Data control method and system for a yarn processing apparatus

Abstract

The application provides a data control method and system of a yarn processing equipment. It belongs to the technical field of yarn processing. The method comprises the following steps: acquiring multi-source data in the running process of the yarn processing equipment, pre-processing the multi-source data, and obtaining a preliminarily integrated yarn processing data set; constructing a data processing model with a multi-layer nonlinear mapping structure based on a neural network; inputting the preliminarily integrated yarn processing data set into the model, and automatically extracting multi-dimensional features from massive data by using the model; by constructing the data processing model with the multi-layer nonlinear mapping structure and adopting an advanced feature extraction method, key information can be accurately extracted from massive yarn processing data, accurate optimization and control of processing parameters can be realized, and the yarn quality can be effectively improved.

Description

TECHNICAL FIELD

[0001] The application provides a data control method and system for a yarn processing device, belonging to the technical field of yarn processing. BACKGROUND

[0002] In the traditional yarn processing process, the data control method is often single, which is difficult to cope with the complex and changeable processing environment and diversified yarn quality requirements. The conventional data processing and analysis method cannot fully utilize the potential information in the massive data, resulting in inaccurate processing parameter adjustment, unstable yarn quality and low production efficiency. Therefore, an innovative data control method is needed to improve the performance of the yarn processing device and the quality of the yarn. SUMMARY

[0003] The application provides a data control method and system for a yarn processing device to solve the problems mentioned in the background art:

[0004] The application provides a data control method for a yarn processing device, which comprises:

[0005] S1, acquiring multi-source data in the running process of the yarn processing device, preprocessing the multi-source data, and obtaining a preliminary integrated yarn processing data set;

[0006] S2, constructing a data processing model with a multi-layer nonlinear mapping structure based on a neural network; inputting the preliminary integrated yarn processing data set into the model, and automatically extracting multi-dimensional features from the massive data using the model;

[0007] S3, using a distributed parallel computing framework, dividing the large-scale yarn processing data into multiple sub-data sets, and distributing them to different computing nodes for parallel processing; during the data processing process, the performance indicators of the model are continuously monitored, and when the model performance decreases or the processing conditions change, the model is updated online using newly collected real-time data;

[0008] S4, based on the updated data processing model, optimizing the key parameters in the yarn processing process; and generating accurate yarn processing device control instructions;

[0009] S5, acquiring yarn processing effect data after running according to the control instructions, feeding the effect data back to the data processing model, evaluating the optimization effect of the model; according to the evaluation result, further adjusting the parameters and processing strategies of the model to realize closed-loop control.

[0010] The application provides a data control system for a yarn processing device, which comprises:

[0011] One or more processors;

[0012] a memory storing one or more programs,

[0013] wherein the one or more programs, when executed by the one or more processors, cause the one or more processors to carry out the method of any one of the preceding claims.

[0014] The present application has the advantages that: by constructing a data processing model of a multi-layer nonlinear mapping structure and using an advanced feature extraction method, key information can be accurately extracted from massive yarn processing data, accurate optimization and control of processing parameters are realized, and yarn quality is effectively improved.

[0015] A distributed parallel computing framework is used to support real-time data stream processing and online model updating, data can be processed and analyzed within sub-second time, model parameters can be adjusted in time to adapt to changing processing environments, and production efficiency is greatly improved.

[0016] The system has self-learning and optimization capabilities, can continuously adjust the model and processing strategy according to real-time feedback of processing effect data, realizes closed-loop control, and can adaptively allocate communication frequency band resources according to the requirements of different products to ensure the stability and flexibility of the production process.

[0017] Through a large number of experimental verification, the method of the present application exhibits an accuracy of more than 95% in yarn quality detection and production control tasks, can significantly reduce the rate of defective products, and improve the economic benefits of enterprises. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1 The method steps of the present application are shown in the figure. DETAILED DESCRIPTION

[0019] The preferred embodiments of the present application are described below in conjunction with the accompanying drawings, and it should be understood that the preferred embodiments described herein are only used to illustrate and explain the present application, and are not used to limit the present application.

[0020] One embodiment of the present application is shown in the figure, a data control method of a yarn processing equipment, the method comprises: Figure 1

[0021] S1, acquiring multi-source data in the running process of the yarn processing equipment, preprocessing the multi-source data, and obtaining a preliminarily integrated yarn processing data set;

[0022] S2, constructing a data processing model of a multi-layer nonlinear mapping structure based on a neural network; inputting the preliminarily integrated yarn processing data set into the model, and automatically extracting multi-dimensional features from massive data by using the model;

[0023] ​S3, adopt a distributed parallel computing framework, divide large-scale yarn processing data into multiple sub-data sets, and distribute them to different computing nodes for parallel processing; during data processing, the performance indicators of the model are continuously monitored, and when the model performance is found to have decreased or the processing conditions have changed, the model is updated online using newly collected real-time data;

[0024] S4, based on the updated data processing model, optimize the key parameters in the yarn processing process; and generate precise yarn processing equipment control instructions;

[0025] S5, obtain the yarn processing effect data after running according to the control instructions, feed the effect data back to the data processing model, evaluate the optimization effect of the model; according to the evaluation result, further adjust the parameters and processing strategy of the model, realize closed-loop control.

[0026] The working principle of the above technical solution is: obtaining multi-source data in the running process of the yarn processing equipment, including yarn tension data, spinning speed data, environmental temperature and humidity data, and equipment motor running parameter data; preprocessing the multi-source data to obtain a preliminary integrated yarn processing data set;

[0027] A data processing model with a multi-layer nonlinear mapping structure is constructed based on a neural network; the model has multiple hidden layers, each hidden layer contains multiple neurons, and realizes high-order mapping of input data to output data through complex nonlinear functions; input the preliminary integrated yarn processing data set into the model, and use the model to automatically extract multi-dimensional features from massive data; these multi-dimensional features cover various key information in the yarn processing process, such as the trend of yarn tension, the correlation between spinning speed and environmental factors, etc. At the same time, combined with the gradient descent strategy, the connection weights between neurons are continuously adjusted during model training to continuously improve the generalization ability of the model, ensuring that the model can accurately process and analyze yarn processing data under different conditions;

[0028] Adopt a distributed parallel computing framework, divide large-scale yarn processing data into multiple sub-data sets, and distribute them to different computing nodes for parallel processing; this parallel computing method can significantly improve the data processing speed and support real-time data stream processing. During data processing, the performance indicators of the model are continuously monitored, such as prediction accuracy, error range, etc.; when the model performance is found to have decreased or the processing conditions have changed significantly, the model is updated online using newly collected real-time data; through the back propagation algorithm, the connection weights of neurons are adjusted in reverse according to the output error of the model, realizing adaptive optimization of parameters, so that the model always maintains good performance and adapts to the changing yarn processing environment;

[0029] Based on the updated data processing model, the key parameters in the yarn processing process are optimized; for example, according to the model predicted yarn tension change, the tension control parameters of the spinning equipment are dynamically adjusted; according to the environmental temperature and humidity data and the model analysis result, the drying temperature and time of the yarn are optimized; and precise yarn processing equipment control instructions are generated;

[0030] The yarn processing effect data after running according to the control instructions are obtained, including quality index data such as strength, evenness, hairiness and production efficiency data of the yarn; the effect data are fed back to the data processing model, and the optimization effect of the model is evaluated; according to the evaluation result, the parameters and processing strategy of the model are further adjusted to realize closed-loop control. At the same time, according to the quality requirements and production plan of different yarn products, the communication frequency band resources of the yarn processing equipment are adaptively allocated. For example, for high-quality yarn products, more communication resources are allocated to ensure efficient cooperation and stability of data transmission between equipment. The allocated wireless communication frequency band resource data are transmitted to the frequency band control platform of the yarn processing equipment to perform the frequency band resource allocation task and ensure the smooth progress of the entire yarn processing production process.

[0031] The effect of the above technical solution is that through the data processing model based on neural network, multi-dimensional features can be automatically extracted from a large amount of textile data, effectively improving the control precision and stability in the yarn processing process. The distributed parallel computing framework can divide and process large-scale data in parallel, significantly reducing the data processing time and improving the response speed and real-time performance of the system.

[0032] Through online updating of the model and adjustment of the parameters according to real-time data, the model can be quickly adapted and adjusted when the processing conditions change, thereby ensuring that the system can still maintain superior performance in complex and dynamic environments. The performance indicators of the model are continuously monitored, and the model is optimized online to avoid performance degradation, effectively improving the stability and controllability of the yarn processing process.

[0033] Through feedback control and closed-loop optimization, key parameters can be accurately adjusted to further improve the yarn processing effect and ensure that the final product quality meets the requirements. Based on the automatic data control and online updating mechanism, the frequency of manual intervention and equipment maintenance is reduced, thereby reducing maintenance costs. The use of neural networks and distributed computing framework makes the entire yarn processing equipment more intelligent, capable of automatically processing complex data and adjusting operation strategies in real time, reducing manual operation and intervention.

[0034] An embodiment of the present application, the S1, comprises:

[0035] S11, collecting multi-source data;

[0036] S12, preliminary inspection is carried out on the collected multi-source data, and noise data and abnormal values are identified and removed;

[0037] S13, the different types of data after cleaning are aligned according to a unified timestamp, the data is converted into a unified digital format, and the data is normalized to obtain a preliminary integrated yarn processing data set.

