A solvent dyeing method and apparatus incorporating solvent recovery

Abstract

This invention discloses a solvent dyeing method and apparatus incorporating solvent recovery, relating to the field of solvent treatment technology for textile materials. The solvent dyeing method includes three stages: pretreatment, dyeing, and post-treatment. A stepwise distillation method is employed to address the problem of inefficient solvent recovery in traditional simple distillation. By controlling the purity of the recovered solvent, potential problems such as product quality degradation and equipment damage caused by directly using insufficiently pure recovered solvent in production are resolved. An online prediction system is used to predict the purity of the recovered solvent. When the predicted purity value is outside the purity threshold range, the system issues an early warning signal, prompting operators to inspect and repair various equipment components, thus enabling proactive prevention.

Description

Technical Field

[0001] This invention relates to the field of solvent treatment of textile materials, and more particularly to a solvent dyeing method and apparatus that incorporates solvent recovery. Background Technology

[0002] Polyester (polyethylene terephthalate, PET) is a synthetic fiber with low hydrophilicity and a relatively compact fiber structure, therefore disperse dyes are typically used in its dyeing process. Disperse dyes bind to polyester fibers through hydrogen bonds and van der Waals forces. The dyeing process requires high temperature and pressure to improve dye solubility and diffusion rate. However, disperse dyes have extremely low solubility in water, mainly existing as tiny particles suspended in the dye bath, which can easily aggregate and affect dyeing uniformity. Solvents have excellent dissolving power for disperse dyes, dissolving or decomposing dye molecules into smaller particles, increasing the effective dye concentration in the dye bath, reducing dye particle aggregation, ensuring dye bath stability, and enabling effective dyeing at lower temperatures.

[0003] Therefore, in the process of dyeing polyester, in order to improve the solubility and penetration of dyes and improve dyeing efficiency and quality, a certain amount of solvent is added. However, this also faces challenges such as solvent recovery, cost and environmental regulations.

[0004] In existing polyester dyeing processes, waste liquid is typically disposed of directly after dyeing, or the solvent is simply recovered through distillation and then directly reintroduced into production. This inefficient solvent recovery leads to solvent waste, increased production costs, and environmental pollution. Furthermore, during recovery, the purity of the recovered solvent fluctuates due to variations in dyeing and recovery process parameters for each batch, and is affected by parameter changes or equipment malfunctions. The recovered solvent needs to be reused in production; if solvent purity is not controlled, it can lead to potential problems such as decreased product dyeing quality. Therefore, purity testing is necessary. However, when testing reveals that the recovered purity does not meet recovery standards, retrospective inspection of process parameters, processes, and equipment is a reactive and excessively labor-intensive process. Summary of the Invention

[0005] To address the aforementioned problems, the present invention aims to provide a solvent dyeing method and apparatus incorporating solvent recovery. It employs a staged distillation method to solve the problem of inefficient solvent recovery in traditional simple distillation. By controlling the purity of the recovered solvent, it effectively solves the potential problems of product quality degradation and equipment damage caused by directly using insufficiently pure recovered solvent in production. The system is used for predicting the purity of the recovered solvent. When the predicted purity value is outside the purity threshold range, the system issues an early warning signal, prompting operators to inspect and repair various equipment components, thus enabling proactive prevention.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solution:

[0007] A solvent staining method incorporating solvent recovery includes three steps: pretreatment S1, staining S2, and posttreatment S3, wherein:

[0008] Pretreatment S1: The polyester fibers are cleaned in the cleaning reagent contained in the pretreatment tank;

[0009] Dyeing S2: Add the pretreated polyester fibers, dye and solvent to the dyeing kettle in a certain proportion. The ratio of polyester fibers to mixed solvent is 1:5-1:15, and the ratio of dye to polyester fibers is 0.5%-5%. Dyeing is carried out at a temperature of 100-140℃ and a pressure of 0.3-0.5MPa for 30-60 minutes.

[0010] Post-processing S3: Solvent is recovered using a recovery device.

[0011] Furthermore, the cleaning reagent contained in the pretreatment tank is hexane.

[0012] Furthermore, the solvent added to the dyeing vessel during the dyeing process is the main solvent dimethyl sulfoxide (DMSO) and the co-solvent dimethyl carbonate (DMC), with a ratio of 8:2.

