The invention relates to a deep neural network robust traffic prediction method based on multi-modal spatio-temporal data, and the method comprises the steps: obtaining the feature data of a traffic flow state at a certain moment, building a feature data set according to a function, taking historical time series data as the input, capturing an urban road network topological structure through a graph convolution network, and obtaining spatial features; secondly, inputting a gating recursion unit model, obtaining dynamic changes through information transmission between units, and capturing timefeatures; and obtaining a T-GCN output prediction result in a full connection layer mode, and performing Kalman filtering processing on the obtained output data set to obtain a final output predictionresult. Data preprocessing and feature learning are carried out on input vectors of traffic flow data in sequence by adopting a time graph convolutional network model, then, an output result is corrected by Kalman filtering to explore the intrinsic law in the traffic flow data, and finally, the model is sampled to obtain a high-level feature vector to predict the state of the traffic flow.
本发明涉及一种基于多模态时空数据的深度神经网络鲁棒交通预测方法,获取某时刻下交通流状态的特征数据,根据函数构建特征数据集,将历史时间序列数据作为输入,利用图卷积网络捕获城市路网拓扑结构,获得空间特征;然后输入门控递归单元模型,通过单元间的信息传递获得动态变化,捕捉时间特征;以全连接层的方式得到T‑GCN输出预测结果,得到的输出数据集经过卡尔曼滤波处理得到最终的输出预测结果,本发明采用时间图卷积网络模型对交通流数据的输入向量先后进行数据预处理和特征学习,并对输出结果用卡尔曼滤波进行矫正,从中挖掘交通流数据中的本质性规律,最后通过对该模型进行抽样以获取高层次的特征向量来预测交通流的状态。
Deep neural network robust traffic prediction method based on multi-modal spatio-temporal data
基于多模态时空数据的深度神经网络鲁棒交通预测方法
2021-01-29
Patent
Electronic Resource
Chinese
European Patent Office | 2020
|Spatio-Temporal Graph Neural Network for Traffic Prediction Based on Adaptive Neighborhood Selection
Transportation Research Record | 2023
|