According to the daily precipitation of 20 meteorological stations in Haihe Plain from 1955 to 2019, the standardized precipitation index (SPI) on annual scale and seasonal scale was calculated, and the drought frequency and spatial and temporal characteristics of different levels on each scale were analyzed. Taking annual scale SPI as the sample, the normal cloud generator algorithm and the multi-step restoring reverse cloud transformation algorithm were used to construct the cloud model to analyze the randomness and stability of drought. The results showed that the drought frequency in Haihe Plain was mainly between 0.28 and 0.31, with the characteristics of high frequency of light drought and low frequency of heavy drought. The frequency of spring drought was the highest and winter drought was the lowest. The differences between regions and between years were the greatest in summer. The three characteristic parameters of the interannual SPI cloud model were all showed decreasing trends, in which the entropy significantly was decreased and the superentropy was significantly and positively correlated with the entropy, that was, the randomness and inhomogeneity of the SPI distribution showed a consistent trend. In space, the superentropy and entropy of each station showed a very significant negative correlation, the randomness and inhomogeneity showed an opposite trend. The inter-annual differences of cloud characteristics were greater than the inter-site differences, that was, the cloud model can better reflect the randomness and stability of regional inter-annual SPI in space. Haihe Plain tended to be drier in general, and the randomness of SPI of each station was decreased significantly and tended to be stable and uniform.