CN115482666B - Multi-graph convolution neural network traffic prediction method based on data fusion - Google Patents
Multi-graph convolution neural network traffic prediction method based on data fusion Download PDFInfo
- Publication number
- CN115482666B CN115482666B CN202211110596.1A CN202211110596A CN115482666B CN 115482666 B CN115482666 B CN 115482666B CN 202211110596 A CN202211110596 A CN 202211110596A CN 115482666 B CN115482666 B CN 115482666B
- Authority
- CN
- China
- Prior art keywords
- traffic
- traffic flow
- data
- graph
- model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 39
- 230000004927 fusion Effects 0.000 title claims abstract description 26
- 238000000034 method Methods 0.000 title claims abstract description 25
- 230000000737 periodic effect Effects 0.000 claims abstract description 17
- 238000012549 training Methods 0.000 claims description 32
- 238000010276 construction Methods 0.000 claims description 23
- 238000012360 testing method Methods 0.000 claims description 21
- 230000004913 activation Effects 0.000 claims description 15
- 238000013527 convolutional neural network Methods 0.000 claims description 12
- 238000011176 pooling Methods 0.000 claims description 9
- 238000011156 evaluation Methods 0.000 claims description 8
- 230000003213 activating effect Effects 0.000 claims description 6
- 239000011159 matrix material Substances 0.000 claims description 6
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 3
- 238000003064 k means clustering Methods 0.000 claims description 3
- 101150006061 neur gene Proteins 0.000 claims description 3
- 238000005457 optimization Methods 0.000 claims description 3
- 238000007781 pre-processing Methods 0.000 claims description 3
- 238000005096 rolling process Methods 0.000 claims description 3
- 238000000638 solvent extraction Methods 0.000 claims description 3
- 230000008859 change Effects 0.000 claims description 2
- 238000012163 sequencing technique Methods 0.000 claims 1
- 230000008569 process Effects 0.000 abstract description 4
- 230000002708 enhancing effect Effects 0.000 abstract description 2
- 238000000605 extraction Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 239000000284 extract Substances 0.000 description 3
- 101001095088 Homo sapiens Melanoma antigen preferentially expressed in tumors Proteins 0.000 description 2
- 102100037020 Melanoma antigen preferentially expressed in tumors Human genes 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 230000008901 benefit Effects 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G08—SIGNALLING
- G08G—TRAFFIC CONTROL SYSTEMS
- G08G1/00—Traffic control systems for road vehicles
- G08G1/01—Detecting movement of traffic to be counted or controlled
- G08G1/0104—Measuring and analyzing of parameters relative to traffic conditions
- G08G1/0125—Traffic data processing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/762—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using clustering, e.g. of similar faces in social networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/80—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level
- G06V10/806—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- G—PHYSICS
- G08—SIGNALLING
- G08G—TRAFFIC CONTROL SYSTEMS
- G08G1/00—Traffic control systems for road vehicles
- G08G1/01—Detecting movement of traffic to be counted or controlled
- G08G1/0104—Measuring and analyzing of parameters relative to traffic conditions
- G08G1/0137—Measuring and analyzing of parameters relative to traffic conditions for specific applications
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Medical Informatics (AREA)
- Databases & Information Systems (AREA)
- Multimedia (AREA)
- Chemical & Material Sciences (AREA)
- Analytical Chemistry (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Traffic Control Systems (AREA)
Abstract
The invention discloses a multi-graph convolution neural network traffic prediction method based on data fusion, which aims at solving the problem that the space-time characteristics of nodes are not considered when a traffic topological graph is constructed by a current main flow model, and 3D convolution is utilized to form a 3DRepVGG component to extract the space-time characteristics of European traffic flow data and take the space-time characteristics as the characteristics of each node in the traffic topological graph; the periodic gating logic unit is designed to process non-European traffic flow data, so that the capability of the model for extracting time characteristics is improved, the regional traffic condition is identified and quantified by using a clustering algorithm, the generalization capability of the model is improved, and the parameter quantity is reduced; and constructing a plurality of traffic topological graphs according to the regional traffic state, the non-European traffic flow data and the node characteristic data, enhancing the capability of the model in extracting the remote space characteristics, and further improving the prediction precision.
Description
Technical Field
The invention belongs to the field of traffic flow prediction, and relates to a multi-graph convolution neural network traffic prediction method based on data fusion.
