Algorithm of removing outliers in SVR

The outlier and the measurement that an outlier does not fit the theoretical model in the regression problems are defined. The relationship between the theoretical model and the regression model in the regression problem is analyzed. An approximate theorem is proposed and verified by deleting outlier one by one to construct SVR to approximate the theoretical model. An algorithm of detecting outliers in the SVR problems is constructed based on the approximate theorem. The theoretical analysis of the convergence and effectiveness of the proposed algorithm is given. Then, the step-by-step search algorithm is introduced to improve the outlier removing algorithm to remove outliers in SVR with large-scale samples. The theoretical analysis shows that the improved algorithm is convergent and effective. Finally, the samples produced by two test functions and the samples in UCI data set are used for simulation, and the results show that the proposed algorithm is effective and robust.     

基于SAGA-SVR的马铃薯贮藏库温度预测方法

马铃薯贮藏库温度受室外温度、室内马铃薯呼吸释放温度、通风降温等因素的影响难以准确预测,提出了一种改进遗传算法(Genetic Algorithm,GA)优化支持向量回归机(Support Vector Regression,SVR)的马铃薯贮藏库温度预测方法。该方法针对支持向量回归机参数难以选择、容易陷入局部极小的缺点,引入了具有并行性、全局搜索能力强的GA算法,结合局部搜索能力强的模拟退火算法(Simulated Annealing,SA),实现支持向量回归机的自动寻优。以新疆某农产品加工公司马铃薯贮藏库实测温度数据为样本,建立SAGA-SVR马铃薯贮藏库温度预测模型,进行贮藏库温度准确的预测。仿真结果表明,与GA-SVR、反向传播(Back Propagation,BP)温度预测模型的预测结果相比较,SAGASVR预测结果优于GA-SVR、BP预测结果,具有良好的预测效果。该预测方法较好地解决了系统非线性、小样本等问题,为类似应用场合地温度预测提供参考。     

基于时间序列SVR模型的玉米价格预测研究

研究旨在运用时间序列和支持向量回归(support vector regression,SVR)理论,建立短期玉米价格预测模型,为玉米市场监测预警提供技术支持与决策依据。本研究根据玉米价格序列波动的非线性特征,以河南省2010年1月1日—2019年6月30日的月度数据为研究对象,结合混沌性时间序列与SVR理论,研究一种短期玉米价格预测模型。通过相空间重构的方法对价格序列进行处理,利用重构后的数据训练模型,运用网格交叉验证(GridSearchCV)对研究模型进行优化。得到一种基于时间序列支持向量回归的玉米价格预测模型。将研究模型的预测结果进行对比分析,结果显示,研究模型的预测值更贴近真实值,其平均相对误差(MRE)和均方根误差(RMSE)分别为0.006和20.57,优于传统支持向量回归模型的预测结果。研究模型可以为玉米价格预测提供方法依据,且相较于传统预测方法在精度方面具有优势。     

基于主成分与支持向量机的邵阳县烟草亩产量预测

为探索准确预测邵阳县烟草产量的方法,首先对邵阳县70个植烟区土壤样本中的碱解氮、有效磷、速效钾等19个养分指标进行主成分分析,得出邵阳县烟草产量主要受有机质、有效锌、有效硼、有效锰、有效硫、交换性钙、全钾、有效铁和速效钾等9个养分含量的影响。在此基础上,基于支持向量机回归算法SVR对70个植烟区的烟草产量进行回归预测。结果发现,以主成分分析后的9个养分指标作为特征变量预测得到的烟草产量的均方误差明显小于以19个养分指标作为特征变量预测得到的烟草产量的均方误差。同时,对比SVR算法和随机森林回归算法发现,SVR算法的预测精度明显优于随机森林回归算法。基于主成分与支持向量机的回归算法是预测邵阳县烟草产量的有效方法。     

Application of partial least squares regression in the forecast of ground subsidence

Taking into account many influence factors of ground subsidence induced by underground exploitation,based on partial least squares multinomial regression,a forecast analysis on the maximum of ground subsidence is carried out.Taking height,depth,obliquity of coal clay and rigidity coefficient as independent variables,and maximum of ground subsidence as dependent variable,the forecast model of maximum of ground subsidence is obtained.It is found that,Press residual value decreases with the increase of number of latent variables,and the number of latent variables is four by Press residual value versus number of latent variables.The normal regression coefficient of height is the largest in the four influence factors,and this indicates that the influence of height is the largest on maximum of ground subsidence.The determination coefficient of forecast model obtained in this paper is 0.915 7,the error of forecast model is ±10.41%.The following conclusion can be drawn that the model based on partial least squares multinomial regression is a better and feasible non linear method.     

