Self-adaptive control process for environment-friendly film covering of packaging color box

Abstract

The present application relates to the technical field of packaging printing, and discloses a self-adaptive regulation and control process for packaging color box environment-friendly film covering, comprising: realizing differential pressure equalization of thick and thin areas of ink layer, eliminating bubbles and virtual adhesion through full-width laser scanning and cooperation of multi-zone pressure array; realizing tension fluctuation blocking and longitudinal wrinkle suppression through passive attenuation and active micro-tension feed-forward compensation coupling of multi-stage floating roller; realizing disturbance prediction and control through cooperation of LSTM prediction and deep reinforcement learning pre-execution; reducing quality fluctuation and energy consumption; realizing strategy evolution and cloud atlas optimization through linkage of execution feedback and DRL self-learning update; and improving film covering stability.

Description

Technical Field

[0001] This invention relates to the field of packaging and printing technology, specifically to an adaptive control process for environmentally friendly lamination of packaging color boxes. Background Technology

[0002] Packaging boxes are a common packaging form in industries such as consumer electronics, fast-moving consumer goods, and food. The coating layer on the surface of the box directly determines the product's appearance, moisture barrier, and wear resistance. With increasingly stringent regulations such as the "Guidelines for Green Procurement of Packaging Materials" and the "VOC Limit Management Measures," traditional coating processes using PVDC, PVC, or solvent-based adhesives are no longer sufficient to meet the environmental requirements of low VOC, biodegradability, and recyclability. At the same time, differences in box size, substrate thickness, and surface condition lead to frequent quality problems such as coating thickness, bubbles, and delamination, directly affecting the visual effect and protective performance of the packaging. To achieve high-quality, uniform, bubble-free, firmly bonded coating that meets green environmental standards, and to enable rapid switching between different box sizes while reducing energy consumption and waste rate, it is urgent to develop an adaptive closed-loop control environmentally friendly coating process. This process should enable real-time sensing and intelligent adjustment of key parameters such as temperature, pressure, tension, and coating thickness to ensure precise control of coating density and performance.

[0003] Existing adaptive control processes for environmentally friendly lamination of packaging boxes cannot adjust the pressure in different zones based on local thickness differences in the paper. Insufficient pressure in areas with thick ink layers results in air bubbles and poor adhesion, while excessive pressure in areas with thin ink layers causes the adhesive layer to overflow. Traditional single-point constant tension control is unable to absorb fluctuations in paper feed speed, and the film is prone to longitudinal stretching and deformation. Furthermore, pressure control and tension control are disconnected, failing to consider the coupling effect of zone pressure changes on roll gaps and film speed. Moreover, traditional control is mostly a passive response, unable to predict disturbances such as paper joints and batch switching, and lacks self-evolving models with execution feedback. Its practicality has certain limitations. Summary of the Invention

[0004] This invention provides an adaptive control process for environmentally friendly coating of packaging boxes. By constructing a nonlinear mapping model between thickness deviation and pressure, and introducing a tension disturbance factor to achieve pressure-tension coupling compensation, and on this basis, combining quality feedback and parameter adaptive update mechanism, stable control and quality optimization of the coating process are achieved.

[0005] This invention provides the following technical solution: an adaptive control process for environmentally friendly film coating of packaging boxes, comprising:

[0006] Before the paper enters the hot-pressing lamination zone, the paper thickness distribution data is obtained by full-width scanning, the average thickness and the maximum thickness difference rate of the full width are calculated, and the degree of thickness variation is judged accordingly.

[0007] Based on the pressure zoning structure of the hot press roller, the thickness distribution data is divided into multiple zones. The deviation of each zone from the average thickness of the whole width is calculated, and each zone is determined to be a thick zone, a thin zone, or a normal zone.

[0008] Based on the thickness deviation of each zone, differential adjustment is made on the basis of the basic composite pressure, so that the pressure corresponding to the thick zone is increased and the pressure corresponding to the thin zone is decreased, and the adjustment range increases nonlinearly with the increase of thickness deviation.

[0009] During the lamination process, the tension at the composite point and its changing trend are acquired in real time to characterize the current tension fluctuation state and perform tension state detection.

[0010] Based on the tension fluctuation state, the target pressure of each zone is adjusted in a coordinated manner, and pressure-tension coupling compensation is performed so that the disturbance caused by tension fluctuation is compensated by pressure regulation.

[0011] Based on the revised target pressure, closed-loop control is applied to each pressure zone to gradually converge the actual pressure to the target pressure range.

[0012] The bubble rate and wrinkle rate after lamination are obtained through online detection and compared with preset thresholds to determine the quality status of lamination.

[0013] When thickness fluctuations, tension fluctuations, or abnormal quality indicators occur, the pressure regulation and coupling compensation parameters are adaptively updated based on historical operating data.

[0014] Preferably, based on the pressure zoning structure of the hot press roller, the thickness distribution data is divided into multiple zones, and the deviation of each zone relative to the average thickness of the entire width is calculated, specifically as follows:

[0015] The full-width thickness distribution column vector is sliced ​​at equal intervals according to the number of independent pressure zones along the axial direction of the hot press roller to obtain the zone thickness sub-vector. The full-width thickness data is sliced ​​according to the pressure zones to establish the axial mapping relationship between the sampling points and the pressure zones.

[0016] Calculate the average thickness of each partition sub-vector and compare it with the average thickness of the whole area to determine the regional attributes of each partition. Calculate the average thickness of each partition and compare it with the overall mean to quantitatively identify thick, thin and normal ink layer areas.

[0017] Perform sampling point missing rate verification and adjacent sampling point thickness jump detection on each partition sub-vector to detect sampling missing and data jump, and remove invalid data caused by laser occlusion or reflection interference;

[0018] The attribute markers and axial coordinates of each partition area are mapped to a thickness cloud map index table. The output is determined based on data integrity and abnormal jump detection results. The partition attributes and coordinates are integrated to generate an index table. The next step or re-collection is determined based on the verification results.

[0019] Preferably, based on the thickness deviation of each zone, differential adjustments are made on the basis of the basic composite pressure, specifically as follows:

[0020] Using the regional attribute flags of each partition in the thickness cloud map index table as input, and combining the basic composite pressure, the pressure increment of the ink layer thick area compensation and the pressure reduction of the thin area, the target pressure of each partition is analyzed. Based on the thickness cloud map, the basic pressure is mapped to the differentiated target value of each partition, and a benchmark reference for pressure equalization is established.

[0021] The actual pressure in each zone is collected in real time by pressure sensors, and the pressure deviation of each zone is calculated to quantify the degree of deviation between the actual pressure and the target pressure in each zone, providing a basis for subsequent adjustment direction and magnitude.

[0022] For the ink layer thick area corresponding to the area with the regional attribute flag equal to 1, pressure increase judgment is performed, and undervoltage detection and step pressure increase are performed on the ink layer thick area to ensure that the adhesive layer penetrates the ink layer to achieve full activation.

[0023] For the corresponding partitions of thin areas or inkless areas with a region attribute flag of -1, pressure reduction judgment is performed, and overpressure detection and step pressure reduction are performed on thin areas to prevent paper crushing and glue overflow defects.

[0024] Summarize the adjustment status of all partitions, determine whether the partition voltage equalization adaptation is complete, summarize the adjustment requirements of all partitions, determine whether the voltage equalization adaptation has converged globally, and control the process to proceed progressively or cyclically.

[0025] Preferably, tension state detection is performed, specifically as follows:

[0026] The real-time position offset of the first floating roller relative to the mechanical balance position is obtained, its absolute value is calculated and compared with the coarse adjustment absorption threshold, and it is detected whether the real-time position offset of the first floating roller exceeds the large value fluctuation absorption threshold. The first-stage passive damping is determined to be saturated and the subsequent active compensation requirement is triggered.

[0027] Obtain the position offset sequence of the second floating roller within the sliding time window, calculate its residual amplitude and compare it with the fine-tuning attenuation threshold, calculate the residual amplitude of the second floating roller within the sliding time window, and determine whether the low inertia fine-tuning damping is sufficient to attenuate high-frequency disturbances.

[0028] Obtain the current paper feed speed, calculate its time derivative value and compare it with the fluctuation threshold, identify speed abrupt changes and trigger micro-tension feedforward compensation;

[0029] The real-time roll gap height is collected by the hot press roll bearing seat displacement sensor, the roll gap height change is calculated and compared with the allowable threshold, the roll gap height change caused by the zone pressure adjustment is monitored, and it is determined whether the film linear speed and paper feed speed at the composite point are mismatched.

[0030] The system summarizes the over-limit indicators of damping at all levels, and determines whether to simultaneously activate active micro-tension feedforward compensation and roll gap height feedforward correction through logical OR operation, and simultaneously outputs compensation roll speed correction and unwinding angular velocity feedforward commands.

[0031] The real-time tension at the composite point is obtained, the deviation between the real-time tension at the composite point and the target tension is calculated, the deviation between the real-time tension at the composite point and the target tension is detected, and it is determined whether the tension fluctuation after the coordination of each level of damping converges to the allowable threshold.

[0032] Preferably, based on the tension fluctuation state, the target pressure of each zone is adjusted in a coordinated manner to perform pressure-tension coupling compensation, specifically as follows:

[0033] The bubble rate statistics of the online visual inspection system after lamination are obtained, and the bubble rate is compared with the first-level qualified threshold. The proportion of bubble defects in the composite layer is quantified by visual inspection to determine whether the local non-adhesion caused by uneven thickness has been effectively eliminated.

[0034] The wrinkle rate statistics of the online visual inspection system after composite are obtained, and the wrinkle rate is compared with the first-level qualified threshold. The proportion of longitudinal deformation defects of the film is quantified by visual inspection to determine whether the wrinkles caused by tension fluctuations have been effectively suppressed.

[0035] The system summarizes the indicators of excessive bubble rate and excessive wrinkle rate, and uses logical operations to determine whether the thickness sensing zone equal pressure damping collaborative judgment has been passed. Combining the two quality indicators of bubbles and wrinkles, it determines whether the three-level collaborative control effect of the thickness sensing zone equal pressure damping collaborative judgment meets the threshold for progressively entering the depth enhancement predictive tension pressure control.

[0036] When the thickness-sensing partition pressure equalization damping collaborative judgment pass flag equal to 0, it triggers a return based on the thickness deviation of each partition, and performs differentiated adjustment and readjustment based on the basic composite pressure, and records the number of iterations, and simultaneously expands the pressure adjustment step size, and gradually eliminates residual bubbles and viscous defects through incremental pressure correction.

[0037] The iteration adjustment counter is compared with the maximum allowed number of iterations to determine whether an alarm shutdown is triggered, and the upper limit of the automatic iteration adjustment number is limited to prevent infinite loop. If the target is not met after three iterations, the system will switch to manual intervention mode in a timely manner.

[0038] Preferably, based on the corrected target pressure, closed-loop control adjustment is performed on each pressure zone, specifically as follows:

[0039] During the stable operation phase of thickness-sensing partitioned pressure equalization and damping collaborative determination, paper feed speed, composite point tension, partitioned average pressure and roll gap height are continuously collected at a fixed sampling period to construct a multi-dimensional time series data matrix and verify the window length, providing a sample basis that meets the input dimension requirements for the LSTM model;

[0040] Perform missing rate statistics on each dimension of the time series data matrix, and calculate the proportion of missing sampling points in each dimension to identify data incompleteness caused by sensor communication interruption or packet loss.

[0041] The data repair strategy is determined based on the degree of missing data, and interpolation repair or extended collection is selected based on the missing rate threshold to achieve a balance between data integrity and model timeliness.

[0042] Obtain the current paper batch identifier, film type identifier, and paper joint position mark, compare them with the baseline value when the historical window was initially established, monitor batch, film, and joint signals, and forcibly clear the historical window when the operating conditions change abruptly to prevent old operating condition data from contaminating the prediction of new operating conditions.

[0043] The system summarizes three criteria: window length, data integrity, and operational stability. It then performs a prediction model readiness check and, based on these three criteria, decides whether to activate the LSTM model or perform corresponding data repair and reconstruction operations.

[0044] Preferably, the bubble rate and wrinkle rate after lamination are obtained through online detection and compared with preset thresholds to determine the lamination quality status, specifically as follows:

[0045] Based on historical time series data windows, and through forward propagation of the LSTM model, multidimensional prediction sequences for multiple future control cycles are output. Based on historical windows and through forward propagation of the LSTM model, prediction sequences for velocity and thickness distribution in multiple future cycles are output, providing a data basis for disturbance prediction.

[0046] Calculate the maximum step change in the velocity prediction sequence and the maximum difference rate in the thickness prediction sequence to determine whether a deterministic perturbation mode exists and decide whether to activate the agent.