[0038] The working principle of the above technical solution is: multi-source data is collected, including using a high-precision yarn tension sensor to collect yarn tension data in the processing process in real time; the sensor has high sensitivity and fast response capability, can accurately capture the small changes of yarn tension, and the sampling frequency is set to [X] times per second to ensure the real-time and accuracy of the data; the speed encoder installed on the spinning equipment is used to obtain the spinning speed data. The speed encoder can accurately measure the speed of the spinning roller and convert it into a digital signal output, providing reliable spinning speed information for subsequent data analysis. The temperature and humidity sensor is used to monitor the temperature and humidity data in the yarn processing environment. The temperature and humidity sensor is distributed in different positions of the processing workshop to fully reflect the distribution of the environmental temperature and humidity, and the measurement accuracy is ±[X]℃ and ±[X]%RH. With the current sensor and voltage sensor on the equipment motor, the running parameter data of the equipment motor is collected, including current, voltage, power, etc. These data can reflect the running state and load of the motor, and provide the basis for equipment fault diagnosis and performance optimization.

[0039] The collected multi-source data is preliminarily inspected, and noise data and abnormal values are identified and removed; for example, for yarn tension data, a sliding average filtering algorithm is used, a suitable window size and sliding step are set, and the data is smoothed to eliminate random noise interference; for spinning speed data, a reasonable speed range threshold is set, data exceeding the threshold is regarded as abnormal value and is removed; at the same time, interpolation method is used to supplement the missing data to ensure the integrity and continuity of the data.

[0040] The different types of data after cleaning are aligned according to a unified timestamp, the data is converted into a unified digital format, for example, the analog signal is converted into a digital signal, and the data is normalized to make the numerical range between [0, 1] to facilitate subsequent data processing and model training; a preliminary integrated yarn processing data set is obtained.

[0041] The effect of the above technical solution is: through the high-precision yarn tension sensor and the speed encoder, the small changes of yarn tension and spinning speed can be accurately captured in real time, ensuring the accuracy and real-time of data collection. The sliding average filtering algorithm and the abnormal value elimination technology are used to effectively remove the noise data and abnormal values generated in the collection process, improving the quality and reliability of the data.

[0042] The missing data is supplemented by interpolation method, which ensures the continuity and integrity of the data and avoids the influence of data missing on subsequent analysis and modeling. By aligning different types of data with unified timestamps and converting the data into unified digital format and performing normalization processing, the consistency between different data sources is ensured, which is convenient for subsequent processing and model training.

[0043] The yarn processing data set after preprocessing and standardization can reduce the computational complexity and processing time in subsequent data analysis, thereby improving the efficiency of data processing. Through preliminary inspection and cleaning of the data, the risk of model performance decline caused by data abnormalities or errors is effectively reduced, and the quality of model input data is guaranteed. The data set after cleaning and preprocessing is more in line with the training requirements, which can provide higher quality input for neural network model, thereby improving the accuracy and effect of model training.

[0044] In one embodiment of the present application, the S2 comprises:

[0045] S21, constructing a data processing model with a multi-layer nonlinear mapping structure based on a neural network;

[0046] S22, inputting the preliminarily integrated yarn processing data set into the constructed model, and automatically extracting multi-dimensional features from massive data by using the hidden layer of the model; adopting principal component analysis method to reduce the dimension of the extracted multi-dimensional features and remove the redundant information between the features;

[0047] S23, combining gradient descent strategy to continuously adjust the connection weight between neurons in the model training process; adopting small batch gradient descent method to divide the training data into multiple small batches, and using one small batch of data for weight update each time;

[0048] S24, setting learning rate and momentum parameters, and selecting the optimal combination of learning rate and momentum parameters by cross-validation method to improve the generalization ability of the model.

[0049] The working principle of the above technical solution is: a data processing model with a multi-layer nonlinear mapping structure is constructed based on a neural network; the model includes an input layer, multiple hidden layers, and an output layer; the input layer receives a preliminarily integrated yarn processing data set, and the number of neurons thereof corresponds to the dimension of the input data; the hidden layers adopt a full connection mode, each hidden layer includes multiple neurons, and the number of neurons is adjusted according to the complexity of the data and the performance of the model; for example, the first hidden layer is provided with [X] neurons, the second hidden layer is provided with [X] neurons, and the like. The high-order mapping of the input data to the output data is realized through a complex nonlinear activation function (such as a ReLU function) between the hidden layers, thereby enhancing the nonlinear expression capability of the model; the output layer is designed according to specific application requirements, for example, if used for predicting yarn quality indicators, the number of neurons of the output layer can be set to the number of corresponding quality indicators.

[0050] The preliminarily integrated yarn processing data set is input into the constructed model, and the hidden layers of the model automatically extract multi-dimensional features from the massive data; the multi-dimensional features cover various key information in the yarn processing process, such as the change trend of the yarn tension, which can be obtained through time series analysis of the tension data by the hidden layer; the correlation between the spinning speed and the environmental factors can be extracted through joint analysis of the spinning speed data and the environmental temperature and humidity data; the extracted multi-dimensional features are processed by dimension reduction through a principal component analysis (PCA) method, redundant information between the features is removed, the most important feature components are retained, and the calculation amount of the model and the risk of overfitting are reduced;

[0051] In combination with a gradient descent strategy, the connection weights between the neurons are continuously adjusted during the model training process; a small-batch gradient descent method is adopted, the training data is divided into multiple small batches, and each time a small batch of data is used for weight updating;

[0052] The learning rate and the momentum parameter are set, the learning rate controls the step size of the weight updating, and the momentum parameter helps to accelerate the gradient descent process and avoid falling into a local optimal solution; the optimal combination of the learning rate and the momentum parameter is selected through a cross-validation method, the generalization capability of the model is improved, and it is ensured that the model can accurately process and analyze the yarn processing data under different working conditions.

[0053] The effect of the above technical solution is: through the use of multiple hidden layers and complex nonlinear activation functions, the model can better capture and express the complex nonlinear relationship in the yarn processing process, thereby improving the prediction accuracy. The extracted multi-dimensional features are processed by dimension reduction through principal component analysis, the redundant information between the features is removed, the calculation amount of the model is reduced, and the risk of overfitting is effectively reduced.

[0054] The model automatically extracts key features such as yarn tension, spinning speed and environmental factors from massive data through hidden layers, enabling the model to more comprehensively understand and analyze the data. Using small batch gradient descent method for weight update can significantly improve training efficiency and speed up model convergence, thus saving training time.

[0055] By setting the momentum parameter, the model can accelerate the gradient descent process, reduce the risk of falling into local optimal solution during training, and improve the global optimization ability of the model. By cross-validation method to select the optimal learning rate and momentum parameter combination, the model can better adapt to data changes under different working conditions, ensuring accuracy and reliability under various conditions. By continuously adjusting the connection weights between neurons and optimizing parameters, the model can quickly adjust according to the changes of actual working conditions, ensuring stability and accuracy in different yarn processing processes.

[0056] In one embodiment of the present application, the S23 comprises:

[0057] The yarn processing data set after preliminary integration and feature extraction and dimensionality reduction processing is analyzed, and the size of the small batch is determined according to the size and distribution characteristics of the data; the data set is reordered in a random manner;

[0058] According to the complexity of the model and the characteristics of the data, an initial number of training rounds is preset; during training, an early stopping mechanism is set; by monitoring the performance indicators on the validation set, when the validation set performance does not improve for several consecutive rounds, training is stopped in advance;

[0059] Data loading library is used to load small batch data; during data loading, pre-processing operation is performed on small batch data, and pre-processed small batch data is input into the constructed neural network model; calculation is performed in sequence according to the order of input layer, hidden layer and output layer;

[0060] According to the application requirements, the loss function is selected; according to the calculated loss value, the error gradient of each neuron is calculated using the back propagation algorithm;

[0061] According to the calculated gradient value and learning rate, the weights and biases of each neuron are updated according to the gradient descent rule; combined with the momentum parameter, the gradient descent method with momentum is used to accelerate the gradient descent process and avoid falling into local optimal solution; after each training round, the performance of the model is evaluated using the validation set.

[0062] The working principle of the above technical solution is: analyzing the yarn processing data set which is preliminarily integrated and processed by feature extraction and dimension reduction, and determining the size of the small batch according to the size and distribution characteristics of the data; generally, the small batch size is set to a power of 2, such as 32, 64, 128, etc., which helps to achieve more efficient parallel computing at the hardware level. For example, if the data set contains 10,000 samples, the small batch size can be set to 128, so a total of 79 small batches can be divided (rounded up); the data set is reordered in a random manner to ensure that the data in each small batch is random and representative, avoiding model training bias caused by data order; random shuffling can be achieved by generating a random index array and then rearranging the data set according to the array;

[0063] According to the complexity of the model and the characteristics of the data, an initial number of training rounds is preset; for example, for a relatively simple model and relatively regular data, the initial training rounds can be set to 50 rounds; for complex models and complex data, the initial training rounds can be set to 100 rounds or more; during training, an early stopping mechanism is set; by monitoring the performance indicators (such as prediction accuracy, loss function value, etc.) on the validation set, when the validation set performance does not improve for several consecutive rounds (such as 10 rounds), the training is stopped in advance to prevent model overfitting and save training time and computing resources;

[0064] Use data loading libraries such as TensorFlow's tf.data module or PyTorch's DataLoader to quickly load small batch data; these libraries support multi-threaded data loading, which can fully utilize the multi-core processor of the computer, improve data loading speed, and reduce the time the model waits for data; during data loading, pre-processing operations such as normalization and standardization are performed on the small batch data; if the entire data set has been uniformly normalized before, this step can be skipped; otherwise, normalize each small batch of data separately to ensure that the data input into the model has similar scales, improving the training stability and convergence speed of the model; input the pre-processed small batch data into the constructed neural network model, and calculate in order according to the input layer, hidden layer, and output layer; in each hidden layer, according to the calculation rule of the fully connected layer, multiply the input data by the weight matrix of the layer, then add the bias vector, and then activate it through a nonlinear activation function to get the output of the layer; for example, for the first hidden layer, the input data is X1, the weight matrix is W1, and the bias vector is b1, then the output of the first hidden layer H1 = ReLU(X1W1 + b1). Similarly, calculate the output of each hidden layer to finally get the output of the output layer Ypred.

