Based on the daily maximum power load and meteorological data of Xi'an from 2010 to 2013, the variation law of daily maximum power load was analyzed, and the meteorological load was separated from daily maximum power load using the least square method after eliminating the holiday effect and weekend effect of daily maximum load. Then the correlation between meteorological load rate and temperature, relative humidity, total cloud, precipitation and wind speed in Xi'an was established. Based on the data of the period from November to February and the period from June to August from 2010 to 2012, the forecast models between the maximum meteorological load and the main meteorological factors were established by stepwise regression, multiple linear regression and BP neural network, respectively. The daily maximum meteorological load rate of the corresponding time in 2013 was taken as the independent sample test of the forecast effect. The results show that the daily maximum load in Xi'an from 2010 to 2013 has an obvious increasing trend year by year, and there are obvious weekend effect and holiday effect. Temperature is the most significant factor affecting the meteorological load rate, and the BP neural network algorithm with Temperature and Humidity Index (THI) is the best for fitting and forecasting the meteorological load rate.