Silk-covered enameled winding process optimization control method and system based on intelligent analysis

Abstract

The invention relates to the technical field of process optimization control, and discloses a silk-covered enameled winding process optimization control method and system based on intelligent analysis. The method comprises the steps that multi-area production parameters are collected in real time and preprocessed; constructing parameter and product quality mapping through principal component dimensionality reduction, grey correlation and support vector regression; training an Actor-Critic structure by using a depth deterministic strategy gradient network to obtain an adjustment strategy; a non-dominated sorting genetic algorithm is combined with analytic hierarchy process to carry out multi-objective optimization, and a control scheme for balancing mass, efficiency and energy consumption is realized. According to the application, on the basis of considering the complex coupling relationship among the process parameters, multi-objective dynamic balance optimization of product quality, production efficiency and energy consumption is realized, the process parameters can be adaptively adjusted according to the production state and the task demand, and the stability and the optimization degree of the silk-covered enameled winding production process are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of process optimization control, in particular to a silk-covered winding process optimization control method and system based on intelligent analysis. BACKGROUND

[0002] Silk-covered winding is a core component of electrical equipment such as motors, transformers, and inductors, and its production process directly affects the performance and reliability of electrical equipment. The traditional silk-covered winding production process includes annealing, painting, baking, cooling, lubrication, and take-up, etc. Each process has multiple key process parameters that need to be accurately controlled. In the prior art, the process control of silk-covered winding mainly adopts the combination of experience parameter setting and PID control. The operator sets the initial process parameters according to the product specifications and experience, and the control system maintains the parameters within the set range through the feedback loop. Some advanced enterprises also try to introduce data acquisition systems and statistical process control methods to analyze historical production data, optimize process parameter configuration, and improve product quality stability. In recent years, some production enterprises have begun to explore the application of machine learning and optimization algorithms in process parameter optimization, but most of them are limited to single-process or single-target optimization.

[0003] However, the prior art has obvious deficiencies in the optimization control of silk-covered winding process. First, there is a complex coupling relationship and nonlinear influence between process parameters, and traditional experience setting cannot fully consider these complex relationships, resulting in inaccurate parameter configuration. Second, multiple processes in the production process interact with each other, and local optimization of a single process often cannot achieve global optimal results. Third, most existing control methods focus on single-target optimization, such as product quality or production efficiency, making it difficult to find the best balance point between product quality, production efficiency, and energy consumption. In addition, the traditional control strategy lacks adaptability to changes in production environment and process fluctuations, and cannot dynamically adjust process parameters according to real-time state, resulting in low stability of the production process. Finally, the existing technology lacks effective knowledge accumulation and continuous optimization mechanism, and cannot extract experience from historical production data to improve the process control strategy, limiting the continuous optimization of the production process. SUMMARY

[0004] The present application provides a silk-covered winding process optimization control method and system based on intelligent analysis, which realizes dynamic balance optimization of multiple targets such as product quality, production efficiency, and energy consumption based on considering the complex coupling relationship between process parameters, and can adaptively adjust process parameters according to production state and task demand, improving the stability and optimization degree of the silk-covered winding production process.

[0005] In a first aspect, the application provides a silk-covered winding process optimization control method based on intelligent analysis, which comprises: collecting and preprocessing the temperature, humidity, tension, speed and current parameters of multiple regions of the production line in real time to obtain a standardized processed process parameter dataset; based on the process parameter dataset, performing dimension reduction processing on high-dimensional data through principal component analysis, combining gray correlation analysis and support vector regression algorithm to construct a process parameter influence model, and obtaining the mapping relationship between the process parameters and product quality indicators; inputting the mapping relationship into a deep deterministic policy gradient network, training an Actor-Critic network structure by setting a reward function and an experience replay mechanism, and obtaining a dynamic adjustment strategy of the process parameters; optimizing the dynamic adjustment strategy through a non-dominated sorting genetic algorithm, dynamically adjusting the weights of the objective function through an analytic hierarchy process, executing an optimal parameter adjustment sequence through model predictive control, and obtaining a multi-objective optimization control scheme that balances product indicators, production efficiency and energy consumption.

[0006] In a first implementation of the first aspect, the collecting and preprocessing the temperature, humidity, tension, speed and current parameters of multiple regions of the production line in real time to obtain a standardized processed process parameter dataset comprises: configuring temperature sensors, humidity sensors, tension sensors, speed sensors, current sensors and high-precision image acquisition devices in the annealing area, the painting area, the baking area, the cooling area, the lubricating area and the take-up area, acquiring real-time data at a frequency of 10 times per second to obtain multi-region production raw data; transmitting the multi-region production raw data to a central data processing unit, filtering abnormal values and noise interference through edge computing technology to obtain a preliminarily cleaned data stream; applying moving average filtering, median filtering and wavelet transform denoising algorithms to the preliminarily cleaned data stream for data cleaning to obtain denoised process parameters; performing interpolation processing on missing data in the denoised process parameters, filling in data gaps through linear interpolation or polynomial interpolation to obtain a parameter sequence; associating and matching the parameter sequence with product quality indicators to establish a quality database containing conductor diameter, insulating paint film thickness, paint film adhesion, voltage resistance performance, and obtaining parameter-quality correspondence; standardizing the parameter-quality correspondence, mapping each parameter to a unified numerical interval through z-score or min-max method to obtain a standardized processed process parameter dataset.

[0007] In a second implementation of the first aspect, the mapping relationship between the process parameters and the product quality indicators is obtained by dimension reduction processing of high-dimensional data through principal component analysis, combined with grey correlation degree analysis and support vector regression algorithm to construct a process parameter influence model based on the process parameter data set, comprising: applying principal component analysis to the standardized process parameter data set, calculating eigenvalues and eigenvectors, extracting main eigenvectors that can explain data variation, and obtaining a key feature set after dimension reduction; the key feature set after dimension reduction is combined with wavelet packet transform technology for multi-scale decomposition to capture the dynamic change characteristics of temperature, tension and speed parameters in different frequency domains, and obtain time series eigenvectors; a grey correlation degree analysis model between the process parameters and the product quality indicators is constructed based on the time series eigenvectors, the correlation degrees of each process parameter to the film uniformity, adhesion, and insulation performance quality indicators are calculated, and a correlation degree matrix is obtained; the correlation degree matrix is combined with expert experience and historical production data to introduce a fuzzy comprehensive evaluation method, the weight coefficients of each process parameter are set, and a fuzzy relationship matrix of process parameters-product quality is obtained; the annealing temperature, painting speed, paint viscosity, and baking temperature key parameters in the fuzzy relationship matrix are input into the support vector regression algorithm, mapped to a high-dimensional feature space through a Gaussian kernel function, a regression model with a determination coefficient R² not less than 0.92 is trained, and a parameter-quality prediction model is obtained; the time series characteristics of the process parameters in the parameter-quality prediction model are analyzed through an autoregressive integrated moving average model to predict the short-term change trend of the key parameters, a multi-level process parameter influence model library is constructed, and the mapping relationship between the process parameters and the product quality indicators is obtained.

[0008] In a third implementation form of the first aspect, the inputting the mapping relationship into the deep deterministic policy gradient network comprises: modeling a Markov decision process for the mapping relationship between the process parameters and the product quality indicators, defining a state space comprising a process parameter set, an equipment operating state and a semi-finished product quality feature, defining an action space comprising adjustment parameters for a temperature adjustment amplitude, a tension change amount and a painting speed change amount, and obtaining a state-action environment model; constructing a reward function based on the state-action environment model, calculating a weighted sum of a product quality score and a production efficiency, assigning a positive reward value to a high-quality product, assigning a negative reward value to a substandard product, and introducing an energy consumption penalty term to obtain a reinforcement learning training signal; inputting the reinforcement learning training signal into the deep deterministic policy gradient network, constructing a four-layer fully connected Actor network structure, matching the number of input layer neurons with the state dimension, setting two hidden layers with 256 and 128 nodes respectively, configuring a ReLU activation function, using a Tanh activation function in the output layer to map to the action space, and obtaining a parameter adjustment strategy network; inputting the state-action environment model into a four-layer fully connected Critic network structure, receiving state information at the input layer, fusing action information at the second hidden layer, using a ReLU activation function, and calculating Q values at the output layer to obtain a state-action value evaluation network; constructing an experience replay pool with a capacity of 100,000 based on the parameter adjustment strategy network and the state-action value evaluation network, randomly sampling a batch size of 128 samples each time for training, setting a target network update rate τ = 0.001 for soft updating, introducing an Ornstein-Uhlenbeck noise process with an initial noise intensity of 0.5 and a decay coefficient of 0.99, and obtaining an exploration-exploitation balance mechanism; designing a hierarchical priority constraint condition for the exploration-exploitation balance mechanism, limiting the change amplitudes of the temperature and tension key parameters within a safe range, converting the constraint into a component of the reward function by a penalty function method, and obtaining a dynamic adjustment strategy for the process parameters after more than 500,000 cumulative interaction samples.

[0009] In a fourth implementation of the first aspect, the dynamic adjustment strategy is optimized by a non-dominated sorting genetic algorithm, a target function weight is dynamically adjusted by an analytic hierarchy process, an optimal parameter adjustment sequence is executed by a model predictive control to obtain a multi-objective optimization control scheme balancing product indexes, production efficiency and energy consumption, comprising: constructing a multi-objective optimization model based on the dynamic adjustment strategy of the process parameters, defining a target function set including a product quality index function, a production efficiency function and an energy consumption function, wherein the product quality index function is composed of a weighted sum of insulation strength and paint film uniformity indexes, the production efficiency function is composed of effective output per unit time, and the energy consumption function is composed of a standardized sum of electric energy and thermal energy, to obtain an optimization target function set; applying an improved non-dominated sorting genetic algorithm to the optimization target function set, setting a population size of 100, a maximum iteration number of 200, a crossover probability of 0.85, and a mutation probability of 0.15, maintaining population diversity by a non-dominated sorting and crowding distance calculation mechanism, to obtain a Pareto optimal solution set of process parameter configuration; introducing a heuristic operator based on problem characteristics to the Pareto optimal solution set, applying an intelligent crossover operator considering process parameter continuity and a parameter sensitivity oriented adaptive mutation operator to accelerate algorithm convergence, to obtain multiple feasible process parameter configuration schemes; inputting the process parameter configuration scheme into an analytic hierarchy process decision support module, dynamically adjusting weight coefficients of each target function in combination with current production targets, to form a final process parameter optimization scheme and obtain an optimization control sequence; applying a model predictive control strategy based on the optimization control sequence, setting a prediction horizon of 15, predicting system behavior at multiple time steps in the future using the current state of the system, determining an optimal control sequence by solving a rolling optimization problem, to obtain a model predictive control action; introducing a recursive least squares method to the model predictive control action for online parameter updating, configuring a multi-level safety guarantee mechanism including parameter change rate limitation, safety boundary check and emergency intervention logic to ensure safety of the control action, to obtain a multi-objective optimization control scheme balancing product indexes, production efficiency and energy consumption.

[0010] In a fifth implementation of the first aspect, the heuristic operator based on problem characteristics is introduced to the set of Pareto optimal solutions, and intelligent crossover operators considering process parameter continuity and parameter sensitivity oriented adaptive mutation operators are applied to accelerate the convergence of the algorithm and obtain multiple feasible process parameter configuration schemes, including: performing historical data correlation analysis on the process parameters in the set of Pareto optimal solutions, establishing a correlation matrix between the parameters, identifying strong coupling relationships between the parameters, and obtaining a parameter dependency relationship network; designing intelligent crossover operators based on the parameter dependency relationship network, performing collaborative crossover operations on strongly correlated parameters, maintaining the consistency and continuity of parameter changes, and obtaining child individuals that meet process constraints; calculating parameter sensitivity coefficients for each process parameter in the child individuals, quantifying the influence of each parameter on the objective function through partial derivative analysis of historical process data, and obtaining a parameter sensitivity ranking table; constructing an adaptive mutation operator according to the parameter sensitivity ranking table, setting a smaller mutation step size for parameters with high sensitivity and a larger mutation step size for parameters with low sensitivity, and obtaining an adaptive mutation strategy; applying a simulated annealing mechanism to the adaptive mutation strategy, setting the initial temperature to 100 and the cooling coefficient to 0.95, dynamically adjusting the mutation probability according to the algorithm iteration process, maintaining a higher exploratory nature in the early stage, and gradually converging to local fine search in the later stage to obtain an optimized search trajectory; retaining the elite solutions in the optimized search trajectory to the next generation, while performing diversity enhancement processing on solutions with high repetition in the population, introducing chaotic disturbance to avoid premature convergence of the population, and obtaining diversified process parameter configuration schemes that meet the process continuity constraints after 200 iterations.