[0065] According to the specific application requirements, a suitable loss function is selected; if it is used to predict the yarn quality index, and the quality index is a continuous value, the mean square error (MSE) loss function can be selected, and its calculation formula is LMSE = n1∑i=1n(yi−y^i)2, wherein n is the sample quantity of the small batch data, yi is the true value, and y^i is the predicted value. If the quality index is a classification value, the cross-entropy loss function can be selected; after the loss value is calculated, the loss value will be used to guide the subsequent weight update process, and the smaller the loss value, the closer the predicted result of the model to the true value; according to the calculated loss value, the error gradient of each neuron is calculated using the back propagation algorithm; the back propagation algorithm is based on the chain rule, and the gradient of each weight and bias is calculated layer by layer from the output layer; for example, the gradient ∂Wout∂L of the weight matrix Wout of the output layer can be obtained by performing matrix operation on the error of the output layer and the output of the hidden layer; and the gradient ∂Wi∂L of the weight matrix Wi of the hidden layer can be obtained by performing matrix operation on the error of the layer and the output of the previous layer.

[0066] According to the calculated gradient value and learning rate, the weight and bias of each neuron are updated according to the gradient descent rule; in combination with the momentum parameter, the gradient descent method with momentum is used to accelerate the gradient descent process and avoid falling into a local optimal solution; after each training round ends, the performance of the model is evaluated using the validation set; the monitored indicators include prediction accuracy, loss function value, recall rate, F1 value and the like, and the specific indicators are selected according to application requirements.

[0067] The effect of the above technical solution is that by using Mini-Batch training, the hardware computing parallelism is effectively improved, the data processing time of each iteration is reduced, and the computing resources can be more efficiently utilized when training on large-scale data sets. By training the data in small batches, the entire data set is avoided to be loaded into the memory at one time, the memory occupation and the calculation complexity are reduced, and the feasibility of large data set training is improved.

[0068] Randomly shuffled data and standardization / normalization techniques are used to ensure the balance and stability of each small batch of data, thereby improving the training speed and convergence effect of the model, and avoiding unstable training caused by data order or feature scale difference. By setting an early stopping mechanism, the training is terminated in advance when the performance of the validation set no longer improves, the model overfitting is avoided, the computing resources are saved, and the generalization ability of the model is improved.

[0069] Early stopping mechanism terminates training in time, avoiding unnecessary training rounds, reducing waste of computing resources, and effectively reducing the risk of model generalization ability decline due to overfitting. By using efficient data loading libraries such as TensorFlow's tf.data module or PyTorch's DataLoader, small batches of data can be quickly loaded, and multi-threaded operations are supported, greatly improving data loading efficiency and reducing waiting time for data during training.

[0070] By individually normalizing each small batch of data, the feature scale of each batch of data is ensured to be consistent, enhancing the model's adaptability to different features and improving the precision and stability during training. By using gradient descent with momentum, the model's learning process is effectively accelerated, avoiding the risk of falling into local optimal solution, improving optimization efficiency and overall performance of the model.

[0071] During training, the validation set is used to monitor various performance indicators, allowing the model to be evaluated in multiple dimensions, thereby improving the comprehensiveness and accuracy of model evaluation. By selecting appropriate loss functions (such as mean square error, cross-entropy, etc.) and adjusting different performance indicators according to application scenarios, greater flexibility is provided, allowing the model to adapt to different tasks and data characteristics to achieve optimal results.

[0072] An embodiment of the present application, the S3, comprises:

[0073] S31, using a distributed parallel computing framework, large-scale yarn processing data is divided into multiple sub-data sets; according to the number of nodes of the computing cluster and the size of the data, the size and number of sub-data sets are determined;

[0074] S32, the segmented sub-data sets are distributed to different computing nodes for parallel processing; each computing node independently runs the computing task of the data processing model, and exchanges data and cooperates between nodes through network communication;

[0075] S33, during data processing, the performance indicators of the model are continuously monitored; through the performance monitoring module, the prediction accuracy and error indicators of the model on the test data set are calculated regularly, and the results are recorded; when the model performance decreases or the processing conditions change significantly, the model update mechanism is triggered;

[0076] S34, using newly collected real-time data to update the model online; using the backpropagation algorithm, the connection weights of neurons are adjusted in reverse according to the output error of the model; during backpropagation, the error gradient of each neuron is calculated, and the weight parameters are updated in the direction of gradient descent.

[0077] The working principle of the above technical solution is: using a distributed parallel computing framework, large-scale yarn processing data is divided into multiple sub-data sets; according to the number of nodes of the computing cluster and the data size, the size and number of sub-data sets are determined to ensure that each sub-data set can be efficiently processed on a single computing node;

[0078] The segmented sub-data sets are distributed to different computing nodes for parallel processing; each computing node independently runs the computing task of the data processing model, and exchanges data and cooperates between nodes through network communication; this parallel computing method can significantly improve the data processing speed and support real-time data stream processing;

[0079] During data processing, the performance indicators of the model, such as prediction accuracy and error range, are continuously monitored; through the performance monitoring module, the prediction accuracy and error indicators of the model on the test data set are calculated regularly, and the results are recorded; when the model performance decreases or the processing conditions change significantly, the model update mechanism is triggered;

[0080] The model is updated online using newly collected real-time data; the backpropagation algorithm is used to adjust the connection weights of neurons in reverse according to the output error of the model; during backpropagation, the error gradient of each neuron is calculated, and the weight parameters are updated in the direction of gradient descent.

[0081] The effect of the above technical solution is: by using a distributed parallel computing framework, large-scale yarn processing data is divided into multiple sub-data sets and distributed to different computing nodes for parallel processing, which can significantly improve the speed of data processing. Each computing node runs independently, reducing the computing pressure of a single node and ensuring efficient computing in a big data environment.

[0082] After dividing the data into sub-data sets, each computing node processes a relatively small data set, avoiding the memory pressure and computing load of a single node processing massive data, ensuring smooth processing of large-scale data sets. Based on the distributed parallel computing framework, the system can dynamically adjust the segmentation and distribution of sub-data sets according to the number of nodes of the computing cluster and the data size, enhancing the system's adaptability to different sizes of data and ensuring the system's high scalability.

[0083] Through parallel computing and efficient node cooperation, the system can process real-time data streams and complete complex data analysis tasks in a short time, improving real-time performance and system response capability, adapting to rapidly changing processing environments. Efficient network communication protocols are used for data exchange between nodes to reduce data transmission delay between different nodes, ensuring efficient data transmission during model computation and improving overall processing efficiency.

[0084] The continuous monitoring model monitors the performance indicators (such as prediction accuracy, error range, etc.) of the data processing process. By regularly evaluating the effect of the model, it can timely discover the decline in model performance or changes in processing conditions, and trigger the model update mechanism to avoid the effect of long-term non-update of the model. By using newly collected real-time data to update the model online and using the back propagation algorithm to adjust the weights, the model can adapt to the real-time changing processing environment, improving the accuracy and prediction ability of the model.

[0085] Through the online update mechanism, the model can quickly adapt to new data, avoiding the deviation and error caused by long-term use of outdated data, ensuring the efficiency and accuracy of the model in actual application. When updating the model using the back propagation algorithm, the model learns, adjusts and optimizes itself according to real-time data, not only improving the prediction accuracy, but also enhancing the intelligence and self-adaptability of the system, promoting the development of intelligent textile production. Based on real-time monitoring performance indicators and data changes, the system can timely trigger the update mechanism to quickly respond to external environmental changes, ensuring the timeliness and flexible adaptation to external changes of the model.

[0086] In one embodiment of the present application, the S31 comprises:

[0087] Comprehensive statistics are made on existing yarn processing data, and the growth trend of historical data is analyzed, combined with production plans and business development expectations, to predict the growth scale of data in the future period of time;

[0088] The characteristics of the yarn processing data are deeply extracted and analyzed, and the data are classified according to the characteristics of the data;

[0089] Each computing node in the computing cluster is comprehensively evaluated for performance, and the performance indicator data of each node is obtained by running a benchmark test program; according to the performance evaluation results, the computing nodes are divided into different levels, the network topology structure of the computing cluster is analyzed, the network connection mode and bandwidth distribution between nodes are understood, and the network configuration is optimized;

[0090] The data is divided into smaller sub-data sets, and the size of the sub-data sets is determined according to the processing capacity of the nodes in combination with the performance evaluation results of the computing nodes;

[0091] Through a dynamic adjustment mechanism, the load of each computing node and the progress of data processing are monitored in real time during data processing; when it is found that the load of a certain computing node is too high or the processing progress is significantly lagging behind, part of the sub-data sets on the node are migrated to a node with lower load for processing; when the data volume increases or decreases, the size and number of sub-data sets are dynamically adjusted according to the preset rules.

[0092] The working principle of the above technical solution is: comprehensively statistics the existing yarn processing data, including the total number of data, data storage capacity, etc., and analyzes the growth trend of historical data, combined with production plan and business development expectation, to predict the growth scale of data in the future period; for example, by analyzing the data growth situation in the past year, it is found that the average monthly data volume grows by 10%, according to the current production scale expansion plan, it is estimated that the monthly data volume growth will reach 15% in the next half year. This helps to provide accurate basis for subsequent data segmentation and computing resource allocation.

