Intelligent distillate oil property prediction method in combination with multi-center loss

Abstract

The invention discloses a distillate oil property intelligent prediction method in combination with multi-center loss. According to the method, a convolution-attention double-branch network is constructed to carry out classification training on near infrared spectrums, a joint supervision mechanism of cross entropy loss and multi-center loss is introduced in training, and intra-class compact and inter-class separable high-dimensional spectral features are extracted. Then, historical samples are retrieved from a database based on prediction categories, the most similar samples are selected by using cosine similarity to construct a correction set, and standardization processing is performed on spectral features and property labels; and finally, carrying out partial least squares regression modeling on the correction set, extracting latent variables to maximize the correlation between the spectral features and the physicochemical properties, and inputting the feature vectors of an oil sample to be detected into the model to obtain a prediction result. According to the method, the robustness and applicability under the small sample condition are enhanced while the prediction precision is guaranteed, and the method is suitable for rapid property detection and intelligent analysis in the refining process.

Description

Technical Field

[0002] This invention belongs to the field of distillate oil property detection in refining and chemical enterprises, specifically a method for intelligent prediction of distillate oil properties that combines multicenter loss and deep feature regression. Background Technology

[0004] In the refining industry, rapid detection of distillate oil properties directly impacts the optimization of production processes and the quality of final products. Near-infrared spectroscopy, as a non-destructive testing method, has been widely applied to the classification and property detection of distillate oils.

[0005] In our previous research, we applied for patents including "A Method to Improve the Accuracy of Intelligent Classification of Distillate Oils" (application number: 202510486438.3) and "A Method for Predicting the Properties of Distillate Oils Based on Deep Learning Feature Extraction and Partial Least Squares Regression" (202511260761.5). These methods introduce single-center loss during network training, enhancing intra-class compactness and inter-class discriminative power by assigning a center vector to each category. While this method improves classification and prediction accuracy to some extent, actual spectral distributions often exhibit multimodal characteristics. For example, the same type of oil may be produced under different refining conditions, have different raw material sources, or be produced at different times and batches, all of which can lead to multiple clusters of spectral features in the feature space. A single center cannot cover all intra-class variations, resulting in feature learning bias for some samples and reducing the model's discriminative and generalization abilities.

[0006] To overcome the aforementioned shortcomings, this invention proposes an improved method based on multi-center loss. Unlike single-center constraints, the multi-center mechanism allows each category to correspond to multiple centers in the feature space. Constraints are applied by measuring the minimum distance between a sample and all centers of that category, thus better fitting the complex distribution of samples of the same type. This method achieves higher prediction accuracy and stability, providing a more reliable and intelligent means of property detection for refining processes. Summary of the Invention

[0008] This invention addresses the problems existing in the background technology by proposing an intelligent prediction method for distillate oil properties that combines multicenter loss and deep feature regression. The method first trains a deep learning network to extract spectral features through a spectral classification task, and then introduces a joint supervision mechanism of cross-entropy loss and multicenter loss. Next, a calibration set is dynamically constructed based on cosine similarity. Finally, partial least squares (PLS) regression is used to model the features to achieve distillate oil property prediction. The specific steps are as follows:

[0009] Step 1: Read the raw data from the distillate oil spectral database and perform spectral cleaning and data preprocessing: The cleaning stage uses an absorbance threshold-based method to remove outlier data; the preprocessing stage includes baseline correction, band truncation, vector normalization, and SG smoothing to generate a dataset. And in accordance with The proportion of the training set is used to divide the training set. and test set .

