A spinning quality prediction method based on an attention-gru model

Abstract

The application provides a spinning quality prediction method based on an Attention-GRU model, and belongs to the technical field of textiles. The prediction method comprises the following steps: determining N cotton performance indexes, converting the N cotton performance indexes into N-dimensional feature vectors; determining l processing procedures, and determining l+1 N-dimensional feature vectors as input parameters according to the processing sequence; building a single-layer GRU neural network structure, fully learning the input parameters through the GRU neural network structure to capture the dependency relationship in the time series data; establishing an Attention layer, analyzing the importance of the N cotton performance indexes at different times according to the weight size, and constantly updating and iterating the optimal weight parameter; connecting the Attention layer output to a fully connected neural network to form an Attention-GRU spinning quality prediction model, aggregating information through the output end of the fully connected network, and outputting the predicted value of the yarn quality. The prediction method provided by the application is accurate and time-saving.

Description

Technical Field

[0001] This application belongs to the field of textile technology, specifically relating to a method for predicting spinning quality based on the Attention-GRU model. Background Technology

[0002] As a pre-weaving process, spinning directly determines the efficiency of warp and weft yarns, ultimately affecting the quality of the finished fabric. Therefore, it's necessary to determine cotton blending schemes and processing parameters based on yarn quality indicators. However, current spinning workshops are order-driven, planning specific production processes according to order demands. With the diversification of order varieties and processing parameters, the previous method of determining production plans through extensive trial spinning is no longer sufficient to meet the processing needs of enterprises. Furthermore, due to the diversity of raw materials and the multi-stage nature of the process, relying solely on the personal experience of technical personnel will result in significant uncertainty in yarn quality prediction.

[0003] In response to this situation, scholars both domestically and internationally have proposed a series of data-driven yarn quality prediction methods. For example, optimized support vector machine models are used for small-sample spinning process quality prediction. While these models have strong generalization ability, they are sensitive to parameter and kernel function selection. Alternatively, multiple linear regression models are established for cotton lap / sliver and yarn strength and evenness CV quality indices, effectively reducing raw cotton waste. However, these models are not well-suited for complex nonlinear relationships. Another approach is to study the influence of chemical fiber spinning process parameters on winding tension based on grey prediction models, providing a suitable method for spinning tension prediction, but the final prediction accuracy remains insufficient. With the rapid development of deep learning theory, its advantages in areas such as data nonlinear computation are gradually becoming apparent. Researchers are continuously optimizing and widely applying BP neural networks to yarn quality prediction. Some scholars have proposed a four-layer BP neural network with two hidden layers for cotton yarn quality prediction, which improves upon three-layer networks in terms of training steps and average error. Other scholars have optimized the optimization process of BP neural network weights and thresholds using bio-inspired algorithms such as genetic algorithms and firework algorithms, thereby improving the optimization speed and accuracy of yarn quality prediction models. However, the spinning process is a time-series task characterized by close connections between upstream and downstream processes. Most prediction simulations based on backpropagation neural networks primarily rely on the physical properties of the raw cotton as input, rarely considering process parameters. Even when process parameters are included, the impact of the temporal sequence of processing steps on yarn quality cannot be adequately accounted for. In summary, existing spinning quality prediction methods suffer from limited accuracy and computationally intensive processing, limiting their widespread application in cotton yarn production. Summary of the Invention

[0004] The purpose of this application is to provide a spinning quality prediction method based on the Attention-GRU model, so as to solve the problem that existing spinning quality detection methods cannot fully consider the processing steps, resulting in limited accuracy and long processing time.

[0005] To solve the above-mentioned technical problems, this application is implemented as follows:

[0006] Firstly, this application provides a method for predicting spinning quality based on an Attention-GRU model, the prediction method comprising:

[0007] Determine N raw cotton performance indicators and convert the N raw cotton performance indicators into an N-dimensional feature vector;

[0008] Determine l processing steps, and determine l+1 N-dimensional feature vectors as input parameters according to the processing sequence;

[0009] A single-layer GRU neural network structure is constructed, and the input parameters are fully learned through the GRU neural network structure to capture the dependencies in the time series data;

[0010] An Attention layer is established to analyze the importance of the N raw cotton performance indicators at different times based on the weight size, and the optimal weight parameters are continuously updated and iterated.

