Dyeing process optimization method and system based on intelligent control

Abstract

The invention relates to the technical field of intelligent control, and discloses a dyeing process optimization method and system based on intelligent control. The dyeing process optimization method based on intelligent control comprises the following steps: acquiring key parameters of a dyeing process to obtain an original data set; performing preprocessing and feature extraction on the data set to obtain a feature data set; constructing a deep learning prediction model by using the feature data set, and executing multi-objective optimization through a non-dominated sorting genetic algorithm to obtain an optimal process parameter combination; and performing scheme analysis based on the optimal parameters to obtain an intelligent dyeing scheme. Multi-parameter collaborative optimization of the whole process of the dyeing process is realized, dependence on human experience is reduced, dyeing quality consistency is improved, self-learning and self-adaptive capabilities are achieved, optimization strategies can be automatically adjusted according to different fabric characteristics and production conditions, and production efficiency is improved. Therefore, the problems of large quality fluctuation, high energy consumption, poor adaptability and the like in the traditional dyeing process are solved.

Description

Technical Field

[0001] The present application relates to the field of intelligent control technology, and in particular to a dyeing process optimization method and system based on intelligent control. Background Art

[0002] As a key link in the textile industry, the dyeing process, its dyeing quality and process stability directly affect the commercial value and market competitiveness of the product. Traditional dyeing processes mainly rely on workers' experience and fixed process recipes for control, usually using a single parameter control method, that is, setting parameters such as temperature curve, dye ratio, and auxiliary agent addition based on experience, and maintaining these parameters within the set range through a simple PID controller. With the development of automation technology, the dyeing industry has introduced intelligent control technologies based on methods such as fuzzy control and expert systems, realizing automatic adjustment of some parameters, such as temperature control and liquid level control. However, most of these methods independently control a single or a few parameters, lack the ability to collaboratively optimize multiple parameters throughout the dyeing process, and are difficult to adapt to the changes in the characteristics of different types of fabrics and dyes.

[0003] Existing dyeing process control methods have obvious shortcomings: First, traditional empirical control methods are highly dependent on workers' skills, and the setting of process parameters lacks a scientific basis, resulting in large fluctuations in dyeing quality and poor consistency between batches; second, existing automated control systems are mostly single-parameter control, ignoring the complex interactions between dyeing parameters and unable to achieve global optimization; third, existing control methods lack learning and adaptive capabilities, making it difficult to automatically adjust and optimize parameters according to actual dyeing effects; fourth, the collection and utilization of dyeing process data is insufficient, and a large amount of valuable production data has not been fully mined and applied; fifth, existing control systems usually use the same control strategy to handle different types of fabrics and dye combinations, failing to perform differentiated control for specific dyeing process characteristics, reducing the adaptability and flexibility of the process. Summary of the Invention

[0004] The present application provides a dyeing process optimization method and system based on intelligent control, which is used to realize multi-parameter collaborative optimization of the entire dyeing process, reduce dependence on human experience, and improve the consistency of dyeing quality. At the same time, it has self-learning and self-adaptation capabilities, and can automatically adjust the optimization strategy according to different fabric characteristics and production conditions, thereby solving the problems of large quality fluctuations, high energy consumption, and poor adaptability in traditional dyeing processes.

[0005] In the first aspect, the present application provides a dyeing process optimization method based on intelligent control, which includes: collecting key parameters of the dyeing process to obtain an original data set of the dyeing process; performing preprocessing and feature extraction based on the original data set of the dyeing process to obtain a dyeing process feature data set; using the dyeing process feature data set to construct a deep learning prediction model and perform multi-objective optimization to obtain an optimal dyeing process parameter combination, and the multi-objective optimization adopts a non-dominated sorting genetic algorithm; performing dyeing scheme analysis according to the optimal dyeing process parameters to obtain an intelligent dyeing scheme.

[0006] In a second aspect, the present application provides a dyeing process optimization system based on intelligent control, the dyeing process optimization system based on intelligent control comprising:

[0007] The acquisition module is used to collect key parameters of the dyeing process and obtain the original data set of the dyeing process;

[0008] An extraction module is used to perform preprocessing and feature extraction based on the original dyeing process data set to obtain a dyeing process feature data set;

[0009] an optimization module, configured to construct a deep learning prediction model using the dyeing process feature dataset and perform multi-objective optimization to obtain an optimal dyeing process parameter combination, wherein the multi-objective optimization adopts a non-dominated sorting genetic algorithm;

[0010] The analysis module is used to analyze the dyeing scheme according to the optimal dyeing process parameters to obtain an intelligent dyeing scheme.

[0011] In a third aspect, a dyeing process optimization device based on intelligent control is provided, comprising: a memory and at least one processor, wherein instructions are stored in the memory; the at least one processor calls the instructions in the memory so that the dyeing process optimization device based on intelligent control executes the above-mentioned dyeing process optimization method based on intelligent control.

[0012] In a fourth aspect, a computer-readable storage medium is provided, wherein instructions are stored in the computer-readable storage medium, which, when executed on a computer, enables the computer to execute the above-mentioned dyeing process optimization method based on intelligent control.

[0013] In the technical solution provided by the present application, the original dyeing process data set is obtained by collecting the key parameters of the dyeing process, thereby realizing the comprehensive capture and digital expression of the data of the entire dyeing process, laying a data foundation for subsequent intelligent analysis; the dyeing process feature data set is obtained by preprocessing and feature extraction based on the original dyeing process data set, which effectively solves the problems of large data noise and insufficient feature extraction in the traditional dyeing process, and significantly improves the data quality and feature expression ability; the dyeing process feature data set is used to construct a deep learning prediction model and perform multi-objective optimization, which not only overcomes the defects of the limited expression ability of the traditional single model, but also realizes the accurate prediction of dyeing quality through the complementary advantages of multi-layer perceptron processing non-time series features, convolutional neural network capturing local time series patterns, and recurrent neural network modeling long-term dependencies. At the same time, the non-dominated sorting genetic algorithm used in the multi-objective optimization can simultaneously optimize multiple objectives such as color fastness, color difference, dyeing uniformity, color flower rate, dyeing time and energy consumption, and find the best compromise solution; hierarchical control is achieved according to the optimal combination of dyeing process parameters. The dyeing control system with a hierarchical control architecture obtains an intelligent dyeing scheme that can automatically adjust process parameters. The hierarchical control architecture includes a field execution layer, a process control layer and an optimization decision-making layer, which makes the control granularity finer and the response speed more in line with the control requirements, greatly reducing the dependence of traditional dyeing processes on manual experience. Especially in the specific application field of dyeing technology, the present invention provides a special solution for the complex nonlinear characteristics, multi-parameter coupling problems and multi-objective optimization requirements of the dyeing process by integrating multiple deep learning algorithms and non-dominated sorting genetic algorithm features. The core contribution of the algorithm features to the solution is reflected in: the multi-structure fusion characteristics of the deep learning algorithm accurately capture the complex relationship between static characteristics and dynamic timing characteristics in the dyeing process; the multi-objective balance characteristics of the non-dominated sorting genetic algorithm effectively solve the trade-off problem between dyeing quality and production efficiency; the hierarchical characteristics of the hierarchical control architecture perfectly adapt to the requirements of different control granularities and time scales in the dyeing process, and achieve a comprehensive optimization effect of stable dyeing quality, reduced energy consumption and improved production efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0015] Figure 1 Schematic diagram of an embodiment of a dyeing process optimization method based on intelligent control in an embodiment of the present application;

[0016] Figure 2Schematic diagram of an embodiment of a dyeing process optimization system based on intelligent control in an embodiment of the present application;

[0017] Figure 3 It is a schematic block diagram of the structure of the dyeing process optimization equipment based on intelligent control in an embodiment of the present invention. DETAILED DESCRIPTION

[0018] The present application embodiment provides a kind of dyeing process optimization method and system based on intelligent control.Term " first ", " second ", " third ", " fourth " etc. (if existing) in the specification and claims of the application and above-mentioned accompanying drawing are to be used to distinguish similar object, and need not be used to describe specific order or precedence.Should be understood that the data used like this can be interchanged in appropriate circumstances, so that the embodiment described here can be implemented in the order except the content illustrated or described here.In addition, term " comprises " or " have " and any distortion thereof, be intended to cover non-exclusive comprising, for example, comprise the process, method, system, product or equipment of a series of steps or unit and need not be limited to those steps or unit clearly listed, but can comprise not clearly listed or for other step or unit intrinsic to these processes, method, product or equipment.

