Automated parallel traffic simulation analysis method and device

Abstract

The present disclosure relates to an automated parallel traffic simulation analysis method and device. The method comprises: in response to receiving a region delineation instruction, generating an initial digitized road network from a map data source, and performing a topology calibration and connectivity repair operation on the initial digitized road network to generate a digitized road network file; generating structured traffic demand data based on user input; generating a basic traffic scenario file based on the digitized road network file and the traffic demand data, and forming a traffic simulation running scenario based on user configuration of the basic traffic scenario file; generating a plurality of signal timing schemes based on the traffic simulation running scenario; performing traffic simulation based on the plurality of signal timing schemes, analyzing the simulation results to obtain multi-level key performance indicators, visualizing and comparing the multi-level key performance indicators, and generating a natural language evaluation report using a large language model. Thus, the simulation efficiency, ease of use and flexibility of scenario construction can be significantly improved.

Description

Technical Field

[0001] This disclosure relates to the field of intelligent transportation simulation technology, and more specifically, to an automated parallel traffic simulation analysis method and apparatus. Background Technology

[0002] The refined operation and congestion management of urban transportation systems have become core issues. Traffic simulation is a key technology for achieving this goal, providing a scientific basis for the formulation and evaluation of traffic management strategies by simulating real traffic flow in digital space. Among existing technologies, simulation engines such as Urban Traffic Simulation (SUMO) provide powerful underlying computing capabilities and have developed traffic management platforms integrating graphical user interfaces (GUIs). While these platforms integrate modules such as road network configuration, traffic flow generation, signal control, and result analysis, they still have the following limitations in achieving efficient and intelligent traffic analysis and decision-making:

[0003] First, the simulation configuration is complex and the user threshold is high: Traditional traffic simulation software usually requires users to have a strong background in traffic engineering and to perform a lot of tedious manual configuration work, including detailed road network drawing, writing traffic flow path files, and inputting traffic light timing schemes. The entire preparation process is time-consuming and laborious, which seriously hinders the promotion of simulation technology and its application in grassroots traffic management departments.

[0004] Second, the scene generation capability is limited and the flexibility is poor: existing simulation platforms mostly rely on structured parameters or preset templates, which lack flexibility in scene generation. When managers want to test specific, dynamic scenarios, such as "simulating the traffic situation on a main road during holidays after one lane is closed due to construction," existing technologies cannot quickly generate the corresponding simulation scenario through simple natural language commands. Multiple parameters need to be manually adjusted, resulting in low scene construction efficiency.

[0005] Third, the results analysis is time-consuming and lacks depth: After the simulation, the platform usually outputs massive amounts of multi-dimensional performance index data, such as average delay and queue length. Managers need to spend a lot of time interpreting and comparing this raw data, and the analysis focuses more on "which solution is better" without a deep analysis of the root causes behind traffic bottlenecks, making it difficult to form an effective closed loop of optimization strategies.

[0006] Fourth, algorithm integration and comparison are inconvenient: different signal timing optimization algorithms (e.g., the classic Webster method, green wave coordinated control, reinforcement learning-based adaptive control, etc.) are often scattered across different research or commercial software. If users want to compare the advantages and disadvantages of multiple optimization strategies under the same road network and traffic scenario, they usually need to switch between multiple systems, resulting in fragmented operation processes and low comparison efficiency. Summary of the Invention

[0007] There is an urgent need in this field for a traffic simulation technology that can overcome the above-mentioned limitations, realize full-process automation from road network generation and scenario configuration to scheme evaluation, and support virtual-real interaction between the simulation system and the actual system to form a parallel execution closed loop, providing deep intelligent analysis and proactive guidance capabilities for the actual traffic system.

[0008] To address the aforementioned issues, this disclosure proposes an automated parallel traffic simulation analysis method and apparatus, a computing system, and a computer-readable storage medium.

[0009] According to one aspect of this disclosure, an automated parallel traffic simulation analysis method is provided, comprising: in response to receiving a region delineation instruction, generating an initial digital road network from a map data source, and performing topology calibration and connectivity repair operations on the initial digital road network to generate a digital road network file conforming to the simulation engine specifications; generating structured traffic demand data based on user input, wherein the input includes at least one of preset traffic flow patterns, measured traffic flow data, custom drawing, origin-endpoint matrix files, and natural language descriptions, and the traffic demand data includes traffic flow intensity, time period distribution, and vehicle travel patterns; based on... The digitized road network file and the traffic demand data are used to generate a basic traffic scenario file. Based on the user's configuration of the road network and traffic demand parameters included in the basic traffic scenario file, a traffic simulation operation scenario is formed. Based on the traffic simulation operation scenario, multiple signal timing schemes are generated using various optimization algorithms. Traffic simulation is performed based on the multiple signal timing schemes. The simulation results of the multiple signal timing schemes are analyzed to obtain multi-level key performance indicators for each of the multiple signal timing schemes. The multi-level key performance indicators of the multiple signal timing schemes are visualized and compared, and a natural language evaluation report is generated using a large language model.

[0010] Optionally, the area delineation instruction is generated when a user searches for a geographical location name and selects a preset city road network template through the map interaction interface, uses the polygon tool to customize and draw a closed geographical area through the map interaction interface, or selects and loads previously stored digital road network data from a saved road network list.

[0011] Optionally, the topology calibration and connectivity repair operations include: identifying and processing dead ends, dangling edges, and erroneously segmented weakly connected components; performing node merging, geometric resampling, and gap bridging; and performing format and topology consistency verification.

[0012] Optionally, the steps for generating structured traffic demand data based on user input include: responding to user input as a preset traffic flow pattern, generating traffic demand data based on the daily traffic volume distribution curve and the simulation period and traffic flow intensity specified by the user; responding to user input as measured traffic flow data, generating traffic demand data based on the measured traffic flow data using a data back-calculation algorithm; responding to user input as custom drawing, providing an interactive drawing interface to facilitate user drawing of traffic flow curves, and generating corresponding traffic demand data based on the user-drawn traffic flow curves; responding to user input as a start-endpoint matrix file, verifying the file format of the start-endpoint matrix file and converting it into traffic demand data; responding to user input as a natural language description, using a large language model to perform semantic parsing on the input natural language description and generating structured traffic demand data.

[0013] Optionally, the various optimization algorithms include at least two of Webster's method, green band coordination, Synchro optimization, optimization algorithms based on large language models, and reinforcement learning algorithms.

[0014] Optionally, when the input includes a natural language description and the multiple optimization algorithms include optimization algorithms based on a large language model, the step of generating multiple signal timing schemes using multiple optimization algorithms includes: parsing the optimization intent from the user's natural language description, the optimization intent including at least one of optimization objectives based on road corridors, optimization objectives based on specific intersections, and optimization objectives based on global performance; passing the digital road network file and the traffic demand data as input data to the large language model; and performing global optimization calculations through the large language model based on the parsed optimization intent and the input data to output signal timing schemes defining the cycle, phase, and green ratio of each intersection.

[0015] Optionally, the step of visually comparing the multi-level key performance indicators of the multiple signal timing schemes and generating a natural language evaluation report using a large language model includes: in response to a user selecting two or more signal timing schemes, extracting and aligning the key performance indicators at the road network and intersection levels of the selected signal timing schemes; generating and displaying visual charts for comparing the key performance indicators based on the extracted key performance indicators; and using a large language model to comprehensively evaluate the key performance indicators to output the natural language evaluation report, which includes quantitative analysis, comparison of advantages and disadvantages, and improvement suggestions.

[0016] According to another aspect of this disclosure, an automated parallel traffic simulation analysis device is provided, comprising: a road network generation unit configured to, in response to receiving a region delineation instruction, generate an initial digital road network from a map data source, and perform topology calibration and connectivity repair operations on the initial digital road network to generate a digital road network file conforming to the simulation engine specifications; a traffic demand generation unit configured to generate structured traffic demand data based on user input, wherein the input includes at least one of preset traffic flow patterns, measured traffic flow data, custom drawing, origin-endpoint matrix files, and natural language descriptions, and the traffic demand data includes traffic flow intensity, time period distribution, and vehicle travel patterns; and a traffic scene configuration unit. The system is configured to generate a basic traffic scenario file based on the digitized road network file and the traffic demand data, and to form a traffic simulation operation scenario based on the user's configuration of the road network and traffic demand parameters included in the basic traffic scenario file; the signal timing optimization unit is configured to generate multiple signal timing schemes based on the traffic simulation operation scenario using various optimization algorithms; the scheme evaluation unit is configured to perform traffic simulation based on the multiple signal timing schemes, analyze the simulation results of the multiple signal timing schemes to obtain multi-level key performance indicators for each of the multiple signal timing schemes, perform a visual comparison of the multi-level key performance indicators of the multiple signal timing schemes, and generate a natural language evaluation report using a large language model.

