Method for building carbon-based conductive industry model based on petroleum coke detection big data

Abstract

The invention relates to the technical field of carbon-based conductive industry model building, in particular to a carbon-based conductive industry model building method based on petroleum coke detection big data. The method comprises the following steps: collecting petroleum coke data, and preprocessing the petroleum coke data by using an ETL automatic tool; constructing petroleum coke interaction features, extracting petroleum coke high-dimensional features by using a convolutional neural network and introducing a conductive attribute regulation factor, and performing dimension reduction on the petroleum coke high-dimensional features by using t-SNE to obtain petroleum coke low-dimensional features; selecting XGBoost combined with an interpretable linear model as a classification model based on the low-dimensional characteristics of petroleum coke, and constructing a carbon-based conductive industry model; and evaluating and optimizing the carbon-based conductive industry model. According to the carbon-based conductivity industry model building method based on the petroleum coke detection big data, XGBoost and an interpretable linear model are combined, the model classification capability is improved, meanwhile, the result interpretability is guaranteed, and conductivity property adjusting factors are introduced to accurately capture conductivity characteristics.

Description

Technical Field

[0001] The present invention relates to the technical field of carbon-based conductive industry model building, and in particular to a method for building a carbon-based conductive industry model based on petroleum coke detection big data. Background Art

[0002] The method for building a carbon-based conductive industry model based on petroleum coke detection big data aims to accurately predict the conductive properties of carbon-based materials and optimize production process parameters. It combines high-performance XGBoost with a linear model with strong interpretability, and introduces a conductive property adjustment factor to improve the transparency and interpretability of the carbon-based conductive industry model. It also integrates the SHAP interpretability tool with XGBoost to provide a transparent decision-making process. In the existing carbon-based conductive industry model building method, the model has low interpretability, and because the deep neural network itself is a black box model, after combining with XGBoost, the interpretability of the overall model is further reduced. In the construction of the carbon-based conductive industry model that requires high transparency and interpretability, there will be problems with trust and maintenance difficulty. While ensuring high performance, the present invention combines XGBoost with an interpretable linear model to introduce a conductive property adjustment factor to improve the transparency and interpretability of the carbon-based conductive industry model. Therefore, a carbon-based conductive industry model building method based on petroleum coke detection big data is provided. Summary of the invention

[0003] The purpose of the present invention is to provide a method for building a carbon-based conductive industry model based on petroleum coke detection big data, so as to solve the problem raised in the above background technology that since the deep neural network itself is a black box model, after combining with XGBoost, the interpretability of the overall model is further reduced. In the construction of the carbon-based conductive industry model that requires high transparency and interpretability, there will be problems with trust and maintenance difficulty.

[0004] To achieve the above object, the present invention aims to provide a method for building a carbon-based conductive industry model based on petroleum coke detection big data, comprising the following steps: S1. Collect petroleum coke data and pre-process the petroleum coke data using ETL automation tools; S2. Construct interactive features of petroleum coke, use convolutional neural network and introduce conductive property adjustment factor to extract high-dimensional features of petroleum coke, and use t-SNE to reduce the dimension of high-dimensional features of petroleum coke to obtain low-dimensional features of petroleum coke; S3. Based on the low-dimensional features of petroleum coke, XGBoost combined with interpretable linear model was selected as the classification model to build a carbon-based conductive industry model; S4. Evaluate and optimize the carbon-based conductive industry model.

[0005] As a further improvement of the technical solution, in said S1, the petroleum coke data includes physical property data, chemical composition data and electrical conductivity data; Physical property data include: density , particle size distribution and specific surface area ; Chemical composition data include: Carbon content , sulfur content , oxygen content and ash ; Conductive performance data include: conductivity and resistivity ; The ETL automation tool is used to pre-process the petroleum coke data as follows: S1.1. Clean the physical property data, chemical composition data and electrical conductivity data; S1.2. Convert data in different units uniformly and normalize the numerical features of all data; S1.3. Store the preprocessed physical property data, chemical composition data and conductive performance data in a database.

