Urban traffic congestion treatment method and framework based on general big language model

Abstract

The invention relates to an urban traffic congestion treatment method and framework based on a general large language model, and the method comprises the steps: receiving and analyzing the questioning content of a user, and dynamically selecting a bottom-layer multi-expert Agent most related to the questioning content for the first round of interaction; according to the questioning content of the user and the feedback information of the underlying expert Agent of the first round of interaction, judging whether the first round of interaction result meets the user demand or not; if the first-round interaction result does not meet the user requirement, other bottom layer expert Agents continue to be selected for interaction, successive approximation is conducted till a solution meeting the user requirement is found, and a final result is fed back to the user after the treatment process is finished. Therefore, technical means such as Multi-Agent cooperation and a large language model are introduced, comprehensive, efficient and intelligent treatment of the urban traffic congestion problem is realized by implementing a dynamic interaction and feedback process, and a new solution and thought are brought to urban traffic congestion treatment.

Description

Technical Field

[0001] The present invention relates to the field of intelligent transportation technology, and in particular to a method and framework for managing urban traffic congestion based on a universal large language model. Background Art

[0002] With the acceleration of urbanization, urban traffic congestion has become increasingly prominent and has become an important factor restricting urban development. At present, the causes of intersection congestion are becoming more and more complicated, including but not limited to road planning, traffic light control, vehicle flow, driver behavior, weather conditions and other multiple factors. The interaction between these factors makes the causes of congestion extremely complex, making it difficult for a single expert or team to conduct a comprehensive and accurate analysis.

[0003] In traditional traffic congestion management methods, it is usually necessary to manually collect, integrate and analyze data from various dimensions to determine the causes of congestion and formulate corresponding management measures accordingly. However, this method has many shortcomings. First, due to the numerous and complex influencing factors, manual analysis is difficult to cover all details, which may lead to inaccurate or incomplete cause analysis. Secondly, when congestion is found, it is necessary to quickly pull up multiple system platforms for collaborative positioning, which is not only time-consuming and labor-intensive, but may also affect the management effect due to data differences and information delays between platforms.

[0004] Therefore, traditional traffic congestion management methods seem to be powerless in the face of increasingly complex causes of congestion. There is an urgent need for a method that can comprehensively and efficiently detect congestion problems, accurately analyze the causes and quickly implement management measures. Summary of the invention

[0005] 1. Technical issues to be resolved

[0006] In view of the above-mentioned shortcomings and deficiencies of the prior art, the present invention provides a method and framework for urban traffic congestion management based on a universal large language model, which solves the technical problems that traditional traffic congestion management methods are unable to comprehensively and efficiently deal with the complex and changeable causes of congestion, resulting in inaccurate and incomplete cause analysis, and the collaborative positioning process is time-consuming and labor-intensive, and is easily affected by data differences and information delays.

[0007] (II) Technical solution

[0008] In order to achieve the above object, the main technical solutions adopted by the present invention include:

[0009] In a first aspect, an embodiment of the present invention provides a method for managing urban traffic congestion based on a general large language model, comprising:

[0010] Receive and analyze the user's questions, dynamically select and activate the underlying expert agent most relevant to the question content for the first round of interaction;

[0011] Based on the user's question content and the feedback information from the underlying expert agent in the first round of interaction, determine whether the results of the first round of interaction meet the user's needs;

[0012] If the results of the first round of interaction meet the user's needs, the information aggregation and feedback phase will begin;

[0013] If the results of the first round of interaction do not meet the user's needs, then based on the user's needs and the results of the first round of interaction, continue to select other underlying expert agents for at least one round of interaction, gradually approaching until a solution that meets the user's needs is found;

[0014] After the governance process is completed, the information summary feedback agent in the underlying expert agent obtains and integrates the feedback information of all underlying expert agents participating in the interaction and the user's questions, structures them according to the predefined data format, and feeds back the final results to the user in at least one form.

[0015] Optionally, before receiving and parsing the user's question content and dynamically selecting and activating the underlying expert Agent most relevant to the question content for the first round of interaction, the method further includes:

[0016] Build and initialize the traffic congestion management framework, load the general large language model and traffic professional knowledge library;

[0017] Create at least one underlying expert agent in the traffic congestion management framework, and configure each underlying expert agent to perform the set functional tasks;

[0018] Establish a communication mechanism between the collaborative coordinating agent and the underlying expert agent, and set up interfaces between the agents to achieve dynamic selection, activation and interaction of at least one operation;

[0019] Among them, at least one underlying expert agent includes: a congestion detection expert role Agent, which is used to identify the real-time signal scheme of the corresponding intersection based on the checkpoint passing data, output the intersection timing signal scheme information, and determine whether the target section is congested according to the intersection timing signal scheme information, and output the congestion detection result of the target section with time information, congestion degree quantitative label and congestion frequency quantitative label; a congestion cause analysis expert role Agent, which is used to analyze the congestion causes including supply and demand matching, section interference, traffic organization, event detection, road network and scheme, and time and space coordination based on the detection results of the target section; a congestion suggestion expert Agent, which is used to provide targeted congestion suggestions according to the congestion causes; a congestion management expert Agent, which is used to provide management solutions including variable lane switching, reverse variable lane switching and intersection signal scheme optimization according to the congestion situation.

