Repair in the distribution network is a very important part of the power system running, the lean of the distribution network emergency management can not only improve the quality of power supply service, but also reduce the economic losses of power companies. In this paper, a new fault forecasting method of repair in the distribution network is proposed. Firstly, based on historical data, the temperature, wind, the fault of the previous day, the maximum and minimum loads, etc are regarded as dependent variables. and feature mapping and preprocessing are performed on the variables. Then, the Random Forest algorithm is applied to establish the fault prediction model of the repair of distribution network, and to forecast the future failure rate in different regions, different power grids and non-grid faults and different voltage dimensions. The experimental results on the real power data show that the proposed method has better prediction efficiency and accuracy.