In recent years, Role Based Access Control(RBAC) has become the de facto access control model and organizations usually adopt role mining to define a set of roles from existing user-permission assignments. However, there is few approaches to generate roles for Temporal Role Based Access Control (TRBAC), the extension of RBAC with temporal constraints. In this paper, we propose a temporal approximation-based role mining approach for TRBAC which can be applied to large-scale dataset and generate the minimal number of roles with all user-permission assignments preserved by allowing mismatches in temporal components. The experiment results demonstrates the viability of our approach.