考虑噪声影响的短时交通流预测模型及验证

Construction and validation of short-term traffic flow prediction model considering noise effect

为进一步提高交通流预测的精确性,相较于忽略噪声影响的传统预测方法,提出一种基于小波变换的双向长短时记忆神经网络-自回归滑动平均模型的预测模型(WBLA)。WBLA模型首先采用小波变换,将交通流数据分解为特征项及噪声项,在此基础上,对特征项采用双向长短时记忆神经网络(BiLSTM)进行预测,对噪声项采用自回归滑动平均模型(ARMA)进行预测,最后对两项预测结果求和作为最终的预测结果。将未考虑噪声影响的其它基准方法作为对比模型,在美国加州高速公路交通流数据集上进行测试及验证,实验结果表明:WBLA模型同未考虑噪声影响的次好模型相比,MAE、RMSE和MAPE分别下降17.86%、15.98%、16.39%,表明WBLA模型符合实际交通流速度变化趋势,模型合理性得到验证。

In order to further improve the accuracy of traffic flow prediction,compared with the traditional prediction methods which ignore the influence of noise,a bidirectional long short-term memory(BiLSTM)-autoregressive moving average model(ARMA)based on wavelet transform is proposed(WBLA).In WBLA model,the traffic flow data is decomposed into the trend term and the noise term by wavelet transform.BiLSTM is used to predict the trend term,and ARMA is used to predict the noise term.Finally,the sum of the two prediction results was used as the final prediction result.The test and validation were carried out on the highway traffic flow data set in California,USA,and other reference methods without considering noise effects were used as comparison models.The result shows that compared with the second-best model without considering noise,the WBLA model reduces MAE,RMSE and MAPE by 17.86%,15.98%and 16.39%,respectively,which indicates that the WBLA model is more consistent with the actual traffic flow speed change trend,and the rationality of the model is verified.