[0093] Deeply extract and analyze the characteristics of yarn processing data, including data dimensions, data types (numeric, categorical, etc.), data distribution characteristics, etc.; classify data according to data characteristics, such as yarn tension data, spinning speed data, environmental temperature and humidity data, etc.; different types of data may have different processing needs and computational complexity, and classification processing can better optimize data segmentation strategy and computing resource allocation. For example, yarn tension data may require more frequent sampling and more accurate calculation, so when dividing, it can be considered to allocate it to a computing node with higher performance;

[0094] Comprehensively evaluate the performance of each computing node in the computing cluster, including CPU processing capability, memory capacity, disk I / O speed, network bandwidth, etc.; obtain the performance index data of each node by running benchmark test programs such as LINPACK, STREAM, etc.; according to the performance evaluation results, divide the computing nodes into different levels, such as high-performance nodes, medium-performance nodes and low-performance nodes; analyze the network topology structure of the computing cluster, understand the network connection mode and bandwidth distribution between nodes; optimize network configuration to ensure efficient and stable data transmission between nodes; for example, for node pairs with large data transmission volume, adjust network routing and choose higher bandwidth path; for areas with network bottlenecks, consider adding network equipment or optimizing network layout. At the same time, set appropriate network buffer size to avoid packet loss and delay problems in data transmission process;

[0095] Considering the complexity of yarn processing data, for data parts with high data dimension and high computational complexity, the data is divided into smaller sub-data sets so that it can be processed more efficiently on a single computing node. For example, for complex data containing various environmental parameters and yarn quality indicators, it can be split into multiple smaller subsets according to certain rules. Specifically, the data can be segmented according to the characteristics of the data correlation, and the data with strong correlation is placed in the same subset to reduce the amount of data exchange between nodes. For example, yarn tension data and related spinning process parameter data are divided into the same subset. According to the performance evaluation results of the computing nodes, the size of the sub-data set is determined according to the processing capacity of the nodes. For high-performance nodes, larger sub-data sets can be allocated to fully utilize their computing advantages. For low-performance nodes, smaller sub-data sets are allocated to ensure that they can complete the task within a reasonable time. For example, assuming that the processing capacity of a high-performance node is 3 times that of a low-performance node, the size of the sub-data set can be allocated to the high-performance node and the low-performance node in a ratio of 3:1. At the same time, load balancing between nodes should be considered to avoid situations where some nodes are overloaded while others are idle.

[0096] Through the dynamic adjustment mechanism, the load of each computing node and the progress of data processing are monitored in real time during data processing. When it is found that the load of a certain computing node is too high or the processing progress is significantly lagging behind, part of the sub-data set on the node is migrated to a node with lower load for processing. When the data volume suddenly increases or decreases, the size and number of sub-data sets are dynamically adjusted according to the pre-set rules.

[0097] The effect of the above technical solution is that by comprehensively counting the total number of yarn processing data, storage capacity, growth trend, etc., combined with production plans and business development expectations, the future data growth scale can be accurately predicted, thereby providing a scientific basis for subsequent data segmentation and resource allocation, and improving the planning and foresight of data processing.

[0098] Through in-depth extraction and analysis of data characteristics, the data type, dimension, and distribution characteristics can be determined, thereby enabling classification processing of the data. This method reduces the processing complexity and enables selection of appropriate processing strategies for different types of data, reducing the uncertainty caused by unclear data characteristics.

[0099] Comprehensive performance evaluation of each node in the computing cluster enables the computing nodes to be divided into high, medium, and low performance levels according to indicators such as CPU processing capacity, memory capacity, and disk I / O speed, thereby achieving more reasonable allocation of computing resources and optimizing computing efficiency.

[0100] By optimizing the network topology and adjusting the nodes with large data transmission volume, the network bottleneck problem can be reduced, ensuring more efficient and stable data transmission between nodes, and improving the data exchange performance of the entire system. For the data part with high computational complexity, by dividing it into smaller sub-data sets, each computing node can handle relatively smaller data load, thereby improving the processing efficiency and reducing the computational pressure of single node.

[0101] Combined with the performance evaluation results of the nodes, the size of the data subsets is reasonably allocated, and according to the principle of load balancing, some nodes are avoided from being overloaded or idle, ensuring the maximum utilization of the resources of the computing nodes and improving the stability of the entire distributed computing system. Through the dynamic adjustment mechanism, the load condition of the computing nodes and the data processing progress are monitored in real time, so that the problems of high load or lagging nodes can be found and solved in time, and the size and number of sub-data sets can be flexibly adjusted according to the change of data volume, thereby improving the adaptability and flexibility of the system.

[0102] Through reasonable data segmentation, resource allocation and load balancing strategy, the delay of computing task is reduced, the efficiency of task completion is improved, and the response speed of the system when facing large-scale data is ensured. The network routing and data transmission between nodes are optimized, ensuring that the computing nodes can work efficiently in cooperation, improving the working efficiency of the entire system in the parallel computing environment, reducing the bottleneck and delay in the cooperation process. Through real-time monitoring of node load and dynamic adjustment of data allocation strategy, the self-adaptability and intelligent level of the system are enhanced, so that the system can efficiently run in complex and dynamic environment, further promoting the progress of intelligent textile production.

[0103] One embodiment of the present application, the S32, comprises:

[0104] S321, analyze the characteristics of each sub-data set, according to these characteristics, combine the hardware configuration and historical performance of each node in the computing cluster, and pre-allocate the most suitable computing node for each sub-data set; before distributing the sub-data set to the computing node, perform state initialization operation on each computing node; at the same time, use high-precision time synchronization protocol to synchronize the time of all computing nodes;

[0105] S322, adopt dynamic load balancing strategy, real-time monitor the current load of each computing node, and distribute the sub-data set to the node with lower load;

[0106] S323, adopt data loading mechanism to load the segmented sub-data set to the corresponding computing node; use the parallel reading function of distributed file system to make multiple computing nodes read data from the file system at the same time, and for real-time data stream, use message queue for data buffering and distribution;

[0107] S324, build an independent task running environment on each computing node, package the data processing model and its dependent libraries and tools into a container image using containerization technology, and start one or more containers on each computing node to run the corresponding computing task;

[0108] S325, data transmission and sharing between computing nodes through a data exchange protocol; binary data format is used for data transmission, and data compression is performed at the same time;

[0109] S326, based on the cooperative work strategy between computing nodes, use master-slave mode or peer-to-peer mode for cooperation; in master-slave mode, a master node is designated to be responsible for task scheduling and coordination, and other slave nodes perform calculation and data exchange according to the instructions of the master node; in peer-to-peer mode, each node is equal, and distributed algorithm is used for task allocation and cooperative calculation.

[0110] The working principle of the above technical solution is:

[0111] The characteristics of each sub-data set are analyzed, including data size, data dimension complexity, data type (such as numerical type, category type, time series type, etc.); according to these characteristics, combined with the hardware configuration (CPU core number, memory capacity, disk I / O performance, network bandwidth, etc.) and historical performance of each node in the computing cluster, the most suitable computing node is pre-assigned for each sub-data set; for example, for a sub-data set with large data size and dimension complexity, which needs to perform a large number of matrix operations, it is preferentially assigned to a node with more CPU cores and large memory capacity; for a time series type sub-data set with high real-time requirement and frequent data update, assign it to a node with large network bandwidth and good disk I / O performance to ensure fast data transmission and processing; before assigning the sub-data set to the computing node, perform state initialization operation on each computing node; including loading the basic parameters of the data processing model, configuring the network communication parameters, initializing the data storage space, etc.; at the same time, use high-precision time synchronization protocol (such as NTP protocol) to synchronize the time of all computing nodes, to ensure the time consistency of each node in the data processing process;

[0112] Consider the load of the computing nodes, data locality principle, and task priority factors; adopt a dynamic load balancing strategy to monitor the current load of each computing node (such as CPU usage, memory occupancy, etc.), and distribute sub-datasets to nodes with lower load to avoid processing delays caused by node overload; at the same time, consider the data locality principle and try to distribute associated sub-datasets to adjacent or low-cost communication nodes to reduce data transmission between nodes. For example, if some sub-datasets need to be frequently calculated together in subsequent processing, distribute them to nodes within the same rack; according to the priority of the task, give higher allocation priority to important sub-datasets or urgent processing tasks to ensure that critical data can be processed in time;

[0113] Use data loading mechanism to quickly load the segmented sub-datasets to the corresponding computing nodes; use the parallel reading function of distributed file system (such as HDFS, Ceph, etc.) to read data from the file system simultaneously by multiple computing nodes; for real-time data stream, use message queue (such as Kafka, RabbitMQ, etc.) for data buffering and distribution; the producer sends the real-time generated yarn processing data to the message queue, and the computing node as the consumer subscribes and obtains data from the message queue to realize real-time loading and processing of data;

[0114] Build an independent task running environment on each computing node to ensure that the computing tasks of the data processing model can run stably in an isolated environment; use containerization technology (such as Docker, Kubernetes, etc.) to package the data processing model and its dependent libraries and tools into a container image, and each computing node starts one or more containers to run the corresponding computing tasks; containerization technology provides good environment isolation and portability, which can avoid environment conflicts between different tasks, and facilitates task deployment and management;

[0115] Data exchange protocol is used for data transmission and sharing between computing nodes; the protocol should have low delay and high throughput characteristics to meet the needs of real-time data processing; binary data format is used for data transmission to reduce data size and parsing time; at the same time, data compression is used to further reduce network transmission bandwidth occupancy. For example, use Snappy, LZ4, etc. fast compression algorithm for data compression and decompression;

[0116] Based on the cooperative work strategy between computing nodes, it is ensured that each node can coordinate to complete the data processing task; the master-slave mode or peer-to-peer mode is used for cooperation; in the master-slave mode, a master node is designated to be responsible for the scheduling and coordination of the task, and other slave nodes perform calculation and data exchange according to the instructions of the master node; in the peer-to-peer mode, each node is equal, and task allocation and cooperative calculation are performed through a distributed algorithm; for example, when predicting the quality of yarn, multiple nodes can respectively extract features and preliminarily predict different sub-data sets, and then the results are aggregated to a node for final fusion prediction, improving the accuracy and efficiency of prediction.

