基于改进麻雀算法的PCC-DBN-LSTM气温预测模型  

PCC-DBN-LSTM Temperature Prediction Model based on Improved Sparrow Algorithm

作　　者：王冬萌 文斌[1] 李晓燕 徐越 刘书慧[3] 付世军[3] WANG Dongmeng;WEN Bin;LI Xiaoyan;XU Yue;LIU Shuhui;FU Shijun(College of Communication Engineering,Chengdu University of information Technology,Chengdu 610225,China;Huozhou Meteorological Bureau,Huozhou 031400,China;Nanchong Meteorological Bureau,Nanchong 637000,China)

机构地区：[1]成都信息工程大学通信工程学院,四川成都610225 [2]山西省霍州市气象局,山西霍州031400 [3]四川省南充市气象局,四川南充637000

出　　处：《成都信息工程大学学报》2024年第5期527-533,共7页Journal of Chengdu University of Information Technology

基　　金：四川省科学技术厅重点研发资助项目(2023YFN0051)。

摘　　要：气温预测是气象学中的一个重要研究领域。随着气象精准化发展,迫切需要提升气温预测的精准度。为解决传统气温预测算法效果不佳,并且对于多个站点气象数据时空特征提取能力不足,提出一种基于改进麻雀算法优化的皮尔逊积矩相关系数(PCC)-深度置信网络(DBN)-长短时记忆网络(LSTM)的气温预测模型。首先利用Pearson相关系数对众多的气象参数进行选择,DBN网络对输入的多站点气象数据特征进行提取和降维,LSTM对提取的特征进行建模和预测。由于模型初始化参数众多,提出改进麻雀算法优化DBN-LSTM网络参数,提高模型的预测精度和稳定性。实验表明:所提模型的RMSE为0.527,精度高于单一模型和同类模型。Temperature prediction is an important research field in meteorology.With the development of meteorological technology,there is a need to improve the accuracy of temperature predictionurgently.To solve the poor effect of traditional temperature prediction algorithms and the insufficient ability to extract the spatiotemporal characteristics of meteorological data from multiple stations,a multi-station temperature prediction model based on the Pearson product-moment correlation coefficient(PCC)-Deep Belief Network(DBN)-Long Short-Term Memory Network(LSTM)optimized by the improved Sparrow algorithm was proposed.Firstly,Pearson correlation coefficients are used to select numerous meteorological parameters,DBN networks extract and reduce the dimensionality of input multi-site meteorological data,and LSTM modelspredict the extracted features.Due to the numerous initialization parameters of the model,an improved sparrow algorithm is proposed to optimize the parameters of the DBN-LSTM network,improving the prediction accuracy and stability of the model.Experiments show that the RMSE of the proposed model is 0.527,which is lower than that ofa single modeland other similar models.

关 键 词：气温预测 皮尔逊积矩相关系数 深度置信网络 改进麻雀算法 长短时记忆网络 

分 类 号：TP389.6[自动化与计算机技术—计算机系统结构]