[0011] The output of the Attention layer is connected to a fully connected neural network to form an Attention-GRU yarn quality prediction model. Information is aggregated through the output of the fully connected network to output the predicted value of yarn quality.

[0012] This application addresses data such as raw material indicators, process parameters, and yarn quality in actual cotton spinning production. It establishes a spinning quality prediction model that considers the timing of processing using a GRU neural network improved by the Attention mechanism. This improves the problem of low prediction accuracy caused by severe timing issues, helps production personnel to more efficiently determine cotton blending schemes and processing parameters, and can improve enterprise production efficiency. Attached Figure Description

[0013] Figure 1 This is a schematic diagram of a spinning quality prediction method based on the Attention-GRU model provided in an embodiment of this application;

[0014] Figure 2 This is a schematic diagram of the process flow for producing cotton textiles in related technologies;

[0015] Figure 3 This is a schematic diagram of the GRU unit structure provided in an embodiment of this application;

[0016] Figure 4 This is a schematic diagram of the Attention mechanism structure provided in the embodiments of this application;

[0017] Figure 5 This is a schematic diagram of the spinning quality prediction model structure based on Attention-GRU provided in the embodiments of this application;

[0018] Figure 6 This is a comparison chart showing the correlation analysis between the Attention-GRU spinning quality prediction model provided in this application and other models for spinning quality prediction.

[0019] The realization of the objective, functional characteristics and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0021] First, a brief introduction to cotton spinning technology will be given. For example... Figure 2 The diagram shown illustrates the process flow for cotton spinning in related technologies. Cotton spinning is the process of processing cotton fibers into cotton yarn and thread. It is divided into two processes: carding and combing. The main processing steps include carding, drawing, roving, and spinning.

[0022] Based on the characteristics of process manufacturing, it is known that the results of the preceding processes in the spinning production process will affect the subsequent processes. Taking the relationship between the final drawing, roving, and spinning as an example, the weight unevenness and evenness CV value of the spinning yarn determine whether the finished cotton fabric will have weft stripes or warp stripes. Both are affected by the internal fiber structure of the roving. Similarly, to improve the structural rationality of the roving, it is necessary to start with the fiber straightness of the final drawing. In addition, in carding, the carded cotton is made into slivers and then directly used for drawing, while in combing, the carded slivers are pre-drawn, lapped, and then further combed by a combing machine. The yarns from these two processes differ in texture, durability, and evenness. Therefore, when predicting spinning quality, it is necessary not only to consider the performance indicators of the raw cotton and simple process parameters, but also to conduct a time-series analysis of the impact of the processing sequence and preceding processing conditions on the yarn quality.

[0023] The following, in conjunction with the accompanying drawings, provides a detailed description of a spinning quality prediction method based on the Attention-GRU model provided in this application, through specific embodiments and application scenarios.

[0024] Reference Figure 1 The diagram shows a flowchart of a spinning quality prediction method based on the Attention-GRU model provided in an embodiment of this application.

[0025] This application provides a method for predicting spinning quality based on the Attention-GRU model, which includes:

[0026] S101: Determine N raw cotton performance indicators and convert the N raw cotton performance indicators into an N-dimensional feature vector;

[0027] This application's embodiments utilize a gated recurrent unit (GRU) neural network to predict spinning quality. In the prior art, Long Short-Term Memory (LSTM) networks, which effectively improve the gradient vanishing defect of RNN models, are the mainstream method for text processing and time-series prediction. However, the complex structure of LSTM significantly increases the training time of the neural network and reduces its operating efficiency. The GRU neural network used in this application's embodiments effectively improves the model training speed while reducing overfitting and other phenomena compared to LSTM. Figure 3 The above is a schematic diagram of the cell structure of GRU. GRU contains only two gates: update gate and reset gate. The information propagation process inside its cell is shown in Equation (1).

[0028]

[0029] In the formula: x t Input at time t; h t-1 Let r be the hidden layer state at time t-1; σ be the sigmoid activation function; ... t For resetting the gate; ⊙ indicates element-wise multiplication; For candidate information at time t; z t For updating the gate; h t Let W be the hidden layer state at time t; rx W rh W hx W hh W zz W zh The weight parameters of the unit.