[0117] The effect of the above technical solution is that by analyzing the features of each sub-data set and allocating data according to the node hardware configuration, it can be ensured that each sub-data set can be processed on the most suitable node, thereby improving the accuracy and efficiency of the computing task. Through the dynamic load balancing strategy, the load of each computing node is monitored in real time, and the sub-data set can be allocated to the node with lower load, avoiding resource waste and delay caused by node overload, and improving the overall computing efficiency of the system.

[0118] Through high-precision time synchronization protocol (such as NTP protocol) and node state initialization operation, the time consistency of each node in the data processing process is ensured, avoiding data processing errors or inconsistencies caused by time synchronization problems, and improving the stability of the entire system. By using a low-latency, high-throughput data exchange protocol, combined with a fast compression algorithm (such as Snappy, LZ4) for data compression, data transmission time and bandwidth occupancy can be significantly reduced, thereby improving the data transmission efficiency and stability of the system.

[0119] By splitting and quickly loading data to corresponding computing nodes, and using distributed file systems and message queues, real-time data can be quickly loaded and processed, reducing the delay in the data processing process and improving the system response speed. Using containerization technology to isolate the task environment can easily expand the computing nodes, flexibly cope with different scales of computing task demand, and simplify the deployment and management of tasks, improving the scalability of the system.

[0120] According to the data locality principle, associated sub-data sets are allocated to adjacent or low communication cost nodes, which can reduce the communication overhead between nodes and improve the efficiency of data processing. By constructing an independent task running environment and using containerization technology, stable operation of tasks in an isolated environment can be ensured, avoiding environmental conflicts between different tasks, and simplifying system operation and management.

[0121] Adopting master-slave mode or peer-to-peer mode for task cooperation enables each node to be flexibly scheduled and cooperatively calculated according to different task requirements, improves the intelligent level of task scheduling, and ensures the timely completion of key tasks. Through the cooperative working strategy between the computing nodes, whether it is master-slave mode or peer-to-peer mode, the coordination between each node can be ensured, the bottleneck in task allocation and data exchange is reduced, and the overall computing efficiency is improved.

[0122] In one embodiment of the application, the S323 comprises:

[0123] Before loading the segmented sub-data set into the corresponding computing node, each sub-data set is comprehensively and deeply analyzed; based on the analysis result, a loading strategy is tailored for each sub-data set;

[0124] The distributed file system is configured and adjusted, the number of data block replicas and the storage location are set according to the hardware performance and network status of the computing node; and according to the data sharding algorithm, the sub-data set is dynamically divided into more suitable data shards according to the characteristics of the sub-data set and the processing capacity of the computing node; during the parallel reading process, the reading progress and network bandwidth usage of each computing node are monitored in real time, and the allocation of data shards is dynamically adjusted;

[0125] For real-time data streams, message queues are used for data buffering and distribution; and before the data enters the queue, the data is cleaned and converted, invalid data and abnormal values are removed, and the data is uniformly converted into a format that is easy for the computing node to process;

[0126] A mixed data loading mode is constructed; during the calculation process, the loading proportion of historical data and real-time data is dynamically adjusted according to the data processing requirements;

[0127] During the data loading process, a comprehensive monitoring system is established to monitor the speed, accuracy and integrity of data loading in real time; based on the monitoring data, intelligent algorithms are used to dynamically adjust the data loading process.

[0128] The working principle of the above technical solution is that each sub-data set is comprehensively and deeply analyzed before being loaded into the corresponding computing node; in addition to the known data size, data dimension complexity and data type, the distribution characteristics of the data are further analyzed, such as whether the numerical data has skew distribution, whether the category number and distribution of the category data are uniform, etc. At the same time, the association between data is studied to determine whether there are strongly or weakly associated data pairs or data groups; based on the analysis results, a special loading strategy is tailored for each sub-data set; for a numerical sub-data set with a large data size and uniform distribution, considering that it is suitable for parallel processing, a multi-thread parallel reading mode is planned to fully utilize the multi-core resources of the computing node and improve the data loading speed. For category data, if the category number is large and the distribution is uneven, a category block loading strategy is adopted, and the category data with high frequency is loaded first to meet the key computing requirements. For data groups with strong association, they are planned to be loaded together to reduce the data splicing and association operations in the subsequent computing process and improve the overall processing efficiency;

[0129] When using the parallel reading function of the distributed file system (such as HDFS, Ceph, etc.), deep optimization and more efficient data loading are performed; the distributed file system is configured and adjusted according to the hardware performance and network status of the computing node to set the number of data block replicas and storage location; for example, for computing nodes with strong performance, increase the number of data block replicas on nearby storage nodes to reduce network hops during data reading; and according to the data sharding algorithm, the sub-data set is dynamically divided into more suitable data shards according to the characteristics of the sub-data set and the processing capacity of the computing node; these data shards not only consider the size of the data, but also take into account the uniformity and association of the data; during parallel reading, the reading progress and network bandwidth usage of each computing node are monitored in real time, and the allocation of data shards is dynamically adjusted to ensure the load balancing of each computing node and avoid some nodes reading too fast or too slow;

[0130] For real-time data streams, message queues are used for data buffering and distribution. In addition to the traditional method of sending yarn processing data generated by producers to message queues, a smart routing mechanism is introduced. According to the characteristics and priority of real-time data, different routing rules are set for different data. For example, for key quality indicator data, high priority routing is set to ensure that it can be quickly and accurately sent to the corresponding computing node for processing. For general production data, ordinary priority routing is used to reasonably allocate network resources while ensuring data is not lost. Before data enters the queue, data cleaning and format conversion are performed to remove invalid data and outliers, and data is uniformly converted to a format that is easy for computing nodes to process. This can reduce the preparation work of computing nodes when processing data and improve overall processing efficiency. In addition, the persistent storage function of the message queue is used to ensure that real-time data is not lost when the system fails or the network is interrupted, and the processing can continue after the system recovers.

[0131] A hybrid data loading mode is constructed by combining the advantages of distributed file system parallel reading and real-time data stream message queue. For scenarios where both historical data are stored in the distributed file system and real-time data are continuously generated, a phased loading method is used. In the initial stage, historical data are preferentially loaded from the distributed file system in parallel to provide basic data support for computing nodes. The real-time data stream message queue is started to continuously receive and process newly generated real-time data. During the computing process, the loading proportion of historical data and real-time data is dynamically adjusted according to the data processing requirements. For example, when performing real-time quality prediction, the loading proportion of real-time data is appropriately increased to ensure that the prediction results can timely reflect the current production situation. When performing long-term trend analysis, the loading amount of historical data is increased to obtain more comprehensive analysis results. Through this hybrid data loading mode, the advantages of the two data loading methods are fully utilized to meet the data processing requirements in different scenarios.

[0132] During data loading, a comprehensive monitoring system is established to monitor the speed, accuracy and integrity of data loading in real time. By deploying monitoring agents in computing nodes, distributed file systems, message queues and other components, various performance indicator data such as data reading rate, network transmission delay and data loss rate are collected. Based on the monitoring data, intelligent algorithms are used to dynamically adjust the data loading process. When it is found that the data loading speed of a computing node is too slow, the cause is analyzed and appropriate measures are taken, such as adjusting data shard size, optimizing network routing, etc. If data loss or damage is found, it is recovered and reloaded from the source data in time. At the same time, according to the actual situation of data loading, the subsequent data loading strategy is intelligently adjusted to ensure that the entire data loading process is efficient and stable.

[0133] The effect of the above technical solution is that through in-depth analysis of each sub-data set, a tailor-made loading strategy is customized, the loading mode of different types of data is optimized, especially when handling large-scale data, multi-thread parallel reading is adopted, the data loading speed is effectively improved, and the multi-core resources of the computing node are fully utilized. Through the mixed loading mode of the distributed file system and the real-time data stream, the loading proportion of historical data and real-time data is dynamically adjusted, unnecessary resource waste is avoided, and the efficiency of data processing is ensured.

[0134] Through the intelligent routing mechanism and the dynamic data sharding strategy, combined with the distribution characteristics of the data and the hardware performance of the computing nodes, the data can be reasonably distributed, the load balancing in the loading process is improved, and the overload or idleness of some nodes is avoided. Real-time monitoring of the reading progress and network bandwidth usage of each computing node dynamically adjusts the allocation of data shards, ensures the load balancing of each computing node, and avoids the computing bottleneck caused by uneven data distribution or inconsistent processing speed.

[0135] Through cleaning and format conversion of real-time data, invalid data and outliers are removed, data formats are unified, and the interference of data anomalies or inconsistent formats on the processing of computing nodes is reduced, and the data quality is improved. The persistent storage function of the message queue is adopted to ensure that the system will not lose data when a fault or network interruption occurs. After system recovery, data processing can continue from the queue, ensuring the continuity and stability of data processing.

