为更准确预测CO2在临界点附近区域的热物性,分别建立了基于BPNN，SVR和GPR算法的智能模型来预测近临界区CO2的密度、粘度和导热系数，并将3种模型进行比较。结果表明：基于BPNN的密度（R2=0.946 5）和粘度（R2=0.970 2）预测模型相较于其他智能模型精度更高，而基于SVR的导热系数的模型预测精度更高（R2=0.999 7）；所提出的智能模型相较于传统模型中SW密度方程（R2=0.596 6）、Laesecke的粘度方程（R2=0.844 5）和J&H的导热系数方程（R2=0.021 8）的R2提高了14.88%~4 444.5%

In order to more accurately predict the thermophysical properties of CO2 in the region around the critical point,the intelligent models of BPNN,SVR and GPR were developed to predict and compared the density,viscosity and thermal conductivity of CO2 in the nearcritical region,respectively.The results show that the density (R2=0.946 5) and viscosity (R2=0.970 2) prediction models based on BPNN are more accurate than other intelligent models,and the thermal conductivity model based on SVR is more accurate (R2=0.999 7).Compared with the SW density equation (R2=0.596 6),Laesecke′s viscosity equation (R2=0.844 5) and J&H′s thermal conductivity equation (R2=0.021 8) in the traditional model,the R2 of the proposed intelligent model is improved by 14.88% to 4 444.5%.

TK211