第24卷网络预印版(小五宋体)农业工程学报Vol.242008年网络预印版TransactionsoftheCSAE20081基于遗传神经网络的番茄光合速率模型胡瑾,何东健※,张海辉王智永王小春(西北农林科技大学机械与电子工程学院,陕西杨凌712100)摘要:光合速率模型是开展光环境细粒度调控是理论基础,已成为是决定光环境调控成效关键,因此,如何基于智能算法构建高效精准的光合速率模型已成研究关键。由于BP神经网络算法初始权值为任意随机数,在建立回归预测模型容易陷入局部平坦区,现有型(基于神经网络光合速率模的)存在收敛速度慢、训练时间长等的问题。针对上述问题,本文提出了一种基于遗传算法的启发式神经网络黄瓜的光合速率模型,通过遗传算法优化初始权值,有效提高模型的性能。模型建立多因子嵌套试验获得825组黄瓜幼苗光合速率测试数据的基础上,通过BP网络结构构建和数据预处理,基于遗传算法进行网络权值和阈值优化,以及基于LM训练法的网络训练,建立基于遗传神经网络的黄瓜光合速率模型。在此基础上,进一步对比分析本文建立的遗传神经网络的光合速率预测模型与未优化的神经网络预测模型的训练性能和模型精度。试验结果表明,遗传神经网络的光合速率预测模型其训练效果与模型精度均优于神经网络预测模型,模型预测值与实测值间的相关系数为0.987,光合速率绝对误差小于±0.52/umolms关键词(小五、黑体):农业;工程;编辑;科技;论文(小五、楷体。)中图分类号:S126文献标志码:A文章编号:(小五、黑体)TheModelofTomatoPhotosyntheticRateBasedonGeneticNeuralNetworkAbstractThephotosyntheticratemodelisthetheoreticalbasistocarryoutthelightenvironmentfine-grainedcontrol,whichhavebecomethekeytodeterminetheeffectivenessoflightenvironmentcontrol.Therefore,howtobuildefficientprecisionforphotosyntheticratemodelbasedonintelligentalgorithmhasbecomecriticalonthestudy.DuetotheinitialweightsofBPneuralnetworkalgorithmforarbitraryrandomnumber,theestablishmentofregressivepredictionmodeliseasilytrappedinpartialflatarea.Theexistingtypeofphotosyntheticratemodelbasedonneuralnetworkexiststheproblem,suchasslowconvergencespeed,longtrainingtimeandsoon.Inviewoftheaboveproblem,thispaperpresentsaphotosyntheticratemodelofheuristicneuralnetworkforcucumberbasedongeneticalgorithm.Theperformanceofmodelcanbeeffectivelyimprovedbygeneticalgorithmoptimizinginitialweights.Themodelestablishesthemulti-factornestingexperimenttoobtain825groupofcucumberseedlingphotosynthesisratetestdatainthefoundation.ByBPnetworkstructureconstructionanddatapreprocessing,basedonthegeneticalgorithmtooptimizenetworkweightsandthreshold,aswellasbasedonLMtrainingmethodofnetworktraining,thephotosyntheticratemodelofheuristicneuralnetworkforcucumbercanbeestablished.Onthisbasis,trainingperformanceandmodelprecisioncanbefurthercomparedandanalysedbetweenthephotosyntheticrateofgeneticneuralnetworkpredictionmodelonthispaperandunoptimizedneuralnetworkpredictionmodel.Thetestresultsshowedthatthetrainingeffectsandaccuracyforthegeneticneuralnetworkpredictionmodelofphotosyntheticratewerebetterthantheneuralnetworkpredictionmodel.Thecorrelationcoefficientbetweenthemodelpredicteddataandmeasureddatais0.987,andtheabsoluteerrorofphotosyntheticrateofislessthan±0.5.第一作者,第二作者,第三作者,等.中文标题[J].农业工程学报,2014,30():-(小五、黑体)第一作者,第二作者,第三作者,etal.英文标题[J].TransactionsoftheChineseSocietyofAgriculturalEngineering(TransactionsoftheCSAE),2014,30():-.