基于DWT-CNN-LSTM逐日气温预测模型研究  

Daily Temperature Prediction Model Based on DWT-CNN-LSTM

作　　者：樊姝琪 刘慧铭 庄润杰 王诗雨 温永仙[1,2] FAN Shuqi;LIU Huiming;ZHUANG Runjie;WANG Shiyu;WEN Yongxian(College of Computer and Information Sciences,Fujian Agriculture and Forestry University,Fuzhou 350002,China;Institute of Statistics and Applications,Fujian Agriculture and Forestry University,Fuzhou 350002,China)

机构地区：[1]福建农林大学计算机与信息学院,福建福州350002 [2]福建农林大学统计及应用研究所,福建福州350002

出　　处：《热带气象学报》2024年第6期1063-1073,共11页Journal of Tropical Meteorology

基　　金：福建省自然科学基金项目(2021J01126);福建农林大学科技创新专项基金(KFb22094XA)共同资助。

摘　　要：气温的准确预测对人类生产生活、农业等方面至关重要。针对传统气温预测方法难以捕捉数据的动态变化、预测精度差等问题,提出了一种融合离散小波变换(Discrete Wavelet Transformation,DWT)、卷积神经网络(Convolutional Neural Network,CNN)和长短期记忆网络(Long Short-Term Memory,LSTM)的组合气温预测模型。首先,利用DWT对原始气温观测数据进行分解与重构;其次,使用CNN进行特征提取,使用LSTM对提取的特征信息进行处理,以实现气温预测。同时采用均方根误差、平均绝对值误差和决定系数作为评价指标;最后,使用气温观测数据验证所提模型的有效性,并分别与LSTM模型、CNN-LSTM模型和DWTLSTM模型进行对比分析。实验结果表明,与LSTM模型、CNN-LSTM模型和基于离散小波变换的LSTM模型相比,DWT-CNN-LSTM模型分别将RMSE降低了1.00924,1.00274,0.10023,MAE降低了0.91836,0.86265,0.14489,R^(2)提高了0.04703,0.04662,0.00400,验证了该模型在气温预测中的有效性。这一结果为气温预测领域提供了新的参考依据,并有望在未来得到更广泛的应用。Accurate temperature prediction is crucial for human production and life.To address the challenges of traditional temperature prediction methods,which struggle to capture dynamic data changes and often yield poor accuracy,this paper proposed a combined temperature prediction model that integrates discrete wavelet transform(DWT),convolutional neural network(CNN),and long short-term memory network(LSTM).First,the original temperature observation data were decomposed and reconstructed using the DWT.Second,the CNN was used to perform feature extraction,and the LSTM was applied to process the extracted feature information to achieve temperature prediction.Meanwhile,root mean square error(RMSE),mean absolute error(MAE),and coefficient of determination(R^(2))were used as the evaluation indexes.Lastly,temperature observation data were used to validate the effectiveness of the proposed model,and a comparative analysis was conducted using the LSTM model,the CNN-LSTM model,and the DWT-LSTM model.Experimental results show that,compared with the LSTM model,the CNN-LSTM model,and the LSTM model based on discrete wavelet transform,the DWT-CNN-LSTM model reduced the RMSE by 1.00924,1.00274,and 0.10023,respectively,and the MAE by 0.91836,0.86265,and 0.14489 respectively.It also improved R^(2) by 0.04703,0.04662,and 0.00400 respectively.These findings confirm the effectiveness of the model in temperature prediction and provide a new reference for temperature prediction,indicating potential for broader future applications.

关 键 词：气温预测 时间序列 离散小波变换 卷积神经网络 长短期记忆网络 

分 类 号：P456[天文地球—大气科学及气象学]