CN114781894A - Traffic resource distribution method, system and medium based on confidence graph convolution network - Google Patents

Traffic resource distribution method, system and medium based on confidence graph convolution network Download PDF

Info

Publication number
CN114781894A
CN114781894A CN202210479576.5A CN202210479576A CN114781894A CN 114781894 A CN114781894 A CN 114781894A CN 202210479576 A CN202210479576 A CN 202210479576A CN 114781894 A CN114781894 A CN 114781894A
Authority
CN
China
Prior art keywords
traffic
network
nodes
node
probability
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.)
Pending
Application number
CN202210479576.5A
Other languages
Chinese (zh)
Inventor
任勇
杜冠廷
任艳
徐云龙
陈志峰
胥薇
杨艳红
朱斐
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Suzhou University
Original Assignee
Suzhou University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Suzhou University filed Critical Suzhou University
Priority to CN202210479576.5A priority Critical patent/CN114781894A/en
Publication of CN114781894A publication Critical patent/CN114781894A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G06Q10/0631—Resource planning, allocation, distributing or scheduling for enterprises or organisations
    • 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
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • 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
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • 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
    • G06N3/088—Non-supervised learning, e.g. competitive learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/08—Logistics, e.g. warehousing, loading or distribution; Inventory or stock management
    • G06Q10/083—Shipping
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
    • G06Q50/10—Services
    • G06Q50/26—Government or public services

Abstract

The invention discloses a traffic resource distribution method, a system and a medium based on a confidence graph convolution network, wherein the traffic resource distribution method comprises the steps of modeling a current actual traffic network and a traffic station; carrying out feature extraction on traffic network map data generated by modeling through a map embedding method so as to determine feature extraction vectors; and processing the feature extraction vector based on the deep belief network, and transmitting and updating the feature extraction vector through a graph convolution neural network, so as to update the feature representation of the nodes and change the label type of each node to generate a target topological structure, and further automatically distributing traffic resources to corresponding traffic stations according to the target topological structure to realize the automatic distribution of the traffic resources at one time. The invention can respond to the resource allocation problem of the traffic station in time, thereby realizing more optimal dynamic allocation of traffic resources.

