Based on observed total solar radiation, air temperature, relative humidity and air pressure data at representative photovoltaic power stations of Gansu Province, total solar radiation data forecasted by WRF model, and total cloud cover products from FY satellite in 2019, the correlation between total solar radiation and meteorological factors was analyzed, and the prediction ability of WRF model was evaluated, firstly. And on this basis the errors of short-time solar radiation forecast were corrected. The results show that the atmospheric transmissivity was positively correlated with air temperature, and the correlation coefficient was 0.61, while it was negatively correlated with relative humidity, air pressure and total cloud cover, and the correlation coefficients were -0.44, -0.31 and -0.81 in turn. The contribution of total cloud cover to solar radiation attenuation was the most, followed by relative humidity. The deviation of solar radiation forecasted by WRF model was bigger, and the monthly distribution of forecast errors appeared ‘single peak’ pattern, the forecast errors was the biggest in June. The root mean square error (RMSE) of solar radiation forecast was the smallest in winter (45.63 W·m ^-2) and the biggest in summer (240.4 W·m^-2). The forecast ability of WRF model was better on sunny days or partly cloudy days, while it was worse on cloudy days. The forecast errors mainly came from phase bias and system bias. The correction effect of solar radiation forecast considering cloud cover was significant, the RMSE of solar radiation forecast after correction sharply decreased by 101-216.4 W·m^-2 on cloudy days, the average absolute error decreased by 59.5-173.07 W·m^-2, and the RMSE decreased by 1.92-64.23 W·m^-2 in summer with the maximum error.

Based on observation data and numerical forecast data at ZDLYFP photovoltaic power station from March 2017 to February 2019, the inclined plane total solar radiation algorithm was improved, firstly. And on this basis two forecast models of short-term photovoltaic output power were established by using multiple linear regression (MLR) and empirical formula methods, then the forecast results were tested and evaluated. The results are as follows: (1) The inclined plane total solar radiation and temperature were higher correlated with photovoltaic output power in each season, the total correlation coefficients were 0.896 and 0.386, respectively, so they were introduced to forecast model based on MLR method as the predictors of photovoltaic output power. (2) The forecast effect of short-term power improved after the improvement of inclined plane solar radiation algorithm, and the relative root mean square error (RRMSE) of photovoltaic output power forecasted by two models of MLR and empirical formula methods reduced by 0.066 and 0.040, respectively. (3) The total root mean square error (RMSE) of photovoltaic output power obtained by MLR and empirical formula methods were 940.917 kW and 1147.172 kW, respectively, and total RRMSEs were 0.188 and 0.229. In addition, RMSEs and RRMSEs based on MLR method were less than those based on empirical formula method in each month, and the correlation coefficient of the former was slightly higher than that of the latter, which indicated that the forecast effect of MLR method was better and more stable in practical application. (4) The effect of photovoltaic power prediction was obviously distinct under different weather conditions, RRMSEs of two methods increased in turn for sunny weather, cloudy weather, overcast weather, rainy weather, dust weather and snow weather. In general, the effect of photovoltaic power prediction based on MLR method was better under different weather conditions.