PLS and PCR Methods in the Assessment of Coastal Water Quality

Partial least squares regression analysis (PLS) and principal component regression analysis (PCR) were examined asmethodological procedures for assessing the quality of coastalwaters in a tourist area. Four variables related to the trophicstate of waters, namely nitrate, nitrite, ammonia and phosphate were analyzed. The models resulting from PLS and PCR were verysimilar. Both defined three groups of water masses characterizedby different nutrient loadings. These groups were in accordancewith those obtained by numerical classification. The PLS methodwas selected as the optimal model, on the basis of its lowerprediction errors (lower Press and Rmsd values). For managementpurposes, this statistical model allows mesotrophic conditions,reflecting some nutrient enrichment over background conditions,to be characterized and the successful diagnosis of additionalsamples within this context.     

The use of PLS algorithms and quantum chemical parameters derived from PM3 hamiltonian in QSPR studies on direct photolysis quantum yields of substituted aromatic halides

The Partial Least-Squares (PLS) method was used to further study Quantitative Structure-Property Relationship (QSPR) for photohydrolysis quantum yields of selected aromatic halides. The cluster of substituted bromobenzenes and iodobenzenes was further grouped into two clusters, substituted bromobenzenes and substituted iodobenzenes, for which QSPRs were obtained, respectively. The QSPRs obtained by PLS are more significant than those obtained in previous studies. The studies showed that only when the compounds belonging to different groups are studies separately, inherent relationships between molecular properties and their structure characteristics can be revealed. It can be generally concluded that the photohydrolysis quantum yields of the substituted aromatic halides are dependent on the overall characters of the molecules, the characters of the carbon–halogen bonds to be broken in the photolysis, and the character of the halogen atoms to be replaced in the photolysis. Electronegativity has great relevance to the photohydrolysis processes of the compounds. Substituted aromatic halides with large average molecular polarizability and molecular weight values tend to have low photolysis quantum yields.     

Linear free energy relationships for dechlorination of aromatic chlorides by Pd/Fe

Reductive dechlorination rate constants for five chlorobenzenes in the presence of Pd/Fe as catalyst were determined experimentally. Linear free energy relationships (LFER) for the dechlorination rate constants of five chlorobenzenes and three chlorophenols were developed by partial least squares (PLS) regression based on quantum chemical parameters computed by PM3 Hamiltonian. The optimal LFER model obtained is

logk=−1.63+1.46×10⁻³ΔH_f−7.69×10⁻¹E_LUMO

where k stands for the dechlorination rate constants, ΔH_f is the standard heat of formation, and E_LUMO is the energy of the lowest unoccupied molecular orbital. The Q²_cum value of the model is 0.879, indicating good robustness and predictive power of the model.     

15种取代酚对淡水发光菌Q67的毒性及定量构效分析

为了更加准确和便捷地预测各种取代酚类化合物的急性毒性,以淡水发光菌Q67(Vibrio qinghaiensis sp.-Q67)为受试生物,测定了15种典型取代酚的急性毒性;采用logD(正辛醇/水分配系数),LUMO(分子最低空轨道能)和MW(分子量)等取代酚的7种主要结构参数,利用偏最小二乘回归法建立了定量结构-活性相关(quantitative structure-activity relationships,QSAR)模型。结果表明,15种取代酚的EC_(50)在5.76×10~(-6)～1.27×10~(-3)mol·L~(-1)之间,且有很好的剂量-效应关系;QSAR模型的主成分分析显示,-logEC_(50)与logD、LUMO和MW值正相关,且logD对模型的贡献最大,即越容易与Q67菌结合的酚类化合物对其的急性毒性越大;建立的QSAR模型具有较好的预测能力(Q~2_(EXT)=0.91,RMSE=0.49)和较高的稳定性(Q~2_(CUM)=0.58),能够用于预测其他酚类化合物对Q67菌的急性毒性。     

苯酚类化合物光解量子产率与结构性质的定量关系

采用量子化学PM3算法计算得到的苯酚类化合物的量子化学参数,应用偏最小二乘(PLS)算法,建立了能预测苯酚类化合物光解量子产率(Y)的定量结构-性质关系(QSPR)模型.影响苯酚类化合物光解量子产率的主要因素是分子最高占据轨道能(Ehomo)和分子生成热(HOF),logY值随着Ehomo的增大而增大,随着HOF的增大而减小.分子的电负性和(Ehomo-Ehomo)2对logY值也有一定的影响,电负性和(Ehomo-Ehomo)²较小的分子,其logY值也较大.     