[0017] Optionally, the area delineation instruction is generated when a user searches for a geographical location name and selects a preset city road network template through the map interaction interface, uses the polygon tool to customize and draw a closed geographical area through the map interaction interface, or selects and loads previously stored digital road network data from a saved road network list.

[0018] Optionally, the topology calibration and connectivity repair operations include: identifying and processing dead ends, dangling edges, and erroneously segmented weakly connected components; performing node merging, geometric resampling, and gap bridging; and performing format and topology consistency verification.

[0019] Optionally, the traffic demand generation unit is further configured to: generate traffic demand data based on a daily traffic volume distribution curve and the simulation period and traffic flow intensity specified by the user, in response to a preset traffic flow pattern; generate traffic demand data based on measured traffic flow data using a data back-calculation algorithm, in response to measured traffic flow data, in response to user input; provide an interactive drawing interface for users to draw traffic flow curves, and generate corresponding traffic demand data based on the user-drawn traffic flow curves, in response to user input; verify the file format of the origin-endpoint matrix file and convert it into traffic demand data, in response to user input; and perform semantic parsing of the input natural language description using a large language model, in response to user input, in order to generate structured traffic demand data.

[0020] Optionally, the various optimization algorithms include at least two of Webster's method, green band coordination, Synchro optimization, optimization algorithms based on large language models, and reinforcement learning algorithms.

[0021] Optionally, when the input includes a natural language description and the multiple optimization algorithms include optimization algorithms based on a large language model, the signal timing optimization unit is further configured to: parse the optimization intent from the user's natural language description, the optimization intent including at least one of optimization objectives based on road corridors, optimization objectives based on specific intersections, and optimization objectives based on global performance; pass the digital road network file and the traffic demand data as input data to the large language model; and perform global optimization calculations based on the parsed optimization intent and the input data through the large language model, outputting a signal timing scheme defining the cycle, phase, and green ratio of each intersection.

[0022] Optionally, the scheme evaluation unit is further configured to: extract and align key performance indicators at the road network and intersection levels of the selected signal timing schemes in response to the user selecting two or more signal timing schemes; generate and display visualization charts for comparing the key performance indicators based on the extracted key performance indicators; and perform a comprehensive evaluation of the key performance indicators using a large language model to output the natural language evaluation report, which includes quantitative analysis, comparison of advantages and disadvantages, and improvement suggestions.

[0023] According to another aspect of this disclosure, a computing system is provided that includes at least one computing device and at least one storage device for storing instructions, wherein the instructions, when executed by the at least one computing device, cause the at least one computing device to perform the automated parallel traffic simulation analysis method as described above.

[0024] According to another aspect of this disclosure, a computer-readable storage medium is provided for storing instructions, wherein when the instructions are executed by at least one computing device, the at least one computing device causes the at least one computing device to perform the automated parallel traffic simulation analysis method as described above.

[0025] By adopting this disclosure, simulation efficiency and ease of use can be significantly improved. Through full-process automation and innovative natural language command interaction, users are freed from tedious parameter configuration, greatly lowering the barrier to entry and enabling non-professionals to quickly build and run complex traffic simulation tasks. It enables highly flexible scenario construction by utilizing the semantic understanding capabilities of a large language model to automatically convert from macroscopic, qualitative natural language descriptions to microscopic, quantitative simulation parameters, making simulation scenario construction more intuitive, rapid, and flexible. It deepens the decision-making value of simulation results by introducing intelligent analysis and multi-scheme comparison functions from the large language model, elevating simulation results from "data presentation" to "comprehensive evaluation," revealing the advantages and disadvantages of different schemes and directions for improvement, providing traffic managers with intuitive, data-driven decision support. It breaks through the traditional mirror model of static simulation by introducing the concept of parallel systems, establishing a parallel execution mechanism of virtual-real interaction between the simulation system and the actual system, enabling optimization and evaluation to form a closed-loop iteration, driving the continuous evolution of the simulation system and signal control schemes. Attached Figure Description

[0026] The above and / or other objects and advantages of this disclosure will become clearer from the following description of embodiments in conjunction with the accompanying drawings, wherein:

[0027] Figure 1 This is a flowchart illustrating an automated parallel traffic simulation analysis method according to an exemplary embodiment of the present disclosure;

[0028] Figure 2 This is a schematic diagram illustrating the overall framework of an automated traffic simulation analysis system according to exemplary embodiments of the present disclosure;

[0029] Figure 3 This is a flowchart illustrating an automated traffic network generation method according to an exemplary embodiment of the present disclosure;

[0030] Figure 4 This is a flowchart illustrating an automated traffic demand generation method according to an exemplary embodiment of the present disclosure;

[0031] Figure 5 This is a flowchart illustrating an automated traffic scene configuration and editing method according to exemplary embodiments of the present disclosure;

[0032] Figure 6 This is a flowchart illustrating an automated signal timing optimization method according to an exemplary embodiment of the present disclosure;

[0033] Figure 7 This is a flowchart illustrating an automated traffic control scheme evaluation method according to exemplary embodiments of the present disclosure;

[0034] Figure 8 This is a block diagram illustrating an automated parallel traffic simulation analysis apparatus according to an exemplary embodiment of the present disclosure;

[0035] Figure 9 This is a block diagram illustrating a computing system including at least one computing device and at least one storage device of storage instructions according to an exemplary embodiment of the present disclosure. Detailed Implementation

[0036] The following description, in conjunction with the accompanying drawings, provides specific embodiments to aid the reader in gaining a comprehensive understanding of the methods, apparatus, and / or systems described herein. However, upon understanding this disclosure, various changes, modifications, and equivalents of the methods, apparatus, and / or systems described herein will become apparent. For example, the order of operations described herein is merely illustrative and is not limited to those orders set forth herein, but may be altered as will become clear upon understanding this disclosure, except for operations that must occur in a specific order. Furthermore, for clarity and conciseness, descriptions of features known in the art may be omitted.

[0037] To address the problems of complex simulation configuration, difficulty in scene generation, and insufficient result analysis in existing technologies, this disclosure proposes an automated parallel traffic simulation and signal control optimization method. This method uses natural language commands and a visual interface to achieve full-process automation from traffic modeling and control optimization to scheme evaluation, and supports virtual-real interaction between simulation and actual systems.

[0038] Figure 1 This is a flowchart illustrating an automated parallel traffic simulation analysis method according to an exemplary embodiment of the present disclosure.

[0039] like Figure 1As shown, in step S101, in response to receiving a region delineation instruction, an initial digitized road network is generated from a map data source (e.g., OpenStreetMap), and topology calibration and connectivity repair operations are performed on the initial digitized road network to generate a digitized road network file that conforms to the simulation engine specifications. In the example, the region delineation instruction is generated when the user searches for a geographic location name and selects a preset city road network template through the map interactive interface, uses the polygon tool to customize and draw a closed geographic area through the map interactive interface, or selects and loads previously stored digitized road network data from a saved road network list. In the example, the topology calibration and connectivity repair operations include: identifying and processing dead-end roads, hanging edges, and incorrectly segmented weakly connected components; performing node merging, geometric resampling, and gap bridging; and performing format and topology consistency checks. This step aims to solve the problem of complex manual modeling by automatically generating an initial digitized road network from a map data source by receiving a region delineation instruction from the user on, for example, a graphical user interface (GUI), and further performing topology calibration and connectivity repair to generate a logically consistent road network data that can be directly used for simulation.

[0040] In step S102, structured traffic demand data is generated based on user input. User input includes at least one of the following: preset traffic flow patterns, measured traffic flow data, custom plots, origin-endpoint matrix files, and natural language descriptions. Traffic demand data includes traffic flow intensity, time period distribution, and vehicle travel patterns. In this example, in response to the user input of a preset traffic flow pattern, traffic demand data is generated based on the daily traffic volume distribution curve and the user-specified simulation time period and traffic flow intensity. In response to the user input of measured traffic flow data, traffic demand data is generated based on the measured traffic flow data using a data back-calculation algorithm. In response to the user input of custom plots, an interactive plotting interface is provided to allow users to plot traffic flow curves, and corresponding traffic demand data is generated based on the user-plotted traffic flow curves. In response to the user input of an origin-endpoint matrix file, the file format of the origin-endpoint matrix file is verified and converted into traffic demand data. In response to the user input of natural language descriptions, a large language model is used to perform semantic parsing on the input natural language descriptions and generate structured traffic demand data. This step aims to address the issue of inflexible scene generation, supporting multiple generation methods (including preset traffic flow patterns, reverse engineering from measured data, custom-drawn traffic flow curves, OD matrix upload, and parsing natural language descriptions through Large Language Model (LLM), thereby transforming the user's high-level intent into accurate and executable simulation input.