[0006] As a further improvement of the technical solution, in S2, based on petroleum coke data, petroleum coke interactive features are constructed, a convolutional neural network is used to extract high-dimensional features of petroleum coke, and t-SNE is used to reduce the dimension of the high-dimensional features of petroleum coke to obtain low-dimensional features of petroleum coke. The specific method steps are as follows: S2.1. Construct the interaction characteristics of petroleum coke based on physical property data, chemical composition data and electrical conductivity data; S2.2, using convolutional neural network and introducing conductive property adjustment factor to extract high-dimensional features of petroleum coke; S2.3. Use t-SNE to reduce the dimensionality of petroleum coke high-dimensional features to obtain low-dimensional features of petroleum coke.

[0007] As a further improvement of the technical solution, in S2.1, based on the physical property data, chemical composition data and electrical conductivity data, the petroleum coke interaction characteristics are constructed, and the specific method steps are as follows: S2.1.1. Constructing physical property interaction features: Particle size distribution and specific surface area Interaction features: ; density and particle size distribution Interaction features: ; S2.1.2. Constructing chemical composition interaction features: Carbon content and sulfur content Interaction features: ; Sulfur content and oxygen content Interaction features: ; Oxygen content and ash Interaction features: ; S2.1.3. Constructing interactive features of conductive properties: Conductivity and resistivity Interaction features: ; S2.1.4. Integrate all interactive features to construct petroleum coke interactive features: ; in, It is the interactive characteristic of petroleum coke.

[0008] As a further improvement of the technical solution, in S2.2, a convolutional neural network is used and a conductive property adjustment factor is introduced to extract high-dimensional features of petroleum coke. The specific method steps are as follows: S2.2.1. Input the petroleum coke interaction features into the convolutional neural network and apply the convolution operation to extract the local features of petroleum coke: ; in, is the convolution operation; is the convolution kernel; is the bias term; S2.2.2. Density-based , particle size distribution and specific surface area , calculate the conductive property adjustment factor and introduce it into the convolution operation: ; ; in, For the first convolution operation The first output of the activation function; is the conductive property adjustment factor; is the density weight; is the particle size distribution weight; is the specific surface area weight; S2.2.3. Continue stacking convolutional layers and apply a second convolution operation to extract higher-level features: ; in, For the second convolution operation The second output of the activation function; S2.2.4. Repeat the convolution operations of S2.2.1-S2.2.2 until k convolution operations are applied to obtain the high-dimensional features of petroleum coke: ; in, It is the high-dimensional characteristics of petroleum coke; Features extracted by the first convolution operation; Features extracted for the second convolution operation; Features extracted by the kth convolution operation.

[0009] As a further improvement of the technical solution, in S2.3, t-SNE is used to reduce the dimension of the high-dimensional features of petroleum coke to obtain low-dimensional features of petroleum coke. The specific method steps are as follows: S2.3.1. Based on high-dimensional features of petroleum coke , use t-SNE to calculate the similarity of each petroleum coke high-dimensional feature to all other petroleum coke high-dimensional features: ; in, is the high-dimensional feature of the i-th petroleum coke; is the j-th high-dimensional feature of petroleum coke; is the kth petroleum coke high-dimensional feature; is the L2 norm of the i-th petroleum coke high-dimensional feature and the j-th petroleum coke high-dimensional feature; is the L2 norm of the i-th petroleum coke high-dimensional feature and the k-th petroleum coke high-dimensional feature; is the scale parameter of t-SNE; represents the sum of all k-th petroleum coke high-dimensional features, and ; is the similarity between the i-th petroleum coke high-dimensional feature and the j-th petroleum coke high-dimensional feature; S2.3.2. Define the optimization objective function of t-SNE: Optimization objective function: ; in, is the similarity between the high-dimensional feature of the i-th petroleum coke and the high-dimensional feature of the j-th petroleum coke in the low-dimensional space; S2.3.3, t-SNE solves the optimization objective function, minimizes the similarity of the two high-dimensional features of petroleum coke in the low-dimensional space, and obtains the low-dimensional features of petroleum coke: ; in, It is the low-dimensional feature of petroleum coke; To solve the optimization objective function for t-SNE, the high-dimensional features of petroleum coke are reduced in dimension.

