Machine Learning-Based Prediction of River Flow Discharge for Improved Hydropower Planning in Indian River Basins

River flow discharge is a prerequisite in hydropower planning, scheduling, and operation because it is difficult to predict it accurately. In India, where more than 45, 000 MW of installed hydropower capacity relies on the highly seasonal and spatially variable river flows, the constraints of classical hydrological models have long been a thorn in the flesh of reliable energy generation forecasting. This paper will conduct a comparative analysis of six machine learning (ML) algorithms, namely Artificial Neural Networks (ANN), Long Short-Term Memory (LSTM) networks, Random Forest (RF), Support Vector Regression (SVR), XGBoost, and Multiple Linear Regression (MLR) to predict monthly stream flow in four large Indian river basins, namely, Ganga, Brahmaputra, Godavari, It used a 30-year period of discharge records, precipitation, temperature and evapotranspiration of the Central Water Commission (CWC), India Meteorological Department (IMD) and Water Resources Information System (WRIS) of India. Nash-Sutcliffe Efficiency (NSE), Root Mean Square Error (RMSE), Percentage Bias (PBIAS) and coefficient of determination (R2) were used to measure model performance. The findings show that the LSTM model had the best accuracy (NSE = 0.93, RMSE = 31.7 m 3 s) when compared to all other models, especially in monsoon seasons of high flows. The results underscore how deep learning methods can greatly improve the accuracy of discharge predictions, which will allow hydropower planners to optimize reservoir operations and dispatch generation. It makes recommendations on how to incorporate ML-based prediction systems in national hydropower management systems.