基于机器学习的蜡油性质智能预测方法

Intelligent Prediction Method of Wax Properties Based on Machine Learning

蜡油是炼油工艺过程中的重要副产品,具有较高的附加值。原油经过不同工艺加工可获得多种石蜡产品,产品的关键性质(如熔点、含油量、针入度等)受其碳组成含量和结构的影响较大。产品性质的传统检测方法有色谱法(GC)、质谱法(MS)、近红外法(IR)以及色谱质谱联用法(GC-MS)等。但是这些方法检测耗时长,设备操作比较复杂,难以快速实现产品性质的预测。为能快速得到蜡油关键性质,收集整理了样本检测数据,对数据结构进行了预处理,建立了基于随机森林和XGBoost的蜡油关键性质智能预测模型,通过对蜡油3个关键性质的预测对比分析两个模型性能,XGBoost模型收敛性较好,3个关键性质训练结果的R2均已达到0.75以上,可用于指导蜡油产品性质的预测。

Wax oil is an important by-product in refining process and has high added value.Many kinds of paraffin products can be obtained from crude oil processed by different processes.The key properties of the products(such as melting point,oil content,penetration degree,etc.)are greatly affected by their carbon composition and structure.The traditional methods for detecting product properties include chromatography(GC),mass spectrometry(MS),near infrared spectroscopy(IR),and chromatography-mass spectrometry(GC-MS).However,these methods take a long time to detect,and the equipment operation is complicated,so it is difficult to quickly realize the prediction of product properties.In order to quickly obtain the key properties of wax oil,the sample detection data were collected,the data structure was preprocessed,and the intelligent prediction model of key properties of wax oil based on random forest and XGBoost was established.The performance of the two models was compared and analyzed by predicting the three key properties of wax oil,and the XGBoost model had good convergence.The R2 of the training results of the three key properties was above 0.75,so the XGBoost model can be used in the prediction of wax oil product properties.