A knowledge graph analysis method for traffic overload control
By establishing a static road network map and a dynamic spatiotemporal map of traffic flow on road segments, and combining the spatiotemporal map convolutional neural network, the trend of illegal behavior is predicted and the deployment strategy of overload control facilities is iteratively optimized. This solves the problems of low prediction accuracy and lack of iterative optimization in existing technologies, and realizes an efficient overload control deployment scheme.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SHENZHEN URBAN TRANSPORT PLANNING CENT CO LTD
- Filing Date
- 2026-03-10
- Publication Date
- 2026-06-09
AI Technical Summary
Existing technologies fail to effectively integrate dynamic spatiotemporal traffic flow data in traffic enforcement against overloading, resulting in low prediction accuracy and a lack of iterative optimization mechanisms, thus failing to effectively assist in the formulation of overloading control strategies.
By establishing a static road network map and a dynamic spatiotemporal map of traffic flow on road segments, a spatiotemporal graph convolutional neural network is designed. Combined with a multi-layer graph convolutional neural network and a long short-term memory network, the trend of illegal behavior is predicted, and the deployment strategy of overload control facilities is iteratively optimized to achieve multi-objective optimization.
It has achieved the integration and updating of multi-source heterogeneous data, improved the accuracy of violation prediction, optimized the overall effect of the overload control facility deployment scheme, and dynamically balanced the violation risk and deployment cost.
Smart Images
Figure CN122175083A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of traffic overload control technology, and in particular to a knowledge graph analysis method for traffic overload control. Background Technology
[0002] Knowledge graphs are a structured representation of knowledge that expresses entities, attributes, and relationships between entities in graphical form. They have wide applications across various industries, including e-commerce, healthcare, financial services, the internet, and the Internet of Things. Knowledge graphs also play a crucial role in the transportation sector, including intelligent traffic management, route planning, public transport optimization, traffic safety and risk management, vehicle-to-everything (V2X) connectivity, and urban planning. However, research in areas such as traffic enforcement related to overloading, such as deployment of controls and source management, is currently limited.
[0003] Existing technologies only use static knowledge graphs to analyze traffic violations, without combining dynamic spatiotemporal traffic flow data (such as time-segmented vehicle volume and detour frequency), resulting in low prediction accuracy. Furthermore, spatiotemporal graph convolution is only used for traffic flow prediction, without being linked to overload control facility deployment strategies, and lacks an iterative optimization mechanism. At the same time, due to the diverse and insufficiently correlated data sources in this field, it also faces the challenges of multimodal and heterogeneous data, posing significant difficulties for the analysis and application of knowledge graphs. Summary of the Invention
[0004] The purpose of this invention is to address the limitations of applying knowledge graphs in the field of traffic overload control enforcement. By providing a knowledge graph analysis method for traffic overload control deployment, this invention can systematically acquire, represent, analyze, and store relevant data, information, and knowledge related to traffic control enforcement, thereby effectively solving the limitations of applying knowledge graphs in the field of traffic overload control enforcement and better assisting in enforcement. To achieve the above objectives, this invention is implemented through the following technical solutions: A knowledge graph analysis method for traffic overload control includes the following steps: S1. Establish a static road network map and a dynamic spatiotemporal map of road segment traffic flow; the static road network map is a knowledge graph that includes road segment entities, road segment connection relationships and static attributes, and the dynamic spatiotemporal map of road segment traffic flow is a time-series knowledge graph that includes traffic flow time-series attributes in different time periods; S2. Design a spatiotemporal graph convolutional neural network to predict and infer the trend of violations on each road segment in the static road network map and the dynamic spatiotemporal map of traffic flow; the spatiotemporal graph convolutional neural network integrates a road network coding module, a spatiotemporal graph convolutional coding module and an inference module; The implementation of step S2 includes: Step S21: Encode the static road network map using the road network coding module and extract the road network representation; wherein, the road network coding module is: Formula 1 In the formula, For the static diagram of the road network, X r A represents a road segment node in a static road network diagram. r This represents a road segment node in a static road network diagram; Road segment node A in the static road network diagram r Matrix in real vector space express, Represents the space of real numbers. The dimension is A dimensional real vector space, static map of the road network The number of road segment nodes, The dimension of the road segment node attributes; Step S22: Encode the dynamic spatiotemporal map of road segment traffic flow using a spatiotemporal graph convolutional coding module and extract the