In order to improve the accuracy of flood forecasting in the upper Huaihe River Basin, the cointegration theory and error correction model (ECM) in econometrics were introduced for real-time correction of flood forecasting. Meanwhile, for solving the problem that the autoregressive (AR) model is not able to simulate the non-stationary sequence and the problem of sequential correlation, the autoregressive error correction model (ARECM) was constructed based on the concept of error correction. With the Huaihe River Basin above the Lutaizi station used as the study area, 1- to 3-order AR model, ECM and ARECM were constructed respectively to correct the simulation results of the distributed vertical mixed runoff generation model. The correction results of different correction models regarding 10 floods' forecasting from 2003 to 2014 in the upper Huaihe River Basin were analyzed and compared using five evaluation indexes, including the correction effect evaluation coefficient, deterministic coefficient, relative error of flood peak, relative error of runoff depth, and error of time to flood peak. The results show that the forecasted flood in the upper Huaihe River Basin is effectively corrected by the three real-time correction models. The correction performance of the AR model is relatively worse, and after two floods that have non-stationary error sequences are excluded, the average correction effect evaluation coefficient is 0.20; ECM can effectively correct forecasted floods, and the average correction effect evaluation coefficient is 0.76; the correction performance of ARECM is better as compared to the traditional AR model, and the correction effect of flood peak is significantly improved, with the average correction effect evaluation coefficient being 0.98, demonstrating that ARECM can be well applied to real-time correction of flood forecasting in the upper Huaihe River Basin.