A semantic uncertainty-based road network spatial site selection method and system

Abstract

The application relates to the technical field of traffic site selection, and discloses a road network space site selection method and system based on semantic uncertainty, which comprises the following steps: generating Beta distribution shape parameters corresponding to OD flow according to an OD flow data set and a research area POI data set; constructing a hierarchical binary tree decision environment based on city road network topology data, an OD trajectory data set and the Beta distribution shape parameters, and generating flow composition fingerprint features corresponding to each tree node; determining a state space, an action space and a marginal contribution reward mechanism based on sequence priority of a plurality of intelligent agents based on the above processing; initializing double-head Critic network and Actor network of the plurality of intelligent agents, generating experience trajectories through interaction between the intelligent agents and the environment, calculating a risk adjustment advantage function of uncertainty perception, updating network parameters based on Gaussian negative logarithm likelihood loss and PPO truncation agent loss, and outputting a final facility deployment road section set after iterative training until convergence.

Description

Technical Field

[0001] This invention relates to the field of traffic location technology, and in particular to a method and system for spatial location selection of road networks based on semantic uncertainty. Background Technology

[0002] Flow capture location is an important branch of spatial optimization research, aiming to maximize the interception or service coverage of origin-destination (OD) flows by optimizing the deployment location of facilities in the road network. It has wide applications in areas such as urban law enforcement resource allocation, surveillance equipment deployment, and emergency rescue station planning. In real-world scenarios, OD flows are formed by movement between origin and destination points. Vehicles typically traverse multiple road segments, leading to spatial dependencies in facility deployment. Distributing facilities across different OD flows along their respective segments allows for the interception of multiple traffic flows, resulting in coordinated spatial interception. However, deploying facilities upstream or downstream of the same OD flow path results in repeated interception of the same flow, leading to redundant coverage. Therefore, adjusting the location of a facility can affect the overall traffic capture, and the interception gains cannot be simply linearly accumulated, making the flow capture location problem essentially an NP-hard combinatorial optimization problem.

[0003] Numerous scholars have conducted research on the location selection problem for traffic flow capture. Early studies largely relied on the assumption of definitive OD (Original Discharge) flow paths; however, this assumption fails to reflect the complexities of real-world travel scenarios, leading to discrepancies between selected location schemes and the actual optimal solutions. As research has deepened, scholars have gradually incorporated path selection uncertainties caused by factors such as vehicle range and driver decision preferences into location selection studies, significantly improving the realism of the selected schemes. Current research has yielded many results, effectively addressing the problem of OD flow coverage optimization in road networks. However, human travel activities are highly purposeful, and OD flows possess different target attributes from different location selection task perspectives. If coverage optimization is based solely on indicators such as the quantity of OD flows, it becomes difficult to capture OD flows with specific semantic meanings in specific task scenarios, resulting in discrepancies between the location selection results and the actual task objectives (such as drunk driving enforcement and hazardous materials transportation supervision).

[0004] Furthermore, existing methods face the challenge of excessively high decision space dimensionality when dealing with large-scale road networks, making it difficult to efficiently search for globally optimal facility combinations. Simultaneously, the credit allocation problem in multi-facility collaborative deployment (i.e., how to fairly and effectively assess the marginal contribution of each facility) remains unresolved, leading to suboptimal site selection schemes. Therefore, how to achieve efficient and collaborative multi-facility spatial site selection in large-scale road networks while simultaneously modeling the semantic uncertainty of OD flows is a pressing technical problem to be solved in this field. Summary of the Invention

[0005] Aimed at at least in solving one of the technical problems existing in the prior art, the present invention provides a road network spatial location method and system based on semantic uncertainty, so as to solve the problems of low location efficiency and misalignment between location scheme and target demand in existing road network spatial location methods.

[0006] One aspect of the present invention provides a road network spatial location method based on semantic uncertainty, comprising: S1, based on the OD flow dataset and the POI dataset of the study area, quantifies the uncertainty of the occurrence probability of the target semantics of each OD flow and generates the Beta distribution shape parameter corresponding to each OD flow; S2, based on urban road network topology data, OD trajectory dataset and Beta distribution shape parameters, constructs a hierarchical binary tree decision environment and generates flow composition fingerprint features corresponding to each tree node; S3 defines the state space, action space, and marginal contribution reward mechanism based on sequence priority for multi-agent systems, based on the binary tree decision environment, Beta distribution shape parameters, and the number of facilities to be deployed. S4 initializes the multi-agent dual-head Critic network and Actor network, generates experience trajectories through the interaction between the agents and the environment, calculates the risk adjustment advantage function for uncertainty perception, updates the network parameters based on Gaussian negative log-likelihood loss and PPO truncated agent loss, and outputs the final set of facility deployment road segments after iterative training until convergence.

