权重提取与Dempster多重融合的凝汽器真空预测  

Condenser vacuum prediction based on weight extraction and Dempster multiple fusion

作　　者：夏琳琳[1] 台金娟 刘惠敏[2] 王丹[1] 文磊[1] 

机构地区：[1]东北电力大学自动化工程学院,吉林吉林132012 [2]青岛农业大学机电工程学院,山东青岛266109

出　　处：《沈阳工业大学学报》2015年第3期329-334,共6页Journal of Shenyang University of Technology

基　　金：吉林省科技厅青年科研基金资助项目(20130522171JH)

摘　　要：为了解决单一网络预测结果不准确的问题,提出一种由BP、Elman及RBF三网络组合的预测模型,并引入模糊软集理论进行"判断证据"的权重提取以及D-S的多证据融合.以某电厂连续4天实测的现场参数构成样本空间,经主成分分析降维及权重提取后,采用Dempster组合规则下置信函数三重融合结果对随后一天的真空值进行预测.结果表明,与单一网络预测模型相比,组合预测模型的平均绝对误差和均方根误差均显著减小,融合精度更高.In order to solve the prediction uncertainty of single network,a combined prediction model composed of three networks of BP,Elman and RBF was proposed,and the weight extraction of judge evidences and the fusion of multi-evidences based on D-S theory were performed through introducing the fuzzy soft set theory. In addition,the sample space was established with the measured field parameters in 4continuous days for certain power plant. After the procedures of dimensionality reduction and weight extraction,the vacuum value prediction in the following day was performed with the triple fusion results of belief functions with Dempster combining rule. The results showthat compared with the single network prediction model,the average absolute error and RM SE of combined prediction model obviously reduce,and the fusion accuracy is higher.

关 键 词：凝汽器真空值 神经网络 DEMPSTER组合规则 模糊软集 主成分分析 组合预测模型 权重提取 数据融合 

分 类 号：TK264.11[动力工程及工程热物理—动力机械及工程]