[0136] Through the establishment of a comprehensive monitoring system, the speed, accuracy and integrity of data loading are monitored in real time to ensure that problems can be found and measures can be taken in time to avoid hidden dangers in the data loading process. By dynamically adjusting the number of data block replicas and storage locations according to the hardware performance of the computing nodes and network conditions, the data reading delay caused by network bottlenecks is reduced, and the stability of data transmission is improved.

[0137] The distributed file system parallel reading and intelligent routing mechanism enable the computing nodes to fully utilize storage and network resources, improving the overall performance and response speed of the system. Through flexible loading strategies and distributed system configuration adjustments, this solution can adapt to different scales of production environments, and as the data volume and computing demand increase, the system can be seamlessly expanded to ensure efficient data processing.

[0138] An embodiment of the present application, the S326, comprises:

[0139] The nature of the data processing task is analyzed, and the node resources and capabilities are evaluated;

[0140] According to the task requirements and node capabilities, the advantages and disadvantages of master-slave mode and peer-to-peer mode are considered, and the most suitable collaborative mode or a mixed mode combining the two modes is selected; according to the selected collaborative mode, a collaborative planning scheme is developed;

[0141] In the master-slave mode, a master node is elected from the computing nodes through a weighted election algorithm based on node performance indicators; after the election is completed, the master node is initialized;

[0142] The master node decomposes the entire data processing task into multiple sub-tasks according to the requirements and characteristics of the task; after receiving the sub-tasks allocated by the master node, the slave nodes start the corresponding computing tasks in the local environment; the master node continuously monitors the task execution of each slave node and adjusts the task allocation strategy according to the monitoring results;

[0143] A distributed task allocation algorithm is used to evenly distribute data processing tasks to each node; after receiving the allocated tasks, each node independently performs calculation and processing; during the calculation process, nodes interact and cooperatively calculate data as needed to jointly complete complex data processing tasks;

[0144] A mode switching and fusion strategy is developed to dynamically adjust the collaborative mode according to the different stages and characteristics of the task; comprehensive performance monitoring is performed on the collaborative work in the mixed mode, and the collaborative strategy is optimized and adjusted according to the monitoring results.

[0145] The working principle of the above technical solution is as follows: the nature of the data processing task is analyzed, including task size, complexity, real-time requirement, and data dependency relationship, etc. For example, for large-scale image recognition tasks involving a large amount of image data and complex neural network calculation, high computing power and efficient data transmission are required; for data stream processing tasks in real-time monitoring systems, real-time requirements are extremely high, requiring fast response and processing of newly arrived data; and node resources and capabilities are evaluated, including detailed understanding of the hardware configuration (such as CPU core number, memory capacity, disk I / O performance, network bandwidth, etc.) and software environment (such as operating system, data processing framework version, etc.) of each node in the computing cluster, as well as the historical performance of each node. Through performance testing and monitoring tools, the running data of the nodes under different loads is collected to provide a basis for the selection of collaborative mode.

[0146] According to the task requirements and node capabilities, the advantages and disadvantages of master-slave mode and peer-to-peer mode are considered, and the most suitable collaborative mode is selected or a hybrid mode combining the two modes is adopted; for large and complex tasks with clear task scheduling and coordination requirements, the master-slave mode can provide more centralized management and control; for some distributed computing tasks, the peer-to-peer mode can better exert the autonomy and parallel processing capability of the nodes. For example, in a large-scale machine learning training task, the master-slave mode can be used, with the master node responsible for model parameter updating and task allocation, and the slave nodes responsible for specific model training calculation; at the same time, in the peer-to-peer mode, nodes can exchange and share data quickly, improving training efficiency; according to the selected collaborative mode, a collaborative planning scheme is developed, and the roles, responsibilities and interaction modes of each node in the task execution process are clarified. The planning scheme should include task allocation strategy, data transmission path, communication protocol between nodes, etc., to ensure that each node can clearly understand its own task and collaboration with other nodes;

[0147] In the master-slave mode, a master node is elected from the computing nodes through a weighted election algorithm based on node performance indicators; the master node is responsible for the scheduling, coordination and monitoring of the entire task; after the election is completed, the master node is initialized, loading the task management module, data scheduling module and communication management module, etc., and configuring relevant parameters and strategies;

[0148] The master node decomposes the entire data processing task into multiple subtasks according to the requirements and characteristics of the task. When decomposing the task, the principle of data locality and the computing capacity of the nodes are fully considered, and the associated subtasks are distributed to adjacent or computing capacity matching nodes as much as possible. For example, when performing a large-scale data sorting task, the master node can divide the data into multiple data blocks according to certain rules, and then distribute the sorting task of each data block to different slave nodes. After receiving the subtask distributed by the master node, the slave node starts the corresponding computing task in the local environment. During the task execution process, the slave node monitors the progress and resource usage of the task in real time and feeds back this information to the master node regularly. If the slave node encounters problems or abnormal situations during the execution of the task, such as data errors, insufficient computing resources, etc., it should report to the master node in time for the master node to make corresponding adjustments and processing. The master node continuously monitors the task execution of each slave node, including task progress, resource utilization, error information, etc. According to the monitoring results, the master node can dynamically adjust the task allocation strategy, such as migrating part of the task on the slave node with high load to the node with low load, to ensure the execution efficiency and stability of the entire task. At the same time, the master node is responsible for coordinating the data transmission and sharing between the slave nodes to ensure the accuracy and consistency of the data. In the peer-to-peer mode, all nodes are equal in status, and each node performs initialization operation after starting, loads the necessary collaborative computing module and communication module. Then, the node registers with other nodes in the cluster, shares its basic information such as node ID, hardware configuration, available resources, etc., so that other nodes can understand the overall situation of the cluster.

[0149] Distributed task allocation algorithms, such as consistent hashing-based algorithms or market mechanism-based task allocation algorithms, are used to evenly distribute data processing tasks to each node; these algorithms can automatically achieve reasonable task allocation according to the node's ability and task requirements, avoiding uneven node load. For example, when using a consistent hashing algorithm, the task key is mapped to a ring through a hash function, and then the task is allocated to the nearest node according to the node's location; in peer-to-peer mode, frequent data exchange and sharing between nodes are required to meet the needs of collaborative computing. Efficient data exchange protocols and binary data formats are used for data transmission to reduce data size and parsing time. At the same time, data compression is performed to further reduce network transmission bandwidth occupancy. For example, nodes can exchange data through point-to-point communication or broadcast communication, and select the appropriate communication method according to the characteristics and needs of the data; after receiving the allocated task, each node independently performs calculation and processing; during the calculation process, nodes interact and collaboratively compute data as needed to jointly complete complex data processing tasks; for example, in distributed machine learning training, each node can train the local data model, then fuse the training results of each node through a distributed algorithm (such as the average gradient descent algorithm) to update the global model parameters. Finally, the calculation results of each node are aggregated and fused to obtain the final processing result.

[0150] In some complex data processing tasks, a single collaborative mode may not meet the task requirements, so a hybrid mode can be used to combine master-slave mode and peer-to-peer mode. Develop mode switching and fusion strategies to dynamically adjust the collaborative mode according to the different stages and characteristics of the task; for example, in the initialization stage of the task, the master-slave mode can be used for task decomposition and allocation; in the execution stage of the task, the peer-to-peer mode can be used for parallel computing to improve computing efficiency; in the end stage of the task, switch back to the master-slave mode for result aggregation and integration; in the hybrid mode, due to the switching of multiple collaborative modes and the flow of data between different nodes, data consistency is particularly important. Distributed consensus algorithms (such as Paxos algorithm or Raft algorithm) are used to ensure data accuracy and consistency during task execution. At the same time, a data version management mechanism is established to record and track data modifications and updates to facilitate correct handling in the event of data conflicts; comprehensive performance monitoring of collaborative work in the hybrid mode is performed, including task execution time, node resource usage, data transmission delay, etc.; based on the monitoring results, collaborative strategies are optimized and adjusted, such as optimizing task allocation algorithms, adjusting data transmission paths, and optimizing node communication protocols, to improve the performance and efficiency of the entire computing cluster.

[0151] The effect of the above technical solution is that in the master-slave mode, through reasonable decomposition of tasks and allocation of resources, the computing power of each node can be maximized, and the time of overall task execution can be reduced. Especially in large-scale machine learning training, when the master node allocates tasks to the slave nodes, the processing efficiency can be improved according to data locality and node performance scheduling. In the peer-to-peer mode, nodes can quickly exchange data and collaborative computing, avoiding the bottleneck problem caused by centralized scheduling, and ensuring the efficiency of parallel computing. For example, in distributed machine learning training, data sharing and collaborative computing between nodes can accelerate the update of the global model. The hybrid mode combines the centralized scheduling of the master-slave mode and the parallel processing of the peer-to-peer mode, and selects the appropriate collaboration strategy in different stages of the task, further improving the efficiency and flexibility of task execution.

[0152] By dynamically adjusting the task allocation strategy in the master-slave mode, the tasks of nodes with excessive load can be migrated to nodes with lower load in time, reducing resource waste and improving the overall utilization of cluster resources. In the peer-to-peer mode, the resources of nodes can be more reasonably allocated, and each node can participate in computing according to its computing power and task demand, avoiding excessive dependence on a single master node, thereby avoiding centralized waste of resources. The flexibility of the hybrid mode allows dynamic adjustment of task allocation according to node capabilities in different stages of the task, effectively avoiding resource idling or overload, and improving the resource utilization of the cluster.

