Traffic flow prediction method based on gated circulation network
By adopting a gated recurrent network method in traffic flow prediction, combining convolutional neural network and lightweight attention mechanism, the problem of insufficient traffic flow prediction accuracy and efficiency in the prior art is solved, and higher prediction accuracy and computing efficiency are achieved.
Patent Information
- Application Number
- CN202510104853.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-01-23
- Publication Date
- 2025-06-27
AI Technical Summary
Existing traffic flow prediction methods based on recurrent neural networks and hybrid convolutional networks are limited in accuracy, low efficiency, and difficult to effectively process sparse traffic data and nonlinear patterns when modeling long-term dependencies and complex spatial relationships.
The traffic flow prediction method based on the gated recurrent network is adopted, combined with the six-layer convolutional neural network to extract spatial features, enhance long-term dependence modeling capabilities through jump connections, and use lightweight attention mechanisms to improve model efficiency and prediction accuracy.
It significantly improves the flow prediction accuracy in complex traffic environments, optimizes the computing efficiency, especially in long-term prediction, with higher accuracy.
Smart Images
Figure CN120220384A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of traffic flow prediction, and in particular, to a traffic flow prediction method based on a gated recurrent network. Background Art
[0002] With the rapid development of urbanization, traffic flow prediction has become a key research direction in intelligent transportation systems. Existing methods based on recurrent neural networks (RNNs) and hybrid convolutional networks (CNN-LSTMs) face problems of limited accuracy and low efficiency when modeling long-term dependencies and complex spatial relationships. In addition, traditional models are insufficient in dealing with sparse traffic data, dynamic feature capture, and non-linear pattern modeling. To address these challenges, the present invention proposes a traffic flow prediction method based on a gated recurrent network, which combines a six-layer convolutional neural network to extract spatial features, enhances the long-term dependency modeling ability through skip connections, and uses a lightweight attention mechanism to improve model efficiency and prediction accuracy, significantly optimizing the traffic flow prediction performance in complex traffic environments. Summary of the Invention
[0003] In view of this, the object of the present invention is to provide a traffic flow prediction method based on a gated recurrent network, which captures dynamic associations in the traffic road network through a hybrid neural network and an attention-based representation learning module, and optimizes attention calculation through a sparsified self-attention mechanism, thereby achieving efficient prediction of traffic flow.
[0004] To solve the above technical problems, the technical solution of the present invention is: a traffic flow prediction method based on a gated recurrent network, characterized by comprising the following steps:
[0005] S1: Construct a traffic network sequence structure, and use a six-layer one-dimensional convolutional neural network to extract the spatial features of traffic data.
[0006] S2: Design a feature extraction module to optimize the calculation of attention weights through sparsification processing.
[0007] S3: Combine the traffic road network features with the time series structure to generate the input of traffic flow data.
[0008] S4: Design a sequence representation learning module to capture the spatio-temporal correlation of data.
[0009] S5: Deploy the model to perform traffic flow prediction.
[0010] As a preferred embodiment of the present invention, the network constructed in step S1 includes a six-layer one-dimensional convolutional neural network. A pooling layer is connected after each convolutional layer to extract higher-level features and reduce noise. Then, batch normalization technology is used to improve the training performance and stability of the model. Finally, the ReLU activation function and a non-linear activation function are introduced to enhance the method's understanding of the data.
[0011] As a preferred embodiment of the present invention, in the feature extraction module in step S2, padding technology is used to avoid feature loss. Finally, the output of the convolutional neural network is flattened into one-dimensional data as the input for the subsequent module.
[0012] As a preferred embodiment of the present invention, in step S3, time series inputs are generated by the sliding window method and combined with node feature inputs into a gated recurrent network.
[0013] As a preferred embodiment of the present invention, the sequence representation learning module in step S4 uses gated recurrent units with skip connections to extract long-term temporal features of traffic data. At each time step, the gated recurrent unit updates the hidden state of the current time step based on the current input and the hidden state of the previous time step.
[0014] In summary, the present invention has the following beneficial effects:
[0015] Improved prediction accuracy: By combining a gated recurrent network and an attention mechanism, the ability to model long-term dependencies is enhanced, and it performs well in processing complex spatio-temporal traffic data, especially with higher accuracy in long-term temporal prediction.