Description

Traffic resource allocation method, system and medium based on confidence map convolution network
Technical Field
The invention relates to the field of traffic resource allocation, in particular to a traffic resource allocation method, a traffic resource allocation system and a traffic resource allocation medium based on a confidence map convolutional network.
Background
Traffic resources broadly refer to the basic design and combination of all transportation modes under various technical conditions in the transportation industry. The optimized configuration of traffic resources is directly considered as an optimized combination of various transportation means infrastructures. The intelligent traffic resource allocation system is firstly established on the basis of real sampling and data analysis, abstracted modeling is carried out on a real scene, and then an advanced analysis and calculation method is used for analyzing the optimal allocation mode of traffic resources on the basis of a modeling result.
In recent years, on a macroscopic level, the transportation industry of China realizes the leap-type development, a transportation network mainly based on public road transportation and high-speed rail inter-city transportation is laid in a large area in the whole country, transportation hubs are built in all large cities, the cross-provincial transportation modes are gradually enriched, and the passenger capacity and the freight capacity are continuously increased for many years. On a microscopic level, each city has independent local traffic resources, such as shared bicycles, urban rails, buses and the like. However, as the size of the traffic resource network is increased, the network topology becomes more complex, which means that how to efficiently allocate traffic resources becomes a difficult problem. The traditional traffic resource allocation can not dynamically capture the change condition of the traffic network, thereby lacking the capability of dynamic adaptive allocation.
The traditional convolutional neural network can only process Euclidean space data such as images, texts, voices and the like, the data in the fields have translation invariance, and the translation invariance means that local structures with the same size can be obtained by taking any pixel point as a center. The translational invariance allows a globally shared convolution kernel to be defined in the input data space, thereby defining a convolution network. The graph data is non-Europe space data, and the local structure of each node of the graph data is different, so that the translation invariance is no longer satisfied, namely the graph data does not have the translation invariance, and therefore basic operators (convolution and pooling) in the traditional convolutional neural network cannot realize the application function on the graph data.
Therefore, there is a need for a traffic resource allocation method, system and medium based on a confidence map convolutional network that can solve the above problems.
Disclosure of Invention
The invention aims to provide a traffic resource allocation method, a traffic resource allocation system and a traffic resource allocation medium based on a confidence map convolutional network, which solve the problem of dynamic resource allocation in a traffic network.
In order to achieve the purpose, the technical scheme adopted by the invention is as follows:
a traffic resource allocation method based on a belief-plot convolutional neural network, which is used for the automatic allocation of traffic resources, and comprises the following steps:
modeling the current actual traffic network and traffic stations, wherein the process comprises converting the actual traffic network into an edge set of traffic network diagram data and converting partial traffic stations into a node set of the traffic network diagram data;
carrying out feature extraction on traffic network map data generated by modeling through a map embedding method so as to determine a feature extraction vector; the characteristic extraction process comprises the steps of obtaining original characteristic representation of the node, inputting the original characteristic representation into a Skigram model, and processing to obtain a final characteristic extraction vector;
processing the feature extraction vector based on a deep belief network, and transmitting and updating the feature extraction vector through a graph convolution neural network, so as to update the feature representation of the nodes and change the label type of each node to generate a target topological structure, and further automatically distributing traffic resources to corresponding traffic stations according to the target topological structure to realize the automatic distribution of the traffic resources for one time; the processing process based on the deep belief network comprises the steps of carrying out unsupervised pre-training based on RBM and carrying out weight initialization through a CD-k method, so that the probability distribution that the hidden layer belief neurons in the deep belief network are started is calculated, and the weight is updated based on the probability distribution; the graph convolution based neural network propagating and updating process includes aggregating a set of nodes of the traffic network graph data by pooled propagation; the label type of the node represents the tension degree of the current traffic resource of the traffic stop corresponding to the node, and when the tension degree is higher, the traffic resource allocated to the corresponding traffic stop is more.
Further, the traffic resource allocation method further comprises the following steps:
after completing the automatic allocation of the traffic resources for one time, updating the resource allocation condition of the current actual traffic network, and performing the automatic allocation of the traffic resources again based on the updated actual traffic network data; the method for updating the resource allocation condition of the current actual traffic network comprises the following steps:
updating the characteristic matrixes of all traffic nodes according to the change condition of the traffic resources at each traffic station in the actual traffic network after automatic distribution so as to obtain new traffic network diagram data;
carrying out Fourier transform and propagation between nodes and neighbor nodes on the new traffic network graph data by using a GCN model, wherein the characteristics of target nodes related to the target topological structure are aggregated according to the characteristics of the neighbor nodes;
and updating the label type of the target node according to the aggregation result, and starting to perform reverse propagation at the same time, thereby updating the label types of the neighbor nodes of the target node.
