Ball mill granularity soft measurement method based on large time sequence model

Abstract

The invention relates to the technical field of engineering process control, and discloses a ball mill granularity soft measurement method based on a time sequence large model. Constructing a multi-feature fusion module for extracting multi-scale features based on Convld K3, Convld K5 and Convld K1, and constructing a soft measurement model in combination with multi-head attention and a large language module with a fixed weight; the soft measurement model generates a first feature and a query matrix, generates a key matrix and a value matrix according to a fixed weight of the large language module, generates a second feature based on the query matrix, the key matrix and the value matrix, and fuses the second feature with the first feature to obtain a fused feature; and then a granularity prediction result corresponding to the field data is obtained through a large language module, so that the problems that an existing soft measurement model is too high in dependence on large-scale sample data, insufficient in modeling capability for complex nonlinear process parameters and low in prediction precision are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial process control, and in particular to a ball mill particle size soft measurement method based on a time series large model. Background Art

[0002] Ball mills are key equipment in the industrial grinding process, and their discharge particle size is directly related to the grinding effect and process stability. Due to the existing problems of high equipment cost, long response delays, and difficulty in acquiring data for real-time online particle size measurement, the construction of soft sensing models based on other readily available process parameters has become a research hotspot in the industry.

[0003] However, traditional soft measurement methods often require a large amount of sample data to train the model in order to effectively capture complex process characteristics and nonlinear relationships. In actual scenarios where data is scarce, the prediction accuracy is difficult to be satisfactory, and it is impossible to accurately measure the output particle size of the ball mill. Summary of the Invention

[0004] The present invention provides a ball mill particle size soft measurement method based on a time series large model to solve the problems of existing soft measurement models being too dependent on large-scale sample data, insufficient modeling capabilities for complex nonlinear process parameters, and low prediction accuracy.

[0005] In order to achieve the above object, the present invention is implemented through the following technical solutions: The present invention provides a ball mill particle size soft measurement method based on a time series large model, comprising the following steps: Step 1: Based on Convld K3, Convld K5 and Convld K1, a multi-feature fusion module for extracting multi-scale features is constructed. The soft sensing model is constructed based on the query matrix, key matrix and value matrix of multi-head attention combined with the large language module with fixed weights and the multi-feature fusion module; Step 2: Obtain field data from ball mill grinding and input it into the soft measurement model. The soft measurement model generates a first feature and a query matrix based on the field data. The soft measurement model generates a key matrix and a value matrix based on the fixed weights of the large language module. The second feature is generated based on the query matrix, key matrix, and value matrix, and is fused with the first feature to obtain a fused feature. The particle size prediction result corresponding to the field data is then obtained through the large language module.

[0006] By adopting a large language module with fixed weights, the rich semantic knowledge accumulated by the large model during the pre-training phase can be fully utilized to achieve high prediction accuracy and stability even under limited sample data conditions, thereby effectively reducing the requirement for reliance on large amounts of data. By utilizing the Transformer architecture with a multi-head attention mechanism and a large language module, it is possible to fully model the local and global features in the input data from multiple perspectives, effectively capturing the complex nonlinear and dynamic dependencies between the ball mill process parameters, and improving the system's ability to respond to particle size changes under complex working conditions.

[0007] Furthermore, the multi-feature fusion module includes an input layer, a Convld K3 layer, a Convld K5 layer, a ConvldK1 layer, a split layer, a concat layer and a first output layer; The input layer is connected to the Convld K3 layer, the split layer, the cancat layer, the Convld K1 layer and the first output layer in sequence, and the split layer and the cancat layer are further connected through the Convld K5 layer.

[0008] Furthermore, the soft measurement model is constructed in the following manner: a first multi-feature fusion module is set in the query matrix generation branch of the multi-head attention, the fixed weight of the large language model is used as the input of the key matrix and value matrix generation branch, and a second multi-feature fusion module is set in the splicing fusion of the multi-head attention, and the Transformer framework of the large language model is set in the fusion feature extraction granularity prediction result stage.

[0009] A multi-head attention mechanism is used to perform parallel computations on the query matrix, key matrix, and value matrix, capturing both local and global features in the data from multiple perspectives and effectively handling the interactions between complex, nonlinear process parameters. Combined with the Transformer architecture's ability to extract long-range dependencies, the entire system achieves a balance between accuracy and computational efficiency in real-time online prediction, a requirement that existing technologies often struggle to simultaneously address.

