High-stability papermaking dye and performance testing method thereof

Abstract

The invention relates to the technical field of papermaking dye performance testing, and discloses a high-stability papermaking dye and a performance testing method thereof. The method comprises the following steps: collecting dye sample data and assistant influence parameters, and constructing an initial environment variable set; determining a dye dispersion state category by adopting a machine learning clustering algorithm; if the state is aggregation abnormity, migration track data are obtained, and a dynamic process change rule is obtained through analysis; obtaining fiber attachment distribution information based on the rule, analyzing the relevance between the fiber attachment distribution information and auxiliary parameters by using a support vector machine model, and determining migration path optimization points; extracting real-time monitoring indexes from the optimization points, and integrating environmental variable influences by adopting an information fusion technology to obtain a stability quantized value; and if the quantized value is lower than the threshold value, adjusting environment simulation parameters, evaluating the improvement degree, updating the dynamic process model, and generating a real-time monitoring scheme. According to the invention, quantitative tracking and regulation of dye dispersion and migration stability are realized, and the uniformity and stability of the dyeing process are obviously improved.

Description

Technical Field

[0001] This application relates to the field of paper dye performance testing technology, and in particular to a highly stable paper dye and its performance testing method. Background Technology

[0002] In the papermaking industry, to obtain the decorative, identifiable, and functional properties of paper (such as anti-counterfeiting, traceability, and weather resistance), various papermaking dyes are often added to the pulp system or during the surface sizing / coating process. The development path of existing papermaking dyes has generally gone through the following stages: from early single dye systems that focused on hue and color intensity, to high-fastness dyes that take into account comprehensive properties such as washability, lightfastness, and chemical resistance; at the same time, to adapt to high-speed papermaking and complex wet-end chemical conditions, the industry has further developed high-stability dyes and their supporting auxiliary agent systems with the goal of "dispersion stability, controllable migration, and high adhesion efficiency," and has gradually introduced online monitoring and data-driven process control ideas to reduce color difference fluctuations and batch instability.

[0003] However, the wet end of papermaking is a typical multi-component complex system, involving the combined effects of pulp fibers, fillers and fine fibers, retention and filtration aids, sizing agents, reinforcing agents, dispersants / surfactants, metal ions, and water salinity. Furthermore, environmental variables such as temperature, pH, and ionic strength fluctuate significantly with varying operating conditions. In this multi-component environment, dyes are prone to charge shielding, complexation / precipitation, changes in micellar behavior, or particle aggregation, leading to decreased dispersion stability. Simultaneously, the migration and adsorption pathways of dyes on fiber / filler surfaces are synergistically influenced by additives, potentially resulting in a problem of "local enrichment—uneven adsorption—color spots / differences in the finished paper." Current technologies for evaluating the performance of paper dyes largely rely on offline sampling and single-factor index (such as exhaustion rate, color intensity, and residual liquor color) testing. This makes it difficult to reflect the dynamic process of dye dispersion, migration, and adhesion in real-time under a real multi-component environment. Even when online sensing or image detection is introduced, it often remains at the level of phenomenon monitoring. It lacks the ability to couple and analyze fiber adhesion distribution with auxiliary parameters and provide actionable process optimization points and closed-loop correction mechanisms accordingly. This results in lagging process control, reliance on experience for parameter adjustment, and uncertain stability improvement effects.

[0004] Therefore, there is an urgent need for a high-stability paper dye performance testing method for multi-component papermaking environments. This method should be able to identify the dispersion state of dyes under fluctuating operating conditions, perform trajectory analysis of migration behavior under abnormal aggregation conditions, and further establish a correlation between fiber surface adhesion distribution and auxiliary agent influence parameters to extract key parameter combinations that can be used for process control. At the same time, it is also necessary to form a stability quantification index and trigger parameter adjustment, assess the degree of improvement, and update the model when the index is below a threshold. This would enable real-time, quantitative, and closed-loop testing and monitoring of the dispersion and migration stability of paper dyes, thereby improving dye adhesion uniformity and papermaking process stability. Summary of the Invention

[0005] To address the problems of dispersion instability, abnormal aggregation, uncontrollable migration paths, uneven fiber adhesion, and color difference fluctuations in existing paper dyes in multi-component wet-end environments due to fluctuations in temperature, pH, ionic strength, water quality ions, and auxiliary agent systems, as well as the shortcomings of existing performance tests which are mostly offline single-index evaluations, unable to reflect the entire "dispersion-migration-adhesion" process in real time and lacking a closed-loop optimization mechanism, this application provides a highly stable paper dye and its performance testing method, enabling real-time monitoring, quantitative evaluation, and adaptive optimization of dye behavior stability.

[0006] In a first aspect, this application provides a highly stable papermaking dye, comprising: The dye matrix has a chemical structure that is stable in high temperature and acid / alkali environments, and is selected from organic dyes, inorganic dyes or combinations thereof. The dye matrix has good dispersibility. At least one stabilizer that enhances the dye’s tolerance to fluctuations in water quality, temperature and pH during the papermaking process. The stabilizer is selected from polymer stabilizers, surfactant stabilizers, or a combination of polymer stabilizers and surfactant stabilizers. At least one auxiliary agent, which can adjust the adhesion properties and color fastness of the dye, and can adjust the solubility of the dye in acidic or alkaline environments to ensure the stable migration and uniform distribution of the dye in the papermaking process.

[0007] Secondly, this application provides a performance testing method for highly stable papermaking dyes, the method comprising: S1. Collect paper dye sample data and auxiliary influence parameters in a multi-component papermaking environment, extract environmental variable characteristics and related influence characteristics from the collected data, and obtain an initial environmental variable set; S2. Use machine learning clustering algorithms to classify the initial set of environmental variables and determine the dye dispersion state category; S3. When the dye dispersion state category indicates abnormal aggregation, acquire migration behavior trajectory data and analyze the dynamic process change law; S4. Obtain fiber adhesion distribution information based on the dynamic process change law, analyze the correlation between fiber adhesion distribution information and auxiliary agent influence parameters through support vector machine model, and determine the migration path optimization point; S5. Extract real-time monitoring indicators from migration path optimization points, and use information fusion technology to integrate the impact of environmental variables on real-time monitoring indicators to obtain the stability quantification value of paper dye stability. S6. If the stability quantification value is lower than the preset stability threshold, the simulation parameters of the multi-component environment of papermaking are adjusted, and the improvement of dye dispersion and migration behavior after parameter adjustment is evaluated. S7. Update the dynamic process model based on the improvement level data, and generate the final real-time monitoring scheme based on the updated model. The dynamic process model is used to simulate the dispersion and migration behavior of dyes in the multi-component environment of papermaking.

[0008] Compared with the prior art, the beneficial effects of the technical solution of this application are at least as follows: 1. By simultaneously introducing environmental variable characteristics and auxiliary agent-related influence characteristics in a multi-component papermaking environment, a multi-dimensional feature dataset is constructed. Through clustering to identify the dispersion state, the rapid determination of the abnormal transition of dyes from normal dispersion to aggregation is achieved, which improves the adaptability and sensitivity of the test evaluation to fluctuations in operating conditions.

[0009] 2. When clustering anomalies occur, the dynamic process change pattern is obtained based on the time series analysis of migration trajectory and the matching of dynamic feature library, realizing a structured characterization of the dynamic behavior of dye migration, providing interpretable basis for subsequent attachment distribution collection and migration path diagnosis, and reducing the uncertainty of traditional experience judgment.

[0010] 3. By establishing a correlation between fiber surface adhesion distribution information and auxiliary influence parameters through a support vector machine model and extracting key parameter combinations, the migration path optimization points are obtained. This allows performance test results to be directly transformed into executable process control basis, thereby improving dye adhesion uniformity and process stability.

[0011] 4. By generating an influence matrix through information fusion and calculating a stability quantification value, a unified quantitative evaluation scale is formed. When stability is insufficient, simulation parameters are adjusted, the degree of improvement is evaluated, and the dynamic process model is updated. A closed-loop mechanism of "quantitative judgment - parameter adjustment - effect evaluation - model update" is constructed, thereby realizing real-time tracking and adaptive optimization of dye dispersion and migration stability, improving the consistency of paper dyeing quality and the controllability of the production process. Attached Figure Description

[0012] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 This is a flowchart of a high-stability paper dye and its performance testing method according to this application. Figure 2 This is a schematic diagram comparing the migration trajectories of dye particles under normal dispersion and abnormal aggregation states in an embodiment of this application. Figure 3 This is a schematic diagram comparing the dye migration trajectory distribution of the conventional method and the method of the present invention in the embodiments of this application; Figure 4 This is a schematic diagram comparing the technical effects of the method of the present invention and the conventional method in the embodiments of this application. Detailed Implementation

[0014] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a particular order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms “comprising” or “having,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0015] This application provides a highly stable papermaking dye, comprising: The dye matrix has a chemical structure that is stable in high temperature and acid / alkali environments, and is selected from organic dyes, inorganic dyes or combinations thereof. The dye matrix has good dispersibility. At least one stabilizer that enhances the dye’s tolerance to fluctuations in water quality, temperature and pH during the papermaking process. The stabilizer is selected from polymer stabilizers, surfactant stabilizers, or a combination of polymer stabilizers and surfactant stabilizers. At least one auxiliary agent, which can adjust the adhesion properties and color fastness of the dye, and can adjust the solubility of the dye in acidic or alkaline environments to ensure the stable migration and uniform distribution of the dye in the papermaking process.

[0016] In one specific embodiment, the dye matrix is ​​selected from reactive dyes, disperse dyes, or acid dyes; Polymer stabilizers are polyvinyl alcohol, polyacrylic acid, or a combination of polyvinyl alcohol and polyacrylic acid; Surfactant stabilizers include sodium dodecylbenzenesulfonate, hexadecyltrimethylammonium chloride, or a combination of sodium dodecylbenzenesulfonate and hexadecyltrimethylammonium chloride; The adjuvant is one or more chelating agents, metal salts, or regulators.