Further, the method for acquiring the original feature representation of the node comprises the following steps:
repeating the following steps to collect a plurality of sets of node sequences: starting from any node in the graph data and moving among the nodes with equal probability.
Further, the method for processing the original feature representation input into the Skipgram model to obtain a final feature extraction vector includes:
updating partial weight of a hidden layer during back propagation, judging whether to update the weight according to node occurrence probability related to the weight, continuously sampling partial nodes, and controlling the probability of the nodes being sampled;
wherein the probability that the weight is updated is determined according to:
Figure BDA0003627026330000031
wherein, P (v)viIs a node viProbability of occurrence, Z (v)i) Is a node viWeight of (d), Z (v)j) Is a node vjWeight of vi、vjA set of nodes from traffic network map data;
wherein the node occurrence probability is determined according to the following formula:
Figure BDA0003627026330000032
in the formula, P (o)i) For nodes o in the network constructed according to existing weightsiProbability of occurrence, Z (o)i) For nodes o in the network constructed according to the existing weightsiThe weight of (c).
Further, calculating the probability distribution that the hidden layer signaling neuron in the deep confidence network is opened according to the following formula:
Figure BDA0003627026330000033
in the formula (I), the compound is shown in the specification,
Figure BDA0003627026330000034
in order to be a probability distribution,
Figure BDA0003627026330000035
v(0)from a set of nodes of traffic network map data, θ is the learning rate, WjFor the vector matrix to be processed, bjIs the offset.
Further, the deep belief network based process further includes:
and continuing Gibbs sampling (GIBBS sampling) according to the calculated probability distribution, and extracting corresponding values from neurons of a layer in the deep belief network to perform sampling reconstruction, so as to update and obtain new weight and bias, wherein the process is represented by the following formula:
W←W+λ[x(h(0)＝1)|z(0)]z(0)T-x(x(1)＝1|v(1))z(1)T
in the formula, W is a new weight matrix, lambda and T are learning parameters, x is a sampling probability, h(0)Is a hidden layer representation of layer 0, the h(0)The calculation method comprises the steps that after the aggregation operation is finished, results obtained by nodes on different subgraphs are added in the 0 layer, and z is(0)Is the output of layer 0 node level, x(1)Is the sampling probability of layer 1, z(1)V (1) is the output of the level of level 1 nodes, and is the set of level 1 nodes.
Further, the set of nodes of the traffic network map data is aggregated according to the following equation:
Figure BDA0003627026330000041
wherein pool is the pooling operation, s is the s-th layer vector, and θ isThe learning rate, W is the adjacency vector,
Figure BDA0003627026330000042
is a node viK denotes the k-th layer of the neural network, b denotes an offset, viA set of nodes V from traffic network map data.
Further, the updating process based on the graph convolution neural network further includes:
updating the characterization of the node by gradient-dropping a loss function, wherein the loss function is determined by:
Figure BDA0003627026330000043
in the formula, Lg(uz) For the loss function, θ is the learning rate, uz、uvIs a feature vector of the traffic network map data, T is a learning parameter,
Figure BDA0003627026330000044
is a probability distribution.
A traffic resource allocation system based on a belief-plot convolutional neural network, the traffic resource allocation system comprising:
a modeling unit configured to model a current actual traffic network and a traffic stop to generate traffic network map data;
a feature extraction unit configured to perform feature extraction on the traffic network map data generated by modeling by a map embedding method to determine a feature extraction vector;
and the data processing unit is configured to process the feature extraction vector based on a deep belief network, propagate and update the feature extraction vector through a graph convolution neural network, update the feature representation of the nodes and change the label type of each node so as to generate a target topological structure, and automatically allocate traffic resources to corresponding traffic stations according to the target topological structure so as to realize the automatic allocation of primary traffic resources.
A storage medium configured to store a computer program configured to perform the traffic resource allocation method described above.
The invention has the advantages that: the method has the advantages that graph data are used as input, high-precision modeling of the actual traffic network environment is used as a basis, the confidence graph convolutional neural network combining the deep confidence convolutional neural network and the graph convolutional neural network is used for realizing automatic distribution of traffic resources, the distribution efficiency is high, the resource distribution problem of a traffic station can be responded in time, and accordingly more optimal dynamic distribution of the traffic resources is realized.
Drawings
In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings needed to be used in the description of the embodiments will be briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without creative efforts.
Fig. 1 is a schematic diagram of a traffic resource distribution system according to an embodiment of the present invention.
Detailed Description
In order to make the technical solutions of the present invention better understood and more clearly understood by those skilled in the art, the technical solutions of the embodiments of the present invention will be described below in detail and completely with reference to the accompanying drawings. It should be noted that the implementations not shown or described in the drawings are in a form known to those of ordinary skill in the art. Additionally, although examples may be provided herein of parameters including particular values, it should be appreciated that the parameters need not be exactly equal to the respective values, but may approximate the respective values within acceptable error margins or design constraints. It is to be understood that the described embodiments are merely exemplary of the invention, and not necessarily all exemplary embodiments. All other embodiments, which can be obtained by a person skilled in the art without making any creative effort based on the embodiments in the present invention, shall fall within the protection scope of the present invention. In addition, the terms "comprises" and "comprising," and any variations thereof, in the description and claims of this invention are intended to cover a non-exclusive inclusion, such that a process, method, apparatus, article, or device that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed, but may include other steps or elements not expressly listed or inherent to such process, method, article, or device.
