Abstract:The micro meteorological environment generated by local micro topography is one of the important factors causing meteorological forecasting errors, and also an important factor leading to the low accuracy of ice cover forecasting by power grid companies. Using high-precision MODIS system 15s (about 500m) terrain data to drive a mesoscale weather research and forecasting (WRF) model, and using LightGBM to correct the WRF forecast, the improvement effect of ice cover prediction after correction is evaluated through local case studies. The results show that under the assumed conditions, the ice deposition rate of supercooled droplets increases rapidly with decreasing temperature, and then the growth rate remains unchanged. Moreover, the larger the droplet size, the lower the temperature required for complete freezing; The WRF LightGBM correction algorithm effectively improves the accuracy of temperature forecasting in mountainous micro terrain, with an error of less than 2 ℃ between predicted and actual temperatures under typical winter cold wave conditions, and a forecasting accuracy of 76%; In a case of ice cover on a tower, after inputting corrected temperature and relative humidity data, the melting period of the ice cover was eliminated, and the ice cover thickness curve was basically consistent with the actual situation, with a growth rate close to the same.