In order to improve the precision of time interval partition at an isolated control intersection, a kagscv-DBSCAN algorithm based on KD-tree accelerated search and hyperparametric optimization using GridsearchCV was firstly proposed to solve the preliminary time of interval partition plan based on natural data of detectors, and was modified by adding the consideration of impacts of plan transition. Finally, the total vehicle delay, intersection capacity and computational speed were used as evaluation indexes for practical case analysis and comparison with traditional methods. The results show that the proposed method reduces the total vehicle delay by 8.69%, 6.61%, 3.68% and 1.04%, respectively, and improves the capacity by 5.67%, 5.07%, 3.48% and 1.69%, respectively, compared with K-means, PAW, FCM and RSAGA-FCM, and has a significant advantage in computing speed.