[0047] If the perturbation significance flag is assigned a value of 1, the predicted sequence will be input into the deep reinforcement learning agent, and the policy network will output the future multi-cycle pressure zone adjustment increment and compensation roller speed correction sequence.

[0048] The controller preloads the action sequence sequentially, and accumulates the correction amount cycle by cycle based on the current actual value to generate pre-executed control commands, thereby realizing predictive control output before disturbances occur.

[0049] If the disturbance significance flag is set to 0, then the stable operating condition standby judgment is used. When the predicted trend is stable and there is no significant disturbance, the deep enhancement prediction tension pressure control enters low power standby monitoring. The thickness sensing partition equal pressure damping collaborative judgment maintains the basic collaborative regulation to save computing power.

[0050] If the significance of the disturbance is equal to 0, then the current operating condition is determined to require no predictive intervention.

[0051] If the disturbance significance flag is equal to 1, it is determined that the bubble rate and wrinkle rate after coating need to be obtained through online detection and compared with the preset threshold to determine the coating quality status.

[0052] The present invention has the following beneficial effects:

[0053] 1. The adaptive control process for the environmentally friendly lamination of this packaging box involves a laser line scanning sensor performing a full-width thickness scan before the paper enters the hot-press lamination zone. This scan acquires thickness distribution data covering the entire width and calculates the maximum thickness difference rate to assess paper thickness uniformity. If the thickness variation exceeds a preset threshold, local thickness unevenness is identified, triggering the adaptive control process. If the variation does not exceed the threshold, the conventional constant pressure and tension lamination mode is maintained. For paper with thickness unevenness, the axial coordinates of thick and thin ink layers in the thickness distribution matrix are further analyzed to verify data integrity and the presence of abnormal jumps. Under valid data conditions, a thickness cloud map index table corresponding one-to-one with the pressure zones of the hot-press roller is generated, providing a precise spatial mapping benchmark for subsequent zoned pressure equalization. When data is abnormal, the sensor self-cleaning or calibration is triggered until a valid cloud map is obtained. Through full-width thickness pre-scanning and cloud map analysis before paper entry, the risk of thickness unevenness is identified in advance, providing a precise benchmark for zoned pressure equalization and eliminating the potential for localized air bubbles and loose adhesion from the source.

[0054] 2. The adaptive control process of this environmentally friendly lamination for packaging boxes is based on a thickness cloud map index table. Each independent pressure zone of the hot press roller performs differentiated pressure adaptation. The zone corresponding to the thick ink layer performs pressure increase adjustment to ensure that the adhesive layer is fully activated and penetrates the ink layer. The zone corresponding to the thin area or the area without ink layer performs pressure decrease adjustment to avoid adhesive layer overflow or paper crushing. Through the multi-zone independent pressure array, dynamic matching with the paper thickness distribution is achieved. While the pressure is homogenized, a multi-level tension attenuation buffer zone is constructed at the unwinding end. Two-stage floating rollers absorb large-value tension fluctuations and high-frequency residual disturbances respectively. A micro-tension active compensation roller is set at the lamination inlet. Feedforward compensation is performed based on the differential value of the paper feed speed and the change in the roller gap height. The film allowance is released or tightened to maintain constant tension at the lamination point and synchronization with the linear speed. This achieves synergistic optimization of pressure homogenization and tension stability. Through the synergy of independent pressure equalization in each zone and multi-level tension damping, the pressure mismatch caused by uneven paper thickness and the longitudinal deformation caused by film tension fluctuation are solved simultaneously, achieving stable lamination quality.

[0055] 3. The adaptive control process of the environmentally friendly film coating of this packaging box continuously collects time-series data on speed, tension, pressure, and thickness during the stable operation of the basic collaborative control system to construct a historical window. The LSTM model outputs a prediction sequence of speed fluctuation trends and thickness distribution changes for multiple control cycles in the future. When the prediction sequence shows a deterministic disturbance mode, the deep reinforcement learning agent outputs the pressure zone adjustment increment and tension compensation roller speed correction sequence in advance. The controller pre-executes these sequences before the actual disturbance occurs, achieving prediction-as-control. After the prediction is executed, the bubble rate, wrinkle rate, and system energy consumption are detected online. If the quality is further improved compared to the basic control and the energy consumption does not increase significantly, the optimal parameters are sent back to the cloud knowledge graph. If the quality is not improved or the prediction deviation is too large, the system immediately reverts to the basic collaborative control mode and triggers the local self-learning update of the DRL model. After correcting the prediction deviation, the system re-enters the time-series acquisition and prediction cycle. Through deep reinforcement learning predictive control and cloud knowledge graph feedback, pre-adjustment before disturbance and cross-device group evolution are achieved, ensuring composite quality while avoiding additional energy consumption burden.

[0056] 4. The adaptive control process of the environmentally friendly film coating of this packaging box achieves precise pressure distribution through nonlinear mapping of thickness deviation, reducing the generation of local bubbles. Through tension-pressure coupling compensation, it suppresses wrinkle defects caused by tension fluctuations. Through the parameter adaptive update mechanism, it achieves dynamic optimization under different working conditions. Attached Figure Description

[0057] Figure 1 This is a flow chart of the adaptive control process for the environmentally friendly film coating of packaging boxes according to the present invention.

[0058] Figure 2 The flowchart for the collaborative determination of thickness sensing zone equalization damping as described in claims 3-6 of this invention is shown below.

[0059] Figure 3 This is a flowchart of the depth-strengthening predictive tension and pressure control process according to claims 7-9 of the present invention;

[0060] Figure 4 This is a schematic diagram of the device structure of the present invention. Detailed Implementation

[0061] Example 1: An adaptive control process for environmentally friendly lamination of packaging boxes, see reference. Figure 1 ,include:

[0062] Before the paper enters the hot-pressing lamination zone, the paper thickness distribution data is obtained by full-width scanning, the average thickness and the maximum thickness difference rate of the full width are calculated, and the degree of thickness variation is judged accordingly.

[0063] Based on the pressure zoning structure of the hot press roller, the thickness distribution data is divided into multiple zones. The deviation of each zone from the average thickness of the whole width is calculated, and each zone is determined to be a thick zone, a thin zone, or a normal zone.

[0064] Based on the thickness deviation of each zone, differential adjustment is made on the basis of the basic composite pressure, so that the pressure corresponding to the thick zone is increased and the pressure corresponding to the thin zone is decreased, and the adjustment range increases nonlinearly with the increase of thickness deviation.

[0065] During the lamination process, the tension at the composite point and its changing trend are acquired in real time to characterize the current tension fluctuation state and perform tension state detection.

[0066] Based on the tension fluctuation state, the target pressure of each zone is adjusted in a coordinated manner, and pressure-tension coupling compensation is performed so that the disturbance caused by tension fluctuation is compensated by pressure regulation, thereby realizing the synergistic compensation relationship between the pressure demand caused by thickness difference and tension disturbance.

[0067] Based on the revised target pressure, closed-loop control is applied to each pressure zone to gradually converge the actual pressure to the target pressure range.

[0068] The bubble rate and wrinkle rate after lamination are obtained through online detection and compared with preset thresholds to determine the quality status of lamination.

[0069] When thickness fluctuations, tension fluctuations, or abnormal quality indicators occur, the pressure regulation and coupling compensation parameters are adaptively updated based on historical operating data.

[0070] Specifically, the calculation of the average thickness across the entire width and the maximum thickness variation rate, and the determination of the degree of thickness variation based on these, are as follows:

[0071] Before the paper enters the hot-pressing lamination zone, at a fixed scanning station, a laser line scanning sensor performs a full-width scan along the paper's axial direction at equal intervals, collecting data. The thickness values ​​of discrete sampling points are used to construct a full-width thickness distribution column vector. Dense sampling is then performed along the width to construct a thickness digital matrix, providing the original data foundation for ink layer thickness area identification and difference calculation.

[0072] ;

[0073] In the formula, This is a column vector representing the thickness distribution across the entire width of the image, with superscripts... This represents the transpose of a vector. For the first Paper thickness measurement values ​​at each axial sampling point , The total number of equally spaced sampling points along the width of the paper is determined by the sensor resolution and the paper width. This is the transpose operator, representing a mathematical operation that converts a row vector into a column vector;

[0074] With thickness distribution column vector Based on this, calculate the average thickness across the entire width. Maximum thickness and minimum thickness This establishes a baseline statistic for thickness fluctuation analysis, extracts the mean and extreme values ​​of thickness, quantifies the overall thickness level and absolute fluctuation range, and provides a benchmark reference for calculating the difference rate.

[0075] ;

[0076] ;

[0077] ;

[0078] In the formula, This is the arithmetic mean of the thickness across the entire sheet, reflecting the overall thickness level of the paper. for The maximum thickness among the sampling points corresponds to the area with the thickest ink layer. for The minimum thickness among the sampling points corresponds to the area with no ink layer or the thinnest area of ​​the paper. From arrive The summation symbol, and In order to be in Value to Operators for finding the maximum and minimum values ​​within a range;

[0079] With average thickness Using this as a benchmark, calculate the maximum thickness difference. The ratio of the average thickness to the normalized maximum thickness variation rate is obtained. This eliminates the dimensional influence caused by different paper basis weights, normalizes the absolute thickness difference into a relative difference rate, eliminates the dimensional interference of paper basis weights, and makes different batches of paper comparable:

[0080] ;

[0081] In the formula, The maximum thickness variation rate characterizes the relative fluctuation of paper thickness. This represents the maximum thickness difference across the entire width. This reflects the absolute height difference between thick and thin areas of the ink layer. The average thickness across the entire width is used as a normalization benchmark, with the unit being mm; ×100% is the percentage conversion factor, which converts the decimal ratio into a percentage form.

[0082] Maximum thickness difference rate Compared with the preset benchmark threshold Perform numerical comparisons and output Boolean-type variability risk flags. This serves as a switch for whether to activate adaptive regulation or maintain the normal mode:

[0083] ;

[0084] In the formula, This is a risk flag for variability; a value of 1 indicates a risk of uneven thickness distribution, while a value of 0 indicates a uniform thickness distribution. The preset maximum thickness difference rate benchmark threshold is determined by the paper type and lamination process window;

[0085] If the maximum thickness difference rate is greater than the preset benchmark threshold, that is If the paper exhibits localized thickness unevenness, with significant differences between thick and thin ink layers, then the variability risk marker is activated. The value is assigned to 1, and the thickness distribution data is divided into multiple partitions according to the pressure partition structure of the hot press roller. The deviation of each partition relative to the average thickness of the whole width is calculated, and the thickness sensing-partition pressure equalization-multi-level tension damping coordinated control is started.

[0086] If the maximum thickness difference rate is less than or equal to the preset benchmark threshold, that is... If the paper thickness distribution is uniform and there is no risk of pressure mismatch, then the variability risk flag is set. The value is set to 0, and the normal constant pressure and constant tension coating mode is maintained without activating adaptive control;

[0087] The thickness distribution data was obtained through laser line scanning.

[0088] Example 2 is an improvement upon Example 1. (See attached document for details.) Figure 2 The adaptive control process for the environmentally friendly film coating of this packaging box divides the thickness distribution data into multiple zones based on the pressure zoning structure of the hot press rollers, and calculates the deviation of each zone from the average thickness of the entire surface. Specifically:

[0089] Full-width thickness distribution column vector According to the number of independent pressure zones along the axial direction of the hot press roller Perform isometric slicing to obtain Each partition thickness sub-vector slices the full-width thickness data according to pressure partitions, establishing an axial mapping relationship between sampling points and pressure partitions:

[0090] ;

[0091] In the formula, For the first The thickness sub-vector corresponding to each pressure zone. This represents the total number of independently controllable pressure zones along the axial direction of the hot press roller. For the first Each partition in the column vector The starting sampling point index in the data. , For the first Each partition in the column vector The index of the terminating sampling point in the middle, , The number of sampling points corresponding to each pressure zone. ;

[0092] For each partition subvector Calculate the average thickness of the partition and the average thickness of the entire width Numerical comparisons are performed to determine the regional attributes of each zone, the average thickness of each zone is calculated and compared with the overall mean, and thick, thin, and normal ink layer areas are quantitatively identified.

