School of Energy and Power Engineering, Beihang University
School of Environmental Science and Engineering, Hainan University
Research Center for Advanced Energy and Carbon Neutrality, Internation Research Institute forMultidisciplinary Science, Beihang University

废塑料热解制油(如航空燃料)与合成气(一氧化碳和氢气)是回收利用废塑料的重要途径。原料类型、工况条件等因素对热解产物产生重要影响,这使得热解过程的反应机理较为复杂,需通过大量实验数据探究反应规律,且实验成本高。机器学习具有数据处理量大、便于提炼统计规律的优势,可降低成本与研究难度。因此,基于多种机器学习算法,利用以无催化和分子筛催化剂催化为主体的数据构建模型,对原料热解进行研究。结果表明,在选取的几种模型中,梯度提升回归算法(GBR)对热解油产率的预测具有最好的拟合性能(R2=0.91,RMSE=7.78),而自适应提升算法(AdaBoost)对热解气产率的预测具有最好的拟合性能(R2=0.83,RMSE=6.42),因此用于预测反应条件。通过排列重要性分析与单依赖性分析,在加热速率约为20℃/min、温度为500℃时,油料的产率较高。同时,对热解油产率与反应温度、加热速率和反应时间3个条件进行了双依赖性分析。量化了加热速率、热解温度等反应条件对废塑料热解油气产率的影响,为废塑料回收的生产实践提供了理论基础。

The conversion of waste plastics into oil (aviation fuel) and syngas(carbon monoxide andhydrogen) through pyrolysis offers an efficient means of recycling and reusing these plastics. Factorssuch as feedstock types and working conditions have an important impact on pyrolysis products, whichmakes the reaction mechanism of pyrolysis process more complex, so it is necessary to explore the reac⁃tion nature through a large number of experimental data, and the experimental cost is high. Machinelearning has the advantages of large data processing volume and easy extraction of statistical laws,which can reduce costs and research difficulties. A machine-learning approach was applied to utilizedata from non-catalytic and molecular sieve catalytic processes and to build a model for analyzing rawmaterial pyrolysis. The Gradient Boosting Regression (GBR) algorithm has the best fitting performancefor predicting oil yield (R = 0.91, RMSE = 7.78), while the adaptive boosting algorithm (AdaBoost)has the best fitting performance for predicting gas yield (R = 0.83, RMSE = 6.42), enabling accurateprediction of reaction conditions. It was found that optimal oil yield occurred at a heating rate of approx⁃imately 20 ℃ / min and a temperature of 500 ℃ through importance ranking and single dependency an⁃alyses. Additionally, a dual dependency analysis of oil yield with reaction temperature, heating rate,and reaction time was conducted. This study quantified the effects of heating rate, pyrolysis temperatureand other reaction conditions on the oil and gas yield of plastic pyrolysis, which provides a theoreticalbasis for the production practice of waste plastic recycling.

国家自然科学基金资助项目(U23B20166,22206011)

谌思罕,袁志龙,王晔,等.基于机器学习的废塑料热解制燃料模型构建研究[J].能源环境保护,2024,38(5):127-134.

CHEN Sihan, YUAN Zhilong, WANG Ye, et al. Study of model construction of fuel production from waste plasticpyrolysis based on machine learning[J]. Energy Environmental Protection, 2024, 38(5): 127-134.