Background
Currently, most traffic flow prediction models treat traffic flow data as an European data structure, i.e. a city is divided into grid areas of regular shape, each grid value representing the number of vehicles in the area. The benefit of this European structure is that it is straightforward and intuitive, and the traffic volume per zone can be clearly understood, but the drawbacks are also apparent, and the structure is not indicative of zone-to-zone vehicle flow. Along with the construction of expressways in cities, vehicles can flow more conveniently across areas, and dependency relationship can be generated between two areas which are far apart. Deployment of a large number of advanced sensors on roads also makes it possible to collect inter-regional vehicle flow information. By means of the vehicle flow information between the areas, the change of urban traffic can be more intuitively known. The areas will not only affect adjacent areas, but also vehicle flows with areas that are far apart, and even geographically adjacent areas will not be affected by the areas in certain time periods, and new traffic flow data will often exhibit non-European structural features, preserving effective long-range spatial correlation.
In general, european traffic flow data contains fewer long-range spatial features and because of the shared nature of the convolution kernel parameters, the field of view of the model extraction features is smaller. In recent years, in a model using european traffic flow data as a data set, in order to extract a distant spatial correlation, researchers often adopt two methods of hole convolution and deepening network depth, and although this method can obtain global spatial features to a certain extent, the possibility of extracting invalid information by the model is increased, so that the prediction accuracy of the model is affected. In addition, when facing a training set of a small scale, the prediction model with a deeper structure often has problems such as gradient explosion or gradient disappearance. In summary, it is insufficient to predict regional traffic data for the next time step using only European traffic flow data, and non-European traffic flow data is required to construct a traffic topology map. However, since traffic flow also has multiple characteristics, constructing a non-European topology map with only a single characteristic is not sufficient to fully characterize traffic flow data. In addition, node features in non-European data generally only have features such as flow or speed, but have no space-time features, and most mainstream prediction models often ignore the features. Therefore, how to effectively process non-European structures and consider various characteristics and node characteristics of traffic flow to accurately and comprehensively characterize traffic flow data to construct an accurate traffic flow prediction model has yet to be studied.
Disclosure of Invention
Aiming at the defects of the prior art, the invention discloses a multi-graph convolution neural network traffic prediction method based on data fusion, so as to realize efficient and accurate prediction of large-scale traffic flow data. The main content is as follows: (1) Aiming at the space-time characteristics of nodes are not considered when a traffic topological graph is constructed by a current main flow model, 3D convolution is utilized to form a 3DRepVGG component to extract the space-time characteristics of European traffic flow data, the space-time characteristics are used as each node characteristic (2) in the traffic topological graph, a periodic gating logic unit is designed to process non-European traffic flow data, the capacity of the model for extracting time characteristics is improved, the regional traffic condition is identified and quantified by a clustering algorithm, the generalization capacity of the model is improved, and the quantity of parameters is reduced. (3) And constructing a plurality of traffic topological graphs according to the regional traffic state, the non-European traffic flow data and the node characteristic data, enhancing the capability of the model in extracting the remote space characteristics, and further improving the prediction precision.
The invention comprises two steps: constructing a multi-graph convolution neural network traffic prediction model based on data fusion, and training and testing the multi-graph convolution neural network traffic prediction model based on data fusion.
Step 1: multi-graph convolution neural network traffic prediction model construction based on data fusion
The construction of the multi-graph convolution neural network traffic prediction model based on data fusion comprises 3 steps: non-European node characteristic information is obtained, a traffic topological graph is constructed, and a multi-graph convolution neural network is constructed;
Step 1-1: non-European node characteristic information acquisition
3D convolution is utilized to form a 3DRepVGG component to extract the space-time characteristics of European traffic flow data and serve as non-European node characteristic information;
The 3DRepVGG component contains three layers RepVGG operations, each layer RepVGG operation contains two 3D convolutions, one 3D hole convolution and activation operations; the formula for the 3DRepVGG component is as follows:
Wherein, sigma is the activation operation, The output of the operations for layer RepVGG, which is also the input of the operations for layer RepVGG, conv 1 is a 3D convolution of 1 x 1 structure, conv 3 is a 3D convolution of a 3x 3 structure and dconv 3 is a 3D hole convolution of a 3x 3 structure.
The spatio-temporal feature Y N is finally obtained by the 3DRepVGG component and serves as a non-euro node feature V.
Step 1-2: traffic topology construction
In order to fully express the traffic flow characteristics, the prediction accuracy of the model is further improved, and a traffic flow similarity topological graph and a traffic flow influence topological graph are respectively constructed to comprehensively describe the characteristics of traffic flow. The construction of the traffic topology map is divided into 4 steps: non-European traffic flow direction data acquisition, regional traffic state division, traffic flow direction similarity topological graph construction and traffic flow direction influence topological graph.