[0017] Specifically, through a three-layer structure of "dye matrix + stabilizer + auxiliaries", the controllability of dye dispersion, migration and fiber adhesion can still be maintained under conditions of fluctuations in temperature, pH, water ionic strength and coexistence of multiple auxiliaries.

[0018] First, there is the dye matrix. The wet end environment of papermaking is characterized by "high shear, complex ions, pH fluctuations, and numerous fillers / fine molecules." If the chemical structure of the dye is sensitive to acids, alkalis, or temperature, it is prone to hydrolysis, dissociation changes, or sudden changes in solubility, leading to aggregation, sedimentation, or undesirable adsorption with fibers / fillers. The dye matrix in this invention possesses a chemical structure that is stable in high-temperature and acid / alkali environments, and can be selected from organic dyes, inorganic dyes, or combinations thereof, ensuring that the dye maintains basic chemical stability and dispersibility under fluctuating operating conditions. Preferably, reactive dyes, disperse dyes, or acid dyes can be selected, each corresponding to different dissolution / coloring mechanisms: reactive dyes are more likely to form chemical bonds or strong affinity adsorption, disperse dyes rely more on the stability of the dispersion system, and acid dyes are more sensitive to changes in charge state, thus requiring subsequent stabilizers to "passivate" the impact of environmental fluctuations on the dispersion state.

[0019] Secondly, at least one stabilizer is required. The stabilizer's role is not to provide color, but rather to inhibit aggregation and improve resistance to salt, temperature, and pH fluctuations through colloid / interface regulation. Stabilizers can be polymeric, surfactant-based, or a combination of both, exhibiting two typical stabilization mechanisms: polymeric stabilizers (such as polyvinyl alcohol (PVA), polyacrylic acid (PAA), or a combination thereof) primarily enhance the anti-flocculation ability of the dispersion system through steric hindrance, thickening, and adsorption layer formation, especially slowing down effective particle collision and adhesion when ionic strength increases; surfactant-based stabilizers (such as sodium dodecylbenzenesulfonate (SDBS), hexadecyltrimethylammonium chloride (CTAC), or a combination thereof) stabilize dye particles by reducing interfacial tension, providing electrostatic repulsion, or forming micelle coatings. The significance of combining both lies in achieving a synergistic effect of "electrostatic stability + steric hindrance stability": when salt levels increase leading to enhanced electrostatic shielding, the polymer steric hindrance can still maintain dispersion; when temperature fluctuations or increased shear occur, the rapid interfacial response of the surfactant can inhibit transient aggregation.

[0020] Secondly, at least one auxiliary agent is required. Auxiliaries are more focused on process performance control, aiming to regulate the adhesion efficiency, migration controllability, and final color fastness of dyes on the fiber surface. Simultaneously, they adjust solubility under acidic / alkaline conditions to avoid problems such as excessively rapid adsorption leading to localized enrichment or decreased solubility causing deposition. Auxiliaries can be chelating agents, metal salts, or regulators: chelating agents bind metal ions in water, reducing the risk of complexation precipitation between metal ions and dyes / surfactants; metal salts can regulate the charge environment and adsorption driving force under certain conditions; regulators are used to stabilize pH or ionic strength, keeping migration and adsorption within a controllable window.

[0021] Overall, this composition achieves stable migration and uniform distribution in a multi-component papermaking system through "the matrix providing color and basic stability, the stabilizer providing dispersion stability, and the additives providing migration / adhesion and environmental adaptation regulation," thereby improving dyeing consistency and process robustness.

[0022] This invention also provides a performance testing method for highly stable papermaking dyes, such as... Figure 1 As shown, the flowchart for performance testing of highly stable paper dyes includes the following steps: S1. Collect paper dye sample data and auxiliary influence parameters in a multi-component papermaking environment, extract environmental variable characteristics and related influence characteristics from the collected data, and obtain an initial environmental variable set.

[0023] In one specific embodiment, the process of performing step S1 may specifically include the following steps: Data on dye samples and parameters affecting auxiliaries in the multi-component environment of papermaking are collected using sensors or measuring devices. Environmental variable features were extracted from dye sample data, including water quality ion features, temperature change features, pH value variable features and / or ion strength features. Relevant influence characteristics were extracted from the parameters affecting auxiliaries. These characteristics included the influence of auxiliaries on dye affinity, the critical micelle concentration of auxiliaries, and / or the complexing ability of auxiliaries on metal ions. The environmental variable characteristics are integrated with the relevant impact characteristics to obtain the initial environmental variable set.

[0024] Specifically, the wet end system of papermaking simultaneously contains fibers, fillers, dissolved salts, and various auxiliaries. Dyes in the system may experience changes in ionic bonding conditions due to variations in water ions and ionic strength, or may undergo dissolution-aggregation equilibrium migration due to temperature and pH fluctuations, resulting in dispersion drift and uneven fiber adhesion. Offline sampling and single-point testing are difficult to cover this multivariate coupled process in both time and space. Therefore, it is necessary to synchronously collect dye sample data and auxiliary influence parameters on the same time axis, align them with the same data structure, and complete characterization to form an initial set of environmental variables that can be directly used for subsequent clustering and discrimination, in order to address the problem of maintaining dispersion stability and migration uniformity in real time under multi-component environments.

[0025] The data acquisition stage can be located in the mixing tank, white water loop, or bypass circulation unit downstream of the dosing point. Sensors or measuring devices synchronously acquire dye sample data and auxiliary agent influence parameters within the same operating window. The acquisition process establishes data alignment based on a timestamp sequence. Dye sample observations include the concentration of dissolved dye, dispersed dye scattering signal, particle size distribution characterization signal, or equivalent turbidity signal in the slurry system, and are linked to readings from online thermometers, pH electrodes, conductivity meters, ion-selective electrodes, Zeta potential meters, or ion chromatography rapid detection modules. Auxiliary agent parameter observations include the dosage of retention aids, dispersants, surfactants, chelating agents, metal salts or regulators, as well as the concentration of active ingredients, dosing point location, and dilution ratio. Due to shear disturbances and sampling delays in the wet section, the sampling frequency of different devices is resampled during the data cleaning stage and time-aligned with the window width. The median or moving average is used within the window to suppress transient spikes, while retaining abrupt change edges to avoid masking the signs of aggregation initiation. Missing values ​​are interpolated by adjacent windows or by interpolation based on process continuity constraints.

[0026] Environmental variable characteristics were calculated from water quality ions, temperature changes, pH values, and ionic strength information in the dye sample data. Water quality ion characteristics not only represent the concentration of individual ions but also reflect the comprehensive impact on dye charge shielding and complexation risk. Therefore, concentration vectors were formed for divalent metal ions such as calcium and magnesium, multivalent ions such as aluminum and iron, and anions such as chloride and sulfate, and charge equivalent and ion competition indices were calculated at each time point. Temperature change characteristics were expressed using the rate of change of temperature over time or with operating condition gradients, rather than using single-point temperatures, to characterize the disturbances to solubility and micelle structure caused by heating or cooling. Similarly, pH value characteristics should not only take instantaneous pH; window difference and interval deviation were used. The interval deviation was calculated from the distance of pH relative to the process setting interval, to characterize the changes in ionization state caused by acid-base fluctuations. Ionic strength characteristics were estimated by combining conductivity and ion concentration, using a weighted average of the squared ion concentrations to construct an intensity index that reflects both salinity levels and the amplification effect of divalent ions on electrostatic shielding.

[0027] The relevant influence characteristics are derived from the influence parameters of auxiliaries. The goal is to transform the dosage into a characterization of its effect on dye behavior, thereby addressing the problem of uncontrollable dispersion and migration caused by changes in the auxiliary system. The influence characteristics of auxiliaries on dye affinity mainly originate from retention and filtration aids, reinforcing agents, or cationic auxiliaries. An affinity index is constructed by coupling the auxiliary charge density, dosage, and slurry surface potential. The affinity index can be expressed as the product of the dosage and charge density, combined with the Zeta potential shift, to obtain the affinity component at each time point, reflecting the change in the driving force for dye adsorption to fibers or fine molecules. The critical micelle concentration characteristics of auxiliaries, specifically for surfactant-based auxiliaries, are calculated by converting the dosage concentration into a ratio relative to the critical micelle concentration. For example, ,in, for The ratio of relative critical micelle concentrations at time 1 to 2. for The effective concentration of the surfactant in the system at any given time is obtained by converting the dosage, dilution ratio, and system volumetric flow rate (e.g., by multiplying the dosage of the diluted adjuvant solution by the effective concentration after dilution and dividing by the system volumetric flow rate). Cm is the estimated equivalent critical micelle concentration of the adjuvant under the current salinity and temperature conditions (e.g., by inputting the current salinity characterization and temperature reading into a pre-calibrated "salinity-temperature-critical micelle concentration" correspondence (database / fitting function), and interpolating to output the equivalent critical micelle concentration under this condition). This is used to characterize whether micelle formation and solubilization capabilities have entered abrupt change regions, thereby assessing the risk of dye being encapsulated or released by micelles. The complexation ability of auxiliaries to metal ions is characterized for chelating agents or dispersants. A complexation margin index is constructed by comparing the effective concentration of the chelating agent with the equivalent amount of the target metal ion. Insufficient complexation margin makes it easier for dye to complex and deposit with metal ions or for the auxiliaries to fail. The complexation margin at each time point is calculated from the difference between the concentration converted from the chelating agent dosage and the equivalent amount of the metal ion, and is linked to a pH-related correction factor to reflect the influence of the acid-base environment on the complexation balance.

