Traffic resource dynamic allocation method based on confidence map convolutional network

Abstract

The invention discloses a traffic resource dynamic allocation method based on a confidence map convolutional network. The traffic resource dynamic allocation method comprises the following steps: modeling a current actual traffic network and traffic stations; performing feature extraction on traffic network graph data generated by modeling through a graph embedding method to determine a feature extraction vector; processing the feature extraction vector based on a deep belief network, and spreading and updating through a graph convolutional neural network, thereby updating the feature representation of the node and changing the label type, used for representing the tension degree of traffic resources, of each node to generate a target topological structure; and automatically allocating traffic resources to the corresponding traffic stations according to the target topological structure so as to realize one-time automatic allocation of the traffic resources, and performing automatic allocation of the traffic resources again based on the updated actual traffic network data. According to the invention, the resource allocation problem of the traffic station can be timely responded, so that the traffic resources can be better and dynamically allocated.

Description

Technical Field

[0001] This invention relates to the field of traffic resource allocation, and in particular to a method for dynamic allocation of traffic resources based on confidence graph convolutional networks. Background Technology

[0002] In a broad sense, transportation resources refer to the basic design and combination of all modes of transportation under various technological conditions within the transportation industry. The optimal allocation of transportation resources is directly considered as the optimal combination of infrastructure for various transportation modes. Intelligent transportation resource allocation systems are first established based on real-world sampling and data analysis. This involves abstracting and modeling real-world scenarios, and then using advanced analytical algorithms based on the modeling results to analyze the optimal allocation of transportation resources.

[0003] In recent years, at the macro level, my country's transportation industry has achieved leapfrog development. A transportation network primarily based on public roads and high-speed rail intercity transport has been widely deployed across the country. Major cities have established transportation hubs, and inter-provincial transportation options have become increasingly diverse. Passenger and freight volumes have both achieved continuous year-on-year growth for many years. At the micro level, each city possesses independent local transportation resources, such as shared bicycles, urban rail transit, and buses. However, with the expansion of the transportation resource network and its increasing complexity, the efficient allocation of transportation resources has become a challenge. Traditional transportation resource allocation methods cannot dynamically capture changes in the transportation network, thus lacking the ability for dynamic adaptive allocation.

[0004] Traditional convolutional neural networks (CNNs) can only process Euclidean space data, such as images, text, and speech. This type of data possesses translation invariance, meaning that local structures of the same size can be obtained centered at any pixel. Translation invariance allows for the definition of globally shared convolutional kernels in the input data space, thus defining the convolutional network. However, graph data is non-Euclidean space data, where the local structure of each node is different, making translation invariance no longer satisfied. Therefore, graph data lacks translation invariance, and the basic operators in traditional CNNs (convolution and pooling) cannot be applied to graph data. Summary of the Invention

[0005] The purpose of this invention is to provide a method for dynamic allocation of transportation resources based on confidence graph convolutional networks to address the problem of dynamic resource allocation in transportation networks.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0007] A method for dynamic allocation of traffic resources based on a confidence graph convolutional neural network is provided for the automated allocation of traffic resources. The method includes:

[0008] Modeling the current actual traffic network and traffic stations involves converting the actual traffic network into an edge set of traffic network graph data and converting some traffic stations into a node set of traffic network graph data.

[0009] The feature extraction vector is determined by using graph embedding methods to extract features from the traffic network graph data generated by modeling.

[0010] The feature extraction vector is processed based on a deep belief network and then propagated and updated through a graph convolutional neural network to update the feature representation of the nodes and change the label type of each node to generate the target topology. The label type of the node represents the current traffic resource tension of the traffic station corresponding to that node. When the tension is higher, more traffic resources are allocated to the corresponding traffic station.

[0011] Traffic resources are automatically allocated to the corresponding traffic stations according to the target topology, so as to realize the automatic allocation of traffic resources.

[0012] After completing an automated allocation of traffic resources, the current resource allocation status of the actual traffic network is updated, and the automated allocation of traffic resources is performed again based on the updated actual traffic network data.