[0013] Furthermore, the solvents recovered in the post-processing include hexane, dimethyl sulfoxide (DMSO), and dimethyl carbonate (DMC).

[0014] A solvent dyeing apparatus incorporating solvent recovery includes a pretreatment tank, a dyeing vessel, and a recovery device. The pretreatment tank is connected to the dyeing vessel, and the dyeing vessel is connected to the recovery device.

[0015] Furthermore, the recovery device includes a collection device, a filtration device, and a distillation device. The dyeing vessel is connected to the collection device, which is connected to the filtration device, which is connected to the distillation device.

[0016] Furthermore, the filtration device includes a two-stage filtration system consisting of a 10μm pore size filter membrane and a 0.2μm precision filter membrane. The collection device is connected to the 10μm pore size filter membrane, which is connected to the 0.2μm precision filter membrane.

[0017] Furthermore, the distillation apparatus includes a primary distillation unit, a secondary distillation unit, and a tertiary distillation unit. A 0.2μm precision filter membrane is connected to the primary distillation unit, the primary distillation unit is connected to the secondary distillation unit, and the secondary distillation unit is connected to the tertiary distillation unit.

[0018] Furthermore, the primary distillation unit is a primary distillation column equipped with a heating kettle and a condensation system; the secondary distillation unit is a secondary distillation column equipped with a precision temperature control system; and the tertiary distillation unit is a vacuum distillation column equipped with a vacuum system and made of corrosion-resistant materials.

[0019] Furthermore, the solvent staining apparatus also includes an online prediction system, which includes a data collection module, a data preprocessing module, a model selection and training module, a model evaluation and interpretation module, and a model deployment and dynamic update module, which are connected in sequence.

[0020] The data collection module is used to collect historical data on solvent dyeing and recycling related production processes.

[0021] The data preprocessing module includes a data cleaning submodule, a feature transformation submodule, and a feature filtering submodule;

[0022] The data cleaning submodule processes missing values ​​and abnormal data;

[0023] The feature conversion submodule performs Z-score normalization on continuous variables such as temperature and pressure, and uses one-thermal encoding on discrete variables such as dye type and filter batch.

[0024] The feature filtering submodule is used to remove redundant features to simplify the model;

[0025] The model selection and training module includes a dataset partitioning submodule, a model selection submodule, and a model training submodule.

[0026] The dataset partitioning submodule is divided into training set, validation set and test set in a 7:2:1 ratio, and SMOTE oversampling or class-weighting is used to handle sample imbalance.

[0027] The model selection submodule uses Gradient Boosting Regression Tree (GBRT) to predict the specific value of solvent purity;

[0028] The model training submodule uses 5-fold cross-validation to avoid overfitting and uses grid search or Bayesian optimization to tune hyperparameters.

[0029] The model evaluation and interpretation module includes an indicator evaluation submodule and a model interpretation submodule;

[0030] The indicator evaluation submodule uses mean squared error (MSE) and mean absolute error (MAE) to evaluate the deviation between the predicted value and the true purity.

[0031] The model interpretation submodule uses SHAP values ​​or feature importance analysis to identify key influencing factors;

[0032] The model deployment and dynamic update module includes a deployment method submodule and a dynamic update submodule;

[0033] The deployment method submodule encapsulates the model as an API interface, embeds it into the production monitoring system, receives process parameters in real time, and outputs specific values ​​of solvent purity. When the predicted specific value of solvent purity is not within the purity threshold range, an alarm is triggered. If the predicted specific value of solvent purity meets the purity threshold range, the actual recycling work is carried out normally.

[0034] The dynamic update submodule updates the model using incremental learning after collecting 50 batches of new data, adapts to process fluctuations, periodically verifies model performance, and retrains the model when the recall rate is below 90%.

[0035] Compared with the prior art, the present invention has at least one of the following technical effects:

[0036] 1. This invention recovers solvents from the waste liquid after dyeing and the residual solution on the surface of the dyed polyester fibers. The solvent recovery system is more complete, reducing the impact of solvents on the dyed polyester fibers and preventing problems such as inadequate quality control of polyester fibers.