[0153] In the master-slave mode, the master node is responsible for the scheduling and monitoring of the entire task, and can discover and handle node failures or exceptions in time during task execution, improving the fault tolerance of the system. If a slave node fails, the master node can reallocate tasks to other nodes to ensure task execution. In the peer-to-peer mode, all nodes are equal, so even if a node fails, other nodes can still execute tasks, avoiding single-point failure of the system. The self-organizing and decentralized characteristics of nodes enhance the reliability of the entire system. In the hybrid mode, the master-slave mode and the peer-to-peer mode can be switched, and when the master node fails, the system can quickly switch to the peer-to-peer mode to ensure that the task is not interrupted.

[0154] In the master-slave mode, the master node ensures the accuracy and consistency of data by continuously monitoring and coordinating the transmission and sharing of data. If the slave node encounters a problem, the master node can adjust the strategy in time to avoid the flow of incorrect data to other nodes. In the peer-to-peer mode, data consistency is ensured through efficient data exchange protocols and distributed consensus algorithms (such as Paxos or Raft algorithms), ensuring data consistency between nodes and avoiding incorrect calculations due to data conflicts. In the hybrid mode, by reasonably designing task scheduling and data management strategies, the consistency and accuracy of data can be ensured when switching between different collaboration modes, thereby reducing problems caused by data conflicts.

[0155] The master-slave mode can increase more slave nodes to improve processing capacity by decomposing tasks into multiple sub-tasks and assigning them to different slave nodes. As the cluster size expands, the master-slave mode can flexibly expand to support more computing tasks. The decentralized nature of the peer-to-peer mode allows each node to independently process tasks, so the system can flexibly add new nodes without considering the bottleneck problem caused by centralized scheduling, and has good horizontal scalability. The hybrid mode can dynamically adjust the collaboration strategy according to the task requirements and node computing capabilities, so that when the cluster size expands, task scheduling can be optimized to improve the scalability of the system.

[0156] Through reasonable decomposition and distribution of tasks, the master-slave mode can dynamically adjust the scheduling strategy according to the characteristics of the task and the resource status of the node, making the task execution more flexible and controllable. The peer-to-peer mode allows each node to self-schedule according to real-time computing requirements and resource status, thereby improving the adaptability of the system and the flexibility of the task. The hybrid mode can flexibly select the scheduling mode according to the different stages of the task, improving the schedulability of the overall task. Especially when dealing with complex distributed computing tasks, it can handle different stages of task requirements by switching collaboration modes.

[0157] In one embodiment of the present application, the S4 comprises:

[0158] S41, based on the updated data processing model, predicting key parameters in the yarn processing process;

[0159] S42, according to the model prediction result, using a genetic algorithm to optimize the key parameters;

[0160] S43, generating yarn processing equipment control instructions according to the optimized key parameters; converting the control instructions into digital signals or analog signals that can be recognized by the equipment, and sending them to the corresponding equipment through the communication interface.

[0161] The working principle of the above technical solution is: based on the updated data processing model, the key parameters in the yarn processing process are predicted; for example, the model is used to predict the change of yarn tension, according to the input current yarn tension data, spinning speed data and environment temperature and humidity data, the model outputs the change trend of yarn tension in the future period of time; the drying temperature and time of the yarn and other parameters are predicted. Combined with the environmental temperature and humidity data and the model analysis result, the optimal parameters of yarn drying under different environmental conditions are considered, which provides a scientific basis for the parameter adjustment of the equipment;

[0162] According to the prediction result of the model, the genetic algorithm is used to optimize the key parameters; taking yarn tension control as an example, the target range of yarn tension is set, and the tension control parameters of the spinning equipment are adjusted through the optimization algorithm, so that the predicted yarn tension is as close as possible to the target range; for the optimization of yarn drying temperature and time, considering the factors such as the material and moisture content of the yarn, combined with the prediction result of the model, the optimal parameter combination is found to improve the drying efficiency and reduce the energy consumption under the premise of ensuring the quality of the yarn;

[0163] According to the optimized key parameters, the yarn processing equipment control instruction is generated; the control instruction is converted into digital signal or analog signal which can be recognized by the equipment, and is sent to the corresponding equipment through the communication interface. For example, the tension control parameter is converted into the adjustment instruction of motor speed, which is sent to the motor controller of the spinning equipment, so as to ensure the operation of the equipment under the best parameters and improve the quality of the yarn.

[0164] The effect of the above technical solution is: through the sub data set allocation based on node hardware configuration and feature analysis, the processing accuracy of each sub data set is improved, and the accuracy and efficiency of the calculation task are ensured. Through the dynamic load balancing mechanism, the load of the calculation node is monitored and adjusted in real time, the resource waste caused by node overload is reduced, and the calculation efficiency of the system is improved.

[0165] High-precision time synchronization protocol (such as NTP) and node initialization operation ensure the consistency between all nodes of the system, avoid data processing errors caused by different time synchronization, and enhance the stability of the system. Low delay, high throughput transmission protocol and compression algorithm are used to reduce data transmission time and bandwidth occupation, improve data transmission efficiency and stability of the system.

[0166] Through the combination of data segmentation, fast loading to nodes, distributed file system and message queue technology, the delay in data processing process is reduced, and faster response speed is ensured. The application of containerization technology improves the scalability of the system, allows the calculation node to be quickly increased when the demand changes, and simplifies the task deployment and management.

[0167] By assigning relevant sub-datasets to nodes with low communication costs, the communication overhead between nodes is reduced, thereby improving the efficiency of data processing. Through the independence of the task environment and the use of containerization technology, the operation and maintenance of the system is simplified, task environment conflicts are avoided, and system management is more efficient.

[0168] The application of task coordination mode (master-slave mode or peer-to-peer mode) improves the intelligence of system scheduling, ensures that each node can flexibly cooperate according to task requirements, and improves the efficiency of task completion. Through the cooperative work strategy between computing nodes, the bottleneck in task allocation and data exchange is reduced, and the overall computing efficiency is improved.

[0169] In one embodiment of the present application, the S5 comprises:

[0170] S51, obtaining yarn processing effect data after running according to the control instruction, measuring various quality indicators of the yarn by using a yarn quality detection device;

[0171] S52, recording the running time and production of the device and other data through the production management system to calculate the production efficiency index; and arranging and storing the obtained processing effect data;

[0172] S53, feeding the effect data into the data processing model to evaluate the optimization effect of the model; using multiple evaluation indexes to calculate the error between the model prediction value and the actual measured value;

[0173] S54, drawing an error distribution graph to intuitively show the prediction error of the model under different working conditions; according to the evaluation result, further adjusting the parameters and processing strategy of the model;

[0174] S55, according to the quality requirements and production plan of different yarn products, adaptively allocating the communication frequency band resources of the yarn processing equipment; transmitting the allocated wireless communication frequency band resource data to the frequency band control platform of the yarn processing equipment, and the frequency band control platform adjusts and configures the communication frequency band of the equipment according to the received data to perform the frequency band resource allocation task.

[0175] The working principle of the above technical solution is: obtaining yarn processing effect data after running according to the control instruction, including yarn strength, uniformity, hairiness and other quality index data and production efficiency data; using yarn quality detection equipment such as strength tester, Uster yarn evenness tester, etc. to measure various quality indicators of the yarn;

[0176] Through the production management system, the running time and production of the device and other data are recorded to calculate the production efficiency index; and the obtained processing effect data is arranged and stored;

[0177] The effect data is fed back to the data processing model to evaluate the optimization effect of the model; various evaluation indexes such as mean square error (MSE), mean absolute error (MAE), etc. are used to calculate the error between the model prediction value and the actual measurement value;

[0178] An error distribution diagram is drawn to visually show the prediction error of the model under different working conditions; according to the evaluation results, the parameters and processing strategies of the model are further adjusted, such as adjusting the number of hidden layer neurons, learning rate, etc. of the model, or adding new feature extraction methods, to realize closed-loop control and continuously improve the performance and prediction accuracy of the model;

[0179] According to the quality requirements and production plans of different yarn products, the communication frequency band resources of the yarn processing equipment are adaptively allocated; for example, for high-quality yarn products, due to the high requirements for equipment collaboration and data transmission stability during production, more communication resources are allocated, such as using higher bandwidth communication frequency bands or increasing the number of communication channels, to ensure efficient collaboration and stable data transmission between devices; for yarn products with urgent production plans, communication resources are preferentially allocated to ensure the smooth progress of the production process; the allocated wireless communication frequency band resource data is transmitted to the frequency band control platform of the yarn processing equipment, and the frequency band control platform adjusts and configures the communication frequency bands of the equipment according to the received data to perform frequency band resource allocation tasks and ensure the smooth progress of the entire yarn processing production process.

[0180] The effect of the above technical solution is that by obtaining the quality data of the yarn after processing in real time, the strength, evenness, hairiness, etc. of the yarn can be comprehensively monitored to ensure that the yarn quality is within a controllable range. By feeding the yarn quality data back to the data processing model and using various evaluation indexes to calculate the prediction error of the model, prediction bias can be discovered and corrected in a timely manner, thereby reducing quality fluctuations during production.

[0181] By recording and calculating the device running time and yield data in the production management system, the production efficiency monitoring and adjustment are optimized, so that the production efficiency can be effectively improved and low-efficiency links can be adjusted in a timely manner. Using a closed-loop control mode, the prediction accuracy of the data processing model under different working conditions is gradually improved through continuous optimization of the model (such as adjusting parameters, adding feature extraction methods, etc.), which reduces bias and improves prediction accuracy.

[0182] According to the quality requirements and the urgency of the production plans of the yarn products, the communication frequency band resources of the devices are adaptively allocated, the configuration of the communication resources is optimized, and the production bottlenecks caused by insufficient or wasted resources are reduced. For high-quality yarn products, by allocating higher bandwidth communication frequency bands or increasing the number of communication channels, the efficient collaboration and data transmission stability between devices are ensured, thereby improving the overall production quality.

