基于多维度LSTM模型的短时交通流预测

Short-term traffic flow prediction based on multi-dimensional LSTM model

提出一种基于长短时记忆神经网络(Long Short-Term Memory,LSTM)的交通流预测模型,不同于单一因素预测,模型深入探究时间占有率等因素对预测结果的影响,从而进行多维度的短时交通流预测。最后以长沙市某实地数据对模型预测结果的精确性进行检验。研究结果表明:在以10 min为间隔预测中,与时间占有率组合的多维度因素速度预测和流量预测的平均绝对误差相较单一因素分别由4.6 km/h降至2.78 km/h,9.65辆降至5.8辆。加入时间占有率等其他因素后,模型预测的精度显著提高。

Different from single-factor prediction,a traffic flow prediction model based on Long Short-Term Memory was developed to explore the influence of time occupancy and other factors on the prediction results,so as to conduct multi-dimensional short-term traffic flow prediction.A real-world traffic data in Changsha was used to test the accuracy of the model.The results show that in the 10-minute interval prediction,compared with the single-dimensional model,the MAE of the speed in the multi-dimensional model can be substantially reduced from 4.6 km/h to 2.78 km/h,and that of volumes from 9.65 to 5.8.Therefore,taking other factors as time occupancy into consideration can significantly improve the accuracy of the traffic prediction models.