[0011] In a sixth implementation form of the first aspect, the inputting the process parameter configuration scheme into the analytic hierarchy process decision support module, dynamically adjusting the weight coefficients of each objective function in combination with a current production target, forming a final process parameter optimization scheme, and obtaining an optimized control sequence, comprises: constructing a three-level analytic hierarchy structure for the multiple feasible process parameter configuration schemes, the first level being a total target layer, the second level being a criterion layer including three indexes of product quality, production efficiency and energy consumption, and the third level being a scheme layer including multiple parameter configuration schemes, to obtain an analytic hierarchy judgment matrix; performing pairwise comparison based on the analytic hierarchy judgment matrix, constructing a relative importance comparison matrix between indexes of each criterion layer by using a 1-9 scale method, and calculating the maximum eigenvalue and the corresponding eigenvector by a characteristic value method to obtain the weight coefficients of the criterion layer; performing consistency check on the weight coefficients of the criterion layer, calculating a consistency index CI and a random consistency ratio CR, considering that the judgment matrix has satisfactory consistency when the CR is less than 0.1, otherwise readjusting the judgment matrix, and obtaining a consistency check result of the weight coefficients of the indexes of the criterion layer; adjusting the relative weights of the indexes of the criterion layer according to the consistency check result, automatically increasing or decreasing the weight coefficients through fuzzy rules in combination with a current production task demand, increasing the weight of production efficiency in an emergency order, increasing the weight of the quality index in high-end product production, and obtaining a dynamic weight distribution scheme; applying the dynamic weight distribution scheme to comprehensive evaluation of each process parameter configuration scheme, calculating the weighted total score of each scheme under each objective function, and sorting according to the total score to select the configuration scheme with the highest comprehensive score, and obtaining an optimized process parameter combination; and performing time sequence expansion on the optimized process parameter combination, converting the static process parameters into time sequence parameter adjustment strategies according to the dynamic characteristics of the production process, forming an optimal parameter adjustment trajectory including future 15 time steps, and obtaining an optimized control sequence.

[0012] In a second aspect, the application provides a wire-coated enameled winding process optimization control system based on intelligent analysis, comprising: An acquisition module is configured to acquire and preprocess temperature, humidity, tension, speed and current parameters of multiple regions of a production line in real time, and obtain a standardized process parameter dataset; A dimension reduction module is configured to perform dimension reduction processing on high-dimensional data by principal component analysis based on the process parameter dataset, construct a process parameter influence model by combining gray correlation degree analysis and support vector regression algorithm, and obtain a mapping relationship between process parameters and product quality indexes; A training module is configured to input the mapping relationship into a deep deterministic policy gradient network, train an Actor-Critic network structure by setting a reward function and an experience replay mechanism, and obtain a dynamic adjustment strategy for process parameters; The execution module is used for non-dominated sorting genetic algorithm optimization of the dynamic adjustment strategy, dynamic adjustment of target function weight by combining with analytic hierarchy process, optimal parameter adjustment sequence by model predictive control, and a multi-objective optimization control scheme balancing product indexes, production efficiency and energy consumption.

[0013] In a third aspect, an intelligent analysis-based silk-covered winding process optimization control device is provided, comprising a memory and at least one processor, the memory storing instructions; the at least one processor invokes the instructions in the memory to enable the intelligent analysis-based silk-covered winding process optimization control device to perform the intelligent analysis-based silk-covered winding process optimization control method described above.

[0014] In a fourth aspect, a computer readable storage medium is provided, the computer readable storage medium storing instructions, when running on a computer, enabling the computer to perform the intelligent analysis-based silk-covered winding process optimization control method described above.

[0015] The technical scheme provided in the application comprises the following steps: collecting and preprocessing the parameters of temperature, humidity, tension, speed and current in multiple regions of the production line in real time, establishing a comprehensive process parameter dataset, laying a solid data foundation for subsequent intelligent analysis, applying principal component analysis based on the process parameter dataset to reduce the dimension of high-dimensional data, effectively eliminating data redundancy, and extracting key features that significantly affect product quality; combining grey correlation analysis and a support vector regression algorithm to construct a process parameter influence model, accurately quantifying the influence degree and mapping relationship of each process parameter on product quality indicators, overcoming the limitations of traditional methods in dealing with nonlinear relationships between parameters, wherein the kernel function mapping mechanism of the support vector regression algorithm enables the model to capture complex nonlinear relationships, and a high-precision prediction ability with a determination coefficient R 2 not less than 0.92 provides a reliable basis for subsequent optimization; inputting the mapping relationship into a deep deterministic policy gradient network, fully utilizing the advantages of deep learning in continuous action space, and realizing efficient training of the Actor-Critic network structure through the design of a reasonable reward function and an experience replay mechanism, solving the limitations of traditional reinforcement learning in continuous control problems, breaking the time correlation between samples through the experience replay mechanism, and greatly improving the learning efficiency, and the deterministic policy gradient characteristic of the DDPG algorithm makes it particularly suitable for dealing with precise parameter regulation problems in silk-covered wire production, and through the learning of 500,000 interaction samples, the network obtains a highly intelligent process parameter dynamic adjustment strategy; the dynamic adjustment strategy is optimized by a non-dominated sorting genetic algorithm, which fully utilizes the outstanding advantages of the NSGA-II algorithm in dealing with multi-objective optimization problems, and the non-dominated sorting and congestion degree calculation mechanism ensures the diversity and uniform distribution of the solution set, and the parameter sensitivity-oriented adaptive mutation operator and the intelligent crossover operator considering the continuity of process parameters further improve the convergence speed and solution quality of the algorithm; the weights of the objective function are dynamically adjusted by combining the analytic hierarchy process, the product quality, production efficiency and energy consumption are flexibly balanced according to different production demands, and intelligent decision-making of the production process is realized; the optimal parameter adjustment sequence is executed through model predictive control, the static optimization result is converted into a dynamic control strategy, and the prediction time domain design and rolling optimization mechanism effectively cope with system disturbances and model uncertainties. The application organically combines various artificial intelligence algorithms, fully considers the matching of algorithm characteristics and silk-covered wire winding production process characteristics, such as principal component analysis to solve high-dimensional data redundancy, support vector regression to deal with nonlinear mapping relationships, deep reinforcement learning to realize continuous action space control, NSGA-II algorithm to balance multi-objective optimization, and analytic hierarchy process to support dynamic decision-making, each algorithm makes a unique contribution to specific functional links, forming a complete intelligent process optimization control system, and significantly improving the quality stability, process optimization degree and intelligent level of silk-covered wire winding production. BRIEF DESCRIPTION OF DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiment description. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can be obtained by those skilled in the art without any creative effort based on these drawings.

[0017] Figure 1 An embodiment of the method for optimizing and controlling the silk-covered winding process based on intelligent analysis in the embodiments of the present application; Figure 2 An embodiment of the system for optimizing and controlling the silk-covered winding process based on intelligent analysis in the embodiments of the present application; Figure 3 An embodiment of the structure schematic diagram of the device for optimizing and controlling the silk-covered winding process based on intelligent analysis in the embodiments of the present application. DETAILED DESCRIPTION

[0018] The embodiments of the present application provide a method and system for optimizing and controlling the silk-covered winding process based on intelligent analysis. The terms "first", "second", "third", "fourth" and the like (if any) in the specification and claims of the present application and the above drawings are used to distinguish similar objects, and do not necessarily mean a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the term "comprising" or "having" and any variation thereof is intended to cover non-exclusive inclusion, for example, a process, method, system, product or device including a series of steps or units does not necessarily limit to those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0019] For the convenience of understanding, the specific process of the embodiments of the present application will be described below. Please refer to Figure 1 An embodiment of the method for optimizing and controlling the silk-covered winding process based on intelligent analysis in the embodiments of the present application includes: Step S101, real-time acquisition and preprocessing of temperature, humidity, tension, speed and current parameters of the production line multi-region, to obtain a standardized processed process parameter data set; Step S102, based on the process parameter data set, dimension reduction processing of high-dimensional data is performed through principal component analysis, and a process parameter influence model is constructed by combining gray correlation degree analysis and support vector regression algorithm, to obtain the mapping relationship between the process parameters and the product quality indicators; Step S103, input the mapping relationship into a deep deterministic policy gradient network, train an Actor-Critic network structure by setting a reward function and an experience replay mechanism, and obtain a dynamic adjustment strategy of the process parameters; Step S104, perform a non-dominated sorting genetic algorithm optimization on the dynamic adjustment strategy, combine an analytic hierarchy process to dynamically adjust a target function weight, execute an optimal parameter adjustment sequence through model predictive control, and obtain a multi-objective optimization control scheme balancing product indexes, production efficiency, and energy consumption.

[0020] It can be understood that the execution subject of the present application can be a silk package varnish package winding process optimization control system based on intelligent analysis, and can also be a terminal or a server, which is not limited here. The server is taken as an example for description of the embodiments of the present application.

[0021] Specifically, when collecting parameters in multiple areas of the production line, temperature sensors, humidity sensors, tension sensors, speed sensors, current sensors, and high-precision image acquisition devices are configured in the annealing area, the painting area, the baking area, the cooling area, the lubricating area, and the take-up area, respectively, and the collection frequency is 10 times per second. After these raw data are transmitted to the central data processing unit through industrial Internet of Things devices, the edge computing technology performs preliminary filtering on the obvious abnormal values and noise. Subsequently, moving average filtering, median filtering, and wavelet transform denoising algorithms are used to clean the data. The moving average filtering replaces the original point with the average value of multiple consecutive data points, the median filtering selects the median value within the window, and the wavelet transform removes high-frequency noise through multi-scale decomposition. For missing data, linear interpolation or polynomial interpolation methods are used to fill in the missing data to ensure data integrity. Then, the complete parameter sequence is associated and matched with the product quality indicators to establish a quality database, and finally, the z-score or min-max method is used for standardization processing to map each parameter to a unified numerical interval, obtaining the standardized process parameter data set. Principal component analysis is applied to reduce the dimensionality of high-dimensional data, and by calculating the eigenvalues and eigenvectors of the covariance matrix, the main eigenvectors that can explain the data variation are extracted. The key feature set after dimensionality reduction is combined with wavelet packet transform technology for multi-scale decomposition to capture the dynamic change characteristics of temperature, tension, and speed parameters in different frequency domains. Based on the time series eigenvectors, a grey correlation degree analysis model is constructed to calculate the correlation degree of each process parameter to the quality indicators of film uniformity, adhesion, and insulation performance. The grey correlation degree analysis quantifies the influence of process parameters on quality indicators by calculating the correlation coefficient between the reference sequence and the comparison sequence. The correlation matrix is combined with expert experience and historical production data to introduce a fuzzy comprehensive evaluation method, and the weight coefficients are set. The key parameters of annealing temperature, painting speed, paint viscosity, and baking temperature in the fuzzy relationship matrix are input into the support vector regression algorithm, which is trained through the Gaussian kernel function mapping to the high-dimensional feature space, with a determination coefficient R² not less than 0.92. The time series characteristics of process parameters in the parameter-quality prediction model are analyzed through the autoregressive integrated moving average model to predict the short-term trend of key parameters, and finally a multi-level process parameter influence model library is constructed to obtain the mapping relationship between process parameters and product quality indicators.

[0022] The mapping relationship is input into a deep deterministic policy gradient network for control policy construction. First, the mapping relationship between the process parameters and the product quality indicators is modeled as a Markov decision process, and the state space and action space are defined. The state space includes the process parameter set, the equipment operating state and the semi-finished product quality characteristics, and the action space includes the adjustment parameters of the temperature adjustment amplitude, the tension change amount and the painting speed change amount. Based on the state-action environment model, a reward function is constructed, which is calculated by weighting the product quality score and the production efficiency, and a positive reward value is given to high-quality products and a negative reward value is given to unqualified products, and an energy consumption penalty term is introduced. Reinforcement learning training signals are input into a deep deterministic policy gradient network to construct a four-layer fully connected Actor network structure, the number of input layer neurons matches the state dimension, the two hidden layers have 256 and 128 nodes respectively, and ReLU activation function is configured, and Tanh activation function is used in the output layer. The state-action environment model is input into a four-layer fully connected Critic network structure to evaluate the value function of the state-action pair. Based on the parameter adjustment strategy network and the state-action value evaluation network, an experience replay pool with a capacity of 100000 is constructed, a batch size of 128 samples is randomly sampled each time, the target network update rate τ is set to 0.001 for soft updating, and an Ornstein-Uhlenbeck noise process is introduced to realize the exploration-exploitation balance. The exploration-exploitation balance mechanism is designed with hierarchical priority constraints to limit the change amplitude of the temperature and tension key parameters within a safe range, and the constraints are converted into a component of the reward function by a penalty function method. After more than 500,000 interactive samples are accumulated, the dynamic adjustment strategy of the process parameters is obtained.