[0013] Furthermore, following any one or a combination of the aforementioned technical solutions, the method also includes updating the resource allocation of the current actual transportation network in the following ways:

[0014] The feature matrix of all traffic nodes is updated based on the changes in traffic resources at each traffic station in the actual traffic network after the automated allocation, so as to obtain new traffic network map data.

[0015] The new traffic network graph data is subjected to Fourier transform and propagation between nodes and neighboring nodes, wherein the features of the target node related to the target topology are aggregated based on the features of its neighboring nodes;

[0016] The label type of the target node is updated based on the aggregation result, and backpropagation is started to update the label types of the target node's neighboring nodes.

[0017] Furthermore, following any one or a combination of the aforementioned technical solutions, feature extraction is performed on the traffic network graph data to determine the feature extraction vector, including: obtaining the original feature representation of the node, and inputting the original feature representation into a Skipgram model for processing to obtain the final feature extraction vector; wherein, the original feature representation of the node is obtained by repeatedly starting from any node in the graph data and transferring between the nodes with equal probability to collect multiple sets of node sequences;

[0018] And / or,

[0019] The process of processing the feature extraction vector based on the deep belief network includes:

[0020] Unsupervised pre-training is performed based on RBM and weights are initialized using the CD-k method to calculate the probability distribution of the activation of the hidden layer confidence neurons in the deep belief network, and then the weights are updated based on the probability distribution.

[0021] The process of propagation and updating based on graph convolutional neural networks includes aggregating the node set of the traffic network graph data through pooling propagation.

[0022] Furthermore, following any one or a combination of the aforementioned technical solutions, the method for processing the original feature representation into a Skipgram model to obtain the final feature extraction vector includes:

[0023] During backpropagation, some weights of the hidden layer are updated, and the probability of updating the weights is determined based on the occurrence probability of nodes associated with the weights. The higher the occurrence probability of a node, the higher the probability that the weight associated with it will be updated. Some nodes are sampled, and the probability of the node being sampled is controlled. In partial sampling, the higher the frequency of a node, the higher the probability that it will be used as the starting point for sampling.

[0024] Furthermore, following any one or a combination of the aforementioned technical solutions, the probability of the weight being updated is determined according to the following formula:

[0025]

[0026] In the formula, P(v) vi For node v i The probability of occurrence, Z(v) i ) is node v i The weights, Z(v) j ) is node v j The weights, v i v j A set of nodes derived from traffic network graph data;

[0027] The probability of the occurrence of the node is determined according to the following formula:

[0028]

[0029] In the formula, P(o i ) represents node o in the network constructed based on existing weights. i The probability of occurrence, Z(o) i ) represents node o in the network constructed based on existing weights. i The weight.

[0030] Furthermore, following any one or a combination of the aforementioned technical solutions, the probability distribution of the activation of hidden layer confidence neurons in the deep belief network is calculated according to the following formula:

[0031]

[0032] In the formula, For probability distribution, v (0) The set of nodes derived from traffic network graph data, where θ is the learning rate and W is the learning rate. j Let b be the vector matrix to be processed. j This is the bias value;

[0033] And / or,

[0034] The node set of the traffic network graph data is aggregated according to the following formula:

[0035]

[0036] In the formula, pool represents the pooling operation, s is the vector of the s-th layer, θ is the learning rate, and W is the adjacency vector. For node v i The prediction result, k represents the k-th layer of the neural network, b is the offset, v i The set of nodes V comes from the traffic network graph data.

[0037] Furthermore, following any one or a combination of the aforementioned technical solutions, the deep belief network-based processing procedure further includes:

[0038] Based on the calculated probability distribution, Gibbs sampling continues, and corresponding values ​​are extracted from the neurons of the explicit layers in the deep belief network for sampling and reconstruction, thereby updating and obtaining new weights and biases. The process is represented by the following formula:

[0039] W←W+λ[x(h (0) =1)|z (0) ]z (0)T -x(x (1) =1|v(1))z(1)T

[0040] In the formula, W is the new weight matrix, λ and T are the learning parameters, x is the sampling probability, and h is the weight matrix. (0) For the hidden layer representation of layer 0, the h (0) The calculation method includes summing the results obtained by the node on different subgraphs after the aggregation operation at level 0, z. (0) For the output at the level 0 node, x (1) z is the sampling probability of the first layer. (1) v(1) is the output of the first-level node, and v(1) is the set of nodes in the first level.

