The topology structure of the new power system with multiple distributed power sources in the distribution network has diversity and variability, which affects the real-time and accuracy of topology identification. To address this issue, a distribution network partition topology identification method based on an improved deep forest algorithm is proposed. Construct a distribution network topology identification framework that combines topology, using switch state matrices to describe the topology structure for physical identification and dimensionality reduction; Propose a feature selection and topology identification method based on an improved deep forest algorithm, which obtains region topology identification of historical and unknown topologies through parallel offline training in zones. The improved deep forest model obtains real-time region switch state matrix labels through online application, forming a distribution network switch state matrix to achieve system topology identification. The test results of the distribution network example system have verified the effectiveness of the proposed method.