基于SARIMA-GA-Elman组合模型的短时交通流预测方法

针对单一模型无法深入挖掘交通流复杂的线性和非线性特征方面的局限性以及神经网络模型在训练时收敛速度缓慢等问题,提出了一种基于SARIMA-GA-Elman的组合预测模型.该组合模型有效地融合了季节性差分自回归滑动平均(seasonal autoregressive integrated moving average,SARIMA)模型良好的线性拟合能力和Elman递归神经网络强大的非线性映射能力;在预测过程中首先基于SARIMA滚动预测时间序列的线性分量,然后使用SARIMA模型的预测误差序列建立Elman-RNN构建非线性误差模型;此外在训练非线性误差模型的过程中使用经过二进制编码的遗传算法(genetic algorithm,GA)优化Elman-RNN,旨在提升Elman-RNN的训练效率,最后把两个模型的预测结果加权组合得到最终的预测值.实验结果表明,该组合模型在预测精度和鲁棒性方面相比单一模型都有较为明显的提升.

Research on short-term traffic flow forecasting method based on SARIMA-GA-Elman combined model

Aiming at the limitations of a single model that can not dig into the complex linear and nonlinear characteristics of traffic flow and the slow convergence of neural network models during training, a combined prediction model based on SARIMA-GA-Elman is proposed. The combined model effectively combines the excellent linear fitting ability of the seasonal autoregressive integrated moving average (SARIMA) model and the powerful nonlinear mapping ability of the Elman recurrent neural network. In the prediction process, the linear component of the time series is firstly predicted based on the SARIMA rolling, and then the prediction error sequence of the SARIMA model is used to establish the Elman-RNN to construct the nonlinear error model. In addition, in the process of training the nonlinear error model, the binary-coded genetic algorithm (GA) is used to optimize the Elman-RNN in order to improve the training efficiency of Elman-RNN. Finally the prediction results of the two models are weighted and combined to obtain the final prediction value. Experimental results show that the combined model has a significant improvement in prediction accuracy and robustness compared to a single model.