[0037] 2. The stepwise distillation of the present invention can achieve continuous operation, avoid the fluctuations of intermittent operation affecting product purity, increase the throughput of large plants, and achieve efficient solvent recovery;

[0038] 3. This invention uses a heat pump system to transfer the heat of the steam at the top of the tower to the bottom of the tower for heating, which significantly reduces energy consumption;

[0039] 4. The online prediction system of the present invention establishes a gradient boosting regression tree (GBRT) to predict the specific value of solvent purity according to the preset process parameters, and compares it with the preset purity threshold range of the recovered solvent. If it meets the purity threshold range, the recovery operation is carried out according to the preset process parameters. When the predicted specific value of solvent purity is not within the purity threshold range (such as 97%-99%), an alarm is triggered, prompting the operator to adjust the parameters (such as reducing the tower temperature) or check the equipment. Prediction enables prevention in advance, avoiding blind recovery that leads to non-compliance. Prevention in advance reduces the passivity of post-event detection and effectively reduces the workload of post-event processing.

[0040] 5. The online prediction system of this invention calculates the prediction confidence interval (e.g., 95%-100% confidence interval) of GBRT using the Bootstrap method. It uses whether the actual value exceeds the confidence interval as the criterion, dynamically adjusting the threshold according to model error. This upgrades GBRT point prediction to "interval prediction with uncertainty," achieving more accurate judgment by combining the compliance range and process parameter fluctuations. Compared to a single threshold, the confidence interval better reflects the reliability of the prediction, providing a more comprehensive basis for decision-making regarding "whether compliance is achieved" and "whether there are risks of unreasonable processes or equipment failures," thus improving prediction accuracy. Attached Figure Description

[0041] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0042] Figure 1 This is a flowchart of a solvent dyeing method combining solvent recovery according to the present invention;

[0043] Figure 2 This is a structural diagram of a solvent dyeing device incorporating solvent recovery according to the present invention;

[0044] Figure 3 This is a module architecture diagram of the online prediction system of the present invention;

[0045] Figure 4 This is a module architecture diagram of the prediction confidence interval system of the present invention.

[0046] Explanation of reference numerals in the attached figures

[0047] Pretreatment tank-1; dyeing vessel-2; recovery device-3; collection device-4; filtration device-5; distillation device-6; 10μm pore size filter membrane-7; 0.2μm precision filter membrane-8; primary distillation column-9; secondary distillation column-10; vacuum distillation column-11; online prediction system-12; data collection module-13; data preprocessing module-14; model selection and training module-15; model evaluation and interpretation module-16; model deployment and dynamic update module-17; prediction confidence interval system-18; Bootstrap integrated model library construction module-19; single sample confidence interval calculation module-20; judgment module-21. Detailed Implementation

[0048] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0049] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0050] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0051] As used in this application specification and the appended claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrase "if determined" or "if detected [the described condition or event]" may be interpreted, depending on the context, as meaning "once determined," "in response to determination," "once detected [the described condition or event]," or "in response to detection [the described condition or event]."

[0052] Furthermore, in the description of this application and the appended claims, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0053] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0054] Example 1