[0028] Environmental variable characteristics and related influencing characteristics are integrated simultaneously to form an initial environmental variable set. The integration method combines splicing and normalization, so that each time point corresponds to a multi-dimensional feature vector. The first half of the vector is the environmental characteristic component, and the second half is the auxiliary agent effect component. All components are standardized before unifying their dimensions. The standardization is based on the mean and dispersion of each component within the production set period to form dimensionless values, thereby ensuring that clustering or subsequent models are not biased due to unit differences. The integrated initial environmental variable set maintains the synchronous correspondence between "multi-component environment - auxiliary agent system - dye sample". Changes in the dispersed state signal in the dye sample can be traced back to the combined state of auxiliary agent effect components such as temperature change component, ionic strength component and critical micelle concentration characteristics of auxiliary agent at the same time, solving the technical problem of difficulty in locating influencing factors when anomalies caused by wet end fluctuations occur. At the same time, it enables the co-occurrence relationship between aggregation anomalies and features such as sudden increase in ionic strength, pH deviation or micelle ratio crossing threshold to be identified, reducing the lag caused by judgment based on a single indicator.

[0029] Preferably, there are two sources of deviation in the above-mentioned construction features: First, the calculation of the equivalent critical micelle concentration Cm and affinity index depends on empirical assumptions, which leads to unstable estimation of the micelle threshold and adsorption driving force of the same auxiliary agent under different salinity, temperature or water quality conditions; Second, there is a process residence time from the addition point to the sampling point where the auxiliary agent enters the system. If the same timestamp is still used for direct pairing, the "input auxiliary agent change" and the "output system response" will be misaligned, causing the correspondence between the feature and the dye state to be distorted.

[0030] When constructing the initial set of environmental variables, to further improve the accuracy of the correspondence between characteristics and dye states in the time dimension, a delay correction can be applied to the auxiliary influence parameters. Specifically, for each type of auxiliary, it is time-shifted according to the residence time δ from the addition point to the sampling point, so that the auxiliary influence parameters are adjusted accordingly. - With sample observation Establish causal correspondences to reduce feature distortion caused by logistics delays.

[0031] Furthermore, to improve the stability and interpretability of relevant influencing characteristics (especially the critical micelle concentration characteristics of auxiliaries and the influence characteristics of auxiliaries on dye affinity) under different operating conditions, the following dynamic modeling method can be adopted: For the critical micelle concentration characteristic Cm of the additive, the actual critical micelle concentration of the additive under different combinations of salinity and temperature can be determined by online small-scale experiments. Then, a lookup table model or fitting function model with salinity and temperature as input and Cm as output can be constructed, so that Cm is transformed from a fixed value into a state variable that is dynamically adjusted according to the working conditions.

[0032] To assess the influence of auxiliaries on dye affinity, a regression calibration model can be constructed to replace simple product calculations. For example, within a sampling window, the amount of auxiliaries added, auxiliary charge density, the zeta potential of the system, and the cation charge requirement of the fiber can be simultaneously collected as input variables. The uniformity or adsorption rate of dye adhesion on the fiber can be used as the calibration target. A regression model (such as multiple linear regression or partial least squares regression) can be trained to fit the mapping relationship between the input variables and the calibration target. The predicted value or intermediate score output by the trained regression model is defined as a dynamic affinity characterization for online feature calculation. This method can more comprehensively reflect the influence of the system's charge state and adsorption capacity on affinity, thereby improving the stability of the correspondence between this feature and dye dispersion / adhesion behavior.

[0033] S2. Use machine learning clustering algorithms to classify the initial set of environmental variables and determine the category of dye dispersion state.

[0034] In one specific embodiment, the process of performing step S2 may specifically include the following steps: A machine learning clustering algorithm is used to perform cluster analysis on a multidimensional feature dataset consisting of multiple environmental variable features from an initial set of environmental variables. Based on the cluster analysis results, the corresponding dye dispersion state categories are determined. The dye dispersion state categories are selected from a set including normal dispersion state and anomalous aggregation state. Normal dispersion state is the state where the characteristic value is within the standard range, and anomalous aggregation state is the state where the characteristic value deviates from the standard range and indicates that the dye has aggregated.

[0035] Specifically, the initial environmental variable set is formed by integrating the environmental variable characteristics and auxiliary influence-related characteristics under the same time index of the papermaking wet end. It records the joint state of temperature change, pH value variable, ionic strength, water quality ionic composition, as well as auxiliary affinity, critical micelle behavior, complexation margin and other characteristics within the same operating window. Cluster analysis uses this set as input to solve the technical problem that offline testing cannot timely determine whether the dye has changed from dispersion to aggregation at the moment of operating condition fluctuation.

[0036] Before data is clustered, sample correspondences are established, constructing a multidimensional feature dataset in matrix form with time as the index. Columns correspond to environmental variable features and additive-related influence features, respectively. The environmental variable column includes ionic strength features, temperature change features, pH deviation features, and ionic composition equivalent features. The additive column includes affinity influence features, micelle ratio features, and complexation margin features. Because the different feature dimensions differ significantly and have different distribution patterns, the multidimensional feature dataset undergoes homoscalation to eliminate the impact of differences in feature dimensions on distance measurement and ensure the stability of the algorithm under varying production process conditions. Homoscalation employs a combination of quantile truncation and standardization. Quantile truncation compresses extreme outliers in each column to a set upper limit to preserve trends while reducing the impact of occasional noise. Standardization converts each column into a zero-mean, unit-variance representation, making the difference in temperature change and the nonlinear estimate of ionic strength comparable in distance calculation. Because data distribution drift can occur in the wet end of papermaking due to process batch changes, shutdowns and restarts, and changes in additive formulations, the standardized benchmarks for the training and operation periods are kept consistent. The benchmarks are derived from statistics within a stable production range to avoid misjudging anomalies due to overall drift in state identification during batch changes.

[0037] The selection of clustering algorithms revolves around "unsupervised division of stable dispersed and clustered anomalous sample groups". Distance metrics can be Euclidean distance or Mahalanobis distance combined with density information. Algorithms can be K-means, hierarchical clustering or density-based clustering methods. Anomalous samples in the wet end of papermaking usually account for a low proportion and exhibit sparse density characteristics. Density-based clustering is suitable for identifying sparse anomalous groups. K-means is more applicable when the sample distribution is approximately spherical. Hierarchical clustering is suitable when the sample size is small or when the clustering level needs to be interpreted.

[0038] During cluster analysis, each record is mapped to a feature space and formed into several clusters according to distance or density rules. Each cluster represents a common combination of "environment-auxiliary agent-system response" states. Samples within a cluster exhibit similar patterns in dimensions such as ionic strength, pH deviation, temperature change, and micelle ratio. Differences between clusters correspond to different impact paths of operating condition fluctuations on the dispersed system. If K-means is used, the preset number of clusters is two, or the number of clusters that maximizes inter-class separation and minimizes intra-class dispersion is selected from the candidate clusters using the silhouette coefficient and elbow criterion. If density clustering is used, the separation of "dense normal regions" and "sparse abnormal regions" is controlled by the neighborhood radius and minimum sample number parameters. The parameters are determined by the nearest neighbor distance distribution of stable production period data, ensuring that most stable operating condition samples form the main cluster while deviation samples are assigned to small clusters or noise classes. The clustering output is a cluster label for each t record. The cluster label itself is not necessarily directly equivalent to "normal dispersion" or "abnormal aggregation." The semantic assignment stage needs to establish a mapping relationship between the cluster label and the standard range and aggregation indication. The standard range is determined during the stable production period or formulation validation period. An interval threshold is established for each characteristic column. The threshold source can be a quantile interval or a process-defined allowable fluctuation range. The threshold covers fluctuations in temperature, pH deviation, and ionic strength under normal dispersion conditions, and is combined with the stable intervals of scattering or particle size characterization signals to form a joint constraint. An increase in scattering signal and an increase in particle size correspond to an aggregation trend, while a stable scattering signal corresponds to stable dispersion. For each cluster, the proportion of each characteristic column within the cluster falling into the threshold and the statistical characteristics of the scattering or turbidity signal are calculated. When the samples within the cluster are within the threshold in most characteristic dimensions and the scattering or particle size characterization signal is within a stable interval, the cluster is mapped to a normal dispersion state. When the samples within the cluster exhibit systematic deviations in dimensions such as ionic strength, pH deviation, temperature change, or micelle ratio, accompanied by enhanced scattering, increased equivalent turbidity, or a rightward shift in the particle size characterization signal, the cluster is mapped to an anomalous aggregation state. To ensure that occasional out-of-bounds errors in a single dimension do not trigger misjudgments, the mapping rule adopts a joint criterion. The joint criterion requires that at least two feature dimensions deviate from the threshold or the deviation exceeds the set boundary at the same time, and that the scattering or turbidity characterization shows corresponding changes, so as to avoid misjudgments caused by only instantaneous temperature disturbances or brief pH fluctuations.

[0039] During the runtime, when new data arrives, the feature vector of the new record is obtained by preprocessing and standardization transformation consistent with the training period. This vector is then fed into the established clustering model or the cluster center is updated based on online incremental clustering rules. The cluster label is output, and the dye dispersion state category is obtained through the mapping relationship from cluster to state. This provides a dispersion state judgment when operating conditions fluctuate and provides an entry point for subsequent migration trajectory analysis triggered by aggregation anomalies.

[0040] This processing chain integrates multi-source process data under the same time index and forms state categories through unsupervised partitioning, avoiding the lag and untraceability caused by relying on a single offline indicator. At the same time, it uses standard range and system response signals to perform semantic calibration on the clustering results, so that the dispersion state category corresponds to the dye aggregation risk in the wet end of papermaking, supporting the determination of dye dispersion state when production conditions change.

