New type of power system has become the emerging topics to meet the China’s carbon emission target at 2030 and 2060. Among many key technology innovations within the new type of power system domain, one critical challenge to face is the coordination among the generation, network infrastructures, load and energy storage systems (ESS). To manage and operate these grid resources in an economic, secure and reliable manner, the corresponding static and dynamic model parameters need to be estimated accurately and effectively. Deep reinforcement learning (DRL) is one of the most trending artificial intelligence techniques in recent years and has led to many advanced applications in power systems, especially in control related topics. Since DRL itself is also a powerful tool in optimization, it has shown promising results in parameter estimation problem. Therefore, in this paper, we systematically review the DRL methods adopted for the power system model parameter estimation, including load models and generator models. To further promote the applications of DRL in the related areas for academic and industry researchers, future research directions are also highlighted.