[0007] According to the semantically uncertain road network spatial location method, S1 includes: Input target semantic corresponding POI weight vector The number of Points of Interest (POIs) in the neighborhood of the origin or destination of the OD flow and the total number of POIs in the neighborhood are counted. The expected value of the target semantic probability of the OD flow is calculated using the Sigmoid function. for: ; in, The first in the neighborhood of the OD flow The number of POIs The total number of POIs in the neighborhood, of which Total number of POI categories; Calculate the functional mixing degree of POI in the OD stream for: ; ; in, This represents the total number of semantic categories of POIs covered within the region. Indicates the first Class POI in the first The percentage of the neighborhood of an OD flow; Couple the POI mixture to the total number of samples to calculate semantic confidence. for: ; in, This represents the total number of POIs within this neighborhood. By reparameterizing, the expected probability and confidence level are mapped to shape parameters controlling the Beta distribution. The calculation method is as follows: ; ; in, This represents the shape parameter that controls the Beta distribution.

[0008] According to the semantically uncertain road network spatial location method, S2 includes: The start and end sets of the full OD trajectories are extracted, and the K-Means algorithm is used to cluster the start and end sets to obtain the start cluster set. Clustering set with endpoint Each OD trajectory is encoded into a binary tuple based on the starting point cluster set and the ending point cluster set. ,in , , The total number of clusters starting from the starting point. The endpoint is the total number of clusters; The proportion of different types of origin-destination (OD) flows on each road segment is statistically analyzed to generate the flow fingerprint characteristics of the road segment. for: ; in, This indicates that the person passed through this section of road and belonged to... Area to Number of trajectories in the region This represents the total number of all tracks passing through this road segment; The original directed road network is transformed into a dual graph, where road segments are mapped to dual nodes, and the topological connections between road segments are mapped to dual edges. The Metis graph cutting algorithm is then used to perform hierarchical partitioning of the dual graph, which satisfies the following relationship: ; in, Indicates the strength of topological association. This represents the cutting cost, which is the sum of the weights of the dual edges that are cut. This represents minimizing the cost of cutting; and This represents two dual nodes in a dual graph. Represents the set of nodes in the subgraph obtained from the segmentation; Represents the edge ( () represents a dividing edge that spans two subgraphs. Let be the set of edges of the dual graph; Hierarchical binary tree structure for building road networks from top to bottom The binary tree structure The root node is used to represent the complete road network, the internal nodes are used to represent subnetworks at different scales, and the leaf nodes are used to represent physical candidate road segments. Output hierarchical binary tree decision environment for road network The fingerprint features are formed by the flow corresponding to each tree node.

[0009] According to the semantically uncertain road network spatial location method described above, S3 includes: A one-hot encoded identity representation is generated for each agent. The local observation state of the agent at a binary tree node is the concatenation of the features of the left and right child nodes and the identity representation, where the local observation state satisfies the following relationship: ; in, One-hot encoding to characterize an intelligent agent, Features of the left node For the characteristics of the right node, the local observation state includes the number of road segments, traffic flow, and flow within the node's jurisdiction, which constitute fingerprint features. The agent performs binary selection in the action space of non-leaf nodes, and the decision terminates when the agent moves down from the root node to a leaf node. The leaf node reached by the agent is the facility deployment location. Semantic reward values ​​are sampled from the Beta distribution of the intercepted OD streams and evaluated sequentially by agent number. Rewards are given only to OD streams not intercepted by preceding agents; penalties are applied if subsequent agents select already occupied leaf nodes. The agent's reward... The calculation method is as follows: ; in, Represents intelligent agents The set of marginal flow contributions made to the system. For overlapping penalty terms, Represents OD flow The semantic basis reward for sampling.

[0010] According to the semantically uncertain road network spatial location method, in S4, each agent independently initializes a dual-head Critic network and an Actor network, and the network parameters are not shared; the dual-head Critic network simultaneously outputs the expected reward of the state. With return volatility ,in Represents intelligent agents The local observation status includes the number of road segments, traffic flow, and flow characteristics within the node's jurisdiction, forming a fingerprint feature.

