Studying the characteristics of temperature differences on bridge decks and their prediction models can provide decision-making basis for traffic management departments to predict severe weather and reduce traffic accidents. Based on observation data from three traffic meteorological stations on the Tianxingzhou Bridge section in Wuhan over the past three years, including the minimum temperature, air temperature, wind speed, precipitation, etc., for every five minutes, the daily differences in the minimum temperature between the bridge deck and the road surface, the hourly variation characteristics of typical weather cases, and the temperature change patterns under different weather conditions are analyzed. The prediction models for the minimum temperature of the bridge deck are established by using multiple linear regression and BP (Back Propagation) neural network methods, and the models are driven and tested using intelligent grid minimum temperature prediction products. The results indicate that due to differences in engineering structure, pavement material, geographical environment, and environmental meteorological factors, the temperature of the bridge deck is usually lower than that of the pavement, and the temperature difference between the two is the largest under sunny conditions. The speed at which the temperature on the bridge deck drops below freezing point is faster, and the duration of low temperature maintenance is longer. Both multiple linear regression and BP neural network methods can achieve good prediction results. Among them, BP method is more suitable for scenarios that require high prediction accuracy, while multiple linear regression method is suitable for applications that require high prediction accuracy.

Based on the daily maximum power load values in Jingzhou, Jingmen, Yichang, Xianning and Suizhou areas of Hubei Province from 2008 to 2019 and the meteorological data of national meteorological observation stations in the same period, the relationships between the change rate of maximum meteorological load (L_pm), four comfort indexes such as temperature and humidity index (I), meteorological sensitive load index (MSLI), human comfort (ET) and somatosensory temperature index (Te) and temperature were analyzed. The daily maximum power load forecasting models were established based on the above four comfort indexes by using multiple regression and BP neural network method. The results show that the L_pm was positively correlated with temperature and the above four comfort indexes in summer, negatively correlated in winter, and the correlation was significantly higher in summer than in winter. The changes of the above four comfort indexes integrating temperature, humidity and wind speed could cause the change of the L_pm, and this change was more obvious in summer, especially in July and August. The errors of BP neural network model and multiple regression model were basically controlled within the requirements of the power department. The prediction effect of BP neural network was better than that of multiple regression. In the later business application, it was suggested to select ET index in Jingmen and Xianning, and four indexes in other cities can be used.