[0020] Optionally, receiving and parsing the user's question content, and dynamically selecting and activating the underlying expert Agent most relevant to the question content for the first round of interaction includes:

[0021] Receiving question content input by the user through the user interface;

[0022] Use the pre-built large language model to parse the user's questions, extract keywords and semantic information, and determine the user's needs;

[0023] According to the analyzed user needs, the underlying multi-expert agent most relevant to the question content is dynamically selected through predefined rules;

[0024] Send activation instructions to the selected underlying expert agent to activate the selected underlying multi-expert agent for the first round of interaction;

[0025] The predefined rules include:

[0026] If the question involves congestion identification, the congestion detection expert Agent is activated;

[0027] If the question is about the cause of congestion, the congestion cause analysis expert Agent will be activated;

[0028] If congestion management suggestions are needed, the congestion suggestion expert agent is activated;

[0029] If it involves the implementation of the congestion management plan, the congestion management expert Agent will be activated.

[0030] Optionally, if the result of the first round of interaction does not meet the user's needs, then based on the user's needs and the result of the first round of interaction, continue to select other underlying expert agents for at least one round of interaction, and gradually approach until a solution that meets the user's needs is found, including:

[0031] If the results of the first round of interaction do not meet the user's needs, identify the information points that need to be supplemented based on the results of the first round of interaction and the user's needs;

[0032] Select a bottom-level expert agent that is most relevant to the information point to be supplemented for the next round of interaction;

[0033] Deliver user requirements and the results of the previous round of interactions to the underlying expert agent selected in the current round, and receive and process the feedback information from the underlying expert agent;

[0034] According to the feedback information of the bottom-level expert agent selected in the current round, it is judged again whether the current interaction result meets the user's needs;

[0035] If the current interaction result meets the user's needs, the result summary phase will begin;

[0036] If the current interaction result does not meet the user's needs, other relevant expert agents are selected to continue the next round of interaction. In each round of interaction, the selection strategy for the underlying expert agent is dynamically adjusted according to the feedback information of the previous round, so as to gradually approach until a solution that meets the user's needs is found.

[0037] In a second aspect, an embodiment of the present invention provides a framework for managing urban traffic congestion based on a general large language model, including:

[0038] A collaborative coordination agent, used to execute the method described above;

[0039] At least one underlying multi-expert Agent is used to select corresponding algorithms according to data sources of different modes to implement urban traffic congestion control measures including congestion detection, congestion cause analysis, congestion suggestions and control plan formulation.

[0040] Optionally, at least one underlying expert Agent includes:

[0041] The congestion detection expert role Agent is used to identify and analyze the real-time signal scheme of the corresponding intersection based on the acquired checkpoint passing data, output the intersection timing signal scheme information, and determine whether the target section is congested based on the intersection timing signal scheme information, and output the congestion detection result of the target section with time information, congestion degree quantitative label and congestion frequency quantitative label;

[0042] Congestion cause analysis expert role Agent, used to analyze the causes of congestion including supply and demand matching, section interference, traffic organization, event detection, road network and scheme, and time and space coordination based on congestion detection results;

[0043] A congestion suggestion expert Agent, used to provide at least one congestion suggestion according to the cause of congestion;

[0044] The congestion management expert Agent is used to provide a management solution including variable lane switching, reverse variable lane switching and intersection signal solution optimization according to at least one congestion suggestion.

[0045] Optionally, the congestion detection expert role Agent includes:

[0046] The intersection timing signal scheme information detection sub-agent is used to automatically obtain and process the all-day checkpoint vehicle passing data, generate a vehicle passing data set based on the vehicle flow direction classification standard through time division and vehicle flow direction aggregation, use autocorrelation analysis and ordered clustering to identify the signal cycle and divide the control period, and use Fourier series to perform curve fitting to identify the maximum point, combine the time data corresponding to each phase switching point of the fitting curve to compare and analyze, determine the phase sequence and phase duration of each signal phase, and then integrate and output the intersection timing signal scheme information including cycle, period, phase, phase sequence and duration;

[0047] The congestion detection result output sub-agent is used to perform time correction on the license plate capture data from the electronic police at the checkpoint based on the intersection timing signal scheme information, and then perform correction after interference removal, clustering, and the introduction of at least one correction factor related to the travel speed, to obtain the real-time travel speed of the section without being affected by the red light, and combine the obtained section length information to determine whether the corresponding section is congested within a given time period, and output the congestion detection result of the target section with time information, congestion degree quantitative label, and congestion frequency quantitative label.

