基于案例推理和智能算法的石漠化治理模式优选研究

【作者】 刘占明；

【导师】 蒋忠诚； 胡宝清；

【作者基本信息】 广西师范学院 ， 自然地理学， 2011， 硕士

【摘要】 石漠化是目前岩溶学领域最突出的环境问题和研究热点。喀斯特石漠化是指在亚热带喀斯特地域环境背景下,受人类不合理活动的驱动,引起土壤严重侵蚀、基岩大面积裸露、土地生产力急剧下降、地表出现类似荒漠景观的土地退化过程。目前我国西南喀斯特地区一批石漠化生态治理项目正在试点实施,并提出了一些治理模式,但石漠化综合治理进展仍然比较缓慢,还存在不少问题。如何利用新思路、新手段与新方法准确、客观而有效地防治石漠化灾害与维护喀斯特生态环境健康安全,是跨学科的重大课题之一。基于相似性的推理越来越受到重视,出现案例推理(Case‐Based Reasoning,CBR)概念,并且基于案例推理的基本思想符合人的推理模式。许多领域基于规则的系统无能为力,但基于案例的系统却几乎不受应用领域的限制,案例推理中知识表示是以案例为基础,案例的获取比规则获取要容易,大大简化知识获取,对过去的求解结果进行复用,可以提高对新问题的求解效率。在CBR推理中,根据新案例与案例库中现有案例的相似度来确定其石漠化综合治理类型的隶属度,在推理过程中使用隶属度可以减少许多不确定性,使得分类结果更合理,对比不同方法的分类效果显示,所建立的案例库可以被重复使用。本研究对当前西南地区石漠化治理中较为成功的治理模式进行了初步分类,并初步建立了喀斯特石漠化治理案例指标体系。实现了使用GABPANN模型对石漠化治理案例进行初步推理,并取得了较为明显的效果。BP神经网络是在梯度算法的基础上推导出来的,进化学习收敛速度较慢,容易陷入局部最优值。由于BP神经网络初始权值与阈值是随机取值,导致网络容易陷入局部最优值,并且每次训练后的网络预测输出都有差别。本研究把BP神经网络与遗传算法有机结合,用遗传算法寻找网络初始最优权值和阈值,可以弥补初始权值和阈值随机取值的缺陷,使网络能够更精确的预测系统输出,从而使预测推理的结果更为准确。同时,本论文根据所建立的GA‐BP神经网络模型,在石漠化治理模式案例推理中进行了实证分析,并取得了较为理想的效果。本论文所提出的预测推理模型将对未来的西南地区喀斯特石漠化治理提供了一种新的思路和方法,可以为制定石漠化治理模式决策提供参考。更多还原

【Abstract】 Karst rocky desertification is currently the most prominent environmentalissues and research focus in field of karst. Karst rocky desertification refers to thesubtropicalkarstregionsenvironmentalbackground,drivenbyunreasonablehumanactivities,causing serious soil erosion,large areas of exposed bedrock,landproductivitysharplyreduceandsurfacesimilardesertlandscapeoflanddegradationprocess.CurrentlyagroupofsouthwestChinakarstrockydesertificationareaprojectisinthepilotimplementation,andmadeanumberofgovernance,buttherockydesertification progress is still slow,there are still many problems. How use newideasandnewmethodisaccurate,objectiveandeffectivecontrolandmaintenanceof karst rocky desertification environmental health safety. All,is one of theinterdisciplinarymajorissue.Similarity‐based reasoning more and more attention,there CBR (Case‐BasedReasoning,CBR)conceptandthebasicideaofcase‐basedreasoningisconsistentwithhumanreasoningpatterns.Rule‐basedsystemsinmanyareascandonothing,butthecase‐basedsystem,applicationsalmostwithoutlimit,case‐basedreasoninginknowledgerepresentationisbasedoncase‐based,caseforeasieraccessthantherule,greatlysimplifyingtheknowledgeacquisition forthepast.Theresultswerere‐used,canimprovetheefficiencyofsolvingnewproblems. IntheCBRreasoning,accordingtonewcasesandexistingcasesinthecasebasetodeterminethesimilarityofrockydesertificationtypesofmembership,inthereasoningprocesscanreducethe use of membership are many uncertainties,making the classification resultsmorereasonable,comparingdifferentmethodsofclassificationresultsshowthattheestablishedcaselibrarycanbereused.Thesouthwestenregionofthecurrentdesertificationcontrolmoresuccessfulmodelofgovernanceinapreliminaryclassification,andtheinitialestablishmentofindex system of karst rocky desertification control cases. Achieved using theGA‐BP‐ANNmodeloftheinitialcaseofdesertificationcontrolreasoning,andmadeamoresignificantresults. BP neural network is based on the gradient algorithm derived,evolutionaryconvergencerateisslow,easytofallintolocaloptimum.SincetheinitialweightsofBP‐neural network is a random value and the threshold,resulting in network isvulnerabletofallingintolocaloptimalvalue,andaftereachtrainingsessiontherearedifferences between the network predicted output. The BP neural network andgeneticalgorithmcombination,theinitialnetworkusinggeneticalgorithmtofindtheoptimalweightsandthresholdvalues,canmakeupfortheinitialweightsandthresholdvaluesofrandomdefects, sothenetworkcanmoreaccuratelypredictthesystemoutput,andtheresultsofreasoningmoreaccurateprediction.Meanwhile,this paper established the GA‐BP neural network base case‐based reasoning indesertification control mode in the empirical analysis,and achieved a moresatisfactoryresults.TheproposedreasoningmodelwillforecastthefuturegovernanceofSouthwestChina Karst , a new way of thinking and method for the development ofdesertificationgovernancemodelcanprovidereference.更多还原