[0055] A solvent staining method incorporating solvent recovery, such as Figure 1-2As shown, the solvent dyeing method includes three steps: pretreatment S1, dyeing S2, and posttreatment S3. Pretreatment S1 involves briefly cleaning polyester fibers in a pretreatment tank 1 containing hexane, whose highest boiling point is 68.7℃. Hexane removes oil, impurities, and spinning auxiliaries from the surface of the polyester fibers, improving the fibers' adsorption capacity for dye. Dyeing S2 involves adding the pretreated polyester fibers, dye, and solvent to a dyeing kettle 2 in a specific ratio: polyester fiber to mixed solvent is 1:5-1:15, and dye to polyester fiber is 0.5%-5%. Dyeing is carried out at a temperature of 100-140℃ and a pressure of 0.3-0.5 MPa for 30-60 minutes. Stirring ensures the dye is evenly dispersed and penetrates into the fibers. The solvent is primarily dimethyl sulfoxide (DMSO) and co-solvent dimethyl carbonate (DMC), with a ratio of 8:2. DMSO has a boiling point of 189°C, and DMC has a boiling point of 90.3°C. Due to its higher boiling point, DMSO maintains a stable high-temperature environment during dyeing process S2, enhancing the swelling and penetration of the polyester fibers, promoting better diffusion and binding of dye molecules into the fiber interior. Simultaneously, it works synergistically with DMC to optimize the solvent system's solubility and dyeing efficiency, ensuring more uniform and effective dyeing. Post-treatment S3 utilizes a recovery device 3 to recover the solvent, which includes hexane, DMSO, and DMC. The recovery device 3 includes a collection device 4, a filtration device 5, and a distillation device 6. The collection device 4 collects the waste liquid after dyeing and the residual solution on the surface of the dyed polyester fibers. The filtration device 5 includes a two-stage filtration system. First, fiber debris is removed through a 10μm pore size filter membrane 7, and then undissolved dye particles are filtered through a 0.2μm precision filter membrane 8, ensuring the residual liquid is clear. The distillation device 6 includes a primary distillation unit, a secondary distillation unit, and a tertiary distillation unit. The primary distillation unit is a primary distillation column 9, equipped with a heating kettle and a condensation system. The temperature at the top of the primary column is controlled at 68-69℃, and the temperature at the bottom of the primary column does not exceed 80℃, ensuring hexane volatilization while preventing premature volatilization of dimethyl carbonate (DMC). The secondary distillation unit is a secondary distillation column 10, equipped with a precision temperature control system. The temperature at the top of the secondary column is controlled at 90-91℃, and the temperature at the bottom of the secondary column does not exceed 120℃, ensuring dimethyl carbonate (DMC) volatilization while preventing overheating of dimethyl sulfoxide (DMSO). The distillation column is equipped with a heat pump system to transfer the heat of the vapor at the top of the column to the bottom for heating, significantly reducing energy consumption. The three-stage distillation unit is a vacuum distillation column 11, equipped with a vacuum system and made of corrosion-resistant materials. The filtered filtrate enters the first-stage distillation column 9, where, after heating, the low-boiling-point hexane evaporates preferentially. The vapor rises to the top of the column and is condensed and liquefied to obtain hexane.The residue from the first-stage distillation column enters the second-stage distillation column 10, where it is heated and dimethyl carbonate (DMC) evaporates and condenses at the top of the second-stage column to obtain dimethyl carbonate (DMC). The residue from the second-stage distillation column enters the vacuum distillation column 11, where it is heated under reduced pressure, dimethyl sulfoxide (DMSO) evaporates, and is collected by condensation to obtain dimethyl sulfoxide (DMSO).

[0056] Working principle: A solvent dyeing method and apparatus that combines solvent recovery. The working process of the apparatus is as follows: waste liquid is collected into a waste liquid collection tank. The waste liquid is filtered through a two-stage filtration system to remove impurities. Taking advantage of the different volatilization capabilities of the components in the waste liquid, corresponding distillation equipment is selected. The temperature of the top and bottom of the column is strictly controlled to ensure that the solvent evaporates while reducing energy consumption and avoiding affecting the solvent in the next stage. The target solvent is obtained through step-by-step distillation.

[0057] Example 2

[0058] Online prediction system 12, such as Figure 3 As shown, the system includes a data collection module 13, a data preprocessing module 14, a model selection and training module 15, a model evaluation and interpretation module 16, and a model deployment and dynamic update module 17, which are connected in sequence.

[0059] The data collection module 13 collects historical data on solvent dyeing and recovery processes, including fiber type, fiber weight, fiber diameter, initial surface oil content, spinning auxiliaries, hexane cleaning time, cleaning temperature, initial hexane purity, hexane content, dimethyl sulfoxide (DMSO) content, dimethyl carbonate (DMC) content, initial ratio of DMSO to DMC, dyeing temperature, pressure, dyeing time, dye type and concentration, dye liquor viscosity, dye liquor pH, pressure difference before and after filtration, filter membrane flux, top temperature, bottom temperature, reflux ratio, heating power, vacuum degree, distillation pressure of the first to third stage distillation, and purity and impurity content of each solvent after recovery (such as dye residue, fiber debris, other solvents, ppm). The above data is detected by corresponding sensors or manually input through the human-machine interface unit of this module. For example, gas chromatography is used to detect the original concentration of each solvent in the filtered waste liquid and the final concentration of the solvent obtained after distillation, and sensors are used to monitor the temperature and pressure of each stage of distillation in real time.

