分布式智能养老养护系统

沈阳大学硕士学位论文I摘要随着老龄化社会的逐步趋近，老年人的养老问题显得越来越重要。目前国内大部分养老院设施落后、人员管理不规范、采用简单的信息管理系统，子女不能知道老人在养老院是否被很好的照顾。分布式智能养老系统正是为解决传统养老院分散的管理系统信息孤立、养老服务设施不能满足实时看护的需求提出的。系统结合了RFID技术和视频联动监控的优势，实时监护老人的状况，子女也能随时了解父母的健康状况。在养老院中给老人佩戴RFID腕带，实时采集老人的体征信息。由于RFID腕带不停的被读写，系统中存在大量的数据。同时，养老院的监控体系不断采集老人的图像信息。为了能知道老人是否被很好的照顾，需要对两类信息进行处理分析。数据融合技术正是解决多属性数据得到最优决策的有效工具。国内外针对数据融合的算法有很多，其中神经网络算法具有良好的容错能力和自适应性，对系统的先验概率分布要求较小，可以处理不完备、不精确的信息。结合分布式智能养老系统数据的多源性，海量性特点，要求数据处理时的实时性，准确性的特点，以及BP神经网络具有实现简单以及在一定范围内具有较高识别精度的特点，本文采用BP神经网络算法作为研究。大量的原始数据直接作为神经网络的输入，导致网络结构复杂化，训练时间也变长。冗余数据和噪声数据也对融合的性能造成影响。粗糙集对处理不完整、不确定及冗余多的数据有很强的优势，能有效地提取最少、最佳特征，运算量小且精度高，不存在因个人主观性而导致的不准确性。综合两者的优点，先利用粗糙集对数据进行约简，将约简后的数据作为神经网络的输入，降低网络复杂度，减少训练时间，同时使系统具有一定的容错和抗干扰能力。图像信息经过处理，用K值代表人体的姿态特征。粗糙集约简前，首先进行数据的离散化处理。传统的BP算法采用基于梯度的最速下降法，使网络具有易陷入局部极小值、学习过程收敛速度缓慢等缺点。利用附加动量法对传统BP算法进行改进，使梯度在同一方向上增加权值和阈值的修正量，保证算法向收敛方向进行。关键词：分布式智能养老系统，数据融合，粗糙集，BP神经网络沈阳大学硕士学位论文II沈阳大学硕士学位论文IIITheApplicationofDataFusionTechniqueBasedonBPNeuralNetworktoDistributedIntelligentPensionSystemAbstractWithanageingsocietygraduallyapproaches,thepensionissueofelderlyhasbecomeincreasinglyimportant.Familiesdon’tknowwhethertheirolderpeoplearewelltakencareofbecauseofthenursingduetopoorfacilitiesinthenursinghomes,lackofstandardizationandcombinationofonlyfewinformationmanagementsystems.Thedistributedintelligentpensionssystemisadoptedtosolvetheproblemmeetingrealtimecareneedwhichisunattainableusingscatteredandisolatedinformationmanagementsystemsinpresentnursinghomesandold-ageservicefacilities.Withthisdistributedsystemofficerscannotonlymonitorphysicalconditionsinreal-time,butcanalsoprovideinformationtofamiliesaboutthehealthoftheelderlythroughthecombineduseofRFIDtechnologyandvideoCoordinatedSupervisoryControlmonitoring.PeopleinthenursinghouseswearRFIDwristbandstogathersignsinformationinrealtime.ThereprovideshugedatatothesystemduetoRFIDwristbandsbeingreadingcontinuously.Simultaneouslythevideosurveillancesystemcollecttheimagesoftheelderlyinreal-time.Inordertoknowwhethertheelderlyarewelltakenof,thesystemneedstodealwiththesedata.Datafusionisapowerfultooltosolvetheproblemofgetstheoptimaldecisionfrommultipleattributedata.Therearedifferentkindsofalgorithmofdatafusionhomeandabroad.Theneuralnetworkshaveadvantageatfaulttoleranceandresilience,whichmeanshavelesserrequirementsofthesystem’spriorprobabilitytodealwithincompleteorinaccurateinformation.Combinedwithmulti-sourceandmasscharacteristicsofdatainthedistributedsystemanddataprocessingreal-timeaccuracycharacteristics,alsoaccordingtobasicusewhichhascharacteristicsofhighrecognitionaccuracytosomeextent,BPneuralnetworkalgorithmischoseninthisarticle.However,takinglargeamountsofdatadirectlyasinputtotheneuralnetworkbeforedatafusionmakestheneuralnetwork’sinputdimensionbecometoohighandstructurallycomplicatedrequiringalongertrainingtime.Redundantpropertiesandnoiseshavebadinfluenceontrainingresults.RoughSethasastrongadvantageatdealwithincomplete,uncertainandredundantdatawhichmeansnotonlycaniteffectivelygetthebestfeaturewithlessoperationandhighprecisionbutalsoremovetheinaccuracymadebypersonalsubjectivity.Integratingadvantagesofbothbycombingroughsetwithneuralnetworks.First,usingrough沈阳大学硕士学位论文IVsetfordatareductionandletthereductionofdataasinputtotheneuralnetwork.Thewholeprocessreducesnetworkcomplexity,therebyreducingtrainingtimenetworkwhilemakingthewholesystemacertaindegreeoffaulttoleranceandanti-jammingcapabilities.Theattitudeofthehumanbodyfeaturesreducedtokrepresentsafterdealwithimagesinformationbasedonforegroundimageinformationprocessing.Firstthedataneedstobeprocessedbeforedataprocessing.ThetraditionalBPalgorithmbasedonthegradientofthesteepestdescentmethodwhichmakesthenetworkeasilytrappedintolocalminimaandslowlyconvergenceoflearningprocess.InthispaperselectadditionalmomentumalgorithmischosentoimprovethetraditionalBPalgorithm.Theadvancedalgorithmincreasestheamountofamendmentstotheweightsandthresholdsinthesamegradientdirectiontoensurealgorithmwentinconvergencedirection.KeyWords:Distributedpensionintelligentsystem，Datafusion，Roughset，BPNeuralnetworkalgorithm沈阳大学硕士学位论文1目录摘要...................................................................................................................................IABSTRACT...........................................................................................................................III1绪论......................................................................................................................................11.1研究背景.......................................................................................................................11.2问题的提出与解决思路..............................................................................................21.2.1问题的提出...........................................................................................................21.2.2解决思路...............................................................................................................21.3研究意义和研究现状.........................................................................