Step 1-2-1: non-European traffic flow data acquisition
The periodic gating logic unit is designed to obtain periodic non-European traffic flow data. The period gating logic unit consists of two pooling layers, a one-dimensional rolling and activating operation, and the formula of the unit is as follows:
Wherein X NEur is non-European traffic flow data, AP is average pooling operation, MP is maximum pooling operation, delta 1 and delta 2 are self-defined learnable variables, sigma is activating operation, CNN is one-dimensional convolution, Is periodic traffic flow data.
Step 1-2-2: regional traffic state partitioning
And dividing the traffic state of the area by using a K-means clustering algorithm. The method comprises the following specific steps:
(1) Traffic flow data preprocessing. The number of vehicles flowing out of the area, the number of vehicles flowing in the area, and the number of area connections (the number of areas flowing with other vehicles) are obtained from the non-European traffic flow data, and a3 XH X W sample set is formed.
(2) Initializing a cluster class center. Setting the number K of needed cluster types, and randomly generating K cluster type centers Ck, K E (1..K) by an algorithm.
(3) Sample variance values are calculated. And calculating the difference value between the sample point and the cluster center point according to the Euclidean distance, wherein the class to which each sample belongs is the class represented by the cluster center with the smallest distance.
(4) Updating the cluster center. The average value of all samples in the current cluster is calculated and used as a new cluster center.
(5) Repeating the step (3) and the step (4) until all sample points are not changed in category.
And finally obtaining the original cluster labels C k of the areas.
Step 1-2-3: traffic flow similarity topology graph construction
And constructing a traffic flow similarity topological graph by the node characteristic data, the periodic traffic flow data and the regional traffic state data. The definition of the traffic flow similarity value is as follows:
Wherein, Is the traffic state of the vehicle outflow zone,/>Is the traffic state of the vehicle inflow region, C max is the maximum value of the clusters, k all is the number of the clusters, and the total traffic state classification value of the current traffic state cluster is the upper part of the cluster. e S (i, j) is the similarity value of the traffic flow direction of the area i to the area j, namely the edge feature in the traffic topology. E s is the traffic flow similarity of the city, which is composed of the traffic flow similarity values of all areas. Finally, the traffic flow similarity topology is denoted as G s＝(V,Es).
Step 1-2-4: traffic flow direction influencing topological graph construction
And constructing a traffic flow similarity topological graph by the node characteristic data and the periodic traffic flow data. The traffic flow direction influence degree value is defined as follows:
Where e (i,j) is the vehicle flow data between region i and region j, For the total flow value of the area i, the numerator is the difference value of the two, and e P (i, j) is the traffic flow direction influence degree value of the area i and the area j. E P is the degree of influence of the traffic flow direction of the city, which is composed of the traffic flow direction influence values of all areas. Finally, the traffic flow direction impact topology is denoted as G P＝(V,EP).
Step 1-3: multi-graph convolution neural network construction
After two traffic topological graphs are constructed, a multi-graph convolution neural network is designed to extract the characteristics of the two topological graphs.
The multi-graph convolutional neural network is composed of a graph convolutional neural network, an activation operation and a full connection layer. The two topological graphs are respectively subjected to graph convolution neural network and activation operation to extract features, and then the features are fused and further learned through a full connection layer to obtain the features of traffic flow data. The specific formula is as follows:
YG＝Θ(Θ(o1×YS+o2×YP))
wherein, Q S is a node degree matrix obtained according to the traffic flow direction similarity topological graph G S, and Q P is a degree matrix obtained according to the traffic flow direction influence topological graph G p; f is the total operator of the convolution operation and the activation function, and two operations are needed for G S and G p to obtain Y S and Y P, respectively. Θ is a fully connected operation, o 1 and o 2 are custom learnable variables, and Y G is a feature of the traffic topology map.
And finally, fusing the features extracted by the 3DRepVGG component with the features of the traffic topological graph to obtain a final prediction result.
Y＝q1×YN+q2×YG
Wherein q 1 and q 2 are custom learnable variables.
Step 2: multi-graph convolution neural network traffic prediction model training and testing based on data fusion
Training and testing of a multi-graph convolution neural network traffic prediction model based on data fusion are divided into two steps: model training and model testing;
Step 2-1: model training
The traffic flow data is divided into a training set and a testing set, after the training set data are ordered according to time sequence, part of data are extracted as input according to the requirements of each component in each training round.
The Adam optimization algorithm is adopted to adjust parameters during model training; when the training times reach the set k value, the model stops training;
step 2-2: model testing
And predicting in the test set by using the trained model, and comparing the obtained predicted value with a real observed value to obtain the prediction precision.
