A regional power grid supply load fine prediction method based on two-stage reduction comprises the following steps that 1, historical data of regional power consumption loads, small hydro-thermal power, new energy power generation and regional mutual supply power and historical and prediction data of time-sharing zoning meteorology are collected; Step 2, decomposing the grid supply load into a combination of a regional power consumption load, small hydro-thermal power, new energy power generation and regional mutual supply power by adopting a two-stage reduction method; Step 3, introducing time-sharing partition meteorological information according to characteristics and influence factors of different types of loads, constructing different prediction models, and obtaining a prediction result of each type of load curve based on each type of load prediction model and a refined meteorological prediction value of a prediction day; And step 4, combining the load prediction curves of various types, and restoring to obtain a network supply load prediction curve of the prediction day.