[0010] As a further improvement of the technical solution, in S3, based on the low-dimensional features of petroleum coke, XGBoost combined with an interpretable linear model is selected as a classification model to construct a carbon-based conductive industry model. The specific method steps are as follows: S3.1. Based on low-dimensional features of petroleum coke , define the feature matrix and target variable, and divide the training set and test set; S3.2. Train the XGBoost classification model and interpretable linear model based on the training set to build a carbon-based conductive industry model.

[0011] As a further improvement of the technical solution, in S3.1, based on the low-dimensional features of petroleum coke , define the feature matrix and target variable, divide the training set and test set, the specific steps are as follows: S3.1.1, define the feature matrix and target variable: ; ; in, is the feature matrix; is the target variable, indicating that the target value needs to be predicted; is the original target variable; S3.1.2, divide the training set and test set: ; in, is the feature data in the training set; is the feature data in the test set; is the target variable in the training set; is the target variable in the test set; is the function for partitioning into training set and test set; Indicates that the proportion of the test set is 20%; The random seed to be set.

[0012] As a further improvement of the technical solution, in S3.2, an XGBoost classification model and an interpretable linear model are trained based on the training set to construct a carbon-based conductive industry model, as follows: S3.2.1. Use the training set to train the XGBoost model, and use the trained XGBoost model to predict the test set to obtain the prediction results: ; ; in, The prediction result of the XGBoost model on the test set; is the original XGBoost model; is the XGBoost model after training; S3.2.2. The prediction results of the XGBoost model on the test set Input into the interpretable linear model and train the interpretable linear model: ; in, is the feature data in the new training set; ; ; in, is the original interpretable linear model; is the interpretable linear model after training; To make the prediction results explainable; S3.2.3, the final carbon-based conductive industry model is as follows: ; ; .

[0013] As a further improvement of this technical solution, in S4, the carbon-based conductive industry model is evaluated and optimized as follows: S4.1. Use recall, F1-score and AUC-ROC indicators to evaluate the carbon-based conductive industry model; S4.2. Use the early stopping method to optimize the carbon-based conductive industry model.

[0014] Compared with the prior art, the present invention has the following beneficial effects: 1. In the method of building a carbon-based conductive industry model based on petroleum coke detection big data, by introducing a conductive property adjustment factor, the complex relationship between the physical properties and conductive properties of petroleum coke can be fully explored, and the features can be dynamically adjusted through a convolutional neural network, thereby improving the model's sensitivity and accuracy to conductivity-related features.

[0015] 2. In the carbon-based conductive industry model building method based on petroleum coke detection big data, by combining the XGBoost classification model with a highly interpretable linear model, it is possible to improve the interpretability and transparency of the model while maintaining high-performance prediction capabilities. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 The figure is a flow chart of the overall method of the present invention. DETAILED DESCRIPTION

[0017] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0018] Example: See Figure 1 As shown, this embodiment provides a method for building a carbon-based conductive industry model based on petroleum coke detection big data, including the following steps: S1. Collect petroleum coke data and pre-process the petroleum coke data using ETL automation tools; In this embodiment S1, the petroleum coke data includes physical property data, chemical composition data and electrical conductivity data; Physical property data include: density , particle size distribution and specific surface area ; Chemical composition data include: Carbon content , sulfur content , oxygen content Ash ; Conductive performance data include: conductivity and resistivity ; The ETL automation tool is used to pre-process the petroleum coke data as follows: S1.1. Clean the physical property data, chemical composition data and electrical conductivity data; S1.2. Convert data in different units uniformly and normalize the numerical features of all data; S1.3. Store the preprocessed physical property data, chemical composition data and conductive performance data in a database.