[0023] Multi-objective optimization and adaptive control are implemented for dynamic adjustment strategy. Multi-objective optimization model is constructed based on dynamic adjustment strategy of process parameters. Objective function set is defined including product quality index function, production efficiency function and energy consumption function. Improved non-dominated sorting genetic algorithm is applied to optimization objective function set. Population size is set as 100, maximum iteration number is set as 200, crossover probability is set as 0.85, mutation probability is set as 0.15. Population diversity is maintained by non-dominated sorting and crowding distance calculation mechanism. Pareto optimal solution set of process parameter configuration is obtained. Heuristic operator based on problem characteristics is introduced to Pareto optimal solution set. Intelligent crossover operator considering continuity of process parameters and adaptive mutation operator oriented to parameter sensitivity are applied. Process parameter configuration scheme is input into analytic hierarchy process decision support module. Weight coefficients of each objective function are dynamically adjusted according to current production target. Model predictive control strategy is applied based on optimal control sequence. Prediction horizon is set as 15. Optimal control sequence is determined by solving rolling optimization problem. Recursive least squares method is introduced to model predictive control action for online parameter update. Multi-level safety protection mechanism is configured including parameter change rate limitation, safety boundary check and emergency intervention logic. Multi-objective optimization control scheme balancing product index, production efficiency and energy consumption is obtained.

[0024] In the embodiment of the application, by real-time collection and preprocessing of the temperature, humidity, tension, speed and current parameters of the production line in multiple regions, a comprehensive process parameter dataset is established, laying a solid data foundation for subsequent intelligent analysis; based on the process parameter dataset, principal component analysis is applied to reduce the dimension of high-dimensional data, effectively eliminating data redundancy and extracting key features that significantly affect product quality; the process parameter influence model constructed by combining grey correlation analysis and support vector regression algorithm accurately quantifies the influence degree and mapping relationship of each process parameter on product quality indicators, overcoming the limitations of traditional methods in handling nonlinear relationships between parameters, and the kernel function mapping mechanism of the support vector regression algorithm enables the model to capture complex nonlinear relationships, with a high-precision prediction ability of a determination coefficient R² not less than 0.92, providing a reliable basis for subsequent optimization; the mapping relationship is input into the deep deterministic policy gradient network, fully utilizing the advantages of deep learning in continuous action space, and through the design of a reasonable reward function and an experience replay mechanism, the efficient training of the Actor-Critic network structure is realized, solving the limitations of traditional reinforcement learning in continuous control problems, and the experience replay mechanism breaks the temporal correlation between samples, significantly improving the learning efficiency, while the deterministic policy gradient characteristic of the DDPG algorithm makes it particularly suitable for handling precise parameter regulation problems in silk-covered wire production, and through 500,000 interaction samples learning, the network obtains a highly intelligent process parameter dynamic adjustment strategy; the non-dominated sorting genetic algorithm optimization is performed on the dynamic adjustment strategy, fully utilizing the outstanding advantages of the NSGA-II algorithm in handling multi-objective optimization problems, and the non-dominated sorting and congestion degree calculation mechanism ensures the diversity and uniform distribution of the solution set, and the parameter sensitivity-oriented adaptive mutation operator and the intelligent crossover operator considering the continuity of process parameters further improve the convergence speed and solution quality of the algorithm; the dynamic adjustment of the objective function weight is combined with the analytic hierarchy process, and according to different production demands, the product quality, production efficiency and energy consumption are flexibly balanced, realizing the intelligent decision-making of the production process; the optimal parameter adjustment sequence is executed through model predictive control, converting the static optimization result into a dynamic control strategy, and the prediction time domain design and rolling optimization mechanism effectively cope with system disturbances and model uncertainties. The present application organically combines various artificial intelligence algorithms, fully considers the matching of algorithm characteristics and silk-covered wire winding production process characteristics, such as principal component analysis to solve high-dimensional data redundancy, support vector regression to handle nonlinear mapping relationships, deep reinforcement learning to realize continuous action space control, NSGA-II algorithm to balance multi-objective optimization, and analytic hierarchy process to support dynamic decision-making, each algorithm makes a unique contribution to specific functional links, forming a complete intelligent process optimization control system, significantly improving the quality stability, process optimization degree and intelligent level of silk-covered wire winding production.

[0025] In a specific embodiment, the process of performing step S101 can specifically include the following steps: Temperature sensors, humidity sensors, tension sensors, speed sensors, current sensors, and high-precision image acquisition devices are configured for the annealing area, the painting area, the baking area, the cooling area, the lubricating area, and the take-up area to obtain real-time data at a sampling frequency of 10 times per second, thereby obtaining multi-area production raw data; The multi-area production raw data is transmitted to a central data processing unit, and edge computing technology is used to filter abnormal values and noise interference, thereby obtaining a preliminarily cleaned data stream; The preliminarily cleaned data stream is subjected to data cleaning by using moving average filtering, median filtering, and wavelet transform denoising algorithms, thereby obtaining denoised process parameters; Interpolation processing is performed on missing data in the denoised process parameters, and data gaps are filled by linear interpolation or polynomial interpolation, thereby obtaining a parameter sequence; The parameter sequence is associated and matched with product quality indicators, a quality database containing conductor diameter, insulating paint film thickness, paint film adhesion, and voltage resistance performance is established, and a parameter-quality correspondence relationship is obtained; The parameter-quality correspondence relationship is subjected to standardization processing, and each parameter is mapped to a unified numerical interval by using a z-score or a min-max method, thereby obtaining a standardized process parameter data set.

[0026] Specifically, in the production process of wire-coated winding, a data acquisition system first deploys multiple types of sensors in six key production areas. PT100 temperature sensors are installed in the annealing area to monitor the winding annealing temperature curve, viscosity sensors and flow sensors are configured in the painting area to record the paint state, infrared temperature sensor arrays are set in the baking area to measure the baking temperature distribution, temperature and humidity sensors are deployed in the cooling area to monitor the cooling environment parameters, tension sensors are used in the lubricating area to monitor the winding tension changes, and a high-speed camera system is configured in the take-up area to capture winding surface defects. Each sensor samples at a frequency of 10 Hz, forming a multi-dimensional time series data stream containing parameters such as temperature, humidity, tension, speed, and current. After these production raw data are transmitted to the central data processing unit through an industrial-grade 5G network, the edge computing device performs preliminary data screening. The edge computing device uses a triple abnormality detection algorithm to filter data in real time: the Z-score method based on statistical distribution identifies abnormal points deviating by more than 3σ; the local outlier factor (LOF) algorithm based on density detects local anomalies; and the threshold judgment based on rules eliminates data points that are obviously outside the process range. This process identifies and labels abnormal values from the raw data, generating a preliminarily cleaned data stream.

[0027] When performing deep noise reduction processing on the preliminary cleaned data stream, different filtering algorithms are applied for different parameter characteristics. For temperature parameters, a 7-point moving average filter is applied to smooth short-term fluctuations; for tension and speed parameters, a 5-point median filter is applied to remove transient peak interference; for image data and complex waveform data, a four-layer wavelet decomposition based on db4 wavelet basis is applied, and the first 50% energy components in the low-frequency approximation coefficients and high-frequency detail coefficients are retained, and the reconstructed signal retains the key features. The three filtering processes combine their respective advantages to generate noise-reduced process parameters. When there is missing data in the process parameter sequence, the system automatically selects an interpolation algorithm based on the characteristics of the missing data. For short-term missing data (<5 seconds), a cubic spline interpolation is applied to maintain the smoothness of the curve; for medium-length missing data (5-30 seconds), data is predicted and filled based on an autoregressive model; for long-term missing data (>30 seconds), data patterns of similar working conditions in history are used to fill in. After filling in all the data gaps through interpolation, a complete and continuous parameter sequence is formed, ensuring the integrity of the data in time.

[0028] When matching the parameter sequence with the product quality indicators, the system establishes a time correspondence database. The quality detection data of each batch of products (conductor diameter, insulating paint film thickness, paint film adhesion, voltage resistance performance) are matched with the process parameters during the production process of the batch, and the time stamps are accurately aligned. The time delay factor in the process flow is also considered, such as the process delay of about 40 seconds from painting to baking. The system constructs a database containing parameter-quality correspondence, recording the mapping relationship between each set of process parameters and the final product quality indicators.

[0029] When standardizing the parameter-quality correspondence, the system processes parameters of different dimensions and numerical ranges. For continuous numerical parameters such as temperature and current, Z-score standardization is applied to make the data mean value 0 and variance 1; for parameters with clear upper and lower limits such as speed and tension, Min-Max standardization is applied to map the data to the [0, 1] interval; for discrete parameters, they are converted to numerical vectors through one-hot encoding. After converting all parameters to a unified numerical interval, a standardized process parameter data set is formed, with a structure of (n, m) matrix, n being the number of time points and m being the feature dimension, containing the standardized values of all process parameters.

[0030] In a specific embodiment, the process of performing step S102 can specifically include the following steps: Applying principal component analysis to the standardized process parameter data set, calculating eigenvalues and eigenvectors, extracting main eigenvectors that can explain data variation, and obtaining a reduced key feature set; The reduced dimension key feature set is subjected to multi-scale decomposition in combination with wavelet packet transform technology to capture dynamic change characteristics of temperature, tension, and speed parameters in different frequency domains, thereby obtaining a time series feature vector; A grey correlation degree analysis model between process parameters and product quality indicators is constructed based on the time series feature vector, correlation degrees of each process parameter to paint film uniformity, adhesion, and insulation performance quality indicators are calculated, and a correlation degree matrix is obtained; The correlation degree matrix is subjected to fuzzy comprehensive evaluation in combination with expert experience and historical production data, weight coefficients of each process parameter are set, and a fuzzy relationship matrix of process parameters and product quality is obtained; Key parameters of annealing temperature, painting speed, paint viscosity, and baking temperature in the fuzzy relationship matrix are input into a support vector regression algorithm, are mapped to a high-dimensional feature space through a Gaussian kernel function, a regression model with a determination coefficient R² not less than 0.92 is trained, and a parameter-quality prediction model is obtained; Time series characteristics of process parameters in the parameter-quality prediction model are analyzed through an autoregressive integrated moving average model, short-term change trends of key parameters are predicted, a multi-level process parameter influence model library is constructed, and a mapping relationship between process parameters and product quality indicators is obtained.

[0031] Specifically, principal component analysis is applied to the standardized process parameter data set, a data covariance matrix is first calculated, the covariance matrix reflects the correlation between different process parameters. The eigenvalues and eigenvectors of the matrix are calculated, the eigenvalues represent the variance contribution rate of the principal components, and the eigenvectors constitute the direction of the principal components. The eigenvalues are sorted in descending order, and the first few eigenvectors with an accumulated contribution rate of more than 85% are selected as the principal components. In the production of silk-covered coil winding, 8 principal components are extracted from the original 30+ process parameters, these principal components linearly combine the original parameters, and can explain most of the total variation of the data. Principal component analysis projects high-dimensional data into low-dimensional space, reduces data redundancy and dimensionality problems, and obtains a reduced dimension key feature set. When the reduced dimension key feature set is subjected to multi-scale decomposition in combination with wavelet packet transform technology, the Daubechies wavelet basis function is selected to decompose the time series data for 5 layers. Wavelet packet transform is an extension of wavelet transform, which not only decomposes the low-frequency part, but also subdivides the high-frequency part, and can provide good localization ability in time and frequency domains. After decomposing the temperature, tension, and speed parameters, the energy distribution characteristics in different frequency bands are analyzed, the low-frequency part reflects the overall change trend of the parameters, the medium-frequency part corresponds to the fluctuations in the normal process, and the high-frequency part corresponds to noise or abnormal fluctuations. By calculating the energy proportion and entropy value characteristics of each frequency band, the dynamic change characteristics of the parameters in different frequency domains are quantified, and a time series feature vector is formed. This processing process extracts the dynamic characteristics of the process parameters over time, rather than just focusing on static numerical values.

[0032] When constructing the grey correlation degree analysis model based on the time series feature vector, the reference sequence and the comparison sequence are first determined. The reference sequence is the product quality index (paint film uniformity, adhesion, insulation performance), and the comparison sequence is the time series features of each process parameter. By calculating the grey correlation degree between the reference sequence and the comparison sequence, the influence degree of each parameter on the quality index is quantified. The calculation of the grey correlation degree includes three steps of data non-dimensionalization, correlation coefficient calculation, and correlation degree calculation. For each pair of reference sequence and comparison sequence, the correlation coefficient of all corresponding points is calculated, and then the average value is taken as the correlation degree. Finally, the correlation degree matrix is obtained, and each element of the matrix represents the influence degree of a specific process parameter on a specific quality index, and the numerical range is between 0 and 1, and the larger the value, the more significant the influence.