In one embodiment of the invention, a traffic resource allocation method based on a belief-map convolutional neural network is provided, and the traffic resource allocation method is used for automatic allocation of traffic resources.
First, the current actual traffic network and traffic stations are modeled. In the embodiment, an actual traffic network is converted to generate an edge set of traffic network map data, and a part of traffic stations are converted to generate a node set of traffic network map data;
secondly, feature extraction is carried out on the traffic network map data generated by modeling through a map embedding method, so that a feature extraction vector is determined. In this embodiment, the process of feature extraction includes obtaining an original feature representation of a node, and the obtaining method includes repeatedly performing a step of starting from any node in graph data and performing equal probability transition between nodes, so as to collect multiple sets of node sequences, and then inputting the original feature representation, that is, the collected node sequences, to a Skipgram model to process so as to obtain a final feature extraction vector, where the specific processing includes updating a partial weight of a hidden layer during back propagation, determining whether to update the weight according to a node occurrence probability related to the weight, continuing to sample a partial node, and controlling a probability that the node is sampled.
And then, processing the feature extraction vector based on the deep belief network, and propagating and updating the feature extraction vector through a graph convolution neural network, so as to update the feature representation of the nodes and change the label type of each node to generate a target topological structure, and further automatically distributing traffic resources to corresponding traffic stations according to the target topological structure to realize the automatic distribution of the traffic resources at one time. In this embodiment, the processing procedure based on the deep belief network includes performing unsupervised pre-training based on RBM and performing weight initialization by using a CD-k method, thereby calculating probability distribution of the hidden layer belief neurons being turned on in the deep belief network, then continuing GIBBS sampling (GIBBS sampling) according to the calculated probability distribution, and extracting corresponding values from neurons in the display layer in the deep belief network to perform sampling reconstruction, thereby updating to obtain new weights and offsets; the graph convolution neural network based propagation and update process includes aggregating a set of nodes of the traffic network graph data by pooled propagation, wherein the update process further includes updating the feature representations of the nodes by gradient descent of a loss function.
Note that the tag type of a node indicates the degree of tension of the current traffic resource of the traffic stop corresponding to the node, and as the degree of tension is higher, the traffic resources allocated to the corresponding traffic stop are more. In this embodiment, the label types of the setting node include 1, 2, and 3, where 1 represents a resource shortage, 2 represents a resource moderation, and 3 represents a resource sufficiency. If the label type of a certain transportation station is 1, the transportation station needs to invest (or more) transportation resources to the transportation station; if the label of the transportation station is 3, it means that no (or reduced) transportation resources are invested into the transportation station. The label type can be set according to the actual application, and the protection scope of the present invention is not limited by the above 1, 2, 3.
Finally, after completing the automatic allocation of the traffic resources for one time, updating the resource allocation condition of the current actual traffic network, and performing the automatic allocation of the traffic resources again based on the updated actual traffic network data; the method for updating the resource allocation condition of the current actual traffic network comprises the following steps:
updating the characteristic matrixes of all traffic nodes according to the change condition of the traffic resources at each traffic station in the actual traffic network after automatic distribution so as to obtain new traffic network diagram data;
carrying out Fourier transform and propagation between the nodes and neighbor nodes on new traffic network graph data by using a GCN model, wherein the characteristics of target nodes related to a target topological structure are aggregated according to the characteristics of the neighbor nodes;
and updating the label type of the target node according to the aggregation result, and starting to perform reverse propagation at the same time, thereby updating the label types of the neighbor nodes of the target node. The above processes are circulated, so that the continuous dynamic automatic allocation of the traffic resources of the traffic station is realized, and the problem of station resource shortage or overflow can be solved in time.
In one embodiment of the present invention, the traffic resource allocation method includes the steps of:
(one) reality modeling
Firstly, a traffic network and important nodes in reality are modeled. According to the structural characteristics of the graph network, the traffic network is abstracted into an edge set E { E } of graph data1,e2,e3,...,enWhere E denotes the total set of edges, E1,e2,e3,……，enRepresenting n edges, e represents any one edge, and important traffic nodes are abstracted into nodes V { V } of graph data1,v2,v3...vnV denotes the total set of nodes, V1,v2,……，vnRepresenting n nodes, with v representing any one node. According to the interworking relation between networks, the nodes have interconnecting edges, and the relation is described by an adjacency matrix W.
The step can obtain network data of the traffic resource network in reality, and obtains a point set, an edge set and an adjacency matrix to describe the traffic resource network.
(II) feature extraction
The traffic resource allocation method uses graph embedding to perform feature extraction on graph data generated by a modeling result.
