Abstract: Check dam break flood poses a major threat to the lives and property of the downstream people.Fast and accurate prediction of the peak flow of the dam break provides important technical support for scientific assessment of the disaster consequences and emergency response of the dam break.Based on the survey data collection of earth-rock dam break cases,domestic and foreign scholars have adopted the method of regression analysis to propose a series of parameter models for predicting the peak flow of earth-rock dam break.However,the physical mechanism of compacted loess check dams is different from that of earth-rock dams,and the peak flow prediction model cannot be directly applied,at present,there is no corresponding prediction model for the peak flow of check dam break.Therefore,based on the database of 137 domestic check dam break case information,by analyzing the relationship between different types,checkation area,dam storage capacity and the peak flow of the dam break,and comparing the physical models of check dam break with different compaction degrees,seven influencing parameters of dam height (H_d),dam width (W_d),dam volume (V_d),storage capacity (V_l),breach depth (H_w),average width of breach (W_b) and average particle size of silt dam (d₅₀) are extracted,and the prediction model of the peak flow of the check dam break considering the erosion characteristics and geometric structure was constructed.Sixty check dam break cases with complete data are selected to validate the model and compared with 13 sets of typical parameter models available at home and abroad.The calculation results show that the root mean square error RMSE and correlation coefficient R² of the proposed radial-based neural network prediction model are 421.877m³/s and 0.949,respectively,Compared with other models,it has certain advantages,and its reliability is verified by an example,which has certain guiding significance for the risk prevention and control of the check dam project.