The evaluation index of the model performance is as follows:
mean absolute error (Mean Absolute Error, MAE): the result is the average of the absolute errors between the actual and predicted values, and the formula is as follows:
Mean absolute percent error (Mean Absolute Percentage Error, MAPE): the result is the average of the absolute percentage error between the actual and predicted values, as follows:
Root mean square error (Root Mean Square Error, RMSE): the result is the arithmetic square root of the mean square error between the actual and predicted values, as follows:
after the model obtains three evaluation indexes, the model is compared with the current main stream model, and the performance of the model is proved to be in the leading position.
The method of the invention, namely the multi-graph convolution neural network traffic prediction model based on data fusion, can realize better extraction of the characteristics of traffic flow data under the condition of using European traffic flow data and non-European traffic flow data at the same time. Based on the trained parameter model, a high-precision prediction result can be obtained. Meanwhile, the invention has the following characteristics:
1) The important characteristic of non-European traffic flow data is the long-distance spatial characteristic, in the method of the invention, a multi-graph convolution neural network is formed by utilizing the graph convolution neural network and the full-connection layer, and the characteristic of the traffic topological graph is extracted by utilizing the spectrogram theory, so that the model can model urban traffic more comprehensively.
2) At present, most researches do not consider the space-time characteristics of the nodes, the flow or the speed of the nodes are generally used as single characteristics of the nodes, the space-time characteristics of the nodes are ignored, and the model cannot comprehensively learn the characteristics of traffic flow data. The invention adopts 3DRepVGG components to extract the space-time characteristics of European traffic flow data and takes the space-time characteristics as the space-time characteristics of the nodes, so that the model can learn the characteristics of the nodes more comprehensively and further improve the prediction precision.
3) Aiming at the defect of insufficient time feature extraction capability of the graph convolutional neural network, the invention designs the periodic gating logic unit to firstly process non-European traffic flow data to obtain periodic traffic flow data to construct a traffic topological graph, and improves the time feature extraction capability of the model, thereby helping the model to obtain more accurate prediction.
4) Aiming at the characteristic that a single traffic flow topological graph cannot comprehensively express traffic flow data, the traffic flow similarity topological graph and the traffic flow influence topological graph are respectively constructed, traffic flow is respectively modeled from two angles of traffic flow similarity and traffic flow influence degree, and the model is helped to learn the characteristics of different traffic flow characteristics and traffic flow influence degrees of similar areas, so that the prediction accuracy of the model is further improved.
Drawings
Fig. 1 is a general flow chart of the present invention.
Fig. 2 is a model flow diagram of the present invention.
Fig. 3 is a block diagram of 3DRepVGG of the present invention.
FIG. 4 is a block diagram of the cycle gating logic component of the present invention.
Fig. 5 is a multi-graph convolutional neural network structure of the present invention.
Detailed Description
As shown in fig. 1, the present invention comprises two steps: constructing a multi-graph convolution neural network traffic prediction model based on data fusion, and training and testing the multi-graph convolution neural network traffic prediction model based on data fusion.
Step 1: multi-graph convolution neural network traffic prediction model construction based on data fusion
The specific flow is shown in fig. 2, and the construction of the multi-graph convolution neural network traffic prediction model based on data fusion comprises 3 steps: non-European node characteristic information is obtained, a traffic topological graph is constructed, and a multi-graph convolution neural network is constructed;
Step 1-1: non-European node characteristic information acquisition
As shown in fig. 3, 3D convolution is used to form a 3DRepVGG component to extract the space-time characteristics of the european traffic flow data and to serve as non-european node characteristic information;
The 3DRepVGG component contains three layers RepVGG operations, each layer RepVGG operation contains two 3D convolutions, one 3D hole convolution and activation operations; the formula for the 3DRepVGG component is as follows:
Wherein, sigma is the activation operation, The output of the operations for layer RepVGG, which is also the input of the operations for layer RepVGG, conv 1 is a 3D convolution of 1 x 1 structure, conv 3 is a 3D convolution of a 3x 3 structure and dconv 3 is a 3D hole convolution of a 3x 3 structure.
The spatio-temporal feature Y N is finally obtained by the 3DRepVGG component and serves as a non-euro node feature V.
Step 1-2: traffic topology construction
In order to fully express traffic flow characteristics and further improve the prediction accuracy of the model, the invention respectively constructs a traffic flow similarity topological graph and a traffic flow influence topological graph to comprehensively describe the characteristics of traffic flow. The construction of the traffic topology map is divided into 4 steps: non-European traffic flow direction data acquisition, regional traffic state division, traffic flow direction similarity topological graph construction and traffic flow direction influence topological graph.
Step 1-2-1: non-European traffic flow data acquisition
As shown in fig. 4, the period gating logic is designed to obtain periodic non-european traffic flow data. The period gating logic unit consists of two pooling layers, a one-dimensional rolling and activating operation, and the formula of the unit is as follows:
Wherein X NEur is non-European traffic flow data, AP is average pooling operation, MP is maximum pooling operation, delta 1 and delta 2 are self-defined learnable variables, sigma is activating operation, CNN is one-dimensional convolution, Is periodic traffic flow data.