基于遗传算法的PLS分析在QSAR研究中的应用

将遗传算法和偏最小二乘法结合应用于定量结构活性关系研究中 ,进行变量选择和建立最终模型 .以较少的变量个数包含较多的变量信息 ,且变量间没有线性相关问题 ,因而得到较好的 QSAR模型 .将这种方法应用于氯代酚和单取代苯 2种系列化合物 ,可以得到几种常规多元回归分析方法不能得到的质量较高的 QSAR模型 .     

珠江口海域总无机氮的遥感提取研究

利用珠江口海域2个航次36个站位的实测遥感反射率和总无机氮(TIN)数据,基于偏最小二乘回归思想,建立了珠江口海域无机氮浓度估算的遥感模型.建模中考虑了6种比值法.结果发现,不同波段与第6波段比值所建立的模型精度最高,故选用该比值法所建立的模型作为珠江口海域无机氮遥感提取模型.利用该模型得到的珠江口海域18站位无机氮浓度和实测值相比,其总体平均相对误差为9.06%,相对误差最大值为18.2%.将该模型用于1998年12月31日的SeaWiFS资料,得到了较好的估算结果.与1998年12月31日的实测资料进行对比分析,实测值与估算值的平均相对误差为31.9%,说明该模型的稳定性较好,能够适用于不同时相的SeaWiFS图像,可为珠江口海域无机氮含量的分布以及富营养化问题研究提供了实时监测的依据.     

Emission Factors Determination of Euro III 1,200- to 1,400-cc Petrol Passenger Cars with a PLS Multivariate Regression Model

This paper presents emission factors of a class of passenger cars obtained by applying a statistical model developed to evaluate average emission factors based on driving cycle emission measurements. A multivariate regression method based on principal components, namely, the partial least squares (PLS) method, is applied to calculate the model. The method was applied to emission data from a sample of petrol Euro III 1,200- to 1,400-cc passenger cars taken from the ARTEMIS database. A vehicle effect analysis showed that vehicle effect is considerable, in some cases comparable to or greater than the driving cycle effect. Determination of emission factors is obviously affected by these aspects. Thus, the CO2 PLS model fit results are good, CO, HC and NOX more or less sufficient. PLS-predicted quantities were compared with corresponding quantities estimated by a multiple regression model (GLM) based on a quadratic polynomial equation of sub-cycle overall mean speed. GLM goodness of fit was poorer than PLS ones. A validation effort of models is in progress, which is considering the ARTEMIS database extended with tests performed within other national or international projects. In this way, an extended population of combinations of vehicles and driving cycles will provide a better calculation of models and emission factors.     

酚类化合物对硝化颗粒污泥活性抑制的定量结构-活性相关研究

测定了酚类化合物对硝化颗粒污泥活性抑制的logIC50值,以量子化学参数为自变量,应用偏最小二乘法（PLS）,建立了酚类化合物对硝化颗粒污泥活性抑制的定量结构-活性相关（QSAR）模型。模型所提取的PLS主成分所能解释的因变量总方差的比例Qc2um为0.820,表明模型具有较好的稳定性和预测能力。模型的结果表明,影响酚类化合物对硝化颗粒污泥活性抑制的主要因素是logkow、CCR和Ehomo,酚类化合物对硝化颗粒污泥活性抑制的logIC50随着分子logkow的增大而减小,随着Ehomo和CCR的增大而增大。     

城市生活污水排放量的影响因子分析及关联性研究

生活污水排放系统复杂、影响因素多种多样,从其内涵出发,分析主要影响因子,共选择3大类14个影响因子,它们之间存在着严重相关性问题,为了解决多重相关性问题,引入偏最小二乘回归方法,该方法可以有效克服多重相关性,并能够实现多种数据分析方法的综合应用;而人工神经网络具有学习和记忆能力,将二者相关联,可以较好地解决非线性问题。为检验影响因子选择的合理性和方法的适用性,以郑州市为例,对生活污水排放量和影响因子进行定量分析。结果表明,主要影响因子的选择合理,拟合和预测精度均较好。