[0041] Furthermore, following any one or a combination of the aforementioned technical solutions, the process of updating based on graph convolutional neural networks further includes:

[0042] The feature representation of the node is updated by applying gradient descent to the loss function, wherein the loss function is determined by the following equation:

[0043]

[0044] In the formula, L g (u z Let θ be the loss function, θ be the learning rate, and u be the learning rate. z u v Let T be the feature vector of the traffic network graph data, and T be the learning parameters. It represents a probability distribution.

[0045] According to another aspect of the present invention, a traffic resource dynamic allocation system based on a confidence graph convolutional neural network is provided. The traffic resource dynamic allocation system allocates traffic resources based on any one or a combination of the above technical solutions. The traffic resource dynamic allocation system includes:

[0046] The modeling unit is configured to model the current actual traffic network and traffic stations to generate traffic network map data;

[0047] The feature extraction unit is configured to extract features from the modeled traffic network graph data using a graph embedding method to determine the feature extraction vector.

[0048] The data processing unit is configured to process the feature extraction vector based on a deep belief network and propagate and update it through a graph convolutional neural network, thereby updating the feature representation of the node and changing the label type of each node to generate a target topology. Then, based on the target topology, traffic resources are automatically allocated to the corresponding traffic stations to achieve automated allocation of traffic resources.

[0049] According to another aspect of the present invention, a storage medium is provided, the storage medium being configured to store a computer program, the computer program being configured to execute the dynamic allocation method for traffic resources as described in any one or a combination of the above technical solutions.

[0050] The advantages of this invention are: it takes graph data as input, is based on high-precision modeling of the real traffic network environment, and uses a confidence graph convolutional neural network that combines deep confidence convolutional neural networks and graph convolutional neural networks to realize the automatic allocation of traffic resources. The allocation efficiency is high, and it can respond to the resource allocation problem of traffic stations in a timely manner, thereby achieving a better dynamic allocation of traffic resources. Attached Figure Description

[0051] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0052] Figure 1 This is a schematic diagram of the dynamic allocation system for traffic resources provided in an embodiment of the present invention. Detailed Implementation

[0053] To enable those skilled in the art to better understand the present invention and its objectives, technical solutions, and advantages, the technical solutions in the embodiments of the present invention are clearly and completely described below with reference to specific embodiments and accompanying drawings. It should be noted that implementations not illustrated or described in the accompanying drawings are forms known to those skilled in the art. Furthermore, while this document provides examples of parameters containing specific values, it should be understood that the parameters need not be exactly equal to the corresponding values, but can approximate the corresponding values ​​within acceptable error tolerances or design constraints. Obviously, the described embodiments are merely some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort should fall within the scope of protection of the present invention. In addition, the terms "comprising" and "having," and any variations thereof, in the specification and claims of this invention are intended to cover non-exclusive inclusion. For example, a process, method, apparatus, product, or device that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to these processes, methods, products, or devices.

[0054] In one embodiment of the present invention, a method for dynamic allocation of traffic resources based on a confidence graph convolutional neural network is provided. This method is used for the automated allocation of traffic resources.

[0055] First, the current actual traffic network and traffic stations are modeled. In this embodiment, the actual traffic network is transformed into an edge set of traffic network graph data, and some traffic stations are transformed into a node set of traffic network graph data.

[0056] Secondly, feature extraction is performed on the traffic network graph data generated by the modeling using graph embedding methods to determine the feature extraction vector. In this embodiment, the feature extraction process includes obtaining the original feature representation of the nodes. This method involves repeatedly starting from any node in the graph data and transferring between nodes with equal probability to collect multiple sets of node sequences. Then, the original feature representation, i.e., the collected node sequences, is input into the Skipgram model for processing to obtain the final feature extraction vector. The specific processing includes updating some weights of the hidden layer during backpropagation, determining whether to update the weights based on the probability of node occurrence related to the weights, continuing to sample some nodes, and controlling the probability of nodes being sampled.

