Due to the fact that existing traffic flow forecasting methods cannot completely reflect the real time traffic situation of the road network, a new method of short-term traffic flow prediction is proposed based on deep learning in this study. Firstly, in order to improve the efficiency of data processing, a method of road network data compression is proposed based on correlation analysis and CX decomposition. Secondly, the traffic flow data are divided into trend term and random fluctuation term by spectral decomposition method, and the influence of trend term on prediction accuracy is removed. Finally, by combining a deep belief network model and a kernel extreme learning machine classifier as the prediction model, the essential characteristics of the traffic flow data are extracted by using DBN at the bottom of the network, and the extracted results are input into the kernel extreme learning machine to predict the traffic flow. The actual regional road network traffic flow data are tested to verify the effectiveness of the proposed short-time network traffic flow forecasting method. The results show that the proposed method can not only save 90% of the running time but also the average prediction accuracy of each road section can reach 92%.

中文翻译：

---

基于深度学习的路网短期交通流量预测

针对现有交通流量预测方法不能完全反映路网实时交通状况的事实，提出一种基于深度学习的短期交通流量预测新方法。首先，为提高数据处理效率，提出了一种基于相关分析和CX分解的路网数据压缩方法。其次，利用谱分解法将交通流量数据分为趋势项和随机波动项，消除了趋势项对预测精度的影响。最后，通过结合深度信念网络模型和内核极限学习机分类器作为预测模型，使用网络底部的DBN提取交通流数据的基本特征，然后将提取的结果输入到内核极限学习机中，以预测流量。测试了实际的区域道路网络交通流量数据，以验证所提出的短期网络交通流量预测方法的有效性。结果表明，该方法不仅可以节省90％的运行时间，而且每个路段的平均预测精度可以达到92％。