[0010] Furthermore, in step 2, after acquiring the field data, feature extraction is performed to obtain statistical features, and then the statistical features are input into the soft sensor model; The statistical characteristics include central tendency characteristics, dispersion characteristics, distribution characteristics and data distribution characteristics; The central tendency characteristics include mean and median; The dispersion characteristics include variance, standard deviation, root mean square error and bias; The distribution characteristics include maximum value, minimum value, first quartile, second quartile, third quartile, kurtosis, skewness and percentile; The data distribution characteristics include autocorrelation coefficients.

[0011] Furthermore, the soft measurement model specifically includes a data branch, a multi-head branch, and an output branch based on a large language module; The data branch generates a first feature based on a statistical feature of the field data; The multi-branch generates a query matrix based on the statistical features of the field data, generates a key matrix and a value matrix according to the fixed weights of the large language module, and then generates a second feature based on the query matrix, the key matrix and the value matrix; The output branch obtains the first feature and the second feature and fuses them to obtain a fused feature, and extracts the long-distance dependency of the fused feature through the large language module to generate a granularity prediction result corresponding to the field data.

[0012] Furthermore, the data branch includes a statistical feature layer, a prompt layer, a tokenizer layer, an embedding layer, and a linear layer in sequence; The statistical feature layer is based on the central trend feature, discreteness feature, distribution feature and statistical features of the data distribution feature of the field data and is transmitted to the prompt layer. The prompt layer converts the statistical features into prompt words and transmits them to the tokenizer layer for word segmentation processing, and then generates the first feature through vector conversion of the embedding layer and linear processing of the linear layer.

[0013] By comprehensively extracting multiple statistical features, this data is converted into prompt words, which are then processed through a tokenizer to form model input. This approach, combining statistical features with prompt word construction, allows the model to incorporate prior knowledge during the feature representation stage, significantly outperforming traditional methods that rely solely on numerical data input.

[0014] Furthermore, the multi-head branch includes a normalization layer, an unfold layer, a first multi-feature fusion module, a first linear layer, a second linear layer, an MHA layer, and a second feature fusion module; The normalization layer normalizes the data and transmits it to the unfold layer. The unfold layer extracts sliding window features from the normalized data. The first multi-feature fusion module performs feature conversion based on the sliding window features to generate a query matrix. The first linear layer generates a key matrix based on the fixed weights of the large language module, and the second linear layer generates a value matrix based on the fixed weights of the large language module; The MHA layer performs multi-head attention mechanism calculation based on the query matrix, key matrix and value matrix and combines it with the second feature fusion module to generate the second feature.

[0015] Furthermore, the output branch is composed of a large language module, including a concat layer, a large language module, a third linear layer and a second output layer; The concat layer, the large language module, the third linear layer and the second output layer are connected in sequence; The concat layer obtains the first feature and the second feature, fuses them to obtain a fused feature, and transmits it to the large language module. The large language module extracts the encoding vector of the long-distance dependency relationship of the fused feature and transmits it to the third linear layer. The third linear layer generates a granularity prediction result corresponding to the field data based on the encoding vector and outputs it through the second output layer.

[0016] By freezing the pre-trained weights, the model can significantly reduce the computational consumption during training and inference while maintaining high prediction accuracy, thereby meeting the strict real-time and stability requirements of industrial control systems and achieving online prediction and rapid response.

[0017] Furthermore, the soft measurement model constructs an objective function based on the actual particle size value of the ball mill grinding output and the predicted particle size value predicted by the soft measurement model; The objective function MSE is expressed by the following formula: ; in, is the total number of samples; It is The true granularity value of each sample; It is The predicted granularity value of the samples.

[0018] Furthermore, the training process of the soft sensor model adopts adaptive learning and regularization mechanism to optimize the model.

