基于PSO-SVM的咪唑类离子液体捕集CO_2性能预测

Predictable study on CO_2 capture performance of the imidazolium ionic liquids based on PSO-SVM

CO_2是主要的温室气体,大量CO_2的存在严重影响着环境,而咪唑类离子液体具有独特的气体选择溶解性,在CO_2的捕集分离中有非常好的应用前景。基于支持向量机方法,结合粒子群优化算法(PSO-SVM)建立了咪唑类离子液体捕集CO_2性能的理论预测模型,该模型包含温度、压力、密度、黏度和表面张力5个主要参数。根据PSO算法,得到模型的最优参数为惩罚参数C=100,不敏感损失参数ε=11.699 7,核函数的宽度γ=0.279 2;SVM算法得出训练集的相关系数r=0.993,均方根误差RMSE=12.012,平均绝对误差AAE=4.117,测试集r=0.999,RMSE=4.766,AAE=3.028。对预测模型进行了评价验证以及稳定性分析,明确了咪唑类离子液体捕集CO_2性能的主要影响因素及其重要程度。

This paper is inclined to devote itself to a predictable study on CO2 capture performance of the imidazolium ionic liquids based on PSO - SVM （ particle swarm optimization and support vector machine）. In this paper, there exist ten kinds of imidazolium ionic liquids with ninety-five groups of experimental data being used as the modeling samples. As the main greenhouse gas, the large number of CO2 seriously affects the environment safety. For its unique gas solubility, it would be necessary to apply the imidazolium ionic liquids to the prospective potential of the capture of CO2. Based on the method of PSO - SVM, we have worked out the optimal parameters （C, 6, T） and established a model for predicting CO2 capture performance of imidazolium ionic liquids, which has been made up of five main parameters, that is, the temperature, the pressure, the density, the viscosity and the surface tension. In accordance with PSO method, we have worked out the optimal parameters of the model as C = 100, e = 11. 699 7 and γ= 0. 279 2, respectively. At the same time, we have also worked out r, RMSE and AAE of the training set were 0.993, 12.012 and 4. 117 respectively in accordance with the SVM method, and those of the testing set were 0. 999, 4. 766 and 3. 028, respectively. Furthermore, we have also made a detailed discussion over the predictability of the model, in which the internal interaction coefficient Q2Loo is 0. 983 while that of external one Q2ext is 0. 998. Besides, we have also found the residual error scattering randomly on the both sides of the zero calibration in the plot of residuals. The above results of our investigation and analysis show that the model we have proposed enjoys a very high stability and predictability. The main influential factors of CO2 capture performance of the imidazolium ionic liquids can be determined, and the important order of which is the pressure, temperature, density, viscosity and surface tension on the basis of MMDI （method for measure of descriptor importance） method. And, therefore, it can be concluded that the model we have proposed can serve as a new effective method for prediction of CO2 capture performance of imidazolium ionic liquids in the engineering application.