[0033] The correlation degree matrix is combined with expert experience and historical production data to introduce the fuzzy comprehensive evaluation method and establish the fuzzy weight vector. First, the process experts evaluate the importance of each parameter according to their experience (such as "extremely important", "very important", "generally important", etc.), and then convert the language evaluation into fuzzy membership. At the same time, based on the historical production data, the point biserial correlation coefficient of each parameter and the product yield is calculated, and the expert evaluation and data statistics results are weighted and fused to form the weight coefficient of the process parameter. Through fuzzy synthesis operation, the correlation degree matrix and the weight vector are multiplied to obtain the fuzzy relationship matrix of process parameters and product quality. The fuzzy relationship matrix represents the mapping relationship between process parameters and quality indicators under fuzzy conditions, overcoming the limitations of traditional deterministic models.

[0034] The key parameters such as annealing temperature, painting speed, paint viscosity, and baking temperature in the fuzzy relationship matrix are input into the support vector regression algorithm to establish an accurate prediction model of process parameters and quality indicators. Support vector regression first maps the input features to a high-dimensional feature space through a Gaussian kernel function. In the high-dimensional space, a linear regression function is constructed, and a non-sensitive loss function and a regularization term are introduced to balance the fitting ability and generalization ability of the model. Through cross-validation optimization of kernel parameters and regularization parameters, a high-precision parameter-quality prediction model is trained, and the determination coefficient of the model is not less than 0.92, indicating that the model can explain more than 92% of the variation of the quality index.

[0035] The time series characteristics of the process parameters in the parameter-quality prediction model are analyzed by using an autoregressive integrated moving average model. The autoregressive integrated moving average model is divided into three parts: an autoregressive part, an integral part and a moving average part. First, the stationarity of the time series data is tested. If it is not stationary, it is converted into a stationary sequence by difference operation. Then, the order of the model is determined, and the optimal order combination is selected by information criterion. The autoregressive integrated moving average model is fitted according to the determined order, which is used to predict the short-term trend of the key parameters. The prediction results are combined with the previous analysis to construct a multi-level process parameter influence model library containing static mapping relationship and dynamic prediction ability, and finally a complete mapping relationship between process parameters and product quality indicators is formed.

[0036] In the actual silk-covered winding production, the principal component analysis processes the standardized data set including more than 30 parameters such as temperature, humidity and tension, extracts 8 principal components, the first principal component is mainly composed of annealing temperature and cooling rate, and the contribution rate reaches one third, the second principal component is mainly composed of paint viscosity and coating speed, and the contribution rate is close to one fourth. After wavelet packet transform is applied to these principal components, it is found that the annealing temperature has high energy ratio in the low frequency band, indicating that the temperature change is relatively gentle; while the tension parameter has high energy ratio in the medium frequency band, showing periodic fluctuation characteristics. The gray correlation degree analysis result shows that the annealing temperature has high correlation degree with the uniformity of paint film, and the coating speed has second correlation degree with the adhesion. The fuzzy comprehensive evaluation method integrates the experience evaluation of many experienced engineers and the recent production data to determine the weight coefficients of the parameters. After the key parameters are input into the support vector regression algorithm, the trained prediction model has high prediction accuracy for the paint film thickness, and the determination coefficient meets the standard. The autoregressive integrated moving average model analysis shows that the optimal model of the annealing temperature can accurately predict the temperature change trend in the near future, providing predictive guidance for the control system, and effectively solving the problem that the traditional method cannot accurately capture the dynamic change characteristics of the process parameters.

[0037] In a specific embodiment, the process of performing step S103 can specifically include the following steps: The mapping relationship between the process parameters and the product quality indicators is modeled by a Markov decision process, the state space is defined to include the process parameter set, the equipment operating state and the semi-finished product quality characteristics, the action space is defined to include the adjustment parameters of the temperature adjustment amplitude, the tension change amount and the coating speed change amount, and a state-action environment model is obtained; Based on the state-action environment model, a reward function is constructed, the weighted sum of the product quality score and the production efficiency is calculated, a positive reward value is given to the high-quality products, a negative reward value is given to the unqualified products, and an energy consumption penalty term is introduced, and a reinforcement learning training signal is obtained; The reinforcement learning training signal is input into a deep deterministic policy gradient network, a four-layer fully connected Actor network structure is constructed, the number of input layer neurons is matched with the state dimension, two hidden layers are set to 256 and 128 nodes respectively, a ReLU activation function is configured, a Tanh activation function is used in the output layer to map to the action space, and a parameter adjustment policy network is obtained; The state-action environment model is input into a four-layer fully connected Critic network structure, the input layer receives state information, the second hidden layer fuses action information, a ReLU activation function is used, and the output layer calculates Q values to obtain a state-action value evaluation network; Based on the parameter adjustment policy network and the state-action value evaluation network, an experience replay pool with a capacity of 100000 is constructed, a batch size of 128 samples is randomly sampled each time for training, a target network update rate τ = 0.001 is set for soft updating, and an Ornstein-Uhlenbeck noise process is introduced, with an initial noise intensity of 0.5 and a decay coefficient of 0.99, to obtain an exploration-exploitation balance mechanism; A hierarchical priority constraint condition is designed for the exploration-exploitation balance mechanism, the change range of temperature and tension key parameters is limited within a safe range, a penalty function method is used to convert the constraint into a component of the reward function, and after more than 500,000 interactive samples are accumulated, a dynamic adjustment strategy for process parameters is obtained.

[0038] Specifically, the wire enamel winding production environment is abstracted as a Markov decision process. Markov decision process is a mathematical model describing sequential decision problems, including state set, action set, state transition function and reward function. In the wire enamel winding production, the state space contains three parts of information: the process parameter set includes the current values of key parameters such as annealing temperature, painting speed, paint viscosity and baking temperature; the equipment running state includes the main machine speed, motor load, paint circulation pressure and other equipment indicators; the semi-finished product quality characteristics include the wire diameter deviation, surface defect number, temporary measurement value of insulation resistance and other online detection data. The action space defines the adjustment range of key process parameters, such as annealing zone temperature adjustment range ±10℃, tension change amount ±5N, painting speed change amount ±2m / min, etc. By recording a large amount of production operation data, a state transition model is established, which describes the probability distribution of the system moving to the next state after executing a specific action in the current state, thereby forming a complete state-action environment model. Based on the state-action environment model, a reward function is constructed, and a reasonable reward mechanism is designed to guide the control strategy to develop in the optimization direction. The reward function considers both product quality score and production efficiency, and integrates them by weighted summation. The product quality score is calculated according to the uniformity of paint film, adhesion, insulation performance and other indicators. When the quality indicators meet the A-grade product standard, a high positive reward value +10 is given; when the quality indicators meet the B-grade product standard, a medium positive reward value +5 is given; when the quality indicators meet the C-grade product standard, a low positive reward value +2 is given; when unqualified products are produced, a negative reward value-15 is given, which strongly inhibits the production of unqualified products. The production efficiency part calculates the reward according to the output per unit time. For every 1% increase in the ratio of output to standard capacity, the reward value increases by +0.5. At the same time, an energy consumption penalty term is introduced. For every 1% increase in the ratio of energy consumption to standard consumption, the reward value decreases by-0.3, promoting energy-saving production. Through this multi-dimensional reward design, a reinforcement learning training signal is generated to guide the control strategy to find the best balance point among quality, efficiency and energy consumption.

[0039] The reinforcement learning training signal is input into the deep deterministic policy gradient network to build a neural network structure suitable for continuous control problems. The deep deterministic policy gradient network is an algorithm combining deep learning and reinforcement learning, which is particularly suitable for handling control problems in continuous action space. Its core is the Actor-Critic architecture, where the Actor is responsible for generating action strategies, and the Critic is responsible for evaluating action values. A four-layer fully connected Actor network structure is specifically constructed, with the number of input layer neurons matching the state dimension. For a silk package varnishing winding production environment containing 30 state variables, 30 input neurons are set; the hidden layers are set to 256 and 128 nodes respectively, forming a funnel-shaped structure to extract state features layer by layer; all hidden layers are configured with ReLU activation functions to effectively alleviate the gradient vanishing problem; the output layer corresponds to the action space dimension, using the Tanh activation function to map the output to the [-1, 1] interval, and then through linear transformation to map to the actual action range, such as mapping the temperature adjustment range from [-1, 1] to [-10℃, +10℃], to obtain the parameter adjustment strategy network.

[0040] The state-action environment model is input into the Critic network structure to build an action value evaluation module. The Critic network also adopts a four-layer fully connected structure, with the input layer receiving state information, the size being the same as that of the Actor network input layer; the first hidden layer contains 256 nodes to process state features; the second hidden layer is special in that it fuses action information, concatenating the action vector output by the Actor network with the first hidden layer output and inputting them together into the second hidden layer, which is set to 128 nodes; ReLU activation function is also used to enhance non-linear expression ability; the output layer has only one neuron without using activation function, directly outputting the Q value estimate corresponding to the state-action pair, i.e. the long-term cumulative reward expectation of executing a specific action in the current state, thereby obtaining the state-action value evaluation network. Through this structure design, the Critic network can evaluate the action value generated by the Actor network to guide the optimization direction of the Actor strategy.

[0041] An experience replay mechanism is constructed based on the parameter adjustment strategy network and the state-action value evaluation network to improve learning efficiency and stability. Experience replay is an important technique in reinforcement learning, which stores and reuses historical experience samples to break the temporal correlation between samples and improve learning efficiency. The specific implementation includes: constructing an experience replay pool with a capacity of 100,000 to store state transition quadruples (current state, executed action, reward obtained, next state); randomly sampling a batch of 128 samples from the replay pool each time to reduce sample correlation; setting the target network update rate τ = 0.001 and using a soft update strategy to slowly update the target network parameters to avoid unstable training; introducing an Ornstein-Uhlenbeck noise process to generate time-dependent exploration noise, with an initial noise intensity of 0.5 and a decay factor of 0.99 to gradually reduce the exploration level as training progresses, forming a smooth transition from exploration to utilization and achieving a complete exploration-exploitation balance mechanism.

[0042] A hierarchical priority constraint is designed for the exploration-exploitation balance mechanism to ensure the safety and rationality of control operations. In silk-covered winding production, excessive adjustment of some process parameters can lead to product quality problems or even equipment damage. A hierarchical constraint design is adopted: the highest priority constraint is the equipment safety limit, such as a temperature change rate of no more than 5℃ / min and a tension change of no more than 2N / s; the next highest priority constraint is the product quality guarantee, such as maintaining the annealing temperature within the process window of 220±20℃; the lower priority constraint is energy efficiency optimization, such as controlling the total energy consumption to be no more than 110% of the standard value. The penalty function method is used to convert the constraints into a component of the reward function, introducing a large negative reward when the control action violates the constraints, with the degree proportional to the severity of the violation. After 500,000 interaction samples of training, the network gradually learns the control strategy that meets the constraints of each layer, and finally obtains a process parameter dynamic adjustment strategy that can optimize product quality and production efficiency while ensuring equipment safety and energy saving.

[0043] In practical application, the method is applied in the production process of a certain specification of enameled wire. First, the production environment is modeled as a Markov decision process with a state space containing 32 variables and an action space containing 4 key parameter adjustment amounts. In the reward function design, +10 reward is given for A-level products, -15 punishment is given for unqualified products, 0.5 reward is added for each 1% increase in production efficiency, and 0.3 reward is reduced for each 1% increase in energy consumption. During the training process, the initial stage explores a rich strategy, and the temperature adjustment range is large, from -8℃ to +8℃. As the training progresses, the network gradually learns the rule that temperature mutation has a significant impact on the uniformity of the paint film, and the adjustment strategy gradually becomes smooth, eventually forming a stable control mode with an average adjustment range of ±2.5℃. When detecting that the paint viscosity is rising, the network automatically adopts a coordinated adjustment strategy of reducing the coating speed and slightly increasing the baking temperature, rather than simply adjusting a single parameter. This dynamic and coordinated adjustment capability is difficult to achieve with traditional control methods, and it effectively solves the technical problems of complex coupling between parameters in enameled winding production and the lack of intelligent decision support for process parameter adjustment.