Firstly, an ANRL model is used for obtaining an original characteristic representation, the operation is specifically that the nodes are moved from any one point in the graph in an equal probability manner, the steps are repeated for a limited time, and finally a series of node sequences are collected. When enough sequences are collected, the sequences are input into a Skigram model in a mode similar to text processing, and finally an embedded vector is obtained.
The traffic resource allocation method introduces negative sampling to accelerate the calculation speed so as to improve the homogeneity of the embedded vector and the performance of structural equivalence, specifically, partial weight of a hidden layer is updated during reverse propagation so as to reduce the calculated amount, meanwhile, whether the weight is updated depends on the occurrence probability of nodes related to the weight, and the probability that the weight related to the node is updated is higher when the occurrence probability of the node is higher. Wherein the probability that the weight is updated is determined according to:
Figure BDA0003627026330000071
wherein, P (v)viIs a node viProbability of occurrence, Z (v)i) Is a node viWeight of (d), Z (v)j) Is a node vjWeight of vi、vjA set of nodes from traffic network map data;
on the basis, sampling is continued on part of nodes in the network so as to save sampling time. In partial sampling, the probability of a node being sampled needs to be controlled, and the higher the frequency of occurrence of the node is, the lower the probability of sampling the node as a starting point is, that is, the node which has been repeatedly sampled does not need to be re-sampled, wherein the probability of occurrence of the node is determined according to the following formula:
Figure BDA0003627026330000072
in the formula, P (o)i) For nodes o in the network constructed according to existing weightsiProbability of occurrence, Z (o)i) For nodes o in the network constructed according to existing weightsiThe weight of (c). Thus, the feature extraction vector W is finally obtained as the extraction result.
The method comprises the steps of carrying out dimensionality reduction extraction on features in the graph network generated in the step (I) and preparing for obtaining a network data representation matrix in the step (III) described below, wherein a feature extraction tool used in the step is an ANRL model, the ANRL model is used for modeling node attributes and relations through a dual-target network, and finally obtained graph embedding vectors are limited by attribute and relation training data to obtain a graph embedding vector integrating attribute and relation information.
(III) obtaining data representation of graph network
Inputting the feature extraction result into a vector result W obtained by deep belief network processing2And fourier transformed.
Firstly, carrying out unsupervised pre-training based on RBM, and carrying out weight initialization by using a CD-k method:
Figure BDA0003627026330000081
where W is a weight vector, a is a bias vector of the visible layer, and b is a bias vector of the hidden layer.
Then, W is assigned to the explicit layer, and the probability that it causes the implicit layer signaling neuron to be turned on is calculated according to the following formula:
Figure BDA0003627026330000082
in the formula (I), the compound is shown in the specification,
Figure BDA0003627026330000083
in order to be a probability distribution,
Figure BDA0003627026330000084
v(0)a set of nodes from traffic network map data, θ is the learning rate, WjFor the vector matrix to be processed, bjIs the offset.
And then, continuing Gibbs sampling (GIBBS sampling) according to the probability distribution obtained by calculation to extract corresponding values from the neurons in the display layer for sampling reconstruction, and updating to obtain new weight and bias:
W←W+λ[x(h(0)＝1)|z(0)]z(0)T-x(x(1)＝1|v(1))z(1)T
in the formula, W is a new weight matrix, lambda and T are learning parameters, x is a sampling probability, h(0)Is a hidden layer representation of layer 0, h(0)The calculation method of (1) is that after the aggregation operation of the layer 0 is finished, the results obtained by the nodes on different subgraphs are added, z is(0)Is the output of layer 0 node level, x(1)Is the sampling probability of layer 1, z(1)And v (1) is the output of the level 1 node, and is the node set of the level 1.
The processing model used in this step is a deep belief network, as shown in fig. 1, the deep belief network is composed of a plurality of neurons, the component elements are RBM restricted Boltzmann machines, the structure of the deep belief network is restricted to two layers, namely a visible layer and a hidden layer, the layers are connected, but no connection exists between units in the layers, and the hidden layer units are trained to capture the correlation of high-order data expressed in the visible layer.
(IV) propagation
In this phase, the target node is propagated, specifically, by a propagation function. The propagation function may be an average propagation or a pooled propagation. In this embodiment, pooling propagation is used to continually aggregate data:
Figure BDA0003627026330000085
wherein pool is the pooling operation, s is the s-th layer vector, θ is the learning rate, W is the adjacent vector,
Figure BDA0003627026330000086
is a node viK denotes the k-th layer of the neural network, b denotes an offset, viA set of nodes V from traffic network map data.
Specifically, first, performing a nonlinear transformation on the adjacent point expression vector of the target vertex, then performing a pooling operation (maximum pooling or mean pooling), splicing the obtained result with the expression vector of the target vertex, and finally performing a nonlinear transformation to obtain a k-th layer expression vector of the target vertex.
(V) update
The updating of the node characteristics is achieved in this phase by gradient descent of the loss function. In this embodiment, the loss function is as follows:
Figure BDA0003627026330000091
in the formula, Lg(uz) For the loss function, θ is the learning rate, uz、uvIs a feature vector of the traffic network map data, T is a learning parameter,
Figure BDA0003627026330000092
is a probability distribution. Continuously updating L using a gradient update approachg(uz) Until convergence, the loss function provided based on this embodiment enables adjacent vertices to have similar vector representations, while allowing the representations of separate vertices to have greater discrimination.