[0010] Step 2: Construct a neural network structure that integrates convolutional and attention branches. Generate high-dimensional feature representations of distillate oil through a feature concatenation module. Train the network using a joint supervision mechanism of cross-entropy loss and multi-center loss. The composite loss function used for network training... :

[0011]

[0012] Among them, the central loss It represents the minimum Euclidean distance between a high-dimensional feature and the centers of multiple high-dimensional features belonging to its category. Indicates the first The high-dimensional features output by the neural network after processing the sample belong to the first... Class, d is the feature dimension, Represents the first in the feature space The class of One central vector, For category The central set, with a batch size of M. For loss weights; Let cross-entropy be the loss function. It is a parameter matrix The List, It is the first Class deviation terms, It is a deviation term. The number of classes is used. The central loss during network training employs a mini-batch update strategy, and hyperparameters are set accordingly. The learning rate is used to adjust the center update.

[0013] Step 3, train the set The loss function is minimized in the input network using the AdamW algorithm. The model's weight parameters are adjusted to learn the nonlinear mapping relationship between spectral absorbance and distillate oil category.

[0014] Step 4, using the test set To evaluate the generalization ability of the trained model, determine whether the model's classification accuracy on the test set is greater than [previous value]. If the condition is met, proceed to step 5; otherwise, modify the hyperparameters and proceed to step 3.

[0015] Step 5: Input the near-infrared spectrum of the oil sample to be tested into the trained convolutional-attention dual-splitting network to extract high-dimensional feature vectors.

[0016] Step 6: Based on the predicted category obtained in Step 5), retrieve historical samples belonging to the same category from the spectral database. Select the 50 most similar samples according to the cosine similarity metric to construct a calibration set. The cosine similarity calculation formula is:

[0017]

[0018] Where Z and V are the high-dimensional feature vectors of the sample to be tested and the samples in the database, respectively. The Euclidean norm of a vector. This represents the dot product operation of vectors.

[0019] Step 7: Standardize the spectral data and property labels of the calibration set, including mean centering and unit variance scaling for both the spectral data and property labels.

[0020]

[0021] in, For the spectral matrix, The property matrix, For the sample size, Let be the dimension. For the quantity of properties. , Let represent the mean and standard deviation of the column vectors of the spectral matrix, respectively. , These represent the mean and standard deviation of the column vectors of the property matrix, respectively.

[0022] Step 8: Apply partial least squares regression to the standardized calibration set to extract latent variables and maximize spectral features. With property data Based on the covariance between them, a property prediction model is established.

[0023] Step 9: Input the high-dimensional feature vector of the oil sample to be tested into the learned PLS regression (property prediction) model, and output the physicochemical property prediction results of the oil sample.

[0024] Beneficial effects:

[0025] This invention discloses an intelligent prediction method for distillate oil properties combining multicenter loss. This method utilizes a spectral classification task to pre-train a neural network and introduces a joint supervision mechanism of cross-entropy loss and multicenter loss, further combining it with PLS regression to achieve property modeling under small sample conditions. Compared with single-center loss, the multicenter loss of this invention can establish multiple feature centers for each category, better characterizing the multimodal distribution characteristics of similar samples, achieving higher prediction accuracy and stability, and providing a more reliable intelligent property detection method for refining processes. Attached Figure Description

[0027] Figure 1 This is a flowchart of the intelligent prediction method for distillate oil properties based on multicenter loss according to the present invention.

[0028] Figure 2 This is a structural diagram of the one-dimensional convolutional self-attention network Conformer1D.

[0029] Figure 3 This is a schematic diagram of multicenter loss.

[0030] Figure 4 This is a graph showing the loss versus accuracy during model training.

[0031] Figure 5 This is a framework diagram for predicting the properties of gasoline. Detailed Implementation

[0033] The embodiments of the present invention are described in detail below. These embodiments are implemented based on the technical solution of the present invention, and provide detailed implementation methods and specific operation processes. However, the scope of protection of the present invention is not limited to the following embodiments.

[0034] By implementing near-infrared spectroscopy rapid analysis technology in multiple refining and chemical plants in China, the company has collected a large amount of distillate oil spectral data. Using the data in the company's database, the effectiveness of the intelligent prediction method for distillate oil properties combined with multi-center loss proposed in this invention is verified.

