Load interval prediction conducting probabilistic analysis for load power quantifies the impact of uncertain factors accurately. Compared with traditional point prediction,interval prediction is beneficial to the safety and stability of the power system,and it reflects the trend of load changes more intuitively. It is proposed a rolling bus load interval prediction method based on variational mode decomposition (VMD) and long short-term memory neural network quantile regression (LSTMQR) in this paper. First of all,the bus load is decomposed into a series of subsequences with different frequency characteristics by VMD. After that,the optimal rolling steps of different subsequences are determined and LSTMQR is used to predict power intervals of different subsequences. Finally,the interval predictions of different subsequences are reconstructed to obtain the original load prediction results. It is verified by 220 kV and 10 kV bus load data to obtain that the proposed method above has a significant improvement in prediction accuracy and interval width by comparing with traditional interval prediction models.