[0041] Preferably, in the above technical solution, clustering is an unsupervised partitioning method, and the output is several cluster labels. If the number of clusters, distance metric, feature weights, and mapping rules from cluster to state are not defined, it is easy for the same working condition to be assigned to different clusters in different batches, or for abnormal samples to be absorbed into the main cluster, leading to missed detections. At the same time, if the criterion of "feature value within / out of standard range" does not specify the source and update mechanism of the standard range, benchmark drift will occur when facing seasonal changes in water quality, changes in additive formulation, or changes in paper type, resulting in false alarms or missed alarms. To further improve the stability of cluster analysis, the robustness of state discrimination, and the adaptability of the system to long-term process changes, in a preferred embodiment, the following "clustering-criteria-calibration" closed-loop mechanism may also be included: Clustering is limited to binary clusters or density clustering is used to output the main cluster and outliers, with the center of the main cluster as the normal baseline. For each feature, an interval threshold based on stable production period data is established and updated by a sliding window. The threshold update trigger condition is bound to the paper type / formula switching event. The mapping from cluster to state adopts a joint rule, requiring that samples within a cluster be judged as agglomeration anomalies only when at least two indicators such as ionic strength, pH deviation, and temperature change exceed the limits and the dispersion characterization signal (scattering / turbidity / particle size) shows a consistent change. At the same time, cluster stability test and retraining trigger condition are introduced. When the silhouette coefficient or the radius of the main cluster exceeds the threshold, the clustering model and standard range are updated, thereby reducing cross-operating condition drift and improving the consistency of anomaly identification. For example, the update process of the clustering model and the standard range is as follows: collect recent data with a sliding window, calculate the silhouette coefficient and the radius of the main cluster. If the threshold is exceeded, re-perform feature standardization and re-run clustering with the window data to obtain the new main cluster center and cluster boundary. The old model is retained as the fallback baseline. The standard range is based on the samples in the new main cluster. For each feature, the quantile interval or the mean ± multiple dispersion is calculated to form a threshold range and a buffer band is set. Then, the statistics are updated recursively by window during the operation. If a paper type or formula switching event occurs, the window is reset and the threshold range is recalibrated.

[0042] S3. When the dye dispersion state category indicates abnormal aggregation, acquire migration behavior trajectory data and analyze the dynamic process change law.

[0043] In one specific embodiment, the process of performing step S3 may specifically include the following steps: When the dye dispersion state category is determined to be an aggregation anomaly state, the migration behavior trajectory data associated with the aggregation anomaly state is obtained. The migration behavior trajectory data includes the sequence of the position of dye particles in the multi-component environment of papermaking changing over time. Time series analysis methods are used to analyze migration behavior trajectory data and identify patterns of change in the data; The change pattern is matched with a preset migration dynamic feature library, and the dynamic process change law is determined based on the matching result. The dynamic process change law is used to qualitatively characterize the migration dynamic behavior of dyes in the multi-component environment of papermaking.

[0044] Specifically, when the dye dispersion state is determined to be an anomalous aggregation, the migration and attachment of dye particles in the wet end of papermaking will change from uniform diffusion to local enrichment or agglomeration deposition. Traditional offline sampling can only observe color difference or residual liquid changes after the fact and cannot locate the time period and migration path of aggregation. Therefore, this step uses the anomalous aggregation as a trigger condition to transform the migration behavior into a calculable trajectory time series, which is used to characterize the dynamics of anomalous migration and provide guidance for subsequent fiber attachment distribution collection.

[0045] The trigger time of aggregation anomaly is denoted as λ. A trajectory acquisition window is established around λ, covering the backtracking segment before λ and the tracking segment after λ, in order to capture the transition process from dispersion to aggregation. The acquisition object within the window is the spatial position of observable dye particles or dye aggregates in a multi-component environment. The spatial position can be obtained from online microscopic imaging, laser scattering localization, fluorescence tracer imaging, or image particle tracking module of the bypass flow cell. The acquisition results are organized as a sequence of position changes over time. The sequence elements are two-dimensional or three-dimensional coordinate vectors. The coordinate vectors are bound to the slurry flow rate, shear characterization, and process variables such as temperature, pH, and ionic strength at the same time, so that the trajectory sequence not only includes the geometric path, but also retains the correspondence with the fluctuation of the operating conditions. To avoid trajectory breaks caused by imaging noise and occlusion, the position sequence undergoes trajectory reconstruction and denoising before entering the analysis. The reconstruction process uses the maximum displacement constraint and velocity continuity constraint of adjacent frames as matching rules to associate candidate points of the same particle at adjacent time points. At the breakpoints, interpolation based on the motion model is used to complete the process. At the same time, median filtering is used to reduce the impact of pixel jitter on displacement calculation. Each trajectory in the trajectory sequence corresponds to a continuous motion record of a particle or agglomerate, and a traceable correspondence is maintained through the trajectory number and the trigger window λ.

[0046] Time series analysis constructs migration feature sequences based on the coordinate sequences of each trajectory. Migration features include displacement sequences, velocity sequences, acceleration sequences, and path curvature sequences. The displacement sequence is obtained from the difference between adjacent coordinates, the velocity sequence is obtained by removing positions and sampling intervals, the acceleration sequence is obtained from the velocity difference, and the path curvature is obtained from the geometric relationship of three points or the rate of change of the tangent vector. All of the above sequences are kept consistent in time according to the time index and can be aligned with the rate of change features of environmental variables at the same time. In the change pattern recognition stage, the migration feature sequence is transformed into pattern descriptors. These descriptors can employ a combination of sliding window statistics and event detection. Sliding window statistics include the mean velocity, velocity variance, displacement skewness, and mean curvature within the window. Event detection captures abrupt change points and stage transition points. Abrupt change points can be obtained through cumulative sum testing or threshold-based change point detection. Stage transition points are used to divide the trajectory into sub-segments such as stable migration segments, deceleration and aggregation segments, and backflow oscillation segments. Each sub-segment outputs a set of parameter vectors, containing the time and location of the abrupt change, the magnitude of the abrupt change, the duration of the stable segment, and the oscillation frequency. These parameter vectors correspond to trajectory numbers, thus achieving a ensemble representation of patterns across multiple trajectories. To summarize the change patterns of multiple trajectories into dynamic process change patterns, the pattern descriptors are aggregated at the trajectory level. The aggregation method is based on the pattern occurrence ratio and co-occurrence relationship. The proportion of each type of pattern within the trigger window is calculated, and its synchronicity with process variable fluctuations is recorded, forming a pattern fingerprint for matching the feature library.

[0047] The migration dynamic feature library stores pattern fingerprints of different migration dynamic behaviors in the form of template sets. Each template contains several constraints and allowable deviation ranges. The constraints cover velocity decay patterns, residence time distribution, path curvature features, and mutation point density, etc. They can also include correlation markers with ionic strength or pH deviations to distinguish between agglomeration deceleration caused by charge shielding and reciprocating oscillations caused by shear perturbation. The matching process calculates the similarity between the pattern fingerprint to be matched and the templates in the database. The similarity can be obtained by weighted distance or dynamic time-warped distance. The weight is used to highlight the feature dimensions that are more sensitive to aggregation, such as the magnitude of speed drop and the increase in residence time. A similarity score is obtained for each template, and the one with the smallest score or the largest similarity is selected as the matching result. The matching result outputs the corresponding dynamic process change pattern category and the key behavioral representation statement under the category, such as category labels such as "deceleration enrichment type", "reciprocating oscillation type", and "intermittent sudden aggregation type". The category label is bound to the trigger time λ and the trajectory set, which serves as the orientation basis for subsequent fiber attachment distribution collection. This allows the collection area to focus on the dense residence area or curvature change area rather than random sampling, solving the technical problems of difficulty in localization and lag in processing after anomalies occur. At the same time, this change pattern provides a behavioral explanation framework for subsequent correlation analysis, making there a traceable mediating variable between changes in auxiliary parameters and changes in migration path.

[0048] like Figure 2 As shown, the left side represents the particle trajectory in a normal dispersed state. The two-dimensional motion trajectories of multiple particles exhibit random diffusion without a significant dominant direction near the origin. The trajectories intertwine, and the sampling points are relatively evenly distributed, reflecting the system's stable dispersion and minimal disturbance to migration behavior. The right side represents the particle trajectory in an anomalous aggregation state. The particle trajectory starts from the aggregation center (red square) and migrates radially or in bundles. Within the aggregation-affected area (red shaded circle), particles exhibit dwell time, circling, and path deflection, reflecting the uneven migration and local enrichment trend caused by aggregation. Green dots represent sampling locations at equal time intervals, and blue squares represent trajectory endpoints. These are used to generate state variables such as migration speed, dwell time ratio, and aggregation risk, and to provide a basis for subsequent directional sampling of attachment distribution.

[0049] Preferably, trajectory acquisition is highly dependent on imaging quality and particle visibility. High slurry turbidity can easily lead to occlusion, causing trajectory breakage and affecting the detection of change points. To further improve the reliability of trajectory data in complex slurry environments, a constant optical path and a dilution bypass branch can be added to the bypass flow cell to stabilize imaging. The bypass flow cell is used for online imaging or particle tracking. The constant optical path reduces light transmission changes and scattering noise caused by slurry concentration fluctuations by fixing the flow cell thickness and optical path length, thus stabilizing image contrast. The dilution bypass branch dilutes the sampling branch at a fixed ratio without changing the main process condition judgment, reducing turbidity and occlusion probability, and reducing trajectory breakage and mismatch. Simultaneously, multi-source trajectory fusion can be introduced to verify the consistency between the temporal characteristics of the imaging trajectory and the scattered particle size signal. The fractured segments are aligned and corrected according to the particle size surge event. The imaging trajectory provides path information of particle positions over time, while the scattered particle size signal synchronously reflects changes in the system's particle size distribution. When a sudden drop in velocity or an increase in residence time occurs during a continuous trajectory period, it usually corresponds temporally to a sudden increase in particle size or an increase in scattering intensity. The consistency verification compares the temporal position and direction of change of the two types of signals to determine whether the trajectory change is synchronized with the particle size event. If the imaging is fractured due to occlusion, a time window of the particle size surge event can be searched before and after the fracture. The time index of the fractured segment is aligned to this event, and interpolation is performed to complete the sequence according to the motion trend before and after the fracture, ensuring that the trajectory sequence remains synchronized with the actual aggregation process.