[0093] ;

[0094] ;

[0095] In the formula, For the first The average thickness of each partition For the first The region attribute flags of each partition. The thickness deviation threshold for thick and thin areas is determined by the difference between the ink layer thickness and the paper substrate thickness. The average thickness across the entire width;

[0096] If the region attribute flag is output as 1, that is... Then determine the first Each zone is a thick ink layer zone;

[0097] If the region attribute flag is output as -1, that is... Then determine the first Each partition is a thin partition;

[0098] If the region attribute flag is output as 0, that is... Then determine the first Each partition is a normal partition;

[0099] For each partition sub-vector, perform sampling point missing rate verification and thickness jump detection between adjacent sampling points to detect sampling missing and data jumps, and remove invalid data caused by laser occlusion or reflection interference:

[0100] ;

[0101] ;

[0102] In the formula, The missing rate of full-area sampling points. This represents the number of invalid sampling points caused by obstruction or reflection failure during laser line scanning. For the first Abnormal transition flags for each partition This is an existential quantifier, indicating that there exists at least one element within the interval that satisfies the condition. It is a universal quantifier, indicating that all elements within the interval satisfy the condition. This represents the absolute difference in thickness between adjacent sampling points. The abnormal jump detection threshold is determined by the paper surface roughness and the sensor noise level;

[0103] If at least one Make Then determine the first An abnormal transition exists in one partition; abnormal transition flag. The value is assigned to 1;

[0104] If for all All Then determine the first No abnormal transitions were observed in any of the partitions; abnormal transition flags were displayed. The value is assigned to 0;

[0105] Set the regional attribute flags for each partition The thickness contour map index table is mapped to the axial coordinate position of the data. The output is determined based on the data integrity and abnormal jump detection results. The index table is generated by integrating the partition attributes and coordinates. The next step or re-collection is determined based on the verification results.

[0106] ;

[0107] ;

[0108] In the formula, For thickness contour map index table, by A column vector consisting of the region attribute flags and axial coordinates of each partition. For the first The axial center coordinates of each pressure zone in the paper width direction. As a marker of data validity, The allowable threshold for the sampling point missing rate is determined by the sensor's reliability level. This is the logical AND operator, indicating that both the preceding and following conditions must be satisfied. The logical OR operator indicates that at least one of the preceding and following conditions must be met. For all The sum of abnormal transition flags for each partition;

[0109] like Less than or equal to and If the value is 0, then the data validity flag is set. A value of 1 is assigned to determine that the axial coordinates of the thick and thin ink layers in each partition have been effectively resolved, the thickness difference data is complete and there are no abnormal jumps, and at this point, a thickness contour map index table is generated. Based on the thickness deviation of each zone, differential adjustments are made on the basis of the basic composite pressure;

[0110] like Greater than ,or If the value is greater than 0, then the data validity flag is set. If the value is set to 0, it is determined that the parsing has failed or there is a data anomaly. At this time, the sensor self-cleaning or calibration program is triggered. After the calibration is completed, the thickness distribution data is re-divided into multiple partitions according to the pressure partitioning structure of the hot press roller, and the deviation of each partition relative to the average thickness of the whole width is calculated.

[0111] If the data validity flag is not clear after three consecutive re-executions. If the assigned value is still 0, the sensor is determined to be continuously faulty, triggering an alarm and shutdown.

[0112] The partitions correspond one-to-one with the independent axial pressure partitions of the hot press roller.

[0113] This embodiment also provides that, based on the thickness deviation of each zone, a differentiated adjustment is made on the basis of the basic composite pressure, specifically as follows:

[0114] Thickness contour map index table Regional attribute flags for each partition As input, combined with basic composite pressure Incremental pressure compensation in thick ink layer areas and thin-area pressure reduction The target pressure for each zone is analyzed, and the base pressure is mapped to the differentiated target value for each zone based on the thickness contour map, thus establishing a benchmark reference for pressure equalization adaptation.

[0115] ;

[0116] ;

[0117] In the formula, For the first The target pressure for each partition The basic composite pressure is determined by the paper basis weight and the film type. For the first Pressure adjustment amount for each partition The pressure increment compensation for thicker ink layers is determined by the ink layer thickness and the activation requirements of the binder layer. The pressure reduction for thin areas is determined by the upper limit of paper pressure resistance and the requirement to prevent adhesive layer overflow. For the first Regional attribute flags for each partition;

[0118] If the region attribute flag If the value is equal to 1, then determine the first... Each zone represents the ink layer thickness area, with pressure correction amount. Assign the value to the first parameter, that is ;

[0119] If the region attribute flag If the value is equal to -1, then determine the first... Each zone is a thin zone, with pressure correction amount. Assign the value to the second parameter, that is ;

[0120] If the region attribute flag If the value is 0, then determine the first... Each zone is a normal zone, with pressure correction. The value is assigned to the third parameter, which is 0;

[0121] Real-time pressure data is collected using pressure sensors in each zone. The pressure deviation of each zone is calculated, and the degree of deviation between the actual pressure and the target pressure in each zone is quantified, providing a basis for subsequent adjustment direction and magnitude.

[0122] ;

[0123] In the formula, For the first Pressure deviation of each partition For the first The current actual pressure of each zone is collected by pressure sensors;

[0124] For regional attribute markers For ink layer thickness areas equal to 1, a pressure boosting judgment is performed. For these thick ink layer areas, undervoltage detection and step-by-step pressure boosting are applied to ensure that the adhesive layer penetrates the ink layer and achieves full activation.

[0125] ;

[0126] ;

[0127] In the formula, For the first The boost pressure determination flag for each partition. This is the pressure regulation dead zone threshold, used to avoid frequent adjustments caused by small fluctuations. The single pressure adjustment step size is determined by the actuator's response speed and control accuracy. For the first Pressure command values ​​for each partition;

[0128] If the pressure deviation of the zone is greater than the pressure regulation dead zone threshold, that is... Then determine the first The actual pressure in some zones did not reach the compensation pressure required to penetrate the ink layer, posing a risk of insufficient adhesive layer activation. (Pressure increase judgment indicator) Assign a value of 1 and execute pressure boosting regulation; pressure command value. Under current real pressure Increase the single pressure adjustment step size based on the existing model. ;

[0129] If the pressure deviation of the zone is less than or equal to the pressure regulation dead zone threshold, that is... Then determine the first The pressure in each zone has reached the target; pressure increase indicator. The value is assigned to 0, which is the pressure command value. Maintain current actual pressure constant;

[0130] For regional attribute markers For areas with a thickness equal to -1 or no ink layer, a pressure reduction judgment is performed. Overpressure detection and step-by-step pressure reduction are then applied to the thin areas to prevent paper crushing and adhesive overflow defects.

[0131] ;

[0132] ;

[0133] In the formula, For the first Decompression criteria for each partition, This is the upper limit parameter for paper pressure resistance; exceeding this value may result in adhesive overflow or paper crushing. For logical OR operator, For logical AND operator;

[0134] If the current actual pressure Greater than the paper's pressure resistance limit parameter or pressure deviation of the zone Less than the negative pressure regulation dead zone threshold Then determine the first If the actual pressure in any zone exceeds the paper's pressure limit or is higher than the target pressure, there is a risk of adhesive overflow or paper crushing. Pressure reduction judgment indicator. Assign a value of 1 and execute pressure reduction adjustment; pressure command value. Under current real pressure Reduce the single pressure adjustment step size based on this. ;

[0135] If the current actual pressure Less than or equal to the paper's maximum pressure resistance parameter And the pressure deviation of the partition Greater than or equal to the negative pressure adjustment dead zone threshold Then determine the first The pressure in each zone is within the safe range and has reached the target; pressure reduction judgment indicator. The value is assigned to 0, which is the pressure command value. Maintain current actual pressure constant;

[0136] Summary of all The system monitors the adjustment status of each partition, determines whether partition voltage equalization adaptation is complete, summarizes all partition adjustment requirements, determines whether voltage equalization adaptation has globally converged, and controls the flow to proceed incrementally or cyclically.

[0137] ;

[0138] In the formula, To establish a convergence flag for the partitioned voltage equalization adaptation. From arrive The summation symbol, For the first The boost judgment flag for each partition is only applicable to... The selected partition is valid; all other partitions are forced to have their values ​​set to 0. For the first The decompression determination flag for each partition is only applicable to... The selected partition is valid; all other partitions are forced to be set to 0.

[0139] If the partitioned voltage equalization adaptation convergence flag is output as 1, that is... Then determine all The pressure of each zone has been matched with the thickness contour map requirements, and tension status detection has been performed.

[0140] If the partitioned voltage equalization adaptation convergence flag is output as 0, that is... If at least one zone has not yet completed pressure matching, the actual pressure will be collected again in real time by the pressure sensors in each zone. And calculate the pressure deviation of each zone, so as to re-acquire the actual pressure and perform adjustment until the zone pressure equalization and adaptation convergence flag output is 1;

[0141] The pressure regulation employs a nonlinear regulation strategy, which increases the pressure variation amplitude as the thickness deviation increases.

[0142] This embodiment also provides a method for detecting tension state, specifically:

[0143] Obtain the real-time position offset of the first floating roller relative to the mechanical equilibrium position. The absolute value is calculated and compared with the coarse adjustment absorption threshold. It is then checked whether the real-time position offset of the first floating roller exceeds the large fluctuation absorption threshold to determine whether the first-stage passive damping is saturated and to trigger subsequent active compensation requirements.

[0144] ;

[0145] In the formula, This represents the position offset of the first floating roller, relative to its mechanical equilibrium position. This is the absolute value of the position offset. The coarse adjustment absorption threshold of the first floating roller is determined by the floating roller's stroke and its ability to attenuate large amplitude fluctuations. This is a coarse adjustment exceeding the limit indicator;

[0146] If the absolute value of the position offset Greater than the coarse adjustment absorption threshold of the first floating roller If the offset of the first floating roller exceeds the absorption range of large fluctuation, the coarse adjustment over-limit flag will be triggered. The value is assigned to 1;

[0147] If the absolute value of the position offset Less than or equal to the first floating roller coarse adjustment absorption threshold If the first floating roller is within the coarse adjustment absorption range, the coarse adjustment over-limit flag will be displayed. The value is assigned to 0;

[0148] Obtain the position offset sequence of the second floating roller within the sliding time window, calculate its residual amplitude and compare it with the fine-tuning attenuation threshold, calculate the residual amplitude of the second floating roller within the sliding time window, and determine whether the low-inertia fine-tuning damping is sufficient to attenuate high-frequency disturbances:

[0149] ;

[0150] ;

[0151] In the formula, For the second floating roller at time Position offset The length of the sliding time window is determined by the period of the high-frequency disturbance. and In the time window The maximum and minimum value operators within the scope, This represents the residual amplitude of the second floating roller. The attenuation threshold for the second floating roller is fine-tuned and determined by the low inertia design objective. To fine-tune the over-limit indicator;

[0152] If the residual amplitude of the second floating roller Greater than the fine-tuning attenuation threshold of the second floating roller If the second floating roller has unattenuated high-frequency residual disturbance, the fine-tuning over-limit flag will be activated. The value is assigned to 1;

[0153] If the residual amplitude of the second floating roller Less than or equal to the second floating roller fine-tuning attenuation threshold If the high-frequency disturbance of the second floating roller has been sufficiently attenuated, the fine-tuning over-limit indicator will be activated. The value is assigned to 0;

[0154] Get the current paper feed speed Calculate its time derivative and compare it with the fluctuation threshold to identify velocity abrupt changes and trigger micro-tension feedforward compensation:

[0155] ;

[0156] ;

[0157] In the formula, The current paper feed speed is collected by the encoder. The differential value of the paper feed speed represents the rate of change of speed. The differential symbol, The absolute value of the velocity derivative. The differential fluctuation threshold of the paper feed speed is determined by the allowable fluctuation range of the tension. This is a speed fluctuation indicator;

[0158] If the absolute value of the velocity derivative Greater than the differential fluctuation threshold of paper feed speed If this is detected, it indicates a sudden fluctuation in the paper feed speed, and the speed fluctuation indicator is... The value is assigned to 1;

[0159] If the absolute value of the velocity derivative Less than or equal to the differential fluctuation threshold of paper feed speed If the paper feed speed changes smoothly, then the speed fluctuation indicator is considered stable. The value is assigned to 0;

[0160] The real-time roll gap height is collected by a hot-press roll bearing seat displacement sensor. The change in roll gap height is calculated and compared with an allowable threshold. The change in roll gap height caused by zone pressure adjustment is monitored to determine whether there is a mismatch between the film linear velocity and the paper feed speed at the lamination point.

[0161] ;

[0162] ;

[0163] In the formula, The current real-time roll gap height is collected by a displacement sensor. The reference roll gap height is determined by the sum of the nominal paper thickness and the film thickness. This represents the change in roll gap height, reflecting the deviation in roll gap height caused by zoned pressure adjustment. The allowable threshold for roll gap height variation is determined by the elastic elongation of the film. This is a sign of roll gap mismatch.

