基于GRU的未来24小时高低温预报技术研究

Research on high and low temperature forecast technology for the next 24 hours based on GRU

温度作为重要的气象要素,关乎民生和生产,其中最高温和最低温更是引人关注,但相关的研究却鲜有涉及。文中基于GRU模型提出了一种未来24小时高低温的AI预报算法,并针对数据缺失情况设计了5种数据处理方法,利用实况和多种模式资料制作AI训练数据集,以过去72小时实况数据和模式未来24小时预报数据为输入。实验表明,该方法能够有效提高高低温的预报精度,最高温预报误差为1.59℃,最低温预报误差为1.19℃,预报精度高于EC模式和预报员的预报精度,尤其是最低温预报精度提升比较明显,对预报员具有较好的预报指导意义。

Temperature is an important meteorological element related to people’s livelihood and production,and the highest and lowest temperature among them attract people’s attention,but the relevant research is rather little.Based on GRU model,this paper proposes an AI prediction algorithm for high and low temperature in the next 24 hours,and five data processing methods are designed based on the missing data.AI training data sets are made using live and multiple mode data,taking the past 72 hours of live data and the forecast data of the next 24 hours as input.The experiment shows that this method can effectively reduce the prediction error of high and low temperatures.The prediction error of the highest temperature is 1.59℃,and the prediction error of the lowest temperature is 1.19℃.The prediction accuracy is higher than that of the EC model and the forecasters.In particular,the prediction accuracy of the lowest temperature is significantly improved,which has a good guiding significance for the forecasters.