[0048] Optionally, the congestion cause analysis expert role Agent includes:

[0049] The supply-demand matching index sub-agent is used to analyze the peak traffic of different road levels based on the congestion detection results, and output the road section traffic demand results according to the traffic sorting and the preset traffic threshold, and calculate the peak hour coefficient according to the maximum traffic volume in the selected peak period, evaluate the duration of the peak based on the peak hour coefficient, and identify the peak travel distribution results;

[0050] The cross-section interference sub-agent is used to collect and analyze the vehicle trajectory data of the target section, calculate the difference in the total amount of vehicle data of the upstream and downstream sections, and determine the interference of the attraction point; determine the interference of the section opening on the traffic flow according to the road grade of the section and the upstream and downstream intersection information of the adjacent sections; use the GPS location data of the bus station to count the number and frequency of buses arriving at the station within the set time period to determine the bus line and bus gathering interference; and comprehensively evaluate and output the interference of bus stops and bus lines on the traffic path according to the road grade of the section, the upstream and downstream intersection information of the adjacent sections, the number of bus lines and the location of the bus stations;

[0051] Traffic organization detection sub-agent is used to obtain and analyze the forward flow and reverse flow data of the target section within a preset time period, calculate the relative flow difference and tidal degree, so as to identify whether there is a tidal phenomenon in the import flow direction of the target point section, and output the description information of the import flow direction tidal phenomenon; according to the changes in the traffic volume in different directions of the section, determine whether there is a tidal phenomenon in both directions of the target section, and output the description information of the tidal phenomenon in both directions of the section; and by comparing the number of upstream import lanes and downstream export lanes, output the matching information of the upstream import and downstream export traffic capacity of the target section;

[0052] The event detection sub-agent is used to determine whether a traffic safety incident occurs on the target road section based on the obtained illegal accident information, and output the event description information of the target road section with time and location;

[0053] A road network and solution query sub-agent, used to query and provide road network and signal solution description information including intersection model, intersection channelization, neighbor relationship and real-time signal solution from at least one data source;

[0054] The spatiotemporal coordination sub-agent is used to determine the speed of the upstream and downstream of the target section, the length of the target section, the average speed during the congestion period of the upstream of the target section, the average speed during the congestion period of the downstream of the target section, and to obtain the optimal vehicle trajectory data by marking the path coordination direction, completing the missing trajectory data and connecting the trajectories of the acquired vehicle passing data; then, combining the intersection equipment type and the non-stop travel time, analyzing the parking situation of each intersection, calculating the path parking rate, and comparing the parking rate of the current section with that of the adjacent sections; when the speed of the upstream and downstream of the target section, the length of the target section, the average speed during the congestion period of the upstream of the target section, the average speed during the congestion period of the downstream of the target section, and the parking rate of the current section with that of the adjacent sections meet the set incoherence conditions, outputting the conclusion that the upstream and downstream sections of the target section are spatiotemporally incoherent; and, calculating the speed variation coefficient of multiple adjacent cycles of the target section by the ratio of the speed standard deviation to the speed average value, and if the variation coefficient meets the set unevenness conditions, outputting the conclusion that the target section is uneven.

[0055] Optionally, the congestion suggestion expert agent outputs the following suggestions based on the results returned by each sub-agent in the congestion cause analysis expert role agent after executing the corresponding process:

[0056] If the demand indicated by the traffic demand result of a road section is greater than the set demand threshold, a suggestion is outputted to set up an induction display screen upstream of the target road section for traffic diversion;

[0057] For peak travel concentration indicated by the peak travel distribution results, output suggestions for staggered travel or use of public transportation;

[0058] For attraction point interference, the output includes setting up an induction display screen upstream of the target road section to induce non-arrival traffic flow to choose an alternative path, the recommended location and number of openings in the attraction point area, and suggestions for concentrating the suction flow of the inducing road section to a fixed area;

[0059] For the disruption of road openings to traffic flow, the output includes suggestions for installing hard barriers, adding entry lanes, re-channeling lanes, marking waiting areas, and setting up green wave belts between adjacent intersections;

[0060] For bus route and bus cluster interference, the output includes suggestions for adjusting bus routes and arrival times;

[0061] For the interference of bus stops and bus lines on traffic routes, output suggestions for moving the location of bus stops upstream;

[0062] For the presence of tidal phenomena in the import flow direction description information, the output increases and switches to variable lanes, adjusts the intersection signal cycle duration, and reallocates the green light duration for each flow direction;