And (4) completing the step (four) and the step (five) through a graph convolution neural network, and mainly realizing the step (four) and the step (five) through a graph convolution neural network structure model. In the process of propagation, the target node continuously updates the feature representation of the node, generalizes the feature sequence of the updated node, and changes the label category of each node. Finally, automatic distribution of traffic resources can be completed through the label type of each node, a reasonable distribution node sequence, namely a target topological structure, is provided, and scientific support is provided for optimal decision of the traffic resources.
Specifically, a traffic resource allocation operation is performed once by taking a certain traffic network G as an example. First, assume that a certain station in the traffic network G has a tag type of 2, which indicates that the node v corresponding to the station has moderate resources. After one round of resource reallocation is carried out, the resource allocation condition of the traffic network G is updated, so that the node characteristic matrix is updated, and at the moment, the graph convolution neural network model starts a new round of node propagation, node characteristic aggregation and node characteristic update process, which specifically comprises the following steps: updating the feature matrix of all traffic nodes in the G according to the change condition of each node resource in the traffic network G; after the characteristic matrix of the traffic node is updated, carrying out Fourier transform on the updated traffic network graph by the graph convolution neural network model, carrying out propagation between the node and a neighbor node, and in the process, aggregating the characteristics of a target node according to the characteristics of the neighbor node; and after the aggregation is finished, the label type of the target node is updated according to the aggregation result, at the moment, the back propagation process starts, and the label types of the neighbor nodes of the target node are also updated. After a round of label updating is finished, the label type of each traffic station represents the tension degree of the current resources, and if the name of a traffic node with a label of 1 is given to represent that the traffic node resources are in short supply, the traffic node needs to be automatically allocated; if the label type of the node v is updated from 2 to 1, the node v is indicated to be in short supply of resources, and therefore the resource allocation work is automatically carried out on the traffic station corresponding to the node v.
In one embodiment of the invention, a traffic resource distribution system based on a belief-map convolutional neural network is provided, and comprises a modeling unit, a feature extraction unit and a data processing unit. Wherein the modeling unit is configured to model a current actual traffic network and traffic stops to generate traffic network map data; the feature extraction unit is configured to perform feature extraction on traffic network map data generated by modeling through a map embedding method to determine a feature extraction vector; the data processing unit is configured to process the feature extraction vectors based on the deep belief network, and to propagate and update through the graph convolution neural network, so as to update the feature representation of the nodes and change the label type of each node to generate a target topological structure, and then automatically allocate traffic resources to corresponding traffic stations according to the target topological structure, so as to realize the automatic allocation of primary traffic resources. Specifically, the specific operation of the traffic resource allocation system is described with reference to fig. 1 and the above detailed description of the traffic resource allocation method, which is not repeated herein.
The idea of the embodiment of the traffic resource allocation system and the working process of the traffic resource allocation method in the embodiment belong to the same idea, and the entire content of the embodiment of the traffic resource allocation method is incorporated into the embodiment of the traffic resource allocation system by full reference, which is not described again.
In one embodiment of the present invention, a storage medium is provided that is configured to store a computer program configured to perform the traffic resource allocation method described above.
The idea of the embodiment of the storage medium and the working process of the traffic resource allocation method in the above embodiment belong to the same idea, and the entire contents of the embodiment of the traffic resource allocation method are incorporated into the embodiment of the storage medium by way of full-text reference, which is not described again.
The invention takes graph data as input, takes high-precision modeling of a real environment as a basis, and uses a confidence graph convolution neural network combining a deep confidence convolution neural network and a graph convolution neural network to realize automatic distribution of traffic resources. By providing a modeling method for graph data of real dynamic change, the dynamic change of all traffic resources is abstracted into vector representation, and intelligent automatic allocation of the traffic resources is realized through calculation and fitting of a neural network. The method specifically utilizes a belief map convolutional neural network to dynamically capture the resource change condition in the real traffic network, and realizes the automatic distribution of traffic resources on the basis of the resource change condition. The invention mainly realizes the capture of traffic network resources through traffic network abstraction, namely, an abstract graph structure is created, traffic stations are taken as nodes, the flow of traffic resources among the traffic stations is abstracted as edges, the resource distribution condition is taken as node characteristics to be assigned, the resource change condition is captured through the change of the node characteristics, and the label type is set by taking the shortage degree of the traffic resources as a scale. The belief graph convolutional neural network performs learning prediction, and the node labels can be updated according to the neighbor node feature matrix of the target node, so that the resource distribution condition of the traffic station is updated.
The above description is only for the preferred embodiment of the present invention and is not intended to limit the scope of the present invention, and all modifications of equivalent structures and equivalent processes that can be directly or indirectly applied to other related technical fields using the contents of the present specification and the accompanying drawings are included in the scope of the present invention.

