Study On The Ordinary-/high-temperature And High-pressure Density Of Binary Organic Mixture Via SVR

Density is one of important characteristics of material, each material has its certain density, the densities of different materials are different in general. It also plays a very important role in the process of industrial production, chemistry and craft. With the knowledge of density, we can deduce many other related thermal physical properties for matter. We know that it is not difficult to measure the density of some simple solid and liquid matter under normal temperature and pressure, what we need to do is measure its quality and volume, but it is complex and quite difficult when we want to measure the density of some materials under high-temperature and high-pressure, even extreme conditions. And it is not easy to reach the requirements when the condition is relatively hard, so what we need to explore is to find an effective way to guide the experimental process, and to optimize the experimental design, and to analyze the experimental data and result.In this thesis, based on the theory of supporting vector regression(SVR), we established SVR models and studied the densities of different kinds of binary mixtures with different material ratio and underordinary/high-temperature and high-pressure, the main work are as follows:① According to related experimental data, the PSO-SVR was employed to model and predict the density of the binary mixture with diethyl carbonate and p-xylene at different condition(material ratio, temperature from 288.15 K to 308.15 K,and pressure from 0.1MPa-40MPa). The result indicated that mean absolute percentage error(MAPE)achieved by established SVR model for the 162 training samples is less than 0.03%, the MAPE for 18 test samples is not greater than 0.04%. The correlation coefficient between the measured and calculated indexes approaches 1. These reflected that the accuracy of constructed SVR model is higher in the case of mid/small number of samples and it has strong generalization ability of in middle and small sample. The densities of the binary mixture with diethyl carbonate and p-xylene at different condition(material ratio, temperature from 288.15 K to 308.15 K, and pressure from0.1MPa-40MPa) were calculated based on the established SVR model.② The PSO-SVR was also used to model and predict the density of the binary mixture of diethyl carbonate and p-xylene at different conditions, i.e., material ratio,temperature from 288.15 K to 308.15 K,and pressure from 0.1MPa-40 MPa. The resultillustrated that mean absolute percentage error(MAPE) achieved by established SVR model for the 198 training samples is less than 0.024%, the MAPE for 18 test samples is not greater than 0.04%. The correlation coefficient between the measured and calculated indexes approaches 1. Thus the both MAPEs either for the training samples or the test samples are smaller than that(0.051%) achieved by the modified Tait equation reported in the literature.the changing trend of the densities under different conditions was depicted in 3D drawn by Origin.③ According to the literature reported experimental densities on binary mixture of octane and p-xylene at different conditions, i.e., material ratio, temperature from288.15 K to 308.15 K, and pressure from 0.1MPa-40 MPa,the PSO-SVR was utilized to establish a model andthe calculated data by SVR model was compared with the measured data by experiment. The result revealed that the MAPE achieved by established SVR model for the 180 training samples is less than 0.024%, the MAPE for18 test samples is not greater than 0.112%. Thus the MAPEs for the training samples is smaller than that(0.03%) of modified Tait equation, but the MAPE of the test samples are slight bigger than that of the modified Tait equation.④ Based on the experimental data of decane and p-xylene, we used PSO-SVR to establish model and analysis the density under different pressure, temperature and material ratio, and compared with the predicted result by modified Tait equation. The result showed that the predictive and training ability of SVR is superior to the reported modified Tait equation. The result was that the MAPEs(0.03% and 0.0005%) achieved by established SVR model for the 198 training samples and 18 test samples are less than0.045% achieved by modified Tait equation. The changing trend of the densities under different conditions was depicted in 3D drawn by Origin.⑤ We used SVR to establisha model and analysis the density of binary mixture of propane and decane with the temperature ranging from 344.2K to 513.2K and the pressure ranging from 3.5MPa to 262.5MPa. And compared the calculated results by SVR model and by a modified Tait equation. It is found that prediction performance of SVR model is better than that of modified Tait equation. The result demonstrate that the MAPE achieved by established SVR model for the 210 training samples is less than0.0016%, the MAPE for 23 test samples is not greater than 0.004%. The correlation coefficient between the measured and calculated indexes is appropriately equal to 1.Thus the both MAPEs either for the training samples or the test samples are smaller than that(0.26%) calculated by the modified Tait equation reported in a literature.In this thesis, we used the experimental data of 5 kinds of binary organic mixtures to establish SVR model by supporting vector regression combined particle swarm optimization. The result showed that SVR has strong modeling and predictive ability,and it can provide theoretic guidance for the experimental process to reach purpose for saving human, financial and material resources.