Step 1-2-2: regional traffic state partitioning
And dividing the traffic state of the area by using a K-means clustering algorithm. The method comprises the following specific steps:
(1) Traffic flow data preprocessing. The number of vehicles flowing out of the area, the number of vehicles flowing in the area, and the number of area connections (the number of areas flowing with other vehicles) are obtained from the non-European traffic flow data, and a3 XH X W sample set is formed.
(2) Initializing a cluster class center. Setting the number K of needed cluster types, and randomly generating K cluster type centers Ck, K E (1..K) by an algorithm.
(3) Sample variance values are calculated. And calculating the difference value between the sample point and the cluster center point according to the Euclidean distance, wherein the class to which each sample belongs is the class represented by the cluster center with the smallest distance.
(4) Updating the cluster center. The average value of all samples in the current cluster is calculated and used as a new cluster center.
(5) Repeating the step (3) and the step (4) until all sample points are not changed in category.
And finally obtaining the original cluster labels C k of the areas.
Step 1-2-3: traffic flow similarity topology graph construction
And constructing a traffic flow similarity topological graph by the node characteristic data, the periodic traffic flow data and the regional traffic state data. The definition of the traffic flow similarity value is as follows:
Wherein, Is the traffic state of the vehicle outflow zone,/>Is the traffic state of the vehicle inflow region, C max is the maximum value of the clusters, k all is the number of the clusters, and the total traffic state classification value of the current traffic state cluster is the upper part of the cluster. e S (i, j) is the similarity value of the traffic flow direction of the area i to the area j, namely the edge feature in the traffic topology. E s is the traffic flow similarity of the city, which is composed of the traffic flow similarity values of all areas. Finally, the traffic flow similarity topology is denoted as G s＝(V,Es).
Step 1-2-4: traffic flow direction influencing topological graph construction
And constructing a traffic flow similarity topological graph by the node characteristic data and the periodic traffic flow data. The traffic flow direction influence degree value is defined as follows:
Where e (i,j) is the vehicle flow data between region i and region j, For the total flow value of the area i, the numerator is the difference value of the two, and e P (i, j) is the traffic flow direction influence degree value of the area i and the area j. E P is the degree of influence of the traffic flow direction of the city, which is composed of the traffic flow direction influence values of all areas. Finally, the traffic flow direction impact topology is denoted as G P＝(V,EP).
Step 1-3: multi-graph convolution neural network construction
After two traffic topological graphs are constructed, a multi-graph convolution neural network is designed to extract the characteristics of the two topological graphs.
As shown in fig. 5, the multi-graph convolutional neural network is composed of a graph convolutional neural network, an activation operation, and a full connection layer. The two topological graphs are respectively subjected to graph convolution neural network and activation operation to extract features, and then the features are fused and further learned through a full connection layer to obtain the features of traffic flow data. The specific formula is as follows:
YG＝Θ(Θ(o1×YS+o2×YP))
Wherein, Q S is a degree matrix of each node obtained according to the traffic similarity topological graph G S, and Q P is a degree matrix obtained according to the traffic similarity topological graph G p. f is the total operator of the convolution operation and the activation function, and two operations are needed for G S and G p to obtain Y S and Y P, respectively. Θ is the fully connected operation, o 1 and o 2 are the custom learnable variables, Y G is the characteristics of the traffic topology map
And finally, fusing the features extracted by the 3DRepVGG component with the features of the traffic topological graph to obtain a final prediction result.
Y＝q1×YN+q2×YG
Wherein q 1 and q 2 are custom learnable variables.
Step 2: multi-graph convolution neural network traffic prediction model training and testing based on data fusion
Training and testing of a multi-graph convolution neural network traffic prediction model based on data fusion are divided into two steps: model training and model testing;
Step 2-1: model training
The data set is selected from taxi driving data of 1 month and 1 day in 2015, and New York. The dataset divided new york city into a 10x 20 grid plot, data collected every half hour, data divided into two formats: the first is data of European structure, and regional traffic data of the whole city is defined as tensor form: Where T is the time step and C is the traffic type (input traffic and output traffic). The second type of data is data with a non-European structure, and the traffic flow data among the whole urban areas is defined as tensor form: /(I) Representing the value of the vehicle flow between the region (i, j) and the other region at time step t.
The data of the first 40 days in the data set is used as a training set, and part of data is extracted as input according to the requirements of each component in each training round. The periodicity component extracts periodicity data for 4 time points as input (t=4), the near term component extracts 16 time point data for near term data as input (t=16), and the trend component extracts 16 time point data one week apart from the near term data as input (t=16).