[0011] According to the aforementioned road network spatial location method based on semantic uncertainty, in step S4, the calculation process of the uncertainty-aware risk-adjusted advantage function includes: Time difference error is calculated based on the output of a dual-headed Critic network. for: ; The generalized advantage estimate is calculated as follows: ; The risk-adjusted advantage function is calculated as follows: ; in, Represents intelligent agents exist Momentary rewards Represents intelligent agents exist Local observation state at a given time. This represents the forward offset of the time step. Represents intelligent agents exist Generalized advantage estimation of time, Represents intelligent agents exist The risk-adjusted advantage function at any given time. This represents the average of all return fluctuations within the current training batch. To prevent extremely small constants with a denominator of zero, As a discount factor, For GAE attenuation parameters, These are the parameters for controlling uncertainty.

[0012] According to the semantically uncertain road network spatial location method described above, in S4, the loss function for updating network parameters includes: The Critic network parameter update uses a Gaussian negative log-likelihood loss function. In each iteration, empirical samples of the agent's interaction with the environment are obtained. Based on these empirical samples, a preset number of iterations are performed to update the network parameters. The loss function of the Critic network satisfies the following relationship: ; The Actor network uses PPO to truncate the agent loss, and the loss satisfies the following relationship: ; Where B is the mini-batch size for each gradient update. The first in the batch The state of the sample, Adjust the advantage function for risk. The cumulative return target value is calculated based on the empirical trajectory. This represents the probability ratio between the old and new strategies. For the clipping function, This represents the strategy pruning factor.

[0013] Another aspect of the present invention provides a road network spatial location system based on semantic uncertainty, comprising: The first module is used to quantify the uncertainty of the occurrence probability of the target semantics of each OD stream based on the OD stream dataset and the POI dataset of the research area, and generate the Beta distribution shape parameters corresponding to each OD stream; The second module is used to construct a hierarchical binary tree decision environment based on urban road network topology data, OD trajectory dataset and Beta distribution shape parameters, and generate flow composition fingerprint features corresponding to each tree node. The third module is used to define the state space, action space, and marginal contribution reward mechanism based on sequence priority for multi-agent systems, based on the binary tree decision environment, Beta distribution shape parameters, and the number of facilities to be deployed. The fourth module is used to initialize the dual-head Critic network and Actor network of multi-agents. It generates experience trajectories through the interaction between agents and the environment, calculates the risk adjustment advantage function for uncertainty perception, updates network parameters based on Gaussian negative log-likelihood loss and PPO truncated agent loss, and outputs the final set of facility deployment road segments after iterative training until convergence.

[0014] The beneficial effects of this invention are as follows: This application calculates the semantic expectation value based on POI weighted density to infer the baseline probability of travel intention; simultaneously, by coupling the functional mixture degree of POIs with the total sample size to construct semantic confidence, the reliability of semantic estimation is quantified; this quantification method can accurately characterize the random fluctuation characteristics of the probability of target semantic occurrence, providing realistic uncertainty input for subsequent reward calculation and risk assessment in reinforcement learning, effectively avoiding location errors caused by semantic estimation bias; by aggregating and mapping the semantic features of macroscopic OD flows to microscopic road segments, a "flow composition fingerprint feature" of road segments is generated, enabling the agent to directly identify the OD flow component structure of different road segments, thus enhancing the model. The system possesses strong semantic awareness capabilities. By combining dual graphs with the Metis graph cutting algorithm, a hierarchical binary tree structure for the road network is constructed from top to bottom, transforming the original high-dimensional road network combination decision problem into a sequential decision problem on a binary tree. Each decision step only requires selection between two sub-regions, significantly reducing the dimensionality of the action space while ensuring the topological connectivity of the sub-regions. This allows the agent to efficiently find the optimal solution in the global solution space, avoiding the computational disaster caused by directly traversing all road segment combinations. Furthermore, by utilizing the prior agent's priority to occupy high-value key road segments, and the subsequent agent's proactive avoidance of overlapping coverage areas driven by marginal rewards, focusing on filling coverage blind spots, global collaborative optimization of multi-facility deployment is achieved. In addition, by introducing a dual-headed Critic network that outputs expected returns and returns fluctuations, and designing a risk-adjusted advantage function that perceives uncertainty, the optimization direction can be dynamically adjusted according to the uncertainty of the state during policy updates, avoiding aggressive updates in highly uncertain regions and significantly improving the stability and convergence efficiency of model training. By combining Gaussian negative log-likelihood loss and PPO truncated proxy loss, the accuracy of value estimation and the robustness of policy updates are further guaranteed. Attached Figure Description