[0019] Beneficial effects: The present invention provides a soft-sensing method for ball mill particle size based on a large time-series model. By combining the powerful expressive power of a large language module with the advantages of a multi-head attention mechanism, it can effectively process complex, nonlinear data and provide accurate ball mill particle size predictions. Due to the fixed weights of the large language module, the soft-sensing model not only operates efficiently with limited training resources but also automatically adapts to changing data patterns under different operating conditions. The soft-sensing model exhibits excellent scalability, can handle large datasets, and can be easily expanded to accommodate other complex industrial tasks. This addresses the issues of existing soft-sensing models, such as their excessive reliance on large-scale sample data, insufficient modeling capabilities for complex nonlinear process parameters, and low prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 Schematic diagram of the network structure of the multi-feature fusion module in an embodiment of the present invention; Figure 2 Schematic diagram of the network structure of the soft measurement model in an embodiment of the present invention. DETAILED DESCRIPTION

[0021] The following is a clear and complete description of the technical solutions of the present invention. It should be understood that the embodiments described are only a portion of the embodiments of the present invention, not all of them. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present invention without inventive effort are intended to fall within the scope of protection of the present invention.

[0022] Unless otherwise defined, the technical or scientific terms used in the present invention shall have the usual meanings understood by persons of ordinary skill in the field to which the present invention belongs. The words "first", "second" and similar terms used in the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. Similarly, words such as "one" or "a" do not indicate a quantity limitation, but rather indicate the existence of at least one. Words such as "connected" or "connected" are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the object being described changes, the relative positional relationship also changes accordingly.

[0023] The present invention provides a method for soft-sensing particle size of a ball mill based on a time series large model, comprising the following steps: Step 1: Based on Convld K3, Convld K5 and Convld K1, a multi-feature fusion module for extracting multi-scale features is constructed. The soft sensing model is constructed based on the query matrix, key matrix and value matrix of multi-head attention combined with the large language module with fixed weights and the multi-feature fusion module; For details, please see Figure 1 ,The multi-feature fusion module includes the input layer, Convld K3 layer, Convld K5 layer, ConvldK1 layer, split layer, concat layer and the first output layer; The input layer is connected to the Convld K3 layer, the split layer, the cancat layer, the Convld K1 layer, and the first output layer in sequence. The split layer and the cancat layer are also connected through the Convld K5 layer.

[0024] Here, the multi-feature fusion module first performs preliminary feature extraction on the input signal through a one-dimensional convolution layer with a convolution kernel size of 3, generating a feature map with a higher-dimensional representation. Subsequently, the feature map is divided into two equal parts in the channel dimension to respectively extract and retain information at different scales. One branch continues to use the original features without any transformation, aiming to maintain the original properties of the input data; the other branch obtains local features within a larger receptive field through a convolution layer with a kernel size of 5, thereby capturing richer spatial semantic information. Finally, the outputs of the two branches are cascaded in the channel dimension and fused through a 1×1 convolution to form the final output. This design not only enhances the model's sensitivity to multi-scale features, but also improves the efficiency of feature fusion, helping to better integrate local area information while maintaining the original information, thereby improving overall model performance.

[0025] The construction of the soft measurement model includes: setting up a first multi-feature fusion module in the query matrix generation branch of the multi-head attention, using the fixed weights of the large language model as the input of the key matrix and value matrix generation branch, setting up a second multi-feature fusion module in the splicing fusion of the multi-head attention, and setting up the Transformer framework of the large language model in the fusion feature extraction granularity prediction result stage. Among them, the first multi-feature fusion module and the second multi-feature fusion module are both multi-feature fusion modules, and the first and second are only used here to distinguish; Step 2: Obtain field data from ball mill grinding and input it into the soft measurement model. The soft measurement model generates a first feature and a query matrix based on the field data. The soft measurement model generates a key matrix and a value matrix based on the fixed weights of the large language module. The second feature is generated based on the query matrix, key matrix, and value matrix, and is fused with the first feature to obtain a fused feature. The particle size prediction result corresponding to the field data is then obtained through the large language module.

[0026] In this embodiment, the collected field data include feed rate, grinding noise, water supply, cyclone feed concentration, and cyclone overflow particle size. After collection, the above data are subjected to conventional processing such as feature extraction, data cleaning, and normalization. Feature extraction involves extracting relevant features based on the characteristics of the raw data using methods such as signal processing and statistical analysis. For example, trends and periodicity can be extracted from time series data, and the frequency components of the signal can be extracted from frequency domain data. These extracted features help the model better capture the dynamic laws of ball mill operation, thereby improving prediction accuracy. Data cleaning involves handling missing values and outliers. Common methods include interpolation (for example, linear interpolation), mean padding, or removing outliers to ensure data integrity and accuracy. This prevents data issues from affecting the model learning process. Normalization involves transforming features into a distribution with a mean of 0 and a standard deviation of 1. Normalization scales feature values to a specific range (such as 0 to 1). This helps improve model convergence speed and stability, ensures that the numerical range of each feature is consistent, and prevents certain features from having an excessive impact on the model training process.