[0041] In step S103, a basic traffic scenario file is generated based on the digital road network file and traffic demand data. Furthermore, a traffic simulation operation scenario is formed based on the user's configuration of the road network and traffic demand parameters included in the basic traffic scenario file. In addition, during the formation of the traffic simulation operation scenario, the user can also configure at least one of the following: simulation time period, key areas, traffic events, and weather conditions. This step generates a traffic scenario file based on the road network and demand, which includes at least the road network and traffic demand. It can be further expanded to include elements such as simulation time period, key areas, traffic events, and weather conditions. Users can choose the default configuration or personalize the existing scenario through the interactive interface (e.g., adjust traffic flow (e.g., "increase overall traffic flow by 30%" or "set truck ratio to 20%)", set simulation time period (e.g., "set simulation time to morning peak 7:00–9:00"), add construction or accident events (e.g., "set a two-way road occupancy construction" or "add temporary closure due to an accident at a specific intersection"), simulate rain and snow weather (e.g., "set rainfall intensity to moderate rain during simulation" or "set visibility reduction to 50%)), etc.) to form a traffic simulation operation scenario that meets specific research objectives. These configurations are expanded and evolved in the simulation system, supporting comparison with actual traffic operation status for continuous correction of the simulation model.

[0042] In step S104, based on the traffic simulation operation scenario, multiple signal timing schemes are generated using various optimization algorithms. In the example, the various optimization algorithms include at least two of Webster's method, green wave coordination, Synchro optimization method, optimization algorithms based on large language models, and reinforcement learning algorithms. In the example, when the user's input includes a natural language description and the various optimization algorithms include optimization algorithms based on large language models, the step of generating multiple signal timing schemes using various optimization algorithms includes: parsing the optimization intent from the user's natural language description, the optimization intent including at least one of the following: optimization objectives based on road corridors (e.g., prioritizing green wave passage on specific arterial roads), optimization objectives based on specific intersections (e.g., focusing on optimizing pedestrian waiting time at specific intersections), and optimization objectives based on global performance (e.g., reducing overall average delay while ensuring arterial road traffic efficiency); passing the digitized road network file and traffic demand data as input data to the large language model; and performing global optimization calculations based on the parsed optimization intent and input data through the large language model, outputting signal timing schemes that define the cycle, phase, and green ratio of each intersection. For example, in response to a user selecting the default signal timing mode, a baseline signal timing scheme is calculated based on the structural attributes of the digital road network and generated traffic demand parameters. In response to a user selecting an optimization mode, multiple optimization algorithms are integrated within a unified framework to automatically optimize signal timing according to road network characteristics and traffic demand, generating optimized signal timing schemes or dynamic control strategies. In response to a user selecting a custom mode, a custom signal timing file conforming to a preset format is received or edited and applied to the simulation. Furthermore, the optimization process is not a single run, but rather an iterative process through computational experiments in the simulation system, feeding the optimization results back to the actual system, forming a cyclical optimization that interacts with the real system. For example, a unified framework integrates multiple optimization algorithms and is divided into two execution modes: offline optimization mode, where, before simulation execution, based on the topology of the digital road network and user-generated traffic demand parameters, Webster optimization, green wave optimization, Synchro optimization, or signal timing optimization algorithms based on a large language model are invoked to calculate and generate a static, optimized signal timing scheme in one go; and online optimization mode, where, during simulation execution, reinforcement learning algorithms interact with the simulation engine in real time, dynamically and iteratively adjusting the signal timing scheme based on road network characteristics and changes in traffic demand. When using the signal timing optimization algorithm based on a large language model in the offline optimization mode, its implementation mechanism includes: taking the topology of the digital road network, intersection geometry information, and user-generated traffic demand parameters as input and passing them to the large language model; the model, based on its built-in traffic engineering knowledge and logical reasoning capabilities, performs global optimization calculations and outputs a complete optimized signal timing scheme that defines the cycle, phase, and green ratio of each intersection.When the signal optimization strategy parameters are reinforcement learning adaptive optimization, optimization refers to a method that dynamically performs real-time interaction between the simulation engine and one or more reinforcement learning agents during the simulation execution process. This includes: fusing digital road network data, traffic flow parameters, and initial signal timing schemes to start the simulation; during the simulation execution process, the reinforcement learning agent periodically obtains traffic conditions from the simulation engine as observations and outputs actions to adjust the signal timing, thereby achieving adaptive optimization based on real-time traffic conditions.

[0043] This step aims to solve the problem of the complexity of manual signal timing optimization. It can integrate multiple optimization algorithms under a unified framework and can flexibly execute offline optimization mode (e.g., calling Webster optimization, green band optimization, Synchro optimization, or large language model-based optimization algorithms to generate a static scheme in one go) or online optimization mode (e.g., dynamically generating optimization schemes through real-time interaction with the simulation engine via reinforcement learning) based on the optimization strategy specified by the user or the optimization intent parsed from natural language instructions, in order to obtain an optimized signal timing scheme or dynamic control strategy.

[0044] In step S105, traffic simulation is performed based on multiple signal timing schemes. The simulation results of the multiple signal timing schemes are analyzed to obtain multi-level key performance indicators (KPIs) for each of the multiple signal timing schemes. The multi-level KPIs of the multiple signal timing schemes are visualized and compared, and a natural language evaluation report is generated using a large language model. In the example, in response to the user selecting two or more signal timing schemes, the network-level and intersection-level KPIs of the selected signal timing schemes are extracted and aligned. Based on the extracted KPIs, a visualization chart for comparing the KPIs is generated and displayed. Using a large language model, a comprehensive evaluation of the KPIs is performed to output a natural language evaluation report, which includes quantitative analysis, comparison of advantages and disadvantages, and improvement suggestions. For example, it can store the results of at least two independent simulation runs, which are conducted under the same traffic network and traffic demand conditions but with different signal timing schemes; in response to the user selecting two or more stored signal timing schemes, it can automatically extract and align the key performance indicators of these schemes at the network and intersection levels; generate and display visualization charts (e.g., radar charts or bar charts) for comparing the performance of the schemes; and call a large language model to comprehensively evaluate the quantitative comparison indicators of the schemes and output a natural language report containing advantages and disadvantages analysis and improvement suggestions.

[0045] This step aims to address the cumbersome and incomplete nature of manual analysis and evaluation of different signal timing schemes. After simulation, it automatically performs quantitative statistics on multi-dimensional performance indicators and uses a large language model to compare and analyze simulation results, providing a comprehensive analysis of the results of one or more signal timing schemes. It can compare the key performance indicators of different schemes and generate a natural language report containing advantages, disadvantages, and conclusions, thus intelligently simplifying the tedious data interpretation process and providing intuitive, data-driven support for decision-making. Furthermore, the evaluation results can be used to guide signal timing in actual traffic systems, while also supporting iterative optimization of actual system timing schemes within the simulation system, thus forming a parallel execution closed loop between the simulation and actual systems.

[0046] By adopting the automated parallel traffic simulation analysis method according to exemplary embodiments of this disclosure, the complexity of traffic simulation modeling and the need for manual intervention are reduced, the efficiency and accuracy of signal timing optimization are improved, and the performance of the scheme can be objectively and comprehensively evaluated through quantitative indicators. Furthermore, it can achieve end-to-end automation and parallel execution from simulation environment construction to scheme optimization and decision support report generation within the virtual-real interaction framework of the simulation system and the actual system. It can be widely applied to the research and engineering practice of intelligent traffic management and intelligent signal control.

[0047] Figure 2 This is a schematic diagram illustrating the overall framework of an automated traffic simulation analysis system according to exemplary embodiments of the present disclosure.

[0048] Referring to Figure 2, the automated traffic simulation analysis system according to an exemplary embodiment of this disclosure adopts a front-end and back-end separation technical architecture, which can be logically divided into an interaction and business layer and a support and engine layer. The interaction and business layer is the core of the system, responsible for handling business logic and user interaction, and includes a front-end interaction layer 21 and a back-end service layer 22.

[0049] The front-end interaction layer 21 serves as the direct entry point for user interaction with the system, providing a graphical user interface (GUI) (including components such as the graphical interface, administrator system, and result display). In this example, the front-end interaction layer 21 can be developed based on the Vue.js framework and the Element Plus component library, managed through Pinia for state management, and establishes real-time communication with the back-end service layer 22 via WebSocket to support dynamic monitoring and data display of the simulation process.

[0050] The backend service layer 22 serves as the core business logic processing center, responding to frontend requests and scheduling underlying resources. In this example, the backend service layer 22 can be built based on the Python Flask framework. Furthermore, the backend service layer 22 may include a road network generation module 23, a traffic demand generation module 24, a traffic scenario configuration and editing module 25, a signal timing optimization module 26, and a signal control scheme evaluation module 27. These modules can interact with the frontend by providing a RESTful API.