[0016] Optimized computational efficiency: Redundant calculations are reduced through a lightweight multi-head attention mechanism, and at the same time, the parallel processing characteristics of the convolutional neural network are utilized to improve the overall computational efficiency of the model. BRIEF DESCRIPTION OF THE DRAWINGS
[0017] Figure 1 It is a schematic diagram of an abstract complex traffic network.
[0018] Figure 2 It is a flowchart of a traffic flow prediction method. DETAILED DESCRIPTION OF THE EMBODIMENTS
[0019] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the scope of protection of the present invention.
[0020] Please refer to Figure 2, the present invention proposes a traffic flow prediction method based on a gated recurrent network, which mainly includes the following steps:
[0021] S1: Construct a traffic network sequence structure, and use a six-layer one-dimensional convolutional neural network to extract the spatial features of traffic data.
[0022] S2: Design a feature extraction module, and optimize the calculation of attention weights through sparsification processing.
[0023] S3: Combine the traffic road network features with the time series structure to generate the input of traffic flow data.
[0024] S4: Design a sequence representation learning module to capture the spatio-temporal correlation of data.
[0025] S5: Deploy the model to perform traffic flow prediction.
[0026] In this embodiment, by constructing a traffic flow prediction model, the spatio-temporal features in traffic data can be efficiently captured, and the prediction accuracy and calculation efficiency can be improved. First, the model constructs a traffic network sequence structure and uses a six-layer one-dimensional convolutional neural network to extract the spatial features of traffic data, thereby effectively capturing the spatial relationship between road nodes and providing high-quality input data for subsequent traffic flow prediction. Secondly, a feature extraction module is designed, and sparsification processing is used to optimize the calculation of attention weights, reduce redundant calculations, and improve the calculation efficiency. By combining the traffic road network features and the time series structure, the model generates accurate traffic flow data input and provides rich spatio-temporal features for the sequence representation learning module. In the sequence representation learning module, a gated recurrent unit with skip connections is used to extract the long-term time series features of traffic data. At each time step, the gated recurrent unit updates the hidden state of the current time step through the current input and the hidden state of the previous time step, thereby capturing long-term dependencies. Finally, the model accurately predicts the traffic flow through the prediction module, and with the help of the multi-head attention mechanism, the calculation efficiency is further optimized. By generating time series input through the sliding window method and combining node features to input the gated recurrent network, the model can effectively transmit spatio-temporal information between different time steps.
Claims
1. A traffic flow prediction method based on gated recurrent network, characterized in that The following steps are involved: S1: Construct a traffic network sequence structure and use a six-layer one-dimensional convolutional neural network to extract the spatial features of traffic data; S2: Design a feature extraction module to optimize attention weight calculation through sparse processing; S3: Combine traffic network characteristics with time series structure to generate traffic flow data input; S4: Design a sequence representation learning module to capture the spatiotemporal correlation of data; S5: Deploy the model to perform traffic flow prediction.
2. A traffic flow prediction method based on gated recurrent network according to claim 1, characterized in that: The network constructed in step S1 includes six layers of one-dimensional convolutional neural network, each convolutional layer is connected to a pooling layer to extract higher-level features and reduce noise, and then batch normalization technology is used to improve the training performance and stability of the model. Finally, the ReLU activation function is used and a nonlinear activation function is introduced to enhance the method's understanding of the data.
3. The traffic flow prediction method based on gated recurrent network according to claim 1 is characterized in that: The feature extraction module in step S2 uses padding technology to avoid feature loss, and finally flattens the output of the convolutional neural network into one-dimensional data as input for subsequent modules.
4. The traffic flow prediction method based on gated recurrent network according to claim 1 is characterized in that: The step S3 generates a time series input by a sliding window method and inputs the gated recurrent network in combination with the node features.
5. The traffic flow prediction method based on gated recurrent network according to claim 1 is characterized in that: The sequence representation learning module in step S4 uses a gated recurrent unit with a jump connection to extract long-term time series features of traffic data. At each time step, the gated recurrent unit updates the hidden state of the current time step according to the current input and the hidden state of the previous time step.