[0019] For ease of understanding, the specific process of the embodiment of the present application is described below. Figure 1 In the embodiments of the present application, an embodiment of the dyeing process optimization method based on intelligent control includes:

[0020] Step S101: collecting key parameters of the dyeing process to obtain an original data set of the dyeing process;

[0021] Step S102: preprocessing and feature extraction are performed based on the original dyeing process data set to obtain a dyeing process feature data set;

[0022] Step S103: constructing a deep learning prediction model using the dyeing process feature dataset and performing multi-objective optimization to obtain the optimal dyeing process parameter combination, wherein the multi-objective optimization adopts a non-dominated sorting genetic algorithm;

[0023] Step S104: Analyze the dyeing scheme according to the optimal dyeing process parameters to obtain an intelligent dyeing scheme.

[0024] It is understandable that the execution subject of this application can be a dyeing process optimization system based on intelligent control, or a terminal or a server, which is not limited here. The embodiment of this application is described by taking the server as the execution subject as an example.

[0025] Specifically, key parameters of the dyeing process are collected to obtain the original data set of the dyeing process. This step sets up multiple sensor collection points to monitor key parameters such as dye concentration, dye solution temperature, dyeing time, liquid ratio, pH value, auxiliary agent addition amount and dyeing machine speed in real time. These sensor collection points are connected to an industrial-grade programmable logic controller to form an on-site control unit to achieve preliminary data processing. Specifically, for dye solution temperature collection, a PT100 temperature sensor is used with an accuracy of not less than 0.1°C; for pH value collection, a pH sensor is used with an accuracy of not less than 0.01; for concentration collection, a concentration sensor is used with an accuracy of not less than 0.1%. The collected data is transmitted to the central data server via industrial Ethernet, and a dual-machine hot backup mechanism is established to ensure data integrity and reliability. For example, in the three-bath dyeing process of silk, cotton, polyester and nylon four-in-one fibers, the hydrogen peroxide parameters are collected at 40°C in the cloth boiling and dyeing bath stage; the 130X40 and cotton dissolving treasure parameters are recorded in the emulsifier and dye addition stage; the temperature curve is recorded at a heating rate of 1.5 degrees per minute until it reaches 80°C in the heating stage; the parameters of the 80°C insulation state are recorded in the special cleaning stage; the cooling curve from 80°C to 40°C is recorded in the cooling stage; the parameters of 40°CX20 washing water are recorded in the deurethane and water-washing cotton rolling stages.

[0026] Preprocessing and feature extraction were performed on the original dyeing process dataset to obtain a dyeing process feature dataset. An adaptive anomaly detection algorithm was used to clean the raw data, identifying and processing missing values, outliers, and duplicate data. Missing values were imputed using linear interpolation, k-nearest neighbor interpolation, or forward filling, depending on the data type. Outliers were identified using the 3σ rule or boxplot rule and corrected or eliminated based on the trends of the preceding and following data. For duplicate data, the first valid data point collected was retained and subsequent duplicates were deleted. After data cleaning, the data was standardized using a nonlinear transformation, converting parameters of different dimensions to a standard normal distribution. Subsequently, a filter was applied to the standardized data for time series noise reduction. The Savitzky-Golay filter was used to smooth the time series data for each parameter, with a filter window width of 15 data points and a polynomial order of 3 to achieve a balance between filtering effectiveness and edge fidelity. Wavelet coefficients and statistical features are then extracted from the smoothed data to form a multidimensional feature matrix. This matrix includes time-domain features such as mean, standard deviation, peak, valley, slope, skewness, and kurtosis, as well as frequency-domain features such as power spectral density, primary frequency components, and band energy distribution, obtained through fast Fourier transforms. Dimensionality reduction techniques are then applied to the multidimensional feature matrix to compress the features, resulting in a low-dimensional feature set. This low-dimensional feature set is then fused with the dyeing process mechanism features to generate a dyeing process feature dataset containing no fewer than 30 feature dimensions, with each dyeing batch forming a multidimensional feature record.

[0027] A deep learning prediction model was constructed using a dyeing process feature dataset and multi-objective optimization was performed to obtain the optimal dyeing process parameter combination. The dyeing process feature dataset was divided into training, validation, and test sets in a ratio of 7:2:1. A multi-structure deep learning model consisting of a multilayer perceptron, a convolutional neural network, and a recurrent neural network was then constructed based on the training set. The multilayer perceptron contained four hidden layers with 128, 256, 128, and 64 neurons, respectively, and used a Reluctant Unit (ReLU) activation function. The convolutional neural network consisted of three convolutional blocks, each containing two convolutional layers and one max pooling layer with a 3×1 kernel size. The initial number of channels was 32, and the number of channels doubled after each convolutional block. The recurrent neural network adopted a bidirectional architecture, consisting of two layers of gated recurrent units with a hidden state dimension of 128, and an attention mechanism added to the output layer. The model was optimized using the validation set to obtain a candidate dyeing quality prediction model. Performance evaluation and validation were then performed on the test set, resulting in a deep learning prediction model capable of predicting color fastness, color difference, levelness, and color fringing. Based on a deep learning prediction model, multi-objective optimization variables and constraints were set. The variables included 12 key parameters, including dye bath temperature, dye ratio, and holding time. A non-dominated sorting genetic algorithm was used to calculate these variables to achieve multi-objective optimization. The algorithm used real number encoding, a population size of 100, 500 generations of evolution, tournament selection, simulated binary crossover with a crossover probability of 0.9 and a distribution index of 20, and polynomial mutation with a mutation probability of 1 / 12. The Pareto optimal solution set was obtained through the optimization algorithm, and the best compromise solution was selected from this set using the weighted TOPSIS method, resulting in the optimal combination of dyeing process parameters.

[0028] Dyeing plans are analyzed based on the optimal dyeing process parameters to generate an intelligent dyeing plan. The optimal dyeing process parameter combination is then mapped and transformed through process data to produce a standardized control parameter matrix. Parameters are first sorted according to the process flow to form a process parameter sequence, including parameters for the dye bath, emulsification, heating, holding, cooling, and deurethane steps. This process parameter sequence is then dimensionally normalized, unifying the parameters of different physical quantities to the range of 0–1 to form a unified-dimensional process parameter matrix. Based on the unified-dimensional process parameter matrix, a correlation diagram between process variables is constructed to generate a parameter correlation network. The inter-parameter dependencies are quantified using the Pearson correlation coefficient. Cluster analysis is performed on the parameter correlation network to identify highly coupled parameter groups. Joint optimization transformation is then performed to generate a decoupled parameter matrix. The decoupled parameter matrix is then restructured according to control hierarchy requirements to produce a standardized control parameter matrix. The standardized control parameter matrix is then subjected to hierarchical parsing to obtain parameter sets for the field execution layer, the process control layer, and the optimization decision layer, respectively. Based on these three parameter sets, a process execution control strategy is constructed to form a dynamic control execution plan, which is then applied to the dyeing process flow. Through real-time parameter monitoring and feedback adjustment, an intelligent dyeing plan is obtained.

