Attention-GRU在PM_(2.5)浓度预测中的应用研究  

Research on the Application of Attention-GRU in PM_(2.5) Concentration Prediction

作　　者：张黎鹏 刘庆杰[1] ZHANG Lipeng;LIU Qingjie(Institute of Disaster Prevention,Langfang 065201,China)

机构地区：[1]防灾科技学院,河北廊坊065201

出　　处：《现代信息科技》2025年第4期74-79,86,共7页Modern Information Technology

摘　　要：针对PM_(2.5)浓度预测问题,选取北京市顺义监测站的每小时空气质量数据及其对应的气象数据作为研究样本,提出了一种融合多头注意力机制的GRU模型(Attention-GRU)。该模型利用门控循环单元(GRU)捕捉时间序列中与目标特征的长期依赖关系,并通过多头注意力策略来优化多特征与PM_(2.5)浓度的权重分布,关注影响较大的特征因素,从而提升预测的准确性。实验结果表明,与传统方法相比,融合多头注意力机制的GRU模型在均方根误差(RMSE)、平均绝对百分比误差(MAPE)和决定系数(R^(2))等指标上表现优异,验证了该方法的有效性和优越性。Regarding the problem of PM_(2.5) concentration prediction,hourly air quality data and corresponding meteorological data from the Shunyi monitoring station in Beijing are selected as research samples,and a GRU model (Attention-GRU) integrating a Multi-head Attention Mechanism is proposed.The model makes use of the Gated Recurrent Unit (GRU) to capture the long-term dependency relationship with the target feature in the time series.Moreover,it optimizes the weight distribution of multiple features and PM_(2.5) concentration through a multi-head attention strategy,focusing on the feature factors with greater influence,so as to improve the prediction accuracy.Experimental results indicate that compared with traditional methods,the GRU model integrating a Multi-head Attention Mechanism performs outstandingly in the indicators such as Root Mean Square Error (RMSE),Mean Absolute Percentage Error (MAPE),and coefficient of determination (R~2),validating the effectiveness and superiority of this proposed method.

关 键 词：多头注意力机制 PM_(2.5)预测 门控循环单元(GRU) 空气质量 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]