[0050] Furthermore, if the migration dynamic feature library is fixed and not updated, it will become mismatched when switching paper types or auxiliary agent systems. To further adapt the migration dynamic feature library to process changes, improvement data can be used as feedback. Newly emerging pattern fingerprints can be added to the library using an incremental update method, and the template threshold can be adaptively adjusted. The improvement data refers to the quantitative change results obtained by re-evaluating the dye dispersion and migration behavior before and after parameter adjustment. It comes from newly collected sample data and calculates indicators such as dispersion uniformity, migration speed, and adhesion deviation. The difference or percentage change is then calculated with the corresponding indicators before adjustment to form an "improvement percentage / improvement magnitude" record. At the same time, it is linked to the auxiliary agent parameters and environmental conditions of the current adjustment to determine the response characteristics of a certain migration pattern under new paper types or new auxiliary agent systems.

[0051] S4. Obtain fiber adhesion distribution information based on the dynamic process change law, analyze the correlation between fiber adhesion distribution information and auxiliary agent influence parameters through support vector machine model, and determine the migration path optimization point.

[0052] In one specific embodiment, the process of performing step S4 may specifically include the following steps: Based on the dynamic process change law, fiber adhesion distribution information on the surface of pulp fiber / paper fiber is collected in a directional manner from the multi-component environment of papermaking. The fiber adhesion distribution information includes the adhesion density and / or position distribution data of dye on the fiber surface. The correlation between fiber adhesion distribution information and auxiliary agent influence parameters was analyzed using a support vector machine model to obtain quantitative results of the correlation. Based on the correlation quantification results, key parameter combinations that have a significant impact on dye migration behavior are identified, and these key parameter combinations are determined as migration path optimization points.

[0053] Specifically, the dynamic process variation patterns are derived from the time series analysis of migration trajectories after aggregation anomalies are triggered. These patterns include information such as the time boundaries of migration behavior, indications of densely populated areas, moments of sudden velocity drops, and backflow oscillation zones. This information is used to address the problem of random sampling locations and delayed sampling times when uneven adhesion occurs in the papermaking wet end, making it impossible to correlate adhesion results with migration anomalies. Therefore, when collecting fiber adhesion distribution information, a sampling index is established based on the spatiotemporal indications output by this pattern. The sampling index is denoted as α, which consists of a time window and spatial region encoding near the trigger moment. The time window covers the interval before and after the abrupt change point to capture the transmission process from migration anomalies to adhesion enrichment. The spatial region corresponds to the flow channel location or slurry layer region where trajectory dwell density is high or curvature abrupt changes are concentrated. Fiber samples are sampled and fixedly stored in the bypass branch according to α, ensuring that each α corresponds to a fiber surface observation result and is aligned with process records such as additive addition records, temperature changes, pH deviations, and ionic strength under the same α. The fiber adhesion distribution information is obtained by fiber surface imaging or surface signal measurement. The imaging method can be fluorescence imaging or reflection imaging to obtain the dye signal matrix on the fiber surface. The signal matrix is ​​mapped to a grid according to the fiber surface coordinate system. The grid cell is denoted as β. β corresponds to a local area on the fiber surface. The pixel intensity on β is converted into an adhesion amount characterization after background subtraction and illumination correction. The adhesion density is obtained by integrating the adhesion amount per unit area. The position distribution is obtained by the spatial distribution of the adhesion amount on the β grid. An adhesion patch set is formed by threshold segmentation. The patch set is used to calculate the patch center coordinates, patch area distribution, patch spacing distribution and spatial autocorrelation structure. To prevent subsequent correlation analysis from falling into the problem of "output being deterministically calculated from input, leading to the neglect of auxiliary parameters," the attachment distribution information is split into two types of data objects. One type consists of directly measured original and intermediate representation quantities, including the attachment density field, patch center set, and patch area set. The other type is the distribution response index used for modeling. The distribution response index selects only intermediate representation quantities that can reflect the attachment enrichment morphology and are not used as output targets, such as the proportion of local excessive attachment, patch size quantile, patch spacing quantile, fractal dimension of the density field, or spatial entropy. This allows the distribution response index to characterize the differences in the enrichment morphology left by migration anomalies on the fiber surface, while avoiding the direct input of an index component that will be used as output, thus preventing target leakage. Each α corresponds to a vector in the distribution response index space. The vector components maintain a traceable correspondence with the original distribution quantity and are bound to the migration dynamic label of the same α, making different dynamic process types corresponding to different attachment morphologies usable supervisory information.

[0054] The parameters affecting adjuvants are organized into parameter vectors under the same α index. These vectors include process inputs such as dosage, concentration, application point, and dilution ratio, as well as derived quantities derived from these inputs, such as affinity, micelle ratio, and complexation margin. The time alignment of the parameter vectors can be corrected according to the residence time from the application point to the sampling point, ensuring that the parameter vector at time α corresponds to the input state that causes the adhesion response at that time. This solves the mismatch problem caused by the lag in adjuvant action. The distributed response index vector and the adjuvant parameter vector are paired one-to-one at α to form a set of sample pairs. This set of sample pairs provides the data foundation for support vector machine modeling. The input in each sample pair consists only of the adjuvant parameter vector or a combination of the adjuvant parameter vector and a migration dynamic label. The output consists of one or more of the distributed response indices. The migration dynamic label is used to distinguish the response differences of the same parameter under different migration modes and can be used as a discrete input component or for group training of the model. The Support Vector Machine (SVM) model task employs a regression approach. The regression objective is not a comprehensive uniformity value directly calculable from the same input, but rather the distribution response index itself, such as the proportion of local over-attachment or the quantile difference between patch spacing. These objectives do not have a deterministic functional relationship with the additive parameters, and changes in the objective reflect the response intensity of the fiber surface enrichment morphology after changes in the additive system. Therefore, the model output must rely on the additive parameters to obtain a smaller error. Model training uses a set of sample pairs as training data. Before training, the components of each additive parameter are scaled uniformly, and components with strong collinearity are regularized or subjected to principal component transformation to maintain numerical stability. A radial basis function (RBF) kernel is used to express nonlinear coupling relationships. Hyperparameters are selected within the training set through cross-validation and determined jointly by regression error and stability constraints. After model training, the correlation quantification results are derived from the model. This is achieved through sensitivity analysis or importance assessment. Sensitivity analysis involves perturbing a specific additive parameter component within its process tolerance range and calculating the magnitude of change in the regression output. The magnitude of change is statistically analyzed across multiple samples to obtain the influence strength score of that parameter on the adhesion response index. Importance assessment involves replacing a parameter component and observing the increase in regression error to obtain a contribution score. A higher score indicates that the parameter is more critical to explaining the adhesion distribution response. The score set forms a correlation quantification table and corresponds one-to-one with the parameter components. If the output target includes multiple distribution response indicators, a score set is formed for each indicator and weighted and fused according to migration dynamic labels to obtain the "comprehensive influence strength on different aspects such as enrichment morphology, patch size, and spacing structure." This ensures that the correlation quantification results can cover the multidimensional characteristics of the adhesion distribution without being limited by a single indicator.

[0055] The identification of key parameter combinations is based on the correlation quantification table and combined with the parameter interaction construction. The interaction is obtained through two-parameter linkage perturbation. The output change caused by linkage perturbation minus the single-parameter perturbation change is the interaction strength. Parameter pairs with interaction strength exceeding the threshold are used as candidate combinations. The candidate combinations are then expanded into multi-parameter combinations under process constraints. The process constraints include the upper limit of addition, pH control range, controllable ion strength range and slurry flow boundary. The combination screening also considers the applicability of migration dynamic tags to avoid misusing parameter combinations that are only effective in a certain migration mode to other modes. The migration path optimization point is defined here as the key parameter combination and its corresponding parameter setting value or control window. The control window is obtained by back-calculating the attachment response index output by the model to the target range. The back-calculation can be obtained in the parameter combination space by grid search or constraint optimization, so that the window boundary corresponds to the feasible parameter range under the target conditions such as "reduction of attachment enrichment ratio, convergence of patch scale, and stabilization of spacing structure". The migration path optimization point is bound to the α index and migration dynamic label and output to the subsequent steps for monitoring index extraction and stability quantification. This allows real-time monitoring to no longer stop at observing the attachment results but to correspond to the deviation and adjustment direction of the adjuvant parameter combination. Thus, the problem of difficulty in locating the root cause of uneven attachment and color difference fluctuation is transformed into a traceable diagnosis and executable adjustment basis for parameter combination.

[0056] S5. Extract real-time monitoring indicators from migration path optimization points, and use information fusion technology to integrate the impact of environmental variables on real-time monitoring indicators to obtain the stability quantification value of paper dye stability.

[0057] In one specific embodiment, the process of performing step S5 may specifically include the following steps: Data scanning was performed on migration path optimization points to extract real-time monitoring indicators related to dye behavior; Information fusion technology is used to integrate the effects of temperature change, pH value and ionic strength on real-time monitoring indicators. An influence matrix is ​​generated based on the fusion results. The influence matrix records the change range of each real-time monitoring indicator under different combinations of variables. Based on the influence matrix, the stability quantification value of papermaking dye stability was calculated using the geometric mean method.

[0058] Specifically, the migration path optimization points were identified in the preceding correlation analysis as key auxiliary agent influence parameter combinations and their corresponding parameter settings or control windows, and an index correspondence was established with the migration dynamic type and fiber adhesion distribution response. This correspondence enabled the papermaking wet-end dye stability to no longer rely solely on offline paper sample color difference or residual liquor color judgment, but to construct calculable indicators within the process around "which parameter combinations need to be monitored and to what extent deviation will trigger risks".

