基于车流量数据的SARIMA和LSTM组合模型的比较研究

Comparative study of SARIMA and LSTM combined models based on a traffic flow data

针对同时具有周期性、长记忆性等多种特征的车流量数据,单一地SARIMA或LSTM模型往往拟合效果不理想,而其组合模型可以弥补单一模型的不足。结合线性和非线性预测方法,文中分别建立了三个SARIMA-LSTM组合模型,随后,对车流量数据进行了预测分析,通过与SARIMA、LSTM两种单模型拟合效果的比较分析表明:1)对含周期性和长记忆性的数据,组合模型的预测效果更优;2)基于MA滤波方法的组合模型三比其他两种方法在提升模型预测精度上表现更好。

For the traffic flow data with various characteristics such as periodicity and long memory,the fitting effect of single SARIMA or LSTM model is often not ideal,but the combined model can make up for the shortcomings of single model.Combined with linear and nonlinear prediction methods,we proposed to use three SARIMA-LSTM combined models to predict and analysis the traffic flow data.Comparing with the single SARIMA and LSTM models showed that:1)For the data containing periodicity and long memory,the prediction effect of the combined model was better;2)The combined model 3 based on MA filtering method performs better than the other two methods in improving the prediction accuracy.