基于改进人工蜂群BP神经网络的PM2.5浓度预测模型  

PM2.5 concentration prediction model based on BP neural network of improved artificial bee colony

作　　者：胡俊 杨辉军[1] 程晨[1] HU Jun;YANG Huijun;CHENG Chen(School of Information Engineering,Anhui Vocational College of International Business,Hefei 231131,China)

机构地区：[1]安徽国际商务职业学院信息工程学院,安徽合肥231131

出　　处：《山东交通学院学报》2020年第4期15-22,共8页Journal of Shandong Jiaotong University

基　　金：2019安徽省教育厅自然科学研究一般项目(2019XJZK01);安徽省教育厅省级质量工程项目(2018jyssf084);安徽省高等学校自然科学研究项目(2018XJZK02)。

摘　　要：为解决传统细颗粒物质(particulate matter,PM)浓度预测模型研究角度片面、非线性程度较高、预测精度不高的问题,建立基于改进人工蜂群反向传播(back propagation,BP)神经网络的PM2.5质量浓度预测模型,将搜索形式与跟随蜂选择概率设为改进角度,优化人工蜂群算法的寻优精度与收敛速率,在BP神经网络模型中引入改进人工蜂群算法,更新网络权重,避免使其陷入局部最小化;依据PM2.5浓度多种影响因素之间的关联性,采用灰色关联分析策略,识别所有因素间的发展趋势依赖程度,选取具有较大关联系数的污染气体,设定其质量浓度、温度及相对湿度为预测模型的变量因子,利用三倍标准差方法舍弃异常数据,根据三位二进制编码,标签化样本数据,通过创建的预测模型,获取PM2.5质量浓度预测结果。仿真分析表明:基于改进人工蜂群BP神经网络的PM2.5质量浓度预测模型的稳定性得到大幅提升,预测精准性具有明显优势。The research angle of the traditional particulate matter(PM2.5)concentration prediction model is too one-sided,resulting in a high degree of nonlinearity,which affects the accuracy of prediction.A PM2.5 concentration prediction model is established based on the back propagation(BP)neural network of the improved artificial bee colony.The search form and the bee selection probability are set as an improved angle,the optimization accuracy and convergence rate of the artificial bee colony algorithm are optimized,an improved artificial bee colony algorithm is introduced in the BP neural network model,and the network weight is updated to avoid the local minimization.According to the correlation in various influencing factors of PM2.5 concentration,a gray correlation analysis strategy is used to identify the degree of dependence of the development trend among all factors.The polluted gas with a larger correlation coefficient is selected to set its concentration,temperature and relative humidity as the variable factors of the prediction model,the triple standard deviation method is used to discard the abnormal data,and then the result of the PM2.5 concentration prediction is obtained through the established prediction model according to the three-bit binary coding and the labeling sample data.The simulation experiments show that the stability of the PM2.5 concentration prediction model based on the BP neural network of the improved artificial bee colony has also been greatly improved,and the prediction accuracy has obvious advantages.

关 键 词：人工蜂群 BP神经网络 PM2.5浓度 变量因子 

分 类 号：X831[环境科学与工程—环境工程]