[0015] Figure 1 This is a schematic flowchart of the road network spatial location method based on semantic uncertainty according to an embodiment of the present invention; Figure 2 This is a flowchart illustrating the model network structure and processing flow of an embodiment of the present invention. Figure 3 This is a block diagram of a road network spatial location system based on semantic uncertainty, according to an embodiment of the present invention. Detailed Implementation

[0016] The embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings. Throughout the description, the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions. In the following description, suffixes such as "module," "part," or "unit" used to denote elements are used only for the purpose of illustrative purposes and have no specific meaning in themselves. Therefore, "module," "part," or "unit" can be used interchangeably. Terms such as "first," "second," etc., are used only to distinguish technical features and should not be construed as indicating or implying relative importance, or implicitly indicating the number of indicated technical features, or implicitly indicating the sequential relationship of the indicated technical features. In the following description, the consecutive reference numerals for method steps are for ease of review and understanding. Adjusting the implementation order of steps, in conjunction with the overall technical solution of the present invention and the logical relationship between the various steps, will not affect the technical effect achieved by the technical solution of the present invention. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0017] Please combine Figure 1 and 2 , Figure 1 This is a schematic flowchart of the road network spatial location method based on semantic uncertainty according to an embodiment of the present invention. Figure 2 This is a flowchart illustrating the model network structure and processing flow of an embodiment of the present invention, wherein... Figure 1 Including but not limited to steps S1~S4: S1, based on the OD flow dataset and the POI dataset of the study area, quantifies the uncertainty of the occurrence probability of the target semantics of each OD flow and generates the Beta distribution shape parameter corresponding to each OD flow.

[0018] In some embodiments, step S1 is specifically shown as S11~S14 as follows: S11, Input the POI weight vector corresponding to the target semantics. The number of Points of Interest (POIs) in the neighborhood of the origin or destination of the OD flow and the total number of POIs in the neighborhood are counted. The expected value of the target semantic probability of the OD flow is calculated using the Sigmoid function. for: ; in, The first in the neighborhood of the OD flow The number of POIs The total number of POIs in the neighborhood, of which Total number of POI categories; S12, Calculate the POI functional mixing degree of the OD stream. for: ; ; in, This represents the total number of semantic categories of POIs covered within the region. Indicates the first Class POI in the first The percentage of the neighborhood of an OD flow; S13, Couple the POI mixture to the total number of samples to calculate the semantic confidence. for: ; in, This represents the total number of POIs within this neighborhood. S14, through reparameterization, maps the probability expectation and confidence level to shape parameters controlling the Beta distribution, calculated as follows: ; ; in, This represents the shape parameter that controls the Beta distribution.

[0019] This embodiment employs a Beta distribution to probabilistically represent the semantics of OD flows, overcoming the limitation of traditional methods that simplify semantic attributes to scalars. It can accurately characterize the random fluctuations in the probability of target semantic occurrence. Specifically, the semantic expectation value is calculated based on the POI weighted density. It can effectively infer the baseline probability of travel intentions through the functional characteristics of land parcels; and construct semantic confidence by combining the functional mixing degree of POIs and the total sample size. This achieves the quantification of the reliability of semantic estimation, especially when the region's POI type is simple (such as a bar cluster area) and there are sufficient samples. Higher values ​​indicate smaller variance in the Beta distribution, leading to more stable semantic estimation; when the POI mix in a region is high (e.g., a mixed commercial area), The values ​​are low, the variance is large, and the semantic uncertainty is higher. This quantification method provides realistic uncertainty input for subsequent reward calculation and risk assessment in reinforcement learning, avoiding location errors caused by semantic estimation bias.

[0020] S2 constructs a hierarchical binary tree decision environment based on urban road network topology data, OD trajectory dataset and Beta distribution shape parameters, and generates flow composition fingerprint features corresponding to each tree node.

[0021] In some embodiments, step S2 specifically includes S21 to S25: S21. Extract the start-point and end-point sets of the full OD trajectory, and use the K-Means algorithm to cluster the start-point and end-point sets to obtain the start-point cluster set. Clustering set with endpoint Each OD trajectory is encoded into a binary tuple based on the starting point cluster set and the ending point cluster set. ,in , , The total number of clusters starting from the starting point. The total number of clusters is the endpoint.

[0022] S22, statistically analyze the proportion of various OD flows on each road segment to generate the flow composition fingerprint characteristics of the road segment. for: ; in, This indicates that the person passed through this section of road and belonged to... Area to Number of trajectories in the region This represents the total number of all tracks that pass through this road segment.

