Abstract：Aiming at the limitations of temperature data from a single meteorological station,a deep fusion prediction system for multisource temperature data was constructed.A hybrid architecture combining Convolutional Neural Network (CNN) and Long ShortTerm Memory (LSTM) was adopted to collaboratively process multidimensional temperature information such as ground observation,satellite remote sensing,and numerical prediction.The attention mechanism was used to dynamically allocate the weights of different data sources to improve fusion accuracy,and a temperature change rate adaptive loss function was designed to optimize the prediction during periods of drastic temperature changes.Experimental results show that the prediction accuracy of the model during periods of rapid temperature changes is significantly superior to traditional methods,with the mean absolute error (MAE) ≤ 0.8℃.In addition,the model can adaptively identify and correct abnormal data to ensure the reliability of output results,providing a new technical path for accurate temperature prediction.
岳昊. 基于深度学习的多源气温数据融合与精确预报模型研究[J]. 气象水文海洋仪器, 2026, 43(5): 52-55.
Yue Hao. Research on multisource temperature data fusion and accurate prediction model based on deep learning. Meteorological Hydrological and Marine Instrument, 2026, 43(5): 52-55.
2026, Vol. 43 