[0029] In the embodiments of the present application, the original dyeing process data set is obtained by collecting the key parameters of the dyeing process, thereby realizing the comprehensive capture and digital expression of the data of the entire dyeing process, laying a data foundation for subsequent intelligent analysis; preprocessing and feature extraction are performed based on the original dyeing process data set to obtain a dyeing process feature data set, which effectively solves the problems of large data noise and insufficient feature extraction in the traditional dyeing process, and significantly improves the data quality and feature expression ability; the dyeing process feature data set is used to construct a deep learning prediction model and perform multi-objective optimization, which not only overcomes the defects of the limited expression ability of the traditional single model, but also realizes the accurate prediction of dyeing quality through the complementary advantages of multi-layer perceptron processing non-time series features, convolutional neural network capturing local time series patterns, and recurrent neural network modeling long-term dependencies. At the same time, the non-dominated sorting genetic algorithm used in the multi-objective optimization can simultaneously optimize multiple objectives such as color fastness, color difference, dyeing uniformity, color flower rate, dyeing time and energy consumption, and find the best compromise solution; hierarchical control is achieved according to the optimal dyeing process parameter combination. The dyeing control system of the architecture obtains an intelligent dyeing scheme that can automatically adjust the process parameters. The hierarchical control architecture includes a field execution layer, a process control layer and an optimization decision layer, which makes the control granularity finer and the response speed more in line with the control requirements, greatly reducing the dependence of traditional dyeing processes on manual experience. Especially in the specific application field of dyeing process, the present invention provides a special solution for the complex nonlinear characteristics, multi-parameter coupling problems and multi-objective optimization requirements of the dyeing process by integrating multiple deep learning algorithms and non-dominated sorting genetic algorithm features. The core contribution of the algorithm features to the solution is reflected in: the multi-structure fusion characteristics of the deep learning algorithm accurately capture the complex relationship between static characteristics and dynamic timing characteristics in the dyeing process; the multi-objective balance characteristics of the non-dominated sorting genetic algorithm effectively solve the trade-off problem between dyeing quality and production efficiency; the hierarchical characteristics of the hierarchical control architecture perfectly adapt to the requirements of different control granularities and time scales in the dyeing process, and achieve a comprehensive optimization effect of stable dyeing quality, reduced energy consumption and improved production efficiency.

[0030] In a specific embodiment, the process of executing step S101 may specifically include the following steps:

[0031] Multiple sensor collection points are set for dye concentration, dye liquor temperature, dyeing time, liquid ratio, pH value, auxiliary agent addition amount and dyeing machine speed to obtain a multi-dimensional parameter collection network;

[0032] The multi-dimensional parameter acquisition network is connected to an industrial-grade programmable logic controller to obtain a field control unit. The programmable logic controller is equipped with a processor of no less than 100MHz;

[0033] The field control unit is used to execute a data filtering algorithm to process the collected data to obtain parameter data;

[0034] Transmit parameter data to the central data server via industrial Ethernet to obtain parameter data stream;

[0035] Set up a dual-machine hot backup mechanism for the parameter data stream to obtain a data storage solution;

[0036] Based on the data storage solution, the parameters of the entire dyeing process are continuously collected to obtain the original dyeing process data set. The original dyeing process data set contains parameter records of all processes from boiling the cloth and dyeing bath to deurethane washing and rolling the cotton.

[0037] Specifically, during the dyeing process, multiple sensor collection points are deployed for key parameters such as dye concentration, dye bath temperature, dyeing time, liquid ratio, pH value, additive dosage, and dyeing machine speed, creating a multidimensional parameter collection network. For example, during the dye bath boiling stage, the hydrogen peroxide parameters must be precisely monitored to maintain a temperature of 40°C. During the heating stage, a steady temperature increase rate of 1.5°C per minute must be maintained until reaching 80°C. During the holding stage, the temperature stability at 80°C must be monitored. During the cooling stage, the entire process from 80°C to 40°C must be recorded. The data collected by these sensors directly impacts dyeing quality and process control accuracy.

[0038] The multidimensional parameter acquisition network is connected to an industrial-grade programmable logic controller (PLC), forming a field control unit. These controllers are equipped with processors running at least 100MHz, ensuring sufficient computing power to handle the high-frequency data acquisition tasks. PLC systems typically have at least 8MB of memory, at least 32 digital inputs and outputs, and at least 8 analog inputs and outputs to meet the control requirements of complex dyeing processes. Sensors communicate with the PLC via a standard 4-20mA signal output and an RS485 bus, ensuring stable and reliable signal transmission. The field control unit processes the raw acquired data using a data filtering algorithm to remove significant noise and outliers. These filtering algorithms, including moving average processing, median filtering, and outlier detection, provide preliminary cleanup of the raw sensor data. For example, if a temperature sensor exhibits abnormal short-term fluctuations, a moving average algorithm can smooth these fluctuations. Similarly, if a pH sensor occasionally displays abnormal readings, a median filter can effectively eliminate these interferences. These processing steps result in more reliable and consistent parameter data.

[0039] After preliminary processing, parameter data is transmitted to a central data server via Industrial Ethernet, forming a continuous parameter data stream. Industrial Ethernet utilizes the OPC UA communication protocol to ensure real-time and reliable data transmission. The data stream is recorded with an accuracy of at least 10 data points per second, capturing subtle changes during the dyeing process, such as temperature fluctuations, chemical reaction changes, and physical state transitions.

[0040] A dual-machine hot backup mechanism is set up for parameter data streams to establish a robust data storage solution. In the event of a failure in the primary system, the backup system can seamlessly take over data collection tasks in no more than 3 seconds, preventing data loss. Data storage uses a relational database structure to store historical data and is equipped with a real-time database to process real-time data streams. The server has a storage capacity of no less than 10TB, sufficient to record long-term production data. Based on this data storage solution, parameters from the entire dyeing process are continuously collected to obtain a complete raw dyeing process data set. The data set contains parameter records for all processes, starting from boiling the cloth and dye bath, through emulsifier addition, dye addition, heating, heat preservation, cooling, and finally deurethane washing and rolling of cotton. In the silk and cotton blend dyeing process, the dyeing cycle typically generates tens of thousands of data points, covering key information such as temperature curves, pH value changes, and dye concentration changes at each stage of the process.

[0041] In a specific embodiment, the process of executing step S102 may specifically include the following steps:

[0042] Perform data cleaning on the original data set of dyeing process to obtain the integrity cleaned data set;

[0043] The integrity cleansing data set is standardized through nonlinear transformation to obtain balanced standard data;

[0044] Applying a filter to the equalized standard data to perform time series noise reduction processing to obtain smoothed data;

[0045] Extract wavelet coefficients and statistical features from smoothed data to obtain a multi-dimensional feature matrix;

[0046] Perform dimensionality reduction technology on the multi-dimensional feature matrix to compress the features and obtain a low-dimensional feature set;

[0047] The low-dimensional feature set is fused with the dyeing process mechanism features to obtain the dyeing process feature dataset.

