To enable the power grid to smoothly pass through the summer load peak, it is necessary to predict the peak load of distribution network lines 1-2 months in advance before the summer load peak. This provides data support for the equipment department to develop and implement capacity expansion and reconstruction projects in a planned manner. This paper proposes a distribution network line peak load forecasting method based on XGBoost. The method comprehensively considers various factors including meteorological, time and spring load factors, and analyzes the correlation coefficients between summer load and various factors, and accordingly determines the eigenvalues of prediction samples. The K-means algorithm is used to cluster the line load growth trends to identify the target lines with heavy load in the future. The XGBoost algorithm is used to predict the peak load of the lines. The predice results on an actual urban distribution network verify the prediction accuracy of method. Comparisons with other methods demonstrate that the proposed method has the advantages of smaller computation scale and faster prediction speed. This work is supported by the Science and Technology Project of State Grid Corporation of China “Load Forecast and Promotion of Distribution Network Power Supply Potential Based on Ubiquitous Power Internet of Things and Big Data Analysis Application” (No. 5211HZ19014P).