搜索
IBP神经网络算法摘要人工神经网络，是由大量处理单元（神经元）组成的非线性大规模自适应动力系统。它具有自组织，自适应和自学习能力，以及具有非线性、非局域性，非定常性和非凸性等特点。它是在现代神经科学研究成果的基础上提出的，试图通过模拟大脑神经网络处理，记忆信息的方式设计一种新的机器使之具有人脑那样的信息处理能力。作为人工智能的重要组成部分，人工神经网络有较大的应用潜力。本文阐述了神经网络的发展、现状及其原理，介绍了神经网络在航空航天业、国防工业、制造业等诸多方面的应用。BP神经网络是目前应用较多的一种神经网络结构。它能以任意精度逼近任意非线性函数，而且具有良好的逼近性能，并且结构简单，是一种性能优良的神经网络。本文阐述了BP神经网络的基本原理，详细分析了标准BP算法、动量BP算法以及学习率可变的BP算法等几种流行的BP神经网络学习算法,详细的介绍了这几种算法的优缺点，并给出了各种算法的仿真程序，通过仿真结果对各种算法进行比较后有针对性的提出了BP算法的一种改进——变梯度BP算法。对于改进的BP算法，本文不仅从理论方面对其进行了深入的分析，还介绍了该算法的详细思路和具体过程，将算法训练后的BP神经网络运用到函数逼近中去。仿真结果表明，这种改进方案确实能够改善算法在训练过程中的收敛特性，而且提高收敛速度，取得令人满意的逼近效果。关键词:人工智能；BP神经网络；变梯度法；改进IIAbstractArtificialneuralnetwork,bythelargenumberofprocessingunits(neurons)composedoflarge-scaleadaptivenonlineardynamicsystems.Itisself-organization,adaptiveandself-learningability,aswellasnon-linear,non-local,non-steadyandnon-convexandsoon.Itisinmodernneuroscienceresearchonthebasisoftheresults,tryingtosimulatethebrainnetworkprocessing,memoryaboutthemeanstodesignanewmachinesothatitisthehumanbrain,astheinformationprocessingcapability.Asanimportantcomponentofartificialintelligence,artificialneuralnetworkshavegreaterpotentialapplications.Thispaperdescribesthedevelopmentofaneuralnetwork,thestatusquoanditsprinciples,introducedaneuralnetworkintheaerospaceindustry,defenseindustry,manufacturingandmanyotheraspectsoftheapplication.BPneuralnetworkismoreofaneuralnetworkstructure.Approachingitwithanyprecisionarbitrarynonlinearfunction,butalsohasagoodapproximationperformance,andsimplestructure,isagoodperformanceofneuralnetworks.Inthispaper,BPneuralnetworkthebasicprinciples,detailedanalysisofthestandardBPalgorithm,momentumBPalgorithmandthevariablerateoflearning,suchasBPalgorithmseveralpopularBPneuralnetworklearningalgorithm,describedindetailtheadvantagesanddisadvantagesofthesedifferentalgorithmsAndgivesavarietyofalgorithmsimulationprogram,throughthesimulationresultsofthevariousalgorithmstocomparetargetedafterBPmadeanimprovedalgorithm-BPchangegradientalgorithm.BPtoimprovethealgorithm,thepapernotonlyfromthetheoreticalaspectsoftheirin-depthanalysis,alsodescribedthealgorithmisdetailedideasandspecificprocess,methodoftrainingtouseBPneuralnetworktofunctionapproximation.Thesimulationresultsshowthatthisimprovementprogrammeistoimprovethetrainingalgorithmintheprocessofconvergencecharacteristics,andimprovetheconvergencerate,asatisfactoryapproximation.IIIKeywords:Artificialintelligence;BPneuralnetwork;changegradientmethod;improveIV目录第一章绪论·······················································································11.1人工神经网络的发展史·····································································11.2人工神经网络的应用·········································································3第二章人工神经网络的基本原理及模型··········································72.1神经网络构成的基本原理·································································72.1.1人工神经元模式···········································································72.1.2连接权值····················································································72.1.3神经网络状态··············································································82.1.4神经网络的输出···········································································82.2神经网络的结构················································································82.3神经网络的特点··············································································102.4神经网络的学习方式·······································································112.5几种典型的神经网络·······································································11第三章BP神经网络算法的改进及其仿真研究································153.1BP算法的数学描述········································································153.2BP网络学习算法············································································183.2.1标准BP算法·············································································18V3.2.2动量BP算法·············································································203.2.3学习率可变的BP算法·································································213.3BP算法的缺陷···············································································223.4BP算法的一种改进——变梯度BP算法·······································233.4.1共轭梯度法················································································233.4.2改进共轭梯度法··········································································243.5BP网络应用实例············································································263.5.1一般BP算法及其改进算法训练过程的不同·····································263.5.2BP神经网络的函数逼近······························································27结束语································································································28参考文献····························································································29致谢····························································································30附录····························································································301第一章绪论