[0057] Then, the feature extraction vectors are processed based on a deep belief network and propagated and updated through a graph convolutional neural network, thereby updating the feature representation of the nodes and changing the label type of each node to generate the target topology. Based on the target topology, traffic resources are automatically allocated to the corresponding traffic stations to achieve automated allocation of traffic resources. In this embodiment, the deep belief network processing includes unsupervised pre-training based on RBM and weight initialization using the CD-k method to calculate the probability distribution of the activation of confidence neurons in the hidden layers of the deep belief network. Then, Gibbs sampling is performed based on the calculated probability distribution, and corresponding values ​​are extracted from the neurons in the explicit layers of the deep belief network for sampling and reconstruction, thereby updating the new weights and biases. The propagation and update process based on the graph convolutional neural network includes aggregating the node set of the traffic network graph data through pooling propagation. The update process also includes updating the feature representation of the nodes by applying gradient descent to the loss function.

[0058] It should be noted that the label type of a node indicates the current level of traffic resource scarcity at the corresponding transportation station. The higher the scarcity level, the more traffic resources are allocated to the corresponding transportation station. In this embodiment, the label types for nodes are set to 1, 2, and 3, where 1 represents scarce resources, 2 represents moderate resources, and 3 represents abundant resources. If a transportation station's label type is 1, it indicates that (or more) traffic resources need to be allocated to that station; if the label type is 3, it indicates that (or less) traffic resources need to be allocated to that station. The label types can be set according to actual applications and are not limited to the scope of protection of this invention by the above-mentioned 1, 2, and 3.

[0059] Finally, after completing one round of automated traffic resource allocation, the current resource allocation status of the actual traffic network is updated, and automated traffic resource allocation is performed again based on the updated actual traffic network data. The methods for updating the current resource allocation status of the actual traffic network include:

[0060] The feature matrix of all traffic nodes is updated based on the changes in traffic resources at each traffic station in the actual traffic network after automatic allocation, so as to obtain new traffic network map data.

[0061] The GCN network model is used to perform Fourier transform on new traffic network graph data and propagation between nodes and neighboring nodes. The features of target nodes related to the target topology are aggregated based on the features of their neighboring nodes.

[0062] Based on the aggregation results, the label type of the target node is updated, and backpropagation begins to update the label types of the target node's neighboring nodes. This process is repeated to achieve continuous dynamic and automated allocation of traffic resources at transportation stations, effectively addressing issues of resource shortages or overflows.

[0063] In one embodiment of the present invention, the dynamic allocation method for traffic resources includes the following steps:

[0064] (I) Reality Modeling

[0065] First, we model the real-world transportation network and key nodes. Based on the structural characteristics of graph networks, the transportation network will be abstracted into a graph data edge set E{e1,e2,e3,...,e...}. n}, where E represents the total set of edges, e1, e2, e3, ..., e n Let n be the number of edges, and e represent any edge. Important traffic nodes will be abstracted as nodes V{v1,v2,v3...v...} in the graph data. n}, where V represents the total set of nodes, v1, v2, ..., v nLet represent n nodes, and v represent any one of the nodes. Based on the interconnectivity of the network, there will be edges connecting the nodes, which are described using an adjacency matrix W.

[0066] This step can obtain network data of the real-world transportation resource network and derive the set of points, the set of edges, and the adjacency matrix to describe the transportation resource network.

[0067] (II) Feature Extraction

[0068] This method for dynamic allocation of transportation resources uses graph embedding to extract features from the graph data generated by the modeling results.

[0069] First, an ANRL model is used to obtain the original feature representation. Specifically, starting from any point in the graph, the nodes are transitioned with equal probability. This process is repeated a finite number of times to collect a series of node sequences. Once enough sequences have been collected, they are input into a Skipgram model in a manner similar to text processing to obtain the final embedding vectors.

