中文摘要：

大量温室气体CO2的存在严重影响环境,而咪唑型离子液体具有独特的气体溶解性,在CO2捕集方面的应用较为广泛。基于定量结构-性质相关性（QSPR）原理提出了一种新的描述符——拓扑指数（Topological Index,TI）描述符,研究了咪唑类离子液体捕集CO2的性能与其拓扑指数描述符之间的内在定量关系。应用遗传算法获得与捕集量最为密切相关的一组拓扑指数描述符,将其与温度和压力一起作为输入参数,分别采用多元非线性回归算法及支持向量机算法建立了咪唑类离子液体捕集CO2性能与其拓扑指数描述符之间的非线性模型。通过多元非线性回归算法得出训练集和测试集的决定系数分别为0.771和0.754,由支持向量机算法得出训练集和测试集的决定系数分别为0.990和0.981。对预测模型进行了评价验证及稳定性分析,结果表明,两种模型均具有良好的稳定性能和预测能力。根据拓扑指数描述符所建立的预测模型为工程应用提供了一种预测咪唑类离子液体捕集CO2性能的有效方法。

英文摘要：

The given paper takes it as the target to make a predictive exploration of the dioxide captivity of the imidazole ionic liquids based on the novel developed topological indexes in hoping to reduce the negative impact of carbon dioxide on the living and industrial environment.As is known,imidazole ionic liquids enjoy marvelous prospect of potential application to capturing CO2 due to itsu nique gas solubility.For the research purpose,we have chosen19 kinds of imidazole ionic liquids and 380 groups of experimental data as the modeling samples.And,then,we have made quantitative investigations for the relation between the CO2 capturing performance and the topological index descriptors of imidazole ionic liquids based on the relation between the quantitative structure vs the property（RQSP） theory.Furthermore,careful calculation of the various topological indexical parameters allows us to work out their respective contributions to the CO2 capture performance and succeed in choosing a set of significant descriptors as the input variables by using the genetic algorithm method.What is more,we have also succeeded in modeling the likely non-linear relations between the selected topological index descriptors and the CO2 capture performance with the multiple non-linear regression（MNR）method and the supporting vector machine（SVM）.The results of our study have thus proven that the R^2 of the training set and the testing set should be equal to 0.771,0.754 by MNR,and 0.990,0.981 as the result of calculation done by SVM,respectively,which has promoted the verification of the stability and the predictive power of the model,whereas the residual errors remain randomly distributed on the both sides of the zero calibration in the plot of residuals.Thus,both of the aforementioned models have gained their solid stability and predictive power.Besides,we have also examined the MADS and MMDI methods respectively so as to identify the corresponding significance of each topological index descriptor in contributing to the CO2 captu