微粒群算法优化样本集在汽油调合设计中的应用  

Optimization of training set based on particle swarm optimization in gasoline blending design

作　　者：高世博[1] 张运陶[1] 刘金迪 白春艳 郑伟 

机构地区：[1]西华师范大学应用化学研究所,四川南充637002 [2]南充炼油化工总厂,四川南充637000

出　　处：《计算机与应用化学》2007年第10期1363-1366,共4页Computers and Applied Chemistry

基　　金：四川省企业信息化项目计划:[2005-199-16]

摘　　要：提出PSO-v-SVR方法建立计算机辅助调合汽油辛烷值预测模型的新思路,采用微粒群算法(PSO)对原始样本集随机抽样并加以优化获得优化的训练集,再以v-支持向量回归(v-SVR)对样本进行训练和预测。用PSO-v-SVR方法对某炼油厂的汽油调合生产数据进行研究,用选出的最佳训练集构成的模型对44组预测样进行预测,实测辛烷值与预测值误差绝对值AE≥0.3的样本数仅为16,平均绝对误差MAE=0.293;明显优于直接用全部原始样本作训练集建模AE≥0.3的样本数26个, MAE=0.366,以及按文献[12]用前期80组样本作训练集建模AE≥0.3的样本数25个,MAE=0.350的预测结果。研究表明,本文的思路可以较大幅度提高模型预测准确性,在化工生产优化和软测量建模中具有推广应用价值。A new approach is proposed to establish a forecasting model of computer-assisted blending octane value in gasoline by PSO-v-SVR. Firstly samples are randomly selected from the original sample set and then optimized to get an optimal training set by PSO. Secondly v-SVR is used to train the optimal training set and then predict the forecasting sample set. A study was carried out with a petroleum refinery＇s productive data of blending the octane value in gasoline. 44 forecasting set is predicted with the model set by the optimal training set. The predicted values are compared with the practical producing values： AE≥0. 3 are 16 times and MAE is 0. 293, which are better than the result obtained with the model all the original samples set that AE≥0. 3 are 26 times, MAE is 0. 366,and are also superior to the reference result that AE≥0. 3 are 25 times, MAE is 0. 350. The approach could improve the accuracy of the predicting model greatly and can be generalized in optimization of the chemical production or soft-sensing modeling.

关 键 词：微粒群算法 随机抽样 优化 训练样本集 汽油调合 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] TE6[自动化与计算机技术—控制科学与工程]