Abstract: Based on the meteorological observation data in non-flood discharge period after Xiangjiaba impoundment, the scikit-learn machine learning algorithm (K neighbor regression, linear regression, decision tree regressoin, linear SVR regression and artificial neural network) was used to establish the temperature prediction model (TPM) of each meteorological station in the dam area of Xiangjiaba hydropower station through sample training and cross validation. The model was applied to quantitatively analyze the influence of flood discharge atomization on the temperarture of the region behind the dam from the aspeccts of temporal and spatial variation and influence degree. The results show that the temperature behind the dam is less affected by the flood discharge of Xiangjiaba Hydropower Station, and the influence degree decreases rapidly with the increase of the distance from the flood discharge orifice. The riverside weather station nearest to the flood discharge outlet is most affected by the flood discharge atomization at 12~18 o 'clock daily, and reaches its peak at 13 o 'clock, and the temperature influence value is mainly within-2.0 °C.