[0023] S23 transforms the original directed road network into a dual graph, where road segments are mapped to dual nodes, and the topological connections between road segments are mapped to dual edges. The Metis graph cutting algorithm is then used to perform hierarchical partitioning of the dual graph, satisfying the following relationship: ; in, Indicates the strength of topological association. This represents the cutting cost, which is the sum of the weights of the dual edges that are cut. This represents minimizing the cost of cutting; and This represents two dual nodes in a dual graph. Represents the set of nodes in the subgraph obtained from the segmentation; Represents the edge ( () represents a dividing edge that spans two subgraphs. Let be the set of edges of the dual graph.

[0024] It should be noted that, in this embodiment, converting the original directed road network into a dual graph includes mapping road segments in the original directed road network to dual nodes and mapping the topological connection relationships between road segments in the original directed road network to dual edges.

[0025] S24, a hierarchical binary tree structure for building road networks from top to bottom. The binary tree structure The root node is used to represent the complete road network, the internal nodes are used to represent subnetworks at different scales, and the leaf nodes are used to represent physical candidate road segments.

[0026] S25, Output road network hierarchical binary tree decision environment The fingerprint features are formed by the flow corresponding to each tree node.

[0027] This embodiment addresses the issues of dimensionality explosion and semantic perception loss in reinforcement learning decision-making under large-scale road networks by constructing a road network environment with semantic-topological coupling. On one hand, the road segment flow composition fingerprint feature aggregates and maps the semantic features of macroscopic OD flow to specific road segments, enabling the agent to directly identify the OD flow component structure of different road segments. Ablation experiments verify that after removing the flow composition fingerprint feature, the model reward remains stagnant at a low level of about 150, failing to form an effective location strategy, proving that this feature is the core support for the model's semantic perception capability. On the other hand, the hierarchical binary tree constructed based on Metis graph cutting transforms the originally high-dimensional road network combination decision problem into a sequential decision problem on a binary tree. Each decision step only requires selection between two sub-regions, significantly reducing the dimensionality of the action space while ensuring the topological connectivity of the sub-regions. This allows the agent to efficiently find the optimal solution in the global solution space, avoiding the computational disaster of directly traversing all road segment combinations.

[0028] S3 defines the state space, action space, and marginal contribution reward mechanism based on sequence priority for multi-agent systems, based on the binary tree decision environment, Beta distribution shape parameters, and the number of facilities to be deployed.

[0029] In some embodiments, step S3 specifically includes S31 to S34: S31, Generate a one-hot encoded identity representation for each agent. The local observation state of the agent at a binary tree node is the concatenation of the features of the left and right child nodes and the identity representation, where the local observation state satisfies the following relationship: ; in, One-hot encoding to characterize an intelligent agent, Features of the left node The characteristics of the right node include the number of road segments, traffic flow, and flow within the node's jurisdiction, which constitute fingerprint features.

[0030] S32, the agent performs binary selection in the action space of non-leaf nodes, and the agent moves down layer by layer from the root node, and the decision terminates when it reaches a leaf node. The leaf node reached by the agent is the facility deployment location.

[0031] S33, semantic reward values ​​are sampled from the Beta distribution of the intercepted OD streams and evaluated sequentially by agent number. Only OD streams not intercepted by preceding agents are rewarded; if a subsequent agent selects an already occupied leaf node, a penalty is imposed. The agent's reward... The calculation method is as follows: ; in, Represents intelligent agents The set of marginal flow contributions made to the system. For overlapping penalty terms, Represents OD flow The semantic basis reward for sampling.

[0032] It should be noted that rewarding OD streams that are not intercepted by preceding agents means allocating corresponding rewards to a preset number of OD streams that have not been intercepted in sequence.

[0033] The technical solution in this embodiment achieves efficient collaborative site selection for multiple facilities through the targeted design of the three elements of Multi-Agent Development (MDP). The local observation strategy limits the perception range of the agent to the sub-region corresponding to the current tree node, avoiding the computational burden of processing the high-dimensional OD matrix of the entire road network; the binary action space is fully adapted to the binary tree decision environment, enabling the agent's movement to cover all candidate road segments and ensuring the completeness of the understanding space; the marginal contribution reward mechanism based on sequence priority fundamentally solves the credit allocation problem in multi-agent collaboration, with preceding agents prioritizing high-value key road segments, and subsequent agents actively avoiding overlapping coverage areas and focusing on filling coverage blind spots under the drive of marginal rewards.