ARMA模型与回归模型的在农药用量预测中的应用

近年来,中国环境污染越来越严重多地区出现严重的虫灾现象,引起人们对残留农药影响农产品安全的广泛关注。为了更加准确的预测农药需求,在预测模型的基础上再对残差序列做ARMA模型对预测模型进行调整。首先建立农药用量与时间的回归模型,通过分析残差序列是否为白噪声,对于非白噪声的残差序列建立ARMA模型。结果表明：单一的回归模型均方残差为25.39,而ARMA模型与回归模型结合后的均方残差为13.79,均方残差减少了11.60,提高了预测精度。调整后的模型是显著的,估计的参数也比较合适。该模型比常用的Logistic 模型的均方残差降低了17.73,有效的提高了预测精度。2012、2013、2014年中国农药用量预测分别为181.068万t,183.64万t,187.96万t。     

Experiment Investigation on Separating Characteristic for Heavy Metal in Cumulate MSW Under 135 Meters Submersed Area of Three-gorge Reservoir

The separating characteristic of element in cumulate MSW is a very important base to investigate the synthetically managing technology and estimate the water environment effects in Three Gorge Reservoir (TGR). The experiment to marinate the cumulate MSW under 135 meters for TGR in water is finished, and the same time, the Separating Distributing Coefficient (SDC) and the Separating Velocity Rate (SVR) of each heavy metal in water are presented, too. The experiment results show that the SDC of Cd is the smallest, the Ratio for Water and MSW (RWM) and marinating time play an important role in SDC and SVR, the SDC and SVR change smaller while the RWM get to 400, the separating content of heavy metal can reach 0.75 during three weeks, SVR decreases more with marinate time's increasing, and SVR gets an limit value with smaller RWM. The results can be a base for deeply studying the managing technology and the cumulate MSW under 135 meters for TGR.     

黄烷酮类衍生物的抗菌活性QSAR研究

为了研究黄烷酮衍生物分子结构与杀菌活性的关系,本文以34种黄烷酮衍生物的杀菌活性为指标,以量子化学计算软件PCLIENT提取化合物的初始分子描述符,经高维特征筛选以及多轮末尾淘汰进行特征选择,获得了14个重要分子描述符,并基于保留描述符构了建多元线性回归、支持向量回归模型,获得了较高的预测精度。最后基于支持向量机非线性解释体系,对保留分子描述符的作用进行解析,各描述符对影响黄烷酮衍生物的杀菌活性的重要性依次为SssO > MATS6e > SeaC2C2aa > EEig04d > EEig03x > C-028 > R7v+ > L2e > X4A > Mor25m > G1m > Mor17m >BEHv1 > RDF045m,为指导黄烷酮类杀菌剂的合成提供了重要指导。     

云烟品牌云南原料基地土壤、烟叶氯素含量特征研究

分析云烟品牌云南原料基地土壤及烟叶氯素特征,开展区划研究,为云烟品牌原料基地养分科学管理和烟叶工业可用性提高提供科学依据。以云烟品牌省内原料基地昆明、曲靖、红河、保山、文山5州（市）37县1130份土壤样品和中部烟叶（C3F等级）样品为材料,对其氯含量特征进行了分析和区划研究。结果表明：云烟品牌省内原料基地中部烟叶氯含量平均值为0.29%,氯含量＜0.3%的分布频率高达71.33%,0.3%-0.8%适宜区间的样品分布频率仅有22.75%。土壤氯平均含量为15.37mg/kg,样品分布频率最高的是＜10.0mg/kg区间,占61.68%,适宜区间10.0-30.0mg/kg样品分布频率仅为28.23%。水溶性氯含量最高的海拔段是1800-2000m,最低的是＜1600m;中壤土的水溶性氯含量最高,轻粘土和中粘土含量较低。云烟品牌云南原料基地中,仅有28.23%的土壤样本和22.75%的烟叶样本在适宜范围内,有61.68%的土壤样本和71.33%的烟叶样本氯含量低于适宜范围。     

基于支持向量回归与地统计学的多维时间序列分析

基于地统计学与支持向量回归,建立一种快速定阶、既反映样本集动态特征,又体现环境因子影响的高精度非线性多维时间序列预测方法（GS-SVR）。对带趋势时间序列平稳化后,先基于地统计学后效时间长度进行因变量快速定阶;再以支持向量机基于最小原则非线性筛选自变量,继以主成分分析消除自变量之间的信息冗余;最后以一步预测法检验GS-SVR的有效性。2个农业科学实例结果显示,GS-SVR在所有参比模型中预测精度最高,稳定性最好。GS-SVR能快速、准确实现模型定阶,是一种融合时间序列分析和回归分析的非线性多维时间序列分析方法,并具非线性、避免过拟合、避免局部最小、泛化能力优异等优点,在农业科学、生态学、经济学等多维时间序列预测领域有较广泛的应用前景。     