[0164] If the absolute value of the change in roll gap height Greater than the allowable threshold for roll gap height variation If the change in roll gap height causes the film linear speed and paper feed speed to be out of sync, then the roll gap mismatch indicator is triggered. The value is assigned to 1;

[0165] If the absolute value of the change in roll gap height Less than or equal to the allowable threshold for roll gap height variation If the roll gap height is within the synchronization range, the roll gap mismatch indicator will be displayed. The value is assigned to 0;

[0166] The system summarizes the over-limit indicators at all levels of damping and uses logical OR operations to determine whether to simultaneously initiate active micro-tension feedforward compensation and roll gap height feedforward correction, while simultaneously outputting compensation roll speed correction and unwinding angular velocity feedforward commands:

[0167] ;

[0168] ;

[0169] ;

[0170] In the formula, The damped cooperative start flag is generated by the logical OR operator. The result is that a value of 1 indicates that at least one level of damping exceeds its limit and active compensation needs to be activated, while a value of 0 indicates that all levels of damping are within the stable range. The reference speed for the micro-tension active compensation roller. This is the commanded value for the compensating roller speed. The velocity feedforward compensation coefficient is determined by the film's elastic modulus and the transmission ratio of the compensation roller. The reference angular velocity for the unwinding servo motor. This is the unwinding servo angular velocity command value. The roll gap height feedforward correction factor is determined by the film allowance release requirement. The current unwinding diameter is updated in real time by the diameter calculation model;

[0171] If the damping coordination start flag If the value equals 1, it indicates that at least one level of damping has exceeded the limit, and active micro-tension feedforward compensation and roll gap height feedforward correction must be initiated simultaneously. At this time, the command value of the compensation roll speed is at the reference speed of the active micro-tension compensation roll. Based on the superposition The unwinding servo angular velocity command value is at the unwinding servo motor reference angular velocity. Based on the superposition ;

[0172] If the damping coordination start flag If the value equals 0, it is determined that all damping parameters are in the stable range, and there is no need to activate active compensation. The command value of the compensation roller speed maintains the reference speed of the micro-tension active compensation roller. The unwinding servo angular velocity command value maintains the reference angular velocity of the unwinding servo motor. ;

[0173] Acquire the real-time tension at the composite point, calculate its deviation from the target tension, detect the deviation between the real-time tension at the composite point and the target tension, and determine whether the tension fluctuation after the coordination of each level of damping converges to the allowable threshold.

[0174] ;

[0175] ;

[0176] In the formula, The real-time tension at the composite point is collected by a tension sensor. The target stable tension is determined by the tensile properties of the membrane material. This is the absolute value of the tension deviation. The allowable threshold for tension fluctuation is determined by the wrinkle control requirements. This is an indicator of tension stability;

[0177] If the absolute value of the tension deviation Less than or equal to the allowable tension fluctuation threshold If the tension fluctuation is controlled within the allowable range, and the multi-stage tension damping provides synergistic stability, then the tension stability indicator is... The value is assigned to 1;

[0178] If the absolute value of the tension deviation Greater than the allowable threshold for tension fluctuation If the tension fluctuation exceeds the limit, the tension stability indicator will be invalidated. Set the value to 0, and reacquire the real-time position offset of the first floating roller relative to the mechanical balance position. Calculate its absolute value and compare it numerically with the coarse-tuned absorption threshold to re-acquire damping parameters at each level and perform adjustments until the tension stabilizes. It equals 1;

[0179] Among them, tension fluctuation and its rate of change are characterized by the real-time tension value and its rate of change.

[0180] This embodiment also provides, based on the tension fluctuation state, a linkage correction for the target pressure of each zone, performing pressure-tension coupling compensation, specifically as follows:

[0181] Obtain the bubble rate statistics from the online visual inspection system after lamination, compare the bubble rate with the first-level acceptance threshold, quantify the proportion of bubble defects in the composite layer through visual inspection, and determine whether the localized loose adhesion caused by uneven thickness has been effectively eliminated.

[0182] ;

[0183] In the formula, The bubble rate is obtained by an online visual inspection system that identifies defects in the full-page image of the laminated paper and calculates the percentage of bubble areas. The first-level bubble rate threshold is determined by the lamination quality standard and the paper-film combination process window. This indicates an excessive bubble rate.

[0184] If the bubble rate Greater than the first-level bubble rate qualification threshold If the bubble rate exceeds the standard after compounding, the bubble rate exceeding the limit will be indicated by the bubble rate indicator. The value is assigned to 1;

[0185] If the bubble rate Less than or equal to the first-level bubble rate qualification threshold If the bubble rate is within the acceptable range, the bubble rate exceeding the limit will be marked as acceptable. The value is assigned to 0;

[0186] The wrinkle rate statistics of the online visual inspection system after lamination are obtained, and the wrinkle rate is compared with the first-level qualified threshold. The proportion of longitudinal deformation defects in the film is quantified by visual inspection to determine whether the wrinkles caused by tension fluctuations have been effectively suppressed.

[0187] ;

[0188] In the formula, The wrinkle rate is obtained by using an online visual inspection system to perform edge detection on the longitudinal texture of the composite film and to statistically analyze the wrinkle line density. The first-level wrinkle rate qualification threshold is determined by the coating tension stability standard and the tensile properties of the membrane material. This indicates that the wrinkle rate exceeds the limit.

[0189] If wrinkle rate Greater than the first-level wrinkle rate qualification threshold If the wrinkle rate exceeds the standard after lamination, the wrinkle rate exceeding the limit will be indicated. The value is assigned to 1;

[0190] If wrinkle rate Less than or equal to the first-level wrinkle rate qualification threshold If the wrinkle rate is within the acceptable range, it is considered to be within the acceptable range; otherwise, the wrinkle rate exceeds the limit. The value is assigned to 0;

[0191] The system summarizes the indicators for excessive bubble rate and excessive wrinkle rate, and uses logical operations to determine whether the thickness-sensing zoned pressure equalization and damping collaborative judgment has been passed. Combining the two quality indicators of bubbles and wrinkles, it determines whether the three-level collaborative control effect of the thickness-sensing zoned pressure equalization and damping collaborative judgment meets the threshold for progressively entering depth-enhanced predictive tension pressure control.

[0192] ;

[0193] In the formula, This is a flag indicating whether the thickness sensing zone's equalizing damping coordination judgment has passed; a value of 1 indicates that the judgment has passed, and a value of 0 indicates that the judgment has failed. For logical AND operator, For logical OR operator;

[0194] If the bubble rate exceeds the limit, the indicator will be displayed. =0 and wrinkle rate exceeds the limit mark If the values ​​are both zero, meaning the bubble rate and wrinkle rate are within the acceptable range, then the basic three-level coordinated control has effectively solved the problems of uneven thickness and tension fluctuation. This indicates that the thickness sensing zone pressure equalization damping coordinated judgment has passed. The value is assigned to 1, and closed-loop control is performed on each pressure zone based on the corrected target pressure.

[0195] If the bubble rate exceeds the limit, the indicator will be displayed. Equal to 1 or wrinkle rate exceeding the limit. If the value equals 1, meaning at least one of the bubble rate or wrinkle rate exceeds the standard, then the quality is deemed to have failed to meet the standard after the thickness sensing partition equal pressure damping collaborative judgment is executed. The thickness sensing partition equal pressure damping collaborative judgment is then considered passed. The value is assigned to 0;

[0196] When the thickness sensing zone equal pressure damping coordination judgment passes the flag When the value equals 0, a return is triggered based on the thickness deviation of each partition. Differential adjustments are then made based on the basic composite pressure, and the iteration count is recorded. Simultaneously, the pressure adjustment step size is increased, and residual bubbles and viscous defects are gradually eliminated through incremental pressure correction.

[0197] ;

[0198] ;

[0199] ;

[0200] In the formula, A counter for thickness-sensing zone equalization damping collaborative determination iterative adjustment is used to record the number of times differential adjustments and readjustments are performed based on the base composite pressure, according to the thickness deviation of each zone. The initial value is 0. For the first The ink layer thickness area is compensated for by the pressure increment during the next iteration. For the first The pressure reduction in the thin region during the next iteration The pressure regulation step size amplification factor is determined by the actuator's limit stroke and safety margin, and ;

[0201] Iteratively adjust the counter The system compares the value with the maximum allowed number of iterations to determine whether to trigger an alarm and stop the system. It limits the upper limit of the number of automatic iterations to prevent infinite loops and switches to manual intervention mode in a timely manner if the target is not met after three iterations.

[0202] ;

[0203] In the formula, For the iterative over-limit alarm flag, a value of 1 indicates that an alarm has been triggered, and a value of 0 indicates that iteration is allowed to continue.

[0204] If the counter is adjusted iteratively If the value is greater than 3, it is determined that the thickness sensing zone equal pressure damping coordination judgment still fails after three consecutive return adjustments, and the iterative over-limit alarm flag is triggered. Setting the value to 1 will trigger an alarm prompting manual intervention.

[0205] If the counter is adjusted iteratively If the number of iterations is less than or equal to 3, it is determined that the number of iterations has not exceeded the limit, and the alarm flag for exceeding the iteration limit is displayed. A value of 0 is assigned, returning a differential adjustment based on the thickness deviation of each zone, on top of the basic composite pressure, and carrying the amplified pressure increment. and decompression Re-execute the partitioned pressure equalization adaptation and perform tension state detection again until the thickness-sensing partitioned pressure equalization damping coordination judgment passes the flag. It equals 1;

[0206] The pressure correction amount is dynamically adjusted according to the tension fluctuation range.

[0207] Example 3 is an improvement upon Example 2. (See attached document for details.) Figure 3 In this embodiment, closed-loop control adjustment is performed on each pressure zone based on the corrected target pressure, specifically as follows:

[0208] During the stable operation phase determined by the thickness-sensing partitioned pressure equalization and damping, paper feed speed, composite point tension, partitioned average pressure, and roll gap height are continuously collected at a fixed sampling period to construct a multi-dimensional time-series data matrix and verify the window length, providing a sample foundation that meets the input dimension requirements for the LSTM model.

[0209] ;

[0210] ;

[0211] In the formula, The data is a time-series matrix with dimensions N×4. This is the current historical data window length. For a moment The state vector, For a moment Paper feed speed, For a moment The composite point tension, For a moment The average pressure of each zone, For a moment Roll gap height, This indicates that the window length meets the standard. The minimum historical window length required for the LSTM prediction model is determined by the number of neurons in the model's input layer.

[0212] If the current historical data window length Greater than or equal to the minimum history window length required by the LSTM prediction model If the historical data window length meets the input requirements of the prediction model, then the window length meets the standard. The value is assigned to 1;

[0213] If the current historical data window length Smaller than the minimum historical window length required by the LSTM prediction model If the data window is insufficient, the window length meets the standard. The value is assigned to 0;

[0214] For time series data matrix The system performs missing rate statistics across various dimensions, calculating the percentage of missing sampling points for each dimension to identify data incompleteness issues caused by sensor communication interruptions or packet loss.

[0215] ;

[0216] ;

[0217] ;

[0218] In the formula, For dimension Data missing rate, For dimension In the window The number of missing sampling points within, The maximum missing rate among the four dimensions. This is the operator for finding the maximum value in a set of dimensions. This is a marker for missing data.

[0219] If the maximum missing rate If the value is greater than 0, it indicates that the data is missing; this is the missing data flag. The value is assigned to 1;

[0220] If the maximum missing rate If the value is 0, the data is considered complete and without missing data; the missing data flag is used. The value is assigned to 0;

[0221] The data repair strategy is determined based on the degree of missing data, and interpolation repair or extended data collection is selected based on the missing rate threshold, achieving a balance between data integrity and model timeliness.

[0222] ;

[0223] In the formula, This is a flag indicating whether interpolation is feasible. The maximum missing rate threshold for allowing interpolation completion is determined by the continuity requirements of the time series data;

[0224] If the maximum missing rate Greater than 0 and less than or equal to the maximum missing rate threshold allowed for interpolation completion If the degree of missing information is within the interpolable range, then the interpolation feasibility flag is set. The value is assigned to 1, and linear interpolation or spline interpolation is performed to complete the missing data. After completion, the time series data matrix is ​​re-applied. Statistical analysis of missing rates across various dimensions;

[0225] If the maximum missing rate Greater than the maximum missing rate threshold allowed for interpolation completion If the missing information is too severe to reliably interpolate, then the interpolation feasibility flag is set. The value is set to 0, the data acquisition time is extended until the window length meets the requirements again, then the multi-dimensional time series data matrix is ​​reconstructed and the window length is verified.