[0034] S4 initializes the multi-agent dual-head Critic network and Actor network, generates experience trajectories through the interaction between the agents and the environment, calculates the risk adjustment advantage function for uncertainty perception, updates the network parameters based on Gaussian negative log-likelihood loss and PPO truncated agent loss, and outputs the final set of facility deployment road segments after iterative training until convergence.

[0035] In some embodiments, each agent independently initializes a dual-head Critic network and an Actor network, and the network parameters are not shared; the dual-head Critic network simultaneously outputs the expected reward of the state. With return volatility ,in Represents intelligent agents The local observation status includes the number of road segments, traffic flow, and flow characteristics within the node's jurisdiction, forming a fingerprint feature.

[0036] In some embodiments, the calculation process of the risk-adjusted advantage function based on uncertainty perception includes steps S41-43: S41, Calculation of time difference error based on dual-headed Critic network output for: ; S42, the generalized dominance estimate is calculated as follows: ; S43, the risk-adjusted advantage function is calculated as follows: ; in, Represents intelligent agents exist Momentary rewards Represents intelligent agents exist Local observation state at a given time. This represents the forward offset of the time step. Represents intelligent agents exist Generalized advantage estimation of time, Represents intelligent agents exist The risk-adjusted advantage function at any given time. This represents the average of all return fluctuations within the current training batch. To prevent extremely small constants with a denominator of zero, As a discount factor, For GAE attenuation parameters, These are the parameters for controlling uncertainty.

[0037] In some embodiments, the Critic network parameter update employs a Gaussian negative log-likelihood (GNLL) loss function. During each iteration, empirical samples of the agent's interaction with the environment are acquired, and a preset number of iterations are performed based on these empirical samples to update the network parameters. The Critic network loss function satisfies the following relationship: ; The Actor network uses PPO to truncate the agent loss, and the loss satisfies the following relationship: ; Where B is the mini-batch size for each gradient update. The first in the batch The state of the sample, Adjust the advantage function for risk. The cumulative return target value is calculated based on the empirical trajectory. This represents the probability ratio between the old and new strategies. For the clipping function, This represents the strategy pruning factor.

[0038] Understandably, each agent in this embodiment employs a dual-head Critic network with independent parameters, simultaneously addressing the scalability problem of centralized reinforcement learning and the problem of modeling payoff fluctuations in uncertain environments. On one hand, independent network parameters enable each agent to learn differentiated location strategies, avoiding the gradient fuzziness problem that occurs in the centralized value network of the MAPPO algorithm when the number of agents increases.

[0039] Figure 3 This is a block diagram of a road network spatial location system based on semantic uncertainty according to an embodiment of the present invention. The diagram includes a first module 31, a second module 32, a third module 33, and a fourth module 34.

[0040] The system comprises four modules: The first module quantifies the uncertainty of the target semantic occurrence probability of each OD flow based on the OD flow dataset and the POI dataset of the study area, generating the Beta distribution shape parameter for each OD flow; the second module constructs a hierarchical binary tree decision environment based on urban road network topology data, OD trajectory dataset, and Beta distribution shape parameter, generating the flow composition fingerprint feature corresponding to each tree node; the third module defines the state space, action space, and marginal contribution reward mechanism based on sequence priority for the multi-agent system based on the binary tree decision environment, Beta distribution shape parameter, and the number of facilities to be deployed; and the fourth module initializes the dual-head Critic network and Actor network of the multi-agent system, generates empirical trajectories through the interaction between the agents and the environment, calculates the risk adjustment advantage function for uncertainty perception, updates the network parameters based on Gaussian negative log-likelihood loss and PPO truncated agent loss, and outputs the final set of facility deployment road segments after iterative training until convergence.

[0041] For example, through the collaboration of the first to fourth modules, this embodiment of the invention can quantify the uncertainty of the target semantic occurrence probability of each OD flow based on the OD flow dataset and the POI dataset of the research area, generating the Beta distribution shape parameter corresponding to each OD flow; construct a hierarchical binary tree decision environment based on urban road network topology data, OD trajectory dataset, and Beta distribution shape parameter, and generate the flow composition fingerprint feature corresponding to each tree node; define the state space, action space, and marginal contribution reward mechanism based on sequence priority of the multi-agent based on the binary tree decision environment, Beta distribution shape parameter, and the number of facilities to be deployed; initialize the dual-head Critic network and Actor network of the multi-agent, generate experience trajectories through the interaction between the agent and the environment, calculate the risk adjustment advantage function for uncertainty perception, update the network parameters based on Gaussian negative log-likelihood loss and PPO truncated surrogate loss, and output the final set of facility deployment road segments after iterative training until convergence. Through the above methods, this application can solve the problems of low site selection efficiency and misalignment between site selection schemes and target requirements in existing road network spatial location methods.