[0030] Therefore, in a GRU neural network, the reset gate selectively ignores the output information from the previous time step based on the current input, while the update gate determines the extent to which the output information from the previous time step is written into the current output information. In this way, GRU continuously forgets unimportant historical information and remembers important new information to better capture dependencies in time-series data.

[0031] In this application, considering that the raw cotton performance indicators and process parameters of each process are actually independent time series, in order to couple these feature information that affect cotton yarn quality, this application refers to the word vector method in natural language processing. Assuming there are N raw cotton performance indicators, they are converted into N-dimensional feature vectors. This is used as the first input. Since the GRU neural network is actually a sequence of the same network at different times, the input dimension must be the same each time. However, the number of key process parameters varies in each step of the spinning process and is often less than N. Therefore, 0 is used to pad to N bits. To meet the input requirements.

[0032] S102: Determine l processing steps, and determine l+1 N-dimensional feature vectors as input parameters according to the processing order;

[0033] The processing parameters are input sequentially according to the processing order. The length of the final model's input sequence is l+1 (the number of processes is l, plus the input of the raw cotton feature vector), and the dimension of the input vector is N.

[0034] In addition, to facilitate model training, the min-max normalization method is used to normalize the original input parameters to the range (-1, 1), and the calculation formula is Equation (2).

[0035]

[0036] In the formula: x is the original input data, x max x min ...

[0037] S103: Construct a single-layer GRU neural network structure, and fully learn the input parameters through the GRU neural network structure to capture the dependencies in the time series data;

[0038] In this step, a single-layer GRU neural network structure is built to fully learn the input feature information in order to capture the dependencies in the time-series data. Since the yarn quality prediction of this model uses data from the past l+1 time steps to predict the data for the next time step, decoding only requires one operation. The output of the GRU layer includes the encoder's hidden state vector H(h1,h2,…,hn) and the decoder's first hidden state h′1.

[0039] S104: Establish an Attention layer to analyze the importance of the N raw cotton performance indicators at different times based on the weight size, and continuously update and iterate to find the optimal weight parameters;

[0040] In practice, yarn quality varies constantly due to various factors. The performance indicators of raw cotton and the degree of influence of different processing steps on yarn quality often differ. GRU neural networks fail to distinguish these feature inputs, making it difficult to capture key input information. Attention mechanisms, which simulate the resource allocation mechanism of the human brain, concentrate attention on key locations, thereby reducing attention to non-key locations. Attention mechanisms are generally used to improve SeqtoSeq models. SeqtoSeq, proposed in machine translation, involves concatenating two RNNs (LSTM, GRU). The input RNN is called the encoder, which converts the input sequence into hidden states before feeding them into another RNN, the decoder, thus mapping to a variable-length output sequence. This encoder-decoder training method not only solves the problem of fixed input and output lengths in traditional tasks but also further improves training efficiency. Figure 4 The SeqtoSeq model with the Attention mechanism is shown. The hidden states in the encoder and the hidden states of a certain unit in the decoder are correlated to obtain the weight values. The hidden states of each hidden layer in the encoder are then weighted and summed. Finally, the sum is concatenated with the hidden state of the unit to obtain the predicted output with the Attention mechanism.

[0041] Specifically, the Attention layer analyzes the importance of feature information at different times based on the weight values ​​and continuously updates and iterates to find the optimal weight parameters. The calculation formula of the Attention mechanism is shown in Equation (3).

[0042]

[0043] Where: h t Let H be the hidden state at time t, where t = 1, 2, ..., n, and scroe be the similarity function used to calculate h. t The cosine similarity score e between h′1 and h′1 t ;softmax is a normalized exponential function, where e t The weights a converted to each hidden state t ; contact is the concatenation function; s is the predicted output with the Attention mechanism added.

[0044] This application embodiment uses an attention mechanism to assign weights to the hidden layer states of the GRU neural network, thereby enabling the model to adaptively focus on the input features that have the most significant impact on yarn quality and improve the model's prediction accuracy.

[0045] S105: Connect the output of the Attention layer to a fully connected neural network to form an Attention-GRU yarn quality prediction model. Aggregate information through the output of the fully connected network to output the predicted value of yarn quality.

