基于Adam 注意力机制的PM2:5浓度预测方法

大气与环境光学学报 ›› 2021, Vol. 16 ›› Issue (2): 117-126.

基于Adam 注意力机制的PM2:5浓度预测方法

张怡文, 袁宏武∗, 孙鑫, 吴海龙, 董云春

安徽新华学院信息工程学院, 安徽合肥 230088

收稿日期:2019-12-16 修回日期:2020-11-29 出版日期:2021-03-28 发布日期:2021-03-28
通讯作者: E-mail: yuanhongwu@axhu.edu.cn E-mail:yuanhongwu@axhu.edu.cn
作者简介:张怡文 (1980 - ), 女, 安徽阜阳人, 硕士, 教授, 主要从事污染物预测模型的研究。 E-mail: yiwenzh@ustc.edu.cn
基金资助:
Supported by Anhui University Provincial Natural Science Research Project (安徽高校自然科学研究项目, KJ2019A0877), Anhui Provincial Quality Engineering Grassroots Teaching and Research Office Demonstration Project (安徽省省级质量工程基层教研室示范项目, 2018JYSSF111)

PM2:5 Concentration Prediction Method Based on Adam′s Attention Model

ZHANG Yiwen, YUAN Hongwu∗, SUN Xin, WU Hailong, DONG Yunchun

College of Information Engineering, Anhui Xinhua University, Hefei 230088, China

Received:2019-12-16 Revised:2020-11-29 Published:2021-03-28 Online:2021-03-28
Contact: wu hongyuan E-mail:yuanhongwu@axhu.edu.cn

摘要/Abstract

摘要： 大气 PM2:5 浓度是一种具有较强时序特征的数据, 故目前关于 PM2:5 浓度的预测多选择 RNN、 LSTM 等序列模型进行。但由于 RNN、 LSTM 等模型对不同时刻输入的数据都采用相同的权重进行计算, 不符合类脑设计, 造成 PM2:5 浓度预报准确率较低。针对以上问题, 提出一种基于 Adam 注意力机制的 PM2:5 预测方法 (AT-RNN 和AT-LSTM), 该方法首先通过 Adam 算法寻找 RNN 或 LSTM 的最优参数并在 Encoder 阶段引入注意力机制, 将注意力权重分配给具有时间序列特征的输入, 再进行 Decoder 解析和预测。通过实验, 对比了 BP、 RNN、 LSTM 和AT-RNN、 AT-LSTM 预测合肥市 PM2:5 浓度的效果。结果表明, 基于 Adam 注意力模型的预测方法准确率优于其它方法, 证明该方法在污染物预测中的有效性。

关键词: PM2:5, 神经网络, Adam 注意力模型

Abstract: Atmospheric PM2:5 concentration is a kind of data with strong time series characteristics, so currently the prediction of PM2:5 concentration is mostly based on RNN, LSTM and other sequence models. However, RNN, LSTM and the other similar models use the same weight to calculate the input data at different times, which is not in line with the brain-like design, resulting in the low accuracy of PM2:5 concentration prediction. In view of the above problems, a PM2:5 prediction method (AT-RNN and AT-LSTM) based on Adam attention mechanism is proposed. This method firstly looks for the optimal parameters of RNN or LSTM through Adam algorithm, and introduces attention mechanism in Encoder stage to assign attention weight to input with time series characteristics, and then carries out Decoder analysis and prediction. Through the experiment, the prediction effects of BP, RNN, LSTM and AT-RNN and AT-LSTM on PM2:5 concentration in Hefei city were compared. The results show that the prediction method based on Adam attention model is more accurate than other methods, which proves the effectiveness of this method in pollutant prediction.

Key words: PM2:5, neural networks, Adam attention model

中图分类号:

X511

张怡文, 袁宏武∗, 孙鑫, 吴海龙, 董云春. 基于Adam 注意力机制的PM2:5浓度预测方法[J]. 大气与环境光学学报, 2021, 16(2): 117-126.

ZHANG Yiwen, YUAN Hongwu∗, SUN Xin, WU Hailong, DONG Yunchun. PM2:5 Concentration Prediction Method Based on Adam′s Attention Model[J]. Journal of Atmospheric and Environmental Optics, 2021, 16(2): 117-126.