[0042] Furthermore, although the invention has been described in the context of functional modules, it should be understood that, unless otherwise stated, one or more of the described functions and / or features may be integrated into a single physical device and / or software module, or one or more functions and / or features may be implemented in a separate physical device or software module. It is also understood that a detailed discussion of the actual implementation of each module is unnecessary for understanding the invention. Rather, considering the properties, functions, and internal relationships of the various functional modules in the apparatus disclosed in the embodiments of the invention, the actual implementation of the module will be understood within the scope of conventional skill of an engineer. Therefore, those skilled in the art can implement the invention as set forth in the claims using ordinary techniques without excessive experimentation. It is also understood that the specific concepts disclosed are merely illustrative and are not intended to limit the scope of the invention, which is determined by the full scope of the appended claims and their equivalents.

[0043] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0044] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can include, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0045] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0046] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0047] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0048] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

[0049] The above is a detailed description of the preferred embodiments of the present invention, but the present invention is not limited to the embodiments described. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the present invention, and these equivalent modifications or substitutions are all included within the scope defined by the claims of this application.

Claims

1. A road network spatial location method based on semantic uncertainty, characterized in that, include: S1, based on the OD flow dataset and the POI dataset of the study area, quantifies the uncertainty of the occurrence probability of the target semantics of each OD flow and generates the Beta distribution shape parameter corresponding to each OD flow; S2, based on urban road network topology data, OD trajectory dataset and Beta distribution shape parameters, constructs a hierarchical binary tree decision environment and generates flow composition fingerprint features corresponding to each tree node; S3 defines the state space, action space, and marginal contribution reward mechanism based on sequence priority for multi-agent systems, based on the binary tree decision environment, Beta distribution shape parameters, and the number of facilities to be deployed. S4. Initialize the dual-head Critic network and Actor network of multi-agents. Generate experience trajectories through the interaction between agents and the environment. Calculate the risk adjustment advantage function for uncertainty perception. Update network parameters based on Gaussian negative log-likelihood loss and PPO truncated agent loss. Iterate until convergence and output the final set of facility deployment road segments. S1 includes: Input target semantic corresponding POI weight vector The number of Points of Interest (POIs) in the neighborhood of the origin or destination of the OD flow and the total number of POIs in the neighborhood are counted. The expected value of the target semantic probability of the OD flow is calculated using the Sigmoid function. for: in, The first in the neighborhood of the OD flow The number of POIs The total number of POIs in the neighborhood, of which Total number of POI categories; Calculate the functional mixing degree of POI in the OD stream for: in, This represents the total number of semantic categories of POIs covered within the region. Indicates the first Class POI in the first The percentage of the neighborhood of an OD flow; Couple the POI mixture to the total number of samples to calculate semantic confidence. for: in, This represents the total number of POIs within this neighborhood. By reparameterizing, the expected probability and confidence level are mapped to shape parameters controlling the Beta distribution. The calculation method is as follows: in, This represents the shape parameter that controls the Beta distribution; S2 includes: The start and end sets of the full OD trajectories are extracted, and the K-Means algorithm is used to cluster the start and end sets to obtain the start cluster set. Clustering set with endpoint Each OD trajectory is encoded into a binary tuple based on the starting point cluster set and the ending point cluster set. ,in , , The total number of clusters starting from the starting point. The endpoint is the total number of clusters; The proportion of different types of origin-destination (OD) flows on each road segment is statistically analyzed to generate the flow fingerprint characteristics of the road segment. for: in, This indicates that the person passed through this section of road and belonged to... Area to Number of trajectories in the region This represents the total number of all tracks passing through this road segment; The original directed road network is transformed into a dual graph, where road segments are mapped to dual nodes, and the topological connections between road segments are mapped to dual edges. The Metis graph cutting algorithm is then used to perform hierarchical partitioning of the dual graph, which satisfies the following relationship: in, Indicates the strength of topological association. This represents the cutting cost, which is the sum of the weights of the dual edges that are cut. This represents minimizing the cost of cutting; and This represents two dual nodes in a dual graph. Represents the set of nodes in the subgraph obtained from the segmentation; Represents the edge ( () represents a dividing edge that spans two subgraphs. Let be the set of edges of the dual graph; Hierarchical binary tree structure for building road networks from top to bottom The binary tree structure The root node is used to represent the complete road network, the internal nodes are used to represent subnetworks at different scales, and the leaf nodes are used to represent physical candidate road segments. Output hierarchical binary tree decision environment for road network The fingerprint features are formed by the flow corresponding to each tree node.

2. The road network spatial location method based on semantic uncertainty according to claim 1, characterized in that, S3 includes: A one-hot encoded identity representation is generated for each agent. The local observation state of the agent at a binary tree node is the concatenation of the features of the left and right child nodes and the identity representation, where the local observation state satisfies the following relationship: in, One-hot encoding to characterize an intelligent agent, Features of the left node For the characteristics of the right node, the local observation state includes the number of road segments, traffic flow, and flow within the node's jurisdiction, which constitute fingerprint features. The agent performs binary selection in the action space of non-leaf nodes, and the decision terminates when the agent moves down from the root node to a leaf node. The leaf node reached by the agent is the facility deployment location. Semantic reward values ​​are sampled from the Beta distribution of the intercepted OD streams and evaluated sequentially by agent number. Rewards are given only to OD streams not intercepted by preceding agents; penalties are applied if subsequent agents select already occupied leaf nodes. The agent's reward... The calculation method is as follows: in, Represents intelligent agents The set of marginal flow contributions made to the system. For overlapping penalty terms, Represents OD flow The semantic basis reward for sampling.

3. The road network spatial location method based on semantic uncertainty according to claim 1, characterized in that, In step S4, each agent independently initializes a dual-head Critic network and an Actor network, and the network parameters are not shared; the dual-head Critic network simultaneously outputs the expected reward of the state. With return volatility ,in Represents intelligent agents The local observation status includes the number of road segments, traffic flow, and flow characteristics within the node's jurisdiction, forming a fingerprint feature.

4. The road network spatial location method based on semantic uncertainty according to claim 3, characterized in that, In step S4, the calculation process of the risk-adjusted advantage function based on uncertainty perception includes: Time difference error is calculated based on the output of a dual-headed Critic network. for: The generalized advantage estimate is calculated as follows: The risk-adjusted advantage function is calculated as follows: in, Represents intelligent agents exist Momentary rewards Represents intelligent agents exist Local observation state at a given time. This represents the forward offset of the time step. Represents intelligent agents exist Generalized advantage estimation of time, Represents intelligent agents exist The risk-adjusted advantage function at any given time. This represents the average of all return fluctuations within the current training batch. To prevent extremely small constants with a denominator of zero, As a discount factor, For GAE attenuation parameters, These are the parameters for controlling uncertainty.

5. The road network spatial location method based on semantic uncertainty according to claim 4, characterized in that, In step S4, the loss function for updating network parameters includes: The Critic network parameter update uses a Gaussian negative log-likelihood loss function. In each iteration, empirical samples of the agent's interaction with the environment are obtained. Based on these empirical samples, a preset number of iterations are performed to update the network parameters. The loss function of the Critic network satisfies the following relationship: The Actor network uses PPO to truncate the agent loss, and the loss satisfies the following relationship: Where B is the mini-batch size for each gradient update. The first in the batch The state of the sample, Adjust the advantage function for risk. The cumulative return target value is calculated based on the empirical trajectory. This represents the probability ratio between the old and new strategies. For the clipping function, This represents the strategy pruning factor.

6. A road network spatial location system based on semantic uncertainty according to any one of claims 1-5, characterized in that, include: The first module is used to quantify the uncertainty of the occurrence probability of the target semantics of each OD stream based on the OD stream dataset and the POI dataset of the research area, and generate the Beta distribution shape parameters corresponding to each OD stream; The second module is used to construct a hierarchical binary tree decision environment based on urban road network topology data, OD trajectory dataset and Beta distribution shape parameters, and generate flow composition fingerprint features corresponding to each tree node. The third module is used to define the state space, action space, and marginal contribution reward mechanism based on sequence priority for multi-agent systems, based on the binary tree decision environment, Beta distribution shape parameters, and the number of facilities to be deployed. The fourth module is used to initialize the dual-head Critic network and Actor network of multi-agents. It generates experience trajectories through the interaction between agents and the environment, calculates the risk adjustment advantage function for uncertainty perception, updates network parameters based on Gaussian negative log-likelihood loss and PPO truncated agent loss, and outputs the final set of facility deployment road segments after iterative training until convergence.

Images