The identification of power load characteristics is important for ensuring the accuracy of power flow simulation results and effects of active load control. Because of the large number of load nodes in real power systems, the characteristics of these load nodes are different. In load modeling, load-related grid analysis, and active control, it is difficult and unnecessary to obtain the voltage characteristics of all load nodes. Thus, it is important to classify the nodes according to the approximate load characteristics. To solve the low accuracy of load classification and inaccurate identification of load characteristics, which is due to the lack of load characteristics in data extraction, this paper proposes a method that combines a Deep Belief Network (DBN) and a Support Vector Machine (SVM) for power grid load node classification. The method uses a DBN for data reconstruction and an SVM as a classifier. Finally, the method is verified in an IEEE 10-machine 39-node system. The results show that the method can divide load nodes into five categories according to the load components and based on static load characteristics, and the accuracy can reach 96%. This method provides support for subsequent load modeling and active control.