[0060] The data preprocessing module 14 includes a data cleaning submodule, a feature transformation submodule, and a feature filtering submodule;

[0061] The data cleaning submodule fills in missing values ​​(such as missing reflux ratios in a certain batch) with the mean and / or median, identifies abnormal data (such as erroneous values ​​such as a sudden increase in tower top temperature to 100°C) using the IQR method (interquartile range), and corrects or removes them in combination with process logic.

[0062] The feature transformation submodule performs Z-score standardization (mean 0, standard deviation 1) on continuous variables such as temperature and pressure to avoid the model being affected by the dimensions, and uses one-hot encoding for discrete variables such as dye type and filter batch.

[0063] The feature filtering submodule removes redundant features (such as "device number", which is irrelevant to the recycling effect) to reduce model complexity.

[0064] The model selection and training module 15 includes a dataset partitioning submodule, a model selection submodule, and a model training submodule;

[0065] The dataset partitioning submodule is divided into a training set (for model learning), a validation set (for evaluating model performance), and a test set (for evaluating generalization ability) in a 7:2:1 ratio. If the samples are imbalanced (e.g., 90% of the samples meet the criteria), SMOTE oversampling (adding non-compliant samples) or class-weighting is used to adjust the class weights.

[0066] The model selection submodule uses Gradient Boosting Regression Tree (GBRT) to predict the specific value of solvent purity;

[0067] The model training submodule uses 5-fold cross-validation (dividing the training set into 5 parts and validating them in turn) to avoid overfitting, and uses grid search or Bayesian optimization to tune hyperparameters, such as n_estimators in random forest and learning_rate in XGBoost.

[0068] Model evaluation and interpretation module 16 includes an indicator evaluation submodule and a model interpretation submodule;

[0069] The indicator evaluation submodule uses mean squared error (MSE) and mean absolute error (MAE) to evaluate the deviation between the predicted value and the true purity (e.g., MAE ≤ 0.5% indicates that the prediction error is within 0.5%).

[0070] The model interpretation submodule uses SHAP values ​​or feature importance analysis to identify key influencing factors, such as: "The influence weight of the first-stage distillation column top temperature (68-69℃) on hexane purity reaches 30%", and "When the second-stage distillation reflux ratio is >1.2, the DMC compliance rate is improved by 25%", and outputs a visualization report (such as a partial dependency graph) to guide the optimization of process parameters (such as stabilizing the column top temperature at 68.5℃ to improve the hexane recovery rate).

[0071] The model deployment and dynamic update module 17 includes a deployment method submodule and a dynamic update submodule;

[0072] The deployment method submodule encapsulates the model as an API interface, embeds it into a production monitoring system (such as SCADA), receives process parameters (such as distillation temperature, pressure, etc.) in real time, and outputs specific values ​​of solvent purity (such as 95% purity of dimethyl sulfoxide (DMSO)).

[0073] When the predicted solvent purity value is outside the purity threshold range (e.g., 97%-99%), an alarm is triggered, prompting the operator to adjust parameters (e.g., reduce the tower temperature) or check the equipment. If the predicted solvent purity value is within the purity threshold range, the actual recovery operation proceeds normally.

[0074] The dynamic update submodule updates the model with incremental learning (such as the update method of XGBoost) after collecting 50 batches of new data to adapt to process fluctuations (such as changes in raw material batches), periodically verifies the model performance, and retrains the model (with the need to supplement new samples) when the recall rate is below 90%.

[0075] This application uses a gradient boosting regression tree (GBRT) to predict the specific value of solvent purity based on pre-set real-time process parameters. The predicted value is then compared to a pre-set purity threshold range for the recovered solvent. If the value falls within the threshold range, recovery is performed according to the pre-set process parameters. When the predicted solvent purity value is outside the threshold range (e.g., 97%-99%), an alarm is triggered, prompting operators to adjust parameters (e.g., lower the reboiler temperature) or inspect the equipment. This predictive approach prevents substandard situations caused by blind recovery. Real-time process parameters include fiber type, fiber weight, and fiber diameter. The parameters to be considered include: diameter, initial surface oil content, spinning aids, hexane cleaning time, cleaning temperature, initial hexane purity, hexane content, dimethyl sulfoxide (DMSO) content, dimethyl carbonate (DMC) content, initial ratio of DMSO to DMC, dyeing temperature, pressure, dyeing time, dye type and concentration, theoretical dye liquor viscosity (selective input), theoretical dye liquor pH (selective input), pressure difference before and after the filter membrane, filter membrane flux, top temperature of the first to third stage distillation column, bottom temperature of the column, theoretical reflux ratio (selective input), heating power, vacuum degree, and distillation pressure.