The parameters are adjusted by adopting an Adam optimization algorithm during model training, the first-order exponential decay rate is set to 0.9, the second-order exponential decay rate is set to 0.999, the learning rate is set to 1e-3, and the weight decay is set to 0.005. When the number of training times reached 600 times set, the model stopped training.
Step 2-2: model testing
And predicting in the test set by using the trained model, and comparing the obtained predicted value with a real observed value to obtain the prediction precision.
The evaluation index of the model performance is as follows:
The data of the first 40 days in the data set is used as a test set, a trained model is used for prediction in the test set, the obtained predicted value is compared with the actual observed value to obtain the prediction precision, and three indexes of MAE, MAPE and RMSE are adopted as the evaluation indexes of the model performance. After the model obtains three evaluation indexes, the three evaluation indexes are compared with the current mainstream model. The model provided by the invention is in the leading position on most indexes, the data of three indexes of the model are 5.51, 0.27 and 14.09 respectively, and experimental results prove that the invention is effective.
Claims (4)
1. The traffic prediction method of the multi-graph convolution neural network based on data fusion is characterized by comprising the following steps of: the method specifically comprises the following steps:
step one: multi-graph convolution neural network traffic prediction model construction based on data fusion
The construction of the multi-graph convolution neural network traffic prediction model based on data fusion comprises 3 steps: non-European node characteristic information is obtained, a traffic topological graph is constructed, and a multi-graph convolution neural network is constructed; the traffic topological graph consists of a traffic flow direction similarity topological graph and a traffic flow direction influence topological graph;
the multi-graph convolution neural network is constructed specifically as follows:
After two traffic topological graphs are constructed, a multi-graph convolution neural network is designed to extract the characteristics of the two topological graphs;
The multi-graph convolution neural network consists of a graph convolution neural network, an activation operation and a full connection layer; the two topological graphs are respectively subjected to graph convolution neural network and activation operation to extract features, and then the features are fused and further learned through a full connection layer to obtain the features of traffic flow direction data; the specific formula is as follows:
YG＝Θ(Θ(o1×YS+o2×YP))
Wherein, Q S is a node degree matrix obtained according to the traffic flow direction similarity topological graph G S, and Q P is a degree matrix obtained according to the traffic flow direction influence topological graph G p; f is the total operator of the graph convolution operation and the activation function, and G S and G p can obtain Y S and Y P through two operations respectively; Θ is a fully connected operation, o 1 and o 2 are custom learnable variables, and Y G is a feature of a traffic topology map;
finally, fusing the features extracted by the 3DRepVGG component with the features of the traffic topological graph to obtain a final prediction result;
Y＝q1×YN+q2×YG
Wherein q 1 and q 2 are self-defined learnable variables, and Y N is a characteristic obtained by non-European node characteristic information;
step two: multi-graph convolution neural network traffic prediction model training and testing based on data fusion
Training and testing of a multi-graph convolution neural network traffic prediction model based on data fusion are divided into two steps: model training and model testing;
The non-European node characteristic information is obtained specifically as follows:
3D convolution is utilized to form a 3DRepVGG component to extract the space-time characteristics of European traffic flow data and serve as non-European node characteristic information;
The 3DRepVGG component contains three layers RepVGG operations, each layer RepVGG operation contains two 3D convolutions, one 3D hole convolution and activation operations; the formula for the 3DRepVGG component is as follows:
Wherein, sigma is the activation operation, The output of the operations for layer RepVGG, which is also the input of the operations for layer RepVGG, conv 1 is a 3D convolution of 1 x 1 structure, conv 3 is a 3D convolution of a3 x 3 structure, dconv 3 is a 3D hole convolution of a3 x 3 structure;
The spatio-temporal feature Y N is finally obtained by the 3DRepVGG component and serves as a non-euro node feature V.