[0226] Obtain the current paper batch identifier, film type identifier, and paper joint location marker, compare them with the baseline values ​​when the historical window was initially established, monitor batch, film type, and joint signals, and forcibly clear the historical window when operating conditions change abruptly to prevent old operating condition data from contaminating new operating condition predictions.

[0227] ;

[0228] ;

[0229] ;

[0230] ;

[0231] In the formula, This is the current paper batch identifier. This refers to the paper batch identifier used when the history window was initially established. This is the current membrane material type identifier. This is the membrane material type identifier used when the historical window was initially established. This marks the location of the paper splice; a value of 1 is assigned when a splice is detected. This is an indicator function that outputs 1 if the condition is true, and 0 otherwise. This is a paper batch switching indicator. Replace the markings on the membrane material. For joint inspection marking, As a marker of sudden changes in operating conditions, For logical OR operator;

[0232] If the current paper batch identifier This is not equivalent to the paper batch identifier when the history window was initially created. If a paper batch change is detected, then the paper batch change flag is displayed. The value is assigned to 1;

[0233] If the current paper batch identifier Equivalent to the paper batch identifier when the history window is initially created. Then the paper batch switching flag The value is assigned to 0;

[0234] If the current membrane material type is identified This is not equivalent to the membrane type identifier when the history window was initially created. If the membrane material replacement is detected, then the membrane material replacement indicator will be displayed. The value is assigned to 1;

[0235] If the current membrane material type is identified Equal to the membrane material type identifier when the history window is initially created. Then the membrane material replacement mark The value is assigned to 0;

[0236] If the paper joint position is marked If the value is 1, a paper splice is detected; this is the splice detection flag. The value is assigned to 1;

[0237] If the paper joint position is marked If the value is 0, then the connector inspection mark is... The value is assigned to 0;

[0238] If the paper batch changeover indicator Equal to 1 or membrane material replacement mark Equal to 1 or connector inspection mark If the value is 1, it indicates a sudden change in the operating condition. (Operating condition change flag) Set the value to 1 to clear the history window and re-collect data;

[0239] If the paper batch changeover indicator =0 and membrane material replacement mark Equal to 0 and connector inspection mark If the value is 0, the operating condition is considered stable; otherwise, a sudden change in operating condition is indicated. The value is assigned to 0;

[0240] The system considers three criteria: window length, data integrity, and operational stability. It then performs a prediction model readiness check, combining these three factors to determine whether to activate the LSTM model or perform corresponding data repair and reconstruction operations.

[0241] ;

[0242] In the formula, This serves as a sign that the predictive model is ready. For logical AND operator, This refers to all other situations besides the conditions mentioned above;

[0243] If the window length meets the standard, the sign indicates that... The flag is equal to 1 and the data is missing. =0 and the indicator of sudden change in operating conditions If the value equals 0, it indicates that the historical data window length meets the requirements, the data is complete and without missing data, and there have been no sudden changes in the operating conditions; this marks the readiness of the prediction model. The value is set to 1, which activates the LSTM prediction model and obtains the bubble rate and wrinkle rate after coating through online detection. The values ​​are then compared with the preset threshold to determine the coating quality status.

[0244] If the operating condition changes suddenly, the flag If the value is 1, it is determined that a sudden change in the working condition has occurred. The historical window is cleared and data is collected again. After the window is rebuilt, closed-loop control adjustment of each pressure zone is performed based on the corrected target pressure.

[0245] If the window length meets the standard, the sign indicates that... =0 and the indicator of sudden change in operating conditions If the value is 0, it is determined that the data window is insufficient, the data acquisition time is extended, and the closed-loop control adjustment of each pressure zone is re-executed based on the corrected target pressure after the window length reaches the target.

[0246] If the window length meets the standard, the sign indicates that... The flag is equal to 1 and the data is missing. The flag is equal to 1 and the interpolation is feasible. Equal to 1 and the indicator of sudden change in operating conditions If the value is 0, then perform data interpolation to complete the data, and then re-compute the time series data matrix. Statistical analysis of missing rates across various dimensions;

[0247] If the window length meets the standard, the sign indicates that... The flag is equal to 1 and the data is missing. The flag is equal to 1 and the interpolation is feasible. =0 and the indicator of sudden change in operating conditions If the value is 0, it is determined that the missing value is too severe to reliably interpolate. The data acquisition time is extended, and the process is repeated after the data is complete, based on the corrected target pressure, to perform closed-loop control adjustment on each pressure zone.

[0248] The closed-loop control adopts an iterative adjustment method based on pressure deviation and sets an adjustment dead zone to avoid frequent fluctuations.

[0249] This embodiment also provides that the bubble rate and wrinkle rate after lamination are obtained through online detection and compared with a preset threshold to determine the lamination quality status, specifically:

[0250] Based on historical time series data windows After forward propagation through the LSTM model, multiple future values ​​are output, i.e. The multidimensional prediction sequence for each control cycle, based on historical windows and forward propagated via LSTM, outputs the predicted sequence for future multi-cycle velocity and thickness distribution, providing a data foundation for disturbance prediction.

[0251] ;

[0252] ;

[0253] In the formula, For the future A column vector of predicted paper feed speed for each control cycle. For the first Predicted paper feed speed at any given time. For the first The full-width thickness distribution prediction column vector at time t. For the first Time of the first The predicted thickness value for each sampling point. The predicted step size, i.e., the total number of future control cycles, is determined by the dimension of the LSTM output layer. Mark the current time. It is the transpose operator;

[0254] Calculate the maximum step change in the velocity prediction sequence and the maximum difference rate in the thickness prediction sequence to determine whether a deterministic perturbation mode exists and decide whether to activate the agent.

[0255] ;

[0256] ;

[0257] ;

[0258] ;

[0259] ;

[0260] In the formula, To predict the maximum step change in the velocity sequence, In order to be in Values ​​range from 1 to Maximum value operator within the range For the first The average thickness prediction at any given time. For the first Time-based thickness prediction variance rate for Maximum difference rate in thickness prediction within each cycle The velocity step disturbance threshold is determined by the allowable tension fluctuation range. The thickness variation disturbance threshold is determined by calculating the average thickness of the entire area and the maximum thickness difference rate, and based on this, a benchmark threshold for judging the degree of thickness variation. The indicator for predicting a joint or batch changeover is obtained by extrapolating from the paper joint position marker. A value of 1 indicates that a joint or batch changeover approach has been predicted, while a value of 0 indicates that none has been predicted. As a marker of the significance of the disturbance, For logical OR operator, For logical AND operator;

[0261] If the velocity prediction sequence has the largest step change Greater than the velocity step disturbance threshold or the maximum difference rate of thickness prediction within the period Greater than the thickness variation perturbation threshold Or predict connector or batch switching indicator If the value is 1, the predicted sequence is determined to show a deterministic perturbation pattern, and the perturbation significance is indicated by... The value is assigned to 1;

[0262] If the velocity prediction sequence has the largest step change Less than or equal to the velocity step disturbance threshold And the maximum difference rate of thickness prediction within the period Less than or equal to the thickness variation perturbation threshold And predict connector or batch switching flags If the value is 0, the predicted trend is considered to be stable without significant disturbance. The significance of the disturbance is indicated by... The value is assigned to 0;

[0263] If the disturbance is significant If a value of 1 is assigned, the predicted sequence will be... and The input is a deep reinforcement learning agent, which, after passing through a policy network, outputs a sequence of future multi-cycle pressure zone adjustment increments and compensation roller speed corrections:

[0264] ;

[0265] ;

[0266] In the formula, For the future A matrix of action sequences for each control cycle. For the first Action vector for each control cycle For the first The first control cycle Pressure adjustment increments for each pressure zone For the first The speed correction amount of the micro-tension active compensation roller in each control cycle;

[0267] The controller preloads the action sequence sequentially, accumulating correction values ​​cycle by cycle based on the current actual values ​​to generate pre-executed control commands, achieving predictive control output before disturbances occur.

[0268] ;

[0269] ;

[0270] In the formula, For the current moment The actual pressure of each pressure zone This represents the actual rotational speed of the micro-tension active compensation roller at the current moment. For the first Time of the first Pre-execution pressure instructions for each pressure partition For the first Pre-execution speed command for the time compensation roller;

[0271] If the disturbance is significant If the value is set to 0, then the stable operating condition standby judgment is used. When the predicted trend is stable and there is no significant disturbance, the deep enhanced predicted tension and pressure control enters low-power standby monitoring. The thickness-sensing partitioned pressure equalization damping collaborative judgment maintains the basic collaborative regulation to save computing power.

[0272] ;

[0273] In the formula, To further enhance the activation status flag for predictive tension and pressure control;

[0274] If the disturbance is significant If the value is 0, it is determined that no predictive intervention is needed for the current working condition, and the activation status flag for deeply enhanced predictive tension and pressure control is activated. When the value is set to 0, the deep enhancement predictive tension and pressure control enters a low-power standby monitoring state, and the basic coordinated regulation is maintained by the thickness sensing partition equal pressure damping collaborative judgment.

[0275] If the disturbance is significant If the value equals 1, it is determined that the bubble rate and wrinkle rate after lamination need to be obtained through online detection and compared with the preset threshold to determine the lamination quality status, and the activation status flag of the depth enhancement predictive tension pressure control is activated. When the value is set to 1, the controller outputs the preloaded action sequence in sequence before the disturbance actually occurs. Based on the current actual value, the correction amount is accumulated cycle by cycle to generate the pre-execution command generated by the pre-execution control command, thus realizing prediction as control.

[0276] The bubble rate and wrinkle rate were obtained through an online visual inspection system.

[0277] This embodiment also provides that, when thickness fluctuations, tension fluctuations, or abnormal quality indicators occur, the pressure regulation and coupling compensation parameters are adaptively updated based on historical operating data, specifically as follows:

[0278] Obtain the bubble rate after the execution of deep-strength predictive tension pressure control predictive control. wrinkle rate and system energy consumption Simultaneously, it reads the bubble rate benchmark value stored in the thickness sensing partition equalization damping collaborative determination stage. Wrinkle rate benchmark value and energy consumption benchmark values Simultaneously, the system collects mass and energy consumption parameters after the execution of depth-enhanced predictive tension and pressure control, and calls upon the benchmark values ​​from the thickness-sensing zone equalization and damping collaborative judgment stage to provide a comparison benchmark for calculating the improvement magnitude.

[0279] ;

[0280] In the formula, To further enhance the predictive control performance of tension and pressure control, the bubble rate after execution is statistically obtained from an online visual inspection system. To further enhance the prediction of wrinkle rate after tension and pressure control, it is obtained statistically by an online visual inspection system. To further enhance the overall system energy consumption after the execution of predictive tension and pressure control, it is obtained by integrating the power consumption within the predictive control cycle using the power metering module. The bubble rate benchmark value is used when the thickness sensing partition equalization damping collaborative determination stage is passed. It is stored at the time of the determination of whether the thickness sensing partition equalization damping collaborative determination (i.e., the content in claims 3-6) is passed. The wrinkle rate benchmark value is used when the thickness-sensing partition equalization damping collaborative determination stage is passed. It is stored at the time of the determination of whether the thickness-sensing partition equalization damping collaborative determination (i.e., the content in claims 3-6) is passed. The baseline value of the system's comprehensive energy consumption for the corresponding period in the thickness-sensing, zone-equalizing, and damping collaborative determination stage is stored by the power metering module. This is an assignment direction symbol, indicating that data is passed from the data source on the right to the variable on the left.

[0281] Calculate the improvement in bubble rate Improvement in wrinkle rate And compare it with the improvement threshold to determine whether predictive control brings about substantial quality improvement:

[0282] ;

[0283] ;

[0284] ;

[0285] In the formula, To improve the bubble rate, To improve the wrinkle rate, The threshold for judging the improvement of bubble rate is determined by the target of improving coating quality. The threshold for judging the improvement of wrinkle rate is determined by the target for improving coating quality. As a mark of quality improvement, For logical AND operator, For logical OR operator;

[0286] If the bubble rate is improved Greater than or equal to the bubble rate improvement judgment threshold And the improvement in wrinkle rate Greater than or equal to the wrinkle rate improvement judgment threshold The bubble rate and wrinkle rate are further reduced when the thickness sensing zone equal pressure damping synergistic judgment is used alone, which is a sign of quality improvement. The value is assigned to 1;

[0287] If the bubble rate is improved Less than the bubble rate improvement judgment threshold Or the amount of wrinkle improvement Less than the wrinkle rate improvement judgment threshold If the quality indicators are not improved or only a single improvement fails to meet the overall requirements, the quality improvement indicator will be determined. The value is assigned to 0;

[0288] The incremental system energy consumption of computationally enhanced predictive tension and pressure control relative to thickness-sensing partitioned equal pressure damping co-determination is calculated. Furthermore, by comparing the energy consumption with the allowable threshold for increased energy consumption, the energy cost of predictive control is quantified to ensure that quality improvement does not come at the expense of excessive energy consumption.

