valve hysteresis state, the dead zone boundary is determined based on the dead zone state, and the target equivalent opening is converted into a mechanical opening control command.

2. The adaptive control method for the opening degree of a casting valve according to claim 1, characterized in that, Establishing the equivalent opening hidden state includes dividing the valve motion branches according to the increasing direction, decreasing direction, and direction switching time of the control command sequence; Calculate the command change, valve position arrival amount, initial response time of casting output, and stable time of casting output in each valve motion branch; The dead zone boundary is determined based on the change in the command and the arrival of the valve position; the hysteresis state is determined based on the first response time of the pouring output and the stable time of the pouring output; and the flow gain per unit opening is determined based on the arrival of the valve position and the corresponding pouring output increment. The equivalent opening hidden state is composed of the valve motion branch, dead zone boundary, hysteresis state, and unit opening flow gain.

3. The adaptive control method for the opening degree of a casting valve according to claim 1, characterized in that, Online correction of the equivalent opening hidden state includes: determining the delay action interval of each control command on the predicted casting output based on the hysteresis state, and aligning the process data with the corresponding control command according to the delay action interval; The sign retention and direction of change of the cumulative material balance residual are calculated over multiple consecutive sampling periods. The residuals that meet the preset continuity conditions are defined as continuous residuals, and the residuals that do not meet the preset continuity conditions are defined as fluctuating residuals. The unit aperture flow gain and the dead zone boundary are updated only using the continuous residual, while keeping the equivalent aperture hidden state parameter within the sampling period corresponding to the fluctuation residual unchanged.

4. The adaptive control method for the opening degree of a casting valve according to claim 1, characterized in that, Converting the target equivalent opening into the mechanical opening control command includes: determining the candidate opening change amount based on the deviation between the target equivalent opening and the current equivalent opening latent state; Based on the current valve movement direction, select the corresponding positive or negative mapping relationship from the valve hysteresis state, and combine it with the dead zone boundary to map the candidate opening change amount into the candidate mechanical opening increment; The allowable range of variation of the candidate mechanical opening increment is limited based on the parameter confidence of the equivalent opening hidden state; The candidate mechanical opening increments are superimposed with the current mechanical opening to generate a mechanical opening control command.

5. The adaptive control method for the opening degree of a casting valve according to claim 1, characterized in that, Determining the hysteresis state includes: taking the moment when each control command exceeds the dead zone boundary as the start of the response event, taking the moment when the process data first deviates from the pre-event reference interval as the start of the output response, and taking the moment when the slope of the change in the process data continuously enters the stable interval as the end of the output response. Based on the response event start point, output response start point, and output response end point, construct the response waiting segment and the effective action segment; When the effective action segments of adjacent response events overlap, the pouring output increment within the overlapping interval is split according to the time decay weight corresponding to the change in each control command, and the hysteresis state corresponding to each response event is updated using the split pouring output increment.

6. The adaptive control method for the opening degree of a casting valve according to claim 3, characterized in that, Updating the unit opening flow gain and the dead zone boundary using the persistent residual includes: determining whether the source of the residual belongs to flow gain offset or dead zone boundary offset based on the distribution difference of the persistent residual during the process of increasing and decreasing the opening. When the continuous residual under the same opening change direction accumulates proportionally with the opening change, the unit opening flow gain is updated according to the corresponding control command within the delay action interval. When the persistent residuals after the direction switch occur in the interval where the mechanical opening changes but the casting output does not respond, update the dead zone boundary of the corresponding opening change direction. Apply directional consistency constraints to the updated parameters to ensure that the dead zone boundaries in the direction of increasing opening and decreasing opening are recorded independently, while sharing the flow gain change trend within the continuous opening interval.

7. The adaptive control method for the opening degree of a casting valve according to claim 4, characterized in that, Each change in control command and its corresponding casting output response is determined as a response event. Based on the degree of matching between the predicted casting output and the actual casting output in the most recent response events, hysteresis state confidence, dead zone boundary confidence and flow gain confidence are generated respectively. The upper limit of the opening increment is determined according to the minimum value among the three confidence levels, and the sign of the opening increment is determined according to the deviation direction between the target equivalent opening and the current equivalent opening. When any confidence level is lower than the preset confidence threshold, the candidate mechanical opening increment is decomposed into a continuous sequence of trial increments, and the next trial increment is generated after the casting output response corresponding to the previous trial increment ends.

8. The adaptive control method for the opening degree of a casting valve according to claim 1, characterized in that, Also includes: For each response event, an event state vector is established, consisting of the valve motion branch, mechanical opening range, response waiting time, effective duration, and unit opening output increment. The state transition direction is determined based on the difference between the state vectors of adjacent events, and the hysteresis state and the unit opening flow gain are synchronously corrected only when the state transition directions of multiple consecutive response events are consistent. When the state transition directions are inconsistent, the corresponding response event is marked as a transient event and excluded from the parameter update sequence; When a subsequent response event has the same migration direction as the transient event, the transient event is reintroduced into the parameter update sequence.

9. The adaptive control method for the opening degree of a casting valve according to claim 7, characterized in that, The update weights for the fast and slow correction channels are assigned based on the reliability of the parameters. The fast calibration channel adjusts the candidate mechanical opening increment based on the valve position arrival amount and response waiting time of a single response event, thereby correcting the current mechanical opening control command. The slow correction channel corrects the unit opening flow gain and the dead zone boundary based on the cumulative material balance residual corresponding to the continuous response events; When any confidence level is lower than a preset confidence threshold, only the fast correction channel is enabled and the slow correction channel is frozen; When all confidence levels reach the preset confidence threshold and the direction of change of the continuous residuals is consistent, the update weights of the two correction channels are configured according to the proportion of each confidence level.

10. An adaptive control system for the opening degree of a casting valve, characterized in that, Includes data acquisition interface, memory, and processor; The data acquisition interface is used to acquire the control command sequence, valve position feedback sequence, and process data characterizing the actual pouring output of the pouring valve. The memory stores instructions that can be executed by the processor; The processor is used to: establish an equivalent openness hidden state, including valve hysteresis state, dead zone state, and unit opening flow gain, based on the response relationship between the control command sequence and the valve position feedback sequence. Based on the predicted casting output from the equivalent opening hidden state, and according to the cumulative material balance residual between the predicted casting output and the process data, the equivalent opening hidden state is corrected online. Based on the corrected equivalent opening hidden state, the target pouring trajectory is inverted into the target equivalent opening, and the target equivalent opening is converted into a mechanical opening control command based on the valve hysteresis state and the dead zone state.