[0027] Statistical characteristics include central tendency characteristics, dispersion characteristics, distribution characteristics and data distribution characteristics; The central tendency characteristics include mean and median; mean is the average value of the data, and median is the middle value after the data is sorted. If there is an even number of them, the middle two numbers are taken and the mean is taken.

[0028] Dispersion characteristics include variance, standard deviation, root mean square error, and bias; Distribution characteristics include maximum value, minimum value, first quartile, second quartile, third quartile, kurtosis, skewness and percentiles; the first quartile means 25% of the data is less than this number, the second quartile is 50%, and the third quartile is 75%. Percentiles divide the data into 100 equal points to indicate the relative position of the data.

[0029] Data distribution characteristics include autocorrelation coefficients.

[0030] For the network structure of the soft sensor model, see Figure 2 ,The soft measurement model specifically includes data branches, multi-head branches, and output branches based on ,large language modules; The data branch generates a first feature based on statistical features of the field data; The multi-branch generates a query matrix based on the statistical characteristics of the field data, generates a key matrix and a value matrix based on the fixed weights of the large language module, and then generates a second feature based on the query matrix, key matrix and value matrix; The output branch obtains the first feature and the second feature and fuses them to obtain the fused feature. The long-distance dependency relationship of the fused feature is extracted through the large language module to generate the granular prediction result corresponding to the field data.

[0031] The data branch includes the statistical feature layer, prompt layer, tokenizer layer, embedding layer and linear layer in sequence; The statistical feature layer transmits the central trend characteristics, discrete characteristics, distribution characteristics and statistical characteristics of the data distribution characteristics of the field data to the prompt layer. The prompt layer converts the statistical features into prompt words and transmits them to the tokenizer layer for word segmentation processing. The first feature is then generated through vector conversion in the embedding layer and linear processing in the linear layer.

[0032] Through statistical feature extraction, prompt word construction, data normalization and multi-stage feature fusion process, the present invention achieves a comprehensive analysis of multi-dimensional data, which not only retains the key information of the original data, but also improves the semantic expression and modeling capabilities of the data through prompt word and multi-scale feature fusion, ensuring that the prediction model fully captures the inherent laws of the data.

[0033] The multi-head branch includes a normalization layer, an unfold layer, a first multi-feature fusion module, a first linear layer, a second linear layer, an MHA layer, and a second feature fusion module; The normalization layer normalizes the field data and transmits it to the unfold layer. The unfold layer extracts sliding window features from the normalized data. The first multi-feature fusion module performs feature conversion based on the sliding window features to generate a query matrix. Specifically, the first linear layer generates a key matrix based on the fixed weights of the large language module, and the second linear layer generates a value matrix based on the fixed weights of the large language module; The MHA layer performs multi-head attention mechanism calculation based on the query matrix, key matrix and value matrix and combines it with the second feature fusion module to generate the second feature.

[0034] The output branch is based on the large language module, including the concat layer, the large language module, the third linear layer and the second output layer; The concat layer, the large language module, the third linear layer, and the second output layer are connected in sequence; The concat layer obtains the first feature and the second feature, fuses them to obtain the fused feature and transmits it to the large language module. The large language module extracts the encoding vector of the long-distance dependency of the fused feature and transmits it to the third linear layer. The third linear layer generates the granularity prediction result corresponding to the field data based on the encoding vector and outputs it through the second output layer.

[0035] Regarding the training of the soft measurement model, the soft measurement model constructs an objective function based on the actual particle size value produced by the ball mill and the predicted particle size value predicted by the soft measurement model; The objective function MSE is expressed by the following formula; ; in, is the total number of samples; It is The true granularity value of each sample; It is The predicted granularity value of the samples.

[0036] The training process of the soft sensor model uses adaptive learning and regularization mechanism to optimize the model.