[0059] Data scanning is organized using a sampling index γ, which corresponds to the additive dosing records, environmental variable readings, and derived parameter values ​​such as affinity characterization, micelle ratio characterization, and complexation margin characterization within the same sampling period. The set of key parameter dimensions corresponding to the migration path optimization points is denoted as Δ. Each dimension in Δ is bound to a control window or setpoint. This window is derived from the parameter range of "reducing the attachment enrichment response" in the previous step and is constrained by the process allowable range. The reason why real-time monitoring indicators need to be extracted from the migration path optimization points is that the optimization points themselves are descriptions of parameter combinations and their target ranges, which cannot be directly used for continuous online judgment. Moreover, the dimensions and sensitivities of different parameters vary greatly. If they are not converted into indicators of a unified scale, it is difficult to integrate them with temperature changes, pH deviations, and ionic strength perturbations into the fusion calculation and form a stability quantification value. Based on this, the current value of the Δ dimension is read for each γ, and a one-dimensional deviation index is calculated. The one-dimensional deviation index is obtained from the "difference between the current value and the set value" or the "distance from the current value to the control window boundary". It is then normalized using the window width or allowable fluctuation range to form a dimensionless deviation ratio, and an out-of-bounds flag is generated to indicate whether the window has been exceeded. A combined deviation index is constructed based on the one-dimensional deviation index to characterize the risk of migration path instability caused by multi-parameter linkage deviations. The combined deviation index is synthesized by weighting each deviation ratio according to the influence weight in the correlation quantification result. The weight comes from the sensitivity score or contribution score in the previous step, making the parameter more sensitive to the attachment distribution response contribute more. Considering the residence time from the dosing point to the sampling point, the dosing amount or concentration data is time-shifted according to the residence time before calculating the deviation index, so that the index at time γ corresponds to the input state that triggers the dye behavior at that time, thus forming a real-time monitoring index vector aligned with environmental variables according to γ.

[0060] In the information fusion stage, a fusion input is constructed using monitoring indicator vectors and environmental variable vectors. The fusion result is organized into an influence matrix, where rows correspond to monitoring indicator components and columns correspond to environmental variable combination states. Environmental variable combination states are obtained by interval coding of temperature changes, pH deviations, and ionic strength characterization quantities. The interval coding is determined by the process setting interval and historical fluctuation distribution, allowing each γ to be mapped to an environmental state number. The environmental state number reflects the combined state of "temperature rise or fall trend, acid-base deviation direction and magnitude, and salinity level." Influence matrix elements characterize the magnitude of the impact of a monitoring indicator on dye stability surrogate quantities under a given environmental state. Stability surrogate quantities are derived from observable dispersion and migration characteristics within the process, such as the rate of change of scattering signal, the magnitude of equivalent turbidity change, the magnitude of rightward shift in particle size distribution, or the residence time ratio of migration trajectories. These surrogate quantities correspond to anomalous aggregation and non-uniform adhesion and can be continuously acquired within the process, addressing the control gap caused by the lag in final paper sample inspection. The surrogate quantities are collected synchronously with γ or aligned to the same timestamp. The matrix is ​​constructed using samples of the same environmental state number within a historical sliding window as a set. For each monitored indicator component, the response amplitude between it and the stability surrogate quantity is calculated. The response amplitude can be characterized by the absolute value of the regression coefficient, the mutual information strength, or the conditional mean difference. The conditional mean difference is given as the "difference between the surrogate quantity mean of the out-of-bounds samples and the non-out-of-bounds samples," enabling the matrix elements to reflect the sensitivity of the indicator to triggering stability degradation under that environmental state. To maintain comparability between different indicators and different environmental states, the matrix elements are normalized to the baseline amplitude under normal dispersion conditions. After normalization, the matrix elements are dimensionless amplification factors. The larger the matrix element, the more significant the deviation of the indicator under that environmental state affects the stability surrogate quantity. Thus, the integration and solidification of the influence of environmental variables on the indicator is formed into a structured matrix.

[0061] The stability quantification value for each γ is calculated jointly by the influence matrix and real-time monitoring indicators. The calculation first selects the corresponding column of the influence matrix based on the environmental state number corresponding to γ, obtaining the influence coefficient corresponding to each monitoring indicator component. Then, the influence coefficient and the deviation ratio of the indicator under that γ are combined to form a risk factor sequence. The risk factor is a dimensionless quantity reflecting the comprehensive strength of the indicator deviation under the amplified influence of the current environmental state. The risk factor can be the product of the deviation ratio and the influence coefficient, or a weighted mapping value of the two. To avoid a single risk factor excessively dominating the overall evaluation and to maintain the synergistic constraints of multiple indicators, the risk factor sequence is synthesized using a geometric mean to obtain the stability quantification value. The geometric mean is obtained by multiplying each risk factor and taking the square root, resulting in a single value. This ensures that any key indicator exceeding its limit under a high-impact environmental state significantly pulls on the quantification value. Simultaneously, the combined effect of multiple slight deviations can also be cumulatively reflected, thus transforming "instability caused by multiple factors under multi-component environmental fluctuations" into a quantification sequence that can be updated over time and applied one-to-one with γ.

[0062] Preferably, to further improve the robustness of the influence matrix estimation and enhance the computational robustness of the stability quantization value, the following optimization measures can be adopted: The influence matrix depends on the discrete encoding of environmental states. If the encoding is too fine, it will lead to sample sparsity and fluctuations in matrix elements. If the encoding is too coarse, it will weaken the continuous effect of temperature and salinity. Therefore, when encoding environmental states, kernel weighted estimation can be used instead of hard binning. For example, a Gaussian kernel function can be used to weight the samples according to the similarity between the samples and the center of the environmental states in order to smoothly calculate the matrix elements.

[0063] Geometric mean is sensitive to zero values ​​and carries the risk of double-weighting for highly correlated indicators. Therefore, a lower limit can be applied to the risk factor sequence before calculating the geometric mean. In addition, if there is a high correlation between the monitoring indicators, an orthogonal transformation (such as principal component analysis) can be performed on them to obtain uncorrelated risk factor components, and then the geometric mean can be applied to these components.

[0064] S6. If the stability quantification value is lower than the preset stability threshold, the simulation parameters of the multi-component environment of papermaking are adjusted, and the improvement of dye dispersion and migration behavior after parameter adjustment is evaluated.

[0065] In one specific embodiment, the process of performing step S6 may specifically include the following steps: Determine whether the stability quantification value is lower than the preset stability threshold. If so, map the stability quantification value to the adjustment instructions for the simulation parameters of the multi-component environment of papermaking according to the preset adjustment strategy. The simulation parameters include at least temperature change, pH value and ionic strength. After adjusting the corresponding simulation parameters according to the adjustment instructions, new dispersion uniformity and migration rate indices are calculated by re-collecting dye sample data in the adjusted multi-component environment. The new dispersion uniformity and new migration rate indices are compared with the corresponding data before parameter adjustment to obtain quantitative improvement data. The quantitative improvement data represents the degree of improvement in dye dispersion and migration behavior after adjustment.

[0066] Specifically, the stability quantification value is calculated in the previous stage by real-time monitoring indicators and the influence matrix of environmental variables, and a time series is formed by sampling index γ. This series is used in the wet end of papermaking to characterize the overall stability of dye dispersion and migration behavior under the current multi-component working conditions. When the quantification value is lower than the preset stability threshold, it means that the deviation of key parameters is amplified under the conditions of temperature change, pH deviation and ionic strength, and is synchronous with the deterioration trend of dispersion characterization and migration characterization. At this time, relying solely on operational experience for adjustment is prone to over-adjustment or lag, resulting in continuous uneven aggregation diffusion and fiber adhesion. Therefore, step S6 uses the quantification value as a trigger signal to map the "degree of instability" into an executable simulation parameter adjustment command, and obtains the improvement data by re-sampling and comparison after adjustment, thereby solving the problems of "lack of closed-loop verification, unquantifiable adjustment effect, and difficulty in convergence of parameter settings under working condition disturbance".

[0067] The threshold judgment is based on the quantized value at the current moment, denoted as ψ, and compared with the stable threshold. The stable threshold can be derived from the lower quantile of the quantized value distribution under normal distributed operating conditions or the minimum stable level allowed by the process. The comparison result generates a binary trigger flag and triggers a preset adjustment strategy. The preset adjustment strategy is stored in the form of a rule base or mapping function. Its input is ψ and its deviation from the threshold. It can also be combined with the current environmental variable status number to distinguish different instability mechanisms such as "high salinity and low pH" and "low salinity and high temperature". The strategy output is an adjustment instruction vector for the simulation parameters. The dimension of the adjustment instruction vector corresponds to three types of simulation parameters: temperature change, pH and ionic strength. It can also include adjustment step size, adjustment direction and adjustment upper limit. The step size is determined by the deviation of ψ in stages, so that slight instability corresponds to small step size fine adjustment and significant instability corresponds to large step size correction. Temperature adjustment commands can be manifested as setting a target temperature or correcting the temperature slope; pH adjustment commands can be manifested as correcting the amount of acid or alkali added or the target pH setting; and ionic strength adjustment commands can be manifested as correcting the amount of dilution water added, salt added, or chelating agent added. All three types of commands are executed in a simulated environment or a bypass pilot environment to avoid directly disturbing the main process. The parameters executed in the simulated environment correspond one-to-one with the commands, and after execution, a stabilization waiting period is entered. The length of the waiting period is determined by the residence time and mixing time to ensure that the adjustment effect covers the sampling points. To ensure the comparability of data before and after adjustment, the baseline data before adjustment uses the sample from the most recent stable window under the same paper type and the same auxiliary agent system as the reference window. The statistical values ​​of dispersion uniformity and migration speed within the reference window are used as baseline indicators and bound to the reference window index. After adjustment, dye sample data are re-collected at the same sampling location and with the same sampling period. The collected data includes dissolved dye concentration, scattering or turbidity characterization, particle size distribution characterization, and necessary trajectory positioning data. The dispersion uniformity index is calculated from the particle size distribution or turbidity sequence. The calculation method can be characterized by the dispersion of the particle size distribution, the variance of the turbidity, or the coefficient of variation of the scattering signal. The migration speed index is calculated from the migration trajectory data. The trajectory data is stored in the form of position over time series. The speed is obtained by dividing the difference between adjacent positions by the sampling interval. The mean or quantile value of multiple trajectories is taken to form the migration speed statistics of the sampling window.

