Abstract:

It introduces the process of using XGBoost algorithm model for data processing and model prediction of straight-run asphalt properties. Based on the operating data of atmospheric and vacuum distillation units and crude oil evaluation data of refining enterprises, combined with production experience to expand the data characteristics, a data-driven regression model is established for the collected 128 asphalt product analysis data. The gradient lifting tree model is used to fully explore the correlation between crude oil properties and asphalt properties, and intelligently predict key indicator properties such as softening point and ductility(10 ℃) of asphalt products under different target penetration degrees. The coefficient of determination for predicting the softening point of asphalt is greater than 0.77. By comparison, the data features that are expanded can effectively improve the predictive ability of the model. The ability of model prediction under different data volumes is also analyzed, and with the continuous accumulation of data, the prediction potential is great. This study can provide enterprises with the crude oil ratio to produce high-grade asphalt, assist enterprises in optimizing the global production process and achieving cost reduction and efficiency increase.

Key words: crude oil properties, straight-run asphalt properties, intelligent predictive models, feature expansion, softening point prediction, ductility predicting, XGBoost

摘要：

介绍了采用XGBoost算法模型对直馏沥青性质的数据处理及模型预测建立的过程。以炼油企业常减压装置运行数据、原油评价数据为基础，结合生产经验扩展数据特征，对收集到的128条沥青产品分析数据建立数据驱动的回归模型，利用梯度提升树模型充分挖掘原油性质与沥青性质间的关联，智能预测沥青产品在不同目标针入度下的软化点、延度(10℃)等关键指标性质，其中沥青软化点预测决定系数大于0.77。经对比，拓展的数据特征能有效提高模型预测能力。同时分析了不同数据下模型预测的能力，随着数据的不断积累，模型预测潜力较大。该模型可为企业提供生产高标号沥青所需要的原油配比，协助企业优化全局生产流程，实现降本增效。

关键词: 原油性质, 直馏沥青性质, 智能预测模型, 特征扩增, 软化点预测, 延度预测, XGBoost,