一种EMD和DE-BPNN组合优化的短时交通流预测方法  

A Short-Term Traffic Flow Forecasting Method Based on EMD and DE-BPNN Combined Optimization

作　　者：吴玲玲[1] 尹莉莉 任其亮[1] WU Lingling;YIN Lili;REN Qiliang(School of Traffic&Transportation,Chongqing Jiaotong University,Chongqing 400074,China)

机构地区：[1]重庆交通大学交通运输学院,重庆400074

出　　处：《重庆理工大学学报（自然科学）》2021年第12期155-163,共9页Journal of Chongqing University of Technology：Natural Science

基　　金：国家重点研发计划项目(2018YFB1601001);国家社会科学基金项目(16XJY013)。

摘　　要：针对短时交通流非线性的特点以及BP神经网络(BPNN)在进行短时交通流预测时易陷入局部极小值的缺点,提出一种基于经验模态分解(EMD)和差分进化算法优化BP神经网络(DE-BPNN)的短时交通流预测方法。利用EMD算法将交通时序数据中不同模态的分量逐级分解出来,生成一系列不同尺度的本征模态函数(IMF)和残余量,去除一定噪声影响;借助DE-BPNN算法进行短时交通流预测,并采用美国加利福尼亚州高速公路交通流数据,对该方法进行验证和预测精度测试。实验结果表明:采用EMD分解后的交通流预测结果更为精确,相比其他预测方法,其预测结果的MAE值分别提升了50.07%、49.36%、18.68%;MSE值分别提升了52.46%、47.84%、12.37%;MAPE值分别提升了52.11%、51.08%、35.09%;MSPE值分别提升了56.36%、52.59%、43.53%。Aiming at the non-linear characteristics of short-term traffic flow and the shortcomings of BP neural network(BPNN)that tend to fall into local minima when forecasting short-term traffic flow,a short-term traffic flow prediction method based on empirical mode decomposition(EMD)and differential evolution algorithm optimized BP neural network(DE-BPNN)is proposed.First,the EMD algorithm is used to decompose the components of different modes in the traffic time series data step by step,and a series of eigenmode functions(IMF)and residuals of different scales are generated to remove certain noise effects.Secondly,the DE-BPNN algorithm is used to carry out short-term traffic flow prediction,and finally California highway traffic flow data is used to verify the method and test the prediction accuracy.The experimental results show that the traffic flow prediction results after the EMD decomposition are more accurate.Compared with other prediction methods,the MAE of the prediction results is increased by 50.07%,49.36%,and 18.68%;the MSE is increased by 52.46%、47.84%and 12.37%respectively;MAPE increased by 52.11%、51.08%、35.09%;MSPE increased by 56.36%、52.59%、43.53%.

关 键 词：短时交通流预测 经验模态分解算法 BP神经网络 差分进化算法 

分 类 号：U491.14[交通运输工程—交通运输规划与管理]