[0068] The indexes corresponding to the newly collected samples are designated as adjusted indices. After being aligned with the baseline indices according to the same index definition, they are differentially or proportionally calculated to obtain quantitative improvement data. The quantitative improvement data includes at least the improvement in dispersion uniformity and the improvement in migration speed, and can be further converted into percentage improvement for cross-batch comparison. By completing parameter adjustments in a simulated environment and comparing with indices of the same caliber to obtain improvement data, changes in dye dispersion and migration are no longer judged solely by subjective observation, but rather quantitative data characterize whether the adjustments have led to convergence of particle size dispersion, reduction of turbidity fluctuations, and restoration of migration speed to an acceptable range.

[0069] like Figure 3 As shown, the left figure shows the dye migration trajectory under the traditional method. The trajectory is randomly scattered, and there is obvious trajectory overlap and local dwell in the aggregation regions 1 and 2 shown in the figure. This indicates that the dye is prone to aggregation-induced uneven migration and enrichment under the disturbance of multi-component wet environment. The right figure shows the dye migration trajectory under the method of the present invention. The trajectory is distributed in a band-like convergence along the same main direction. The dispersion range is significantly reduced and no local aggregation region is formed. This indicates that after determining the migration path optimization point and performing parameter adjustment, the consistency and stability of the dye migration direction are improved, and the fluctuation of state variables such as migration speed and dwell ratio is reduced, which provides a basis for the adaptive update of the dynamic process model.

[0070] Preferably, to further improve the accuracy of closed-loop adjustments, the completeness of evaluations, and the transferability of parameters from the simulation environment to the main production process, the following optimization measures can be adopted: If the adjustment strategy only maps the step size to the stability quantification value, it may lead to misadjustment of different factors with the same value under different instability mechanisms. Therefore, the input of the adjustment strategy can also include the environmental state number and the migration dynamic category, and constraint priority can be introduced to limit the adjustment order of different simulation parameters, thereby improving the pertinence and effectiveness of the adjustment.

[0071] If only dispersion uniformity and migration speed are used to assess the degree of improvement, the improvement in adhesion distribution may be overlooked. Therefore, after adjustment, the fiber adhesion response proxy can be collected simultaneously and used as a third improvement component to make the model update more consistent with the "migration-adhesion" link, thereby ensuring that the improvement assessment covers the complete link of dye behavior.

[0072] Since there are scale differences between the simulation environment and the main process, simulation parameters can be calibrated using dwell time similarity criteria and shear condition similarity criteria, making adjustment instructions more transferable and thus improving the transferability of adjustment parameters from the simulation environment to the main process.

[0073] S7. Update the dynamic process model based on the improvement level data, and generate the final real-time monitoring scheme based on the updated model. The dynamic process model is used to simulate the dispersion and migration behavior of dyes in the multi-component environment of papermaking.

[0074] In one specific embodiment, the process of performing step S7 may specifically include the following steps: The improvement level data is analyzed to extract the quantitative improvement components. According to the preset mapping rules, the quantified improvement components are converted into adjustments to the internal parameters of the dynamic process model; The corresponding parameters of the dynamic process model are updated based on the adjustment amount to obtain the updated dynamic process model; Run the updated dynamic process model, set or optimize the operating parameters for real-time monitoring based on the model output, and generate a real-time monitoring scheme for tracking the stability of dye behavior in real time.

[0075] Specifically, the improvement data is obtained by comparing indicators such as dispersion uniformity and migration speed before and after parameter adjustment, and is linked to the current environmental conditions, auxiliary parameter combinations, and adjustment instructions. In the papermaking wet end scenario, this data is used to characterize the recoverability and adjustment sensitivity of the same dye system under specific multi-component environmental disturbances. The dynamic process model is used to simulate the dispersion and migration behavior of dyes in the multi-component environment of papermaking and outputs state variables related to aggregation risk, migration speed, residence time ratio, and adhesion enrichment trend.

[0076] The construction and training of the dynamic process model includes: collecting time series samples of input variables such as temperature changes, pH deviations, ionic strength, and key additive parameter combinations; normalizing continuous variables; and introducing operating condition context encodings such as paper type and water quality category to characterize scenario differences. A sliding time window sequence of length L is input into a network structure containing an input encoding module, a two-layer gated recurrent unit (GRU) time-series backbone, and parallel multi-task output heads. The network outputs the aggregation risk probability, migration speed, residence time ratio, and attachment enrichment trend. Specifically, the aggregation risk monitoring signal is generated from the results of dispersed state clustering or the aggregation index threshold; the migration speed and residence time ratio are calculated from the migration trajectory; and the attachment enrichment trend is obtained from the fiber attachment distribution proxy quantity and its rate of change. A multi-task weighted loss function is used to jointly optimize the network parameters, and operating condition quantile statistics are saved for threshold calibration. If the model is not updated with changes in paper type, water quality, and additive system, the model output will gradually deviate from the actual process response, leading to a mismatch between threshold setting, sampling period, and index weight under new operating conditions. Therefore, step S7 uses the improvement degree data as feedback to inject the response amount generated by the closed-loop adjustment back into the model parameters, thereby solving the problems of model solidification leading to non-transferability across operating conditions, difficulty in adaptive monitoring parameters, and long-term drift of abnormal thresholds.

[0077] The data analysis of the improvement level is organized using an index ρ, which corresponds to a complete closed-loop record of "instability triggering - parameter adjustment - resampling evaluation". This record includes the quantified stability value before adjustment, the adjustment command vector, and the changes in dispersion uniformity and migration speed after adjustment. During the analysis phase, these changes are converted into quantified improvement components. These quantified improvement components include at least dispersion uniformity improvement and migration speed improvement components, and may also include adhesion response surrogate quantity improvement components to reflect the transmission effect of migration anomalies to the fiber surface enrichment. The quantified improvement components are expressed numerically as differences or percentages, and their direction is unified through symbolic conventions, ensuring that improvement corresponds to positive changes and deterioration corresponds to negative changes, guaranteeing that subsequent mapping rules have a consistent interpretation of different indicators. The correspondence between the quantified improvement components and the model input variables is established simultaneously during the analysis phase. Model input variables include temperature changes, pH deviations, ionic strength, and combinations of key additive parameters. During analysis, the quantified improvement components are bound to the state number of the current input variable, allowing the same improvement component to be interpreted as a response under a specific input combination, thus providing conditional information for the mapping rules.

[0078] Mapping rules are used to convert quantified improvement components into adjustments for model internal parameters. These internal parameters may include the equivalent charge shielding coefficient (characterizing dispersion stability), the effective collision adhesion coefficient (characterizing aggregation tendency), the equivalent viscosity damping term (characterizing migration hindrance), or the backflow intensity term (characterizing residence zone formation), etc. Internal parameters are not directly equivalent to sensor readings but are used to ensure that the model's output migration velocity, residence time ratio, and dispersion indices are consistent with process observations. Mapping rules are stored in lookup table or function form. The input is the quantified improvement component and its corresponding input variable state number, and the output is a vector of adjustments for each internal parameter. The lookup table method categorizes the improvement components and specifies the adjustment direction and magnitude for each level. The function method treats the improvement component as a continuous independent variable and outputs the adjustment amount through a proportionality coefficient or a saturation function. The proportionality coefficient can depend on the environmental state number to distinguish between situations where the charge shielding term is more sensitive under high salinity conditions and the damping term is more sensitive under low salinity conditions. To avoid conflicting directions from multiple improvement components for the same internal parameter, the mapping rule sets priorities and coupling constraints. Priorities are determined by the contribution of migration speed and dispersion uniformity to clustering risk, while coupling constraints limit the magnitude of simultaneous changes in internal parameters, ensuring the updated model still maintains physical consistency. For example, when the clustering tendency parameter increases, the dispersion stability parameter does not simultaneously increase to an unreasonable range. After obtaining the adjustment amount based on the mapping rule, the dynamic process model is updated record by record ρ. The update method can employ recursive correction, adding the current internal parameter to the adjustment amount and setting upper and lower limits, or exponential sliding update to suppress overcorrection caused by single-time noise. The updated model parameter set is obtained after the update and stored in association with a timestamp, giving the model a traceable version that evolves with operating conditions.

[0079] The updated dynamic process model takes real-time sampled environmental variables and additive parameters as inputs and outputs predicted dispersion states, migration speeds, residence time ratios, and aggregation risks. The model outputs operational parameters for setting or optimizing real-time monitoring. These parameters include at least the sampling period, trigger threshold, influence matrix update window length, and monitoring indicator weights. The sampling period is optimized based on the model's predicted rate of state change; a high rate of change shortens the sampling period to capture abrupt changes, while a low rate of change extends the sampling period to reduce computational burden. The trigger threshold is optimized based on the model's output distribution in the normal dispersion input region, setting the stable threshold as the lower quantile of the normal region's output and adding a buffer band to allow the threshold to adapt to changes in water quality and the additive system. The influence matrix update window length is determined based on the model's predicted drift speed; a fast drift shortens the window to track the new distribution, while a slow drift extends the window to improve statistical stability. The monitoring indicator weights are redistributed based on the model's sensitivity to deviations from different parameters, ensuring that the weighting coefficients of the combined deviation indicators are consistent with the dominant factors under the new operating conditions. The model output forms a closed loop with S5. The stability quantification value generated by S5 and the improvement data generated by S6 continuously provide feedback for model updates. The model updates, in turn, adjust the thresholds, weights, and sampling strategies of the monitoring scheme, so that the real-time monitoring scheme can maintain the traceability of dispersion and migration behavior under conditions of paper type switching, auxiliary agent system switching, and water quality fluctuations. The updated monitoring scheme is output in the form of parameter files or rule sets. The rule sets include key parameter control windows, single-dimensional deviation index calculation methods, combined deviation index weights, environmental state coding boundaries, and sampling strategies for triggering and re-evaluation, for online system execution.

