ConvLSTM模型在空气污染物浓度短时预测中的应用研究  

Research on the Application of ConvLSTM Model in ShortTerm Prediction of Air Pollutant Concentrations

作　　者：田旭光 王丁 张成名 刘宁 左钦文 TIAN Xuguang;WANG Ding;ZHANG Chengming;LIU Ning;ZUO Qinwen(Chemical Defense Institute,Academy of Military Sciences,Beijing 102205,China)

机构地区：[1]军事科学院防化研究院,北京102205

出　　处：《防化研究》2025年第1期47-55,共9页CBRN DEFENSE

摘　　要：本文应用ConvLSTM (Convolutional Long Short-Term Memory)模型解决了空气污染物浓度短时预测的问题。首先基于卷积神经网络和长短期记忆网络对ConvLSTM模型的构建方法进行了探讨,深入剖析了模型的基本构成与结构特性,然后通过实验实例详细展示了该模型在空气污染物浓度预测领域的应用过程,包括数据预处理、模型训练、预测结果分析等。实验结果表明,ConvLSTM模型用于空气污染物浓度短时预测的精度较高(始终保持在0.1%以内),同时也表明模型预测精度与时间序列并非总是正相关。当时间步长在某个值(本文实验中时间步长为10)附近时,模型预测精度较高。本研究可为其他具有类似时空特征数据序列的预测问题提供参考。In this paper,the ConvLSTM(Convolutional Long Short-Term Memory) model has been applied to solve the problem of short-term forecasting of air pollutant concentrations.First,the construction method of the ConvLSTM model was discussed based on convolutional neural networks and long short-term memory networks,delving into the basic composition and structural characteristics of the model;then,the application process of the model in the field of air pollutant concentration forecasting was demonstrated in detail through experimental examples,including data preprocessing,model training and analysis of prediction results.The experimental results indicated that the ConvLSTM model had a high accuracy for short-term forecasting of air pollutant concentrations(always maintained within 0.1%),and also showed that the model's prediction accuracy was not always positively correlated with the time series,but rather that the prediction accuracy was higher when the time step was around a certain value(the time step was 10 in this experiment).This study was expected to provide a reference for other prediction problems involving data sequences with similar spatiotemporal characteristics.

关 键 词：ConvLSTM模型 空气污染物 浓度 短时预测 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]