基于LSTM循环神经网络的PM_(2.5)预测  

PM_(2.5) PREDICTION BASED ON LSTM RECURRENT NEURAL NETWORK

作　　者：白盛楠 申晓留[2] Bai Shengnan;Shen Xiaoliu(School of Control and Computer Engineering,North China Electric Power University,Beijing 102200,China;Research Institute of Energy Internet and Electricity Big Data,North China Electric Power University,Beijing 102200,China)

机构地区：[1]华北电力大学控制与计算机工程学院,北京102200 [2]华北电力大学能源互联网与电力大数据研究所,北京102200

出　　处：《计算机应用与软件》2019年第1期67-70,104,共5页Computer Applications and Software

基　　金：国家自然科学基金项目(71071053);北京市自然科学基金项目(9122021)

摘　　要：PM_(2.5)要素对空气质量影响较大。PM_(2.5)浓度变化是多种因素作用的结果,且过程突发、非线性,具有明显的不确定性,难以使用传统的方法进行预测。针对该问题,以气象、大气污染物因素作为PM_(2.5)预测指标,提出基于LSTM循环神经网络的PM_(2.5)预测模型。使用灰色关联度分析方法对多个气象、大气污染指标进行关联强度分析;对数据进行平滑处理,将时间序列问题处理为监督问题;搭建多变量的LSTM循环神经网络PM_(2.5)预测模型,实现PM_(2.5)日值浓度的准确预测。使用北京市2010年-2017年气象数据和大气污染物数据进行仿真实验,结果表明该模型能够较好地预测PM_(2.5)的日值变化趋势。PM2.5 factor has a great influence on air quality.PM2.5 concentration change is the result of many factors,and the process is sudden,nonlinear,and has obvious uncertainty,so it is difficult to use traditional methods to predict.To solve this problem,a PM 2.5 prediction model based on LSTM recurrent neural network was proposed with meteorological and atmospheric pollutant factors as PM2.5 prediction indicators.We used the grey correlation analysis method to analyze the correlation intensity of several meteorological and atmospheric pollution indicators.The data was smoothed and the time series problem was treated as a supervisory problem.We built a multi-variable LSTM recurrent neural network PM2.5 prediction model to achieve accurate prediction of PM2.5 concentration.We used the meteorological data and air pollutant data of Beijing from 2010 to 2017 to simulate the experiment.The results show that the model can better predict the daily variation trend of PM2.5.

关 键 词：空气质量 PM2.5预测 灰色关联度 循环神经网络 LSTM 

分 类 号：TP39[自动化与计算机技术—计算机应用技术]