[0019] S2. Construct interactive features of petroleum coke, use convolutional neural network and introduce conductive property adjustment factor to extract high-dimensional features of petroleum coke, and use t-SNE to reduce the dimension of high-dimensional features of petroleum coke to obtain low-dimensional features of petroleum coke; In this embodiment, a convolutional neural network is used as a deep learning model to extract high-dimensional features from the interactive features of petroleum coke. The convolutional neural network mainly captures local patterns and complex structures in the data through convolution operations. The convolutional neural network learns important features in the petroleum coke data and deepens the expression ability of the model by stacking multiple convolutional layers, so that the model can extract richer features. In this process, the convolutional neural network deeply learns the interactive features of the physical properties, chemical composition and conductive properties of petroleum coke, thereby obtaining a more comprehensive feature representation. The conductivity adjustment factor is a dynamic adjustment factor calculated based on the physical characteristics of petroleum coke, such as density, particle size distribution, and specific surface area. It aims to adjust the degree of influence of the conductivity of petroleum coke based on these characteristics. By adjusting the output of the convolution operation, the conductivity adjustment factor can further optimize the performance of the convolutional neural network and improve the model's sensitivity to conductivity and prediction accuracy. t-SNE is a nonlinear technology for data dimensionality reduction. It is mainly used to map high-dimensional data to low-dimensional space. It achieves data dimensionality reduction by minimizing the difference between the probability distribution between points in high-dimensional space and the probability distribution between points in low-dimensional space, and maintains the structure and similarity of the data in low-dimensional space.

[0020] In this embodiment, t-SNE is used to reduce the dimensionality of the high-dimensional features of petroleum coke extracted by the convolutional neural network, so that the final features can be more intuitively represented in the low-dimensional space, thereby simplifying the subsequent model training process; Through the combination of convolutional neural network, conductive property adjustment factor and t-SNE, the complex features in petroleum coke data can be effectively extracted, adjusted and reduced in dimension for building a carbon-based conductive industry model; In this embodiment S2, based on petroleum coke data, petroleum coke interactive features are constructed, a convolutional neural network is used to extract high-dimensional features of petroleum coke, and t-SNE is used to reduce the dimension of the high-dimensional features of petroleum coke to obtain low-dimensional features of petroleum coke. The specific method steps are as follows: S2.1. Construct the interaction characteristics of petroleum coke based on physical property data, chemical composition data and electrical conductivity data; S2.2, using convolutional neural network and introducing conductive property adjustment factor to extract high-dimensional features of petroleum coke; S2.3. Use t-SNE to reduce the dimensionality of petroleum coke high-dimensional features to obtain petroleum coke low-dimensional features.

[0021] In this embodiment S2.1, based on the physical property data, chemical composition data and electrical conductivity data, the petroleum coke interaction characteristics are constructed, and the specific method steps are as follows: S2.1.1. Constructing physical property interaction features: Particle size distribution and specific surface area Interaction features: ; density and particle size distribution Interaction features: ; S2.1.2. Constructing chemical composition interaction features: Carbon content and sulfur content Interaction features: ; Sulfur content and oxygen content Interaction features: ; Oxygen content Ash Interaction features: ; S2.1.3. Constructing interactive features of conductive properties: Conductivity and resistivity Interaction features: ; S2.1.4. Integrate all interactive features to construct petroleum coke interactive features: ; in, It is the interactive characteristic of petroleum coke.