[0051] The support and engine layer provides basic computing capabilities, data storage, and external service calls for upper-layer business applications. The support and engine layer includes a simulation engine (28), data storage (29), and third-party service interfaces (30).

[0052] The simulation engine 28, as the core computing unit at the platform's bottom layer, provides fundamental capabilities such as road network analysis, traffic flow simulation, and signal control simulation. It supports both the verification of traffic demand generation and signal optimization and scheme evaluation. In one embodiment, the open-source microscopic traffic simulation software SUMO can be used, and its TraCI interface can be used to interact with the backend service layer 22 in real time (e.g., transmitting vehicle position and speed, or receiving traffic light status adjustment commands). This allows for computational experiments within the simulation system and supports online parallel simulations by comparing the simulation with the actual system's operating status.

[0053] Data storage 29 is responsible for persistently storing various types of data required for the system's operation. In this example, an SQLite database can be used to store road network metadata (e.g., ID, name, region GeoJSON), user operation logs, and other structured data; a file system can be used to store preset road network templates, user-defined road network files, and temporary simulation result files.

[0054] Third-party service interface 30 is used to call external professional services to enhance system functionality. For example, it can call map services such as Mapbox API or Gaode Map API to provide the front end with base map display and geographic information interaction capabilities; it can call Large Language Model (LLM) API to support natural language generation for traffic scenarios and intelligent analysis of simulation results; it can also reserve data interfaces with actual traffic management systems to feed back the optimization results of the simulation system to the actual system, forming a closed-loop execution of virtual and real interaction, and supporting the implementation of parallel traffic control.

[0055] Figure 3 This is a flowchart illustrating an automated traffic network generation method according to an exemplary embodiment of the present disclosure.

[0056] As described above, users can input area delineation commands in a variety of flexible ways, for example, through the map interaction interface provided by the front-end interaction layer 21.

[0057] like Figure 3 As shown, the automated traffic network generation method according to an exemplary embodiment of this disclosure begins at step S301. Next, the source of the area delineation instruction can be determined in step S302. In one example, the user can select a preset road network (S303), for example, by selecting a preset city road network template from a drop-down list in the map interface. At this point, the corresponding pre-processed road network file (e.g., a .osm format file) can be directly loaded from the data storage 29. In another example, the user can customize the drawing area (S304). For example, the user draws a custom closed geographical area on the map using a polygon tool. After drawing, the front-end application sends the vertex geographic coordinate sequence of the polygon (e.g., a GeoJSON format polygon object) to the back-end service layer 22. In another example, the user can choose to load a pre-stored road network (S305). For example, the user calls a saved road network list through the interface, which is read from an SQLite database in the data storage 29; after the user selects one, the corresponding road network file can be loaded based on the stored metadata (e.g., file path and region GeoJSON), and its coverage area can be highlighted on the map. Regardless of the method used to delineate the area, the road network generation module 23 will ultimately obtain an area range or an initial road network file through OpenStreetMap (OSM) data acquisition and parsing (S306). If it is necessary to generate a road network from the area range, the vector data of the area can be obtained through a third-party map service (e.g., OpenStreetMap), and an automated processing flow can be initiated.

[0058] When performing OSM data acquisition and parsing (S306), after obtaining the user-selected preset template, manually drawn polygon area, or existing road network index, a third-party map service is invoked to obtain the road vector data of the target area. Preferably, filtering is performed based on road level and attributes (e.g., highway= oneway= lanes= maxspeed= The simulation process removes irrelevant elements (e.g., waterways, green spaces), and splits and normalizes multi-segment polylines, retaining the road centerline geometry and key metadata (e.g., road alignment, direction, number of lanes, speed limit, bridge / tunnel markers). When processing anomalies occur, a retry strategy can be triggered for automatic repair.

[0059] In step S307, nodes and edges can be extracted and transformed. For example, tool libraries such as OSMNX can be used to identify road intersections and endpoints as nodes, and road centerline segments between adjacent nodes as edges, constructing an initial graph structure road network. Preferably, the WGS84 geographic coordinates are first projected to a plane coordinate system (e.g., UTM or local EPSG) that matches the target area, and then converted to Cartesian coordinates (unit: meters) used internally by the simulation engine, retaining information such as directionality (including one-way / two-way), number of lanes, speed limit, and whether it is a bridge / tunnel in the edge attributes. At the same time, the entry / exit sets and turning probabilities of intersections are calculated to provide a basis for subsequent phase design and conflict relationship determination.

[0060] In step S308, neighbor node merging can be performed. To simplify the road network topology and improve simulation efficiency, topology merging can be performed on low-spacing multi-node intersections. For example, candidate nodes can be clustered according to a preset distance threshold and node degree, a representative node can be selected, and nodes within the same cluster can be merged into the representative point; the edges involved in the merging are reconnected and geometrically resampled to keep the road direction and length approximately unchanged; at the same time, a mapping table of original and new nodes and edges is maintained for traceability. Furthermore, "cross-level merging prohibition" constraints can be set for bridges / ramp / elements at different elevation levels to avoid mistakenly merging nodes at different levels into the same intersection.

[0061] In step S309, connectivity repair (dead-end path handling) can be performed, and connectivity checks are executed on the generated graph structure. For example, for "dead-end paths" and hanging edges with a degree of 1 that are not located on the boundary, pruning or splicing repair is performed according to rules; for small weakly connected components caused by incorrect segmentation, the main connected components are retained and noise components are cleaned up; if necessary, gaps are automatically "bridged" based on road geometry and orientation consistency, and a repair log is recorded. After the repair is completed, shortest path reachability and strong / weak connectivity checks are performed again. If anomalies still exist, the anomaly list is submitted to the subsequent step S312 for error location and automatic repair.

[0062] In step S310, the final generation of the simulated road network can be performed to generate one or more verified digital road network files (e.g., SUMO's .net.xml file) that conform to the simulation engine 28 specifications. These files contain all the necessary node and edge information, serving as the core components of the simulation system and forming a mapping relationship with the road network structure of the actual traffic system.

[0063] In step S311, the generated file is checked for format and topology consistency. If the check fails in step S311, error location and automatic repair are performed in step S312, and the process is returned to step S310 after repair.

[0064] If the verification in step S311 passes, the road network file is stored in, for example, data storage 29 in step S313, and the road network ID and file path are output in step S314 for subsequent scenario configuration and simulation calls.

[0065] The automated traffic network generation method according to exemplary embodiments of the present disclosure enables the rapid and accurate construction of digital road networks for simulation tasks.

[0066] Figure 4 This is a flowchart illustrating an automated traffic demand generation method according to an exemplary embodiment of the present disclosure.

[0067] like Figure 4 As shown, in step S401, the scene generation process is started.

[0068] In step S402, the source of the demand is first determined. Users can choose from the following five methods:

[0069] Preset traffic flow pattern (S403): Select the preset simulation time period (e.g., morning peak 7:00-9:00) and traffic flow intensity (e.g., "low flow", "medium flow", "high flow") on the front-end interface, and the system will automatically generate vehicle travel plans based on the built-in daily traffic demand curve.

[0070] Real-time traffic flow reverse calculation mode (S404): Users upload or access real-time traffic flow data, and the system converts it into structured traffic demand parameters through a reverse calculation algorithm;

[0071] Custom drawing mode (S405): Users draw traffic flow curves through the front-end interactive interface, and the system generates corresponding traffic demand files based on the drawing results;

[0072] OD / Demand File Upload Mode (S406): Users upload OD matrix files or traffic demand files (.rou.xml) that conform to SUMO specifications. The system will perform demand conversion on the OD matrix or directly perform legality verification on the .rou.xml file.

[0073] Natural Language Input Mode (S407): The user inputs a natural language description in the GUI (e.g., "Simulating from 7:00 to 9:00, the traffic flow on the main road conforms to the normal distribution of the morning peak, with the peak occurring around 8:00, and the peak value being about 2500 vehicles / hour"). The system calls the large language model to parse the input natural language description into structured parameters.

[0074] Next, in step S408, in natural language input mode, the system calls LLM to parse the input, extract key information, and convert it into a JSON structured object. For example, the input "simulating traffic flow on the main road from 7:00 to 9:00, with a normal distribution, reaching a peak of 2500 vehicles / hour at 8:00" can be parsed into JSON containing time_range and flow_distribution fields. The time_range field represents the time interval, and the flow_distribution field represents the traffic flow changes at different times.

[0075] In step S409, the demand generator performs trimming and flow adjustment based on the analysis results, the built-in baseline traffic volume curve, or the OD matrix input by the user, generates a vehicle travel plan that meets the user's needs, and conducts calculation experiments in the simulation system. At the same time, it retains the alignment interface with the actual traffic data for virtual-real comparison and iterative correction.

