【摘要】 提出了一种语境特征注入的交通流速度预测模型，该方法通过循环神经网络和深度置信网络分别学习和提取交通流速度数据的时间序列特征和语境特征，使用向量融合机制将提取到的语境特征注入时间序列特征中，生成新的融合特征，用于交通流速度预测.实验结果表明，所提模型能够反映交通流模式的真实状况，准确捕捉到交通流速度的随机变化，具有较好的预测性能.在5 min和10 min等6个时间尺度上，模型的预测误差均优于其他对比模型.更多还原

【Abstract】 A traffic flow speed prediction model based on context feature injection is proposed. The time series features and context features of traffic flow speed data are learned and extracted through recurrent neural networks and deep belief networks,respectively. The vector fusion mechanism is used to inject the extracted context features into the time series features to generate new fusion features,which are used for traffic flow speed prediction. The experimental results show that the proposed model can reflect the real state of the traffic flow speed and accurately capture the stochastic changes of traffic speed,and has good predictive performance. On six time scales,such as 5 min and 10 min,the prediction error of the model is better than that of other comparison models.更多还原