To address the issue of simulation bias in hydrological models caused by insufficient modeling or unmodeled behaviors under complex environmental conditions, an inflow forecasting method for hydropower stations based on machine learning and Bayesian model averaging (BMA) was proposed. Based on the Hydromad model, the inflow and residual sequences were calculated; three machine learning models, namely random forest (RF) model, XGBoost model, and support vector regression (SVR) model, were used to capture the nonlinear characteristics of the residuals. Subsequently, the multi-model prediction results were fused via BMA to achieve post-processing error correction and uncertainty assessment of the simulation errors. The case study results of Liyuan Hydropower Station on Jinsha River indicate that the proposed method can significantly improve the fitting accuracy of the Hydromad model in peak flood responses and the rising and falling water stages. After correction, indicators of inflow prediction such as NSE and KGE are significantly improved; the water volume bias is effectively controlled; the overall prediction accuracy is obviously improved. This method can provide more scientific technical support for inflow forecasting and operational decision-making of hydropower stations.