杭州师范大学学报（自然科学版）2025，Vol.24Issue(4)：347-355,364,10.DOI:10.19926/j.cnki.issn.1674-232X.2024.08.121
基于多层注意力机制的多元空气污染物浓度预测方法研究
Research on Multivariate Air Pollutant Concentration Prediction Based on Multi-layer Attention Mechanism
摘要
Abstract
To address the challenges in air pollutant concentration prediction,such as single variable types,difficulties in spatial-temporal feature extraction,and complex dependencies among dimensions(variables),this study proposed a multi-layer attention mechanism-based framework for multivariate air pollutant concentration prediction(spatial-temporal-dimensional Transformer,STD-Transformer).The model utilized a multi-modal segment embedding method,integrating spatial-temporal and dimensional feature embedding techniques for data preprocessing.By constructing a three-layer attention mechanism(temporal,spatial and dimensional),the model efficiently extracted multi-dimensional spatial-temporal features and focuses on critical information.The input included multi-site air quality data and geographical location information,enabling simultaneous predictions of PM2.5,PM10,NO2,SO2 and other pollutant concentrations.Experimental results demonstrated that the STD-Transformer reduced the mean absolute error(MAE)by 11.46％compared to existing advanced methods in multi-step predictions(1 to 12 hours)for multiple pollutants,highlighting its superior performance in handling complex environmental data.关键词
多元空气污染物浓度预测/注意力机制/变量依赖性/时空特征Key words
multivariate air pollutant concentration prediction/attention mechanism/variable interdependence/spatio temporal features分类
资源环境引用本文复制引用
王相朝,刘俊,蔡博涵..基于多层注意力机制的多元空气污染物浓度预测方法研究[J].杭州师范大学学报（自然科学版）,2025,24(4):347-355,364,10.基金项目
国家自然科学基金项目(62273126). （62273126）