[0048] Specifically, data cleaning is performed on the original data set of the dyeing process to handle missing values, outliers, and duplicate data. In the dyeing process, missing or abnormal data often occur due to sensor failure or signal interference. For missing values, different completion methods are used according to the data type: for continuously changing parameters such as temperature, linear interpolation is used to fill short-term missing values; for parameters with large fluctuations such as pH value, k-nearest neighbor interpolation is used to refer to data under similar working conditions for filling; for parameters in a stable state, forward filling is used to replace them with the nearest valid value. For outliers, the 3σ rule is applied to identify data points that are beyond the normal fluctuation range. For example, in the constant temperature stage of 80℃, if the temperature suddenly jumps to 95℃, it will be marked as an anomaly and corrected according to the characteristics of the process curve. For duplicate data, the valid value collected for the first time is retained, and subsequent duplicate records are eliminated to avoid data redundancy affecting subsequent analysis.

[0049] The integrity-cleaned dataset was normalized using a nonlinear transformation to obtain balanced standard data. Since the dyeing process involves multiple parameters with different dimensions, such as temperature (°C), time (min), pH, and concentration (g / L), standardization is necessary to ensure comparability. Appropriate nonlinear transformation methods were selected based on the characteristics of the parameters: z-score normalization was used for parameters with a near-normal distribution; logarithmic transformation and subsequent normalization were used for parameters with skewed distributions, such as dye concentration; and Min-Max normalization was used to map parameters with clear upper and lower limits, such as pH, to the interval [0, 1]. This normalization process places the parameters of different physical quantities on the same scale, facilitating subsequent model learning of the relative importance of each parameter. A filter was applied to the balanced standard data to perform time series noise reduction, resulting in smoothed data. Time series of dyeing process parameters often contain high-frequency noise, which affects data quality. A Savitzky-Golay filter was applied to smooth the time series data, with a window width of 15 data points and a polynomial order of 3. This filter effectively removes noise while preserving the original data trend. For example, in the process of dyeing temperature rising from 40℃ to 80℃, the original data may have high-frequency fluctuations of ±0.5℃. After filtering, a smoother temperature rise curve is obtained, while retaining key inflection point information such as changes in the heating rate.

[0050] Wavelet coefficients and statistical features are extracted from the smoothed data to obtain a multidimensional feature matrix. Wavelet transform is used to capture the characteristics of the data at different time scales and identify key patterns in the process. At the same time, statistical features are extracted, including mean, standard deviation, peak, valley, slope, skewness, and kurtosis, which are indicators that describe the distribution and variation characteristics of parameters. For example, the mean and standard deviation of the temperature curve in the insulation stage can be extracted to quantify the stability of temperature control; the slope characteristics of the curve in the heating stage can be extracted to characterize the accuracy of the heating rate control; the peak and valley values of the pH value curve can be extracted to reflect the key change points of the chemical reaction process. In addition, the frequency domain features, including power spectral density, main frequency components, and frequency band energy distribution, are obtained through fast Fourier transform to reveal the periodic variation pattern of the data.

[0051] Dimensionality reduction techniques are applied to the multidimensional feature matrix to compress the features and obtain a low-dimensional feature set. The original features extracted during the dyeing process are high in dimension, contain redundancy and noise, and require dimensionality reduction. Principal component analysis (PCA) or nonlinear dimensionality reduction techniques such as t-SNE are applied to map high-dimensional features to a low-dimensional space, preserving the data structure while reducing computational complexity. For example, for a complete dyeing process, the original extracted features may have more than 100 dimensions, which can be compressed to 20-30 main features through dimensionality reduction technology while retaining more than 90% of the information, effectively reducing the complexity of subsequent modeling.

[0052] By integrating a low-dimensional feature set with the dyeing process mechanism characteristics, a dyeing process feature dataset is generated. While data-driven features can capture statistical patterns, mechanistic features extracted in combination with process expertise can more comprehensively characterize the dyeing process. Mechanistic features include indicators that directly reflect process performance, such as temperature rise rate, thermal stability index, cooling rate, pH stabilization time, and dye adsorption rate. For example, for the dyeing of silk-cotton blended fabrics, the hydrogen peroxide activity half-life feature is extracted from the cloth-boiling dye bath stage; the dispersion stability index is extracted from the emulsifier stage; the integral value of the deviation between the actual heating curve and the ideal curve is extracted from the heating stage; and the root mean square value of temperature fluctuation is extracted from the thermal holding stage. These mechanistic features are integrated with the aforementioned data-driven low-dimensional feature set to form a dyeing process feature dataset containing more than 30 feature dimensions, providing a comprehensive feature representation for subsequent deep learning model training.

[0053] In a specific embodiment, the process of executing step S103 may specifically include the following steps:

[0054] The dyeing process feature dataset was divided into training set, validation set and test set in a ratio of 7:2:1;

[0055] Based on the training set, a multi-structure deep learning model consisting of a multi-layer perceptron, a convolutional neural network, and a recurrent neural network was constructed to obtain a basic model for dyeing quality prediction. The multi-layer perceptron contains four hidden layers, the convolutional neural network contains three convolutional blocks, and the recurrent neural network adopts a bidirectional structure.

[0056] The basic model for staining quality prediction is optimized using the validation set to obtain a candidate model for staining quality prediction;

[0057] The performance of candidate dyeing quality prediction models was evaluated and verified using the test set, resulting in a deep learning prediction model for predicting color fastness, color difference, dyeing uniformity, and color fringing rate.

[0058] Based on the deep learning prediction model, multi-objective optimization variables and constraints are set to obtain the dyeing process parameter optimization variables, including dye bath temperature, dye ratio and holding time;

[0059] The non-dominated sorting genetic algorithm is used to calculate the optimization variables of dyeing process parameters to obtain the Pareto optimal solution set, and the weighted TOPSIS method is used to screen the compromise solution from it to obtain the optimal dyeing process parameter combination.

[0060] Specifically, the dyeing process feature dataset was divided into training, validation, and test sets in a ratio of 7:2:1. This division ratio ensures sufficient data samples for model training while retaining an appropriate amount of data for model validation and final performance evaluation. For dyeing data, this division method needs to consider the balance of data distribution to ensure that different types of dyeing process conditions and results are reasonably distributed across the three datasets to avoid training bias. Based on the training set, a multi-structure deep learning model consisting of a multi-layer perceptron, a convolutional neural network, and a recurrent neural network was constructed to obtain a basic model for dyeing quality prediction. The multi-layer perceptron (MLP) consists of four hidden layers with 128, 256, 128, and 64 neurons, respectively. It uses the ReLU activation function, primarily processing non-temporal features. The convolutional neural network (CNN) consists of three convolutional blocks, each containing two convolutional layers and one max-pooling layer. The convolution kernel size is 3×1, and the initial number of channels is 32, which doubles after each convolution block. This is primarily used to extract local characteristic patterns of the dyeing parameters. The recurrent neural network (RNN) employs a bidirectional architecture, consisting of two layers of gated recurrent units with a hidden state dimension of 128, primarily capturing the long-term temporal dependencies of the dyeing process. This multi-architecture design leverages the strengths of different neural network architectures, enabling simultaneous processing of both static and dynamic temporal features of the dyeing process.

[0061] The basic dye quality prediction model was optimized on the validation set to obtain a candidate dye quality prediction model. The Adam optimizer was used for optimization, with an initial learning rate of 0.001 and a cosine annealing schedule. The learning rate was reduced to 0.5 every 200 epochs, for a total of 1000 epochs. An early stopping strategy was employed, stopping training after 50 consecutive epochs without improvement in the validation set loss. The loss function was designed as a weighted sum of the mean squared error and the mean absolute error, with a weighting ratio of 7:3 to balance the penalties for large and small errors. By adjusting the network structure, optimizing hyperparameters, and applying regularization techniques such as dropout, the model's performance on the validation set was continuously improved, ultimately resulting in a candidate dye quality prediction model.