[0046] The Attention-GRU spinning quality prediction method provided in this application uses GRU neural network and Attention mechanism for modeling. Since GRU training time is relatively short and Attention mechanism can effectively highlight key information among the factors affecting yarn quality, it can reduce computation time while ensuring prediction accuracy.

[0047] Furthermore, after establishing the Attention-GRU spinning quality prediction model, the loss function can be further used to optimize the model output. The loss is calculated against the actual quality y, and the Adam optimizer is selected to optimize the model parameters. Adam designs independent adaptive learning rates for different parameters by calculating the first and second moment estimates of the gradient, thereby iteratively updating the weights of the neural network based on the training data, so that the output value of the loss function reaches the optimum. The loss function of the model adopts the mean squared error (MSE) algorithm, as shown in Equation (4).

[0048]

[0049] In the formula: n is the number of samples, and y i E represents the predicted value and the actual value of the i-th sample, respectively. mse The extent of the loss.

[0050] After optimizing the model parameters using the Adam optimizer, the optimized parameters can be fed into the Attention-GRU spinning quality prediction model to further optimize the model, so that the predicted value output by the model in the next spinning quality prediction is closer to the actual value, thereby improving the prediction accuracy.

[0051] Furthermore, to verify the effectiveness of the prediction model, this application selects spinning quality data from a cotton textile enterprise in Shijiazhuang City and uses the PyTorch library in Python to build an Attention-GRU model for experimental verification. The prediction results are then compared with those of GRU, LSTM, and BP neural networks without the Attention mechanism. Additionally, two loss functions representing regression error are selected for quantitative evaluation of the model: Mean Relative Error (MAPE) and Root Mean Square Error (RMSE). Smaller values ​​for both indicate more accurate quality predictions. The formulas are as follows:

[0052]

[0053]

[0054] Considering the differences in spinning methods, processes, and spinning equipment for different yarn varieties, yarn variety C27.8 is selected as the experimental subject for this section. The process route for this variety is pure cotton carding, and the main spinning equipment involved is: TC5-1 carding machine, FA306 drawing frame, RSB-D45c drawing frame, FA468E roving frame, and JWF1516JM spinning frame.

[0055] The model selects yarn breaking strength as the output parameter, which is an important indicator of yarn quality and directly determines the yarn's processing performance and final application. In cotton spinning, the main factors affecting yarn strength are fiber properties and yarn structure. Higher fiber strength, finer fineness, and longer length result in higher yarn strength. Yarn strength decreases when folds, buckles, or hooks occur. Both of these factors largely depend on the process parameters of each step in the spinning process. Therefore, taking the carded yarn process as an example, this analysis examines the impact of each step on yarn strength:

[0056] 1) The speed of the licker-in roller and cylinder of the carding machine is one of the main factors affecting the carding quality. Increasing the speed can effectively increase the carding speed and area, thereby reducing neps and impurities. However, increasing the speed will exacerbate the increase in the short fiber content, especially in cotton yarn with fibers shorter than 16 mm. Short fiber content For every 1% increase, the strength of the cotton yarn decreases by 1-2%.

[0057] 2) Drawing has the function of drawing and stretching cotton fibers and mixing them evenly. The spinning speed of the drawing frame can affect the yarn strength by influencing the uniformity of cotton fiber mixing.

[0058] 3) The twist coefficients in the roving and spinning processes have a parabolic relationship with the breaking strength of cotton yarn. As the twist coefficient increases, the cohesion between cotton fiber molecules increases, but at the same time, due to the twisting effect, the axial force decreases, causing the fibers to break at different times. In addition, the spindle speed in both processes is also an important factor affecting yarn strength.

[0059] Therefore, in selecting input indicators, six parameters, including micronaire value, were used as raw cotton performance indicators, while the collected key spinning equipment information was used as process parameter indicators for each process. The training samples for the network model were selected based on the indicators (all cotton yarns were spun in the same production environment). Table 1 shows some of the sample data used in this section.

[0060] Table 1 Partial Sample Data

[0061]

[0062] To effectively compare prediction results, the three neural networks, BP, LSTM, and GRU, need to maintain consistency in the selection of basic parameters. However, as a feedforward neural network, the dimension and size of its input values ​​differ from those of recurrent neural networks, as shown in Table 2.