[0076] In step S410, the system generates a traffic demand file (e.g., SUMO's .rou.xml) and prepares it for validity verification.

[0077] In step S411, the generated or uploaded traffic demand file is validated for legality, and the syntax, consistency of referenced objects, and logical rationality are checked.

[0078] If the verification result in step S411 fails, error location and repair are performed in step S412, and the process is returned to regenerate or manually modify. If the verification result in step S411 passes, the traffic demand file is stored in the data storage module 29 in step S413 and bound to the corresponding road network ID to ensure the consistency between the road network and traffic flow. In step S414, the final scene parameters and file paths are output for the simulation engine to call, enabling subsequent signal timing optimization and scheme evaluation.

[0079] The automated traffic demand generation method according to exemplary embodiments of this disclosure utilizes Large Language Model (LLM) and multi-source data fusion to achieve intelligent and rapid generation of traffic scenarios. The generated demands serve as input to the simulation system and can also be compared and corrected with actual system observation data, providing a foundation for virtual-real interaction.

[0080] Figure 5 This is a flowchart illustrating an automated traffic scene configuration and editing method according to an exemplary embodiment of the present disclosure.

[0081] like Figure 5 As shown, in step S501, the traffic scenario configuration process is initiated.

[0082] Next, users can select one or more of the following configuration items based on the default scenario:

[0083] Configure simulation period (S502): Specify the time interval for simulation operation, such as morning peak 7:00–9:00 and evening peak 17:00–19:00;

[0084] Select key areas / subnets (S503): Designate certain key areas or subnets within the road network as the focus of analysis to reduce interference from irrelevant areas;

[0085] Add traffic events (S504): For example, setting up road construction on a certain road section, or adding an accident event at a certain intersection;

[0086] Configure weather / visibility (S505): For example, set environmental conditions such as rainfall intensity and reduced visibility in foggy weather;

[0087] Demand fine-tuning (S506): Make corrections based on the generated traffic demand, including overall traffic flow adjustment, modification of vehicle type ratios, or scaling of the OD matrix.

[0088] Subsequently, in step S507, the system merges the user-selected configuration items with the basic road network and demand data to generate a complete traffic scenario file, and runs the calculation experiment in the simulation system. At the same time, it retains the alignment interface with the actual traffic operation status for parallel verification and iterative optimization.

[0089] In step S508, the generated traffic scenario file undergoes consistency verification, including file syntax checking, parameter validity verification, and scenario logic consistency detection. If the verification fails in step S508, error location and repair are performed in step S509, prompting the user to re-edit configuration items if necessary. If the verification passes in step S508, the scenario file is stored in the data storage module in step S510, and the corresponding road network ID and demand ID are bound, while a unique scenario ID is generated. In step S511, the scenario file path is output for subsequent use by the simulation engine to achieve signal timing optimization and scheme evaluation.

[0090] The automated traffic scenario configuration and editing method according to exemplary embodiments of this disclosure can further construct a complete traffic simulation scenario file based on the generated traffic network and traffic demand.

[0091] Figure 6 This is a flowchart illustrating an automated signal timing optimization method according to an exemplary embodiment of the present disclosure.

[0092] As shown in Figure 6, the process starts (S601) and enters the optimization mode determination (S602). Based on the determination result, it enters either the offline optimization branch (S603–S607) or the online optimization branch (S608–S614). Finally, it completes the database entry and binding (S617) and outputs the scheme ID / file path (S618).

[0093] The offline optimization branch (S603–S607) includes: Algorithm selection (S603), where the user selects one of the following algorithms: fixed timing, Webster, green wave, Synchro, LLM, etc.; Parameter acquisition and constraint checking (S604), which automatically summarizes road network / traffic flow parameters and phase constraints (e.g., conflicts, pedestrian phases, minimum green light, etc.), and prompts for completion if any items are missing; Scheme calculation (S605), which calls the corresponding optimization engine to solve candidate schemes; Conflict / coordination check (S606), which checks phase conflicts, arterial coordination, and cycle consistency; and Timing file generation (S607), which outputs a signal timing file (e.g., tls.xml) that can be used by the simulation engine.

[0094] In the embodiments of this disclosure, the offline optimization algorithms include, but are not limited to, the Webster optimization algorithm, the green wave optimization algorithm, the Synchro software computation optimization algorithm, and the LLM interactive optimization algorithm.

[0095] The Webster optimization algorithm is a signal timing optimization algorithm for isolated intersections. First, it automatically extracts the traffic flow for each critical phase i of the target intersection from road network data and traffic flow data. (Unit: pcu / h) and saturation flow rate (Unit: pcu / h). Then, calculate the flow ratio for each key phase. And sum to obtain the total flow ratio of the key phase. Combined with the total lost time at the intersection (Including start-up losses for each phase and yellow light duration), the system uses Webster's formula to calculate the optimal signal period. Finally, the effective green light time will be... The green light duration for each phase is determined by allocating traffic according to the flow ratio of each key phase. The algorithm aims to minimize the total delay at the intersection.

[0096] The green wave optimization algorithm is primarily applied to urban arterial roads or multiple consecutive intersections specified by the user. When the user selects this mode and specifies a green wave... When determining the trunk route at an intersection, the system first determines a common signal cycle for that trunk route. Then, using the first intersection as a reference, calculate the downstream... The intersection relative to the upstream first Ideal phase difference at each intersection Its calculation formula is ,in It is the distance between two adjacent intersections. This is the user-defined desired green wave speed. By sequentially calculating and setting the phase difference of all downstream intersections, a coordinated timing scheme is generated to reduce the number of times vehicles stop on the main road.

[0097] Synchro software's optimization algorithm not only considers traffic flow but also integrates detailed intersection geometry data (e.g., lane width, turning radius), complex phase settings (e.g., pedestrian-only phases, overlapping phases), and conflict relationships among different traffic flows (e.g., motor vehicles, non-motorized vehicles, pedestrians). These parameters are input into the optimization engine for Level of Service (LOS) analysis and Performance Index (PI) optimization based on the Highway Capacity Manual (HCM) model. This enables the generation of a highly refined and globally coordinated timing scheme for the entire road network or a specified coordination area.

[0098] The LLM interactive optimization algorithm automatically parses the optimization objectives and constraints expressed by the user in natural language (e.g., arterial priority, cycle limits, minimum / maximum green light, no pedestrian demotion, target speed, phase taboos, etc.) into a structured configuration (objective function, weights, decision variables, and hard / soft constraints) through prompting engineering and parameter normalization mechanisms. This is then integrated with road network geometry and traffic flow statistics to form a solvable problem. A hybrid solution process of "LLM generation + classical optimization" is then employed. First, a large language model generates a draft of candidate timings (e.g., cycle, phase sequence, effective green light for each phase, and arterial phase difference, etc.). Next, the system performs mode / domain and physical constraint reviews (e.g., green light agreements, yellow + all-red clearing time, no intersection in the conflict matrix, unified cycle in coordination zones, pedestrian clearing time, etc.), automatically filling in missing or inconsistent terms and correcting constraint projections. Then, a heuristic / integer programming / rule-based computation kernel completes global coordination and refinement; if infeasible, feasible alternative suggestions are provided and iteratively corrected. The final output is a timing scheme file that meets the constraints and can be directly used for simulation, while retaining the generation log and key hyperparameters to support reproduction and manual review.

[0099] The online optimization branch (S608–S614) interacts with the engine in real time during simulation and dynamically iterates to adjust the timing. It mainly includes the following steps: State extraction (S608): obtaining state variables such as queue length, waiting time, current phase, and traffic flow from the simulation engine; Policy inference (S609): outputting actions based on the current policy (e.g., Actor or Q network); Action delivery (S610): applying actions such as phase switching / holding to the simulation environment via TraCI; Reward calculation (S611): constructing immediate rewards based on metrics such as delay, queueing, and number of stops; Policy update (S612): updating policy parameters online / offline according to the selected algorithm; Convergence / duration determination (S613): stopping after reaching the training rounds or performance threshold; Policy export (S614): exporting model weights / policy files for subsequent reproduction or deployment.

[0100] The basic principle is to model the intersection's signal controller as an agent; the agent makes decisions at each time step. Obtain the current state of the environment from simulation engine 28. Then according to its strategy Choose an action To execute; after executing the action, the simulation environment transitions to a new state. and give back a reward. The goal of an intelligent agent is to learn an optimal policy through repeated interactions with its environment. To maximize long-term cumulative rewards ,in It is a discount factor.