[0289] ;

[0290] ;

[0291] In the formula, This represents the system energy consumption increment, reflecting the additional energy burden introduced by predictive control. An allowable threshold for energy consumption is added, determined by energy-saving operation constraints. This is an energy consumption compliance mark;

[0292] If the system energy consumption increment Less than or equal to the allowable threshold for energy consumption increase If predictive control does not introduce a significant additional energy consumption burden, the energy consumption compliance mark is obtained. The value is assigned to 1;

[0293] If the system energy consumption increment Exceeding the allowable threshold for energy consumption increase If the system energy consumption is significantly increased, the energy consumption compliance mark will be invalidated. The value is assigned to 0;

[0294] By summarizing quality improvement indicators and energy consumption compliance indicators, it is determined whether the enhanced predictive tension and pressure control is maintained or has failed. Combining both quality and energy consumption indicators, it is determined whether the predictive control has sustainable operational value.

[0295] ;

[0296] In the formula, To maintain the deep enhancement predictive tension and pressure control flag, a value of 1 indicates that the deep enhancement predictive tension and pressure control is maintained, while a value of 0 indicates that the deep enhancement predictive tension and pressure control has failed and a rollback is triggered.

[0297] If the quality improvement mark Equal to 1 and energy consumption qualified mark If the value is 1, it indicates that the bubble rate and wrinkle rate are further reduced compared to the thickness-sensing partition pressure equalization damping collaborative judgment, and the system energy consumption does not increase significantly. This indicates that the depth-enhanced predictive tension and pressure control maintains its indicator. Assign a value of 1 and execute the optimal parameter cloud knowledge graph back transmission;

[0298] If the quality improvement mark Equal to 0 or energy consumption qualified mark If the value is 0, it indicates that the quality indicator has not improved or the prediction deviation is too large, resulting in excessive energy consumption. This underscores the need for enhanced predictive tension and pressure control to maintain the indicator. The value is assigned to 0, and rollback triggering and DRL local self-learning update are executed;

[0299] Execute optimal parameters and feed back to the cloud-based knowledge graph; when deep reinforcement predicts tension and pressure control maintains the flag. When the value equals 1, the optimal action sequence and operating condition features of the current control cycle are extracted, encapsulated into knowledge nodes, and transmitted back. The verified optimal control parameters and operating condition features are then encapsulated and transmitted back to the cloud, supporting the continuous evolution of cross-device group knowledge graphs.

[0300] ;

[0301] In the formula, This refers to the vector of knowledge nodes in the cloud. This is the optimal action sequence, i.e., the predicted sequence. and The input is a deep reinforcement learning agent, which, after passing through a policy network, outputs a sequence of actions from the DRL agent that improves quality and meets energy consumption requirements in the future multi-cycle pressure zone adjustment increment and compensation roller speed correction sequence. , This is a feature vector representing the operating conditions, including paper batch, film type, thickness contour features, and speed fluctuation features. The comprehensive score is calculated by weighting the improvement in bubble rate and wrinkle rate with the increase in energy consumption. It is the transpose operator;

[0302] Execution rollback triggering and DRL local self-learning update, when deep reinforcement predicts tension pressure control maintains the flag. When the value equals 0, rollback and model update are performed, triggering the rollback of the deep reinforcement prediction tension and pressure control to the thickness-aware partition equal pressure damping collaborative judgment, and using the prediction bias as training samples to drive the local update of the DRL policy network to correct the model bias:

[0303] ;

[0304] ;

[0305] ;

[0306] In the formula, This is the rollback trigger flag; a value of 1 indicates an immediate rollback to the basic collaborative control mode based on the thickness-sensing partition equalization damping collaborative judgment. Let DRL be the local self-learning loss function. For the first The actual observed value of the paper feed speed at any given moment. The reward adjustment weight coefficient is determined by the policy network update rate. For the first The actual reward value for each control cycle is calculated based on the measured results of quality and energy consumption. For policy network parameters gradient operator, Set the DRL update completion flag;

[0307] If the gradient converges, the local self-learning update of the DRL model is considered complete. The DRL update completion indicator is... The value is assigned to 1;

[0308] If the gradient does not converge, then the DRL update is considered complete during training. Set the value to 0 and continue gradient descent until convergence;

[0309] The decision to re-enter the deep enhancement predictive tension and pressure control loop is based on the DRL update completion flag. This ensures that the DRL model completes bias correction before re-entering deep enhancement predictive tension and pressure control, avoiding repeated failures and forming a closed-loop evolution mechanism.

[0310] ;

[0311] In the formula, To re-enter the target pressure zone based on the revised target pressure, the closed-loop control adjustment flag is set for each pressure zone.

[0312] If the DRL update is complete, the flag indicates that the update is complete. If the value equals 1, it indicates that the local self-learning update of the DRL model has been completed, and the system re-enters the closed-loop control adjustment process for each pressure zone based on the corrected target pressure. Assigning a value of 1 clears the historical window for depth-enhanced predictive tension pressure control and re-enters the closed-loop control adjustment of each pressure zone based on the corrected target pressure, and performs timing data acquisition and prediction model readiness determination.

[0313] If the DRL update is complete, the flag indicates that the update is complete. If the value is 0, the model is considered to be still in training, and the process re-enters the stage of closed-loop control adjustment for each pressure zone based on the corrected target pressure. The value is assigned to 0, and the system will wait for the update to complete before re-evaluating.

[0314] Parameter updates are achieved through historical data modeling or learning algorithms.

[0315] In this embodiment, by coordinating full-width laser scanning with a multi-zone pressure array, differentiated pressure equalization is achieved between thick and thin areas of the ink layer, eliminating bubbles and viscous adhesion. By coupling passive attenuation of multi-stage floating rollers with active micro-tension feedforward compensation, tension fluctuation blocking and longitudinal wrinkle suppression are achieved. By coordinating LSTM prediction with deep reinforcement learning pre-execution, pre-disturbance prediction and control are achieved, reducing quality fluctuations and energy consumption. By linking execution feedback with DRL self-learning updates, strategy evolution and cloud-based graph optimization are achieved, improving coating stability.

Claims

1. An adaptive control process for environmentally friendly lamination of packaging boxes, characterized in that: include: Before the paper enters the hot-pressing lamination zone, the paper thickness distribution data is obtained by full-width scanning, the average thickness and the maximum thickness difference rate of the full width are calculated, and the degree of thickness variation is judged accordingly. Based on the pressure zoning structure of the hot press roller, the thickness distribution data is divided into multiple zones. The deviation of each zone from the average thickness of the whole width is calculated, and each zone is determined to be a thick zone, a thin zone, or a normal zone. Based on the thickness deviation of each zone, differential adjustment is made on the basis of the basic composite pressure, so that the pressure corresponding to the thick zone is increased and the pressure corresponding to the thin zone is decreased, and the adjustment range increases nonlinearly with the increase of thickness deviation. During the lamination process, the tension at the composite point and its changing trend are acquired in real time to characterize the current tension fluctuation state and perform tension state detection. Based on the tension fluctuation state, the target pressure of each zone is adjusted in a coordinated manner, and pressure-tension coupling compensation is performed so that the disturbance caused by tension fluctuation is compensated by pressure regulation, thereby realizing the synergistic compensation relationship between the pressure demand caused by thickness difference and tension disturbance. Based on the revised target pressure, closed-loop control is applied to each pressure zone to gradually converge the actual pressure to the target pressure range. The bubble rate and wrinkle rate after lamination are obtained through online detection and compared with preset thresholds to determine the quality status of lamination. When thickness fluctuations, tension fluctuations, or abnormal quality indicators occur, the pressure regulation and coupling compensation parameters are adaptively updated based on historical operating data.

2. The adaptive control process for environmentally friendly film coating of packaging boxes according to claim 1, characterized in that: Calculate the average thickness and maximum thickness variation rate across the entire width, and determine the degree of thickness variation accordingly. Before the paper enters the hot-pressing lamination area, at a fixed scanning station, the laser line scanning sensor performs a full-width scan along the paper axis at equal intervals, collects the thickness values ​​of discrete sampling points, and constructs a full-width thickness distribution column vector. The thickness digital matrix is ​​constructed by densely sampling along the width of the paper, providing the original data basis for ink layer thickness area identification and difference calculation. Based on the thickness distribution column vector, the average thickness, maximum thickness and minimum thickness of the whole area are calculated to form the benchmark statistics for thickness fluctuation analysis. The mean and extreme values ​​of thickness are extracted to quantify the overall thickness level and absolute fluctuation range, providing a benchmark reference for the calculation of the difference rate. Using the average thickness as a benchmark, the ratio of the maximum thickness difference to the average thickness is calculated to obtain the normalized maximum thickness difference rate. This eliminates the dimensional influence caused by different paper basis weights, normalizes the absolute thickness difference to the relative difference rate, eliminates the dimensional interference of paper basis weights, and makes different batches of paper comparable. The maximum thickness difference rate is compared with a preset benchmark threshold, and a Boolean variability risk flag is output as a switch to activate adaptive control or maintain the normal mode. If the maximum thickness difference rate is greater than the preset benchmark threshold, it is determined that there is local thickness unevenness in the paper, and the difference between the thick and thin areas of the ink layer is significant. At this time, the variability risk flag is assigned a value of 1, and the thickness distribution data is divided into multiple zones according to the pressure zoning structure of the hot press roller. The deviation of each zone relative to the average thickness of the whole width is calculated, and the thickness sensing-zone pressure equalization-multi-level tension damping coordinated control is started. If the maximum thickness difference rate is less than or equal to the preset benchmark threshold, it is determined that the paper thickness distribution is uniform and there is no risk of pressure mismatch. At this time, the variability risk flag is set to 0, and the normal constant pressure and constant tension lamination mode is maintained without starting adaptive control. The thickness distribution data was obtained through laser line scanning.

3. The adaptive control process for environmentally friendly film coating of packaging boxes according to claim 1, characterized in that: Based on the pressure zoning structure of the hot press roller, the thickness distribution data is divided into multiple zones, and the deviation of each zone relative to the average thickness of the entire width is calculated, specifically: The full-width thickness distribution column vector is sliced ​​at equal intervals according to the number of independent pressure zones along the axial direction of the hot press roller to obtain the zone thickness sub-vector. The full-width thickness data is sliced ​​according to the pressure zones to establish the axial mapping relationship between the sampling points and the pressure zones. Calculate the average thickness of each partition sub-vector and compare it with the average thickness of the whole area to determine the regional attributes of each partition. Calculate the average thickness of each partition and compare it with the overall mean to quantitatively identify thick, thin and normal ink layer areas. If the area attribute flag is output as 1, then the partition is determined to be a thick ink layer area; If the region attribute flag is output as -1, then the partition is determined to be a thin region; If the area attribute flag is 0, the partition is determined to be a normal area; Perform sampling point missing rate verification and adjacent sampling point thickness jump detection on each partition sub-vector to detect sampling missing and data jump, and remove invalid data caused by laser occlusion or reflection interference; Map the attribute markers and axial coordinates of each partition area to a thickness cloud map index table, and make output judgments based on data integrity and abnormal jump detection results. Integrate partition attributes and coordinates to generate an index table, and decide whether to proceed to the next step or re-collect data based on the verification results. If the data validity flag is set to 1, it is determined that the axial coordinate positions of the thick and thin areas of the ink layer in each partition have been effectively resolved, the thickness difference data is complete and there are no abnormal jumps. At this time, a thickness cloud map index table is generated and differential adjustment is made based on the thickness deviation of each partition on the basis of the basic composite pressure. If the data validity flag is set to 0, it is determined that the parsing failed or there is a data anomaly. At this time, the sensor self-cleaning or calibration program is triggered. After the calibration is completed, the thickness distribution data is re-divided into multiple partitions according to the pressure partitioning structure of the hot press roller, and the deviation of each partition relative to the average thickness of the whole width is calculated. If the data validity flag is still 0 after three consecutive re-executions, the sensor is determined to be continuously faulty, triggering an alarm and shutdown. The partitions correspond one-to-one with the independent axial pressure partitions of the hot press roller.