[0035] The overall system flowchart of the present invention is as follows: Figure 1 As shown, the specific steps include:

[0036] 1) Read the raw data from the distillate oil spectral database. Table 1 shows some of the spectral data and labels.

[0037] Table 1. Spectral absorbance measurement data

[0038]

[0039] The raw labeled spectral data is cleaned and preprocessed to generate a dataset. Data cleaning involved threshold-based cleaning of spectral outliers; data preprocessing included baseline correction, wavenumber band truncation, vector normalization, and SG smoothing. Table 2 shows some of the outlier spectral data, with absorbance far exceeding the threshold of 2.5. Table 3 shows the preprocessed spectral data. The cleaned and preprocessed datasets were then compared. The training set was divided into two parts at an 8:2 ratio. and test set .

[0040] Table 2. Abnormal spectral data

[0041]

[0042] Table 3. Raw spectral data and preprocessed spectral data

[0043]

[0044] 2) The architecture of the one-dimensional convolutional self-attention network Conformer1D is as follows: Figure 2 As shown, a high-dimensional feature representation of distillate oil is generated through a feature concatenation module, and the network is trained using a joint supervision mechanism of cross-entropy loss and multi-center loss. A schematic diagram of the multi-center loss is shown below. Figure 3 As shown. The composite loss function used for network training. :

[0045]

[0046] Among them, the central loss It represents the minimum Euclidean distance between a high-dimensional feature and the centers of multiple high-dimensional features belonging to its category. Indicates the first The high-dimensional features output by the neural network after processing the sample belong to the first... Class, d is the feature dimension, Represents the first in the feature space The class of One central vector, For category The central set, with a batch size of M. For loss weights; Let cross-entropy be the loss function. It is a parameter matrix The List, It is the first Class deviation terms, It is a deviation term. The number of classes is denoted by α. The center loss during network training employs a mini-batch update strategy, and a hyperparameter α is set to adjust the learning rate for center updates.

[0047] 3) The training set In the input network, set the initial learning rate of the network. The learning rate decay rate is Batch size is 256, regularization coefficient is... The discard rate was 0.5%. , The training process consists of 100 epochs. Based on the defined loss function, the AdamW algorithm is used to minimize the loss function. . Figure 4 The graph shows the loss versus accuracy during model training. The classification accuracy reached 99.51%, meeting the requirements and exceeding the 99.17% of a single center.

[0048] 4) such as Figure 5 As shown, taking gasoline as an example, this category involves several key physicochemical properties, including density at 20℃, initial boiling point, 10% evaporation temperature, 50% evaporation temperature, 90% evaporation temperature, final boiling point, research octane number, and anti-knock index. The near-infrared spectrum of the oil sample to be tested is input into a trained convolutional-attention dual-branch network to extract a high-dimensional feature vector. Based on the feature vector, the category of the oil sample to be tested is first predicted, and it is successfully identified as gasoline. Next, historical samples of the same category are retrieved from the spectral database, and based on the cosine similarity metric, the top 50 samples most similar to the oil sample to be tested are selected, thus constructing a calibration set to provide data support for subsequent partial least squares regression modeling.

[0049] 5) Standardize the spectral data and property labels of the calibration set, including mean centering and unit variance scaling of the spectral data and property labels respectively.

[0050] 6) Partial least squares regression is applied to the standardized calibration set to extract latent variables in order to maximize spectral features. With property data Based on the covariance between them, a property prediction model is established.

[0051] 7) Input the high-dimensional feature vector of the oil sample to be tested into the trained PLS regression model to obtain the predicted physicochemical properties of the oil sample. To verify the effectiveness of the method, this invention compares single-center training and multi-center training under the same conditions, and calculates the root mean square error (RMSE), relative root mean square error (rRMSE), and coefficient of determination based on the property prediction results. Evaluation indicators such as ( ) were used, and the results are shown in Tables 4, 5 and 6, respectively.