Claims (10)

1. A traffic resource distribution method based on a belief plot convolutional neural network is characterized in that the traffic resource distribution method is used for automatically distributing traffic resources, and the traffic resource distribution method comprises the following steps:
modeling the current actual traffic network and traffic stations, wherein the process comprises converting the actual traffic network into an edge set of traffic network diagram data and converting partial traffic stations into a node set of the traffic network diagram data;
carrying out feature extraction on traffic network map data generated by modeling through a map embedding method so as to determine feature extraction vectors; the characteristic extraction process comprises the steps of obtaining original characteristic representation of the node, and inputting the original characteristic representation into a Skipgm model to be processed to obtain a final characteristic extraction vector;
processing the feature extraction vector based on a deep belief network, and transmitting and updating the feature extraction vector through a graph convolution neural network, so as to update the feature representation of the nodes and change the label type of each node to generate a target topological structure, and then automatically allocating traffic resources to corresponding traffic stations according to the target topological structure to realize the automatic allocation of primary traffic resources; the processing process based on the deep belief network comprises the steps of carrying out unsupervised pre-training based on RBM and carrying out weight initialization through a CD-k method, so that the probability distribution that the hidden layer belief neurons in the deep belief network are started is calculated, and the weight is updated based on the probability distribution; the graph convolution based neural network propagation and update process includes aggregating a set of nodes of the traffic network graph data by pooled propagation; the label type of the node represents the tension degree of the current traffic resource of the traffic stop corresponding to the node, and when the tension degree is higher, the traffic resource allocated to the corresponding traffic stop is more.
2. The traffic resource allocation method based on the belief graph convolutional neural network of claim 1, wherein the traffic resource allocation method further comprises:
after completing the automatic allocation of the traffic resources for one time, updating the resource allocation condition of the current actual traffic network, and performing the automatic allocation of the traffic resources again based on the updated actual traffic network data; the method for updating the resource allocation condition of the current actual traffic network comprises the following steps:
updating the characteristic matrixes of all traffic nodes according to the change condition of the traffic resources at each traffic station in the actual traffic network after automatic distribution so as to obtain new traffic network diagram data;
carrying out Fourier transform and propagation between nodes and neighbor nodes on the new traffic network graph data by using a GCN model, wherein the characteristics of target nodes related to the target topological structure are aggregated according to the characteristics of the neighbor nodes;
and updating the label type of the target node according to the aggregation result, and starting to perform reverse propagation at the same time, thereby updating the label types of the neighbor nodes of the target node.
3. The method of claim 1, wherein the method of obtaining raw feature representations of nodes comprises:
repeating the following steps to acquire a plurality of sets of node sequences: starting from any node in the graph data and moving among the nodes in an equal probability mode.
4. The traffic resource allocation method based on the belief diagram convolutional neural network as set forth in claim 1, wherein the method for processing the raw feature representation input into a Skipgram model to obtain a final feature extraction vector comprises:
updating partial weight of a hidden layer during back propagation, judging whether to update the weight according to node occurrence probability related to the weight, continuously sampling partial nodes, and controlling the probability of the nodes being sampled;
wherein the probability that the weight is updated is determined according to:
Figure FDA0003627026320000021
wherein, P (v)viIs a node viProbability of occurrence, Z (v)i) Is a node viWeight of (d), Z (v)j) Is a node vjWeight of vi、vjA set of nodes from traffic network map data;
wherein the node occurrence probability is determined according to the following formula:
Figure FDA0003627026320000022
in the formula, P (o)i) For nodes o in the network constructed according to the existing weightsiProbability of occurrence, Z (o)i) For nodes o in the network constructed according to the existing weightsiThe weight of (c).
5. The traffic resource allocation method based on the belief graph convolutional neural network of claim 1, wherein the probability distribution that the hidden layer belief neurons in the deep belief network are turned on is calculated according to the following formula:
Figure FDA0003627026320000023
in the formula (I), the compound is shown in the specification,
Figure FDA0003627026320000024
in order to be a probability distribution,
Figure FDA0003627026320000025
v(0)from a set of nodes of traffic network map data, θ is the learning rate, WjAs a vector matrix to be processed, bjIs the offset.
6. The traffic resource allocation method based on the belief graph convolutional neural network as claimed in claim 1 or 5, wherein the deep belief network based process further comprises:
and continuing Gibbs sampling according to the calculated probability distribution, and extracting corresponding values from neurons of a display layer in the deep belief network to perform sampling reconstruction, so as to update and obtain new weight and bias, wherein the process is represented by the following formula:
W←W+λ[x(h(0)＝1)|z(0)]z(0)T-x(x(1)＝1|v(1))z(1)T
in the formula, W is a new weight matrix, lambda and T are learning parameters, x is a sampling probability, h(0)Is a hidden layer representation of layer 0, the h(0)The computing method comprises the steps of adding results obtained by nodes on different subgraphs after the aggregation operation of the 0 layer is finished, and z(0)Is the output of layer 0 node level, x(1)Is the sampling probability of layer 1, z(1)V (1) is the output of the level of level 1 nodes, and is the set of level 1 nodes.
7. The method of claim 1, wherein the set of nodes of the traffic network map data are aggregated according to the following equation:
Figure FDA0003627026320000031
wherein pool is the pooling operation, s is the s-th layer vector, θ is the learning rate, W is the neighboring vector,
Figure FDA0003627026320000032
is a node viK denotes the k-th layer of the neural network, b is an offset, viA set of nodes V from traffic network map data.
8. The method of claim 1, wherein the updating based on the graph convolution neural network further comprises:
updating the characterization of the nodes by performing a gradient descent on a loss function, wherein the loss function is determined by:
Figure FDA0003627026320000033
in the formula, Lg(uz) For the loss function, θ is the learning rate, uz、uvIs a feature vector of the traffic network map data, T is a learning parameter,
Figure FDA0003627026320000034
is a probability distribution.
9. A traffic resource allocation system based on a belief-plot convolutional neural network, the traffic resource allocation system comprising:
a modeling unit configured to model a current actual traffic network and traffic stations to generate traffic network map data;
a feature extraction unit configured to perform feature extraction on the traffic network map data generated by modeling by a map embedding method to determine a feature extraction vector;
and the data processing unit is configured to process the feature extraction vector based on a deep belief network, propagate and update through a graph convolution neural network, so as to update feature representation of the nodes and change the label type of each node to generate a target topological structure, and then automatically allocate traffic resources to corresponding traffic stations according to the target topological structure to realize automatic allocation of primary traffic resources.
10. A storage medium configured to store a computer program configured to execute the traffic resource allocation method of any one of claims 1 to 8.
CN202210479576.5A 2022-05-05 2022-05-05 Traffic resource distribution method, system and medium based on confidence graph convolution network Pending CN114781894A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210479576.5A CN114781894A (en) 2022-05-05 2022-05-05 Traffic resource distribution method, system and medium based on confidence graph convolution network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210479576.5A CN114781894A (en) 2022-05-05 2022-05-05 Traffic resource distribution method, system and medium based on confidence graph convolution network

