Information content of data and variables and types of weighting in least-squares regression methods

Algorithms are given for evaluating the relative amount of useful information related to a particular parameter which is carried by individual data points and intervals of the variables. The algorithms provide an efficient means of using the information contained in a set of data. Applications to the optimization of weighting in regression methods are described. Several informational and combined informational-statistical types of weighting are studied as a means of improving the accuracy and precision of the parameters obtained by non-linear regression.     

偏最小二乘火焰原子吸收光谱法同时测定汞和钴

本文研究了偏最小二乘原子吸收光谱法同时测定Hg和Co的方法,校正中Co253.649nm对Hg 253.652nm的吸收红重叠干扰,用Hg空心阴极灯同时一合成样中的Hg和Co,方法快速、简单、准确,结果令人满意。     

PP回归方法在预测模型中的应用－用PP回归建立“积累”“消费”模型

本文利用ＰＰ回归方法,建立“积累”“消费”回归模型,并与最小二乘法拟合的线性回归模型相比较,说明用ＰＰ回归建立的模型预测精度较高．     

基于SVM回归模型的混合气体组分种类光谱识别方法

针对混合气体红外光谱分析中无法采用同一模型同时进行混合气体组分浓度的定量分析和组分种类的定性分析的问题,本文提出了基于SVM回归模型的混合气体组分种类光谱识别方法.通过详细推导,证明混合气体组分种类识别完全可以通过组分浓度分析的SVM回归模型来求解,混合气体组分种类识别是一种特殊的回归.实验结果显示,该方法的混合气体组分种类的正确识别率不小于92.5%.     

基于炉口辐射光谱支持向量机回归的转炉终点碳含量检测

转炉终点碳含量在线检测对实现炼钢终点准确控制,提高钢铁产品质量,降低能耗,减少废气排放具有重要意义。针对转炉冶炼终点控制及碳含量检测的难题,研究了一种新的基于炉口辐射光谱分析的非接触式在线碳含量检测方法。方法基于辐射光谱的支持向量机回归(SVR)实现转炉终点前过程的碳含量预测。通过远距离光谱采集系统获取炉口火焰光谱信息,基于冶炼过程炉口火焰辐射光谱变化规律的分析,分别提取了表征辐射光谱整体特征的两个参数即总谱宽和辐射峰值、以及表征发射谱的三个特征波长600,630和775 nm处的幅值作为支持向量机的输入,结合脱碳理论和实测碳值拟合重构的脱碳函数曲线作为支持向量机的输出,利用支持向量机回归方法建立光谱分布与碳含量的关系模型。通过训练样本集和测试集循环优化确定模型最佳参数。设计的仪器和优选的模型已安装在转炉生产现场长时间运行,现场实验结果表明,终点碳含量检测准确率为90.2%,测量时间小于0.3 s,可实时在线检测,能够满足生产需求,为转炉冶炼终点的精确控制提供了重要依据。     

太赫兹光谱技术用于干旱胁迫下大豆冠层含水量检测研究

近年来水资源短缺问题日益严重,部分地区由于农业灌溉用水不足导致庄稼减产农民利益受损。大豆是一种需水量较大的农作物,一旦水分亏缺将直接影响大豆植株的形态和生长发育,从而造成大豆品质降低和产量减少。大豆叶片的水分状况可真实地反映植株水分受土壤水分亏缺的影响程度,因此,大豆冠层叶片水分含量的快速获取成为一种需要。太赫兹辐射在水中的强烈衰减使其成为一种非常灵敏的非接触式探针,可以快速、无损地检测叶片含水量。因此基于太赫兹光谱这一新技术进行大豆冠层叶片含水量的检测研究,用于实时监测田间大豆的健康状况。实验选用中黄13号大豆进行栽培,为尽可能模拟田间不同程度的干旱胁迫状况,将开花期大豆进行5个不同梯度：正常供水、轻度干旱胁迫、中度干旱胁迫、重度干旱胁迫、严重干旱胁迫(分别占田间最大持水量的80%,65%,50%,35%,20%)的水分灌溉,每个梯度设置3个重复。利用人工称重法与便携式土壤水分速测仪结合将土壤含水量调控到各水分梯度要求。然后,将实验大豆植株运回实验室并利用透射式太赫兹时域光谱仪进行样本扫描,每个梯度采集18片冠层叶片,共90个样本,以2∶1的比例分为校正集和预测集。在获取各样本时域光谱数据后,根据Dorney和Duvillaret提出的模型进行了光学参数的提取,得到各样本的吸收系数谱以及折射率谱。定性分析了太赫兹时域光谱、吸收系数、折射率随水分胁迫程度不同的变化情况。实验发现：随着水分胁迫程度的降低,时域光谱的峰值呈不断衰减趋势,且均低于空白参考峰值,同时有明显的时间延迟。吸收系数值随干旱胁迫程度的加剧逐渐降低;折射率值同样随干旱胁迫程度的加剧逐渐降低。并利用偏最小二乘(PLS)和多元线性回归(MLR)方法定量研究了时域光谱、吸收系数、折射率光谱数据与叶片含水率的相关关系。结果表明,太赫兹波对大豆叶片水分差异十分敏感,基于时域光谱最大值和最小值的MLR预测精度最高,预测集相关性(r_p)达-0.939 3,均方根误差(RMSEP)为0.049 5。研究表明太赫兹光谱技术应用于大豆冠层叶片含水量观测具有良好的可行性,为开展大豆冠层含水量信息快速获取,实现科学节水管理与灌溉决策提供了新的检测手段和实验依据。     

The Use of Encoded Structural Data to Predict the Liquid Chromatographic Retention Behavior of Tetradentate Metal Complexes

Abstract

Retention data for tetradentate metal complexes of copper and nickel are reported for microparticulate silica and reversed phase columns. An attempt is made to relate structure and stability constant data to the retention volume by means of multiple linear regression. The results indicate that if the structure is encoded by means of dummy variables it may be possible to predict retention volumes of metal complexes. There is not a strong correlation between the value of the stability constants for the metal diamine and metal ketone complexes and the retention volume.     

p-Hub approach for the optimal park-and-ride facility location problem

Park and Ride facilities (P&R) are car parks at which users can transfer to public transportation to reach their final destination. We propose a mixed linear programming formulation to determine the location of a fixed number of P&R facilities so that their usage is maximized. The facilities are modeled as hubs. Commuters can use one of the P&R facilities or choose to travel by car to their destinations, and their behavior follows a logit model. We apply a p-hub approach considering that users incur in a known generalized cost of using each P&R facility as input for the logit model. For small instances of the problem, we propose a novel linearization of the logit model, which allows transforming the binary nonlinear programming problem into a mixed linear programming formulation. A modification of the Heuristic Concentration Integer (HCI) procedure is applied to solve larger instances of the problem. Numerical experiments are performed, including a case in Queens, NY. Further research is proposed.     

Learning Interactions via Hierarchical Group-Lasso Regularization

We introduce a method for learning pairwise interactions in a linear regression or logistic regression model in a manner that satisfies strong hierarchy: whenever an interaction is estimated to be nonzero, both its associated main effects are also included in the model. We motivate our approach by modeling pairwise interactions for categorical variables with arbitrary numbers of levels, and then show how we can accommodate continuous variables as well. Our approach allows us to dispense with explicitly applying constraints on the main effects and interactions for identifiability, which results in interpretable interaction models. We compare our method with existing approaches on both simulated and real data, including a genome-wide association study, all using our R package glinternet.