[0062] The performance of candidate dyeing quality prediction models was evaluated and validated using a test set, resulting in a deep learning prediction model for color fastness, color difference, dyeing uniformity, and color fringing rate. The test set, a dataset completely independent of the training process, was used to objectively assess the model's generalization ability. Evaluation metrics included mean absolute error, root mean square error, and R-squared. For color fastness prediction, the model achieved a mean absolute error of no more than 0.5; for color difference prediction, the error was no more than 0.3ΔE; for dyeing uniformity prediction, the accuracy was no less than 92%; and for color fringing rate prediction, the error was no more than 2%. The model output included the predicted values for the four quality metrics and their 95% confidence intervals, providing an estimate of the reliability of the predictions.

[0063] Based on a deep learning predictive model, multi-objective optimization variables and constraints were set to obtain the optimized dyeing process parameters. The optimized variables included 12 key process parameters, including dye bath temperature, dye ratio, and holding time. Each variable was assigned a reasonable range: the dye bath temperature range was 30-90°C; the dye ratio was adjusted ±15% according to the target recipe; and the holding time range was 20-40 minutes. Constraints included the physical limitations of the process parameters and the operational limits of the dyeing equipment, ensuring that the optimized results were feasible in the actual process. A non-dominated sorting genetic algorithm was used to calculate the optimized dyeing process parameter variables, resulting in a Pareto optimal solution set. The non-dominated sorting genetic algorithm is an effective method for solving multi-objective optimization problems and can simultaneously optimize multiple objectives, including color fastness, color difference, dyeing levelness, color fringing rate, dyeing time, and energy consumption. The algorithm uses real number encoding, with a population size of 100 and an evolutionary generation number of 500. The selection operation uses the tournament selection method with a tournament size of 3; the crossover operation uses simulated binary crossover with a crossover probability of 0.9 and a distribution index of 20; and the mutation operation uses polynomial mutation with a mutation probability of 1 / 12. The algorithm also incorporates an adaptive population mechanism: when the number of individuals on the Pareto front exceeds 70% of the total population, the population size is increased by 25%; when the number of individuals on the front falls below 30%, the population size is reduced by 15%, but the minimum population size is never less than 60. After the algorithm converges, a series of non-dominated solutions are obtained, forming the Pareto optimal solution set. Finally, the weighted TOPSIS method is used to screen the best compromise solution from the Pareto optimal solution set and determine the optimal dyeing process parameter combination. The TOPSIS method comprehensively considers the weight ratios of six optimization objectives: color fastness, color difference, dyeing levelness, color fringing rate, dyeing time, and energy consumption are weighted in a ratio of 3:2:2:1:1:1, to find the optimal parameter combination in this multi-objective trade-off.

[0064] In a specific embodiment, the process of executing the step of constructing a multi-structure deep learning model including a multi-layer perceptron, a convolutional neural network, and a recurrent neural network based on the training set may specifically include the following steps:

[0065] Add Gaussian noise and random transformation to the training set to obtain an expanded training data set;

[0066] The expanded training data set is forward propagated through a 4-layer fully connected network to obtain a multilayer perceptron with 128, 256, 128, and 64 hidden layer neurons in the 4 hidden layers, respectively.

[0067] The extended training dataset is subjected to feature extraction through three convolutional blocks to obtain a convolutional neural network. The convolution kernel size of the three convolutional blocks is 3×1, the initial number of channels is 32, and the number of channels doubles after each convolutional block.

[0068] The expanded training dataset is sequence modeled using a bidirectional gated recurrent unit to obtain a recurrent neural network. The bidirectional gated recurrent unit contains two layers, the hidden state dimension is 128, and an attention mechanism is added to the output layer.

[0069] The multilayer perceptron, convolutional neural network and recurrent neural network are trained using Adam optimizer and cosine annealing learning rate scheduling to obtain the neural network model;

[0070] The neural network model is stacked and generalized to obtain the basic model for staining quality prediction.

[0071] Specifically, data augmentation was performed by adding Gaussian noise and random transformations to the training set to generate an expanded training dataset. This is particularly critical for dyeing process data. In practice, Gaussian noise with a mean of 0 and a standard deviation of 0.5°C was added to the temperature parameter to simulate sensor measurement errors; random time windows were sliced into time series data to simulate dyeing processes of varying lengths; and random fluctuations of ±3% were added to the dye concentration parameter to simulate small errors in the batching process. These augmentations enable the model to adapt to the various fluctuations encountered in actual production, improving its robustness.

[0072] The expanded training dataset was forward propagated through a four-layer fully connected network to produce a multilayer perceptron. The number of neurons in the four hidden layers of the multilayer perceptron was set to 128, 256, 128, and 64, respectively, exhibiting a structure that first expands and then contracts. The first layer of 128 neurons is responsible for preliminary feature extraction; the second layer is expanded to 256 neurons, increasing the network's expressive power and capturing more complex feature combinations; the third and fourth layers gradually decrease to 128 and 64 neurons, respectively, compressing and abstracting features. Reluctant linear unit (ReLU) activation function addresses the vanishing gradient problem while maintaining computational efficiency. To prevent overfitting, a dropout layer is added after each layer, with a dropout rate set to 0.3. The multilayer perceptron primarily processes non-temporal features of the dyeing process, such as the relationship between static parameters such as dye type and fabric density and dyeing quality.

[0073] The expanded training dataset was subjected to feature extraction using three convolutional blocks to generate a convolutional neural network. Each convolutional block consists of two convolutional layers and one maximum pooling layer, with a kernel size of 3×1, suitable for capturing local patterns in the dyeing parameter time series. The initial number of channels is set to 32, and this number doubles with each convolutional block, with the second convolutional block outputting 64 channels and the third convolutional block outputting 128 channels, gradually extracting higher-level feature representations. The maximum pooling layer uses a pooling size of 2×1, which reduces the dimensionality of the feature map while retaining important features. Convolutional neural networks are particularly well suited for identifying local temporal patterns such as the morphological characteristics of the temperature curve and the patterns of changes in heating and cooling rates during the dyeing process. These patterns are directly related to dyeing uniformity and color fastness.

[0074] The expanded training dataset was subjected to sequence modeling using bidirectional gated recurrent units (BGRUs), resulting in a recurrent neural network (RNN). This recurrent network employs a bidirectional architecture consisting of two layers of gated recurrent units (GRUs) with a hidden state dimension of 128. This bidirectional architecture enables the network to simultaneously consider both past and future information, enabling a more comprehensive understanding of long-term dependencies within time series data. An attention mechanism was incorporated into the output layer to automatically identify and focus on critical moments in the dyeing process, such as sudden temperature changes and pH stabilization, which significantly impact dyeing quality. RNNs are particularly well-suited for capturing dynamic changes throughout the dyeing process, such as dye adsorption and the cumulative effects of temperature fluctuations on color fastness, along with other long-term dependencies.

