摘要
使用传统单一模型预报气温经常出现漏报现象,最终导致预测结果不理想,精度较低。针对单一预报模型稳定性较低,随机性偏高,突发性较多的特点,在深度学习理论的基础上,提出一种采用门控循环单元(GRU)和灰色模型(GM)集成的方法,先分别训练两个模型,再通过权值ω将二者的预测结果进行加权组合,权值ω适当调整模型,改善模型的预报结果,提高模型的预报精度,并加快了运行速度,并且其普遍适用性和应急突发能力得到巨大改善。实验表明,将GRU神经网络加入灰色模型进行气温预报,效果要明显优于单一的模型,其标准差小了近一倍,从而表明实验方法的可行性和有效性。
The use of traditional single model to forecast temperature often results in omissions,which ultimately leads to unsatisfactory prediction results and low accuracy.To deal with the low stability,high randomness,and excessive abrupt occurrence in a single forecast model,the present study proposes a model using gated recurrent unit(GRU)and grey model(GM)based on deep learning theory.The two models are first trained separately,and then the prediction results of the two are combined using weight w.The weight w appropriately adjusts the models to improve prediction results,increase prediction accuracy,and speed up operation.The universal applicability and emergency response capabilities of the integrated model have been greatly improved.Experiments show that the grey model integrated with GRU neural network for temperature forecasting is significantly better than a single model,and its standard deviation is nearly cut by half,which shows the feasibility and effectiveness of the experimental method.
作者
周满国
黄艳国
杨训根
ZHOU Man-guo;HUANG Yan-guo;YANG Xun-gen(School of Electrical Engineering and Automation,Jiangxi University of Science and Technology,Ganzhou 341000,China)
出处
《热带气象学报》
CSCD
北大核心
2020年第6期855-864,共10页
Journal of Tropical Meteorology
基金
国家自然科学基金(72061016)
江西省教育厅科技项目(GJJ160608)共同资助。