[0183] Through the cooperation of dynamic allocation of wireless communication frequency band resources and frequency band control platform, the resource allocation in the production process can be adjusted in real time, ensuring the smooth progress of the yarn production plan and quickly responding to the sudden demand in the production process. Adaptive allocation of communication frequency band resources can effectively avoid production stagnation caused by communication interruption or resource shortage, reducing the risk of system failure and production interruption. Through comprehensive data collection, processing and feedback, intelligent management of the yarn processing process is realized, which not only improves the product quality, but also optimizes the resource allocation, making the production process more efficient and precise. Through adaptive resource allocation for different yarn products, the production plan can be flexibly adjusted according to the actual situation to meet different production requirements, ensuring smooth and efficient production process.

[0184] One embodiment of the present application is a data control system for a yarn processing device, comprising:

[0185] one or more processors;

[0186] a memory for storing one or more programs,

[0187] wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the method of any one of the above.

[0188] Obviously, those skilled in the art can make various modifications and variations to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalent technologies, the present application also intends to include these modifications and variations.

Claims

1. A data control method for yarn processing equipment, characterized in that, The method includes: S1. Acquire multi-source data during the operation of the yarn processing equipment, preprocess the multi-source data, and obtain a preliminary integrated yarn processing dataset; S2. Construct a data processing model with a multi-layer nonlinear mapping structure based on neural networks; input the initially integrated yarn processing dataset into the model, and use the model to automatically extract multi-dimensional features from the massive data; S3. A distributed parallel computing framework is adopted to divide the large-scale yarn processing data into multiple subsets and distribute them to different computing nodes for parallel processing. During the data processing, the performance indicators of the model are continuously monitored. When the model performance is detected to be degraded or the processing conditions are changed, the model is updated online using newly collected real-time data. S4. Based on the updated data processing model, optimize key parameters in the yarn processing process and generate precise control instructions for yarn processing equipment. S5. Obtain yarn processing effect data after running according to control instructions, feed the effect data back to the data processing model, evaluate the optimization effect of the model; based on the evaluation results, further adjust the model parameters and processing strategy to achieve closed-loop control. The S5 includes: S51. Obtain yarn processing effect data after running according to control instructions, and use yarn quality testing equipment to measure various quality indicators of the yarn; S52. Record equipment operating time and output data through the production management system, calculate production efficiency indicators, and organize and store the obtained processing effect data; S53. Feed the effect data back into the data processing model to evaluate the optimization effect of the model; use multiple evaluation indicators to calculate the error between the model's predicted value and the actual measured value. S54. Draw an error distribution diagram to visually display the prediction error of the model under different working conditions; based on the evaluation results, further adjust the model parameters and processing strategies. S55. Based on the quality requirements and production plans of different yarn products, adaptively allocate the communication frequency band resources of the yarn processing equipment; transmit the allocated wireless communication frequency band resources to the frequency band control platform of the yarn processing equipment; the frequency band control platform adjusts and configures the communication frequency band of the equipment according to the received data, and executes the frequency band resource allocation task.

2. The data control method for a yarn processing equipment according to claim 1, characterized in that, S1 includes: S11. Collect data from multiple sources; S12. Conduct a preliminary inspection of the collected multi-source data to identify and remove noisy data and outliers; S13. Align the different types of cleaned data according to a unified timestamp, convert the data into a unified digital format, and normalize the data to obtain a preliminary integrated yarn processing dataset.

3. The data control method for a yarn processing equipment according to claim 1, characterized in that, The S2 includes: S21. A data processing model based on a multi-layer nonlinear mapping structure constructed using neural networks; S22. Input the initially integrated yarn processing dataset into the constructed model, and use the hidden layer of the model to automatically extract multidimensional features from the massive data; use principal component analysis to perform dimensionality reduction on the extracted multidimensional features and remove redundant information between features. S23. Combining gradient descent strategy, continuously adjust the connection weights between neurons during model training; use mini-batch gradient descent method to divide the training data into multiple mini-batches, and use a mini-batch of data to update the weights each time. S24. Set the learning rate and momentum parameters, and select the optimal combination of learning rate and momentum parameters through cross-validation to improve the generalization ability of the model.

4. The data control method for a yarn processing equipment according to claim 3, characterized in that, S23 includes: The yarn processing dataset, which has been preliminarily integrated and processed by feature extraction and dimensionality reduction, is analyzed. The batch size is determined based on the scale and distribution characteristics of the data. The dataset is then reordered using a random shuffling method. Based on the complexity of the model and the characteristics of the data, a predetermined initial number of training rounds is set; during the training process, an early stopping mechanism is implemented; by monitoring the performance metrics on the validation set, training is stopped early when the performance on the validation set no longer improves after several consecutive rounds. A data loading library is used to load small batches of data. During the data loading process, the small batches of data are preprocessed and then input into the constructed neural network model. The calculations are performed sequentially in the order of input layer, hidden layer, and output layer. Select the loss function according to the application requirements; calculate the error gradient of each neuron using the backpropagation algorithm based on the calculated loss value; Based on the calculated gradient values ​​and learning rate, the weights and biases of each neuron are updated according to the rules of gradient descent. Combined with the momentum parameter, the gradient descent method with momentum is used to accelerate the gradient descent process and avoid getting trapped in local optima. After each training round, the performance of the model is evaluated using a validation set.

5. The data control method for a yarn processing equipment according to claim 1, characterized in that, The S3 includes: S31. A distributed parallel computing framework is adopted to divide the large-scale yarn processing data into multiple subsets; the size and number of subsets are determined according to the number of nodes in the computing cluster and the data scale. S32. Distribute the segmented subsets of data to different computing nodes for parallel processing; each computing node independently runs the computational tasks of the data processing model and exchanges data and collaborates with each node through network communication. S33. During data processing, continuously monitor the model's performance metrics; through the performance monitoring module, periodically calculate the model's prediction accuracy and error metrics on the test dataset and record the results; when a decline in model performance or a significant change in processing conditions is detected, trigger the model update mechanism. S34. Update the model online using newly acquired real-time data; use the backpropagation algorithm to adjust the connection weights of neurons in reverse according to the output error of the model; during the backpropagation process, calculate the error gradient of each neuron and update the weight parameters in the direction of gradient descent.

6. The data control method for a yarn processing equipment according to claim 5, characterized in that, S32 includes: S321. Analyze the characteristics of each subset of data. Based on these characteristics, combined with the hardware configuration and historical performance of each node in the computing cluster, pre-allocate the most suitable computing node for each subset of data. Before allocating the subset of data to the computing node, perform state initialization operations on each computing node. At the same time, use a high-precision time synchronization protocol to synchronize the time of all computing nodes. S322. A dynamic load balancing strategy is adopted to monitor the current load of each computing node in real time and distribute the subset of data to nodes with lower load. S323. A data loading mechanism is adopted to load the segmented subsets of data onto the corresponding computing nodes; the parallel reading function of the distributed file system is used to enable multiple computing nodes to read data from the file system simultaneously; for real-time data streams, message queues are used for data buffering and distribution. S324. Build an independent task execution environment on each computing node, and use containerization technology to package the data processing model and its dependent libraries and tools into container images. Each computing node starts one or more containers to run the corresponding computing tasks. S325. Data transmission and sharing between computing nodes are achieved through a data exchange protocol; binary data format is used for data transmission, and the data is compressed simultaneously. S326. Based on the collaborative working strategy between computing nodes, a master-slave mode or a peer-to-peer mode is adopted for collaboration. In the master-slave mode, a master node is designated to be responsible for task scheduling and coordination, and other slave nodes perform calculations and data exchange according to the instructions of the master node. In the peer-to-peer mode, all nodes have equal status and task allocation and collaborative computing are performed through distributed algorithms.

7. The data control method for a yarn processing equipment according to claim 6, characterized in that, S323 includes: Before loading the segmented subsets of data into the corresponding computing nodes, each subset is thoroughly and deeply analyzed; based on the analysis results, a loading strategy is tailored for each subset. The distributed file system is configured and adjusted according to the hardware performance and network conditions of the computing nodes to set the number of data block replicas and storage locations; and according to the data sharding algorithm, the subset is dynamically divided into more suitable data shards based on the characteristics of the subset and the processing capabilities of the computing nodes; during parallel reading, the reading progress and network bandwidth usage of each computing node are monitored in real time, and the allocation of data shards is dynamically adjusted. For real-time data streams, message queues are used for data buffering and distribution; and before the data enters the queue, the data is cleaned and format converted to remove invalid data and outliers, and the data is uniformly converted into a format that is easy for computing nodes to process. Construct a hybrid data loading mode; during the calculation process, dynamically adjust the loading ratio of historical data and real-time data according to the data processing needs; During the data loading process, a comprehensive monitoring system is established to monitor the speed, accuracy, and completeness of data loading in real time; based on the monitoring data, intelligent algorithms are used to dynamically adjust the data loading process.

8. The data control method for a yarn processing equipment according to claim 1, characterized in that, The S4 includes: S41. Based on the updated data processing model, predict key parameters in the yarn processing process; S42. Based on the model prediction results, the key parameters are optimized using a genetic algorithm; S43. Based on the optimized key parameters, generate control instructions for the yarn processing equipment; convert the control instructions into digital or analog signals that the equipment can recognize, and send them to the corresponding equipment through the communication interface.

9. A data control system for yarn processing equipment, comprising: One or more processors; Memory, used to store one or more programs. Wherein, when the one or more programs are executed by the one or more processors, the one or more processors implement the method of any one of claims 1 to 8.

Images