[0076] To improve the accuracy of the predicted solvent purity, a prediction confidence interval is generated through Bootstrap integration, and the result is judged in conjunction with the threshold range. The prediction confidence interval generation system 18, which integrates the Bootstrap GBRT model, includes a Bootstrap integrated model library module 19, a single-sample confidence interval calculation module 20, and a judgment module 21, wherein the Bootstrap integrated model library module 19, the single-sample confidence interval calculation module 20, and the judgment module 21 are connected in sequence.

[0077] The Bootstrap integrated model library module 19 includes a sample resampling submodule, multiple GBRT model training modules, and a model consistency verification submodule.

[0078] The sample resampling submodule generates N different subsets from the original training set using the Bootstrap sampling with replacement method. The sample size of each subset is the same as that of the original dataset, and duplicate samples are allowed. The original training set includes no less than 300 batches of data, and N=50~100.

[0079] Multiple GBRT model training modules train an independent GBRT model based on each subset of the dataset, forming a "model library" containing N models. During training, the hyperparameters of each model, such as the number of trees and the learning rate, are kept consistent, and model diversity is introduced only through sample differences.

[0080] The model consistency verification submodule calculates the prediction error of N models on the test set. If the standard deviation of the error is ≤0.3% (set according to the process accuracy requirements), it indicates that the model library is stable. If the deviation is too large (e.g., standard deviation >0.5%), the number of N needs to be increased (e.g., increased to 100) or the hyperparameters need to be adjusted and retrained.

[0081] The single-sample confidence interval calculation module 20 includes a multi-model parallel prediction submodule, an interval statistics submodule, and an interval validity verification submodule.

[0082] The multi-model parallel prediction submodule takes a new input sample (such as the process parameters of a certain batch) and inputs it into all N GBRT models in the model library to obtain N predicted purity values ​​(denoted as y1, y2, ..., yN).

[0083] The interval statistics submodule sorts the N predicted values ​​in ascending order and takes the quantile corresponding to a specified confidence level (e.g., 95%) as the upper and lower limits of the interval.

[0084] The lower limit L = the 2.5th quantile (i.e., the 0.025 × Nth value after sorting);

[0085] The upper limit U = the 97.5th quantile (i.e., the 0.975 × Nth value after sorting);

[0086] Example: When N = 100, the 3rd value is taken as the lower limit and the 97th value as the upper limit to form a 95% confidence interval.

[0087] The interval validity verification sub-module randomly selects 100 test set samples and calculates whether the confidence interval of each sample contains its true purity value. If the coverage rate ≥ 90% (close to the set 95% confidence level), the interval calculation is valid; otherwise, N needs to be increased or the model needs to be optimized.

[0088] The judgment module 21 includes an overlapping analysis sub-module and an output sub-module for the interval and the compliance range.

[0089] The overlapping analysis sub-module for the interval and the compliance range presets the compliance range as [Tlow, Thigh] (such as hexane [98%, 100%]) and determines it in different cases according to the position of the confidence interval [L, U], including:

[0090] Compliance: L ≥ Tlow (for example, the lower limit 98.2% ≥ 98%), and the interval completely overlaps with the compliance range (such as [98.2%, 99.1%] ⊆ [98%, 100%]).

[0091] Non-compliance, U < Tlow (for example, the upper limit 97.8% < 98%), and the interval is completely below the compliance range.

[0092] Possible compliance: The interval partially overlaps (such as [97.8%, 98.5%], the lower limit < 98% but the upper limit ≥ 98%). At this time, calculate:

[0093] If P ≥ 80%, judge as compliant; if P < 80%, judge as non-compliant, and backtrack to adjust each process parameter or check the equipment, etc.

[0094] The output sub-module is used to display the above classification results (such as "compliant") and the confidence interval together in the monitoring system to assist the operator in making decisions.

