Based on the precipitation forecast data of convection-permitting ensemble prediction system and observation data from April to September in 2017, 2018 and 2019, the effects of 24 h accumulated precipitation forecast of ensemble members in eastern Sichuan Basin are verified and compared, and the performance of different physical parameterization schemes is further discussed. Results show that the ensemble average forecast and probability matching forecast have obvious advantages over the ensemble members. The schemes is relatively good, while the critical success index (CSI) score of Grell 3D ensemble (G3) scheme is relatively low after 48 h. Thompson and Morrison microphysical parameterization schemes are better for light rain than WRF Single-Moment 6-class (WSM6) scheme, while WSM6 scheme is better for moderate rain, and there is no significant difference in the CSI scores of heavy rain and rainstorm among the three schemes. The CSI score of Mellor-Yamada-Janjic (MYJ) and Mellor-Yamada-Nakanishi-Niino (MYNN) boundary layer schemes is higher than that of Yonsei University (YSU) scheme, especially after 48 h. However, YSU scheme is mostly used in combination with G3 scheme, so its low score is mainly affected by G3. The precipitation distribution characteristics of all these schemes are similar to those of observation, but the false strong rain belt is predicted in Huaying Mountain, Wuling Mountain and Dalou Mountain, and the precipitation from the middle to the northeast of Chongqing is underestimated. The precipitation forecast in eastern Sichuan Basin is most sensitive to cumulus parameterization scheme. After adjusting one member’s G3 scheme to KF scheme, not only the forecast score is significantly improved, but also the ensemble spread is increased, and the skill of probability forecast is improved. prediction performance of Kain-Fritsch (KF) and Betts-Miller-Janjic (BMJ) cumulus parameterization