With the development of power system, plenty of DC loads with high demand response potential can exist in virtual power plants, while the research on the identification technology of them is insufficient. Based on the analysis of the load monitoring requirements and distributed resources features, this paper put forward a gate recurrent unit based DC flexible resource identification method. Specially, several neural networks are established to identify the daily load data. This method can effectively reduce the requirement of sampling frequency and feature dimensions, so as to relieve the computational stress and reduce the identification time successfully. An example of a real intelligent building shows that the proposed method presents good identification accuracy for DC flexible resources, and can meet the load monitoring and identification requirements for scheduling strategies.