4. The adaptive control process for environmentally friendly film coating of packaging boxes according to claim 1, characterized in that: Based on the thickness deviation of each zone, differentiated adjustments are made on the basis of the basic composite pressure, specifically as follows: Using the regional attribute flags of each partition in the thickness cloud map index table as input, and combining the basic composite pressure, the pressure increment of the ink layer thick area compensation and the pressure reduction of the thin area, the target pressure of each partition is analyzed. Based on the thickness cloud map, the basic pressure is mapped to the differentiated target value of each partition, and a benchmark reference for pressure equalization is established. If the area attribute flag is equal to 1, the area is determined to be a thick ink layer area, and the pressure correction amount is assigned the first parameter value. If the area attribute flag is equal to -1, the area is determined to be a thin area, and the pressure correction amount is assigned the value of the second parameter. If the area attribute flag is equal to 0, the zone is determined to be a normal zone, and the pressure correction amount is assigned the value of the third parameter. The actual pressure in each zone is collected in real time by pressure sensors, and the pressure deviation of each zone is calculated to quantify the degree of deviation between the actual pressure and the target pressure in each zone, providing a basis for subsequent adjustment direction and magnitude. For the ink layer thick area corresponding to the area with the regional attribute flag equal to 1, pressure increase judgment is performed, and undervoltage detection and step pressure increase are performed on the ink layer thick area to ensure that the adhesive layer penetrates the ink layer to achieve full activation. If the pressure deviation of a zone is greater than the pressure adjustment dead zone threshold, it is determined that the actual pressure of the zone has not reached the compensation pressure required to penetrate the ink layer, and there is a risk of insufficient activation of the adhesive layer. The pressure increase judgment flag is set to 1, and pressure increase adjustment is executed. The pressure command value is increased by the single pressure adjustment step size based on the current actual pressure. If the pressure deviation of a zone is less than or equal to the pressure regulation dead zone threshold, it is determined that the zone pressure has reached the target, the pressure increase judgment flag is set to 0, and the pressure command value remains unchanged at the current actual pressure. For the corresponding partitions of thin areas or inkless areas with a region attribute flag of -1, pressure reduction judgment is performed, and overpressure detection and step pressure reduction are performed on thin areas to prevent paper crushing and glue overflow defects. If the current actual pressure is greater than the paper pressure limit parameter, or the pressure deviation of the zone is less than the negative pressure adjustment dead zone threshold, then it is determined that the actual pressure of the zone exceeds the paper pressure limit or is higher than the target pressure, and there is a risk of adhesive layer overflow or paper crushing. The pressure reduction judgment flag is assigned a value of 1, and pressure reduction adjustment is performed. The pressure command value is reduced by the single pressure adjustment step size based on the current actual pressure. If the current actual pressure is less than or equal to the upper limit parameter of paper pressure and the pressure deviation of the zone is greater than or equal to the negative pressure adjustment dead zone threshold, then the zone pressure is determined to be in the safe range and the target has been reached. The pressure reduction judgment flag is set to 0, and the pressure command value remains unchanged at the current actual pressure. Summarize the adjustment status of all partitions, determine whether the partition voltage equalization adaptation is complete, summarize the adjustment requirements of all partitions, determine whether the voltage equalization adaptation has converged globally, and control the process to proceed or loop. If the convergence flag for partition pressure equalization is 1, it is determined that the pressure of all partitions has matched the thickness contour map requirements, and tension state detection is performed. If the partition pressure equalization and adaptation convergence flag output is 0, it is determined that at least one partition has not yet completed pressure matching. The current actual pressure is collected in real time through the pressure sensors of each partition, and the pressure deviation of each partition is calculated until the partition pressure equalization and adaptation convergence flag output is 1. The pressure regulation employs a nonlinear regulation strategy, which increases the pressure variation amplitude as the thickness deviation increases.

5. The adaptive control process for environmentally friendly film coating of packaging boxes according to claim 1, characterized in that: Tension state detection is performed, specifically as follows: The real-time position offset of the first floating roller relative to the mechanical balance position is obtained, its absolute value is calculated and compared with the coarse adjustment absorption threshold, and it is detected whether the real-time position offset of the first floating roller exceeds the large value fluctuation absorption threshold. The first-stage passive damping is determined to be saturated and the subsequent active compensation requirement is triggered. If the absolute value of the position offset is greater than the coarse adjustment absorption threshold of the first floating roller, it is determined that the offset of the first floating roller exceeds the absorption range of large fluctuation, and the coarse adjustment over-limit flag is assigned a value of 1. If the absolute value of the position offset is less than or equal to the coarse adjustment absorption threshold of the first floating roller, the first floating roller is determined to be within the coarse adjustment absorption range, and the coarse adjustment over-limit flag is assigned a value of 0. Obtain the position offset sequence of the second floating roller within the sliding time window, calculate its residual amplitude and compare it with the fine-tuning attenuation threshold, calculate the residual amplitude of the second floating roller within the sliding time window, and determine whether the low inertia fine-tuning damping is sufficient to attenuate high-frequency disturbances. If the residual amplitude of the second floating roller is greater than the fine-tuning attenuation threshold of the second floating roller, it is determined that there is an unattenuated high-frequency residual disturbance in the second floating roller, and the fine-tuning over-limit flag is assigned a value of 1. If the residual amplitude of the second floating roller is less than or equal to the fine-tuning attenuation threshold of the second floating roller, it is determined that the high-frequency disturbance of the second floating roller has been sufficiently attenuated, and the fine-tuning over-limit flag is assigned a value of 0. Obtain the current paper feed speed, calculate its time derivative value and compare it with the fluctuation threshold, identify speed abrupt changes and trigger micro-tension feedforward compensation; If the absolute value of the speed derivative is greater than the paper feed speed derivative fluctuation threshold, it is determined that there is a sudden fluctuation in the paper feed speed, and the speed fluctuation flag is assigned a value of 1. If the absolute value of the speed derivative is less than or equal to the paper feed speed derivative fluctuation threshold, the paper feed speed is determined to be stable, and the speed fluctuation flag is assigned a value of 0. The real-time roll gap height is collected by the hot press roll bearing seat displacement sensor, the roll gap height change is calculated and compared with the allowable threshold, the roll gap height change caused by the zone pressure adjustment is monitored, and it is determined whether the film linear speed and paper feed speed at the composite point are mismatched. If the absolute value of the change in roll gap height is greater than the allowable threshold for roll gap height change, it is determined that the change in roll gap height causes the film linear speed and the paper feed speed to be out of sync, and the roll gap mismatch flag is assigned a value of 1. If the absolute value of the change in roll gap height is less than or equal to the allowable threshold for roll gap height change, the roll gap height is determined to be in the synchronization range, and the roll gap mismatch flag is assigned a value of 0. The system summarizes the over-limit indicators of damping at all levels, and determines whether to simultaneously activate active micro-tension feedforward compensation and roll gap height feedforward correction through logical OR operation, and simultaneously outputs compensation roll speed correction and unwinding angular velocity feedforward commands. If the damping coordination start flag is equal to 1, it is determined that at least one level of damping is over the limit, and active micro-tension feedforward compensation and roll gap height feedforward correction need to be started simultaneously. If the damping coordination start flag is equal to 0, it is determined that all damping parameters are in the stable range and there is no need to start active compensation. The real-time tension at the composite point is obtained, the deviation between the real-time tension at the composite point and the target tension is calculated, the deviation between the real-time tension at the composite point and the target tension is detected, and it is determined whether the tension fluctuation after the coordination of each level of damping converges to the allowable threshold. If the absolute value of the tension deviation is less than or equal to the allowable threshold for tension fluctuation, then the tension fluctuation is determined to be within the allowable range, the multi-level tension damping is coordinated and stable, and the tension stability flag is assigned a value of 1. If the absolute value of the tension deviation is greater than the allowable threshold for tension fluctuation, the tension fluctuation is determined to be excessive, the tension stability flag is set to 0, the real-time position offset of the first floating roller relative to the mechanical balance position is reacquired, its absolute value is calculated and compared with the coarse adjustment absorption threshold, until the tension stability flag is equal to 1. Among them, tension fluctuation and its rate of change are characterized by the real-time tension value and its rate of change.

6. The adaptive control process for environmentally friendly film coating of packaging boxes according to claim 1, characterized in that: Based on the tension fluctuation state, the target pressure of each zone is adjusted in a coordinated manner, and pressure-tension coupling compensation is performed, specifically as follows: The bubble rate statistics of the online visual inspection system after lamination are obtained, and the bubble rate is compared with the first-level qualified threshold. The proportion of bubble defects in the composite layer is quantified by visual inspection to determine whether the local non-adhesion caused by uneven thickness has been effectively eliminated. If the bubble rate is greater than the first-level bubble rate qualification threshold, the bubble rate after compounding is determined to be excessive, and the bubble rate excess flag is assigned a value of 1. If the bubble rate is less than or equal to the first-level bubble rate qualification threshold, the bubble rate is determined to be within the qualified range, and the bubble rate excess flag is assigned a value of 0. The wrinkle rate statistics of the online visual inspection system after composite are obtained, and the wrinkle rate is compared with the first-level qualified threshold. The proportion of longitudinal deformation defects of the film is quantified by visual inspection to determine whether the wrinkles caused by tension fluctuations have been effectively suppressed. If the wrinkle rate is greater than the first-level wrinkle rate qualification threshold, the wrinkle rate after composite is determined to be excessive, and the wrinkle rate excess flag is assigned a value of 1. If the wrinkle rate is less than or equal to the first-level wrinkle rate qualification threshold, the wrinkle rate is determined to be in the qualified range, and the wrinkle rate exceeding the limit flag is assigned a value of 0. The system summarizes the indicators of excessive bubble rate and excessive wrinkle rate, and uses logical operations to determine whether the thickness sensing zone equal pressure damping collaborative judgment has been passed. Combining the two quality indicators of bubbles and wrinkles, it determines whether the three-level collaborative control effect of the thickness sensing zone equal pressure damping collaborative judgment meets the threshold for progressively entering the depth enhancement predictive tension pressure control. If the bubble rate exceeding the limit flag is equal to 0 and the wrinkle rate exceeding the limit flag is equal to 0, that is, both the bubble rate and the wrinkle rate are within the qualified range, it is determined that the basic three-level coordinated control has effectively solved the problem of uneven thickness and tension fluctuation. The thickness sensing zone pressure equalization damping coordinated judgment is assigned a value of 1 through the flag, and closed-loop control adjustment is performed on each pressure zone according to the corrected target pressure. If the bubble rate exceeds the limit flag is equal to 1 or the wrinkle rate exceeds the limit flag is equal to 1, that is, at least one of the bubble rate and wrinkle rate exceeds the limit, then it is determined that the quality still does not meet the standard after the thickness sensing partition equal pressure damping collaborative judgment is executed, and the thickness sensing partition equal pressure damping collaborative judgment pass flag is assigned to 0. When the thickness-sensing partition pressure equalization damping collaborative judgment pass flag equal to 0, it triggers a return based on the thickness deviation of each partition, and performs differentiated adjustment and readjustment based on the basic composite pressure, and records the number of iterations, and simultaneously expands the pressure adjustment step size, and gradually eliminates residual bubbles and viscous defects through incremental pressure correction. The iteration adjustment counter is compared with the maximum allowed number of iterations to determine whether an alarm shutdown is triggered, and the upper limit of the automatic iteration adjustment number is limited to prevent infinite loop. If the target is not met after three iterations, the system will switch to manual intervention mode in time. If the iterative adjustment counter is greater than 3, it is determined that the thickness sensing zone equal pressure damping coordination judgment still fails after three consecutive return adjustments. The iterative over-limit alarm flag is set to 1, triggering an alarm prompt for manual intervention. If the iterative adjustment counter is less than or equal to 3, it is determined that the number of iterations has not exceeded the limit, the iterative over-limit alarm flag is set to 0, and the process returns to the differential adjustment based on the thickness deviation of each zone and the basic composite pressure, and carries the amplified pressure increment and pressure reduction to re-execute the zone equalization adaptation, and performs tension state detection again until the thickness sensing zone equalization damping collaborative judgment pass flag is equal to 1. The pressure correction amount is dynamically adjusted according to the tension fluctuation range.

