The active response to the climate change is one of the goals of sustainable development. This paper presents a temperature prediction model based on the graph attention mechanism. The attention mechanism on the topology of the temperature sites is used to selectively aggregate the temperature feature of the surrounding area. Then the neural network is used to fit the complex temperature change pattern and forecast the future temperatures. In the experiments, the temperature data of Beijing-Tianjin-Hebei region from 2000 to 2010 are used. A large number of experiments show that with this method more accurate predictions can be made with a small amount of historical temperature data. The model can provide the decision support for the climate prediction and the climate disaster prevention, with an important scientific and practical significance.

temperature prediction  /  graph neural network  /  attention mechanism