To improve basin-scale flood forecasting accuracy and support the integration of flood forecasting and reservoir operation, taking the Jialing River Basin as the study area, an intelligent correction framework for flood forecasting considering reservoir regulation impacts was developed. In the framework, the VIC distributed hydrological model was utilized to simulate the runoff generation and routing processes. The K-nearest neighbor(KNN) algorithm was employed for multi step extrapolated discharge correction based on historical error similarity. A hierarchical nested long short term memory (LSTM) model was introduced to predict reservoir outflow: LSTM1 characterized the basic inflow outflow response, while LSTM2 incorporated water level storage constraints and water balance principles. Dynamic correction of the entire process was achieved through stage by stage coupling across mainstreams, tributaries, and cascade reservoirs. The results show that the VIC model achieves an average NSE of approximately 0.70 and relative errors of 10%~15% at stations with minimal reservoir impact. The KNN correction yields NSE values mostly above 0.90 and relative errors below 10% for short lead times (≤12h), with sustained improvements for longer lead times. Compared with inflow outflow balance and parametric operation methods, the LSTM model better characterizes complex nonlinear regulation along the Tingzikou ReservoirCaojie Station Beibei sequence, significantly enhancing flood forecasting accuracy. This intelligent correction method significantly reduces simulation errors for both reservoir inflows and downstream control station discharge processes.