[0022] In this embodiment S2.2, a convolutional neural network is used and a conductive property adjustment factor is introduced to extract high-dimensional features of petroleum coke. The specific method steps are as follows: S2.2.1. Input the petroleum coke interaction features into the convolutional neural network and apply the convolution operation to extract the local features of petroleum coke: ; in, is the convolution operation; is the convolution kernel; is the bias term; S2.2.2. Density-based , particle size distribution and specific surface area , calculate the conductive property adjustment factor and introduce it into the convolution operation: ; ; in, For the first convolution operation The first output of the activation function; is the conductive property adjustment factor; is the density weight; is the particle size distribution weight; is the specific surface area weight; S2.2.3. Continue stacking convolutional layers and apply a second convolution operation to extract higher-level features: ; in, For the second convolution operation The second output of the activation function; S2.2.4. Repeat the convolution operations of S2.2.1-S2.2.2 until k convolution operations are applied to obtain the high-dimensional features of petroleum coke: ; in, It is the high-dimensional characteristics of petroleum coke; Features extracted by the first convolution operation; Features extracted for the second convolution operation; Features extracted by the kth convolution operation.

[0023] In this embodiment S2.3, t-SNE is used to reduce the dimension of the high-dimensional features of petroleum coke to obtain the low-dimensional features of petroleum coke. The specific method steps are as follows: S2.3.1. Based on high-dimensional features of petroleum coke , use t-SNE to calculate the similarity of each petroleum coke high-dimensional feature to all other petroleum coke high-dimensional features: ; in, is the high-dimensional feature of the i-th petroleum coke; is the j-th high-dimensional feature of petroleum coke; is the kth petroleum coke high-dimensional feature; is the L2 norm of the i-th petroleum coke high-dimensional feature and the j-th petroleum coke high-dimensional feature; is the L2 norm of the i-th petroleum coke high-dimensional feature and the k-th petroleum coke high-dimensional feature; is the scale parameter of t-SNE; represents the sum of all k-th petroleum coke high-dimensional features, and ; is the similarity between the i-th petroleum coke high-dimensional feature and the j-th petroleum coke high-dimensional feature; S2.3.2. Define the optimization objective function of t-SNE: Optimization objective function: ; in, is the similarity between the high-dimensional feature of the i-th petroleum coke and the high-dimensional feature of the j-th petroleum coke in the low-dimensional space; S2.3.3, t-SNE solves the optimization objective function, minimizes the similarity of the two high-dimensional features of petroleum coke in the low-dimensional space, and obtains the low-dimensional features of petroleum coke: ; in, It is the low-dimensional feature of petroleum coke; To solve the optimization objective function for t-SNE, the high-dimensional features of petroleum coke are reduced in dimension.

[0024] S3. Based on the low-dimensional features of petroleum coke, XGBoost combined with interpretable linear model was selected as the classification model to build a carbon-based conductive industry model; In this embodiment, XGBoost is an ensemble learning algorithm based on a gradient boosting tree, which has excellent classification and regression performance. XGBoost can capture nonlinear relationships and complex patterns in data through a weighted combination of multiple weak classifiers. Based on the low-dimensional features of petroleum coke, XGBoost performs feature learning and prediction; The interpretable linear model is a simple and interpretable statistical learning method that fits data through linear relationships and can clearly show the degree of influence of each feature on the target variable. In this embodiment, the interpretable linear model is selected to ensure the transparency and interpretability of the carbon-based conductive industry model. When processing features predicted by XGBoost, the interpretable linear model can provide the contribution value of each input feature to the final output of the model; The role of XGBoost combined with the interpretable linear model is to give full play to XGBoost's powerful nonlinear modeling capabilities while using the advantages of the interpretable linear model to improve the model's interpretation and transparency. First, XGBoost is used for feature learning, and then the output of XGBoost is used as the input feature, combined with the interpretable linear model for modeling. This can not only capture the complex relationship of petroleum coke data, but also ensure that the final prediction results have high interpretability. This combination achieves the complementarity of the two: XGBoost improves the accuracy and complexity of the model, while the interpretable linear model improves the model's transparency and interpretability, so that the carbon-based conductive industry model has both powerful prediction capabilities and can provide a reliable basis for decision-making; In this embodiment S3, based on the low-dimensional features of petroleum coke, XGBoost combined with an interpretable linear model is selected as a classification model to construct a carbon-based conductive industry model. The specific method steps are as follows: S3.1. Based on low-dimensional features of petroleum coke , define the feature matrix and target variable, and divide the training set and test set; S3.2. Train the XGBoost classification model and interpretable linear model based on the training set to build a carbon-based conductive industry model.

