Page 49 - 《橡塑技术与装备》英文版2026年3期

SPECIAL AND COMPREHENSIVE REVIEW

subsequent data analysis. Networks (RNN), to monitor and predict key parameters such
3.1.2.2 Data preprocessing module as temperature and pressure in the pyrolysis process of waste
Perform operations such as cleaning, removing outliers, plastics in real time, thereby achieving precise control over the
filling missing values, and standardization/normalization on the pyrolysis process. Secondly, introduce reinforcement learning
collected data to ensure data quality and enhance the accuracy algorithms to construct a dynamic decision-making system,
and reliability of subsequent analysis. which automatically optimizes the optimal process conditions
3.1.2.3 Model Training and Optimization by simulating the impact of different operation strategies on the
Module pyrolysis effect of waste plastics, thereby improving pyrolysis
Select an appropriate machine learning or deep learning efficiency and product quality. In addition, combine global
model, such as a regression model for predicting product search methods such as genetic algorithms and particle swarm
yield, or utilize reinforcement learning to optimize pyrolysis optimization to solve multivariable optimization problems and
process parameters. Utilize machine learning or deep learning effectively model the complex nonlinear relationships in the
algorithms to establish a predictive model for the pyrolysis pyrolysis system.
process, and continuously iterate and optimize to enhance the Validation methods and measures: Cross-validation,
model's predictive accuracy and adaptability. A/B testing, and other methods are employed to ensure
3.1.2.4 Control strategy formulation module the stability and reliability of the selected algorithm in the
Based on the trained model, generate the optimal actual pyrolysis system. Through extensive training and
pyrolysis process control strategy to achieve precise regulation testing with experimental data, algorithm parameters are
of the pyrolysis process, thereby improving energy conversion continuously adjusted to enhance the model's generalization
efficiency and product quality. ability, ultimately achieving efficient application of intelligent
3.1.2.5 Monitoring and fault diagnosis module algorithms in the field of waste plastic pyrolysis.

Monitor the status of the pyrolysis process in real-time, 3.3 Exploration of integrated design for ai-
quickly identify abnormalities, and provide a fault diagnosis assisted pyrolysis of waste plastics
and early warning mechanism to ensure stable system Based on the integration and application of intelligent
operation. technology, a comprehensive platform capable of efficiently
3.1.2.6 User interaction module processing waste plastic pyrolysis databases, optimizing
Provide an intuitive and user-friendly operation interface, pyrolysis process flows, and designing and implementing
enabling users to easily monitor system status, adjust parameter intelligent control is constructed. The trained model is
settings, view analysis reports, etc., thereby enhancing the integrated into the control system as a decision engine. The
practicality and user experience of the system. platform should have good scalability and flexibility to adapt to
3.2 Exploration of the design and selection the needs of waste plastic pyrolysis facilities of different sizes.
of AI intelligent algorithms Here are some key integrated design strategies:
Selection criteria: The main considerations are its 3.3.1 Data acquisition and preprocessing
adaptability and accuracy in handling complex data, optimizing Integrated sensor network: Deploy various sensors (such
pyrolysis process parameters, predicting pyrolysis product as temperature, pressure, and gas composition sensors) to
distribution, and enhancing system efficiency. collect real-time data during the pyrolysis process.
Selection strategy: Adjust the hyperparameters of the Data cleaning and integration: Utilize AI algorithms to
model through methods such as grid search, random search, clean and integrate the collected data, ensuring its accuracy
or Bayesian optimization to achieve optimal performance. and completeness.
Firstly, through comparative analysis, select deep learning 3.3.2 Construction and optimization of
algorithms with high accuracy prediction capabilities, such as intelligent models
Convolutional Neural Networks (CNN) or Recurrent Neural Machine learning model: Construct a model using

Vol.52，2026 ·3·