基于遗传算法和正则化极限学习机的PM2.5浓度预测研究
PM2.5 Prediction Based on Genetic Algorithm and Regularized Extreme Learning Machine
DOI: 10.12677/CSA.2018.88132, PDF,  被引量    国家自然科学基金支持
作者: 翁福添, 张天乐, 侯木舟*:中南大学,数学与统计学院,湖南 长沙;罗建书:国防科技大学,理学院,湖南 长沙
关键词: 遗传算法正则化极限学习机PM2.5浓度预测Genetic Algorithm Regularized ELM PM2.5 Concentration Prediction
摘要: 环境质量与人们的健康息息相关,一直是研究的热点。本文选取长沙市2017年NO2、PM10等大气数据对PM2.5日均值进行预测,采用BIC准则进行特征选择。在传统的超限学习机(ELM)的基础上,引入正则化项以控制模型的复杂度,并用遗传算法(GA)对模型的输入层权重矩阵和隐含层阈值矩阵进行优化,建立遗传算法和正则化极限学习机(GA-RE-ELM)的PM2.5预测模型。实验表明,该模型相比BP神经网络、超限学习机有更好的精度,均方误差分别降低了35.09%、25.49%,平均绝对误差分别降低了40.86%、30.80%,平均绝对百分误差分别降低了45.49%、31.65%,为PM2.5浓度的预测提供一种新的方法。
Abstract: Environmental quality is closely related to people’s health and has always been a research hotspot. In this paper, the daily average values of PM2.5 are predicted by atmospheric data such as NO2 and PM10 in Changsha City in 2017, and the BIC criterion is used for feature selection. On the basis of the traditional over-limit learning machine (ELM), the regularization term is introduced to control the complexity of the model, and the input layer weight matrix and the hidden layer threshold matrix of the model are optimized by genetic algorithm (GA) to establish the genetic algorithm. Then the PM2.5 prediction model of the regularized limit learning machine (GA-RE-ELM) is built, the experiment shows that the model achieves more state of the art performance than the BP neural network and the over-limit learning machine, the mean square error is reduced by 35.09% and 25.49%, the average absolute error is reduced by 40.86% and 30.80%, and the average absolute percentage error is reduced by 45.49% and 31.65%. Meanwhile, it provides a new method for predicting PM2.5 concentration.
文章引用:翁福添, 张天乐, 侯木舟, 罗建书. 基于遗传算法和正则化极限学习机的PM2.5浓度预测研究[J]. 计算机科学与应用, 2018, 8(8): 1207-1216. https://doi.org/10.12677/CSA.2018.88132

参考文献

[1] 谢心庆, 郑薇. 国内外PM2.5研究进展综述[J]. 电力科技与环保, 2015, 31(4): 17-20.
[2] 喻其炳, 李勇, 白云, 等. 基于聚类分析与偏最小二乘法的支持向量机PM_(2.5)预测[J]. 环境科学与技术, 2017(6): 157-164.
[3] 苏航, 银燕, 朱彬, 等. 中国环渤海地区SO2和NO2干沉降数值模拟及影响因子分析[J]. 中国环境科学, 2012, 32(11): 1921-1932.
[4] Sahanavin, N., Prueksasit, T., Tantrakarnapa, K. 使用路径分析和线性回归确定的高交通区域PM10和PM2.5之间的关系[J]. 环境科学杂志, 2017.
[5] Tong, L., Lau, A.K.H., Kai, S., et al. 基于区域数值模拟的香港空气质量时间序列预测[J]. 地球物理研究大气学报, 2018, 123(3).
[6] 朱亚杰, 李琦, 侯俊雄, 等. 运用贝叶斯方法的PM_(2.5)浓度时空建模与预测[J]. 测绘科学, 2016, 41(2): 44-48.
[7] 李勇, 白云, 李川. 基于小波分析与 BP神经网络的 PM10浓度预测模型[J]. 环境监测管理与技术, 2016, 28(5): 24-28.
[8] 王鹏,张秦, 等. 基于ARIMA和SVM的PM 2.5混合Garch模型,浓度预测[J]. 大气污染研究, 2017, 8(5).
[9] 陈绍炜, 柳光峰, 冶帅, 等. 基于蝙蝠算法优化ELM的模拟电路故障诊断研究[J]. 电子测量技术, 2015, 38(2): 138-141.
[10] 易少强, 何世平, 王杰. 粒子群算法在约束型垫高阻尼结构动力学优化中的应用[J]. 中国舰船研究, 2018(1): 31-37.
[11] 张卫辉, 黄南天, 杨金成, 等. 基于广义S变换和DE-ELM的电能质量扰动信号分类[J]. 电测与仪表, 2016, 53(20): 50-55.
[12] 梅益, 孙全龙, 喻丽华, 等. 基于GA-ELM的铝合金压铸件晶粒尺寸预测[J]. 金属学报, 2017, 53(9): 1125-1132.
[13] Pineda, F.J. (1987) Generalization of Back-Propagation to Recurrent Neural Networks. Physical Review Letters, 59, 2229-2232. [Google Scholar] [CrossRef
[14] 邢高生. 基于系统工程文档的领域知识库构建[D]: [硕士学位论文]. 北京: 北京交通大学, 2017.
[15] Hecht-Nielsen, R. (1989) Theory of the Backpropagation Neural Network. In: Neural Networks for Per-ception, Vol. 2, Harcourt Brace & Co., 593-605.
[16] Huang, G.B., Zhu, Q.Y. and Siew, C.K. (2006) Extreme Learning Machine: Theory and Applications. Neurocomputing, 70, 489-501. [Google Scholar] [CrossRef
[17] Huang, G.B.,Zhu, Q.Y. and Siew, C.K. (2006) Real-Time Learning Capability of Neural Networks. IEEE Transactions on Neural Networks, 17, 863-878. [Google Scholar] [CrossRef
[18] Huang, G.B., Ding, X. and Zhou, H. (2010) Optimization Method Based Ex-treme Learning Machine for Classification. Elsevier Science Publishers B.V., Amsterdam.
[19] 郑鹏, 张琳娜, 赵凤霞. 形状误差评定的拟合操作研究[J]. 计量与测试技术, 2008, 35(12): 1-3.
[20] 王新全, 孙培廷, 邹永久, 等. 基于 GA-BP模型的船舶柴油机排气温度趋势预测[J]. 大连海事大学学报, 2015, 41(3): 73-76.
[21] Burnham, K.P. (2004) Understanding AIC and BIC in Model Selection. Sociological Methods & Research, 33.