Abstract: Based on observed climatic data of precipitation and temperature from 6 meteorological stations in Shanxi province during 1951-2000, the prediction model of the lowest average monthly temperature in Taiyuan city was established with neural network technology and principal component analysis (PCA) method. According to this prediction model, the average relative error to the data series of 1960-1990 was 3.6%, and the prediction accuracy to 1991-2000 was 90%. The results indicated that the prediction model could improve prediction scale and precision significantly, and could take positive effect in reducing and preventing agricultural disasters.

Key words: The lowest average monthly temperature, The lowest average monthly temperature, Neural network, Principal component analysis, Prediction