The existing post-processing methods of precipitation forecast tend to correct the 24 h precipitation averaged in the whole region, ignoring the regional differences of precipitation, and cannot meet the needs of refined forecasting. Based on the evaluation of CMA-MESO 3 km (China Meteorological Administration Mesoscale Model at 3km resolution) precipitation prediction during the flooding season of 2020 in Shanxi province, the frequency matching method was applied to correct the hourly precipitation forecast for specific periods and regions. The correction results were also analyzed in detail. The results showed that: (1) 24-hour precipitation forecast of CMA-MESO 3 km showed consistent false alarm, and the positive errors in the basin were prominent in Shanxi Province. (2) The forecast biases are inconsistent over different periods and altitudes. The precipitation frequency showed large wet biases during the daytime, especially in the area with lower altitude, but had dry bias in the region with higher altitude, for light precipitation at night. (3) After correction for different periods and regions, the mean errors decreased by 76.8% for 24-hour precipitation, the overestimates of heavy rain in the forecast were significantly reduced. (4) By reducing the over-predicted frequency and intensity, the problems of earlier peak of precipitation and false prediction of precipitation peak in the morning were also solved. The diurnal variation characteristics of precipitation was reproduced well. (5) The correction could improve the forecast of precipitation intensity, which could further effectively improve the forecast of terrain distribution characteristics of precipitation. However, the improvement of forecast of terrain distribution deviation of precipitation was limited.