[0044] In a specific embodiment, the process of performing step S104 can specifically include the following steps: A multi-objective optimization model is constructed based on the dynamic adjustment strategy of the process parameters, and a set of objective functions is defined, including a product quality index function, a production efficiency function, and an energy consumption function. The product quality index function is composed of the weighted sum of insulation strength and paint film uniformity index, the production efficiency function is composed of the effective output per unit time, and the energy consumption function is composed of the standardized sum of electrical energy and thermal energy. The set of optimization objective functions is obtained. An improved non-dominated sorting genetic algorithm is applied to the set of optimization objective functions, with a population size of 100, a maximum number of iterations of 200, a crossover probability of 0.85, and a mutation probability of 0.15. The non-dominated sorting and crowding distance calculation mechanism is used to maintain population diversity, and a set of Pareto optimal solutions for process parameter configuration is obtained. A heuristic operator based on problem characteristics is introduced to the set of Pareto optimal solutions, and intelligent crossover operators considering process parameter continuity and parameter sensitivity-oriented adaptive mutation operators are applied to accelerate algorithm convergence, and a variety of feasible process parameter configuration schemes are obtained. The process parameter configuration scheme is input into the analytic hierarchy process decision support module, the weight coefficients of each objective function are dynamically adjusted according to the current production target, and the final process parameter optimization scheme is formed to obtain the optimized control sequence. Based on the optimized control sequence, a model predictive control strategy is applied, with a prediction horizon of 15, the system's current state is used to predict the system's behavior at multiple future time steps, and the optimal control sequence is determined by solving a rolling optimization problem to obtain the model predictive control action. Recursive least square method is introduced to update parameters of the model predictive control action online, and a multi-level safety guarantee mechanism is configured, including parameter change rate limitation, safety boundary check and emergency intervention logic, to ensure safety of the control action and obtain a multi-objective optimization control scheme balancing product indicators, production efficiency and energy consumption.

[0045] Specifically, in the production of silk-covered lacquered winding, the objective function set includes three key dimensions: product quality indicator function, production efficiency function and energy consumption function. The product quality indicator function is composed of two main quality parameters: insulation strength and film uniformity. The insulation strength represents the electrical performance of the enameled wire under the action of electric field, which is obtained by breakdown voltage test. The film uniformity is calculated by film thickness standard deviation. These two indicators are given different weights according to different requirements of product types, such as high-frequency transformer wire which emphasizes insulation strength, and motor winding wire which emphasizes film uniformity. The production efficiency function is directly calculated by the effective output per unit time, i.e. the length or weight of qualified enameled wire produced per hour. The energy consumption function integrates the consumption of electric energy and thermal energy, converts different energy units into unified energy equivalent values through standardization coefficients, and represents the energy consumption per unit product. Through the definition of the three objective functions, a complete optimization objective function set is formed, providing clear evaluation criteria for subsequent multi-objective optimization. The improved non-dominated sorting genetic algorithm is applied to the optimization objective function set. The non-dominated sorting genetic algorithm is an evolutionary algorithm specifically designed to solve multi-objective optimization problems, which can find a set of approximate Pareto optimal solutions in a single run. The algorithm first sets the population size to 100, i.e. maintains 100 candidate solutions simultaneously, each containing a complete configuration scheme of process parameters such as annealing temperature and coating speed. The maximum number of iterations is set to 200, the crossover probability is 0.85, and the mutation probability is 0.15. These parameter settings balance the exploration ability and convergence speed of the algorithm. The core of the algorithm lies in two special mechanisms: non-dominated sorting and crowding distance calculation. Non-dominated sorting divides the individuals in the population into layers according to the dominance relationship. If solution A is not worse than solution B in any objective and at least better than B in one objective, then A dominates B. The first layer is the set of non-dominated solutions, i.e. the set of solutions in the population that are not dominated by any other solution. The second layer is the non-dominated solution after removing the first layer, and so on, forming a complete hierarchical structure of dominance. Crowding distance measures the distribution density of individuals in the solution space. By calculating the sum of the standardized distances between adjacent solutions in each objective function, the uniqueness of each solution is evaluated. The selection operation preferentially selects individuals at low levels, and preferentially selects individuals with large crowding distance when selecting from the same layer, ensuring that the population not only converges to the optimal solution set, but also maintains the diversity of solutions. After 200 iterations, a set of Pareto optimal solutions of process parameter configurations is obtained, which represent the best process parameter combinations under different quality, efficiency and energy consumption trade-offs.

[0046] Heuristic operators based on problem characteristics are introduced to the Pareto optimal solution set, and special genetic operators are designed according to the characteristics of wire enamelled winding production. Firstly, intelligent crossover operator is considered for the continuity of process parameters. The traditional crossover operation of genetic algorithm often leads to parameter mutation, which is unrealistic in wire enamelled wire production because temperature and other parameters cannot change greatly instantaneously. The intelligent crossover operator sets different change rate limits for different parameters by analyzing the physical characteristics and response time of the parameters, such as limiting the change rate of temperature parameters to 5℃ / min, to ensure that the parameter scheme of the offspring after crossover meets the process continuity requirements. Secondly, the adaptive mutation operator is guided by parameter sensitivity. It dynamically adjusts the mutation step size according to the sensitivity of each parameter to the objective function. Parameters with high sensitivity, such as annealing temperature, use smaller mutation step size, while parameters with low sensitivity, such as transport speed, allow larger mutation range. This method avoids excessive disturbance to key parameters while maintaining sufficient exploration. Through the application of these two special operators, the optimization algorithm is more in line with the actual production characteristics, accelerating the convergence to the solution space with practical significance and obtaining multiple feasible process parameter configuration schemes.

[0047] The process parameter configuration scheme is input into the analytic hierarchy process decision support module for final scheme selection and optimization. Analytic hierarchy process is a structured multi-criteria decision-making method that decomposes complex problems into hierarchical structures. In wire enamelled winding production, a three-level hierarchical structure is constructed: the top layer is the overall goal "optimal process parameter configuration", the middle layer is three criteria "product quality", "production efficiency" and "energy consumption", and the bottom layer is multiple alternative parameter configuration schemes. For different production task types, the weight coefficients of each objective function are dynamically adjusted. For example, when producing high-end products, the quality weight is increased to 0.6, the efficiency weight is reduced to 0.3, and the energy consumption weight is set to 0.1; when dealing with urgent orders, the efficiency weight is increased to 0.7, the quality weight is reduced to 0.2, and the energy consumption weight is set to 0.1. Through expert judgment and pair-wise comparison method, the judgment matrix is constructed, the weight of each level index is calculated, and the process parameter optimization scheme with the highest comprehensive score is finally formed. This scheme is converted into a time series form to obtain the optimized control sequence of process parameters over time.

[0048] A model predictive control strategy is applied based on the optimized control sequence to achieve rolling optimization control. Model predictive control is an advanced control method based on model prediction and rolling optimization, which is suitable for multivariable and complex constraint industrial processes. The prediction horizon is set to 15, i.e., the system behavior in the next 15 time steps is predicted, and the prediction time step is 5 seconds, covering a total of 75 seconds of future state. In each control cycle, the change trajectory of each state variable in the next 15 time steps is predicted using the aforementioned trained parameter-quality prediction model based on the current measured state (temperature, pressure, etc.). Taking the current state as the starting point and the optimized control sequence as the initial value of the control input, the control sequence that optimizes the objective function is found by solving the rolling optimization problem. The optimization process takes into account the physical constraints of the actuators and the limitations of the process parameters, such as the temperature change rate not exceeding the manufacturer's specified value. Finally, only the first control action of the optimized sequence is executed, and when the next control cycle arrives, the above process is repeated based on the new measurement value to obtain the model predictive control action. This rolling optimization method effectively deals with system disturbances and model uncertainties.

[0049] Recursive least squares method is introduced to update the parameters of the model predictive control action online, and a multi-level safety guarantee mechanism is configured. Recursive least squares method is an online parameter estimation method that can incrementally update model parameters based on new observation data without storing and repeatedly processing historical data. In the wire coating paint coating line production process, some dynamic characteristics such as heat conduction coefficient will change over time. Recursive least squares method calculates the error and updates the model parameters to keep the control model consistent with the actual system. The safety guarantee mechanism includes three levels: parameter change rate limitation ensures that the control instruction meets the response capability of the physical device, such as the temperature change rate not exceeding 8℃ / min; safety boundary check ensures that the process parameters are within the effective range, such as tension control between 5-25N; emergency intervention logic handles abnormal situations, automatically switches to safety mode when sensor failure or parameter fluctuation is detected, and maintains basic parameter stability. Through the comprehensive application of these mechanisms, a multi-objective optimization control scheme is finally obtained that can optimize production indicators and ensure operation safety.

[0050] In a specific embodiment, the process of introducing a problem-specific heuristic operator step to the Pareto optimal solution set can specifically include the following steps: Performing historical data correlation analysis on the process parameters in the Pareto optimal solution set to establish a correlation matrix between the parameters and identify strong coupling relationships between the parameters to obtain a parameter dependency network; Designing an intelligent crossover operator based on the parameter dependency network to implement collaborative crossover operation on strongly correlated parameters to maintain the consistency and continuity of parameter changes and obtain child individuals that meet process constraints; A parameter sensitivity coefficient of each process parameter in the offspring individual is calculated, the influence degree of each parameter on the objective function is quantified by performing partial derivative analysis on historical process data, and a parameter sensitivity ranking table is obtained; An adaptive mutation operator is constructed according to the parameter sensitivity ranking table, a smaller mutation step is set for a parameter with high sensitivity, a larger mutation step is set for a parameter with low sensitivity, and an adaptive mutation strategy is obtained; A simulated annealing mechanism is applied to the adaptive mutation strategy, an initial temperature is set to 100, a cooling coefficient is set to 0.95, the mutation probability is dynamically adjusted according to the algorithm iteration process, a higher exploratory property is maintained in the early stage, and gradually converges to a local fine search in the later stage, and an optimized search trajectory is obtained; The elite solution in the optimized search trajectory is reserved to the next generation, the solutions with high repetition in the population are subjected to diversity enhancement processing, chaotic disturbance is introduced to avoid premature convergence of the population, and after 200 iterations, a diversified process parameter configuration scheme that meets the process continuity constraint is obtained.

[0051] Specifically, to analyze the historical data correlation of the process parameters in the Pareto optimal solution set, a large amount of historical data in the production process of wire-coated enameled winding needs to be collected. The continuous production data in the past three months are extracted from the production database, including the time series records of key process parameters such as annealing temperature, painting speed, paint viscosity, baking temperature, and the corresponding production status and product quality data. Align these historical data by time to form a multi-dimensional time series data matrix, each row represents a time point, and each column represents a process parameter. By calculating the Pearson correlation coefficient or mutual information value between parameters, the linear or nonlinear correlation degree between parameters is quantified. For parameters such as annealing temperature and baking temperature, the correlation coefficient is high, indicating that the change trend of the two is similar; while for parameters such as annealing temperature and painting speed, the mutual information value is high but the correlation coefficient is not significant, indicating that there is a nonlinear correlation. Organize the correlation calculation results between all parameters into a correlation matrix, and each element in the matrix represents the correlation strength between the corresponding row and column parameters. Apply hierarchical clustering or community detection algorithms to the correlation matrix to group strongly correlated parameters into the same class and form parameter groups. Finally, construct a parameter dependency relationship network graph based on the clustering results, where nodes represent parameters, edges represent correlation strength, and edge thickness or color depth reflects correlation degree, directly showing the coupling structure between process parameters. Based on the parameter dependency relationship network, an intelligent crossover operator is designed to improve the traditional genetic algorithm with simple random crossover. The traditional crossover operator handles each parameter independently, ignoring the mutual influence between parameters, which does not conform to the physical properties of wire-coated enameled winding production. The intelligent crossover operator first identifies strongly correlated parameter groups based on the parameter dependency relationship network, such as annealing temperature, holding time, and cooling rate, which are usually grouped together, and painting speed and paint viscosity, which are grouped together. During the crossover operation, parameters within the same parameter group are subjected to coordinated crossover, i.e., these parameters are processed in the same crossover point or similar crossover manner to maintain the coordination of parameter changes. In specific implementation, an adaptive segmented crossover method is used to set change rate constraints based on the physical properties and change inertia of parameters, such as temperature parameters that cannot fluctuate significantly at adjacent time points. At the same time, considering the response time differences of different parameters, a time lag factor is introduced to ensure that the parameter combination after crossover is physically feasible. Through this coordinated crossover operation, the generated offspring not only inherits the excellent characteristics of the parents, but also meets the constraints of process continuity and parameter coordination, avoiding unreasonable parameter combinations produced by traditional crossover.

