It is of great significance to establish a proper mortality model for population forecasting and human capital. The estimation method of age-period-cohort（APC） model of random mortality was improved, and Ridge regression estimation was introduced on the basis of generalized linear model. Based on the mortality rate of Japanese population aged 0～95 from 1980 to 2020, the generalized linear Poisson regression model and the improved Ridge regression model were used to fit and predict the mortality rate respectively. Model evaluation criteria such as root-mean-square error were used for comparison. The empirical analysis shows that the improved Ridge regression model is superior to the general liner model（GLM） using maximum likelihood estimation in both fitting effect and prediction results.