The invention discloses a traffic flow prediction method and system based on a multi-time-sequence convolutional gated graph neural network. The method mainly comprises the following steps: firstly, dividing traffic flow data into an adjacent sequence and a periodic sequence; secondly, designing a graph convolutional gated network (GGCN) to capture short-term spatial-temporal characteristics of adjacent sequences, and extracting long-term time characteristics of periodic sequences through a multi-time-sequence convolutional network (MTCN) to enhance the short-term spatial-temporal characteristics; in addition, an external feature extraction module based on one-hot coding is designed to improve the external environment perception capability of the model; and finally, designing a fusion output module based on an attention mechanism to obtain a predicted value. Verification is carried out through a traffic flow data set of a certain urban area, the prediction performance of the multi-time-sequence convolutional gated graph neural network (MTCGGN) is better than that of an existing graph convolutional network based on spatio-temporal feature mining, and compared with an attribute enhanced spatio-temporal graph convolutional network (AST-GCN), a time graph convolutional network (T-GCN) and a diffusion convolutional recurrent neural network (DCNN), the root mean square errors of the MTCGGN are reduced by 2.485%, 4.958% and 10.889% respectively.
本发明公开的一种基于多时序卷积门控图神经网络的交通流预测方法及系统,主要包括以下步骤:首先将交通流数据划分为邻近序列和周期序列;然后设计图卷积门控网络(GGCN)以捕获邻近序列的短期时空特征,并通过多时序卷积网络(MTCN)提取周期序列的长期时间特征来增强短期时空特征;此外,设计基于独热编码的外部特征提取模块以提高模型外部环境感知能力;最后,设计基于注意力机制的融合输出模块得到预测值。通过某市区的交通流数据集进行验证,多时序卷积门控图神经网络(MTCGGN)的预测性能优于现有基于时空特征挖掘的图卷积网络,相比属性增强时空图卷积网络(AST‑GCN)、时间图卷积网络(T‑GCN)和扩散卷积循环神经网络(DCRNN),MTCGGN的均方根误差分别降低2.485%、4.958%和10.889%。
Traffic flow prediction method and system based on multi-time-sequence convolutional gated graph neural network
基于多时序卷积门控图神经网络的交通流预测方法及系统
2023-12-15
Patent
Electronic Resource
Chinese
Traffic flow prediction model based on gated time convolutional network
European Patent Office | 2023
|Multi-scale traffic flow prediction method based on graph convolutional neural network
European Patent Office | 2021
|European Patent Office | 2023
|SHORT-TERM TRAFFIC FLOW PREDICTION METHOD BASED ON CAUSAL GATED-LOW-PASS GRAPH CONVOLUTIONAL NETWORK
European Patent Office | 2024
|