Page 50 - 《橡塑技术与装备》英文版2026年2期

HINA R&P TECHNOLOGY AND EQUIPMENT

after pyrolysis, and the pyrolysis conditions. quality of pyrolysis products under specific conditions.
(2) Feature engineering: Based on the problem definition, 2.2 Functional analysis and key technology
select or create features that enable the differentiation of implementation strategy of support vector
different product categories. machine (SVM) model
(3) Model Training: Train a logistic regression model Main function: A kind of very powerful supervised
using historical data to learn the relationship between input learning algorithm, which excels particularly in classification
features and target categories. and regression problems. The core idea of SVM is to construct
(4) Model evaluation: Evaluate the classification a decision boundary by maximizing the interval between data
performance of the model through indicators such as accuracy, points, thereby improving the accuracy of classification. SVM
recall rate, and F1 score. can be used to predict the composition of products from the
(5) Application and optimization: Apply the model to pyrolysis process under different temperatures and reaction
new waste plastic samples to predict the types of products after times. In waste plastic pyrolysis technology, SVM can be used
pyrolysis, guiding the selection and optimization of pyrolysis to distinguish between different types of plastic materials, or to
processes. optimize pyrolysis conditions to maximize the yield of specific
2.1.3 Selection strategy for key technology products.
implementation Key Technical Implementation Strategies: Select
By rationally applying linear regression and logistic a kernel function suitable for linearly separable data, a
regression models, the efficiency of the waste plastic pyrolysis polynomial kernel that captures nonlinear relationships by
process, resource utilization, and environmental benefits can be combining additional features, a radial basis function (RBF)
improved. kernel suitable for nonlinearly separable data and capable
In practical applications, the scikit-learn library in Python of handling high-dimensional spaces, and a Sigmoid kernel
can be utilized to implement these two models. For instance, similar to the activation function in neural networks. Control
LinearRegression can be employed for linear regression, the fault tolerance of the model, use K-fold cross-validation
while LogisticRegression can be used for logistic regression. to evaluate the model's generalization ability, employ
Furthermore, to enhance the predictive performance of the optimization algorithms (such as the SMO algorithm) to
model, tasks such as feature selection, feature scaling, and solve the optimization problem of SVM, select the features
model parameter tuning are typically required. that contribute most to model performance through statistical
Choose an appropriate model based on the specific nature methods or recursive feature elimination (RFE), and use
of the problem. Linear regression is suitable for predicting methods such as principal component analysis (PCA) and
continuous variables, while logistic regression is suitable for singular value decomposition (SVD) to extract new features
binary or multi-class classification problems. Steps such as data from the original features for the model. Deploy the trained
cleaning, feature engineering, standardization, or normalization model to the production environment for real-time prediction,
are crucial to the performance of the model. and evaluate the actual performance of the model through A/B
Use appropriate evaluation metrics and cross-validation testing.
methods to ensure the generalization ability of the model. In 2.3 Main functional analysis and technical
practical applications, the real-time performance of the model implementation strategies of decision tree
needs to be considered, and online learning or incremental and random forest models
learning methods may be required. Main functions: Both decision trees and random forests
The linear regression model is suitable for handling provide visual tools and features to help users understand
small-scale datasets, especially when the relationships between the decision-making process of the model, which have
features are relatively simple and linearly separable. The significant application value in classification and regression
logistic regression model can be used to predict the type or tasks. Decision trees are used to understand how different

·6· Vol.52，No.2