Page 46 - 《橡塑技术与装备》英文版2026年2期

HINA R&P TECHNOLOGY AND EQUIPMENT

for waste plastic pyrolysis has been developed and integrated, the problem. Cross-validation techniques are used to ensure
providing theoretical support for AI-assisted waste plastic the generalization ability of the model and avoid overfitting.
pyrolysis and achieving and enhancing the level of intelligence Model parameters are adjusted through methods such as grid
in waste plastic pyrolysis. search, random search, or Bayesian optimization to improve
The fundamental theoretical system of AI-assisted prediction accuracy. Data is collected in real-time and
pyrolysis of waste plastics is a scientific framework that compared with model predictions. Real-time feedback is used
integrates knowledge from multiple disciplines, including to adjust process parameters and continuously optimize the
artificial intelligence, chemical engineering, environmental pyrolysis process. Interdisciplinary innovation is carried out by
science, and materials science. combining knowledge from multiple fields such as chemistry,
With the development of AI waste plastic pyrolysis, mechanics, and electrical engineering. New materials, sensors,
the theory of AI waste plastic pyrolysis is also evolving control systems, etc. are explored and applied to enhance the
and developing in practical applications. Researchers have intelligence level of the pyrolysis process.
employed modern scientific methods such as scientific 1.3 Development strategy for basic theories
experimental design, data analysis techniques, machine of model fitting and prediction
learning algorithms, and machine learning models to explore By collecting a large amount of experimental data,
the theory of AI waste plastic pyrolysis. The exploration of including pyrolysis temperature, pressure, feedstock properties,
the basic theory of AI waste plastic pyrolysis mainly revolves product characteristics, etc., key features that have a direct
around the following aspects: impact on the pyrolysis process are selected, mathematical
1.1 Development strategy for the basic models are established, and appropriate machine learning or
theory of mathematical models for pyrolysis deep learning models are selected for training. The collected
processes dataset is used to train the selected models, and key parameters
Through theoretical computational chemistry and such as product distribution, yield, and energy consumption
molecular dynamics simulations, we aim to gain a deep under different conditions are predicted. The models are
understanding of the cracking mechanisms of plastic continuously adjusted and optimized in combination with
molecules under different temperatures, pressures, and practical application backgrounds.
oxygen concentrations, including radical chain reactions and 1.4 Development strategy for basic theories
thermodynamic equilibria. We will develop and optimize of feature selection and dimensionality
mathematical models for the pyrolysis process, such as reaction reduction techniques
kinetics models and heat and mass transfer models, to predict By selecting the feature subset that contributes the most
the product distribution and yield under various conditions. to the model's predictive ability from the original feature set,
1.2 Development of basic theory for pyrolysis combined with data cleaning, handling missing values, and
process optimization other dimensionality reduction techniques, machine learning
By collecting key parameters during the pyrolysis can identify the factors that have the greatest impact on the
process, such as temperature, pressure, reaction time, feedstock pyrolysis process. This helps simplify the model, enhance
type, and product composition, outliers and missing values interpretability and practicality, effectively improve the
are removed and stored in an efficient data system. Based on predictive accuracy and efficiency of the AI waste plastic
the physicochemical principles of the pyrolysis process, key pyrolysis system, and reduce the consumption of computational
features that have a direct impact on target optimization are resources
selected or constructed. Statistical methods or machine learning 1.5 Development strategy for basic theories
algorithms are used to screen the features that have the greatest of online learning and adaptive adjustment
impact on optimization results. Appropriate machine learning Through machine learning algorithms, key parameters
or deep learning models are selected based on the nature of of the pyrolysis process are adjusted online in real-time based

·2· Vol.52，No.2