[0070] This dynamic allocation method for traffic resources introduces negative sampling to accelerate computation and improve the homogeneity and structural equivalence of the embedded vectors. Specifically, during backpropagation, some weights of the hidden layer are updated to reduce computational load. Furthermore, whether a weight is updated depends on the probability of the occurrence of the node associated with that weight; the higher the probability of a node's occurrence, the greater the probability of its associated weight being updated. The probability of a weight being updated is determined by the following formula:

[0071]

[0072] In the formula, P(v) vi For node v i The probability of occurrence, Z(v) i ) is node v i The weights, Z(v) j ) is node v j The weights, v i v j A set of nodes derived from traffic network graph data;

[0073] Based on this, sampling continues on a subset of nodes in the network to save sampling time. In partial sampling, the probability of a node being sampled needs to be controlled. The higher the frequency of a node's occurrence, the lower the probability of using it as a starting point for sampling. That is, nodes that have already been repeatedly sampled do not need to be resampled. The probability of a node's occurrence is determined by the following formula:

[0074]

[0075] In the formula, P(o i) represents node o in the network constructed based on existing weights. i The probability of occurrence, Z(o) i ) represents node o in the network constructed based on existing weights. i The weights are determined. Thus, the final feature extraction vector W is obtained as the extraction result.

[0076] This step involves dimensionality reduction and feature extraction from the graph network generated in step (i), preparing for obtaining the network data representation matrix in step (iii) below. The feature extraction tool used in this step is the ANRL model. The ANRL model models node attributes and relationships through a dual-objective network. The resulting graph embedding vector is constrained by the attribute and relationship training data, resulting in a graph embedding vector that integrates attribute and relationship information.

[0077] (III) Obtaining the data representation of the graph network

[0078] The feature extraction results are input into the vector result W2 obtained by deep belief network processing and then subjected to Fourier transform.

[0079] First, unsupervised pre-training based on RBM is performed, and the weights are initialized using the CD-k method:

[0080]

[0081] In the formula, W is the weight vector, a is the bias vector of the visible layer, and b is the bias vector of the hidden layer.

[0082] Then, W is assigned to the explicit layer, and the probability that it makes the hidden layer confidence neurons turn on is calculated according to the following formula:

[0083]

[0084] In the formula, For probability distribution, v (0) The set of nodes derived from traffic network graph data, where θ is the learning rate and W is the learning rate. j Let b be the vector matrix to be processed. j This is the bias value.

[0085] Then, based on the calculated probability distribution, Gibbs sampling is performed to extract corresponding values ​​from the neurons in the visible layer for reconstruction, and new weights and biases are obtained.

[0086] W←W+λ[x(h (0) =1)|z (0) ]z (0)T -x(x (1) =1|v(1))z (1)T

[0087] In the formula, W is the new weight matrix, λ and T are the learning parameters, x is the sampling probability, and h is the weight matrix. (0) h represents the hidden layer of layer 0. (0) The calculation method is as follows: after the aggregation operation is completed, the results obtained by the node on different subgraphs are added together, z. (0) For the output at the level 0 node, x (1) z is the sampling probability of the first layer. (1) v(1) is the output of the first-level node, and v(1) is the set of nodes in the first level.

[0088] The processing model used in this step is a deep belief network, such as... Figure 1 As shown, this deep belief network consists of several neurons, and the constituent element is a restricted Boltzmann machine (RBM). The deep belief network structure is restricted to two layers: a visible layer and a hidden layer. There are connections between layers, but no connections between units within a layer. The hidden layer units are trained to capture the correlation of high-order data represented in the visible layer.

[0089] (iv) Dissemination

[0090] In this stage, the data is propagated to the target node using a propagation function. This function can be either average propagation or pooling propagation. In this embodiment, pooling propagation is used to continuously aggregate the data.

[0091]

[0092] In the formula, pool represents the pooling operation, s is the vector of the s-th layer, θ is the learning rate, and W is the adjacency vector. For node v i The prediction result, k indicates the k-th layer of the neural network, b is the offset, v i The set of nodes V comes from the traffic network graph data.