spatiotemporal representation of the path; wherein, the spatiotemporal graph convolutional coding module is: Formula 4 In the formula, This is a dynamic spatiotemporal map of traffic flow on the road segment. For the road segment nodes in the dynamic spatiotemporal map of road traffic flow, and , The dynamic spatiotemporal graph of traffic flow in road segments is a matrix showing the connectivity and adjacency of road segments. , The number of nodes in the road segment. The dimensions of the node attributes in the dynamic spatiotemporal map of road traffic flow; The dimension is A dimensional real vector space; Step S23: Use the inference module to fuse the road network representation and the path spatiotemporal representation to predict and infer the trend of violations in each road segment over a period of time. S3. Based on the trend of violations in each road section, with the goal of minimizing the degree of violation risk and the constraint of the preset cost of overload control facilities, generate an optimization strategy for the deployment of overload control facilities. S4. Utilize the generated overload control facility deployment optimization strategy to evaluate the evolution of the violation trend. When the violation risk level of each road segment or the overall violation risk level is determined to be higher than the set threshold, update the overload control facility type attribute in the road network static map, and iteratively execute steps S1 to S3 to re-evaluate the evolution of the violation trend until the violation risk level of each road segment or the overall violation risk level is lower than the set threshold, then stop iterating and complete the overload control facility deployment optimization strategy.
[0005] Preferably, the implementation of step S1 includes: Step S11: Based on the latitude and longitude of each overload control point, match each overload control point to the urban road network that is closest to the current overload control point, and associate the overload control points with the urban road network. Step S12: Extract road segment entities from the urban road network as nodes of the road network static graph, extract the road segment connection relationships from the urban road network as edges of the road network static graph, and extract the road segment length, average vehicle speed, node latitude and longitude, and overload control facility attributes on the road segment as node attributes to form a road network static graph. Step S13: Extract road segment entities from the urban road network again as nodes of the road segment traffic flow dynamic spatiotemporal graph, extract the road segment connection relationships from the urban road network as edges of the road segment traffic flow dynamic spatiotemporal graph, and extract the number of vehicles passing through, the number of overloaded vehicles passing through, the number of vehicles detouring through, the main truck types, the main cargo types, the number of overloadings, the maximum overloading range, and the number of detourings within the time slice as node attributes to form a road segment traffic flow dynamic spatiotemporal graph with time sequence attributes.
[0006] Preferably, the attributes of the overload control point facilities in step S12 include the presence or absence of overload control points and the type of overload control points.
[0007] Preferably, the method for encoding the static road network map in step S21 is as follows: Formula 3 In the formula, For the normalized adjacency matrix of the road network coding module, The first in the road network coding module Features of the layer The first in the road network coding module The network weight matrix of the layer, It is a non-linear activation function. For the number of layers in the encoding network, when hour, .
[0008] Preferably, the road network coding module in step S21 further includes a multi-layer graph convolutional neural network, wherein any one layer of the graph convolutional neural network is represented as: Formula 2 In the formula, For the normalized road network adjacency matrix, The first in the road network coding module Features of the layer The first in the road network coding module The network weight matrix of the layer, Let be a nonlinear activation function, at the th In the layer, the feature matrix This is the input to this layer, and This is the output of the layer after transformation by Equation 2, and is used as the first... Layer input; Preferably, the spatiotemporal graph convolutional coding module in step S22 further includes a multi-layer graph convolutional neural network and a long short-term memory network, wherein any one layer of the graph convolutional neural network is represented as: Formula 5 In the formula, The normalized adjacency matrix, For the spatiotemporal graph convolutional coding module at the current time step s, the first... Features of the layer For the spatiotemporal graph convolutional coding module at the current time step s, the first... The weight matrix of a graph convolutional neural network with multiple layers. For nonlinear activation functions, at the th... In the layer, the feature matrix This is the input to this layer, and This is the output of this layer after transformation by Formula 5, and also serves as the first... Layer input, This indicates that all convolutional networks in this layer are in the first... Each time step is performed; The Long Short-Term Memory network is: Formula 6 In the formula, LSTM represents Long Short-Term Memory network. and These represent the current time step. The first module of the spatiotemporal graph convolutional coding module Layer characteristics and next time step The first module of the spatiotemporal graph convolutional coding module Characteristics of the layer.
