CN-121999604-A - Traffic flow prediction method and related device based on abnormal graph structure and space-time cyclic coupling

CN121999604ACN 121999604 ACN121999604 ACN 121999604ACN-121999604-A

Abstract

The application relates to the technical field of traffic flow prediction, in particular to a traffic flow prediction method and a related device based on abnormal graph structure and space-time cyclic coupling. The method comprises the steps of carrying out feature embedding processing on original traffic flow data through the data embedding layer to obtain an embedded feature vector, inputting the original traffic flow data and the embedded feature vector into a first layer space-time coding block of the encoder to obtain a space-time feature vector of the first layer space-time coding block, sequentially executing feature extraction operation on the space-time coding blocks except the first layer space-time coding block until an L-th layer space-time coding block outputs a global space-time feature vector, outputting the high-dimensional output vector to the decoder, and outputting a prediction result for representing future traffic flow through the decoder.

Inventors

ZHENG XIANWEI
QIU JIANQIANG
WU NANJING
XUE LICHUN
XIAO ZIKAI
ZENG LINA
Lu Juanxi
ZHANG JIEFENG

Assignees

佛山大学

Dates

Publication Date: 20260508
Application Date: 20251202

Claims (8)

1. The traffic flow prediction method based on the abnormal graph structure and the space-time cyclic coupling is characterized by being applied to a traffic flow prediction model, wherein the traffic flow prediction model comprises a data embedding layer, an encoder and a decoder, the encoder consists of L layers of space-time coding blocks, the space-time coding blocks comprise a first phase self-adaptive super-network structure, a graph convolution module and a second phase self-adaptive super-network structure, and the method comprises the following steps: Performing feature embedding processing on the original traffic flow data through the data embedding layer to obtain an embedded feature vector; Inputting the original traffic stream data and the embedded feature vector into a first layer space-time coding block of the encoder to obtain a space-time feature vector of the first layer space-time coding block; The space-time coding blocks except the first layer space-time coding block sequentially execute feature extraction operation until the L layer space-time coding block outputs global space-time feature vectors; outputting the high-dimensional output vector to the decoder, and outputting a prediction result used for representing future traffic flow through the decoder; the space-time characteristic vector of the space-time coding block of the first layer is obtained through the following steps: Inputting the embedded feature vector to a first phase self-adaptive super-network structure of the first layer space-time coding block to obtain a time feature output by the first phase self-adaptive super-network structure of the first layer space-time coding block; inputting the time feature and the original traffic flow data to the graph rolling module to obtain a space dimension feature with time positive feedback output by the graph rolling module of the first layer; The spatial dimension characteristics with time positive feedback output by the graph rolling module of the first layer are input into a second phase self-adaptive super-network structure of the space-time coding block of the first layer, so that a space-time characteristic vector output by the second phase self-adaptive super-network structure of the space-time coding block of the first layer is obtained, and the space-time characteristic vector output by the second phase self-adaptive super-network structure of the space-time coding block of the first layer is input into a first phase self-adaptive super-network structure of a space-time coding block of the next layer; Wherein the feature extraction operation includes the steps of: For any space-time coding block which is not the first layer space-time coding block, receiving a space-time feature vector output by a space-time coding block of a previous layer, inputting the space-time feature vector output by the space-time coding block of the previous layer to a first phase self-adaptive super-network structure of a current layer space-time coding block, and obtaining a time feature output by the first phase self-adaptive super-network structure of the current layer space-time coding block; The time characteristics output by the first phase self-adaptive super-network structure of the current layer space-time coding block and the original traffic flow data are input to a graph convolution module of the current layer space-time coding block, so that the space dimension characteristics with time positive feedback output by the graph convolution module of the current layer space-time coding block are obtained; And the spatial dimension characteristic with time positive feedback output by the graph convolution module of the current layer space-time coding block is input into a second phase self-adaptive super-network structure of the current layer space-time coding block, so that a space-time characteristic vector output by the second phase self-adaptive super-network structure of the current layer space-time coding block is obtained, and the space-time characteristic vector output by the second phase self-adaptive super-network structure of the current layer space-time coding block is input into a first phase self-adaptive super-network structure of a next layer space-time coding block of the current layer space-time coder.
2. The traffic flow prediction method based on the anomaly graph structure and the space-time cyclic coupling according to claim 1, wherein inputting the embedded feature vector into the first phase adaptive super network structure of the first layer space-time coding block, obtaining the time feature output by the first phase adaptive super network structure of the first layer space-time coding block, comprises: Determining phase tag information associated with the embedded feature vector; Generating an embedded representation vector according to the embedded feature vector and the phase tag information; Carrying out dynamic convolution on the embedded representation vector to obtain an intermediate feature; Performing gating activation according to the intermediate features to obtain a feature sequence after gating activation; and carrying out average pooling on the characteristic sequence after the gating activation to obtain the time characteristic output by the first phase self-adaptive super-network structure of the first layer space-time coding block.
