Based on daily gas load and meteorological observation data during heating period in Xi’an of Shaanxi Province from 15 November 2009 to 14 March 2019, the variation characteristics of gas load in heating period, holidays and weekends were analyzed. The significant influence factors on gas load were selected by using correlation analysis. And on this basis the daily forecast model of gas load in heating period was established by using multiple linear regression method, then the forecast model was tested. The results show that the natural gas consumption during heating period gradually increased in Xi’an in past 10 years, the daily gas load presented a single-peak pattern change, and the peak appeared in January. The weekend and holidays effects of gas load were obvious during heating period, the gas consumption on weekend and holiday was less than that on working days, and the longer holiday was, the less gas load was. The gas load was significantly and positively correlated with gas load on previous day, while that was significantly and negatively correlated with meteorological factors of the maximum and minimum temperature, mean temperature and human body comfortable degree, and the correlation between heating gas load separated from actual gas load and meteorological factors obviously improved. Based on the above five influence factors, the dynamic forecast model of daily heating gas load was established by using multiple linear regression method. Upon inspection, the average relative error of the model was 3.4%, and the model was more stable in rush hours of using gas, the average relative error was 2.77%, which could meet gas dispatch needs of natural gas companies.

Based on daily maximum and minimum temperature data of 99 national weather stations in Shaanxi, two methods of univariate linear regression and decaying average were used to correct the temperature forecasts of SCMOC (the data of the national meteorological center forecast), GRAPES_Meso (global/regional assimilation and prediction system) and ECMWF (European center for mediumrange weather forecasting). The result show that the prediction accuracy of daily minimum temperature was higher than that of daily maximum temperature for SCMOC, GRAPES_Meso and ECMWF. The accuracy of daily maximum and minimum temperature forecast of SCMOC was obviously highest among them, while that of GRAPES_Meso was lowest. The methods of univariate linear regression and decreasing average could significantly improve the accuracy of air temperature forecasts of GRAPES_Meso and ECMWF, but could not improve the accuracy of SCMOC. The accuracy of ECMWF’s daily maximum and minimum temperature forecast corrected from 2017 to 2019 was higher than that of SCMOC. The accuracy of GRAPES_Meso’s 24hour and 48hour daily maximum temperature prediction corrected in 2019 was higher than that of SCMOC, while the accuracy of daily minimum temperature prediction after correcting was still much lower than that of SCMOC. By comparison, the ability and stability of univariate linear regression method for rectifying air temperature forecast of numerical model were better than that of decreasing average method.

Based on the observation data of hourly temperature，daily maximum and minimum temperature，daily mean total cloud cover and rainfall from ten stations in Shaanxi Province from 2006 to 2010，a forecast method of hourly temperature was established by using linear regression method on the basis of forecast values of daily maximum and minimum temperature and observed values of hourly temperature，which was tested by comparing forecast values with observed values of hourly temperature at ten stations in Shaanxi Province in 2011． The results show that the forecasting ability of the forecast method of hourly temperature under sunny or lightly cloudy conditions was better than that under heavily cloudy or rainy conditions． The forecasting effect of the method was better between 2 o clock
and 18 o clock than that between 19 o clock and 1 o clock of the next day on sunny or lightly cloudy days，and was better between 1 o clock and 10 o clock than that at other time on heavily cloudy or rainy days． When the forecasted daily maximum and minimum temperature were comparatively accurate，the forecast accuracy of hourly temperature was more than 60% at Xi an station． The accuracy was 100% on sunny days and from 96% to 99% on lightly cloudy days between 14 o clock and 17 o clock． But the accuracy on heavily cloudy or rainy days was about 12% ～ 27% lower than that on sunny or lightly cloudy days between 11 o clock and 17 o clock． With the characteristic of diurnal variations of temperature change in different areas，seasons and sky conditions，the method can turn
the forecasted daily maximum and minimum temperature into forecast of hourly temperature well． To some extent the forecast method of hourly temperature has application and extension values．