[0009] Preferably, the method for encoding the dynamic spatiotemporal map of road traffic flow in step S22 is as follows: Formula 7 In the formula, LSTM represents Long Short-Term Memory network. This represents the number of layers in the coding network. It is a non-linear activation function. The normalized adjacency matrix, Indicates the current time step The following features This indicates that all convolutional networks in this layer are in the first... Each step is performed in a specific time step.
[0010] Preferably, the implementation method of the predictive reasoning in step S23 includes: Step S231, for and Perform representation connections to form a fused representation dimension. , Formula 8 In the formula, This represents the concatenation of two vectors, and the resulting fused representation has a dimension of [missing information]. ; Step S232: Use a self-attention module to focus on nodes and filter out road segments that are prone to violations, thereby alleviating the sparsity problem of road segments prone to violations in positive sample nodes. The way the focus is expressed is as follows: Formula 9 In the formula, To The probability normalized vector after focusing, W h1 and W h2 There are two weight matrices. For a fixed activation function, To integrate representation dimensions; Step S233: Perform linear regression on the node values using a linear layer to form the violation trend of road segment nodes. The linear regression method is as follows: Formula 10 In the formula, For a weight matrix, The final road segment node violation trend matrix is obtained. To The probability normalization vector after focusing. To integrate representation dimensions.
[0011] Preferably, the implementation of step S3 includes: Step S31: Calculate and statistically analyze the violation frequency and violation type based on the violation trends of each road segment, and calculate the violation risk level of each road segment in combination with the road segment attributes; Step S32: The user presets the threshold for the degree of violation risk and presets the deployment cost of the overload control facilities; Step S33: Set minimizing the degree of violation risk as the optimization objective and setting the deployment cost as the constraint. Use the neighborhood search algorithm to perform a local search on the current deployment plan. Improve the deployment plan by trying to add, delete or adjust the location of the overload control facilities, and generate a new optimized facility deployment plan.
[0012] Preferably, the violation trend in step S31 includes one or more of the following: overloading, detour, speeding, and driving against traffic.
[0013] The present invention has the following advantages over the prior art: 1. The method of this invention realizes the ability to integrate multi-source heterogeneous data such as road network, truck routes, and violations; the knowledge graph used is easy to expand and update, and different subgraphs can be extracted according to needs, thereby enhancing the scalability of complex scenarios for controlling overloading.
[0014] 2. The predictive reasoning part of the method of this invention, which targets the trend of violations, fully utilizes the spatiotemporal correlation characteristics of the static road network map and the dynamic spatiotemporal map of traffic flow on road segments by combining spatiotemporal convolution algorithms. The deployment scheme for overload control facilities also adopts a multi-iterative multi-objective optimization design to simultaneously optimize the total amount of violations and the deployment cost of overload control facilities, ensuring the overall optimality of the deployment scheme. Furthermore, it is the first to fuse the static road network map and the dynamic spatiotemporal map of traffic flow on road segments, solving the core pain points of data heterogeneity and insufficient correlation in the field of overload control.
[0015] 3. The method of this invention can systematically acquire, represent, analyze and store traffic control and law enforcement related data, information and knowledge, thereby effectively solving the limitations of knowledge graphs in the field of traffic overload control law enforcement and better assisting overload control law enforcement; and proposes a closed-loop mechanism of control strategy, risk assessment, attribute update and iterative optimization, rather than one-time control, to achieve a dynamic balance between violation risk and control cost. Attached Figure Description
[0016] Figure 1 This is a schematic diagram of the execution flow of the method of the present invention. Detailed Implementation
[0017] The present invention will now be further described with reference to the accompanying drawings and specific embodiments: To make the objectives, technical solutions, and advantages of the present invention clearer and more explicit, the present invention will be further described below with reference to the accompanying drawings and embodiments.