[0093] Specifically, a nonlinear transformation is first performed on the neighboring vertex representation vector of the target vertex, followed by a pooling operation (max pooling or mean pooling). The result is then concatenated with the representation vector of the target vertex, and finally, a nonlinear transformation is performed to obtain the k-th layer representation vector of the target vertex.

[0094] (V) Update

[0095] In this stage, node features are updated by applying gradient descent to the loss function. In this embodiment, the loss function is as follows:

[0096]

[0097] In the formula, L g (u zLet θ be the loss function, θ be the learning rate, and u be the learning rate. z u v Let T be the feature vector of the traffic network graph data, and T be the learning parameters. Let L be a probability distribution. Use gradient update to continuously update L. g (u z Until convergence, the loss function provided in this embodiment can make neighboring vertices have similar vector representations, while making the representations of separated vertices have greater discriminative power.

[0098] Steps (iv) and (v) are accomplished using a graph convolutional neural network (GCNN), primarily through a GCNN architecture model. During propagation, the target node continuously updates its feature representations, generalizes and updates the feature sequences of the nodes, and changes the label category of each node. Ultimately, based on the label type of each node, automated allocation of traffic resources can be achieved, providing a reasonable sequence of allocated nodes, i.e., the target topology, thus offering scientific support for optimal traffic resource decision-making.

[0099] Specifically, let's take a traffic network G as an example to perform a traffic resource allocation operation. First, assume that the label type of a certain station in the traffic network G is 2, indicating that the resources of the node v corresponding to that station are moderate. After a round of resource reallocation, the resource allocation of the traffic network G is updated, thereby updating the node feature matrix. At this time, the graph convolutional neural network model starts a new round of node propagation, node feature aggregation, and node feature update process, specifically as follows: The feature matrix of all traffic nodes in G is updated according to the changes in the resources of each node in the traffic network G; after the feature matrix of the traffic nodes is updated, the graph convolutional neural network model performs Fourier transform on the updated traffic network graph and propagates between nodes and their neighbors. During this process, the features of the target node are aggregated according to the features of its neighbors; after the aggregation is completed, the label type of the target node is updated according to the aggregation result. At this time, the backpropagation process begins, and the label types of the target node's neighbors are also updated accordingly. After a round of label updates, the label type of each transportation station represents its current resource scarcity. If a transportation node name is given with a label of 1, it means that this transportation node is short of resources, and resources need to be automatically allocated to it. If the label type of node v is updated from 2 to 1, it means that node v is short of resources, and resource allocation work is automatically carried out for the transportation station corresponding to node v.

[0100] In one embodiment of the present invention, a dynamic traffic resource allocation system based on a confidence graph convolutional neural network is provided. The system includes a modeling unit, a feature extraction unit, and a data processing unit. The modeling unit is configured to model the current actual traffic network and traffic stations to generate traffic network graph data. The feature extraction unit is configured to extract features from the generated traffic network graph data using a graph embedding method to determine feature extraction vectors. The data processing unit is configured to process the feature extraction vectors based on a deep belief network and propagate and update them through a graph convolutional neural network, thereby updating the feature representations of nodes and changing the label type of each node to generate a target topology. Then, based on the target topology, traffic resources are automatically allocated to the corresponding traffic stations to achieve automated traffic resource allocation. Specifically, the specific working method of this dynamic traffic resource allocation system is combined with... Figure 1 The specific explanation of this dynamic allocation method of transportation resources is provided above and will not be repeated here.

[0101] The idea behind this embodiment of the dynamic allocation system for traffic resources is the same as the working process of the dynamic allocation method for traffic resources in the above embodiments. All the contents of the above embodiments of the dynamic allocation method for traffic resources are incorporated into this embodiment of the dynamic allocation system for traffic resources by means of full reference, and will not be repeated here.

[0102] In one embodiment of the present invention, a storage medium is provided, the storage medium being configured to store a computer program, the computer program being configured to execute the above-described dynamic allocation method for traffic resources.