参考文献 12

[1]	Yang Mian, Wang Yin. Spatial-temporal characteristics of PM2:5 and its influencing factors in the Yangtze River Economic
	Belt [J]. China Population·Resources and Environment, 2017, 27(1): 91-100.
	杨冕, 王银. 长江经济带 PM2:5 时空特征及影响因素研究 [J]. 中国人口·资源与环境, 2017, 27(1): 91-100.
[2]	Jiang Yuncong, Yang Yuanjian, Wang Hong, et al. Urban-rural differences in PM2:5 concentrations in the representative cities
	of China during 2015∼2018 [J]. China Environmental Science, 2019, 39(11): 4552-4560.
	姜蕴聪, 杨元建, 王泓, 等. 2015∼2018 年中国代表性城市 PM2:5 浓度的城乡差异 [J]. 中国环境科学, 2019, 39(11):
45	52-4560.
[3]	Liu Shengdong, Shi junnan, Cheng Yong, et al. Review of pollution characteristics of PM2:5 in Chinese representative megacities [J]. Environmental Science Research, 2020, 33(2): 243-251.
	刘晟东, 史君楠, 程勇, 等. 中国典型城市群 PM2:5 污染特征研究进展 [J]. 环境科学研究, 2020, 33(2): 243-251.
[4]	Yao Xuefeng, Ge Baozhu, Zheng Haitao, et al. Spatiotemporal distribution characteristics of PM2:5 concentration and its main
	control factors in China based on multivariate data analysis [J]. Climatic and Environmental Research, 2018, 23(5): 596-606.
	姚雪峰, 葛宝珠, 郑海涛, 等. 基于多元数据分析的我国 PM2:5 浓度及其主控因子的时空分布特征研究 [J]. 气候与环境
	研究, 2018, 23(5): 596-606.
[5]	Wei Chunxuan, Huang He, Zhai Zhenfang, et al. Analysis of meteorological elements and forecast method of haze day in
	Hefei, China [J]. Journal of Atmospheric and Environmental Optics, 2019, 14(6): 419-430.
	魏春璇, 黄鹤, 翟振芳, 等. 合肥市霾日气象要素特征分析及预报方法研究 [J]. 大气与环境光学学报, 2019, 14(6):
41	9-430.
[6]	Luo Hongyuan, Wang Deyun, Liu Yanling, et al. PM2:5 concentration prediction based on two-layer decomposition technique
	and improved extreme learning machine [J]. System Engineering Theory and Practice, 2018, 38(5): 1321-1330.
	罗宏远, 王德运, 刘艳玲, 等. 基于二层分解技术和改进极限学习机模型的 PM2:5 浓度预测研究 [J]. 系统工程理论与实
	践, 2018, 38(5): 1321-1330.
[7]	Qiao Junfei, Cai Jie, Han Honggui. Study on prediction of PM2:5 based on T-S fuzzy neural network [J]. English Translation
	of Control Engineering, 2018, 25(3): 391-395.
	乔俊飞, 蔡杰, 韩红桂. 基于 T-S 模糊神经网络的 PM2:5 预测研究 [J]. 控制工程, 2018, 25(3): 391-395.
[8]	Wu Chunlin, Li Qi, Hou Junxiong, et al. PM2:5 concentration prediction using convolutional neural networks [J]. Mapping
	Science, 2018, 43(8): 72-79.
	吴春霖, 李琦, 侯俊雄, 等. 卷积神经网络的 PM2:5 预报模型 [J]. 测绘科学, 2018, 43(8): 72-79.
[9]	Li Xiaoli, Mei Jianxiang, Zhang Shan. PM2:5 concentration prediction using BP Adaboost neural network based on modified
	particle swarm optimization [J]. Journal of the Dalian University of Technology, 2018, 58(3): 99-106.
	李晓理, 梅建想, 张山. 基于改进粒子群优化 BP Adaboost 神经网络的 PM2:5 浓度预测 [J]. 大连理工大学学报, 2018,
58	(3): 99-106.
[10]	Zhou Shanshan, Li Wenjing, Qiao Junfei. Prediction of PM2:5 concentration based on self-organizing recurrent fuzzy neural