[0063] Table 2 Comparison of Model Parameter Settings

[0064]

[0065]

[0066] The trained different spinning quality prediction models were used to predict 10 sets of test samples, and the results are shown in Table 3. The data in the table shows that the RMSE of the GRU model with the Attention mechanism was reduced by 53.3%, 38.5%, and 36.4% respectively compared to the previous three methods, while the MAPE was reduced by 0.355%, 0.158%, and 0.190% respectively. Furthermore, the runtime of the four training models was recorded during training. The comparison revealed that GRU improved training time by 14.3% compared to LSTM. Therefore, although the prediction performance is similar, the GRU neural network still has the advantage in time efficiency.

[0067] Table 3 Comparison of Model Prediction Results

[0068]

[0069] like Figure 6The figure shown is a correlation analysis comparison of the Attention-GRU spinning quality prediction model provided in this application with other models. To more clearly and intuitively present the quality prediction results, a correlation analysis was performed on the final data of the four models, and the goodness-of-fit R-squared was introduced. 2 This is used to measure the closeness between the actual quality data and the model's predictions. The goodness of fit of the four models in the figure are 0.831, 0.903, 0.905, and 0.954, respectively. The experimental results show that the Attention-GRU model has a higher goodness of fit between the predicted and actual values ​​in terms of yarn quality prediction than the GRU, LSTM, and BP neural network prediction models that do not introduce the Attention mechanism.

[0070] This application provides a method for predicting yarn quality based on an Attention-GRU model. The method involves determining N raw cotton performance indicators and converting them into N-dimensional feature vectors; determining l processing steps and using l+1 N-dimensional feature vectors as input parameters according to the processing order; constructing a single-layer GRU neural network structure to fully learn the input parameters and capture dependencies in time-series data; establishing an Attention layer to analyze the importance of the N raw cotton performance indicators at different times based on weights and continuously updating and iterating to find the optimal weight parameters; connecting the output of the Attention layer to a fully connected neural network to form an Attention-GRU yarn quality prediction model; and aggregating information through the output of the fully connected network to output the predicted yarn quality value. Furthermore, this application further optimizes the model parameters using the Adam optimizer, resulting in a more accurate Attention-GRU yarn quality prediction model. This reduces computational load and improves prediction efficiency while maintaining prediction accuracy, thus improving business efficiency.

[0071] This paper addresses the impact of the temporal sequence of processing steps on cotton yarn quality. It proposes a GRU neural network model improved with an attention mechanism for predicting the breaking strength of cotton yarn. Comparative analysis using actual sample data is conducted. Compared to BP neural networks, GRU and LSTM neural networks can effectively capture the temporal sequence of the spinning process and the influence of process parameters on yarn quality. Furthermore, GRU reduces training parameters while maintaining predictive performance compared to LSTM, thus requiring less training time. In addition, the attention mechanism effectively highlights key information among the factors affecting yarn quality. Overall, the prediction performance of the Attention-GRU model is superior to that of GRU, LSTM, and BP neural networks without the attention mechanism.

[0072] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of an instruction, which contains one or more executable instructions for implementing a specified logical function. In some alternative implementations, the functions marked in the blocks may occur in a different order than those shown in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0073] Note that, unless otherwise explicitly stated, all features disclosed in this specification (including any appended claims, abstract, and drawings) may be replaced by alternative features for achieving the same, equivalent, or similar purpose. Therefore, unless explicitly stated otherwise, each disclosed feature is merely one example of a set of equivalent or similar features. Where used, "further," "preferably," "even further," and "more preferably" are simple starting points for describing another embodiment based on the foregoing embodiments, the combination of which with the foregoing embodiments constitutes the complete configuration of another embodiment. Any combination of several "further," "preferably," "even further," or "more preferably" settings following the same embodiment constitutes yet another embodiment.

[0074] Those skilled in the art should understand that the embodiments of the present invention described above and shown in the accompanying drawings are merely examples and do not limit the present invention. The objectives of the present invention have been fully and effectively achieved. The functions and structural principles of the present invention have been demonstrated and explained in the embodiments, and any variations or modifications may be made to the implementation of the present invention without departing from the stated principles.