[0052] The parameter sensitivity coefficient of each process parameter in the offspring individual is calculated, and the influence degree of parameter change on the objective function is quantitatively evaluated. The parameter sensitivity is an important index in the optimization process, which indicates which parameters have the greatest influence on the objective function and should be focused on. According to the characteristics of silk-covered winding production, the sensitivity is calculated by partial derivative analysis of historical process data. First, enough process parameter-quality target pairs are extracted from historical data to build a regression model. Then, a small perturbation is made to each process parameter, and the change rate of the objective function is observed. The partial derivative value is calculated by difference approximation. To improve the calculation accuracy, the central difference method is used instead of the forward difference method, that is, the influence of parameter increase and decrease is considered at the same time, and the average change rate is taken. The absolute value of the calculated partial derivative is the parameter sensitivity coefficient, which reflects the influence degree of parameter change on the objective function. The sensitivity coefficient is further standardized so that the sum is 1 or the maximum value is 1, which is convenient for comparing the relative importance of different parameters. Finally, the parameter sensitivity ranking table is generated, and each process parameter is ranked from high to low according to the sensitivity, which provides a basis for subsequent variation operation.

[0053] According to the parameter sensitivity ranking table, an adaptive mutation operator is constructed, and different mutation strategies are adopted for parameters with different sensitivities. In traditional genetic algorithms, the same mutation probability and mutation amplitude are used for all genes in the mutation operation, ignoring the characteristic differences of different parameters. The adaptive mutation operator sets a smaller mutation step for parameters with high sensitivity to avoid excessive negative impact on the objective function, and sets a larger mutation step for parameters with low sensitivity to enhance the exploration ability of the algorithm. In specific implementation, a function relationship is designed between the mutation step and the sensitivity, for example, for the annealing temperature with the highest sensitivity, the mutation step is limited to ±2℃; while for the transport speed with lower sensitivity, the mutation amplitude can reach ±10m / min. Considering the differences in physical units and change ranges of parameters, normalization processing is introduced to make the mutation of different parameters comparable in relative scale. During the mutation operation, whether to mutate a parameter is determined according to the set mutation probability. If mutation occurs, a new value is randomly generated near the current value according to the corresponding step, and it is ensured that the new value does not exceed the effective range of the parameter. This method of adjusting mutation characteristics according to sensitivity not only ensures the stability of important parameters, but also enhances the exploration ability of secondary parameters. The adaptive mutation strategy is applied to the simulated annealing mechanism, and the temperature concept is introduced to control the search behavior of the algorithm. Simulated annealing is a heuristic optimization algorithm that simulates the process of a physical annealing system gradually cooling from high temperature to equilibrium state. By introducing the simulated annealing mechanism into the genetic algorithm, the mutation probability and mutation amplitude are controlled through the temperature parameter, realizing the smooth transition from global exploration to local fine search. The initial algorithm temperature is set to 100, indicating the high exploration stage; the cooling coefficient is set to 0.95 to control the temperature drop rate. During the algorithm iteration process, the temperature decreases according to the rule T(k+1)= 0.95 × T(k), where k represents the current generation number. The mutation probability is positively correlated with the temperature, and the mutation probability is larger in the early high-temperature stage and gradually decreases as the temperature decreases. The mutation step also changes with the temperature, allowing large amplitude mutation at high temperature and narrowing the mutation range at low temperature. This dynamic adjustment mechanism enables the algorithm to maintain a high exploration in the early stage, which helps to escape from local optimum; and gradually converges to local fine search in the later stage, improving the accuracy of the solution. By recording the parameter combination and fitness value of the optimal individual in each generation, an optimization search trajectory is formed, which intuitively shows the convergence process of the algorithm.

[0054] The elite solution in the optimization search trajectory is reserved to the next generation, and the solution with high repetition in the population is treated with diversity enhancement. The elite preservation strategy is an important mechanism to ensure that optimization does not regress, that is, the optimal individual or the top individuals in each generation are directly copied to the next generation and are not affected by crossover and mutation. In the silk package varnish winding process optimization, the solution of the first layer of non-dominated sorting in each generation is selected as the elite solution to ensure that the front solution of multi-objective optimization is not lost. For solutions with high similarity in the population, the Euclidean distance or Manhattan distance is calculated to identify solutions close to each other. When the parameter distance between two solutions is less than the preset threshold, they are considered too similar, and one of them is reserved and the rest are replaced by diversity enhancement. Diversity enhancement uses chaotic disturbance method to produce a seemingly random but deterministic sequence using chaotic mapping such as Logistic mapping for controlled disturbance of parameters. The introduction of chaotic disturbance avoids premature convergence of the population to local optimum and enhances the global search ability of the algorithm. After 200 iterations, the population contains both high-quality elite solutions and sufficient diversity, and a set of diversified and process continuity constraint satisfying process parameter configuration schemes are finally selected, providing production managers with multiple optional strategies.

[0055] In the production of wire enamel wire for a certain type of motor, when the process parameters of the Pareto optimal solution set are analyzed for historical data correlation, nearly 3 months of production data are collected, including 10 key process parameters such as annealing temperature, coating speed, and paint viscosity. By calculating the correlation coefficient matrix, it is found that the correlation coefficient between annealing temperature and cooling rate is as high as 0.83, indicating that the two are highly correlated; while the correlation coefficient between annealing temperature and coating speed is low, only 0.21, but the mutual information value is high, indicating that there is a nonlinear correlation. Applying community discovery algorithm, 10 parameters are divided into 3 main parameter groups: temperature group (annealing temperature, baking temperature, cooling rate), coating group (coating speed, paint viscosity, tension) and mechanical group (wire speed, take-up tension, guide roller pressure). Based on this parameter dependence network, when designing the intelligent crossover operator, the parameters in the temperature group are adopted in the same crossover mode, avoiding physically unreasonable parameter combinations, such as the contradictory situation of increasing annealing temperature and decreasing baking temperature. Through partial derivative analysis to calculate the parameter sensitivity, it is found that the sensitivity coefficient of annealing temperature is 0.32, much higher than that of other parameters, the coating speed is 0.25, and the wire speed is the lowest, only 0.05. Accordingly, the adaptive mutation operator is constructed, and the mutation step length of annealing temperature is limited to 30% of the standard step length, while the mutation step length of wire speed is expanded to 150% of the standard step length. After applying the simulated annealing mechanism, the mutation probability is set to 0.3 in the first generation iteration, and the mutation amplitude is large; while to the 150th generation, the temperature drops to 100×0.95^150≈0.5, the mutation probability drops to 0.08, and the mutation amplitude is reduced to 5% of the initial value, focusing on local fine search. Through the elite preservation and chaos diversity enhancement strategy, the obtained parameter configuration scheme contains not only high-quality high-energy fine schemes, but also medium-quality low-energy economic schemes, effectively solving the problems of traditional process parameter optimization that is difficult to balance multiple objectives and the complexity of parameter coupling relationship.

[0056] In a specific embodiment, the process of performing the step of inputting the process parameter configuration scheme into the analytic hierarchy process decision support module can specifically include the following steps: A three-level analytic hierarchy structure is constructed for the multiple feasible process parameter configuration schemes, the first level is the total target layer, the second level is the criterion layer containing product quality, production efficiency and energy consumption three indexes, and the third level is the scheme layer containing multiple parameter configuration schemes, and the analytic hierarchy judgment matrix is obtained; Based on the analytic hierarchy judgment matrix, pairwise comparison is carried out, 1-9 scale method is used to construct the relative importance comparison matrix between the indexes of the criterion layer, and the maximum eigenvalue and its corresponding eigenvector are calculated by eigenvalue method, to obtain the weight coefficient of the criterion layer; Performing consistency check on the criterion layer weight coefficient, calculating consistency index CI and random consistency ratio CR, when CR is less than 0.1, it is considered that the judgment matrix has satisfactory consistency, otherwise, the judgment matrix is adjusted to obtain the consistency check result of the criterion layer index weight; According to the consistency check result, the relative weight of each index in the criterion layer is adjusted, the weight coefficient is automatically increased or decreased through fuzzy rules combined with the current production task demand, the production efficiency weight is increased in emergency order, the quality index weight is increased in high-end product production, and a dynamic weight distribution scheme is obtained; The dynamic weight distribution scheme is applied to the comprehensive evaluation of each process parameter configuration scheme, the weighted total score of each scheme under each objective function is calculated, and the configuration schemes are sorted according to the total score, and the configuration scheme with the highest comprehensive score is selected to obtain the optimized process parameter combination; The optimized process parameter combination is unfolded in time sequence, the static process parameters are converted into time sequence parameter adjustment strategy according to the dynamic characteristics of the production process, the optimal parameter adjustment trajectory containing future 15 time steps is formed, and the optimized control sequence is obtained.

[0057] Specifically, a three-level analytic hierarchy structure is constructed for various feasible process parameter configuration schemes. Firstly, the basic principle of analytic hierarchy process (AHP) is determined. AHP is a multi-criteria decision-making method that decomposes complex decision problems into a hierarchical structure. By decomposing the decision problem into three levels of objectives, criteria and schemes, the complex problem is organized. In the field of wire enamel winding process optimization, the three-level analytic hierarchy structure is constructed. The first level is the overall objective layer, which is determined as "optimal process parameter configuration scheme". The second level is the criteria layer, which includes product quality, production efficiency and energy consumption. These three indicators represent different optimization directions. The third level is the scheme layer, which includes multiple parameter configuration schemes selected from the Pareto optimal solution set. Usually, 20-30 representative schemes are selected into the evaluation system. Through this hierarchical division, the optimization decision-making process is divided into evaluating the performance of each scheme under three criteria, and then the importance of each criterion is integrated to make the final decision. After the hierarchical structure is established, a judgment matrix is constructed to reflect the relative importance of each element. Specifically, it includes the relative importance matrix of each index in the criteria layer and the relative performance matrix of each scheme in the scheme layer under each criterion, forming a complete set of AHP judgment matrices. Based on the AHP judgment matrix, pairwise comparison is carried out, and the 1-9 scale method is used to quantify subjective judgment. The 1-9 scale method is a commonly used quantification method in AHP, where 1 represents equal importance of two elements, 3 represents slightly more important, 5 represents significantly important, 7 represents strongly important, 9 represents extremely important, and 2, 4, 6, 8 are intermediate values for adjacent judgments. In the optimization of wire enamel winding process, according to the production strategy and market demand of the enterprise, the three indicators in the criteria layer are compared pairwise. For example, when comparing quality and efficiency, if quality is considered significantly more important than efficiency, the corresponding matrix element is assigned a value of 5. Correspondingly, the importance of efficiency compared to quality is 1 / 5. Through this way, pairwise comparison of all indicators is completed, forming the criteria layer judgment matrix. Subsequently, the eigenvalue method is used to process the judgment matrix, and the maximum eigenvalue and its corresponding eigenvector are calculated. In the calculation, the characteristic equation |A-λI|=0 is first solved to find the maximum eigenvalue λmax, and then the linear equation system (A-λmaxI)X=0 is solved to obtain the corresponding eigenvector. The eigenvector is normalized so that the sum of the components is 1. The normalized eigenvector components are the weight coefficients of each criterion, which directly reflect the relative importance of each indicator in decision-making.

[0058] The consistency check is performed on the weight coefficients of the criteria layer to ensure the rationality and consistency of the judgment. The consistency check is a key step in the AHP method, which is used to check whether there is a logical contradiction in the judgment matrix. The consistency index CI=(λmax-n) / (n-1) is calculated, where λmax is the maximum eigenvalue of the judgment matrix, and n is the order of the matrix, i.e. the number of criteria. The closer the CI value is to 0, the better the consistency of the judgment matrix. Considering that the subjective judgment of people is not necessarily absolutely consistent, the random consistency ratio CR=CI / RI is introduced, where RI is the random consistency index, which is the average consistency index of the same order random matrix, and the RI value of the 3-order matrix is 0.58. When CR is less than 0.1, the judgment matrix is considered to have satisfactory consistency, and the weight calculation result can be accepted; otherwise, the judgment matrix needs to be adjusted, and the unreasonable judgment values need to be reviewed and corrected until the consistency check is passed. This checking process ensures the rationality of the final weight distribution and the reliability of the decision-making, and provides a basis for the next weight adjustment.

[0059] According to the consistency check result, the relative weight of each index in the criteria layer is adjusted, and the decision weight is dynamically optimized combined with the specific needs of the current production task. The weight of the traditional AHP method is fixed once it is determined and cannot adapt to the changing needs of different production scenarios. To solve this problem, a fuzzy rule automatic adjustment mechanism is introduced to dynamically adjust the weight according to the production task type. Fuzzy rules are a knowledge representation form based on IF-THEN statements, such as "IF order urgency is high THEN increase production efficiency weight". In the implementation process, first, the fuzzy evaluation of the current production task characteristics is performed to determine the attribute values of the task such as urgency, quality requirement and cost sensitivity; then these attribute values are input into the preset fuzzy rule base to trigger the corresponding adjustment rules; finally, the basic weight is adjusted according to the rule execution result. For example, in the case of urgent orders, the production efficiency weight is automatically increased by 25%-50%; in the case of high-end product production, the quality index weight is increased by at least 30%; and in the case of cost-sensitive orders, the weight of energy consumption index is increased accordingly. Through this dynamic adjustment mechanism, the AHP decision-making process can flexibly adapt to the changing needs of different production scenarios, and form a more targeted dynamic weight distribution scheme.