3. The traffic flow prediction method based on abnormal graph structure and space-time cyclic coupling according to claim 2, wherein the time feature and the original traffic flow data are input to the graph convolution module to obtain a spatial dimension feature with positive feedback output by the first layer graph convolution module, comprising: constructing a differential map signal corresponding to the original traffic flow data to detect an abnormal moment when an abnormality exists; obtaining a plurality of abnormal graph structures according to the detected abnormal moment with the abnormality; obtaining a dynamic adjacent fusion matrix according to each abnormal graph structure; Normalizing the dynamic adjacent fusion matrix to obtain a normalized adjacent matrix; And generating a space dimension characteristic with time positive feedback according to the normalized adjacency matrix and the time characteristic.
4. The traffic flow prediction method based on the coupling of an anomaly graph structure and space-time circulation according to claim 3, wherein each anomaly graph structure corresponds to a single adjacency matrix, the obtaining a dynamic adjacency fusion matrix according to each anomaly graph structure comprises: And carrying out weighted fusion on the adjacent matrixes corresponding to the abnormal graph structures to obtain the dynamic adjacent fusion matrix.
5. The traffic flow prediction method based on the anomaly graph structure and the space-time cyclic coupling according to claim 3, wherein the generating a spatial dimension feature with positive feedback in time from the normalized adjacency matrix and the temporal feature comprises: And activating the weight of the graph convolution module, the normalized adjacent matrix and the time feature through a preset nonlinear ReLU activation function to obtain the space dimension feature with the time positive feedback.
6. A traffic flow prediction device based on an anomaly graph structure and space-time cyclic coupling, which is characterized by being applied to a traffic flow prediction model, wherein the traffic flow prediction model comprises a data embedding layer, an encoder and a decoder, the encoder is composed of L layers of space-time coding blocks, the space-time coding blocks comprise a first phase adaptive super-network structure, a graph convolution module and a second phase adaptive super-network structure, and the device comprises: The first processing unit is used for carrying out characteristic embedding processing on the original traffic flow data through the data embedding layer to obtain an embedded characteristic vector; the first output unit is used for inputting the original traffic stream data and the embedded feature vector into a first layer space-time coding block of the encoder to obtain a space-time feature vector of the first layer space-time coding block; a second output unit, configured to sequentially perform feature extraction operations by space-time coding blocks other than the first layer space-time coding block until the L-th layer space-time coding block outputs a global space-time feature vector; A prediction unit for outputting the high-dimensional output vector to the decoder, and outputting a prediction result for representing a future traffic flow through the decoder; the space-time characteristic vector of the space-time coding block of the first layer is obtained through the following steps: Inputting the embedded feature vector to a first phase self-adaptive super-network structure of the first layer space-time coding block to obtain a time feature output by the first phase self-adaptive super-network structure of the first layer space-time coding block; inputting the time feature and the original traffic flow data to the graph rolling module to obtain a space dimension feature with time positive feedback output by the graph rolling module of the first layer; The spatial dimension characteristics with time positive feedback output by the graph rolling module of the first layer are input into a second phase self-adaptive super-network structure of the space-time coding block of the first layer, so that a space-time characteristic vector output by the second phase self-adaptive super-network structure of the space-time coding block of the first layer is obtained, and the space-time characteristic vector output by the second phase self-adaptive super-network structure of the space-time coding block of the first layer is input into a first phase self-adaptive super-network structure of a space-time coding block of the next layer; Wherein the feature extraction operation includes the steps of: For any space-time coding block which is not the first layer space-time coding block, receiving a space-time feature vector output by a space-time coding block of a previous layer, inputting the space-time feature vector output by the space-time coding block of the previous layer to a first phase self-adaptive super-network structure of a current layer space-time coding block, and obtaining a time feature output by the first phase self-adaptive super-network structure of the current layer space-time coding block; The time characteristics output by the first phase self-adaptive super-network structure of the current layer space-time coding block and the original traffic flow data are input to a graph convolution module of the current layer space-time coding block, so that the space dimension characteristics with time positive feedback output by the graph convolution module of the current layer space-time coding block are obtained; And the spatial dimension characteristic with time positive feedback output by the graph convolution module of the current layer space-time coding block is input into a second phase self-adaptive super-network structure of the current layer space-time coding block, so that a space-time characteristic vector output by the second phase self-adaptive super-network structure of the current layer space-time coding block is obtained, and the space-time characteristic vector output by the second phase self-adaptive super-network structure of the current layer space-time coding block is input into a first phase self-adaptive super-network structure of a next layer space-time coding block of the current layer space-time coder.
7. An electronic device comprising a memory and a processor, the memory storing a computer program, wherein the processor, when executing the computer program, implements the traffic flow prediction method based on the anomaly graph structure coupled with spatiotemporal circulation of any one of claims 1 to 5.
8. A computer program product comprising a computer program, characterized in that the computer program is read and executed by a processor of an electronic device, causing the electronic device to perform the traffic flow prediction method based on an anomaly graph structure coupled with a spatiotemporal cycle according to any one of claims 1 to 5.