[0025] In this embodiment S3.1, based on the low-dimensional features of petroleum coke , define the feature matrix and target variable, divide the training set and test set, the specific steps are as follows: S3.1.1, define the feature matrix and target variable: ; ; in, is the feature matrix; is the target variable, indicating that the target value needs to be predicted; is the original target variable; S3.1.2, divide the training set and test set: ; in, is the feature data in the training set; is the feature data in the test set; is the target variable in the training set; is the target variable in the test set; is the function for partitioning into training set and test set; Indicates that the proportion of the test set is 20%; The random seed to be set.

[0026] In this embodiment S3.2, an XGBoost classification model and an interpretable linear model are trained based on the training set to construct a carbon-based conductive industry model, as follows: S3.2.1. Use the training set to train the XGBoost model, and use the trained XGBoost model to predict the test set to obtain the prediction results: ; ; in, The prediction result of the XGBoost model on the test set; is the original XGBoost model; is the XGBoost model after training; S3.2.2. The prediction results of the XGBoost model on the test set Input into the interpretable linear model and train the interpretable linear model: ; in, is the feature data in the new training set; ; ; in, is the original interpretable linear model; is the interpretable linear model after training; To make the prediction results explainable; S3.2.3, the final carbon-based conductive industry model is as follows: ; ; .

[0027] S4. Evaluate and optimize the carbon-based conductive industry model; In this embodiment S4, the carbon-based conductive industry model is evaluated and optimized, as follows: S4.1. Use recall, F1-score and AUC-ROC indicators to evaluate the carbon-based conductive industry model; S4.2. Use the early stopping method to optimize the carbon-based conductive industry model.

[0028] In this embodiment, the recall rate, F1-score and AUC-ROC performance indicators are used to measure the accuracy and effectiveness of the carbon-based conductive industry model. At the same time, the early stopping method is introduced to optimize the generalization ability of the carbon-based conductive industry model, prevent overfitting, and improve the stability of the carbon-based conductive industry model. The recall rate is the ability of the carbon-based conductive industry model to identify actual positive samples, which is an indicator of how many of all the samples that are actually positive are correctly identified; F1-score is the harmonic mean of precision and recall, and is a balance indicator between precision and recall; AUC-ROC is a comprehensive indicator used to evaluate the performance of the carbon-based conductive industry model under all possible classification thresholds; Early stopping is a technique used to prevent overfitting. It is used for models that take a long time to train. The performance of the model is monitored during the training process. When the performance of the validation set does not improve within several training rounds, the training is stopped. This helps to reasonably and safely train the carbon-based conductive industry model.

[0029] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments, and the above embodiments and descriptions are only preferred examples of the present invention, and are not intended to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, and these changes and improvements all fall within the scope of the present invention to be protected.

Claims

1. A method for building a carbon-based conductive industry model based on petroleum coke detection big data, characterized by: The following steps are involved: S1. Collect petroleum coke data and pre-process the petroleum coke data using ETL automation tools; S2. Construct interactive features of petroleum coke, use convolutional neural network and introduce conductive property adjustment factor to extract high-dimensional features of petroleum coke, and use t-SNE to reduce the dimension of high-dimensional features of petroleum coke to obtain low-dimensional features of petroleum coke; S3. Based on the low-dimensional features of petroleum coke, XGBoost combined with interpretable linear model was selected as the classification model to build a carbon-based conductive industry model; S4. Evaluate and optimize the carbon-based conductive industry model.

