A reasonable mortality model is the key to accurately measuring longevity risks. This paper considers the dependence of mortality among different age groups and the autocorrelation and heteroscedastic structure of mortality in each age group. The multivariate Copula and AR(n)-LSV models are used to construct the mortality model. VaR, TVaR, GlueVaR are used to measure longevity risk. The results show that Copula-AR($n$)-LSV characterizes mortality trends and fluctuations better than Lee-Cater model; When mortality in China gradually decline, insurance companies will face increasing longevity risk in the future.