[0060] The dynamic weight allocation scheme is applied to the comprehensive evaluation of each process parameter configuration scheme for final decision selection. For each process parameter configuration scheme at the scheme level, the performance under the three criteria needs to be evaluated first. The product quality score is based on the predicted values of quality indicators such as insulation strength and paint film uniformity of the expected product under the parameter scheme; the production efficiency score is based on the expected output per unit time; and the energy consumption score is based on the expected total consumption of electric and thermal energy. These raw score data usually have different dimensions and ranges, and need to be converted to a unified score scale through standardization processing, such as using the maximum and minimum standardization method to map each score to the 0-100 point interval. After standardization, the aforementioned dynamic weight allocation scheme is applied to calculate the weighted total score of each configuration scheme: total score = quality score x quality weight + efficiency score x efficiency weight + energy consumption score x energy consumption weight. According to the weighted total score, all schemes are ranked, and the scheme with the highest score is selected as the final selection, obtaining the optimized process parameter combination. This multi-criteria comprehensive evaluation method takes into account the balance of quality, efficiency and energy consumption, providing a comprehensive basis for production decision-making.

[0061] The optimized process parameter combination is unfolded in time series to convert static parameters into dynamic control strategies. The actual production process is dynamic, and parameters cannot be set instantaneously, so the device response characteristics and process transition requirements need to be considered. Time series unfolding is to convert a single parameter set point into a parameter trajectory evolving over time, which includes the following steps: first, determine the time discretization step, usually 5-10 seconds, and set the prediction time domain to 15 time steps, covering 75-150 seconds of control process in the future; then, consider the physical characteristics of each parameter and the device response delay, and apply the corresponding dynamic response model to the process parameters, such as the first or second order lag model for temperature parameters, to describe the gradual change from the current value to the target value; then, consider the coupling relationship and sequence requirements between parameters, and coordinate the change trajectories of multiple parameters to ensure the continuity and safety of the process flow; finally, combine the predicted changes and disturbance compensation strategies of the production conditions to fine-tune and optimize the trajectory, forming the final parameter adjustment trajectory. This time series unfolding method converts static optimization results into actual executable dynamic control instructions, laying the foundation for subsequent control execution.

[0062] The method is applied to the production process of silk-covered lacquered wire in a factory. After obtaining 25 Pareto optimal solutions from the non-dominated sorting genetic algorithm, a three-level hierarchy is constructed for decision-making. First, the product quality, production efficiency and energy consumption are evaluated by 1-9 scale evaluation in the criterion layer. Based on the high-quality strategic positioning of the enterprise, the importance of product quality relative to production efficiency is determined as 5, the importance of product quality relative to energy consumption is determined as 7, and the importance of production efficiency relative to energy consumption is determined as 3, forming a judgment matrix. The initial weights of the three indexes are calculated by the eigenvalue method as quality 0.73, efficiency 0.19, and energy consumption 0.08, and the consistency ratio CR is 0.032<0.1, which passes the consistency test. When an order for high-precision motor winding wire is received, due to the extremely high requirement for insulation performance, the system automatically triggers the quality priority rule, further increasing the quality weight to 0.82, and adjusting the efficiency and energy consumption weights to 0.13 and 0.05, respectively. After standardizing the performance of the 25 schemes under the three criteria, the standardized scores of scheme A under the quality, efficiency and energy consumption indexes are 95, 75 and 60 respectively, and the total score is 90.75; the scores of scheme B are 88, 85 and 80 respectively, and the total score is 87.56; the scores of scheme C are 78, 92 and 88 respectively, and the total score is 80.64. Finally, the scheme A with the highest score is selected as the optimization scheme. The process parameters of scheme A (annealing temperature 245℃, coating speed 18m / min, etc.) are expanded in time series, considering the thermal inertia of the annealing furnace, the temperature change from the current 230℃ to the target value 245℃ is calculated according to a first-order response model, and gradually approaches the target value within 15 time steps, while the coating speed is adjusted according to the temperature change to ensure the stability of the paint film quality, and finally a detailed 15-step process parameter adjustment trajectory is formed, effectively solving the technical problems of lack of scientific decision-making basis for process parameter configuration in traditional silk-covered lacquered winding production and difficulty in directly applying static optimization results to dynamic production process.

[0063] The above describes the silk-covered lacquered winding process optimization control method based on intelligent analysis in the embodiments of the present application. The silk-covered lacquered winding process optimization control system based on intelligent analysis in the embodiments of the present application is described below. Please refer to Figure 2 An embodiment of the silk-covered lacquered winding process optimization control system based on intelligent analysis in the embodiments of the present application includes: The acquisition module 201 is used for real-time acquisition and preprocessing of temperature, humidity, tension, speed and current parameters in multiple regions of the production line, to obtain a standardized process parameter data set; The dimension reduction module 202 is used for dimension reduction processing of high-dimensional data based on the process parameter data set through principal component analysis, and constructing a process parameter influence model combining gray correlation degree analysis and support vector regression algorithm, to obtain the mapping relationship between the process parameters and the product quality indexes. a training module 203, configured to input the mapping relationship into a deep deterministic policy gradient network, train an Actor-Critic network structure by setting a reward function and an experience replay mechanism, and obtain a dynamic adjustment strategy of the process parameters; an execution module 204, configured to perform non-dominated sorting genetic algorithm optimization on the dynamic adjustment strategy, combine an analytic hierarchy process to dynamically adjust a target function weight, execute an optimal parameter adjustment sequence through model predictive control, and obtain a multi-objective optimization control scheme balancing product indexes, production efficiency, and energy consumption.

[0064] Through the cooperation of the above-mentioned components, through the real-time collection and preprocessing of the temperature, humidity, tension, speed and current parameters of the production line in multiple regions, a comprehensive process parameter data set is established, which lays a solid data foundation for subsequent intelligent analysis; based on the process parameter data set, principal component analysis is applied to reduce the dimension of high-dimensional data, effectively eliminating data redundancy, and extracting key features that significantly affect product quality; the process parameter influence model constructed by combining grey correlation analysis and support vector regression algorithm accurately quantifies the influence degree and mapping relationship of each process parameter on product quality indicators, overcoming the limitations of traditional methods in dealing with nonlinear relationships between parameters, and the kernel function mapping mechanism of the support vector regression algorithm enables the model to capture complex nonlinear relationships, with a high-precision prediction ability of a determination coefficient R² not less than 0.92, providing a reliable basis for subsequent optimization; the mapping relationship is input into the deep deterministic policy gradient network, which fully utilizes the advantages of deep learning in continuous action space, and through the design of a reasonable reward function and an experience replay mechanism, the efficient training of the Actor-Critic network structure is realized, solving the limitations of traditional reinforcement learning in continuous control problems, and the experience replay mechanism breaks the temporal correlation between samples, greatly improving the learning efficiency, while the deterministic policy gradient characteristics of the DDPG algorithm make it particularly suitable for handling precise parameter regulation problems in silk-covered wire production, and through 500,000 interaction samples learning, the network obtains a highly intelligent process parameter dynamic adjustment strategy; the non-dominated sorting genetic algorithm optimization is performed on the dynamic adjustment strategy, which fully utilizes the outstanding advantages of the NSGA-II algorithm in handling multi-objective optimization problems, and the non-dominated sorting and congestion degree calculation mechanism ensures the diversity and uniform distribution of the solution set, and the parameter sensitivity-oriented adaptive mutation operator and the intelligent crossover operator considering the continuity of the process parameters further improve the convergence speed and solution quality of the algorithm; combined with the analytic hierarchy process to dynamically adjust the objective function weight, the product quality, production efficiency and energy consumption are flexibly balanced according to different production demands, realizing the intelligent decision-making of the production process; through model predictive control to execute the optimal parameter adjustment sequence, the static optimization result is converted into a dynamic control strategy, and the prediction time domain design and rolling optimization mechanism effectively cope with system disturbances and model uncertainties. The present application organically combines various artificial intelligence algorithms, fully considers the matching of algorithm characteristics and silk-covered wire winding production process characteristics, such as principal component analysis to solve high-dimensional data redundancy, support vector regression to handle nonlinear mapping relationships, deep reinforcement learning to realize continuous action space control, NSGA-II algorithm to balance multi-objective optimization, and analytic hierarchy process to support dynamic decision-making, each algorithm makes a unique contribution to specific functional links, forming a complete intelligent process optimization control system, which significantly improves the quality stability, process optimization degree and intelligent level of silk-covered wire winding production.

[0065] The above Figure 2The intelligent analysis based wire coating enameling winding process optimization control system in the embodiment of the present application is described in detail from the perspective of modular functional entities. The intelligent analysis based wire coating enameling winding process optimization control device in the embodiment of the present application is described in detail from the perspective of hardware processing.

[0066] Figure 3 The intelligent analysis based wire coating enameling winding process optimization control device 300 provided by the embodiment of the present application can have great differences due to different configurations or performances, and can include one or more than one processor (central processing unit, CPU) 310 (for example, one or more than one processor) and a memory 320, one or more than one storage medium 330 (for example, one or more than one mass storage device end) storing an application program 333 or data 332. The memory 320 and the storage medium 330 can be temporary storage or persistent storage. The program stored in the storage medium 330 can include one or more than one module (not shown in the figure), and each module can include a series of instruction operations in the intelligent analysis based wire coating enameling winding process optimization control device 300. Further, the processor 310 can be configured to communicate with the storage medium 330, execute a series of instruction operations in the storage medium 330 on the intelligent analysis based wire coating enameling winding process optimization control device 300, so as to realize the steps of the intelligent analysis based wire coating enameling winding process optimization control method described above.

[0067] The intelligent analysis based wire coating enameling winding process optimization control device 300 can further include one or more than one power supply 340, one or more than one wired or wireless network interface 350, one or more than one input and output interface 360, and / or one or more than one operating system 331, such as Windows Serve, Mac OS X, Unix, Linux, FreeBSD, etc. Those skilled in the art can understand that the intelligent analysis based wire coating enameling winding process optimization control device 300 can further include other components, which are not shown in the figure, for example, bus lines, power supply, keyboard, mouse, display, hard disk, etc. Figure 3 The structure of the intelligent analysis based wire coating enameling winding process optimization control device shown in the figure does not constitute a limitation on the intelligent analysis based wire coating enameling winding process optimization control device provided by the present application, and can include more or fewer components than shown in the figure, or combine certain components, or different component arrangements.

[0068] The application further provides a computer readable storage medium, which can be a nonvolatile computer readable storage medium or a volatile computer readable storage medium, and the computer readable storage medium stores instructions, and the instructions make a computer execute the steps of the silk coating wire winding process optimization control method based on intelligent analysis when the instructions are run on the computer.

[0069] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described system, system and unit can refer to the corresponding processes in the foregoing method embodiments, and will not be described here.

[0070] The integrated unit, if realized in the form of a software functional unit and sold or used as an independent product, can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the application or the whole or part of the technical solutions that essentially contribute to the prior art can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a plurality of instructions for making a smart analysis based on a silk coating wire winding process optimization control device (which can be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the method described in each embodiment of the application. The foregoing storage medium includes a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various program code storage media.

[0071] The above embodiments are only used to illustrate the technical solutions of the application, rather than limit them; although the application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacements for some technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the application.

Claims

1. A silk-wrapped enameled winding process optimization control method based on intelligent analysis, characterized in that: The method comprises: Real-time acquisition and pre-processing of temperature, humidity, tension, speed, and current parameters in multiple areas of the production line to obtain standardized process parameter data sets; Based on the process parameter data set, the high-dimensional data is reduced in dimension by principal component analysis, and a process parameter influence model is constructed by combining grey relational analysis and support vector regression algorithm to obtain the mapping relationship between process parameters and product quality indicators; The mapping relationship is input into a deep deterministic policy gradient network, and the Actor-Critic network structure is trained by setting a reward function and an experience replay mechanism to obtain a dynamic adjustment strategy for process parameters; The dynamic adjustment strategy is optimized by non-dominated sorting genetic algorithm, and the objective function weight is dynamically adjusted in combination with the hierarchical analysis method. The optimal parameter adjustment sequence is executed through model predictive control to obtain a multi-objective optimization control scheme that balances product indicators, production efficiency and energy consumption.