2. The method for building a carbon-based conductive industry model based on petroleum coke detection big data according to claim 1 is characterized in that: In said S1, the petroleum coke data includes physical property data, chemical composition data and electrical conductivity data; Physical property data include: density , particle size distribution and specific surface area ; Chemical composition data include: Carbon content , sulfur content , oxygen content and ash ; Conductive performance data include: conductivity and resistivity ; The ETL automation tool is used to pre-process the petroleum coke data as follows: S1.

1. Clean the physical property data, chemical composition data and electrical conductivity data; S1.

2. Convert data in different units uniformly and normalize the numerical features of all data; S1.

3. Store the preprocessed physical property data, chemical composition data and conductive performance data in a database.

3. The method for building a carbon-based conductive industry model based on petroleum coke detection big data according to claim 2 is characterized in that: In S2, based on petroleum coke data, petroleum coke interactive features are constructed, a convolutional neural network is used to extract high-dimensional features of petroleum coke, and t-SNE is used to reduce the dimension of the high-dimensional features of petroleum coke to obtain low-dimensional features of petroleum coke. The specific method steps are as follows: S2.

1. Construct the interaction characteristics of petroleum coke based on physical property data, chemical composition data and electrical conductivity data; S2.2, using convolutional neural network and introducing conductive property adjustment factor to extract high-dimensional features of petroleum coke; S2.

3. Use t-SNE to reduce the dimensionality of petroleum coke high-dimensional features to obtain low-dimensional features of petroleum coke.

4. The method for building a carbon-based conductive industry model based on petroleum coke detection big data according to claim 3 is characterized by: In S2.1, based on the physical property data, chemical composition data and electrical conductivity data, the petroleum coke interaction characteristics are constructed. The specific method steps are as follows: S2.1.

1. Constructing physical property interaction features: Particle size distribution and specific surface area Interaction features: ； density and particle size distribution Interaction features: ； S2.1.

2. Constructing chemical composition interaction features: Carbon content and sulfur content Interaction features: ； Sulfur content and oxygen content Interaction features: ； Oxygen content and ash Interaction features: ； S2.1.

3. Constructing interactive features of conductive properties: Conductivity and resistivity Interaction features: ； S2.1.

4. Integrate all interactive features to construct petroleum coke interactive features: ； in, It is the interactive characteristic of petroleum coke.

5. The method for building a carbon-based conductive industry model based on petroleum coke detection big data according to claim 4 is characterized in that: In S2.2, a convolutional neural network is used and a conductive property adjustment factor is introduced to extract high-dimensional features of petroleum coke. The specific method steps are as follows: S2.2.

1. Input the petroleum coke interaction features into the convolutional neural network and apply the convolution operation to extract the local features of petroleum coke: ； in, is the convolution operation; is the convolution kernel; is the bias term; S2.2.

2. Density-based , particle size distribution and specific surface area , calculate the conductive property adjustment factor and introduce it into the convolution operation: ； ； in, For the first convolution operation The first output of the activation function; is the conductive property adjustment factor; is the density weight; is the particle size distribution weight; is the specific surface area weight; S2.2.

3. Continue stacking convolutional layers and apply a second convolution operation to extract higher-level features: ； in, For the second convolution operation The second output of the activation function; S2.2.

4. Repeat the convolution operations of S2.2.1-S2.2.2 until k convolution operations are applied to obtain the high-dimensional features of petroleum coke: ； in, It is the high-dimensional characteristics of petroleum coke; Features extracted by the first convolution operation; Features extracted for the second convolution operation; Features extracted by the kth convolution operation.

6. The method for building a carbon-based conductive industry model based on petroleum coke detection big data according to claim 5 is characterized by: In S2.3, t-SNE is used to reduce the dimension of the high-dimensional features of petroleum coke to obtain the low-dimensional features of petroleum coke. The specific method steps are as follows: S2.3.