[0075] A multilayer perceptron, convolutional neural network, and recurrent neural network were trained using the Adam optimizer and a cosine annealing learning rate schedule to generate the neural network models. The Adam optimizer combines the advantages of momentum and adaptive learning rate methods, with an initial learning rate of 0.001. The cosine annealing learning rate schedule reduces the learning rate to 0.5 every 200 epochs, effectively mitigating learning oscillations and helping the model find a better local minimum. Training was performed using mini-batch gradient descent with a batch size of 64, along with an early stopping strategy. Training was terminated after no improvement in the validation set loss for 50 consecutive epochs. The loss function was a weighted combination of mean squared error and mean absolute error, with a weight ratio of 7:3, to balance sensitivity to large and small errors. The trained neural network models were integrated using the stacking generalization technique to generate a base model for dye quality prediction. Stacking generalization is a powerful model ensemble method that uses the outputs of multiple base models as new features to train a meta-learner. Specifically, the output features of the multilayer perceptron, convolutional neural network, and recurrent neural network were concatenated and used as input features for the XGBoost model. The XGBoost model parameter settings include a maximum tree depth of 5, a learning rate of 0.05, a subsampling rate of 0.8, and an iteration count of 500. This ensemble method fully utilizes the complementary strengths of various neural networks: multilayer perceptrons excel at processing static feature relationships, convolutional networks excel at capturing local temporal patterns, and recurrent networks excel at modeling long-term dependencies, resulting in a more accurate prediction of dye quality.

[0076] In a specific embodiment, the process of executing step S104 may specifically include the following steps:

[0077] According to the optimal dyeing process parameter combination, process data mapping conversion is performed to obtain a standardized control parameter matrix;

[0078] Perform hierarchical analysis on the standardized control parameter matrix to obtain the field execution layer parameter set, the process control layer parameter set and the optimization decision layer parameter set;

[0079] Based on the field execution layer parameter set, process control layer parameter set and optimization decision layer parameter set, a process execution control strategy is constructed to obtain a dynamic control execution plan;

[0080] The dynamic control execution plan is applied to the dyeing process, and an intelligent dyeing plan is obtained through real-time parameter monitoring and feedback adjustment.

[0081] Specifically, this is converted into a practically applicable intelligent dyeing solution. First, process data mapping and conversion are performed based on the optimal dyeing process parameter combination to obtain a standardized control parameter matrix. This step begins by sorting the parameters according to the process flow sequence to form a process parameter sequence, including parameters related to the cloth dye bath boiling, dye emulsification, heating, holding, cooling, and deurethane processes. This process parameter sequence is then dimensionally normalized, unifying the parameters of different physical quantities to the range of 0-1 to form a unified-dimensional process parameter matrix. A correlation diagram between process variables is then constructed based on this unified-dimensional process parameter matrix. The Pearson correlation coefficient is used to quantify inter-parameter dependencies and identify key parameter combinations that influence each other. Cluster analysis is performed on the parameter correlation network to identify highly coupled parameter groups. Principal component analysis is then used to perform a joint optimization transformation, converting the relevant parameters into an orthogonal representation to obtain a decoupled parameter matrix. The decoupled parameter matrix is then restructured according to the control hierarchy requirements to form a standardized control parameter matrix.

[0082] A hierarchical parsing process is performed on the standardized control parameter matrix to generate parameter sets for the field execution layer, the process control layer, and the optimization decision layer. This hierarchical parsing process decomposes control parameters according to control granularity and time scale, forming a hierarchical control architecture. The field execution layer parameter set includes specific parameters that directly control the dyeing equipment, such as the frequency setpoint of the variable frequency drive, the opening percentage of the proportional control valve, and the power output value of the heater power controller. These parameters require high-frequency updates, typically adjusting on a timescale of seconds. The process control layer parameter set includes intermediate control parameters for the temperature control loop, pH control loop, and dye addition control loop, responsible for achieving precise control of process parameters. For the temperature control loop, a feedforward-feedback composite control strategy is employed, achieving a control accuracy of ±0.2°C. For the pH control loop, a fuzzy adaptive PID control algorithm is employed, achieving a control accuracy of ±0.05. These parameters are optimized and adjusted on a minute-by-minute basis. The optimization decision-level parameter set includes higher-level process parameter optimization objectives and constraints, such as colorfastness optimization, dyeing uniformity requirements, and energy consumption control targets. These parameters are updated at the dyeing batch level. A process execution control strategy is constructed based on the field execution, process control, and optimization decision-level parameter sets, resulting in a dynamic control execution plan. This process execution control strategy utilizes a model-based predictive control architecture, leveraging a deep learning predictive model to predict future process states and calculate the optimal control trajectory. This control strategy is based on the principle of rolling horizon optimization, with a prediction window length of 120 seconds, a control window length of 30 seconds, and a sampling time of 3 seconds. During each optimization cycle, the system predicts process parameter trends over the next 120 seconds based on the current state, calculates the optimal control action for the next 30 seconds, and executes only the control action for the first sampling cycle. The dynamic control execution plan also includes an automatic fault detection and handling mechanism. Upon detecting sensor failure, actuator malfunction, or control loop anomaly, the system automatically switches to a safe control mode to ensure the reliability of the dyeing process.

[0083] The dynamic control execution scheme is applied to the dyeing process flow, and an intelligent dyeing scheme is obtained through real-time parameter monitoring and feedback adjustment. In actual application, the initial set values of the optimal dyeing process parameters are first loaded to start the dyeing process. During dyeing, the system continuously monitors the actual process parameters and compares them with the predicted values and set values. When the deviation between the actual parameters and the predicted values exceeds the preset threshold, real-time optimization calculations are triggered and the control strategy is dynamically adjusted. The intelligent dyeing scheme has a self-learning function. By comparing the differences between the theoretical model prediction values and the actual dyeing effects, it continuously updates and optimizes the parameters of the deep learning prediction model to achieve continuous improvement in control performance. In the dyeing process of silk and cotton blended fabrics, the intelligent dyeing scheme can automatically adjust key parameters such as the heating rate, holding time, and dye addition amount according to the characteristics of different batches of fabrics, and make fine adjustments based on the real-time monitoring of the dyeing status to ensure the consistency and stability of the dyeing quality while optimizing energy consumption and dyeing time.

[0084] In a specific embodiment, the process of performing the process data mapping conversion step according to the optimal dyeing process parameter combination may specifically include the following steps:

[0085] Perform dimension normalization on the process parameter sequence to obtain the process parameter matrix;

[0086] Based on the process parameter matrix, a correlation diagram between process variables is constructed to obtain a parameter correlation network;

[0087] Cluster analysis of the parameter correlation network was performed to obtain highly coupled parameter groups;

[0088] Perform joint optimization transformation based on the highly coupled parameter group to obtain a decoupling parameter matrix;

[0089] The decoupling parameter matrix is restructured according to the control hierarchy requirements to obtain a standardized control parameter matrix. The structural reorganization allocates the parameters to the corresponding control hierarchy and sets the priority order.

[0090] Specifically, the process parameter sequence was dimensionally normalized to obtain a process parameter matrix. Appropriate normalization methods were employed for different physical quantities: for temperature parameters, Min-Max normalization was used to map actual temperature values between 40°C and 95°C to the 0-1 interval; for time parameters, holding times between 0 and 60 minutes were linearly mapped to the 0-1 interval; for concentration parameters, the concentrations of various dyes and auxiliaries were divided by their maximum allowable concentrations; and for pH parameters, actual values between 4.5 and 10.5 were linearly mapped to the 0-1 interval. This unified dimensioning eliminated scale differences between different physical quantities. A correlation diagram between process variables was constructed based on the process parameter matrix, resulting in a parameter correlation network. The inter-parameter dependencies were quantified by calculating the Pearson correlation coefficient. For example, in the dyeing process for silk-cotton blended fabrics, the hydrogen peroxide concentration in the boiling stage and the amount of dissolved cotton added in the emulsification stage showed a positive correlation of 0.72; while the rate of temperature rise in the dye bath and the temperature stability during the holding stage showed a negative correlation of -0.58. In this way, the correlations between all process parameters are organized into a network structure, which clearly shows the mutual influence relationship between the parameters and provides a basis for subsequent parameter optimization.