曲靖烟区烟叶微量元素与土壤pH值关系研究

为了探明云南曲靖植烟土壤pH值对烟叶微量元素含量的影响,在曲靖烟区采集土样和烟叶样各3656份,采用多元统计分析、相关性分析和平滑回归分析研究了曲靖烟区土壤pH值与烟叶微量元素含量的关系。结果表明,土壤平均pH值为6.28,pH值在5.5以下的土壤样本占22.62%,在7.0以上的占24.45%。曲靖烟叶铁和锌含量处于高水平,而硼、钼和氯含量处于低水平。锰含量适宜的烟叶样本比例最高,约有70%的样本处于适中水平。烟叶锰、锌和氯含量与土壤pH值呈极显著负相关关系,烟叶硼、钼和铁含量与其呈极显著正相关关系。曲靖烟区在烟叶生产中应注重结合丰缺的微量元素,科学合理增施硼和钼肥,增加烟叶微量元素,提高烟叶质量。     

基于全波段高光谱的冬小麦生长参数估算方法比较

利用高光谱数据监测作物生长情况具有无损和高效的特点,是现代农业的发展方向。为了简化高光谱数据处理流程,直接利用原始的高光谱反射率完成从建模到估算作物生长参数的全过程,应用于作物长势的实时监测。本文利用偏最小二乘回归（partial least squares regression,PLSR）、支持向量回归（support vector regression,SVR）和前馈神经网络（feedforward neural network,FNN）3种方法,利用全波段高光谱数据分别对冬小麦多个关键生育期（拔节、孕穗、扬花和乳熟期）生长参数（地上部生物量、叶面积指数、全氮含量和叶绿素浓度）进行了估算。比较3种方法的建模及估测效果,发现对于建模集数据,SVR对上述生长参数4个生育期的估测结果R²均值为0.89~0.98,MAPE为1.70%~7.53%,对于验证集数据,R²均值为0.90~0.94,MAPE为4.04%~7.46%,拟合优度和估测精度均超过PLSR和FNN,是估算方法中利用全波段光谱反射率估测冬小麦生长参数的最佳方案。随着无人机载高光谱技术成熟,SVR方法能够用于处理航拍获取的大范围田间高光谱信息,简便快捷地进行建模与参数反演,实时反映作物生长状态。     

湖南烟区中部叶化学成分比较及聚类分析

为研究湖南烟区化学成分,以湖南烟区42个产烟县73个中部叶样品为材料,对化学成分进行比较和聚类。结果表明：湖南烤烟中部叶总糖、还原糖和糖碱比偏高,氯的变异系数较大且含量偏低;各地区间化学成分差异显著,嘉禾和芷江的化学成分基本符合优质烟叶的生产要求;聚类结果将湖南产烟县分为3类,多重比较表明各类间除总氮和氯外其他化学指标差异显著。类别间差异可以为卷烟企业提供不同原料的供给。     

南平烟区植烟土壤氯的烤烟适宜指标研究

为探索南平烟区在烟-稻轮作下植烟土壤氯的烤烟适宜指标,通过盆栽和田间试验,研究土壤氯水平对烟株氯含量与其分配、干物质积累及土壤氯变化的影响。结果表明,当土壤氯含量＞10 mg/kg时,烟株氯与土壤氯呈极显著线性正相关;当土壤氯含量＜10 mg/kg时,土壤氯对烟株氯的影响较小。随土壤氯水平提高,烟株干物质积累量呈抛物线型,烟株吸收的氯显著增加,且更多地分配到叶片中。适量施氯,盆栽烟稻轮作后或大田烤烟收获后,耕作层土壤氯含量已降至基本一致的水平,残氯量与施氯量多少无关;犁底层和心土层氯含量却随施氯量增加而极显著增大。南平烟区植烟土壤氯适宜含量指标为25~35 mg/kg。在土壤氯含量平均为(20.22±17.22) mg/kg的现状下,可以施用氯化钾22.5~69.3 kg/hm2,使烤烟种植当季土壤氯含量提高到适宜水平,满足优质烟叶生产需要。     

基于DEM的山西省农业气候资源集的建立

为了全面了解山西省的农业气候资源现状,利用1961—2010年山西省气象观测资料和站点的地理信息资料,通过采用小网格推算加残差订正的方法,基于DEM建立了山西省农业气候资源集,并以4个气候资源要素为例进行了分析。结果表明：年平均温度、≥0℃的积温、年日照时数和年降水量的多元线性回归复相关系数分别达到了0.9811、0.9855、0.8725和0.8917,再经过残差订正得到的农业气候资源集结果详尽且符合实际,能够详细描述山西省不同地形条件下气候资源的空间分布特征。此结果可以为农业的合理布局和规划提供科学的方法和决策的依据。     