Publications (1)

Publication Number Publication Date
CN114781894A true CN114781894A (en) 2022-07-22

Family

ID=82434521

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210479576.5A Pending CN114781894A (en) 2022-05-05 2022-05-05 Traffic resource distribution method, system and medium based on confidence graph convolution network

Country Status (1)

Country Link
CN (1) CN114781894A (en)

Similar Documents

Publication Publication Date Title
CN101315663B (en) Nature scene image classification method based on area dormant semantic characteristic
CN112685504B (en) Production process-oriented distributed migration chart learning method
CN106875320A (en) The efficient visual analysis method of ship aeronautical data under cloud environment
CN109787821B (en) Intelligent prediction method for large-scale mobile client traffic consumption
CN112329973A (en) Space-time load prediction method based on graph neural network and regional gridding
CN112508181A (en) Graph pooling method based on multi-channel mechanism
CN106407379A (en) Hadoop platform based movie recommendation method
CN117237559A (en) Digital twin city-oriented three-dimensional model data intelligent analysis method and system
CN114723037A (en) Heterogeneous graph neural network computing method for aggregating high-order neighbor nodes
CN113380043B (en) Bus arrival time prediction method based on deep neural network calculation
CN113381888B (en) Control method, device, equipment and storage medium
CN113449878A (en) Data distributed incremental learning method, system, equipment and storage medium
CN112487724A (en) Urban dynamic expansion simulation method based on partition and improved CNN-CA model
CN114781894A (en) Traffic resource distribution method, system and medium based on confidence graph convolution network
CN115423162A (en) Traffic flow prediction method and device, electronic equipment and storage medium
CN107122472A (en) Extensive unstructured data extracting method, its system, DDM platform
CN115907000A (en) Small sample learning method for optimal power flow prediction of power system
CN115080795A (en) Multi-charging-station cooperative load prediction method and device
CN115359250A (en) Cross-domain small sample image semantic segmentation method based on memory mechanism
CN113033653A (en) Edge-cloud collaborative deep neural network model training method
CN112465066A (en) Graph classification method based on clique matching and hierarchical pooling
CN114826921B (en) Dynamic network resource allocation method, system and medium based on sampling subgraph
CN113743168B (en) Urban flyer identification method based on micro-depth neural network search
CN109919203A (en) A kind of data classification method and device based on Discrete Dynamic mechanism
CN117911662B (en) Digital twin scene semantic segmentation method and system based on depth hough voting

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