[0091] Cluster analysis of the parameter correlation network identified highly coupled parameter groups. Spectral clustering was used to analyze the parameter correlation network, grouping parameters with absolute correlation coefficients greater than 0.5 into the same group and identifying parameter subsets with strong correlations. During the dyeing process, several distinct parameter groups typically emerge: temperature control parameters (including heating rate, holding temperature, cooling rate, etc.), chemical dosage parameters (including dye concentration, auxiliary dosage, pH value, etc.), and time control parameters (including the duration of each process step). Joint optimization transformation was performed based on these highly coupled parameter groups to generate a decoupled parameter matrix. For each highly coupled parameter group, principal component analysis (PCA) was applied to transform the related parameters into an orthogonal representation, reducing redundancy and interference between parameters. For example, PCA of the temperature control parameter group yielded a first principal component representing the overall temperature level and a second principal component representing the temperature change rate. Transformation of the chemical dosage parameter group isolated orthogonal components representing the base dye concentration, dye ratio structure, and auxiliary coordination relationship. This joint optimization transformation reduces the degree of coupling between parameters, enabling more independent and precise parameter adjustments during subsequent control processes.

[0092] The decoupling parameter matrix is restructured according to the control hierarchy requirements to obtain a standardized control parameter matrix. During the restructuring process, the parameters are assigned to different control levels based on their control characteristics and time scales: fast-response execution parameters (such as valve opening and heating power) are assigned to the field execution layer; process parameters with medium response times (such as temperature setpoints and pH control targets) are assigned to the process control layer; and long-cycle optimization strategy parameters (such as dyeing recipe optimization and energy consumption control strategies) are assigned to the optimization decision layer. At the same time, within each control level, a priority order is set based on the degree of influence of the parameters on dyeing quality, and parameters with a significant impact on dyeing uniformity are given a higher priority. Through this hierarchical parameter organization, a standardized control parameter matrix with a clear structure and reasonable control granularity is formed.

[0093] The above describes the dyeing process optimization method based on intelligent control in the embodiment of the present application. The following describes the dyeing process optimization system based on intelligent control in the embodiment of the present application. Figure 2 In the embodiments of the present application, an embodiment of the dyeing process optimization system based on intelligent control includes:

[0094] The acquisition module 201 is used to collect key parameters of the dyeing process and obtain an original data set of the dyeing process;

[0095] Extraction module 202, for performing preprocessing and feature extraction based on the original dyeing process data set to obtain a dyeing process feature data set;

[0096] An optimization module 203 is used to construct a deep learning prediction model using the dyeing process feature data set and perform multi-objective optimization to obtain the optimal dyeing process parameter combination. The multi-objective optimization adopts a non-dominated sorting genetic algorithm;

[0097] The analysis module 204 is used to analyze the dyeing scheme according to the optimal dyeing process parameters to obtain an intelligent dyeing scheme.

[0098] Through the collaborative cooperation of the above components, the key parameters of the dyeing process are collected to obtain the original data set of the dyeing process, which realizes the comprehensive capture and digital expression of the data of the whole dyeing process, laying a data foundation for subsequent intelligent analysis; based on the original data set of the dyeing process, preprocessing and feature extraction are carried out to obtain the feature data set of the dyeing process, which effectively solves the problems of large data noise and insufficient feature extraction in the traditional dyeing process, and significantly improves the data quality and feature expression ability; the feature data set of the dyeing process is used to construct a deep learning prediction model and perform multi-objective optimization, which not only overcomes the defects of the limited expression ability of the traditional single model, but also realizes the accurate prediction of dyeing quality through the complementary advantages of multi-layer perceptron processing non-time series features, convolutional neural network capturing local time series patterns, and recurrent neural network modeling long-term dependencies. At the same time, the non-dominated sorting genetic algorithm used in multi-objective optimization can simultaneously optimize multiple objectives such as color fastness, color difference, dyeing uniformity, color flower rate, dyeing time and energy consumption, and find the best compromise solution; classification is achieved according to the optimal combination of dyeing process parameters. The dyeing control system with a layered control architecture obtains an intelligent dyeing solution that can automatically adjust process parameters. The layered control architecture includes a field execution layer, a process control layer, and an optimization decision layer, which makes the control granularity finer and the response speed more in line with the control requirements, greatly reducing the dependence of traditional dyeing processes on manual experience. Especially in the specific application field of dyeing technology, the present invention provides a special solution for the complex nonlinear characteristics, multi-parameter coupling problems and multi-objective optimization requirements of the dyeing process by integrating multiple deep learning algorithms and non-dominated sorting genetic algorithm features. The core contribution of the algorithm features to the solution is reflected in: the multi-structure fusion characteristics of the deep learning algorithm accurately capture the complex relationship between static characteristics and dynamic timing characteristics in the dyeing process; the multi-objective balance characteristics of the non-dominated sorting genetic algorithm effectively solve the trade-off problem between dyeing quality and production efficiency; the hierarchical characteristics of the layered control architecture perfectly adapt to the requirements of different control granularities and time scales in the dyeing process, and achieve a comprehensive optimization effect of stable dyeing quality, reduced energy consumption and improved production efficiency.

[0099] above Figure 2 The dyeing process optimization system based on intelligent control in the embodiment of the present invention is described in detail from the perspective of modular functional entities. The dyeing process optimization device based on intelligent control in the embodiment of the present invention is described in detail from the perspective of hardware processing.

[0100] Figure 3It is a structural diagram of a dyeing process optimization device based on intelligent control provided in an embodiment of the present invention, the dyeing process optimization device 300 based on intelligent control can produce relatively large differences due to different configurations or performances, and can include one or more processors (central processing units, CPU) 310 (for example, one or more processors) and memory 320, one or more storage media 330 (for example, one or more massive storage device ends) storing application programs 333 or data 332. Wherein, memory 320 and storage medium 330 can be temporary storage or persistent storage. The program stored in storage medium 330 can include one or more modules (not shown), and each module can include a series of instruction operations based on the dyeing process optimization device 300 of intelligent control. Further, processor 310 can be arranged to communicate with storage medium 330, and a series of instruction operations in storage medium 330 are executed on dyeing process optimization device 300 based on intelligent control to realize the step of the dyeing process optimization method based on intelligent control.

[0101] The dyeing process optimization device 300 based on intelligent control may further include one or more power supplies 340, one or more wired or wireless network interfaces 350, one or more input and output interfaces 360, and / or one or more operating systems 331, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, etc. It will be understood by those skilled in the art that Figure 3 The structure of the dyeing process optimization device based on intelligent control shown does not constitute a limitation on the dyeing process optimization device based on intelligent control provided by the present invention, and may include more or fewer components than shown in the figure, or combine certain components, or arrange the components differently.

[0102] The present invention also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. The computer-readable storage medium stores instructions, and when the instructions are run on a computer, the computer executes the steps of the dyeing process optimization method based on intelligent control.

[0103] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0104] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a dyeing process optimization device based on intelligent control (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0105] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A dyeing process optimization method based on intelligent control, characterized in that: The method comprises: Collect key parameters of the dyeing process to obtain the original data set of the dyeing process; Preprocessing and feature extraction are performed based on the original dyeing process data set to obtain a dyeing process feature data set; A deep learning prediction model is constructed using the dyeing process feature dataset and a multi-objective optimization is performed to obtain an optimal dyeing process parameter combination, wherein the multi-objective optimization adopts a non-dominated sorting genetic algorithm; A dyeing scheme is analyzed according to the optimal dyeing process parameters to obtain an intelligent dyeing scheme.

