基于神经网络和D-S证据理论的气液两相流流型识别方法.pdf

《基于神经网络和D-S证据理论的气液两相流流型识别方法.pdf》由会员上传分享，免费在线阅读，更多相关内容在应用文档-天天文库。

1、第57卷第3期化工学报VOl.57NO.32006年3月JOurnalOfChemicalIndustryandEngineering(China)March2006妥、妥妥妥妥、妥妥研究论文妥基于神经网络和D-S证据理论的气液妥、妥妥妥妥、妥两相流流型识别方法周云龙!孙斌(东北电力学院动力系9吉林省吉林市132012)摘要:提出一种运用神经网络和D-S(Dempster-Shafer)证据理论的多特征信息融合的气液两相流流型识别方法.对压差波动信号进行4层小波包分解9提取各频带信号的小波包能量和信息熵构造两个特征向量9再利用统计和分形理论提取压差波动信号的3个统计参数和4个分形参数作为另一

2、个特征向量9然后将这些特征向量送入改进的BP神经网络进行训练9从而实现对流型的识别.以初始识别结果作为彼此独立的证据9根据D-S证据融合规则进行融合处理9得到最终的识别结果.以水平管内空气-水两相流流型识别为例9说明了该方法的具体实现过程.结果表明9多特征信息融合比单一特征的识别方法具有更高的识别率.关键词!气液两相流9小波包变换9BP神经网络9D-S证据理论9流型识别中图分类号!O359.1文献标识码!A文章编号!0438-1157(2006)03-0607-07ldentificationmethodof9as-liCuidtWo-PhasefloWPatternsbasedonneur

3、alnetWorkandD-SevidentialtheoryZHOUYunlon9"SUNBin(DePartmentofThermalPoIer9NortheastChinaInstituteofElectricPoIerEngineering9Jilin1320129Jilin9China)Abstract!BasedOntheneuralnetWOrkandtheD-S(Dempster-Shafer)evidentialtheOry9amethOdWasprOpOsedfOridentifyinggas-liCuidtWO-phaseflOWregimes.Firstly9thed

4、ifferentialpressurefluctuatiOnsignalsWeredecOmpOsedintO4levelsbytheWaveletpackettransfOrm.WaveletpacketenergyandinfOrmatiOnentrOpyOfsignalsinvariOusfreCuencybandsWereextractedandtWOeigenvectOrsWerecOnstructedandthenthethreestatisticalparametersandfOurfractalparametersextractedbythestatisticalandthe

5、fractaltheOriesOfthedifferentialpressurefluctuatiOnsignalsWeretakenasanOthereigenvectOr.FurthermOre9theeigenvectOrsWereputintOtheimprOvedBPneuralnetWOrkandtrainedtOrealizetheflOWregimeidentificatiOn.TakingthepreliminaryidentificatiOnastheindependentevidence9afinalidentificatiOnWasObtainedaccOrdingt

6、OtheD-SevidentialfusiOnalgOrithm.Usingtheair-WatertWO-phaseflOWregimeidentificatiOninthehOrizOntalpipeasanexample9theimplementingprOcessOfthismethOdWasdescribedindetail.TheresultsshOWedthatthemethOdOfmulti-characteristicinfOrmatiOnfusiOncOuldachieveahigheridentificatiOnabilitythanthatOfsinglecharac

7、teristic.KeyWords!gas-liCuidtWO-phaseflOW9WaveletpackettransfOrm9BPneuralnetWOrk9D-SevidentialtheOry9flOWpatternidentificatiOn2005-04-08收到初稿92005-07-18收到修改稿.Receiveddate!2005-04-08.联系人及第一作者!周云龙(1960_)9男9博士9