[0052] Table 4. Comparison of RMSE for oil product property prediction under single-center and multi-center strategies

[0053]

[0054] Table 5. Comparison of rRMSE for oil product property prediction under single-center and multi-center strategies

[0055]

[0056] Table 6. Prediction of oil properties under single-center and multi-center strategies Compare

[0057]

[0058] As shown in Tables 4, 5 and 6, compared with the single-center method, the multi-center method performs better in all evaluation indicators and can significantly improve the accuracy and stability of distillate oil property prediction.

Claims

1. A method for intelligent prediction of distillate oil properties combining multicenter loss, characterized in that, This method first trains a deep learning network to extract spectral features through a spectral classification task, and then introduces a joint supervision mechanism of cross-entropy loss and multi-center loss. Next, a calibration set is dynamically constructed based on cosine similarity. Finally, partial least squares (PLS) regression is used to model the features to predict the properties of distillate oils. The specific steps include: 1) Read the raw data from the distillate oil spectral database, clean and preprocess the spectra, and generate a dataset. Divide the training set and test set ; 2) In the convolutional-attention dual-branch fusion neural network, a joint supervision mechanism of cross-entropy loss and multi-center loss is used for training; 3) The training set The loss function is minimized in the input network using the AdamW algorithm. Adjust the model's weight parameters to learn the nonlinear mapping relationship between spectral absorbance and distillate oil categories; 4) Through the test set To evaluate the generalization ability of the trained model, determine whether the model's classification accuracy on the test set is greater than [previous value]. If the conditions are met, training is complete and proceed to step 5); otherwise, modify the hyperparameters and proceed to step 3. 5) Input the near-infrared spectrum of the oil sample to be tested into the trained convolutional-attention dual-branch fusion neural network to obtain the predicted category and extract the high-dimensional feature vector; 6) Based on the predicted category obtained in step 5), retrieve historical samples belonging to the same category from the spectral database, select the 50 most similar samples according to the cosine similarity measure, and construct a calibration set; 7) Standardize the spectral data and property labels of the calibration set, including mean centering and unit variance scaling for the spectral data and property labels respectively; 8) Partial least squares (PLS) regression is applied to the standardized calibration set to extract latent variables and maximize spectral characteristics. With property data Based on the covariance between them, a property prediction model is established; 9) Input the high-dimensional feature vector of the oil sample to be tested into the learned property prediction model, and output the physicochemical property prediction results of the oil sample.

2. The intelligent prediction method for distillate oil properties combining multi-center loss as described in claim 1, characterized in that, The cleaning stage uses an absorbance threshold-based method to remove outlier data; the preprocessing stage includes baseline correction, band clipping, vector normalization, and SG smoothing.

3. The intelligent prediction method for distillate oil properties combining multi-center loss as described in claim 1, characterized in that, A composite loss function, constructed by combining multi-center loss and cross-entropy loss, is used during network training. : Among them, the central loss It represents the minimum Euclidean distance between a high-dimensional feature and the centers of multiple high-dimensional features belonging to its category. Indicates the first The high-dimensional features output by the neural network after processing the sample belong to the first... Class, d is the feature dimension, Represents the first in the feature space The class of One central vector, For category The central set, with a batch size of M. To lose weight, The Euclidean norm of a vector; Let cross-entropy be the loss function. Represents the relationship between the weight matrix and the first... The column vector corresponding to the class It is a parameter matrix The The superscript T indicates the transpose operation of a matrix or vector; It is the first Class deviation terms, It is a deviation term. This represents the number of categories.

4. The intelligent prediction method for distillate oil properties combining multi-center loss as described in claim 1, characterized in that, The central loss during network training employs a mini-batch update strategy, and hyperparameters are set accordingly. The learning rate is used to adjust the center update.

5. The intelligent prediction method for distillate oil properties combining multi-center loss as described in claim 1, characterized in that, The formula for calculating cosine similarity is: Where Z and V are the high-dimensional feature vectors of the sample to be tested and the samples in the database, respectively. The Euclidean norm of a vector. This represents the dot product operation of vectors.

Images