[0037] The Adam optimizer with adaptive learning rate is used during training, and the initial learning rate is set to , and their attenuation coefficients are set as and , the batch size of training is set to 64, the number of training rounds is set to 100, and finally the regularization mechanism is introduced, L2 regularization is adopted, and the regularization coefficient is set to , in order to constrain the model weights, thereby improving its generalization ability and preventing the model from overfitting the training data.

[0038] The soft measurement model is not only applicable to the online soft measurement of ball mill particle size in this embodiment, but also has a wide range of generalization capabilities and can be extended to soft measurement tasks in other process industries such as mineral processing, metallurgy, and chemical industry. It provides a new technical solution for complex industrial data analysis and real-time monitoring, and has good industrial application prospects and economic benefits.

[0039] Finally, the soft measurement model constructed by the ball mill particle size soft measurement method based on the time series large model provided by the present invention is used to predict the same 5000 sets of ball mill data (the conventional large model training sample set includes hundreds of thousands of sets of data), and the mean square error (MSE) and determination coefficient ( ) for horizontal comparison, see Table 1 for details; Table 1: Horizontal comparison results of mean square error and determination coefficient.

[0040]

[0041] As can be seen from the table, the soft sensing model performs best in terms of mean squared error (MSE), with an MSE of 0.035, significantly lower than other traditional methods. Linear regression achieves the highest MSE, reaching 0.065, demonstrating its limitations in handling complex nonlinear relationships. Support vector regression (SVR) and random forest achieve relatively small MSEs of 0.055 and 0.048, respectively, but still lower than the soft sensing model. The neural network achieves an MSE of 0.040, which compares favorably to traditional methods but still falls short of the model presented in this paper.

[0042] In the coefficient of determination ( ), the soft sensor model also performs well. It reached 0.95, far exceeding other methods. The indicator is used to measure the goodness of fit of the model. The closer the value is to 1, the stronger the predictive ability of the model. It is only 0.60, indicating that it is less effective in capturing the complex relationships of the data. is 0.75, slightly higher than linear regression, but still not comparable to deep learning models. is 0.80, the neural network The value is 0.88, which shows their strong fitting ability in some nonlinear tasks, but they all fail to reach the level of soft sensor models.

[0043] According to MSE and Comparison results clearly show that the soft sensing model provides the most accurate and stable predictions for ball mill particle size prediction. Traditional methods such as linear regression and SVR, while applicable in certain situations, significantly underperform large-scale language model (LLM)-based methods when faced with complex nonlinear data. By combining frozen LLM pre-trained weights with a multi-head attention mechanism, the proposed model demonstrates strong generalization and high prediction accuracy when processing complex industrial data, providing a more powerful soft sensing tool for ball mill particle size prediction.

[0044] The above describes in detail the preferred embodiments of the present invention. It should be understood that those skilled in the art can make numerous modifications and variations based on the concepts of the present invention without inventive effort. Therefore, any technical solutions that can be derived by those skilled in the art through logical analysis, reasoning, or limited experimentation based on the concepts of the present invention and the prior art should be within the scope of protection defined by the claims.

Claims

1. A ball mill particle size soft measurement method based on a time series large model, characterized in that: The steps include: Step 1: Based on Convld K3, Convld K5 and Convld K1, a multi-feature fusion module for extracting multi-scale features is constructed. The soft sensing model is constructed based on the query matrix, key matrix and value matrix of multi-head attention combined with the large language module with fixed weights and the multi-feature fusion module; Step 2: Obtain field data from ball mill grinding and input it into the soft measurement model. The soft measurement model generates a first feature and a query matrix based on the field data. The soft measurement model generates a key matrix and a value matrix based on the fixed weights of the large language module. The second feature is generated based on the query matrix, key matrix, and value matrix, and is fused with the first feature to obtain a fused feature. The particle size prediction result corresponding to the field data is then obtained through the large language module.

2. The ball mill particle size soft measurement method based on time series large model according to claim 1 is characterized in that: The multi-feature fusion module includes an input layer, a Convld K3 layer, a Convld K5 layer, a Convld K1 layer, a split layer, a concat layer and a first output layer; The input layer is connected to the Convld K3 layer, the split layer, the cancat layer, the Convld K1 layer and the first output layer in sequence, and the split layer and the cancat layer are further connected through the Convld K5 layer.

