| 异质多模型集成的辛烷值和抗爆系数分级建模 |
| |
| 引用本文： | 李炜,王亚丽,王晓明,李亚洁,梁成龙.异质多模型集成的辛烷值和抗爆系数分级建模[J].石油学报(石油加工),2023,39(2):380-391. |
| |
| 作者姓名： | 李炜 王亚丽 王晓明 李亚洁 梁成龙 |
| |
| 作者单位： | 1. 兰州理工大学 电气工程与信息工程学院，甘肃 兰州 730050；
2. 甘肃省工业过程先进控制重点实验室，甘肃 兰州 730050；
3. 兰州理工大学 电气与控制工程国家级实验教学示范中心，甘肃 兰州 730050；
4. 中国石油 兰州石化分公司 油品储运厂，甘肃 兰州 730060 |
| |
| 基金项目： | 甘肃省青年博士基金项目（2021QB-044）资助 |
| |
| 摘 要： | 针对成品汽油调和过程中研究法辛烷值(RON)和抗爆系数难以实时获取，且传统未考虑二者关联的单一模型精度和适应性不足等问题，提出了2种异质多模型集成的分级预测建模方法(方法Ⅰ和方法Ⅱ)。先采用K折交叉验证法对建模算法进行参数调整与算法优选，分别建立了5个一级辛烷值和二级抗爆系数异质子模型。方法Ⅰ利用加权平均法对异质子模型进行线性集成，并为提升集成模型性能，对目标函数引入L1、L2、L1&L2等3种正则化约束，以此确定子模型最优权重；考虑线性集成可能对非线性过程适应性不足，方法Ⅱ基于堆叠思想建立了非线性集成预测模型。经使用工业生产数据仿真实验研究表明，较传统单一模型以及加权平均模型，考虑抗爆系数对辛烷值依赖提出的2种异质集成分级建模方法，具有更优的性能，可用于成品汽油调和过程中辛烷值和抗爆系数的准确预测，有望解决后期配方模型建立和优化控制的数据缺失问题。
|
| 关 键 词： | 集成 预测 模型 加权平均 堆叠 异质 |
| 收稿时间： | 2021-12-29 |
Hierarchical Modeling for Octane Number and Anti-Knock Coefficient Based on Heterogeneous Multi-Model Integration |
| |
| LI Wei,WANG Yali,WANG Xiaoming,LI Yajie,LIANG Chenglong.Hierarchical Modeling for Octane Number and Anti-Knock Coefficient Based on Heterogeneous Multi-Model Integration[J].Acta Petrolei Sinica (Petroleum Processing Section),2023,39(2):380-391. |
| |
| Authors: | LI Wei WANG Yali WANG Xiaoming LI Yajie LIANG Chenglong |
| |
| Affiliation: | 1.College of Electrical and Information Engineering, Lanzhou University of Technology, Lanzhou 730050, China;
2. Key Laboratory of Gansu Advanced Control for Industrial Processes, Lanzhou 730050, China;
3. National Demonstration Center for Experimental Electrical and Control Engineering Education, Lanzhou University of Technology, Lanzhou 730050, China;4. Oil Storage and Transportation Plant, PetroChina Lanzhou Petrochemical Company, Lanzhou, 730060, China |
| |
| Abstract: | In view of the difficulty in obtaining the research octane number (RON) and anti-knock coefficient during the blending of finished gasoline, as well as the low accuracy and poor adaptability of the traditional single model that fails to consider the correlation between octane number and anti-knock coefficient, this paper proposes two hierarchical predictive modeling methods based on heterogeneous multi-model integration (Method Ⅰ and Method Ⅱ). The K fold cross validation method was adopted for parameter adjustment and optimal selection of modeling algorithms, based on which five heterogeneous submodels for first-order octane number and second-order anti-knock coefficient were established. Specifically, in Method I, the weighted average method was adopted for linear integration of heterogeneous submodels; to improve the performance of the integrated model, three regularization constraints (L1, L2 and L1&L2) were introduced into the objective function, thereby determining the optimal weight of each submodel. Considering that the linear integration may fail to adapt to nonlinear processes, a nonlinear integrated predictive model was established based on the stacking idea in Method Ⅱ. A simulation experiment was performed using industrial production data. The experimental results show that the two hierarchical modeling methods proposed based on heterogeneous integration and considering the dependence of anti-knock coefficient on octane number have better performance than the traditional single model and weighted average model, which can accurately predict octane number and anti-knock coefficient during the blending of finished gasoline, and is expected to solve the problem of data missing in later formulation model establishment and optimization control. |
| |
| Keywords: | integration prediction model weighted average stacking heterogeneous |
|
| 点击此处可从《石油学报(石油加工)》浏览原始摘要信息 |
|
点击此处可从《石油学报(石油加工)》下载全文 |
|