Mixed models including environmental covariables for studying QTL by environment interaction

The study of the phenotypic responses of a set of genotypes in their dependence on the environment has always been an important area of research in plant breeding. Non-parallelism of those responses is called genotype by environment interaction (GEI). GEI especially affects plant breeding strategies, when the phenotypic superiority of genotypes changes in relation to the environment. The study of the genetic basis of GEI involves the modelling of quantitative trait locus (QTL) expression in its dependence on environmental factors. We present a modelling framework for studying the interaction between QTL and environment, using regression models in a mixed model context. We integrate regression models for QTL main effect expression with factorial regression models for genotype by environment interaction, and, in addition, take care to model adequately the residual genetic variation. Factorial regression models describe GEI as differential genotypic sensitivity to one or more environmental covariables. We show how factorial regression models can be generalized to make also QTL expression dependent on environmental covariables. As an illustrative example, we reanalyzed yield data from the North American Barley Genome Project. QTL by environment interaction for yield, as identified at the 2H chromosome could be described as QTL expression in relation to the magnitude of the temperature range during heading. This revised version was published online in July 2006 with corrections to the Cover Date.     

加氯再生水交替灌溉对土壤氮素残留和马铃薯大肠菌群影响

针对再生水灌溉潜在环境污染及食品安全问题,通过田间小区试验,研究了不同灌溉方法、灌水水质和灌水水平对土壤大肠菌群含量、马铃薯块茎表皮和组织内部大肠菌群含量影响。结果表明：收获后,加氯再生水交替灌溉处理表层土壤矿质氮显著高于其他处理,加氯再生水交替灌溉保持马铃薯产量的同时,增加了根层土壤氮素的可利用性及后效性;加氯再生水交替灌溉根层土壤大肠菌群数量显著低于其他灌水处理,尤其是加氯再生水交替灌溉马铃薯组织内部大肠菌群数量仅为46 MPN/100 g,低于马铃薯淀粉标准一级品规定。因此,加氯再生水交替灌溉兼具保障马铃薯卫生安全与节水高效双重功能的灌溉策略。     

不同厂家六号溶剂油对植物油中溶剂残留量测定结果影响的研究

扦取５份大豆油样品,分别使用两个厂家的六号溶剂油制备标准样品,利用峰高定量法和峰面积定量法测定溶剂残留量,测定结果不同。大多数情况下,植物油样品种六号溶剂油组成与标准样品的不同,采用较高的记录仪纸速和峰面积定量,才能使溶剂残留测定结果较为准确。     

烤烟氯含量与有机酸类物质及感官评吸质量的关系分析

为研究烤烟叶片氯含量与有机酸类物质及感官质量的关系,在国内主产烟区采集294 份烟叶样品,采用描述统计法分析烤烟氯和有机酸含量的数量特征,采用聚类分析法研究了烤烟不同氯含量水平下有机酸含量及感官质量的变化。结果表明：烤烟样本氯含量变幅0.08%~1.60%,平均值为0.44%±0.28%;不同烟区之间烤烟叶片氯含量差异显著,不同部位之间烤烟氯含量差异不显著。基于烤烟氯含量的聚类分析将供试样本分为低、中、高3 类：在低类群中,非挥发性有机酸、有机酸总量较低,高级脂肪酸、挥发性有机酸含量中等;在中类群中,有机酸类物质含量均最高;在高类群中,非挥发性有机酸、有机酸总量中等,高级脂肪酸、挥发性有机酸含量较低。烤烟感官质量评吸总分与氯含量呈显著负相关,随氯含量的提高而逐渐降低。     

The residual background genome from a donor within an improved line selected by marker-assisted selection: impact on phenotype and combining ability

A successful marker-assisted selection (MAS)-derived line should express new characteristics derived from the donor, and have a small number of donor chromosomal fragments. Different selection strategies can result in a different proportion of donor residual background within the improved line. The effects of the residual background on phenotype and combining ability mostly remain unknown. In this study, waxy ( wx ) and a restorer gene ( Rf3 ) of S-type cytoplasm sterility were selected as a model for MAS breeding by three strategies, to develop MAS-derived lines with a different level of residual background. Three MAS-derived lines were successfully developed by introgression of the target gene. Furthermore, the genetic background, agriculturally important traits and combining ability of the three MAS-derived lines were assayed. The results suggested that the three MAS-derived lines retained 4–18% of the chromosomal fragment of the donor, respectively, and the residual background impacted on the phenotype of important traits and combining ability of MAS-derived lines. We also showed that whole genome background screening each backcross generation can accelerate recovery of the recurrent parent genome, phenotype and combining ability.