[0103] The concept of the storage medium embodiment is the same as the working process of the dynamic allocation method of traffic resources in the above embodiments. All the contents of the above dynamic allocation method of traffic resources embodiment are incorporated into this storage medium embodiment by means of full reference, and will not be repeated.

[0104] This invention uses graph data as input and high-precision modeling of the real environment as a foundation. It employs a confidence graph convolutional neural network (CNN), combining deep confidence convolutional neural networks and graph convolutional neural networks, to automate the allocation of traffic resources. By providing a modeling method for dynamically changing graph data, it abstracts all dynamic changes in traffic resources into vector representations. Intelligent and automatic allocation of traffic resources is achieved through neural network computation and fitting. Specifically, the confidence graph CNN dynamically captures resource changes in the real-world traffic network, and uses this as the basis for automated allocation. This invention primarily captures traffic network resources through traffic network abstraction. It creates an abstract graph structure with traffic stations as nodes, and the flow of traffic resources between stations is abstracted as edges. Resource distribution is assigned as node features, and changes in node features capture resource changes. Label types are set using the degree of traffic resource scarcity as a scale. The confidence graph CNN performs learning and prediction, updating node labels based on the feature matrix of the target node's neighbor nodes, thereby updating the resource distribution of traffic stations.

[0105] The above description is merely a preferred embodiment of the present invention and does not limit its patent scope. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, whether directly or indirectly applied to other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method for dynamic allocation of traffic resources based on a confidence graph convolutional neural network, characterized in that, The method for automating the allocation of traffic resources includes: Modeling the current actual traffic network and traffic stations involves converting the actual traffic network into an edge set of traffic network graph data and converting some traffic stations into a node set of traffic network graph data. The feature extraction vector is determined by using graph embedding methods to extract features from the traffic network graph data generated by modeling. The feature extraction vector is processed based on a deep belief network and then propagated and updated through a graph convolutional neural network to update the feature representation of the nodes and change the label type of each node to generate the target topology. The label type of the node represents the current traffic resource tension of the traffic station corresponding to that node. When the tension is higher, more traffic resources are allocated to the corresponding traffic station. Traffic resources are automatically allocated to the corresponding traffic stations according to the target topology, so as to realize the automatic allocation of traffic resources. After completing an automated allocation of traffic resources, the current resource allocation status of the actual traffic network is updated, and the automated allocation of traffic resources is performed again based on the updated actual traffic network data.

2. The method for dynamic allocation of traffic resources based on a confidence graph convolutional neural network according to claim 1, characterized in that, This also includes updating the current resource allocation of the actual transportation network in the following ways: The feature matrix of all traffic nodes is updated based on the changes in traffic resources at each traffic station in the actual traffic network after the automated allocation, so as to obtain new traffic network map data. The new traffic network graph data is subjected to Fourier transform and propagation between nodes and neighboring nodes, wherein the features of the target node related to the target topology are aggregated based on the features of its neighboring nodes; The label type of the target node is updated based on the aggregation result, and backpropagation is started to update the label types of the target node's neighboring nodes.

3. The method for dynamic allocation of traffic resources based on a confidence graph convolutional neural network according to claim 1, characterized in that, Feature extraction of the traffic network graph data to determine the feature extraction vector includes: obtaining the original feature representation of the node and inputting the original feature representation into the Skipgram model for processing to obtain the final feature extraction vector; wherein, the original feature representation of the node is obtained by repeatedly starting from any node in the graph data and transferring between the nodes with equal probability to collect multiple sets of node sequences; And / or, The process of processing the feature extraction vector based on the deep belief network includes: Unsupervised pre-training is performed based on RBM and weights are initialized using the CD-k method to calculate the probability distribution of the activation of the hidden layer confidence neurons in the deep belief network, and then the weights are updated based on the probability distribution. The process of propagation and updating based on graph convolutional neural networks includes aggregating the node set of the traffic network graph data through pooling propagation.

