IG-LSTM模型在空气质量指数预测中的应用  

Application of IG-LSTM model in AQI prediction

作　　者：陈岑 田晓丹 武文星 CHEN Cen;TIAN Xiaodan;WU Wenxing(School of Computer,North China University of Science and Technology,Yanjiao,065201,China)

机构地区：[1]华北科技学院计算机学院,北京东燕郊065201

出　　处：《华北科技学院学报》2020年第4期85-91,共7页Journal of North China Institute of Science and Technology

基　　金：国家重点研发计划资助项目(2018YFC0808306);河北省重点研发计划资助项目(19270318D);河北省物联网监控工程技术研究中心资助项目(3142018055);青海省物联网重点实验室资助项目(2017-ZJ-Y21)。

摘　　要：空气质量指数(Air Quality Index,AQI)是衡量空气质量优劣的重要指标,对安全生产和人民生活具有重要的指导作用。但由于空气质量的状态受到多个污染因子的影响,具有很强的时序性,传统回归预测方法的效率和精确度都较低,鉴于此,文中提出了一种基于IG(信息增益)和长短期记忆网络(Long Short-Term Memory,LSTM)的空气质量指数混合预测方法。本文首先基于TensorFlow机器学习框架搭建了动态预测模型,并利用IG(信息增益)减少输入变量的数量以及降低模型的复杂程度,然后在某市2014年至2018年数据集的基础上训练和测试模型,并使用回归评估指标对模型的结果进行量化,通过与传统BP模型、LSTM模型的对比分析,得出IG-LSTM混合预测模型具有更低的预测误差和损失值,能较精准地预测空气质量指数。Air quality index(AQI)is an important index to measure the quality of air,which plays an important role in guiding production safety and people's life.However,the state of air quality is affected by multiple pollution factors,so it has strong time sequence.The efficiency and accuracy of traditional regression prediction methods are low.In view of this,this paper proposes a hybrid prediction method of air quality index based on information gain(IG)and Long Short-Term Memory(LSTM).In this paper,a dynamic prediction model is built based on tensorFlow machine learning framework,and IG is used to reduce the number of input variables and the complexity of the model.Then,the model is trained and tested on the basis of a city's data set from 2014 to 2018,and the results of the model are quantified using regression evaluation indicators.Through the comparison between traditional BP model and LSTM model,the IG-LSTM hybrid prediction model has lower prediction error and loss value and canpredict the air quality index more accurately.

关 键 词：LSTM模型 信息增益 空气质量指数 大气污染预测 

分 类 号：X511[环境科学与工程—环境工程]