Under the combined influence of multiple factors such as inflow from the Yangtze River and the Five Rivers Basin, extreme flood-drought events in Poyang Lake have occurred frequently in recent years, with rapid shifts between floods and droughts observed from 2020 to 2024, leading to increased uncertainty in hydrological regimes. This study integrated LSTM neural networks and the MIKE21 hydrodynamic model to construct three scenarios: measured boundary conditions, scenarios excluding Three Gorges Reservoir (TGR) regulation, and scenarios with average annual basin inflow. A scenario comparison method was systematically applied to analyze the impacts of TGR regulation and basin inflow on extreme flood-drought events in Poyang Lake since 2020. Key findings include: (1) TGR regulation positively mitigated flood peaks, reducing the maximum water level drop at Xingzi Station during extreme floods by 1.35 m. Conversely, basin inflow negatively affected low-water stability, exacerbating the maximum water level drop at Xingzi Station during extreme droughts by 3.26 m. (2) Under combined effects, TGR regulation dominated extreme flood water level variations (reduction of 0.09–0.38 m, contributing 58%–81%), while basin inflow was the primary driver of droughts, significantly lowering dry-season water levels (reduction of 0.13–1.12 m, contributing 35%–100%). (3) During extreme floods, TGR regulation impacted the entire lake area, reducing water levels by up to 1.32–1.38 m with diminishing effects from north to south. During extreme droughts, basin inflow predominantly affected the main channel and southern Fuhe River inflow zone, lowering water levels by 0–3.96 m and reducing water surface area by 516.03 km2. This research provides scientific support for water resource management, extreme disaster early warning, and ecological protection in the middle and lower Yangtze River basin.