基于GRU-CNN模型的云南地区短期气温预测  

Short-Term Temperature Prediction in Yunnan Region Based on GRU-CNN Model

作　　者：刘家辉 梅平[1] 刘长征[2] 刘剑南 LIU Jia-hui;MEI Ping;LIU Chang-zheng;LIU Jian-nan(School of Atomation,Nanjing University of Information Science and Technology,Nanjing Jiangsu 210044,China;National Climate Center,Beijing 100081,China)

机构地区：[1]南京信息工程大学自动化学院,江苏南京210044 [2]国家气候中心,北京100081

出　　处：《计算机仿真》2023年第9期472-476,共5页Computer Simulation

基　　金：国家重点研发计划课题(2017YFC1502403)。

摘　　要：为了利用较少的预测数据提高短期温度预测精度,将CNN(卷积神经网络)与GRU(门控循环单元网络)相结合,开展了深度学习技术在云南地区的日最高最低气温预测的应用;提取了云南地区8个站点1980-2019年的日度最高最低气温数据,将过去三十天的最高最低气温数据进行预处理后输入CNN、GRU与GRU-CNN模型进行训练,最终利用训练好的模型对站点未来三天的最高最低气温进行预测。在利用RMSE(均方根误差)和MAE(平均绝对误差)对预测效果进行评价后,结果显示GRU-CNN模型的预测效果显著优于CNN模型和GRU模型。GRU模型可以提取序列的时间变化特征,而CNN可以提取数据空间变化的深层局部特征,二者的结合提高了模型的适应能力,让模型可以应对诸如气温预测等复杂的深度学习问题。In order to improve short-term temperature prediction accuracy using less prediction data,this paper combined convolutional neural network(CNN)and gated recurrent unit network(GRU)to carry out the application of deep learning techniques for daily maximum and minimum temperature prediction in Yunnan region.The daily degree and minimum temperature data of eight stations in Yunnan region from 1980-2019 were extracted.The maximum and minimum temperature data of the past thirty days were preprocessed and input into CNN,GRU and GRU-CNN models for training.And finally the trained models were used to predict the maximum and minimum temperatures of the stations for the next three days.After evaluating the prediction effect using root mean square error(RMSE)and mean absolute error(MAE),the results show that the GRU-CNN model has significantly better prediction effect than the CNN model and GRU model;the GRU model can extract the time-varying features of the series,while the CNN can extract the deep local features of the spatial variation of the data,and the combination of the two improves the adaptability of the model,allowing the models to cope with complex deep learning problems such as temperature prediction.

关 键 词：气温预测 卷积神经网络 门控循环单元网络 深度学习 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]