2. The dyeing process optimization method based on intelligent control according to claim 1, characterized in that: The key parameters of the dyeing process are collected to obtain the original data set of the dyeing process, including: Multiple sensor collection points are set for dye concentration, dye liquor temperature, dyeing time, liquid ratio, pH value, auxiliary agent addition amount and dyeing machine speed to obtain a multi-dimensional parameter collection network; Connecting the multi-dimensional parameter acquisition network to an industrial-grade programmable logic controller to obtain a field control unit, wherein the programmable logic controller is equipped with a processor of no less than 100 MHz; Utilizing the field control unit to execute a data filtering algorithm to process the collected data to obtain parameter data; Transmitting the parameter data to a central data server via industrial Ethernet to obtain a parameter data stream; Setting a dual-machine hot backup mechanism for the parameter data stream to obtain a data storage solution; Based on the data storage solution, the parameters of the entire dyeing process are continuously collected to obtain an original dyeing process data set, which includes parameter records of all processes from boiling the cloth and dyeing bath to deurethane washing and rolling the cotton.

3. The dyeing process optimization method based on intelligent control according to claim 1, characterized in that: The preprocessing and feature extraction based on the original dyeing process data set to obtain the dyeing process feature data set includes: performing data cleaning on the dyeing process original data set to obtain an integrity cleaned data set; Standardizing the integrity cleansing data set through nonlinear transformation to obtain balanced standard data; Applying a filter to the equalized standard data to perform time series noise reduction processing to obtain smoothed data; Extracting wavelet coefficients and statistical features from the smoothed data to obtain a multidimensional feature matrix; Executing a dimensionality reduction technique on the multidimensional feature matrix to perform feature compression to obtain a low-dimensional feature set; The low-dimensional feature set is fused with the dyeing process mechanism features to obtain the dyeing process feature data set.

4. The dyeing process optimization method based on intelligent control according to claim 1, characterized in that: The method uses the dyeing process feature data set to construct a deep learning prediction model and perform multi-objective optimization to obtain the optimal dyeing process parameter combination, wherein the multi-objective optimization adopts a non-dominated sorting genetic algorithm, including: The dyeing process feature dataset was divided into a training set, a validation set, and a test set in a ratio of 7:2:1; Based on the training set, a multi-structure deep learning model including a multi-layer perceptron, a convolutional neural network and a recurrent neural network is constructed to obtain a basic model for dyeing quality prediction, wherein the multi-layer perceptron includes four hidden layers, the convolutional neural network includes three convolution blocks, and the recurrent neural network adopts a bidirectional structure; Optimizing the staining quality prediction basic model using the validation set to obtain a staining quality prediction candidate model; Using the test set to evaluate and verify the performance of the dyeing quality prediction candidate model, a deep learning prediction model for predicting color fastness, color difference value, dyeing levelness and color fringing rate is obtained; Setting multi-objective optimization variables and constraints based on the deep learning prediction model to obtain dyeing process parameter optimization variables, wherein the dyeing process parameter optimization variables include dye bath temperature, dye ratio, and holding time; The dyeing process parameter optimization variables are calculated using a non-dominated sorting genetic algorithm to obtain a Pareto optimal solution set, and a compromise solution is screened from the Pareto optimal solution set using a weighted TOPSIS method to obtain an optimal dyeing process parameter combination.

5. The dyeing process optimization method based on intelligent control according to claim 4, characterized in that: The multi-structure deep learning model including a multi-layer perceptron, a convolutional neural network and a recurrent neural network is constructed based on the training set to obtain a basic model for dyeing quality prediction, wherein the multi-layer perceptron includes four hidden layers, the convolutional neural network includes three convolution blocks, and the recurrent neural network adopts a bidirectional structure, including: Adding Gaussian noise and random transformation to the training set to obtain an expanded training data set; Performing forward propagation calculation on the expanded training data set through a 4-layer fully connected network to obtain the multilayer perceptron, wherein the number of hidden layer neurons in the 4 hidden layers is 128, 256, 128, and 64, respectively; Performing feature extraction on the expanded training data set through three convolution blocks to obtain the convolutional neural network, wherein the convolution kernel size of the three convolution blocks is 3×1, the initial number of channels is 32, and the number of channels doubles after each convolution block; Performing sequence modeling on the expanded training data set through a bidirectional gated recurrent unit to obtain the recurrent neural network, wherein the bidirectional gated recurrent unit comprises two layers, a hidden state dimension of 128, and an attention mechanism is added to the output layer; The multilayer perceptron, the convolutional neural network, and the recurrent neural network are trained using an Adam optimizer and cosine annealing learning rate scheduling to obtain a neural network model; The neural network model is stacked and generalized to obtain the basic model for staining quality prediction.

6. The dyeing process optimization method based on intelligent control according to claim 1, characterized in that: The dyeing scheme analysis is performed according to the optimal dyeing process parameters to obtain an intelligent dyeing scheme, including: Performing process data mapping conversion according to the optimal dyeing process parameter combination to obtain a standardized control parameter matrix; Performing hierarchical parsing on the standardized control parameter matrix to obtain a field execution layer parameter set, a process control layer parameter set, and an optimization decision layer parameter set; Building a process execution control strategy based on the field execution layer parameter set, the process control layer parameter set, and the optimization decision layer parameter set to obtain a dynamic control execution plan; The dynamic control execution scheme is applied to the dyeing process, and an intelligent dyeing scheme is obtained through real-time parameter monitoring and feedback adjustment.

7. The dyeing process optimization method based on intelligent control according to claim 6, characterized in that: The process data mapping conversion is performed according to the optimal dyeing process parameter combination to obtain a standardized control parameter matrix, including: Sorting the optimal dyeing process parameter combinations according to the process flow sequence to obtain a process parameter sequence, wherein the process parameter sequence includes cloth boiling dye bath parameters, emulsified dye parameters, heating parameters, heat preservation parameters, cooling parameters and deurethane process parameters; Performing dimension normalization processing on the process parameter sequence to obtain a process parameter matrix; constructing a correlation diagram between process variables based on the process parameter matrix to obtain a parameter correlation network; Performing cluster analysis on the parameter correlation network to obtain a highly coupled parameter group; Performing joint optimization conversion according to the highly coupled parameter group to obtain a decoupling parameter matrix; The decoupling parameter matrix is restructured according to the control level requirements to obtain a standardized control parameter matrix. The restructured reorganization allocates parameters to corresponding control levels and sets a priority order.

8. A dyeing process optimization system based on intelligent control, characterized in that: For implementing the dyeing process optimization method based on intelligent control according to any one of claims 1 to 7, the dyeing process optimization system based on intelligent control comprises: The acquisition module is used to collect key parameters of the dyeing process and obtain the original data set of the dyeing process; An extraction module is used to perform preprocessing and feature extraction based on the original dyeing process data set to obtain a dyeing process feature data set; An optimization module is used to construct a deep learning prediction model using the dyeing process feature data set and perform multi-objective optimization to obtain an optimal dyeing process parameter combination, wherein the multi-objective optimization adopts a non-dominated sorting genetic algorithm; The analysis module is used to analyze the dyeing scheme according to the optimal dyeing process parameters to obtain an intelligent dyeing scheme.

9. A dyeing process optimization device based on intelligent control, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and when the processor executes the computer program, the dyeing process optimization method based on intelligent control according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the processor is enabled to perform the dyeing process optimization method based on intelligent control according to any one of claims 1 to 7.

Images