[0075] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this disclosure, and are not intended to limit them. Although this disclosure has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this disclosure.

[0076] The above description is merely an embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of the present invention should be included within the scope of the claims of the present invention.

Claims

1. A method for predicting spinning quality based on the Attention-GRU model, characterized in that, The prediction method includes: Determine N raw cotton performance indicators and convert the N raw cotton performance indicators into an N-dimensional feature vector; Sure Each processing step is determined according to the processing sequence. N-dimensional feature vectors are used as input parameters; A single-layer GRU neural network structure is constructed, and the input parameters are fully learned through the GRU neural network structure to capture the dependencies in the time series data; Specifically, the construction of a single-layer GRU neural network structure, and the full learning of the input parameters through the GRU neural network structure to capture dependencies in time-series data, includes: The hidden state vector of the encoder is output through the single-layer GRU neural network structure. and the first hidden state of the decoder ; An Attention layer is established to analyze the importance of the N raw cotton performance indicators at different times based on the weight size, and the optimal weight parameters are continuously updated and iterated. The output of the Attention layer is connected to a fully connected neural network to form an Attention-GRU yarn quality prediction model. Information is aggregated through the output of the fully connected network to output the predicted value of yarn quality. The method further includes, after connecting the output of the Attention layer to a fully connected neural network, aggregating information through the output of the fully connected neural network, and outputting a predicted value for yarn quality, the method also includes: The loss function is used to calculate the loss between the predicted value of the yarn quality output by the model and the actual value of the yarn quality, and the Adam optimizer is selected to optimize the parameters of the Attention-GRU spinning quality prediction model. Specifically, the step of calculating the loss between the predicted yarn quality value and the actual yarn quality value output by the model using a loss function, and selecting the Adam optimizer to optimize the parameters of the Attention-GRU spinning quality prediction model, includes: Independent adaptive learning rates are designed for different parameters by calculating the first and second moment estimates of the gradient, and the weights of the neural network are iteratively updated based on the training data to optimize the output value of the loss function. The step of determining N raw cotton performance indicators and converting the N raw cotton performance indicators into an N-dimensional feature vector includes: When the number of key process parameters in each step of the spinning process is less than N, 0 is used to fill in the gaps to determine the N raw cotton performance indicators; Specifically, establishing an Attention layer to analyze the importance of the N raw cotton performance indicators at different times based on their weights, and continuously updating and iterating to find the optimal weight parameters, includes: The importance of the N raw cotton performance indicators is calculated using the following formula: in: for middle The hidden state at any given moment. This represents the first hidden state of the decoder. , This is a similarity function used to calculate... and Cosine similarity score between ; To normalize the exponential function, Converted into weights for each hidden state ; This is a concatenation function; The predicted output with the Attention mechanism added.

2. The spinning quality prediction method based on the Attention-GRU model according to claim 1, characterized in that, The step of determining N raw cotton performance indicators and converting the N raw cotton performance indicators into an N-dimensional feature vector further includes: The N raw cotton performance indicators were normalized using the min-max normalization method. between.

3. The spinning quality prediction method based on the Attention-GRU model according to claim 2, Its features are, The N raw cotton performance indicators are normalized using the following formula. between: in: The original input data, These represent the maximum and minimum values ​​of the N raw cotton performance indicators, respectively. The input data has been normalized.

4. The spinning quality prediction method based on the Attention-GRU model according to claim 1, characterized in that, The process of designing independent adaptive learning rates for different parameters by calculating the first and second moment estimates of the gradient, and iteratively updating the weights of the neural network based on the training data to optimize the loss function output, specifically involves: The output value of the loss function is determined using the following formula: in, For the number of samples, and The first The predicted and actual values ​​for each sample This is the output value of the loss function.

5. The spinning quality prediction method based on the Attention-GRU model according to claim 1, characterized in that, The method further includes: calculating the loss between the predicted yarn quality value and the actual yarn quality value output by the model using a loss function, and optimizing the parameters of the Attention-GRU spinning quality prediction model using the Adam optimizer; The optimized model parameters are then input into the Attention-GRU spinning quality prediction model to obtain the optimized spinning quality prediction model.

Images