7. The adaptive control process for environmentally friendly film coating of packaging boxes according to claim 1, characterized in that: Based on the revised target pressure, closed-loop control adjustment is performed on each pressure zone, specifically as follows: During the stable operation phase of thickness-sensing partitioned pressure equalization and damping collaborative determination, paper feed speed, composite point tension, partitioned average pressure and roll gap height are continuously collected at a fixed sampling period to construct a multi-dimensional time series data matrix and verify the window length, providing a sample basis that meets the input dimension requirements for the LSTM model; If the current historical data window length is greater than or equal to the minimum historical window length required by the LSTM prediction model, then the historical data window length is determined to meet the input requirements of the prediction model, and the window length compliance flag is assigned a value of 1. If the current historical data window length is less than the minimum historical window length required by the LSTM prediction model, then the data window is deemed insufficient, and the window length compliance flag is set to 0. Perform missing rate statistics on each dimension of the time series data matrix, and calculate the proportion of missing sampling points in each dimension to identify data incompleteness caused by sensor communication interruption or packet loss. If the maximum missing rate is greater than 0, then the data is determined to be missing, and the missing data flag is assigned a value of 1; If the maximum missing rate is 0, the data is considered complete and without missing data, and the missing data flag is set to 0. The data repair strategy is determined based on the degree of missing data, and interpolation repair or extended collection is selected based on the missing rate threshold to achieve a balance between data integrity and model timeliness. If the maximum missing rate is greater than 0 and less than or equal to the maximum missing rate threshold allowed by interpolation, then the degree of missing data is determined to be within the interpolation range, the interpolation feasibility flag is set to 1, and linear interpolation or spline interpolation is performed to complete the missing data. After completion, the missing rate statistics are re-performed on each dimension of the time series data matrix. If the maximum missing rate is greater than the maximum missing rate threshold allowed for interpolation completion, it is determined that the missing degree is too severe to reliably interpolate, the interpolation feasibility flag is set to 0, the data collection time is extended until the window length meets the requirements again, the multi-dimensional time series data matrix is ​​reconstructed and the window length is verified. Obtain the current paper batch identifier, film type identifier, and paper joint position mark, compare them with the baseline value when the historical window was initially established, monitor batch, film, and joint signals, and forcibly clear the historical window when the operating conditions change abruptly to prevent old operating condition data from contaminating the prediction of new operating conditions. If the current paper batch identifier is not equal to the paper batch identifier when the history window was initially created, then it is determined that a paper batch switch has been detected, and the paper batch switch flag is set to 1. If the current paper batch identifier is equal to the paper batch identifier when the history window was initially created, then the paper batch switching flag is set to 0. If the current membrane material type identifier is not equal to the membrane material type identifier when the history window was initially established, it is determined that a membrane material change has been detected, and the membrane material change flag is assigned a value of 1. If the current membrane material type identifier is equal to the membrane material type identifier when the history window was initially created, then the membrane material replacement flag is assigned a value of 0; If the paper splice position mark is equal to 1, then a paper splice is detected, and the splice detection flag is set to 1. If the paper joint position mark is equal to 0, then the joint detection mark is assigned a value of 0; If the paper batch switching flag is equal to 1, the film material replacement flag is equal to 1, or the joint detection flag is equal to 1, then it is determined that a sudden change in the working condition has occurred. The working condition change flag is assigned a value of 1, the history window is cleared, and data is collected again. If the paper batch switching flag is equal to 0, the film material replacement flag is equal to 0, and the joint detection flag is equal to 0, then the working condition is considered stable, and the working condition change flag is assigned a value of 0. The system summarizes three criteria: window length, data integrity, and operational stability. It then performs a prediction model readiness check and, based on these three criteria, decides whether to activate the LSTM model or perform corresponding data repair and reconstruction operations. If the window length meets the standard flag, the data missing flag is equal to 0, and the working condition change flag is equal to 0, then it is determined that the historical data window length meets the requirements, the data is complete and without missing data, and the working condition has not changed. The prediction model ready flag is assigned to 1, the LSTM prediction model is activated, and the bubble rate and wrinkle rate after film coating are obtained through online detection and compared with the preset threshold to determine the film coating quality status. If the operating condition change flag is equal to 1, it is determined that an operating condition change has occurred. The historical window is cleared and data is collected again. After the window is rebuilt, closed-loop control adjustment of each pressure zone is performed according to the corrected target pressure. If the window length meets the target and the operating condition change flag is 0, the data window is deemed insufficient. The data acquisition time is extended until the window length meets the target. Then, the closed-loop control adjustment of each pressure zone is re-executed based on the corrected target pressure. If the window length meets the standard flag, the data missing flag is equal to 1, the interpolation is feasible flag is equal to 1, and the change in operating conditions flag is equal to 0, then data interpolation completion is performed, and after completion, the missing rate statistics are re-performed on each dimension of the time series data matrix. If the window length meets the standard flag, the data missing flag is equal to 1, the interpolation is feasible flag is equal to 0, and the operating condition change flag is equal to 0, then the missing degree is determined to be too severe to reliably interpolate. The data acquisition time is extended, and after the data is complete, the closed-loop control adjustment of each pressure zone is re-executed based on the corrected target pressure. The closed-loop control adopts an iterative adjustment method based on pressure deviation and sets an adjustment dead zone to avoid frequent fluctuations.

8. The adaptive control process for environmentally friendly film coating of packaging boxes according to claim 1, characterized in that: The bubble rate and wrinkle rate after lamination are obtained through online detection and compared with preset thresholds to determine the lamination quality status. Specifically: Based on historical time series data windows, and through forward propagation of the LSTM model, multidimensional prediction sequences for multiple future control cycles are output. Based on historical windows and through forward propagation of the LSTM model, prediction sequences for velocity and thickness distribution in multiple future cycles are output, providing a data basis for disturbance prediction. Calculate the maximum step change in the velocity prediction sequence and the maximum difference rate in the thickness prediction sequence to determine whether a deterministic perturbation mode exists and decide whether to activate the agent. If the maximum step change in the velocity prediction sequence is greater than the velocity step disturbance threshold, or the maximum difference rate of the thickness prediction within the period is greater than the thickness variation disturbance threshold, or the prediction joint or batch switching flag is equal to 1, then the prediction sequence is determined to show a deterministic disturbance pattern, and the disturbance significance flag is assigned a value of 1. If the maximum step change in the velocity prediction sequence is less than or equal to the velocity step disturbance threshold, and the maximum difference rate of the thickness prediction within the period is less than or equal to the thickness variation disturbance threshold, and the prediction joint or batch switching flag is equal to 0, then the prediction trend is determined to be stable without significant disturbance, and the disturbance significance flag is assigned a value of 0. If the perturbation significance flag is assigned a value of 1, the predicted sequence will be input into the deep reinforcement learning agent, and the policy network will output the future multi-cycle pressure zone adjustment increment and compensation roller speed correction sequence. The controller preloads the action sequence sequentially, and accumulates the correction amount cycle by cycle based on the current actual value to generate pre-executed control commands, thereby realizing predictive control output before disturbances occur. If the disturbance significance flag is set to 0, then the stable operating condition standby judgment is used. When the predicted trend is stable and there is no significant disturbance, the deep enhancement prediction tension pressure control enters low power standby monitoring. The thickness sensing partition equal pressure damping collaborative judgment maintains the basic collaborative regulation to save computing power. If the disturbance significance flag is equal to 0, it is determined that the current working condition does not require predictive intervention. The activation status flag of the deep enhanced predictive tension and pressure control is assigned to 0, and the deep enhanced predictive tension and pressure control enters the low-power standby monitoring state. The basic coordinated regulation is maintained by the thickness sensing partition equalization damping collaborative determination. If the disturbance significance flag is equal to 1, it is determined that the bubble rate and wrinkle rate after coating need to be obtained through online detection and compared with the preset threshold to determine the coating quality status. The depth enhancement prediction tension pressure control activation status flag is assigned to 1. Before the disturbance actually occurs, the controller outputs the preload action sequence in sequence, and accumulates the correction amount on the basis of the current actual value cycle by cycle to generate the pre-execution command generated by the pre-execution control command, so as to realize prediction as control. The bubble rate and wrinkle rate were obtained through an online visual inspection system.

9. The adaptive control process for environmentally friendly film coating of packaging boxes according to claim 1, characterized in that: When thickness fluctuations, tension fluctuations, or abnormal quality indicators occur, the pressure regulation and coupling compensation parameters are adaptively updated based on historical operating data, specifically as follows: The bubble rate, wrinkle rate and system energy consumption after the execution of the deep enhancement predictive tension and pressure control are obtained. At the same time, the bubble rate benchmark value, wrinkle rate benchmark value and energy consumption benchmark value stored in the thickness sensing partition equalization damping collaborative judgment stage are read. The mass and energy consumption parameters after the execution of the deep enhancement predictive tension and pressure control are collected simultaneously, and the benchmark value of the thickness sensing partition equalization damping collaborative judgment stage is called to provide a comparison benchmark for the improvement magnitude calculation. Calculate the improvement in bubble rate and wrinkle rate, and compare them with the improvement threshold to determine whether predictive control brings about substantial quality improvement. If the improvement in bubble rate is greater than or equal to the bubble rate improvement judgment threshold and the improvement in wrinkle rate is greater than or equal to the wrinkle rate improvement judgment threshold, then the bubble rate and wrinkle rate are both determined to be further reduced compared to the thickness sensing partition equal pressure damping collaborative judgment when running alone, and the quality improvement flag is assigned a value of 1. If the improvement in bubble rate is less than the bubble rate improvement threshold or the improvement in wrinkle rate is less than the wrinkle rate improvement threshold, then the quality indicator is determined to have not improved or only a single improvement fails to meet the overall requirements, and the quality improvement flag is assigned a value of 0. The system energy consumption increment of the depth-enhanced predictive tension pressure control relative to the thickness-sensing partitioned equal pressure damping collaborative judgment is calculated and compared with the allowable threshold for energy consumption increase. This quantifies the energy consumption cost of predictive control and ensures that quality improvement does not come at the cost of excessive energy consumption. If the increase in system energy consumption is less than or equal to the allowable threshold for energy consumption increase, then the predictive control is determined not to introduce a significant additional energy consumption burden, and the energy consumption qualification flag is assigned a value of 1. If the increase in system energy consumption exceeds the allowable threshold for energy consumption increase, the system energy consumption is determined to have increased significantly, and the energy consumption qualification flag is assigned a value of 0. By summarizing quality improvement indicators and energy consumption compliance indicators, it is determined whether the in-depth enhanced predictive tension and pressure control is maintained or failed. By combining the two indicators of quality and energy consumption, it is determined whether the predictive control has sustainable operating value. If the quality improvement flag is equal to 1 and the energy consumption qualification flag is equal to 1, then it is determined that the bubble rate and wrinkle rate are further reduced compared with the thickness sensing partition pressure equalization damping collaborative judgment and the system energy consumption has not increased significantly. The deep enhancement prediction tension pressure control maintenance flag is assigned to 1, and the optimal parameter cloud knowledge graph is returned. If the quality improvement flag is equal to 0 or the energy consumption compliance flag is equal to 0, it is determined that the quality indicators have not improved or the prediction deviation is too large, resulting in excessive energy consumption. The deep enhancement prediction tension and pressure control maintenance flag is assigned to 0, and rollback triggering and DRL local self-learning update are executed. The optimal parameters are fed back to the cloud knowledge graph. When the deep reinforcement predictive tension and pressure control maintenance flag is equal to 1, the optimal action sequence and operating condition features of the current control cycle are extracted, encapsulated as knowledge nodes and fed back. The verified optimal control parameters and operating condition features are encapsulated and fed back to the cloud to support the continuous evolution of cross-device group knowledge graphs. The rollback trigger and DRL local self-learning update are executed. When the deep reinforcement prediction tension and pressure control maintenance flag is equal to 0, the rollback and model update are executed. The deep reinforcement prediction tension and pressure control is triggered to roll back to the thickness-aware partition equal pressure damping collaborative judgment, and the prediction deviation is used as training sample to drive the local update of the DRL policy network to correct the model deviation. If the gradient converges, the local self-learning update of the DRL model is determined to be complete, and the DRL update completion flag is assigned a value of 1. If the gradient does not converge, the DRL update completion flag is set to 0 during training, and gradient descent continues until convergence. Based on the DRL update completion flag, determine whether to re-enter the deep reinforcement predictive tension and pressure control loop, ensuring that the DRL model re-enters deep reinforcement predictive tension and pressure control after completing the deviation correction, avoiding repeated failures, and forming a closed-loop evolution mechanism. If the DRL update completion flag is equal to 1, it is determined that the local self-learning update of the DRL model has been completed. Then, the system will re-enter the process of adjusting the closed-loop control of each pressure zone according to the corrected target pressure, assign the flag to 1, clear the history window of the deep enhancement prediction tension pressure control, and re-enter the process of performing closed-loop control adjustment of each pressure zone according to the corrected target pressure, and determine the readiness of the prediction model. If the DRL update completion flag is equal to 0, it is determined that the model is still in training. The process is to re-enter the closed-loop control adjustment flag for each pressure zone based on the corrected target pressure, set the value to 0, and wait for the update to be completed before re-determining. Parameter updates are achieved through historical data modeling or learning algorithms.

Images