3. The ball mill particle size soft measurement method based on time series large model according to claim 1 is characterized in that: The soft measurement model is constructed in the following manner: a first multi-feature fusion module is set in the query matrix generation branch of the multi-head attention, the fixed weight of the large language model is used as the input of the key matrix and value matrix generation branch, a second multi-feature fusion module is set in the splicing fusion of the multi-head attention, and a Transformer framework of the large language model is set in the fusion feature extraction granularity prediction result stage.

4. The ball mill particle size soft measurement method based on time series large model according to claim 1 is characterized in that: In step 2, after acquiring the field data, feature extraction is performed to obtain statistical features, and then the statistical features are input into the soft sensor model; The statistical characteristics include central tendency characteristics, dispersion characteristics, distribution characteristics and data distribution characteristics; The central tendency characteristics include mean and median; The dispersion characteristics include variance, standard deviation, root mean square error and bias; The distribution characteristics include maximum value, minimum value, first quartile, second quartile, third quartile, kurtosis, skewness and percentile; The data distribution characteristics include autocorrelation coefficients.

5. The ball mill particle size soft measurement method based on time series large model according to claim 3 is characterized in that: The soft measurement model specifically includes a data branch, a multi-head branch, and an output branch based on a large language module; The data branch generates a first feature based on a statistical feature of the field data; The multi-branch generates a query matrix based on the statistical features of the field data, generates a key matrix and a value matrix according to the fixed weights of the large language module, and then generates a second feature based on the query matrix, the key matrix and the value matrix; The output branch obtains the first feature and the second feature and fuses them to obtain a fused feature, and extracts the long-distance dependency of the fused feature through the large language module to generate a granularity prediction result corresponding to the field data.

6. The ball mill particle size soft measurement method based on time series large model according to claim 5 is characterized in that: The data branch includes a statistical feature layer, a prompt layer, a tokenizer layer, an embedding layer, and a linear layer in sequence; The statistical feature layer is based on the central trend feature, discreteness feature, distribution feature and statistical features of the data distribution feature of the field data and is transmitted to the prompt layer. The prompt layer converts the statistical features into prompt words and transmits them to the tokenizer layer for word segmentation processing, and then generates the first feature through vector conversion of the embedding layer and linear processing of the linear layer.

7. The ball mill particle size soft measurement method based on time series large model according to claim 5 is characterized in that: The multi-head branch includes a normalization layer, an unfold layer, a first multi-feature fusion module, a first linear layer, a second linear layer, an MHA layer and a second feature fusion module; The normalization layer normalizes the field data and transmits it to the unfold layer. The unfold layer extracts sliding window features from the normalized data. The first multi-feature fusion module performs feature conversion based on the sliding window features to generate a query matrix. The first linear layer generates a key matrix based on the fixed weights of the large language module, and the second linear layer generates a value matrix based on the fixed weights of the large language module; The MHA layer performs multi-head attention mechanism calculation based on the query matrix, key matrix and value matrix and combines it with the second feature fusion module to generate the second feature.

8. The ball mill particle size soft measurement method based on time series large model according to claim 5 is characterized in that: The output branch is based on a large language module and includes a concat layer, a large language module, a third linear layer, and a second output layer; The concat layer, the large language module, the third linear layer and the second output layer are connected in sequence; The concat layer obtains the first feature and the second feature, fuses them to obtain a fused feature, and transmits it to the large language module. The large language module extracts the encoding vector of the long-distance dependency relationship of the fused feature and transmits it to the third linear layer. The third linear layer generates a granularity prediction result corresponding to the field data based on the encoding vector and outputs it through the second output layer.

9. The ball mill particle size soft measurement method based on time series large model according to claim 1 is characterized in that: The soft measurement model constructs an objective function based on the actual particle size value of the ball mill grinding output and the predicted particle size value predicted by the soft measurement model; The objective function MSE is expressed by the following formula: ； in, is the total number of samples; It is The true granularity value of each sample; It is The predicted granularity value of the samples.

10. The ball mill particle size soft measurement method based on time series large model according to claim 1 is characterized in that: The training process of the soft sensor model adopts adaptive learning and regularization mechanism to optimize the model.

Images