GPAformer模型在气温预测中的应用研究  

Research on Application of a GPAformer Model for Temperature Prediction

作　　者：王文豪 李秀芹 WANG Wen-hao;LI Xiu-qin(School of Information Engineering,North China University of Water Resources and Electric Power,Zhengzhou 450046,China)

机构地区：[1]华北水利水电大学信息工程学院,河南郑州450046

出　　处：《计算机技术与发展》2025年第3期187-193,共7页Computer Technology and Development

摘　　要：气温作为关键气象变量,对环境、农业和公共健康具有重要影响,精准预测是应对气候变化的基础。深度学习在气温预测中表现出对非线性关系和复杂模式建模的优势,但在面对多维度气温数据处理及长期依赖关系捕捉方面仍有不足。为此,提出了一种高斯过程驱动的Autoformer(GPAformer)气温预测模型,结合高斯过程算子和自相关机制,通过对气象数据集的处理,增强了对气温时间序列变化的建模能力,提供更精准的预测。在印度德里气象数据集上的7天气温预测实验中,该模型的平均绝对误差(MAE)和均方误差(MSE)分别为0.0506和0.0135,相比Autoformer、Informer、Transformer、LSTM、GRU、MLP、RF和ARIMA模型的MAE分别降低了43.96%、50.05%、62.10%、80.72%、73.51%、78.23%、74.83%和79.57%。结果显示,该模型在捕捉气温变化趋势上具有显著优势,并在进一步验证后有望应用于其他地区的气温预测。Temperature is a key meteorological variable with significant influence on the environment,agriculture,and public health.Accurate forecasting is crucial for mitigating the effects of climate change.While deep learning models have shown promise in capturing the nonlinear relationships and complex patterns in temperature prediction,they often struggle with handling multidimensional temperature data and modeling long-term dependencies.To address these challenges,we introduce the Gaussian Process-driven Autoformer(GPAformer)model for temperature prediction.By integrating Gaussian Process operators with a self-correlation mechanism,the GPAformer enhances the modeling of temperature time series,resulting in more accurate forecasts.In a 7-day prediction experiment using the Delhi meteorological dataset,the GPAformer model achieved a mean absolute error(MAE)of 0.0506 and a mean squared error(MSE)of 0.0135.These represent reductions in MAE of 43.96%,50.05%,62.10%,80.72%,73.51%,78.23%,74.83%and 79.57%compared to several benchmark models,including Autoformer,Informer,Transformer,LSTM,GRU,MLP,RF,and ARIMA.These results highlight the GPAformer model's effectiveness in capturing temperature trends and its potential for broader application in other regions.

关 键 词：气温预测 深度学习 Autoformer 高斯过程 时间序列分析 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]