[0101] In the example, State Defined as a A matrix of dimension, where It refers to the number of import lanes. It refers to the characteristic quantity of each lane, which includes the number of vehicles queuing in the current lane, the waiting time of the vehicle at the front of the queue, and whether the current lane is in a green light phase, etc.

[0102] In the example, regarding actions... The action space A is defined as a discrete set representing the operations that the controller can perform. For example, Maintain the current phase, then immediately switch to the next preset phase. In another example, the action space can be designed as the set of all conflict-free phase combinations. The agent directly selects the next phase to execute.

[0103] In the example, regarding reward... reward function It is designed to provide positive feedback for desired traffic outcomes and negative feedback for undesirable outcomes. For example, it can be defined as the negative value of the change in the total cumulative delay D at the intersection between two decisions, i.e. In this way, actions that reduce delays will receive positive rewards.

[0104] The framework disclosed herein supports the deployment of various reinforcement learning algorithms, such as Deep Q-Network (DQN) and Proximal Policy Optimization (PPO).

[0105] The deep Q-network algorithm is suitable for discrete action spaces. The agent maintains a parameter of... Deep neural networks Used to approximate the optimal action value function To address the instability of traditional Q-learning in high-dimensional state spaces, DQN can introduce Experience Replay and Target Network mechanisms.

[0106] Regarding experience replay, when an agent interacts with its environment, it generates experience tuples. Stored in a fixed-size experience replay pool In training, the system does not use continuous samples, but rather... We learn by randomly sampling a small batch of data. This breaks down the correlation between samples, making the training process more stable and data utilization more efficient.

[0107] Regarding the target network, DQN uses two neural networks with the same structure but different parameters. One is the main network. One is used to select and update actions at each time step. The other is the target network. Its parameters Regularly from the main network It is copied and remains fixed for a period of time. The target network is specifically used to calculate the target Q-value in the loss function. : ,in, The target Q value; The immediate reward received after performing an action; This is a discount factor used to balance the importance of current and future rewards; The next state; The prediction of the target network in the next state Take all possible actions at the time The maximum Q value that can be obtained.

[0108] Proximal policy optimization (PFO) is a policy gradient method renowned for its excellent stability and data efficiency, capable of handling both discrete and continuous action spaces. In this embodiment, the agent employs an Actor-Critic architecture, maintaining a policy network (Actor) for decision-making. A value network (Critic) for evaluating the value of a state. The core idea of ​​PPO is to limit the range of change between the old and new policies by using an objective function when updating the policy, thereby avoiding policy collapse caused by excessively large single-step updates.

[0109] PPO calculates the probability ratio between the old and new strategies. To achieve this constraint: ,in, This is the strategy that needs to be optimized. It's an old strategy for collecting data.

[0110] There are two main variants of PPO to implement policy update constraints:

[0111] 1. PPO with Penalty: This variant adds a penalty term for KL divergence to the objective function to measure the difference between the old and new policies. Objective function As shown below:

[0112] .

[0113] in, It is an estimate of the advantage function. It is a dynamically adjusted penalty coefficient. When the KL divergence between the old and new strategies is too large, It will be increased to strengthen the punishment.

[0114] 2. PPO with Clipping Objective Function (PPO-Clip): This is a more commonly used variant that directly limits the probability ratio through a clipping operation. The range of its objective function. As shown below:

[0115] .

[0116] in, It is a hyperparameter (e.g., 0.2) used to define the range of clipping; The function will use probability ratios Limited to Within the range. The function ensures that the final target is the smaller of the original target and the pruned target. This design effectively prevents the dominance function from being applied. For the correct time Too large, or when When negative This minimizes the impact of small values, thus enabling stable and reliable strategy updates.

[0117] In step S615, a consistency check is performed on the offline generated signal timing file. If it fails, the process proceeds to error location and repair (S616) and returns to step S607 to regenerate the file.

[0118] In step S614, the strategy / model weights obtained through online mode are exported. Subsequently, in step S617, the data is stored in the database and bound, and a scheme ID and reproducible experimental configuration are generated.

[0119] If the verification in step S615 passes, the scheme is bound to the road network / intersection ID and stored in step S617. Then, in step S618, the scheme ID and file / weight path are output for subsequent scheme evaluation.

[0120] The automated signal timing optimization method according to exemplary embodiments of this disclosure can provide users with flexible signal timing optimization capabilities within a unified framework.

[0121] Figure 7 This is a flowchart illustrating an automated traffic control scheme evaluation method according to an exemplary embodiment of the present disclosure.

[0122] After the simulation runs, the simulation engine can output raw result data including vehicle trajectories, detector flow / speed / occupancy, phase switching logs, and road segment speed grids. It can then analyze, aggregate, and perform quality control on these data to generate multi-level key performance indicators (KPIs). Under a unified standard, it can also perform objective comparisons and report outputs between different schemes. The evaluation results can be used to guide signal timing in actual traffic systems and can support iterative optimization of actual system timing schemes in the simulation system, thus forming a parallel execution closed loop between the simulation system and the actual system.

[0123] like Figure 7 As shown, in step S701, the automated signal control scheme evaluation process is initiated.

[0124] In step S702, scheme selection and consistency assurance are performed. For example, the user selects no fewer than two simulation schemes based on the same road network ID for comparison through the front-end interaction layer 21. The consistency between the road network ID, simulation time period, sampling step size, and vehicle type is judged. If a conflict exists, the user is prompted to change the scheme or enter the data alignment process.

[0125] In step S703, KPI extraction is performed. In this example, KPIs can be categorized by level as follows: road network level indicators, used to evaluate the macroscopic traffic performance of the entire simulation area, such as the average driving speed, average delay time, total driving distance, and total number of stops for the entire road network; road segment level indicators, used to evaluate the operation of a specific road, such as traffic flow, average vehicle speed, traffic density, and average travel time for a specific road segment; and intersection level indicators, used to evaluate the traffic efficiency of a single intersection, such as the total number of vehicles passing through a specific intersection, the average queue length of each approach lane, the maximum queue length, and the average waiting time.

[0126] In step S704, caliber / time period alignment is performed. For example, the calculated KPIs of each scheme are read in parallel or missing items are calculated in real time from the raw data. Then, caliber unification (unit, statistical definition, aggregation method) and time alignment (start and end time / peak and flat segmentation / holiday labels) are performed to ensure "same caliber, same time period, same granularity". If inconsistency in granularity is found, the coarsest granularity is downsampled to avoid spurious differences.

[0127] In step S705, a data integrity / consistency check is performed. Integrity and boundary checks are conducted on key fields (e.g., non-negative traffic flow, speed limit, queue size not less than zero, valid phase sequence), and abnormal jumps are detected. For short-term missing data, linear interpolation / last-observed calculation is used; for long-term missing data, recalculation is performed at the road segment / road network layer. Inconsistent data items are recalculated according to a unified definition. If the check in step S705 fails, anomaly localization and repair are performed in step S706. All repair actions are written to the verification log and are traceable; simultaneously, a comparison interface with actual traffic observation data is maintained to facilitate consistency verification between the simulation system and the actual system.

[0128] If the data integrity / consistency check passes in step S705, a comparison calculation is performed in step S707. In the example, direct numerical results are given for each comparison indicator according to the scheme. Summation comparison is performed for total quantity indicators (e.g., total mileage, total number of stops); average comparison is performed for intensity / average indicators (e.g., average speed, average delay); and for hierarchical sets (e.g., multiple road segments / intersections), the overall comparison results and a "Top-N list of road segments / intersections with the greatest differences" are output simultaneously to facilitate the identification of contribution sources. This step does not perform statistical significance testing or complex evaluation, but only outputs an intuitive and verifiable table of differences and improvement rates.

[0129] In step S708, visualization is generated, for example, automatically generating charts for decision-making presentation, including but not limited to: (1) radar charts, used to compare the overall shape of multiple indicators at the road network level; (2) bar / column charts, used to compare the values ​​of single indicators under different schemes; (3) time series line charts, used to show the evolution differences of indicators over time; (4) optional green wave band time interval charts (main road sections), which intuitively present the effects of queue propagation and traffic coordination. The charts are output in vector format and include a data download portal for easy review and reuse.

[0130] In step S709, a structured summary and prompting process is performed. In the example, the aligned comparison results are structured into a machine-readable object (e.g., JSON: containing indicator name, unit, time window, value, change / significance marker, anomaly notes, etc.), and combined with a preset analysis template to generate a prompt. The template guides the generation of explanations for dimensions such as "congestion status, traffic efficiency, coordination effect, pedestrian impact, robustness (anomaly / missing test sensitivity)," avoiding subjective assumptions and requiring explicit citation of evidence bits (field references) in the structured data.

