%0 Journal Article
%T 基于Bi-LSTM模型的短时交通量多步预测研究<br>Research on Multi-Step Forecast of Short-Term Traffic Based on Bi-LSTM Model
%A 向鸿锐
%A 阮泽宇
%A 曹微
%A 肖亚琪
%A 寇笑天
%A 李永翔
%J Open Journal of Transportation Technologies
%P 141-152
%@ 2326-344X
%D 2024
%I Hans Publishing
%R 10.12677/ojtt.2024.133018
%X 多数短时交通量的预测研究仅集中在单步预测上，并且预测时长不足，为使得交通管理与控制等措施发挥更好的效果，提高短时交通量多步预测精度，最大限度提高交通管理决策和出行决策合理性，本文提出一种基于Bi-LSTM模型进行短时交通量多步预测。首先利用Bi-LSTM模型进行单步预测，将得到的预测值与原始值替换，通过递归迭代进行五步的多步长预测。根据本文研究结果表明：Bi-LSTM模型在多步预测中具有一定优势，相比于ARIMA模型和BP神经网络模型，其平均多步RMSE、MAE、MAPE、RMSRE分别降低了11.1085和9.4134、9.7884和7.2474、26.52%和14.91%、25.01%和14.95%。最后得出，Bi-LSTM在交通量多步预测上具有较大优势。<br />
Most short-term traffic volume forecasting studies only focus on single-step forecasting, and the forecasting time is insufficient. In order to make traffic management and control measures play a better role, improve the accuracy of multi-step forecasting of short-term traffic volume, and maximize the rationality of traffic management decision-making and travel decision-making, this paper proposes a multi-step forecasting of short-term traffic volume based on Bi-LSTM model. Firstly, Bi-LSTM model is used for single step prediction, the predicted value is replaced with the original value, and the five-step multi-step prediction is carried out by recursive iteration. According to the results of this study, Bi-LSTM model has certain advantages in multi-step prediction. Compared with ARIMA model and BP neural network model, The average multi-step RMSE, MAE, MAPE and RMSRE decreased by 11.1085 and 9.4134, 9.7884 and 7.2474, 26.52% and 14.91%, 25.01% and 14.95%, respectively. Finally, it is concluded that Bi-LSTM has a great advantage in multi-step traffic volume prediction.
%K 短时交通量多步预测，双向长短时记忆网络，Bi-LSTM<br>Multi-Step Prediction of Short Time Traffic Volume
%K Bidirectional Long Time Memory Network
%K Bi-LSTM
%U http://www.hanspub.org/journal/PaperInformation.aspx?PaperID=87970