线性分解和周期增强Informer的太阳辐射短临预报研究  

LINEAR DECOMPOSITION AND PERIODIC ENHANCEMENT IN SHORT-TERM SOLAR IRRADIANCE FORECASTING WITH INFORMER

作　　者：姚蕊 刘小芳 Yao Rui;Liu Xiaofang(School of Computer Science and Engineering,Sichuan University of Science&Engineering,Yibin 644000,China;Department of Information Technology,Shanxi Yuncheng Vocational and Technical College of Agricultural,Yuncheng 044000,China)

机构地区：[1]四川轻化工大学计算机科学与工程学院,宜宾644000 [2]山西运城农业职业技术学院信息技术系,运城044000

出　　处：《太阳能学报》2025年第2期505-510,共6页Acta Energiae Solaris Sinica

基　　金：高层次创新人才培养专项(B12402005);四川轻化工大学人才引进项目(2021RC16);教育部高等教育司产学合作协同育人项目(202101038016)。

摘　　要：针对辐射周期趋势及外部影响特征捕获不足的问题,提出一种线性分解和周期增强Informer的地表太阳辐射短临预报方法。首先,改进灰色关联度方法,获取历史辐射与多种外部气象因素关联度,提取16种高相关外部气象特征建立高关联特征集,强化捕捉辐射与气象因素之间的复杂关系的能力;其次,在基于Transformer解决方案的基础上引入周期性嵌入层和ReLU激活函数,为模型提供更准确、合理的周期时间特征和辐射变化区间。最后,在Informer后增加平滑序列分解线性层,将Autoformer中的分解方案和FEDformer中的线性层相结合,进一步增强捕捉时序数据中周期性和季节性成分的能力。实验结果表明:该IDL方法结合外部气象特征能极好地提高模型短临预报效果,精度高于近年来基于Transformer系列的解决方案;比DLinear均方误差最高减少30.6%。In response to the inadequacy in capturing trend and periodic features,a method for short arrival prediction of surface solar radiation:s proposed decomposition and periodic-enhancement linear Informer.Firstly,an improved grey relational analysis method captures the correlation between historical irradiance and various external meteorological factors.This method extracts 16 highly correlated external meteorological features,thus enhancing the model′s ability to capture the intricate relationship between irradiance and meteorological factors.Subsequently,the Informer model is augmented with periodic embedding layers and ReLU activation functions,which better represents the periodic variations in solar irradiance and provides more accurate temporal features.Finally,an integration of the decomposition scheme from Autoformer and the linear layer from FEDformer as a decomposition linear layer after the Informer.This amalgamation enhances the model’scapability to capture the periodic trends and seasonal components in time series solar irradiance data.Experimental results demonstrate that the proposed Informer Decomposition Linear model,in conjunction with external meteorological features,remarkably improves short-term forecasting performance,surpassing the accuracy of recently prominent Transformer-based approaches.In comparison to the best-performing DLinear model,the maximum mean squared error is reduced by 30.6%.

关 键 词：太阳辐射 INFORMER TRANSFORMER 平滑序列线性分解 周期嵌入 灰色关联度 

分 类 号：P456.1[天文地球—大气科学及气象学]