Description

Traffic flow prediction method and related device based on abnormal graph structure and space-time cyclic coupling Technical Field The application relates to the technical field of traffic flow prediction, in particular to a traffic flow prediction method and a related device based on abnormal graph structure and space-time cyclic coupling. Background Currently, existing traffic flow prediction relies on a fixed convolution structure, and conventional fixed convolution kernels are difficult to aim at the periodic variation in traffic flow prediction, so that the problem of dynamic adaptation is caused. The traditional convolution structure extracts the characteristics of all period time by using static parameters, so that the model is difficult to accurately capture the difference characteristics of different modes or different states in the periodic characteristics, such as processing the peak, the holiday flow and the like of the workday. Therefore, the conventional traffic flow prediction method has a problem of insufficient prediction accuracy. Disclosure of Invention In order to solve at least one of the problems, an embodiment of the present application provides a traffic flow prediction method and related device based on the coupling of an anomaly graph structure and space-time circulation, which can improve the accuracy of traffic flow prediction. According to an aspect of the embodiment of the application, a traffic flow prediction method based on the coupling of an anomaly graph structure and space-time circulation is provided, and the traffic flow prediction method is applied to a traffic flow prediction model, wherein the traffic flow prediction model comprises a data embedding layer, an encoder and a decoder, the encoder consists of L layers of space-time coding blocks, the space-time coding blocks comprise a first phase self-adaptive super-network structure, a graph convolution module and a second phase self-adaptive super-network structure, and the method comprises the following steps: Performing feature embedding processing on the original traffic flow data through the data embedding layer to obtain an embedded feature vector; Inputting the original traffic stream data and the embedded feature vector into a first layer space-time coding block of the encoder to obtain a space-time feature vector of the first layer space-time coding block; The space-time coding blocks except the first layer space-time coding block sequentially execute feature extraction operation until the L layer space-time coding block outputs global space-time feature vectors; outputting the high-dimensional output vector to the decoder, and outputting a prediction result used for representing future traffic flow through the decoder; the space-time characteristic vector of the space-time coding block of the first layer is obtained through the following steps: Inputting the embedded feature vector to a first phase self-adaptive super-network structure of the first layer space-time coding block to obtain a time feature output by the first phase self-adaptive super-network structure of the first layer space-time coding block; inputting the time feature and the original traffic flow data to the graph rolling module to obtain a space dimension feature with time positive feedback output by the graph rolling module of the first layer; The spatial dimension characteristics with time positive feedback output by the graph rolling module of the first layer are input into a second phase self-adaptive super-network structure of the space-time coding block of the first layer, so that a space-time characteristic vector output by the second phase self-adaptive super-network structure of the space-time coding block of the first layer is obtained, and the space-time characteristic vector output by the second phase self-adaptive super-network structure of the space-time coding block of the first layer is input into a first phase self-adaptive super-network structure of a space-time coding block of the next layer; Wherein the feature extraction operation includes the steps of: For any space-time coding block which is not the first layer space-time coding block, receiving a space-time feature vector output by a space-time coding block of a previous layer, inputting the space-time feature vector output by the space-time coding block of the previous layer to a first phase self-adaptive super-network structure of a current layer space-time coding block, and obtaining a time feature output by the first phase self-adaptive super-network structure of the current layer space-time coding block; The time characteristics output by the first phase self-adaptive super-network structure of the current layer space-time coding block and the original traffic flow data are input to a graph convolution module of the current layer space-time coding block, so that the space dimension characteristics with time positive feedback output by the graph convolution modul