	network [J]. Journal of Intelligent Systems, 2018, 13(4): 21-28.
	周杉杉, 李文静, 乔俊飞. 基于自组织递归模糊神经网络的 PM2:5 浓度预测 [J]. 智能系统学报, 2018, 13(4): 21-28.
[11]	Cui Xianghui, Xie Jianfeng, Zhang Feng, et al. Establishment of PM2:5 prediction model based on deep learning [J]. Beijing
	Surveying and Mapping, 2017, (6): 22-27.
	崔相辉, 谢剑锋, 张丰, 等. 基于深度学习的 PM2:5 预测模型建立 [J]. 北京测绘, 2017, (6): 22-27.
[12]	Fan Junxiang, Li Qi, Zhu Yajie, et al. Aspatio-temporal prediction framework for air pollution based on deep RNN [J].
	Surveying and Mapping Science, 2017, 42(7): 76-83, 120.
	范竣翔, 李琦, 朱亚杰, 等. 基于 RNN 的空气污染时空预报模型研究 [J]. 测绘科学, 2017, 42(7): 76-83, 120.
[13]	Hu Xinchen. Research on Semantic Relationship Classification Based on LSTM [D]. Harbin: Harbin University of Technology,
	2015.
	胡新辰. 基于 LSTM 的语义关系分类研究 [D]. 哈尔滨: 哈尔滨工业大学, 2015.
[14]	Xie Yi, Rao Wenbi, Duan Pengfei, et al. A Chinese POS tagging approach using CNN and LSTM-based hybrid model [J].
	Journal of Wuhan University (Science Edition), 2017, 63(3): 246-250.
	谢逸, 饶文碧, 段鹏飞, 等. 基于 CNN 和 LSTM 混合模型的中文词性标注 [J]. 武汉大学学报 (理学版), 2017, 63(3):
24	6-250.
[15]	Zhao Xiaoqiang, Song Zhaoyang. Adam optimized CNN super-resolution reconstruction [J]. Computer Science and Exploration, 2019, 13(5): 858-865.
	赵小强, 宋昭漾. Adam 优化的 CNN 超分辨率重建 [J]. 计算机科学与探索, 2019, 13(5): 858-865.
[16]	Huang Liwei, Jiang Bitao, Lv Shouye, et al. Survey on deep learning based recommender systems [J]. Journal of Computer
	Science, 2018, 41(7): 1619-1647.
	黄立威, 江碧涛, 吕守业, 等. 基于深度学习的推荐系统研究综述 [J]. 计算机学报, 2018, 41(7): 1619-1647.
[17]	Yang Bo, Su Xiaohong, Wang Yadong. A hybrid algorithm based on attention model [J]. Journal of Software, 2005, (6):
10	73-1080.
	杨博, 苏小红, 王亚东. 基于注意力模型的混合学习算法 [J]. 软件学报, 2005, (6): 1073-1080.
[18]	Yang Guanci, Yang Jing, Li Shaobo, et al. Modified CNN algorithm based on Dropout and ADAM optimizer [J]. Journal of
	Huazhong University of Science and Technology (Natural Science Edition), 2018, 46(7): 122-127.
	杨观赐, 杨静, 李少波, 等. 基于 Dopout 与 ADAM 优化器的改进 CNN 算法 [J]. 华中科技大学学报 (自然科学版), 2018,
46	(7): 122-127.
[19]	Chang Zihan. Electricity Price Prediction Based on Hybrid Model of Adam optimized LSTM Neural Network and Wavelet
	Transform [D]. Lanzhou: Lanzhou University, 2019.
	常子汉. 基于小波变换与 Adam 优化的 LSTM 电价预测研究 [D]. 兰州: 兰州大学, 2019.
[20]	Breathe freely[EB/OL]. (2008-05-16) [2018-12-31]. http:// www.pm25.com.
	绿色呼吸[EB/OL]. (2008-05-16) [2018-12-31]. http:// www.pm25.com/.
[21]	Wu yunyun, Zhu Jianjun, Zuo Tingying. Investigation on the square root of the linear combination of the squre of RMSE and
	smoothness for the Vondrak filter′s evaluation [J]. Journal of Wuhan University (Information Science Edition), 2012, 37(10):
12	12-1214.
	吴芸芸, 朱建军, 左廷英. RMSE 的平方与平滑度的线性组合的平方根作为 Vondrak 滤波评价标准的探讨 [J]. 武汉大学
	学报 (信息科学版), 2012, 37(10): 1212-1214.