基于BP 人工神经网络的大气颗粒物PM10 质量浓度预测.docVIP

第 43 卷第 5 期 2012 年 5 月  中南大学学报(自然科学版) Journal of Central South University (Science and Technology)  Vol.43 No.5 May 2012  基于 BP 人工神经网络的大气颗粒物 PM10 质量浓度预测 石灵芝，邓启红，路婵，刘蔚巍 (中南大学能源科学与工程学院，湖南长沙，410083) 摘要：根据 2008 年长沙市火车站监测点全年大气 PM10 及气象参数的小时平均数据，建立 BP 人工神经网络预测 模型，预测 PM10 小时平均浓度。为证明人工神经网络模型用于预测 PM10 质量浓度的准确性，研究中考虑 2 种预 测模型：多元线性回归模型与人工神经网络模型。研究结果表明：与传统的多元线性回归模型相比，人工神经网 络模型能够捕捉污染物浓度与气象因素间的非线性影响规律，能更好地预测 PM10 质量浓度，拟合优度 R2 有较大 提高；所选取气象参数及污染源强变量能较准确地描述大气 PM10 质量浓度的实时变化，用于 PM10 质量浓度的预 测准确度较高，整体 R2 可达 0.62；人工神经网络预测模型不仅适用于一般污染浓度情况，对于高污染时期 PM10 质量浓度的预测也较为准确。 关键词：BP 人工神经网络；PM10；预测；多元线性回归；高污染 中图分类号：X831 文献标志码：A 文章编号：1672?7207(2012)05?1969?06  Prediction of PM10 mass concentrations based on BP artificial neural network SHI Ling-zhi, DENG Qi-hong, LU Chan, LIU Wei-wei (School of Energy Science and Engineering, Central South University, Changsha 410083, China) Abstract: The back-propagation (BP) artificial neural network model for prediction of PM10 mass concentrations was developed using atmospheric PM10 mass concentration and meteorological data in 2008, which was monitored in Changsha railway-station. In order to show the accuracy of PM10 mass concentration prediction based on artificial neural network, two models were developed: multiple linear regression model and artificial neural network model. The results show that the BP artificial neural network model can be trained to model the highly non-linear relationships between PM10 mass concentration and meteorological parameters, and to provide better results than the traditional multiple linear regression models with much higher goodness of fit (R2 ). The meteorological parameters and emission source variation variables can accurately describe PM10 variation, and thus provide satisfactory prediction results, with R2 of 0.62. In addition, the developed BP artificial neural network model for prediction of PM10 mass concentrations also w