EEMD+BiGRU组合模型在短时交通流量预测中的应用

针对城市交通流随机波动性强、数据中含噪声多导致预测精度下降的问题,提出一种基于集合经验模态分解(ensemble empirical mode decomposition,EEMD)和双向门控循环单元(bidirectional gated recurrent unit,BiGRU)的组合交通流量预测模型,有效地提升了短时交通流预测的精度。模型利用EEMD算法对原始数据进行分解,根据分解所得的本征模函数(intrinsic mode function,IMF)分量绘制噪声能量图谱,去除分量中的噪声,并将去噪后的IMF分量作为BiGRU网络的输入进行训练,再将训练所得的结果进行重构加和,得到最终的预测结果。实验结果表明,未舍弃含有噪声的IMF分量进行重构的预测结果,相比于参考文献中提出的EMD+LSTM模型、LSTM模型和EEMD+LSTM模型,其平均绝对百分误差分别优化了42.36%、61.82%和30.95%；舍弃含有噪声的IMF分量后进行重构的预测结果,其平均绝对百分误差相比于将全部IMF分量进行重构优化了56.62%。

Application of EEMD+BiGRU combination model in short-term traffic flow prediction

Aiming at the problems of high random fluctuation of urban traffic flow and high noise in data, which leads to the decline of prediction accuracy, a combined traffic flow prediction model based on EEMD (ensemble empirical mode decomposition) and BiGRU (bidirectional gated recurrent unit) was proposed, which can effectively improve the accuracy of short-term traffic flow prediction. EEMD algorithm was used to decompose the original data, and the noise energy map was drawn according to the IMF(intrinsic mode function) component to remove the noise in the component. The denoised IMF components were trained as the input of BiGRU network. And the results of training were reconstructed and added to obtain the final prediction result. The experimental results show that,compared with the EMD+LSTM model, LSTM model and EEMD+LSTM model proposed in references, the mean absolute percentage errors are improved by 42.36%, 61.82% and 30.95% when the IMF components containing noise are not abandoned during reconstruction; after abandoning the IMF component containing noise, the mean absolute percentage error is improved by 56.62% compared with the reconstruction of all IMF components.