With the increase of extreme weather events such as cold waves, power transmission line icing has become more and more severe, affecting the safe and stable operation of power systems. Thus, icing prediction has become crucial for power grids. In this study, we collect multi-source data including the historical observations of meteorological elements and transmission line icing in Sichuan during 2017–2019, and develop an artificial intelligence (artificial intelligence)-based integrated model to achieve icing thickness prediction according to meteorological elements. Using the Weather Research and Forecasting model and the three-dimensional variational data assimilation method, we analyze the weather conditions in Sichuan of China during the 2020 winter, and obtain the high-precision meteorological element fields that are related to icing prediction. The forecasted meteorological elements are then combined with the AI-based integrated model to predict icing conditions, assisting in the warning of transmission line icing. The results indicate that the AI-based integrated model displays superior performance on the accurate prediction of icing thickness in the test set, with only two samples having prediction errors of more than 3 mm. Data assimilation can effectively improve the forecast accuracy of meteorological elements near icing observation stations and thus enhance the accuracy of icing thickness prediction. In particular, icing thickness prediction is remarkably improved at Gaoqiao, Laolinkou and Erlangshan stations.

This study evaluates three bias correction methods, Ridge Regression, Random Forest, and FM-CNN-LSTM, to enhance ERA5 shortwave radiation estimates over Hubei Province, located within China’s second-step topographic region. Observations from six photovoltaic stations revealed significant systematic biases in the original ERA5 dataset, particularly overestimations (~50–100 W/m²) of peak radiation and underestimations during cloudy periods. Random Forest consistently provided the best corrections, achieving the highest correlation coefficients (0.84–0.88), lowest RMSE (86.14 W/m² at Wuhan station), and smallest MAE (60.05 W/m² at Wuhan station). Ridge Regression delivered stable but slightly inferior improvements, while FM-CNN-LSTM showed notable inconsistencies. Spatially, Random Forest-corrected data exhibited realistic radiation patterns aligned with complex topography, highlighting northeastern and eastern Hubei (>1350 kWh/m² annually) as ideal regions for photovoltaic development. Overall, the Random Forest method is recommended as the optimal bias correction approach, substantially improving solar radiation estimates critical for renewable energy planning in complex terrains.