2. The silk-covered enameled winding process optimization control method based on intelligent analysis according to claim 1 is characterized in that: The temperature, humidity, tension, speed and current parameters of multiple areas of the production line are collected and preprocessed in real time to obtain a standardized process parameter data set, including: The annealing area, painting area, baking area, cooling area, lubrication area and take-up area are equipped with temperature sensors, humidity sensors, tension sensors, speed sensors, current sensors and high-precision image acquisition devices. Real-time data is acquired at a frequency of 10 times per second to obtain original production data for multiple areas. The multi-region production raw data is transmitted to a central data processing unit, and outliers and noise interference are filtered out through edge computing technology to obtain a preliminarily cleaned data stream; Applying moving average filtering, median filtering and wavelet transform denoising algorithms to the preliminarily cleaned data stream to perform data cleaning to obtain process parameters after denoising; Performing interpolation processing on the missing data in the process parameters after noise reduction, filling the data gaps through linear interpolation or polynomial interpolation to obtain a parameter sequence; Correlating and matching the parameter sequence with product quality indicators, establishing a quality database including conductor diameter, insulation paint film thickness, paint film adhesion, and voltage resistance performance, and obtaining a parameter-quality correspondence; The parameter-quality correspondence is standardized, and each parameter is mapped to a unified numerical interval using a z-score or min-max method to obtain a standardized process parameter data set.

3. The silk-covered enameled winding process optimization control method based on intelligent analysis according to claim 1, characterized in that: Based on the process parameter data set, the high-dimensional data is reduced in dimension by principal component analysis, and the process parameter influence model is constructed by combining grey correlation analysis and support vector regression algorithm to obtain the mapping relationship between process parameters and product quality indicators, including: Applying principal component analysis to the standardized process parameter data set to calculate eigenvalues ​​and eigenvectors, extracting main eigenvectors that can explain data variation, and obtaining a key feature set after dimensionality reduction; The key feature set after dimensionality reduction is subjected to multi-scale decomposition in combination with wavelet packet transform technology to capture the dynamic change characteristics of temperature, tension, and speed parameters in different frequency domains and obtain a time series feature vector; Based on the time series feature vector, a grey correlation analysis model is constructed between process parameters and product quality indicators, and the correlation between each process parameter and the paint film uniformity, adhesion, and insulation performance quality indicators is calculated to obtain a correlation matrix; The fuzzy comprehensive evaluation method is introduced into the correlation matrix in combination with expert experience and historical production data, and the weight coefficient of each process parameter is set to obtain the fuzzy relationship matrix of process parameters and product quality; The key parameters of annealing temperature, painting speed, paint viscosity, and baking temperature in the fuzzy relationship matrix are input into a support vector regression algorithm, mapped to a high-dimensional feature space using a Gaussian kernel function, and a regression model with a determination coefficient R² of not less than 0.92 is trained to obtain a parameter-quality prediction model; The time series characteristics of the process parameters in the parameter-quality prediction model are analyzed using an autoregressive integral moving average model to predict the short-term change trends of key parameters, build a multi-level process parameter impact model library, and obtain the mapping relationship between process parameters and product quality indicators.

4. The silk-covered enameled winding process optimization control method based on intelligent analysis according to claim 1, characterized in that: The mapping relationship is input into a deep deterministic policy gradient network, and the Actor-Critic network structure is trained by setting a reward function and an experience replay mechanism to obtain a dynamic adjustment strategy for process parameters, including: A Markov decision process model is performed on the mapping relationship between the process parameters and the product quality indicators. A state space is defined to include a set of process parameters, equipment operating status, and semi-finished product quality characteristics. An action space is defined to include adjustment parameters for temperature adjustment amplitude, tension change, and painting speed change, thereby obtaining a state-action environment model. A reward function is constructed based on the state-action environment model. By calculating the weighted sum of product quality score and production efficiency, a positive reward value is assigned to high-quality products and a negative reward value is assigned to unqualified products. An energy consumption penalty term is introduced to obtain a reinforcement learning training signal. Input the reinforcement learning training signal into a deep deterministic policy gradient network to construct a four-layer fully connected Actor network structure. The number of neurons in the input layer matches the state dimension. The two hidden layers are set to 256 and 128 nodes respectively. The ReLU activation function is configured. The output layer is mapped to the action space using the Tanh activation function to obtain a parameter adjustment policy network. The state-action environment model is input into a four-layer fully connected critic network structure, where the input layer receives state information, the second hidden layer fuses action information, a ReLU activation function is used, and the output layer calculates the Q value to obtain a state-action value evaluation network; Based on the parameter adjustment strategy network and the state-action value evaluation network, an experience replay pool with a capacity of 100,000 is constructed. Each training session randomly samples a batch size of 128 samples, sets the target network update rate τ = 0.001 for soft update, and introduces the Ornstein-Uhlenbeck noise process with an initial noise intensity of 0.5 and a decay coefficient of 0.99 to obtain an exploration-exploitation balance mechanism. Hierarchical priority constraints are designed for the exploration-exploitation balance mechanism to limit the variation of key parameters such as temperature and tension to a safe range. The constraints are converted into components of the reward function through the penalty function method. After accumulating more than 500,000 interaction samples, a dynamic adjustment strategy for process parameters is obtained.

5. The silk-covered enameled winding process optimization control method based on intelligent analysis according to claim 1, characterized in that: The dynamic adjustment strategy is optimized by a non-dominated sorting genetic algorithm, and the objective function weight is dynamically adjusted in combination with the hierarchical analysis method. The optimal parameter adjustment sequence is executed through model predictive control to obtain a multi-objective optimization control solution that balances product indicators, production efficiency, and energy consumption, including: A multi-objective optimization model is constructed based on the dynamic adjustment strategy of the process parameters, and a set of objective functions is defined to include a product quality index function, a production efficiency function, and an energy consumption function, wherein the product quality index function is composed of a weighted sum of insulation strength and paint film uniformity indicators, the production efficiency function is composed of the effective output per unit time, and the energy consumption function is composed of the standardized sum of electrical energy and thermal energy, thereby obtaining an optimization objective function set; An improved non-dominated sorting genetic algorithm is applied to the optimization objective function set, with the population size set to 100, the maximum number of iterations to 200, the crossover probability to 0.85, and the mutation probability to 0.

15. The population diversity is maintained through non-dominated sorting and crowding distance calculation mechanisms, and the Pareto optimal solution set of the process parameter configuration is obtained; A heuristic operator based on problem characteristics is introduced into the Pareto optimal solution set, and an intelligent crossover operator considering the continuity of process parameters and an adaptive mutation operator guided by parameter sensitivity are applied to accelerate the convergence of the algorithm and obtain multiple feasible process parameter configuration schemes; Input the process parameter configuration plan into the analytic hierarchy process decision support module, dynamically adjust the weight coefficients of each objective function in combination with the current production target, form the final process parameter optimization plan, and obtain the optimized control sequence; Based on the optimized control sequence, the model predictive control strategy is applied, the prediction horizon is set to 15, the system current state is used to predict the system behavior in multiple time steps in the future, the optimal control sequence is determined by solving the rolling optimization problem, and the model predictive control action is obtained; The recursive least squares method is introduced into the model predictive control action for online parameter update, and a multi-level safety assurance mechanism is configured, including parameter change rate limitation, safety boundary check and emergency intervention logic, to ensure the safety of the control action and obtain a multi-objective optimization control scheme that balances product indicators, production efficiency and energy consumption.

6. The silk-covered enameled winding process optimization control method based on intelligent analysis according to claim 5, characterized in that: The Pareto optimal solution set is introduced with a heuristic operator based on problem characteristics, and an intelligent crossover operator considering the continuity of process parameters and an adaptive mutation operator guided by parameter sensitivity are applied to accelerate the convergence of the algorithm and obtain a variety of feasible process parameter configuration schemes, including: Performing historical data correlation analysis on the process parameters in the Pareto optimal solution set, establishing a correlation matrix between parameters, identifying strong coupling relationships between parameters, and obtaining a parameter dependency network; An intelligent crossover operator is designed based on the parameter dependency network to implement a collaborative crossover operation on strongly correlated parameters, thereby maintaining consistency and continuity in parameter changes and obtaining offspring individuals that meet process constraints. Calculating the parameter sensitivity coefficient for each process parameter in the progeny individuals, quantifying the influence of each parameter on the objective function by performing partial derivative analysis on the historical process data, and obtaining a parameter sensitivity ranking table; An adaptive mutation operator is constructed according to the parameter sensitivity ranking table, a smaller mutation step length is set for parameters with high sensitivity, and a larger mutation step length is set for parameters with low sensitivity, to obtain an adaptive mutation strategy; The simulated annealing mechanism is applied to the adaptive mutation strategy, with an initial temperature of 100 and a cooling coefficient of 0.

95. The mutation probability is dynamically adjusted according to the algorithm iteration process, maintaining a high level of exploration in the early stage and gradually converging to a local fine search in the later stage to obtain an optimized search trajectory. The elite solutions in the optimization search trajectory are retained to the next generation. At the same time, the solutions with high repetition in the population are subjected to diversity enhancement processing, and chaotic perturbations are introduced to avoid premature convergence of the population. After 200 generations of iteration, a diversified process parameter configuration scheme that meets the process continuity constraints is obtained.

7. The silk-covered enameled winding process optimization control method based on intelligent analysis according to claim 5, characterized in that: The process parameter configuration scheme is input into the analytic hierarchy process decision support module, and the weight coefficients of each objective function are dynamically adjusted in combination with the current production target to form a final process parameter optimization scheme and obtain an optimization control sequence, including: A three-level hierarchical analysis structure is constructed for the multiple feasible process parameter configuration schemes. The first level is the overall target layer, the second level is the criterion layer including three indicators: product quality, production efficiency, and energy consumption, and the third level is the solution layer including multiple parameter configuration schemes, thereby obtaining a hierarchical analysis judgment matrix. Based on the hierarchical analysis judgment matrix, pairwise comparison is performed, and the relative importance comparison matrix between the indicators of each criterion layer is constructed using the 1-9 scaling method. The maximum eigenvalue and its corresponding eigenvector are calculated using the eigenvalue method to obtain the criterion layer weight coefficient; Perform consistency test on the criterion layer weight coefficient, calculate the consistency index CI and random consistency ratio CR, when CR is less than 0.1, it is considered that the judgment matrix has satisfactory consistency, otherwise the judgment matrix is ​​readjusted to obtain the consistency test result of the criterion layer indicator weight; According to the consistency test results, the relative weights of the indicators in the criterion layer are adjusted. In combination with the current production task requirements, the weight coefficients are automatically increased or decreased through fuzzy rules. The production efficiency weight is increased for urgent orders, and the quality indicator weight is increased for high-end product production, thereby obtaining a dynamic weight distribution scheme. Applying the dynamic weight allocation scheme to the comprehensive evaluation of each process parameter configuration scheme, calculating the weighted total score of each scheme under each objective function, and sorting them according to the total score, selecting the configuration scheme with the highest comprehensive score, and obtaining the optimized process parameter combination; The optimized process parameter combination is expanded in time series. According to the dynamic characteristics of the production process, the static process parameters are converted into a time series parameter adjustment strategy to form an optimal parameter adjustment trajectory covering the next 15 time steps, thereby obtaining an optimized control sequence.

8. A silk-wrapped enameled winding process optimization control system based on intelligent analysis, characterized in that: For realizing the silk-wrapped enameled winding process optimization control method based on intelligent analysis according to any one of claims 1 to 7, the silk-wrapped enameled winding process optimization control system based on intelligent analysis comprises: The acquisition module is used to collect and pre-process the temperature, humidity, tension, speed and current parameters of multiple areas of the production line in real time to obtain a standardized process parameter data set; A dimensionality reduction module is used to perform dimensionality reduction processing on high-dimensional data based on the process parameter data set through principal component analysis, and to construct a process parameter influence model by combining grey correlation analysis and support vector regression algorithm to obtain a mapping relationship between process parameters and product quality indicators; A training module is used to input the mapping relationship into a deep deterministic policy gradient network, train the Actor-Critic network structure by setting a reward function and an experience replay mechanism, and obtain a dynamic adjustment strategy for process parameters; An execution module is used to optimize the dynamic adjustment strategy using a non-dominated sorting genetic algorithm, dynamically adjust the objective function weights in combination with the hierarchical analysis method, and execute the optimal parameter adjustment sequence through model predictive control to obtain a multi-objective optimization control solution that balances product indicators, production efficiency, and energy consumption.

9. A silk-wrapped enameled winding process optimization control device based on intelligent analysis, characterized in that: It 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, it implements the silk-wrapped enameled winding process optimization control method based on intelligent analysis as described in any one of claims 1 to 7.

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 execute the wire-wrapped enameled winding process optimization control method based on intelligent analysis according to any one of claims 1 to 7.

Images