[0080] like Figure 4 As shown, the top left subplot presents a comparison curve of the stability quantification value changing over time. The dashed line represents the stability threshold, and the shaded area represents the abnormal operating condition range. Under abnormal operating conditions, the stability quantification value of the traditional method drops significantly and remains below the threshold for a long period, while the stability quantification value of the method of this invention remains near or above the threshold with small fluctuations. The top right subplot shows a comparison of dispersion uniformity under different test conditions, showing that the method of this invention maintains higher dispersion uniformity under disturbances such as temperature fluctuations, pH changes, high salinity, and additive fluctuations. The bottom left subplot shows a comparison of anomaly detection response time, indicating that the method of this invention significantly shortens the response time in stages such as aggregation initiation detection, migration anomaly identification, uneven adhesion warning, and overall state assessment. The bottom right subplot is a radar chart for comprehensive dyeing quality evaluation, reflecting that the method of this invention is superior to the traditional method in terms of batch consistency, fiber coverage uniformity, color reproducibility, wash / light fastness, and process stability, thus verifying the improvement effect of the method of this invention on the stability testing and process monitoring of paper dyes.

[0081] Figure 3and Figure 4 Traditional methods refer to routine testing and empirical control of paper dye stability and migration behavior: in wet-end production or simulation systems, only routine monitoring of temperature, pH, conductivity, etc. is carried out, and offline sampling is used to measure residual liquor color, exhaustion rate, color intensity, and other results indicators when necessary. The dosage and sequence of additives are adjusted based on experience or single-factor experiments. No trajectory data collection and time-series analysis of dispersion-migration-adhesion are established, no feature library matching and parameter combination optimization are performed, and no stability quantification value or model closed-loop update based on improvement data is formed.

[0082] Preferably, to further improve the objectivity, balance, and stability of the model update process, the following optimization measures can be adopted: The quantified improvement components are converted into adjustment quantities according to the preset mapping rules. However, relying entirely on manually setting the rules can easily introduce subjectivity. Therefore, after accumulating multiple improvement data records, an online parameter identification method can be used to automatically optimize the mapping rules. In this method, the error between the model output and the observed improvement index is used as the cost function, and recursive least squares or Bayesian updates are used to automatically estimate the internal parameter adjustment quantities, thereby improving the objectivity and adaptability of the mapping rules.

[0083] If the improvement data only covers a small number of operating condition combinations, it will lead to model update bias. Therefore, the improvement data can be sampled in layers according to the environmental state number, and when updating the model, the update step size of the parameters corresponding to the sparse environmental states can be reduced, thereby ensuring that the model is updated evenly under all operating conditions and avoiding bias.

[0084] Frequent model parameter updates can cause fluctuations in system thresholds. Therefore, hysteresis and smoothing mechanisms can be introduced into the thresholds and weights of the model parameters. Simultaneously, a rollback version of the model can be set to prevent abnormal updates from degrading system performance, thus ensuring the smoothness of the model update process and system safety. For example, introducing hysteresis and smoothing mechanisms into the thresholds and weights of the model parameters can be achieved by setting a historical state buffer to smooth parameter updates. When new parameter values ​​change significantly, the update step size is limited by comparing the difference between the current and historical values ​​to avoid abrupt changes. The smoothing operation can employ weighted averaging or exponential decay methods to gradually stabilize parameter changes and prevent the system from overreacting to instantaneous fluctuations. For example, setting a rollback version of the model allows saving a copy of the current model after each parameter update. When a new update causes a significant performance degradation, the system automatically reverts to the previous stable version, thus preventing unstable parameter settings from affecting the overall system performance.

[0085] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A highly stable papermaking dye, characterized in that, Paper dyes include: The dye matrix has a chemical structure that is stable in high temperature and acid / alkali environments, and is selected from organic dyes, inorganic dyes, or combinations thereof. The dye matrix has good dispersibility. At least one stabilizer that enhances the dye’s tolerance to fluctuations in water quality, temperature and pH during the papermaking process, wherein the stabilizer is selected from polymer stabilizers, surfactant stabilizers, or a combination of polymer stabilizers and surfactant stabilizers; At least one auxiliary agent, said auxiliary agent being able to adjust the adhesion properties and color fastness of the dye, said auxiliary agent being able to adjust the solubility of the dye in an acidic or alkaline environment, ensuring the stable migration and uniform distribution of the dye in the papermaking process.

2. The high-stability papermaking dye according to claim 1, characterized in that, The dye matrix is ​​selected from reactive dyes, disperse dyes, or acid dyes; The polymer stabilizer is polyvinyl alcohol, polyacrylic acid, or a combination of polyvinyl alcohol and polyacrylic acid; The surfactant stabilizer is sodium dodecylbenzenesulfonate, hexadecyltrimethylammonium chloride, or a combination of sodium dodecylbenzenesulfonate and hexadecyltrimethylammonium chloride. The adjuvant is one or more chelating agents, metal salts, or regulators.

3. A performance testing method for a highly stable papermaking dye, characterized in that, The method includes: S1. Collect paper dye sample data and auxiliary influence parameters in a multi-component papermaking environment, extract environmental variable characteristics and related influence characteristics from the collected data, and obtain an initial environmental variable set; S2. The initial set of environmental variables is classified using a machine learning clustering algorithm to determine the dye dispersion state category; S3. When the dye dispersion state category indicates an aggregation anomaly, acquire migration behavior trajectory data and analyze the dynamic process change law; S4. Obtain fiber adhesion distribution information based on the dynamic process change law, analyze the correlation between the fiber adhesion distribution information and the auxiliary agent influence parameters through a support vector machine model, and determine the migration path optimization point; S5. Extract real-time monitoring indicators from the migration path optimization points, and use information fusion technology to integrate the influence of environmental variables on the real-time monitoring indicators to obtain the stability quantification value of paper dye stability. S6. If the stability quantification value is lower than the preset stability threshold, the simulation parameters of the multi-component papermaking environment are adjusted, and the improvement of dye dispersion and migration behavior after parameter adjustment is evaluated. S7. Update the dynamic process model based on the improvement data, and generate the final real-time monitoring scheme based on the updated model, wherein the dynamic process model is used to simulate the dispersion and migration behavior of dyes in the multi-component environment of papermaking.

4. The method according to claim 3, characterized in that, S1 includes: Data on the dye samples and parameters affecting the additives in the multi-component environment of papermaking are collected using sensors or measuring devices. The environmental variable features are extracted from the dye sample data, including water quality ion features, temperature change features, pH value change features and / or ion strength features; The relevant influence features are extracted from the influence parameters of the auxiliaries. The relevant influence features include the influence characteristics of the auxiliaries on the dye affinity, the critical micelle concentration characteristics of the auxiliaries, and / or the complexing ability of the auxiliaries on metal ions. The environmental variable features are integrated with the relevant influence features to obtain the initial environmental variable set.

5. The method according to claim 3, characterized in that, S2 include: A machine learning clustering algorithm is used to perform cluster analysis on a multidimensional feature dataset composed of multiple environmental variable features from the initial environmental variable set; Based on the cluster analysis results, the corresponding dye dispersion state category is determined. The dye dispersion state category is selected from a set including normal dispersion state and anomalous aggregation state. The normal dispersion state is the state in which the characteristic value is within the standard range, and the anomalous aggregation state is the state in which the characteristic value deviates from the standard range and indicates that the dye has aggregated.

6. The method according to claim 3, characterized in that, S3 includes: When the dye dispersion state category is determined to be an aggregation anomaly state, migration behavior trajectory data associated with the aggregation anomaly state is obtained. The migration behavior trajectory data includes a sequence of the position of dye particles in the multi-component papermaking environment over time. The migration trajectory data was analyzed using time series analysis methods to identify patterns of change in the data; The change pattern is matched with a preset migration dynamic feature library, and the dynamic process change law is determined based on the matching result. The dynamic process change law is used to qualitatively characterize the migration dynamic behavior of dyes in a multi-component papermaking environment.

7. The method according to claim 3, characterized in that, S4 include: Based on the dynamic process change pattern, fiber adhesion distribution information on the surface of pulp fibers / paper fibers is collected directionally from the multi-component environment of papermaking. The fiber adhesion distribution information includes the adhesion density and / or position distribution data of dyes on the fiber surface. The correlation between the fiber adhesion distribution information and the auxiliary agent influence parameters was analyzed using a support vector machine model to obtain the correlation quantification results. Based on the correlation quantification results, key parameter combinations that have a significant impact on dye migration behavior are identified, and these key parameter combinations are determined as the migration path optimization points.

8. The method according to claim 3, characterized in that, S5 include: Data scanning was performed on the migration path optimization points to extract real-time monitoring indicators related to dye behavior; Information fusion technology is used to integrate the effects of temperature change, pH value and ionic strength on the real-time monitoring index. An influence matrix is ​​generated based on the fusion results. The influence matrix records the change range of each real-time monitoring index under different combinations of variables. Based on the influence matrix, the stability quantification value of paper dye stability is calculated using the geometric mean method.

9. The method according to claim 3, characterized in that, S6 include: Determine whether the stability quantification value is lower than the preset stability threshold. If so, according to the preset adjustment strategy, map the stability quantification value to the adjustment instructions for the simulation parameters of the multi-component environment of papermaking. The simulation parameters include at least temperature change, pH value and ionic strength. After adjusting the corresponding simulation parameters according to the adjustment instructions, new dispersion uniformity and migration rate indices are calculated in the adjusted multi-component environment by re-collecting dye sample data. The new dispersion uniformity and new migration rate indices are compared with the corresponding data before parameter adjustment to obtain quantified improvement data, which represents the degree of improvement in dye dispersion and migration behavior after adjustment.

10. The method according to claim 3, characterized in that, S7 includes: The improvement data is analyzed to extract the quantitative improvement components. According to a preset mapping rule, the quantified improvement component is converted into an adjustment amount for the internal parameters of the dynamic process model; The corresponding parameters of the dynamic process model are updated based on the adjustment amount to obtain the updated dynamic process model; Run the updated dynamic process model, set or optimize the operating parameters for real-time monitoring based on the model output, and generate the real-time monitoring scheme for real-time tracking of dye behavior stability.

Images