2. The data fusion-based multi-graph convolutional neural network traffic prediction method of claim 1, wherein the method comprises the steps of: the traffic topology map is constructed specifically as follows:
The construction of the traffic topology map is divided into 4 steps: non-European traffic flow direction data acquisition, regional traffic state division, traffic flow direction similarity topological graph construction and traffic flow direction influence topological graph;
step 1: non-European traffic flow data acquisition
The method comprises the steps of designing a period gating logic unit to obtain periodic non-European traffic flow direction data; the period gating logic unit consists of two pooling layers, a one-dimensional rolling and activating operation, and the formula of the unit is as follows:
Wherein X NEur is non-European traffic flow data, AP is average pooling operation, MP is maximum pooling operation, delta 1 and delta 2 are self-defined learnable variables, sigma is activating operation, CNN is one-dimensional convolution, Is periodic traffic flow data;
Step2: regional traffic state partitioning
Dividing the traffic state of the area by using a K-means clustering algorithm; the method comprises the following specific steps:
(1) Traffic flow data preprocessing
Acquiring the number of vehicles flowing out of a region, the number of vehicles flowing in of the region and the number of region connection, namely the number of regions flowing with other vehicles, from non-European traffic flow data to form a3 XH XW sample set;
(2) Initializing cluster class centers
Setting the number K of needed clusters, and randomly generating K cluster centers Ck and K E (1..K) by an algorithm;
(3) Calculating a sample difference value
Calculating the difference value between the sample point and the cluster center point according to the Euclidean distance, wherein the class to which each sample belongs is the class represented by the cluster center with the smallest distance;
(4) Updating cluster centers
Calculating the average value of all samples in the current cluster, and taking the average value as a new cluster center;
(5) Repeating step (3) and step (4) until all sample points no longer change category;
finally, the original cluster labels C k of all the areas are obtained;
Step 3: traffic flow similarity topology graph construction
Constructing a traffic flow similarity topological graph by node characteristic data, periodic traffic flow data and regional traffic state data; the definition of the traffic flow similarity value is as follows:
Wherein, Is the traffic state of the vehicle outflow zone,/>Is the traffic state of the vehicle inflow region, C max is the maximum value of clusters, k all is the number of clusters, and the total traffic state classification value of the current traffic state cluster is divided; e S (i, j) is the similarity value of the traffic flow direction of the area i to the area j, namely the edge characteristic in the traffic topological graph; e s is the similarity of the traffic flow directions of cities, and is composed of the traffic flow direction similarity values of all areas; finally, the traffic flow similarity topological graph is represented as G s＝(V,Es), and V is a non-European node characteristic;
Step 4: traffic flow direction influencing topological graph construction
Constructing a traffic flow similarity topological graph by the node characteristic data and the periodic traffic flow data; the traffic flow direction influence degree value is defined as follows:
Where e (i,j) is the vehicle flow data between region i and region j, For the total flow value of the region i, the numerator is the difference value of the two, and e P (i, j) is the traffic flow direction influence degree value of the region i and the region j; e P is the influence degree of the traffic flow direction of the city, and is composed of the traffic flow direction influence values of all areas; finally, the traffic flow direction impact topology is denoted as G P＝(V,EP).
3. The data fusion-based multi-graph convolutional neural network traffic prediction method of claim 1, wherein the method comprises the steps of: the model training is specifically as follows:
dividing traffic flow data into a training set and a testing set, sequencing the training set data according to time sequence, and extracting partial data as input according to the requirements of each component in each training round;
The Adam optimization algorithm is adopted to adjust parameters during model training; when the training times reach the set k value, the model stops training.
4. The data fusion-based multi-graph convolutional neural network traffic prediction method of claim 1, wherein the method comprises the steps of: the model test specifically comprises the following steps:
Predicting in the test set by using the trained model, and comparing the obtained predicted value with a real observed value to obtain prediction precision;
The evaluation index of the model performance is as follows:
Average absolute error: the result is the average of the absolute errors between the actual and predicted values, and the formula is as follows:
average absolute percentage error: the result is the average of the absolute percentage error between the actual and predicted values, as follows:
Root mean square error: the result is the arithmetic square root of the mean square error between the actual and predicted values, as follows:
after the model obtains three evaluation indexes, the model is compared with the current main stream model, and the performance of the model is proved to be in the leading position.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211110596.1A CN115482666B (en) | 2022-09-13 | 2022-09-13 | Multi-graph convolution neural network traffic prediction method based on data fusion |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211110596.1A CN115482666B (en) | 2022-09-13 | 2022-09-13 | Multi-graph convolution neural network traffic prediction method based on data fusion |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115482666A CN115482666A (en) | 2022-12-16 |