(inChinesewithEnglishabstract)(小五、TimeNewsRoman)0引言光是植物生长过程中不可或缺的因素之一,由于受覆盖材料、灰尘、太阳高度角及倾角结构遮光等影响,设施光照在秋末到早春季节难以满足作物生长要求,导致作物生长发育减缓,造成落叶、发花数量少、花形花色不正、坐果率低等问题,人工补光技术作为设施光环境调控重要手段,已成为近期研究热点[]。但是现有补光技术未综合设施小气候条件考虑其对光合作用的影响,可能存在补光过度或补光不足。因此,研究多因子融合下的光合速率建模方法,构建多环境因子关联光合速率模型,并以此为理论基础开展是光环境细粒度调控已成为是决定光环境调控成效的基础性问题。2农业工程学报2008年IntroductionLightisoneoftheindispensablefactorsintheprocessofplantgrowth.Duetotheinfluencesuchas,thecovermaterial,dust,thesunaltitudeangleanddipanglestructureshading,Lightingfacilitiesintheautumntoearlyspringmeetstherequirementofcropgrowth,causingsomeproblemssuchas,cropgrowthslowed,deciduous,fewflowers,theabnormalflowercolorandshape,lowfruitrate.Artificiallightasanimportantmeansofcontrollingthefacilitiesenvironmenthavebecomehotspotsinrecentresearch.Buttheexistinglightingtechnologyarenotcomplexthemicroclimateconditiontoconsideritsinfluenceonphotosynthesis,andtheremaybeexcessiveorinsufficientfill-inlight.Sothephotosyntheticrateunderthemultiplefactorfusionmodelingmethodbuildmoreenvironmentalfactorassociatedphotosyntheticratemodel,andtheoreticalbasistocarryoutthefine-grainedcontrolthelightenvironmenthasbecomeisabasicissuetodeterminelightenvironmentalregulationperformance.在光合速率模型方面国内外已开展了相关研究,已建立了包括直角双曲模型、非直角直角双曲模型和指数关系模型在内的多种光合速率模型[26,27]。在此基础上,提出了基于电子输运的光合速率模型[28]、光合速率稳态模型[29],不同氮素下光合作用模型等相关模型研究[31]。上述研究为光合速率建模提供了良好的理论基础,但模型输入变量为生理参数,在常规试验和生产过程中模型参数不易获取,难以直接应用于设施作物光环境调控。近年来,科研工作者开展了以环境参数为变量的光合速率模型研究,出现了番茄光合速率的影响函数[32],光合速率类卡方模型[21];番茄生长发育的非线性动态模拟模型[33]。上述研究均采用多元回归或者线性拟合的方法进行光合速率建模,考虑了不同环境因子之间的关联性,提高了模型的适应性和准确性,但是仍存在拟合度低,拟合公式复杂等不足,不适用于融合多元环境因子的光合速率建模。Intermsofphotosyntheticratemodel,therelatedresearchhavebeenconductedathomeandabroad,establishedavarietyofphotosyntheticratemodelincludingrectangularhyperbolicmodel,thenon-orthogonalrectangularhyperbolicmodelandexponentialrelationmodel(26,27).Onthisbasis,therelatedmodelresearchhavebeenputforward,suchasbasedonthemodelsofthephotosyntheticrateelectrontransport[28]andphotosyntheticrateofsteady-statemodels[29],photosynthesisunderdifferentnitrogenmodels[31].Theabovestudyprovidesagoodtheoreticalbasisformodelingofphotosyntheticrate,butmodelinputvariablesforthephysicalparametersofmodelparametersintheprocessofroutinetestandproductionisnoteasytoget,hardtodirectlyappliedtothefacilitiesandcroponthelightenvironmentregulation.Inrecentyears,theresearcherscarriedouttostudythephotosyntheticratemodelofenvironmentalparameterasavariable,appearingtheinfluencefunctionsoftomatophotosynthesisrate,pho