[0131] In step S710, a comparative analysis report is generated and verified. In the example, the traffic scenario analysis large language model generates a natural language report based on the structured data and templates mentioned above. This natural language report includes statements of strengths / weaknesses at each level, quantitative evidence, improvement suggestions, and conclusions. To ensure traceability, the system performs automatic verification after generation: checking whether the values ​​in the report are consistent with the input, whether the units match, and whether the conclusions are consistent with the significance results; when inconsistencies are found, a write-back repair and regeneration are triggered; simultaneously, the analysis results can be fed back to the simulation system model for adjustment, making it closer to the actual traffic system, thereby forming an iterative optimization of virtual and real interaction.

[0132] In steps S711 and S712, the report and data are exported and archived. In the example, the report is exported along with detailed data for plotting / statistics (e.g., report.md and kpi_compare.csv), and archiving and scheme ID binding are completed in data storage 29. The archived content includes a list of comparison schemes, road network IDs, statistical configurations (time period, granularity, weight), algorithm version and random seed (if any), anomaly repair logs, and significance evaluation results, to facilitate experiment reproduction and auditing, and to serve as a historical reference baseline for comparing and correcting the simulation system and the actual system in parallel traffic systems.

[0133] In step S713, the results are output. For example, the evaluation result index and file path are returned to the front-end interaction layer 21, and a "one-click reproduction" entry is provided (based on the archive list, the chart and report generation configuration is reloaded). In the example, when only a single solution is selected, the system degenerates into a single solution evaluation mode: skipping the difference calculation and significance test, and only outputting the hierarchical indicator overview, bottleneck location, and improvement suggestions.

[0134] The automated signal control scheme evaluation method according to the exemplary embodiments of this disclosure can automatically extract key indicators from simulation results and form a multi-scheme comparison conclusion.

[0135] Figure 8 This is a block diagram illustrating an automated parallel traffic simulation analysis apparatus according to an exemplary embodiment of the present disclosure.

[0136] like Figure 8 As shown, an automated parallel traffic simulation analysis apparatus 800 according to an exemplary embodiment of this disclosure includes: a road network generation unit 801, configured to generate an initial digital road network from a map data source in response to receiving a region delineation instruction, and to perform topology calibration and connectivity repair operations on the initial digital road network to generate a digital road network file conforming to the simulation engine specifications; a traffic demand generation unit 802, configured to generate structured traffic demand data based on user input, wherein the user input includes at least one of preset traffic flow patterns, measured traffic flow data, custom drawings, origin-destination matrix files, and natural language descriptions, and the traffic demand data includes traffic flow intensity, time period distribution, and vehicle travel patterns; and a traffic scene configuration unit 803. 03 is configured to generate a basic traffic scenario file based on digital road network files and traffic demand data, and form a traffic simulation operation scenario based on the user's configuration of the road network and traffic demand parameters included in the basic traffic scenario file; the signal timing optimization unit 804 is configured to generate multiple signal timing schemes based on the traffic simulation operation scenario using various optimization algorithms; the scheme evaluation unit 805 is configured to perform traffic simulation based on multiple signal timing schemes, analyze the simulation results of multiple signal timing schemes to obtain multi-level key performance indicators for each of the multiple signal timing schemes, perform a visual comparison of the multi-level key performance indicators of multiple signal timing schemes, and generate a natural language evaluation report using a large language model.

[0137] In the example, a region delineation instruction is generated when a user searches for a geographic location name and selects a preset city road network template through the map interface, uses the polygon tool to customize and draw a closed geographic area through the map interface, or selects and loads previously stored digital road network data from a saved road network list.

[0138] In the example, topology calibration and connectivity repair operations include: identifying and handling dead ends, dangling edges, and incorrectly segmented weakly connected components; performing node merging, geometric resampling, and gap bridging; and performing format and topology consistency checks.

[0139] In the example, the traffic demand generation unit 802 is further configured to: generate traffic demand data based on the daily traffic volume distribution curve and the simulation period and traffic flow intensity specified by the user, in response to the user's input of a preset traffic flow pattern; generate traffic demand data based on the measured traffic flow data through a data back-calculation algorithm, in response to the user's input of custom drawing; provide an interactive drawing interface to facilitate the user's drawing of traffic flow curves and generate corresponding traffic demand data based on the user-drawn traffic flow curves; verify the file format of the start-end point matrix file and convert it into traffic demand data, in response to the user's input of a natural language description; and perform semantic parsing of the input natural language description using a large language model to generate structured traffic demand data, in response to the user's input of a natural language description.

[0140] In the example, various optimization algorithms include at least two of Webster's method, green band coordination, Synchro optimization, large language model-based optimization algorithms, and reinforcement learning algorithms.

[0141] In the example, when the input includes a natural language description and multiple optimization algorithms, including those based on a large language model, the signal timing optimization unit 804 is further configured to: parse the optimization intent from the user's natural language description, the optimization intent including at least one of optimization objectives based on road corridors, optimization objectives based on specific intersections, and optimization objectives based on global performance; pass the digital road network file and traffic demand data as input data to the large language model; and perform global optimization calculations based on the parsed optimization intent and the input data through the large language model, outputting a signal timing scheme that defines the cycle, phase, and green ratio of each intersection.

[0142] In the example, the scheme evaluation unit 805 is further configured to: extract and align key performance indicators at the road network and intersection levels of the selected signal timing schemes in response to the user's selection of two or more signal timing schemes; generate and display visualization charts for comparing key performance indicators based on the extracted key performance indicators; and use a large language model to comprehensively evaluate the key performance indicators to output a natural language evaluation report, which includes quantitative analysis, comparison of advantages and disadvantages, and improvement suggestions.

[0143] The above combination Figures 1 to 7 The specific operations shown are respectively by Figure 8The corresponding units in the automated parallel traffic simulation analysis device 800 shown are used to execute the operation; specific operational details will not be elaborated here. By employing the automated parallel traffic simulation analysis device according to the exemplary embodiments of this disclosure, the complexity of traffic simulation modeling and the need for manual intervention are reduced, the efficiency and accuracy of signal timing optimization are improved, objective and comprehensive evaluation of scheme performance is achieved through quantitative indicators, and end-to-end automated and parallel execution from simulation environment construction to scheme optimization and decision support report generation can be realized within the virtual-real interaction framework of the simulation system and the actual system. It can be widely applied to research and engineering practice in intelligent traffic management and intelligent signal control.

[0144] Figure 9 This is a block diagram illustrating a computing system including at least one computing device and at least one storage device of storage instructions according to an exemplary embodiment of the present disclosure.

[0145] like Figure 9 As shown, the computing system 900 according to an exemplary embodiment of the present invention includes a computing device 901 and a storage device 902. The storage device 902 stores computer-executable instructions. When the computer-executable instructions are executed by the computing device 901, the automated parallel traffic simulation analysis method described in any of the foregoing embodiments is executed.

[0146] The computing device 901 can be deployed in a server or client, or on a node device in a distributed network environment. Furthermore, the computing device 901 can be a PC, tablet, personal digital assistant, smartphone, web application, or other device capable of executing the aforementioned set of instructions. Here, the computing device is not necessarily a single computing device; it can be any collection of devices or circuits capable of executing the aforementioned instructions (or instruction sets) individually or in combination. The computing device can also be part of an integrated control system or system manager, or can be configured to interconnect with a portable electronic device locally or remotely (e.g., via wireless transmission). In the computing device, the processor includes a central processing unit (CPU), a graphics processing unit (GPU), a programmable logic device, a dedicated processor system, a microcontroller, or a microprocessor. By way of example and not limitation, the processor also includes analog processors, digital processors, microprocessors, multi-core processors, processor arrays, network processors, etc. For example, the processor executes instructions to run various modules of the backend service layer 22, exchanging data with the frontend interaction layer 21, the third-party service interface 30, and the actual traffic management system through a communication interface, thereby establishing a parallel execution mechanism for virtual-real interaction between the simulation system and the actual system.

[0147] According to another aspect of this disclosure, a computer-readable storage medium is provided that stores instructions, which, when executed by at least one computing device, cause the at least one computing device to perform the automated parallel traffic simulation analysis method described in any of the foregoing embodiments. The computer-readable storage medium includes magnetic media such as floppy disks and magnetic tapes, optical media (including optical disc (CD) ROMs and DVD ROMs), magneto-optical media such as floppy discs, hardware devices such as ROMs and RAMs designed for storing and executing program commands, and flash memory. The instructions may include language code executable by a computer using an interpreter and machine language code generated by a compiler.

[0148] By adopting this disclosure, simulation efficiency and ease of use can be significantly improved. Through full-process automation and innovative natural language command interaction, users are freed from tedious parameter configuration, greatly lowering the barrier to entry and enabling non-professionals to quickly build and run complex traffic simulation tasks. It enables highly flexible scenario construction by utilizing the semantic understanding capabilities of a large language model to automatically convert from macroscopic, qualitative natural language descriptions to microscopic, quantitative simulation parameters, making simulation scenario construction more intuitive, rapid, and flexible. It deepens the decision-making value of simulation results by introducing intelligent analysis and multi-scheme comparison functions from the large language model, elevating simulation results from "data presentation" to "comprehensive evaluation," revealing the advantages and disadvantages of different schemes and directions for improvement, providing traffic managers with intuitive, data-driven decision support. It breaks through the traditional mirror model of static simulation by introducing the concept of parallel systems, establishing a parallel execution mechanism of virtual-real interaction between the simulation system and the actual system, enabling optimization and evaluation to form a closed-loop iteration, driving the continuous evolution of the simulation system and signal control schemes.

[0149] The processes, methods, or algorithms disclosed herein can be transmitted to, or implemented by, a processing device, controller, or computer, which may include any existing programmable electronic control unit or a dedicated electronic control unit. Similarly, the processes, methods, or algorithms can be stored in various forms as data and instructions executable by a controller or computer, including but not limited to information permanently stored on non-writable storage media (such as ROM devices) and information variablely stored on writable storage media (such as floppy disks, magnetic tapes, CDs, RAM devices, and other magnetic and optical media). The processes, methods, or algorithms can also be implemented in a software executable object. Optionally, the processes, methods, or algorithms can be implemented wholly or partially using suitable hardware components (such as ASICs, FPGAs, state machines, controllers, or other hardware components or devices) or a combination of hardware components, software components, and firmware components.

[0150] Although this disclosure includes specific examples, it will be apparent to those skilled in the art that various changes in form and detail may be made to these examples without departing from the spirit and scope of the claims and their equivalents. The examples described herein are to be considered merely for descriptive purposes and not for limiting purposes. The description of features or aspects in each example is to be considered applicable to similar features or aspects in other examples. Suitable results may be obtained if the described techniques are performed in a different order, and / or if components in the described system, architecture, apparatus, or circuit are combined in a different manner and / or if components in the described system, architecture, apparatus, or circuit are replaced or supplemented with other components or their equivalents. Therefore, the scope of this disclosure is not limited by the specific embodiments but by the claims and their equivalents, and all variations within the scope of the claims and their equivalents shall be construed as included in this disclosure.

Claims

1. An automated parallel traffic simulation analysis method, characterized by, The automatic parallel traffic simulation analysis method comprises: In response to receiving the region delineation instruction, generating an initial digitized road network from a map data source, and performing a topological calibration and connectivity repair operation on the initial digitized road network to generate a digitized road network file conforming to the simulation engine specification; Based on the input of the user, generate structured traffic demand data, the input including at least one of a preset traffic flow pattern, measured traffic flow data, custom drawing, origin-destination matrix file and natural language description, the traffic demand data including traffic flow intensity, time period distribution and vehicle travel mode; Based on the digitized road network file and the traffic demand data, generate a basic traffic scenario file, and based on the user's configuration of the road network and traffic demand parameters included in the basic traffic scenario file, form a traffic simulation running scenario; Based on the traffic simulation running scenario, generate multiple signal timing schemes using multiple optimization algorithms; Based on the multiple signal timing schemes, perform traffic simulation, analyze the simulation results of the multiple signal timing schemes to obtain multi-level key performance indicators for each of the multiple signal timing schemes, visualize and compare the multi-level key performance indicators of the multiple signal timing schemes, and generate a natural language evaluation report using a large language model, Wherein, based on the input of the user, the step of generating structured traffic demand data includes: in response to the user's input being a preset traffic flow pattern, generating traffic demand data based on the daily traffic volume distribution curve and the user-specified simulation period and traffic flow intensity; in response to the user's input being measured traffic flow data, generating traffic demand data based on measured traffic flow data through a data back calculation algorithm; in response to the user's input being custom drawing, providing an interactive drawing interface to facilitate the user to draw a traffic flow curve, and generating corresponding traffic demand data based on the user-drawn traffic flow curve; in response to the user's input being an origin-destination matrix file, verifying the file format of the origin-destination matrix file and converting it to traffic demand data; in response to the user's input being a natural language description, using a large language model to perform semantic analysis on the input natural language description and generate structured traffic demand data.

2. The automated parallel traffic simulation analysis method of claim 1, wherein, When the user searches for a geographic location name through the map interaction interface and selects a preset city road network template, uses the polygon tool through the map interaction interface to custom draw a closed geographic area, or selects and loads previously stored digitized road network data from a saved road network list, the region delineation instruction is generated.

3. The automated parallel traffic simulation analysis method of claim 1, wherein, The topological calibration and connectivity repair operation includes: identifying and processing broken roads, hanging edges, and incorrectly segmented weakly connected components; performing node merging, geometric resampling, and short gap bridging; performing format and topological consistency checking.

4. The automated parallel traffic simulation analysis method of claim 1, wherein, The multiple optimization algorithms include at least two of the Webster method, green wave coordination, the Synchro optimization method, a large language model-based optimization algorithm, and a reinforcement learning algorithm.

5. The automated parallel traffic simulation analysis method of claim 4, wherein, In the case that the input comprises a natural language description and the plurality of optimization algorithms comprises a large language model-based optimization algorithm, the step of generating a plurality of signal timing plans using a plurality of optimization algorithms comprises: parsing an optimization intention from the user's natural language description, the optimization intention comprising at least one of an optimization target based on a road corridor, an optimization target based on a specific intersection, and an optimization target based on global performance; passing the digitized road network file and the traffic demand data as input data to a large language model; performing global optimization calculation based on the parsed optimization intention and the input data by the large language model, and outputting a signal timing plan defining the cycles, phases and green ratios of each intersection.

6. The automated parallel traffic simulation analysis method of claim 1, wherein, The step of visualizing and comparing the multi-level key performance indicators of the plurality of signal timing plans, and generating a natural language evaluation report using a large language model comprises: In response to the user selecting two or more signal timing plans, extracting and aligning the key performance indicators of the selected signal timing plans at the road network level and the intersection level; Based on the extracted key performance indicators, generating and displaying a visualization chart for comparing the key performance indicators; Using a large language model, making a comprehensive comment on the key performance indicators to output the natural language evaluation report, which includes quantitative analysis, pros and cons comparison, and improvement suggestions.

7. An automated parallel traffic simulation analysis device, characterized by, The automatic parallel traffic simulation analysis device comprises: A road network generation unit configured to generate an initial digitized road network from a map data source in response to receiving a region delineation instruction, and perform topology calibration and connectivity repair operations on the initial digitized road network to generate a digitized road network file conforming to the simulation engine specification; A traffic demand generation unit configured to generate structured traffic demand data based on user input, the input comprising at least one of a preset traffic flow pattern, measured traffic flow data, custom drawing, origin-destination matrix file, and natural language description, the traffic demand data comprising traffic flow intensity, time period distribution, and vehicle travel mode; A traffic scenario configuration unit configured to generate a basic traffic scenario file based on the digitized road network file and the traffic demand data, and form a traffic simulation running scenario based on user configuration of road network and traffic demand parameters included in the basic traffic scenario file; A signal timing optimization unit configured to generate a plurality of signal timing plans using a plurality of optimization algorithms based on the traffic simulation running scenario; A plan evaluation unit configured to perform traffic simulation based on the plurality of signal timing plans, analyze the simulation results of the plurality of signal timing plans to obtain multi-level key performance indicators for each of the plurality of signal timing plans, visualize and compare the multi-level key performance indicators of the plurality of signal timing plans, and generate a natural language evaluation report using a large language model, The traffic demand generation unit is further configured to: in response to the user inputting a preset traffic flow pattern, generate traffic demand data based on a daily traffic volume distribution curve and a simulation time period and traffic flow intensity specified by the user; in response to the user inputting measured traffic flow data, generate traffic demand data based on the measured traffic flow data through a data back calculation algorithm; in response to the user inputting self-defined drawing, provide an interactive drawing interface to facilitate the user to draw a traffic flow curve, and generate corresponding traffic demand data based on the traffic flow curve drawn by the user; in response to the user inputting a start-end matrix file, verify a file format of the start-end matrix file and convert the start-end matrix file into traffic demand data; and in response to the user inputting a natural language description, perform semantic analysis on the input natural language description by using a large language model, and generate structured traffic demand data.

8. A computing system comprising at least one computing device and at least one storage device storing instructions, wherein the computing system is configured to perform operations comprising: The instructions, when executed by the at least one computing device, cause the at least one computing device to perform the automated parallel traffic simulation analysis method according to any one of claims 1-6.

9. A computer-readable storage medium storing instructions, wherein, The instructions, when executed by the at least one computing device, cause the at least one computing device to perform the automated parallel traffic simulation analysis method according to any one of claims 1-6.

Images