神经网络在复杂自相关预测过程中的应用及对比研究  

Study of Neural Network Application and Comparison in the Forecast Process of Complex Autocorrelation

作　　者：禹建丽[1] 黄鸿琦 YU Jian-li HUANG Hong-qi(Zhengzhou University of Aeronautical, School of Management Engineering, Zhengzhou 450046, Chin)

机构地区：[1]郑州航空工业管理学院管理工程学院,河南郑州450046

出　　处：《数学的实践与认识》2016年第19期212-220,共9页Mathematics in Practice and Theory

基　　金：河南省自然科学基金(142102210077);河南省自然科学基金(142102210105);郑州航空工业管理学院研究生教育创新计划基金(2016CX015)

摘　　要：由于PM_(2.5)日均浓度值受外界多重复杂因素的影响,其较强的自相关性使得时间序列模型ARIMA构建难以实现,因此,给出高映射能力的非线性神经网络预测模型,并分别建立基于BP神经网络和GRNN神经网络的预测模型,进行PM_(2.5)浓度预测实验.结果表明,BP神经网络回检过程和检测过程存在不稳定性,预测残差波动较大,而GRNN神经网络检测残差呈完全U型,回检过程和检测过程较稳定,并且GRNN神经网络回检数据拟合度、预测数据精度和运算速度均优于BP神经网络,建模过程更为方便,易于实际应用.Due to the complexity of the daily average of PM2.5 pollutants density from multiple factors,ARIMA（p,d,q） model was difficult to build.Therefore,the high nonlinear mapping ability neural network prediction model was proposed.Secondly,building PM2.5Concentration prediction model based on the BP neural network and GRNN neural network.By comparing the experimental process of two neural network,the results show that the retrieval process and detection process of BP neural network is not stability and predict residual is volatile,while the retrieval process and detection process of GRNN neural network is stable and present completely U,and its the degree of fitting of retrieval data,the prediction data accuracy and computing speed are better than BP neural network,and modeling process is more convenient,so the prediction mold based on GRNN neural network is easily to practical application.

关 键 词：GRNN神经网络 BP神经网络 PM_(2.5) 时间序列分析 预测 

分 类 号：X513[环境科学与工程—环境工程] TP183[自动化与计算机技术—控制理论与控制工程]