| CN115482666B true CN115482666B (en) | 2024-05-07 |
Family
ID=84423449
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211110596.1A Active CN115482666B (en) | 2022-09-13 | 2022-09-13 | Multi-graph convolution neural network traffic prediction method based on data fusion |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115482666B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117935562B (en) * | 2024-03-22 | 2024-06-07 | 山东双百电子有限公司 | Traffic light control method and system based on deep learning |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109948561A (en) * | 2019-03-25 | 2019-06-28 | 广东石油化工学院 | The method and system that unsupervised image/video pedestrian based on migration network identifies again |
| CN112289034A (en) * | 2020-12-29 | 2021-01-29 | 四川高路交通信息工程有限公司 | Deep neural network robust traffic prediction method based on multi-mode space-time data |
| CN112669606A (en) * | 2020-12-24 | 2021-04-16 | 西安电子科技大学 | Traffic flow prediction method for training convolutional neural network by utilizing dynamic space-time diagram |
| CN113450562A (en) * | 2021-05-08 | 2021-09-28 | 浙江工业大学 | Road network traffic state discrimination method based on clustering and graph convolution network |
| CN113657645A (en) * | 2021-07-15 | 2021-11-16 | 杭州电子科技大学 | Attention mechanism-based space-time connection enhanced 3DCNN traffic prediction method |
| CN113823090A (en) * | 2021-09-22 | 2021-12-21 | 南通大学 | Traffic flow prediction method based on multi-space-time graph convolution network |
-
2022
- 2022-09-13 CN CN202211110596.1A patent/CN115482666B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109948561A (en) * | 2019-03-25 | 2019-06-28 | 广东石油化工学院 | The method and system that unsupervised image/video pedestrian based on migration network identifies again |
| CN112669606A (en) * | 2020-12-24 | 2021-04-16 | 西安电子科技大学 | Traffic flow prediction method for training convolutional neural network by utilizing dynamic space-time diagram |
| CN112289034A (en) * | 2020-12-29 | 2021-01-29 | 四川高路交通信息工程有限公司 | Deep neural network robust traffic prediction method based on multi-mode space-time data |
| CN113450562A (en) * | 2021-05-08 | 2021-09-28 | 浙江工业大学 | Road network traffic state discrimination method based on clustering and graph convolution network |
| CN113657645A (en) * | 2021-07-15 | 2021-11-16 | 杭州电子科技大学 | Attention mechanism-based space-time connection enhanced 3DCNN traffic prediction method |
| CN113823090A (en) * | 2021-09-22 | 2021-12-21 | 南通大学 | Traffic flow prediction method based on multi-space-time graph convolution network |
Non-Patent Citations (1)
| Title |
|---|
| "The Traffic Flow Prediction Method Using the Incremental Learning-Based CNN-LTSM Model:The Solution of Mobile Application";Yanli Shao 等;《Mobile Information System》;20210601(第4期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115482666A (en) | 2022-12-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110264709B (en) | Method for predicting traffic flow of road based on graph convolution network | |
| CN109670277B (en) | Travel time prediction method based on multi-mode data fusion and multi-model integration | |
| CN109410577B (en) | Self-adaptive traffic control subarea division method based on space data mining | |
| Fang et al. | FTPG: A fine-grained traffic prediction method with graph attention network using big trace data | |
| CN110310474A (en) | A kind of vehicle flowrate prediction technique and device based on space-time residual error network | |
| CN106228808B (en) | City expressway travel time prediction method based on Floating Car space-time grid data | |
| CN109840660A (en) | A kind of vehicular characteristics data processing method and vehicle risk prediction model training method | |
| CN109871876A (en) | A kind of Freeway Conditions identification and prediction technique based on floating car data | |
| CN106384509A (en) | Urban road driving time distribution estimation method considering taxi operation states | |
| CN105513370B (en) | The traffic zone division methods excavated based on sparse license plate identification data | |
| WO2021013190A1 (en) | Meteorological parameter-based high-speed train positioning method and system in navigation blind zone | |
| JP7226855B2 (en) | Calculation method, system and storage medium for train passing through tunnel | |
| CN114944053A (en) | Traffic flow prediction method based on spatio-temporal hypergraph neural network | |
| CN112613225B (en) | Intersection traffic state prediction method based on neural network cell transmission model | |
| CN111291098B (en) | Track mode mining method and device | |
| CN111292534A (en) | Traffic state estimation method based on clustering and deep sequence learning | |
| CN112966853A (en) | Urban road network short-term traffic flow prediction method based on space-time residual error mixed model | |
| CN113435658B (en) | Traffic flow prediction method based on space-time fusion correlation and attention mechanism | |
| CN112732905B (en) | Traffic accident analysis, prevention and control method and system based on knowledge graph | |
| Chen et al. | A multiscale-grid-based stacked bidirectional GRU neural network model for predicting traffic speeds of urban expressways | |
| CN113327417B (en) | Traffic flow prediction method based on 3D dynamic space-time residual convolution associated network | |
| CN112884014A (en) | Traffic speed short-time prediction method based on road section topological structure classification | |
| CN115482666B (en) | Multi-graph convolution neural network traffic prediction method based on data fusion | |
| CN116681176B (en) | Traffic flow prediction method based on clustering and heterogeneous graph neural network | |
| CN112419711B (en) | Closed parking lot parking demand prediction method based on improved GMDH algorithm |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |