Abstract:Due to the complex runoff and concentration situation, flood forecasting for small and medium-sized catchments is very difficult. In order to improve the accuracy of flood forecasting, this study constructs a distributed model for flood forecasting based on full storage runoff and mixed runoff model respectively, taking the PoDi Basin in Xingtai City, Hebei Province as study case. LSTM, Transformer, Transformer and LSTM combining models (TFLS) were used to construct the correction model, and the differential evolution (DE) algorithm was used to optimize the hyperparameters. Taking the observed rainfall and distributed model simulation results as input, the residuals of observation flow are fitted, and then the simulation results are corrected. The results show that in the 17 flood simulation results, the mixed runoff model performs better than the full-storage runoff model, and the correction effect of TFLS model is better than LSTM and Transformer models. Compared with the mixed runoff model, the number of floods with peek error not exceeding 20% increased from 9 to 12, accounting for 70.6% of all floods. The number of floods with Nash-Sutcliffe coefficient not less than 0.8 has increased from 5 to 9, accounting for 52.9% of all floods. The performance of the TFLS model is better than other models when the observed flow does not exceed 500m3/s, and the performance of LSTM is slightly better than other models when the flow exceed 500m3/s.