1. Based on high-dimensional features of petroleum coke , use t-SNE to calculate the similarity of each petroleum coke high-dimensional feature to all other petroleum coke high-dimensional features: ； in, is the high-dimensional feature of the i-th petroleum coke; is the j-th high-dimensional feature of petroleum coke; is the kth petroleum coke high-dimensional feature; is the L2 norm of the i-th petroleum coke high-dimensional feature and the j-th petroleum coke high-dimensional feature; is the L2 norm of the i-th petroleum coke high-dimensional feature and the k-th petroleum coke high-dimensional feature; is the scale parameter of t-SNE; represents the sum of all k-th petroleum coke high-dimensional features, and ; is the similarity between the i-th petroleum coke high-dimensional feature and the j-th petroleum coke high-dimensional feature; S2.3.

2. Define the optimization objective function of t-SNE: Optimization objective function: ； in, is the similarity between the high-dimensional feature of the i-th petroleum coke and the high-dimensional feature of the j-th petroleum coke in the low-dimensional space; S2.3.3, t-SNE solves the optimization objective function, minimizes the similarity of the two high-dimensional features of petroleum coke in the low-dimensional space, and obtains the low-dimensional features of petroleum coke: ； in, It is the low-dimensional feature of petroleum coke; To solve the optimization objective function for t-SNE, the high-dimensional features of petroleum coke are reduced in dimension.

7. The method for building a carbon-based conductive industry model based on petroleum coke detection big data according to claim 6 is characterized by: In S3, based on the low-dimensional features of petroleum coke, XGBoost combined with an interpretable linear model is selected as a classification model to construct a carbon-based conductive industry model. The specific steps are as follows: S3.

1. Based on low-dimensional features of petroleum coke , define the feature matrix and target variable, and divide the training set and test set; S3.

2. Train the XGBoost classification model and interpretable linear model based on the training set to build a carbon-based conductive industry model.

8. The method for building a carbon-based conductive industry model based on petroleum coke detection big data according to claim 7 is characterized in that: In S3.1, based on the low-dimensional features of petroleum coke , define the feature matrix and target variable, divide the training set and test set, the specific steps are as follows: S3.1.1, define the feature matrix and target variable: ； ； in, is the feature matrix; is the target variable, indicating that the target value needs to be predicted; is the original target variable; S3.1.2, divide the training set and test set: ； in, is the feature data in the training set; is the feature data in the test set; is the target variable in the training set; is the target variable in the test set; is the function for partitioning into training set and test set; Indicates that the proportion of the test set is 20%; The random seed to be set.

9. The method for building a carbon-based conductive industry model based on petroleum coke detection big data according to claim 8 is characterized by: In S3.2, the XGBoost classification model and the interpretable linear model are trained based on the training set to construct a carbon-based conductive industry model, as follows: S3.2.

1. Use the training set to train the XGBoost model, and use the trained XGBoost model to predict the test set to obtain the prediction results: ； ； in, The prediction result of the XGBoost model on the test set; is the original XGBoost model; is the XGBoost model after training; S3.2.

2. The prediction results of the XGBoost model on the test set Input into the interpretable linear model and train the interpretable linear model: ； in, is the feature data in the new training set; ； ； in, is the original interpretable linear model; is the interpretable linear model after training; To make the prediction results explainable; S3.2.3, the final carbon-based conductive industry model is as follows: ； ； 。 10. The method for building a carbon-based conductive industry model based on petroleum coke detection big data according to claim 9 is characterized in that: In S4, the carbon-based conductive industry model is evaluated and optimized as follows: S4.

1. Use recall, F1-score and AUC-ROC indicators to evaluate the carbon-based conductive industry model; S4.

2. Use the early stopping method to optimize the carbon-based conductive industry model.

Images