[0018] Example 1: like Figure 1 As shown, a knowledge graph analysis method for traffic overload control includes the following steps: S1. Establish a static road network map and a dynamic spatiotemporal map of road segment traffic flow; the static road network map is a knowledge graph containing road segment entities, road segment connection relationships, and static attributes, while the dynamic spatiotemporal map of road segment traffic flow is a time-series knowledge graph containing time-series attributes of traffic flow; wherein, the implementation method of step S1 includes: Step S11: Based on the latitude and longitude of each overload control point, match each overload control point to the urban road network that is closest to the current overload control point, and associate the overload control points with the urban road network; the facility attributes of the overload control points include the presence or absence of the overload control point and the type of the overload control point. Step S12: Extract road segment entities from the urban road network as nodes of the road network static graph, extract the road segment connection relationships from the urban road network as edges of the road network static graph, and extract the road segment length, average vehicle speed, node latitude and longitude, and overload control facility attributes on the road segment as node attributes to form a road network static graph. Step S13: Extract road segment entities from the urban road network again as nodes of the road segment traffic flow dynamic spatiotemporal graph, extract the road segment connection relationships from the urban road network as edges of the road segment traffic flow dynamic spatiotemporal graph, and extract the number of vehicles passing through, the number of overloaded vehicles passing through, the number of vehicles detouring through, the main truck types, the main cargo types, the number of overloadings, the maximum overloading range, and the number of detourings within the time slice as node attributes to form a road segment traffic flow dynamic spatiotemporal graph with time sequence attributes.
[0019] S2. Design a spatiotemporal graph convolutional neural network to predict and infer the trends of violations on each road segment in the static road network map and the dynamic spatiotemporal map of traffic flow. The designed spatiotemporal graph convolutional neural network integrates a road network coding module, a spatiotemporal graph convolutional coding module, and an inference module. The implementation method of this step includes: Step S21: Encode the static road network map using the road network coding module and extract the road network representation; wherein, the road network coding module is: Formula 1 In the formula, For the static diagram of the road network, X r A represents a road segment node in a static road network diagram. r This represents a road segment node in the static road network diagram; while road segment node A in the static road network diagram... r It can also be a matrix in real vector space express, Represents the space of real numbers. The dimension is A dimensional real vector space, static map of the road network The number of road segment nodes, This represents the dimension of the road segment node attributes.
[0020] Specifically, the road network coding module also includes a multi-layer graph convolutional neural network, wherein any one layer of the graph convolutional neural network is represented as: Formula 2 In the formula, For the normalized road network adjacency matrix, The first in the road network coding module Features of the layer The first in the road network coding module The network weight matrix of the layer, Let be a nonlinear activation function, at the th In the layer, the feature matrix This is the input to this layer, and This is the output of the layer after transformation by Equation 2, and is used as the first... The input of the layer.
[0021] The method for encoding the static road network map in step S21 is as follows: Formula 3 In the formula, For the normalized adjacency matrix of the road network coding module, The first in the road network coding module Features of the layer The first in the road network coding module The network weight matrix of the layer, It is a non-linear activation function. For the number of layers in the encoding network, when hour, Furthermore, in the In the layer, the feature matrix This is the input to this layer. This is the output of the layer after transformation by Formula 2, and is used as the first... The input of the layer; this process can be regarded as a multi-layer neural network used to encode the static map of the road network, which is essentially a cumulative multiplication of Formula 2 multiple times.
[0022] Step S22: Encode the dynamic spatiotemporal map of road segment traffic flow using a spatiotemporal graph convolutional coding module and extract the spatiotemporal representation of the path; wherein, the spatiotemporal graph convolutional coding module is: Formula 4 In the formula, This is a dynamic spatiotemporal map of traffic flow on the road segment. For the road segment nodes in the dynamic spatiotemporal map of road traffic flow, and , The dynamic spatiotemporal graph of traffic flow in road segments is a matrix showing the connectivity and adjacency of road segments. , The number of nodes in the road segment. The dimensions of the node attributes in the dynamic spatiotemporal map of road traffic flow; The dimension is A real vector space of dimension 1.
