首页|基于注意力常微分方程的PM2.5浓度预测及其可解释性分析

基于注意力常微分方程的PM2.5浓度预测及其可解释性分析

扫码查看
原文链接
  • NETL
  • NSTL
  • 万方数据
可解释的准确预测PM2.5浓度变化可以有助于人类规避暴露风险,对人类健康风险评估和政策实施具有重要意义.目前已有PM2.5浓度预测模型过多专注于提升模型预测精度,但忽略了模型的可解释性,造成模型可复用性和可信任度较差.鉴于此,本文提出了一种兼顾模型预测精度与模型可解释性的注意力时空常微分方程模型(Attentional SpatioTemporal Ordi-nary Differential Equation,ASTODE)用于PM2.5浓度预测任务.具体而言,本文将神经常微分方程集成至PM2.5浓度预测任务中,以提升预测模型的可解释性.此外,针对传统神经常微分方程难以挖掘PM2.5浓度数据中空间依赖关系的挑战,本文提出了一种新颖时空导数网络将传统神经常微分方程扩展到了时空常微分方程.针对传统神经常微分方程难以挖掘PM2.5浓度数据中长期依赖关系的挑战,本文设计了一种时空注意力机制去融合多个时间节点的隐藏状态.本文采用真实的PM2.5浓度数据集对提出的ASTODE模型进行了验证.实验结果表明,ASTODE模型不仅在预测精度上优于或逼近于存在的6个基线方法,并且在可视化的视角下具有良好的可解释性.
An Attentional Ordinary Differential Equation for Predicting PM2.5 Concentration and Its Interpretability Analysis
Accurate and explainable prediction of PM2.5 concentration can help humans avoid exposure risks to air pollution,which is of great significance for human health risk assessment and policy implementation.Currently,the existing PM2.5 concentration prediction models focus on improving the model prediction accuracy without considering model interpretability,resulting in poor model reusability and trustworthiness.Therefore,this paper proposes an Attentional Spatiotemporal Ordinary Differential Equation(ASTODE)model for PM2.5 concentration prediction tasks considering both prediction accuracy and model interpretability.Specifically,this paper integrates the Neural Ordinary Differential Equation(NODE)into the PM2.5 concentration prediction task to improve the interpretability of the prediction model.In addition,to address the challenge of traditional NODE in mining spatial dependencies in PM2.5 concentration data,this paper proposes a novel spatiotemporal derivative network that extends the traditional NODE to spatiotemporal ordinary differential equations.To address the challenges of traditional NODE in mining long-term dependencies in PM2.5 concentration data,this paper proposes a spatiotemporal attention mechanism to fuse hidden states of multiple time nodes.In the experimental section,the proposed ASTODE model is validated using a real PM2.5 concentration dataset.This paper quantifies the prediction errors of the ASTODE model in both temporal and spatial dimensions.By comparing with six existing baseline methods,our proposed ASTODE model obtains a similar or higher prediction accuracy.This paper also analyzes the interpretability of our proposed ASTODE model from a visualization perspective,demonstrating that the proposed ASTODE model balances the prediction accuracy and interpretability to some extent.

PM2.5 concentration predictionair pollutionenergy conservation and emission reductionspatiotemporal predictionattention mechanismNeural Ordinary Differential EquationSpatiotemporal Ordinary Differential Equationmodel interpretability

王培晓、张恒才、张彤、陆锋

展开 >

中国科学院地理科学与资源研究所资源与环境信息系统国家重点实验室,北京 100101

中国科学院大学资源与环境学院,北京 100049

武汉大学测绘遥感信息工程国家重点实验室,武汉 430079

政务大数据应用省部共建协同创新中心,福州 350003

展开 >

PM2.5浓度预测 空气污染 节能减排 时空预测 注意力机制 神经常微分方程 时空常微分方程 模型可解释性

国家重点研发计划国家博士后创新人才支持计划中国博士后科学基金面上项目中国科学院特别研究助理项目国家自然科学基金武汉大学测绘遥感信息工程国家重点实验室开放基金资源与环境信息系统国家重点实验室创新项目

2022YFB3904102BX202303602023M7434544237147023I0308R8A092YA

2024

10.12082/dqxxkx.2024.230678

地球信息科学学报
中国科学院地理科学与资源研究所

地球信息科学学报

CSTPCD北大核心
影响因子：1.004
ISSN：1560-8999
年,卷(期)：2024.26(6)
  • 30