4. The method for dynamic allocation of traffic resources based on a confidence graph convolutional neural network according to claim 3, characterized in that, Methods for processing the original feature representation into the Skipgram model to obtain the final feature extraction vector include: During backpropagation, some weights of the hidden layer are updated, and the probability of updating the weights is determined based on the occurrence probability of nodes associated with the weights. The higher the occurrence probability of a node, the higher the probability that the weight associated with it will be updated. Some nodes are sampled, and the probability of the node being sampled is controlled. In partial sampling, the higher the frequency of a node, the higher the probability that it will be used as the starting point for sampling.

5. The method for dynamic allocation of traffic resources based on a confidence graph convolutional neural network according to claim 4, characterized in that, The probability that the weights will be updated is determined by the following formula: In the formula, P(v) vi For node v i The probability of occurrence, Z(v) i ) is node v i The weights, Z(v) j ) is node v j The weights, v i v j A set of nodes derived from traffic network graph data; The probability of the occurrence of the node is determined according to the following formula: In the formula, P(o i ) represents node o in the network constructed based on existing weights. i The probability of occurrence, Z(o) i ) represents node o in the network constructed based on existing weights. i The weight.

6. The method for dynamic allocation of traffic resources based on a confidence graph convolutional neural network according to claim 3, characterized in that, The probability distribution of the activation of hidden layer confidence neurons in the deep belief network is calculated according to the following formula: In the formula, For probability distribution, v (0) The set of nodes derived from traffic network graph data, where θ is the learning rate and W is the learning rate. j Let b be the vector matrix to be processed. j This is the bias value; And / or, The node set of the traffic network graph data is aggregated according to the following formula: In the formula, pool represents the pooling operation, s is the vector of the s-th layer, θ is the learning rate, and W is the adjacency vector. For node v i The prediction result, k represents the k-th layer of the neural network, b is the offset, v i The set of nodes V comes from the traffic network graph data.

7. The method for dynamic allocation of traffic resources based on a confidence graph convolutional neural network according to claim 1 or 3, characterized in that, The deep belief network-based processing also includes: Based on the calculated probability distribution, Gibbs sampling continues, and corresponding values ​​are extracted from the neurons of the explicit layers in the deep belief network for sampling and reconstruction, thereby updating and obtaining new weights and biases. 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 the new weight matrix, λ and T are the learning parameters, x is the sampling probability, and h is the weight matrix. (0) For the hidden layer representation of layer 0, the h (0) The calculation method includes summing the results obtained by the node on different subgraphs after the aggregation operation at level 0, z. (0) For the output at the level 0 node, x (1) z is the sampling probability of the first layer. (1) v(1) is the output of the first-level node, and v(1) is the set of nodes in the first level.

8. The method for dynamic allocation of traffic resources based on a confidence graph convolutional neural network according to claim 1, 2, or 3, characterized in that, The update process based on graph convolutional neural networks also includes: The feature representation of the node is updated by applying gradient descent to the loss function, wherein the loss function is determined by the following equation: In the formula, L g (u z Let θ be the loss function, θ be the learning rate, and u be the learning rate. z u v Let T be the feature vector of the traffic network graph data, and T be the learning parameters. It represents a probability distribution.

9. A dynamic traffic resource allocation system based on a confidence graph convolutional neural network, characterized in that, The dynamic traffic resource allocation system allocates traffic resources based on the method described in any one of claims 1 to 8, and the dynamic traffic resource allocation system includes: The modeling unit is configured to model the current actual traffic network and traffic stations to generate traffic network map data; The feature extraction unit is configured to extract features from the modeled traffic network graph data using a graph embedding method to determine the feature extraction vector. The data processing unit is configured to process the feature extraction vector based on a deep belief network and propagate and update it through a graph convolutional neural network, thereby updating the feature representation of the node and changing the label type of each node to generate a target topology. Then, based on the target topology, traffic resources are automatically allocated to the corresponding traffic stations to achieve automated allocation of traffic resources.

10. A storage medium, characterized in that, The storage medium is configured to store a computer program, which is configured to execute the dynamic allocation method for traffic resources as described in any one of claims 1 to 8.

Images