[0023] Specifically, the spatiotemporal graph convolutional coding module also includes a multi-layer graph convolutional neural network and a long short-term memory network, wherein any one layer of graph convolutional neural network is represented as: Formula 5 In the formula, The normalized adjacency matrix, For the spatiotemporal graph convolutional coding module at the current time step s, the first... Features of the layer For the spatiotemporal graph convolutional coding module at the current time step s, the first... The weight matrix of a graph convolutional neural network with multiple layers. For nonlinear activation functions, at the th... In the layer, the feature matrix This is the input to this layer, and This is the output of this layer after transformation by Formula 5, and also serves as the first... Layer input, This indicates that all convolutional networks in this layer are in the first... Each step is performed in a specific time step.
[0024] Long Short-Term Memory (LSTM) networks are as follows: Formula 6 In the formula, LSTM represents Long Short-Term Memory network. and These represent the current time step. The first module of the spatiotemporal graph convolutional coding module Layer characteristics and next time step The first module of the spatiotemporal graph convolutional coding module Characteristics of the layer.
[0025] The method for encoding the dynamic spatiotemporal map of road traffic flow in the above steps is as follows: Formula 7 In the formula, LSTM represents Long Short-Term Memory network. This represents the number of layers in the coding network. It is a non-linear activation function. The normalized adjacency matrix, Indicates the current time step The following features This indicates that all convolutional networks in this layer are in the first... The process is performed at multiple time steps; this process is a multi-layer neural network used to encode the dynamic spatiotemporal map of traffic flow on road segments. Let be the number of layers in the coding network, where when hour, .
[0026] Step S23: The road network representation and path spatiotemporal representation are fused using the inference module to predict and infer the trend of violations on each road segment over a period of time. The implementation methods for this prediction and inference step include: Step S231, for and Perform representation connections to form a fused representation dimension. , Formula 8 In the formula, This represents the concatenation of two vectors, and the resulting fused representation has a dimension of [missing information]. ; Step S232: Use a self-attention module to focus on nodes and filter out road segments that are prone to violations, thereby alleviating the sparsity problem of road segments prone to violations in positive sample nodes. The way to express focus is as follows: Formula 9 In the formula, To The probability normalized vector after focusing, W h1 and W h2 There are two weight matrices. For a fixed activation function, To integrate representation dimensions; Step S233: Perform linear regression on the node values using a linear layer to form the violation trend of road segment nodes. The linear regression method is as follows: Formula 10 In the formula, For a weight matrix, The final result is a trend matrix of road segment node violations, which is a The matrix represents the total The future of each section of road The trend of violations at any given moment, ranging from 0 to 1, with a larger value indicating a stronger trend of violations; To The probability normalization vector after focusing. To integrate representation dimensions.
[0027] S3. Based on the trends of violations on each road segment, and with the goal of minimizing the degree of violation risk, and constrained by the preset cost of deploying overload control facilities, generate an optimization strategy for the deployment of overload control facilities; wherein, the implementation method of this step includes: Step S31: Calculate and statistically analyze the violation frequency and violation type based on the violation trends of each road segment, and calculate the violation risk level of each road segment in combination with the road segment attributes; the violation trends include one or more of the following: overloading, detour, speeding, and driving against traffic; Step S32: The user presets the threshold for the degree of violation risk and presets the deployment cost of the overload control facilities; Step S33: Set minimizing the degree of violation risk as the optimization objective and setting the deployment cost as the constraint. Use the neighborhood search algorithm to perform a local search on the current deployment plan. Improve the deployment plan by trying to add, delete or adjust the location of the overload control facilities, and generate a new optimized facility deployment plan.
[0028] S4. Utilize the generated overload control facility deployment optimization strategy to evaluate the evolution of the violation trend. When the violation risk level of each road segment or the overall violation risk level is determined to be higher than the set threshold, update the overload control facility type attribute in the road network static map, and iteratively execute steps S1 to S3 to re-evaluate the evolution of the violation trend until the violation risk level of each road segment or the overall violation risk level is lower than the set threshold, then stop iterating and complete the overload control facility deployment optimization strategy.
[0029] The purpose of this invention is to "evaluate" and "optimize" the deployment plan for overload control facilities. The evaluation criterion is whether the violation risk meets the requirements after the deployment plan is implemented; the optimization goal is to minimize the violation risk after the deployment plan is implemented. Step S1 is a data processing method, step S2 is an evaluation means, step S3 is an optimization method, and step S4 is a cyclical judgment method. If the optimization goal of step S3 is met, the process ends; if the optimization goal of step S3 is not met, the attributes of the road network static map are updated, and the evaluation and optimization process restarts from step S1.