[0095] The prediction confidence interval of GBRT (such as 95% - 100% confidence interval) is calculated by the Bootstrap method, and "whether the actual value exceeds the confidence interval" is used as the judgment basis, so that the threshold is dynamically adjusted according to the model error, upgrading the point prediction of GBRT to "interval prediction with uncertainty", and combining the compliance range and process parameter fluctuations to achieve more accurate judgment. Compared with a single threshold, the confidence interval can better reflect the reliability of the prediction, provide a more comprehensive decision-making basis for "whether it is compliant", "whether there are risks of unreasonable processes or equipment failures, etc.", and improve the accuracy of the prediction.

[0096] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of the present invention.

Claims

1. A combination solvent dyeing apparatus with solvent recovery, characterized in that, It includes a pretreatment tank (1), a dyeing kettle (2) and a recovery device (3). The pretreatment tank (1) is connected to the dyeing kettle (2), and the dyeing kettle (2) is connected to the recovery device (3). The recovery device (3) includes a collection device (4), a filtration device (5) and a distillation device (6). The dyeing kettle (2) is connected to the collection device (4), the collection device (4) is connected to the filtration device (5), and the filtration device (5) is connected to the distillation device (6). It also includes an online prediction system (12), which includes a data collection module (13), a data preprocessing module (14), a model selection and training module (15), a model evaluation and interpretation module (16), and a model deployment and dynamic update module (17). The data collection module (13), the data preprocessing module (14), the model selection and training module (15), the model evaluation and interpretation module (16), and the model deployment and dynamic update module (17) are connected in sequence. The data collection module (13) is used to collect historical data on solvent dyeing and recycling related production processes; The data preprocessing module (14) includes a data cleaning submodule, a feature transformation submodule, and a feature filtering submodule; The data cleaning submodule processes missing values ​​and abnormal data; The feature conversion submodule performs Z-score standardization on continuous variables of temperature and pressure, and uses unique thermal encoding on discrete variables of dye type and filter membrane batch. The feature filtering submodule is used to remove redundant features to simplify the model; The model selection and training module (15) includes a dataset partitioning submodule, a model selection submodule, and a model training submodule; The dataset partitioning submodule is divided into training set, validation set and test set in a 7:2:1 ratio, and SMOTE oversampling or class-weighting is used to handle sample imbalance. The model selection submodule uses gradient boosting regression trees to predict specific values ​​of solvent purity. The model training submodule uses 5-fold cross-validation to avoid overfitting and uses grid search or Bayesian optimization to tune hyperparameters. The model evaluation and interpretation module (16) includes an indicator evaluation submodule and a model interpretation submodule; The indicator evaluation submodule uses mean square error and mean absolute error to evaluate the deviation between the predicted value and the true purity. The model interpretation submodule uses SHAP values ​​or feature importance analysis to identify key influencing factors; The model deployment and dynamic update module (17) includes a deployment method submodule and a dynamic update submodule; The deployment method submodule encapsulates the model as an API interface, embeds it into the production monitoring system, receives process parameters in real time, and outputs specific values ​​of solvent purity. When the predicted specific value of solvent purity is not within the purity threshold range, an alarm is triggered. If the predicted specific value of solvent purity meets the purity threshold range, the actual recycling work is carried out normally. The dynamic update submodule updates the model using incremental learning after collecting 50 batches of new data, adapts to process fluctuations, periodically verifies model performance, and retrains the model when the recall rate is below 90%.

2. The apparatus for solvent dyeing according to claim 1, wherein The filtration device (5) includes a two-stage filtration system, namely a 10μm pore size filter membrane (7) and a 0.2μm precision filter membrane (8). The collection device (4) is connected to the 10μm pore size filter membrane (7), which is connected to the 0.2μm precision filter membrane (8).

3. The apparatus for solvent dyeing according to claim 2, wherein The distillation apparatus (6) includes a primary distillation apparatus, a secondary distillation apparatus and a tertiary distillation apparatus. A 0.2μm precision filter membrane (8) is connected to the primary distillation apparatus, the primary distillation apparatus is connected to the secondary distillation apparatus, and the secondary distillation apparatus is connected to the tertiary distillation apparatus.

4. The apparatus for solvent dyeing according to claim 3, wherein The first-stage distillation unit is a first-stage distillation column (9), equipped with a heating kettle and a condensation system; the second-stage distillation unit is a second-stage distillation column (10), equipped with a precision temperature control system; and the third-stage distillation unit is a vacuum distillation column (11), equipped with a vacuum system, and made of corrosion-resistant materials.

Images