To address the problem of missing runoff data, a daily runoff prediction model (MM-LSTM-BP model) combining a long short-term memory (LSTM) neural network and a back propagation (BP) neural network based on multiple regression models was constructed. In this model, regression models were adopted to extract the linear, nonlinear, temporal, and random characteristics of runoff, and the LSTM neural network and the BP neural network were used in series for the regression simulation of the daily runoff process. The case verification results of the Weihe River Basin indicate that the MM-LSTM-BP model generally performs better than the single regression methods during the verification period; the root mean square error (RMSE) of the daily runoff data decreases by more than 50%, and the Nash efficiency coefficient (NSE) increases to 0.935; the MM-LSTM-BP model has better stability during the normal flow period and the recession period, and the simulation error of the flood peak is reduced by more than 6% during the flood period.