[0030] For those skilled in the art, various other corresponding changes and modifications can be made based on the technical solutions and concepts described above, and all such changes and modifications should fall within the protection scope of the claims of this invention.
Claims
1. A knowledge graph analysis method for traffic overload control deployment, characterized in that, Including the following steps: S1. Establish a static road network map and a dynamic spatiotemporal map of road segment traffic flow; the static road network map is a knowledge graph that includes road segment entities, road segment connection relationships and static attributes, and the dynamic spatiotemporal map of road segment traffic flow is a time-series knowledge graph that includes traffic flow time-series attributes in different time periods; S2. Design a spatiotemporal graph convolutional neural network to predict and infer the trend of violations on each road segment in the static road network map and the dynamic spatiotemporal map of traffic flow; the spatiotemporal graph convolutional neural network integrates a road network coding module, a spatiotemporal graph convolutional coding module and an inference module; The implementation of step S2 includes: Step S21: Encode the static road network map using the road network coding module and extract the road network representation; wherein, the road network coding module is: Official 1 In the formula, For the static diagram of the road network, X r A represents a road segment node in a static road network diagram. r This represents a road segment node in a static road network diagram; Road segment node A in the static road network diagram r Matrix in real vector space express, Represents the space of real numbers. The dimension is A dimensional real vector space, static map of the road network The number of road segment nodes, The dimension of the road segment node attributes; Step S22: Encode the dynamic spatiotemporal map of road segment traffic flow using a spatiotemporal graph convolutional coding module and extract the spatiotemporal representation of the path; wherein, the spatiotemporal graph convolutional coding module is: Official 4 In the formula, This is a dynamic spatiotemporal map of traffic flow on the road segment. For the road segment nodes in the dynamic spatiotemporal map of road traffic flow, and , The dynamic spatiotemporal graph of traffic flow in road segments is a matrix showing the connectivity and adjacency of road segments. , The number of nodes in the road segment. The dimensions of the node attributes in the dynamic spatiotemporal map of road traffic flow; The dimension is A dimensional real vector space; Step S23: Use the inference module to fuse the road network representation and the path spatiotemporal representation to predict and infer the trend of violations in each road segment over a period of time. S3. Based on the trend of violations in each road section, with the goal of minimizing the degree of violation risk and the constraint of the preset cost of overload control facilities, generate an optimization strategy for the deployment of overload control facilities. S4. Utilize the generated overload control facility deployment optimization strategy to evaluate the evolution of the violation trend. When the violation risk level of each road segment or the overall violation risk level is determined to be higher than the set threshold, update the overload control facility type attribute in the road network static map, and iteratively execute steps S1 to S3 to re-evaluate the evolution of the violation trend until the violation risk level of each road segment or the overall violation risk level is lower than the set threshold, then stop iterating and complete the overload control facility deployment optimization strategy.
2. The knowledge graph analysis method for traffic overload control as described in claim 1, characterized in that, The implementation of step S1 includes: Step S11: Based on the latitude and longitude of each overload control point, match each overload control point to the urban road network that is closest to the current overload control point, and associate the overload control points with the urban road network. Step S12: Extract road segment entities from the urban road network as nodes of the road network static graph, extract the road segment connection relationships from the urban road network as edges of the road network static graph, and extract the road segment length, average vehicle speed, node latitude and longitude, and overload control facility attributes on the road segment as node attributes to form a road network static graph. Step S13: Extract road segment entities from the urban road network again as nodes of the road segment traffic flow dynamic spatiotemporal graph, extract the road segment connection relationships from the urban road network as edges of the road segment traffic flow dynamic spatiotemporal graph, and extract the number of vehicles passing through, the number of overloaded vehicles passing through, the number of vehicles detouring through, the main truck types, the main cargo types, the number of overloadings, the maximum overloading range, and the number of detourings within the time slice as node attributes to form a road segment traffic flow dynamic spatiotemporal graph with time sequence attributes.
3. The knowledge graph analysis method for traffic overload control as described in claim 2, characterized in that, The attributes of the overload control point facilities in step S12 include the presence or absence of overload control points and the type of overload control points.
