Asymmetric Probability Distribution Function-Based Distillation Curve Reconstruction and Feature Extraction for Industrial Oil-Refining Processes

A distillation curve is an essential
property for petroleum. Its
features are beneficial for the modeling and optimization of oil-refining
processes. To capture these features with a small number of parameters,
an asymmetric probability distribution function-based distillation
curve reconstruction and feature extraction method is proposed for
the industrial oil-refining process. In our research, the expressive
power of several frequently used probability distribution functions
are first tested with some available distillation data. According
to the statistics, the Kumaraswamy distribution function, one of the
asymmetric probability distribution functions with four parameters,
is identified as the best. Because not all distillation data are directly
obtainable in the industry, the total probability theory-based data
synthesis technique is adopted to estimate the key distillation points
of unsampled streams, especially for the unmeasurable intermediate
products at the outlet of a reaction system. Along with the distillation
curve reconstruction, features of the synthetic distillation data
are extracted by optimizing the parameters of the Kumaraswamy distribution
function using the state transition algorithm. Industrial experiments
were carried out to demonstrate the effectiveness of our proposal.

蒸馏曲线（distillation curve）是石油的一项核心属性，其特征对于炼油过程的建模与优化具有重要作用。为以极少量参数捕捉此类特征，本文针对工业炼油过程提出了一种基于不对称概率分布函数（asymmetric probability distribution function）的蒸馏曲线重构与特征提取方法。本研究首先利用若干现有蒸馏数据集，测试了多种常用概率分布函数（probability distribution function）的表征能力；经统计分析，四参数不对称概率分布函数之一的库玛拉斯瓦米分布函数（Kumaraswamy distribution function）被确定为最优选择。由于工业场景中并非所有蒸馏数据均可直接获取，本文采用基于全概率理论（total probability theory）的数据合成技术，对未采样流股的关键蒸馏点进行估算，尤其针对反应系统出口处难以测量的中间产物。结合蒸馏曲线重构环节，研究通过状态转移算法（state transition algorithm）优化库玛拉斯瓦米分布函数的参数，以此提取合成蒸馏数据的特征。本研究开展了工业实验，验证了所提方法的有效性。