4. The knowledge graph analysis method for traffic overload control as described in claim 1, characterized in that, The method for encoding the static road network map in step S21 is as follows: Official 3 In the formula, For the normalized adjacency matrix of the road network coding module, The first in the road network coding module Features of the layer The first in the road network coding module The network weight matrix of the layer, It is a non-linear activation function. For the number of layers in the encoding network, when hour, .
5. The knowledge graph analysis method for traffic overload control as described in claim 1, characterized in that, The road network coding module in step S21 further includes a multi-layer graph convolutional neural network, wherein any one layer of the graph convolutional neural network is represented as follows: Official 2 In the formula, For the normalized road network adjacency matrix, The first in the road network coding module Features of the layer The first in the road network coding module The network weight matrix of the layer, Let be a nonlinear activation function, at the th In the layer, the feature matrix This is the input to this layer, and This is the output of the layer after transformation by Equation 2, and is used as the first... The input of the layer.
6. The knowledge graph analysis method for traffic overload control as described in claim 1, characterized in that, The spatiotemporal graph convolutional coding module in step S22 further includes a multi-layer graph convolutional neural network and a long short-term memory network, wherein any one layer of the graph convolutional neural network is represented as: Official 5 In the formula, The normalized adjacency matrix, For the spatiotemporal graph convolutional coding module at the current time step s, the first... Features of the layer For the spatiotemporal graph convolutional coding module at the current time step s, the first... The weight matrix of a graph convolutional neural network with multiple layers. For nonlinear activation functions, at the th... In the layer, the feature matrix This is the input to this layer, and This is the output of this layer after transformation by Formula 5, and also serves as the first... Layer input, This indicates that all convolutional networks in this layer are in the first... Each time step is performed; The Long Short-Term Memory network is: Official 6 In the formula, LSTM represents Long Short-Term Memory network. and These represent the current time step. The first module of the spatiotemporal graph convolutional coding module Layer characteristics and next time step The first module of the spatiotemporal graph convolutional coding module Characteristics of the layer.
7. The knowledge graph analysis method for traffic overload control as described in claim 1, characterized in that, The method for encoding the dynamic spatiotemporal map of road traffic flow in step S22 is as follows: Official 7 In the formula, LSTM represents Long Short-Term Memory network. This represents the number of layers in the coding network. It is a non-linear activation function. The normalized adjacency matrix, Indicates the current time step The following features This indicates that all convolutional networks in this layer are in the first... Each step is performed in a specific time step.
8. The knowledge graph analysis method for traffic overload control as described in claim 1, characterized in that, The implementation methods for predictive reasoning in step S23 include: Step S231, for and Perform representation connections to form a fused representation dimension. , Official 8 In the formula, This represents the concatenation of two vectors, and the resulting fused representation has a dimension of [missing information]. ; Step S232: Use a self-attention module to focus on nodes and filter out road segments that are prone to violations, thereby alleviating the sparsity problem of road segments prone to violations in positive sample nodes. The way the focus is expressed is as follows: Official 9 In the formula, To The probability normalized vector after focusing, W h1 and W h2 There are two weight matrices. For a fixed activation function, To integrate representation dimensions; Step S233: Perform linear regression on the node values using a linear layer to form the violation trend of road segment nodes. The linear regression method is as follows: Official 10 In the formula, For a weight matrix, The final road segment node violation trend matrix is obtained. To The probability normalization vector after focusing. To integrate representation dimensions.
9. A knowledge graph analysis method for traffic overload control as described in claim 1, characterized in that, The implementation of step S3 includes: Step S31: Calculate and statistically analyze the violation frequency and violation type based on the violation trends of each road segment, and calculate the violation risk level of each road segment in combination with the road segment attributes; Step S32: The user presets the threshold for the degree of violation risk and presets the deployment cost of the overload control facilities; Step S33: Set minimizing the degree of violation risk as the optimization objective and setting the deployment cost as the constraint. Use the neighborhood search algorithm to perform a local search on the current deployment plan. Improve the deployment plan by trying to add, delete or adjust the location of the overload control facilities, and generate a new optimized facility deployment plan.
10. A knowledge graph analysis method for traffic overload control as described in claim 9, characterized in that, The violation trends in step S31 include one or more of the following: overloading, detour, speeding, and driving against traffic.