[0063] For the presence of tidal phenomena indicated by the bidirectional tidal phenomenon description information of the road section, the output includes suggestions for adding and switching tidal lanes, adjusting the signal cycle duration of the intersection, and reallocating the green light duration of each flow direction;

[0064] If the matching information of the traffic capacity of the upstream entrance and the downstream exit of the target road section indicates a mismatch, a suggestion is output to increase the number of lanes merging into the widening section of the exit road;

[0065] For the conclusion that the upstream and downstream sections of the target section are not spatiotemporally coherent, the output includes suggestions for strengthening queue length monitoring, optimizing signal linkage release, setting up motor vehicle and non-motor vehicle isolation belts in the middle of adjacent intersection sections, and adjusting the number of upstream and downstream lanes to be consistent;

[0066] The conclusion of the traffic irregularity of the target road section is output through suggestions including optimizing queue lanes, monitoring queue length data, setting common cycles at adjacent intersections, and setting green wave belts between adjacent intersections;

[0067] For illegal parking events indicated by the event description information, the output includes suggestions for automatically identifying and setting up smart parking spaces using video technology;

[0068] For accident events indicated by the event description information, the output includes remote video assessment, electronic document submission for handling minor accidents that meet the set conditions, and recommendations for establishing an efficient linkage mechanism for non-minor accidents.

[0069] Optionally, the congestion management expert Agent includes:

[0070] The variable lane switching sub-agent is used to determine the positive deviation of the left-turn flow and the straight-through flow at the same entrance and the reverse deviation after switching a lane based on the intersection model provided by the traffic organization detection sub-agent and the lane-level flow data produced by the supply-demand matching index sub-agent. When the positive deviation and the reverse deviation meet the preset variable lane switching conditions, the recommended lane flow direction is output and automatically sent to the signal control system connected by the communication;

[0071] The tidal lane switching sub-agent is used to determine the positive deviation of the flow in different directions in both directions of the road section and the reverse deviation after switching a lane to the target lane based on the intersection model provided by the traffic organization detection sub-agent and the lane-level flow data produced by the supply-demand matching index sub-agent. When the positive deviation and the reverse deviation meet the preset tidal lane switching conditions, the recommended switching section is output and automatically sent to the signal control system connected by the communication;

[0072] The intersection signal scheme optimization sub-agent is used to analyze the congestion status of all entrance sections according to the intersection model and real-time signal scheme provided by the traffic organization detection sub-agent and the lane-level flow data produced by the supply-demand matching index sub-agent, and determine whether the intersection is congested. When the intersection is congested, according to the phase-level phase sequence S1, signal cycle and green wave coordination path status in the real-time signal scheme, the cycle length and the green light duration of the phase-level phase sequence S1 are adjusted to obtain the optimized signal scheme, and automatically send it to the signal control system connected by communication; wherein, when optimizing the signal scheme, if the current signal cycle does not reach the maximum signal cycle and the intersection is not on the green wave coordination path, the cycle length is increased and the green light duration of the phase-level phase sequence S1 is gradually increased; when optimizing the signal scheme, if the current signal cycle does not reach the maximum signal cycle but the intersection is on the green wave coordination path, the cycle length is kept unchanged, the green light duration of the phase-level phase sequence S1 is gradually increased, and the green light duration of the non-phase sequence phase S1 is reduced.

[0073] (III) Beneficial effects

[0074] The beneficial effects of the present invention are:

[0075] First, the present invention can parse user questions, formulate a plan to respond to user questions, arrange different sub-agents to work in sequence, and prioritize at least one round of interaction by dynamically selecting and activating the underlying multi-expert Agents most relevant to the question content, thereby ensuring the professionalism and accuracy of user questions, thereby improving the efficiency and accuracy of answers.

[0076] If the first round of interaction does not meet the user's needs, the system will continue to select other underlying multi-expert agents for at least one round of interaction based on the user's needs and the results of the first round of interaction, until it gradually approaches and finds a solution that meets the user's needs, and integrates and promptly provides it to the user. The above intelligent overall arrangement makes the traffic congestion management process more orderly and efficient, and can be triggered regularly or manually, so as to respond to and solve traffic congestion problems in a timely manner.

[0077] Secondly, the design of the underlying multi-expert agent is also a highlight of the invention. These agents are responsible for selecting appropriate algorithms for congestion problem identification, cause analysis, suggestions and governance solutions based on data sources of different modalities. This design with clear division of labor and professional expertise enables each agent to play the greatest role in its field, thereby improving the professionalism and accuracy of the entire governance process.

[0078] In addition, the present invention also realizes the scheduling and unlimited expansion of multiple expert agents by introducing a large language model (LLM). This not only enhances the flexibility and scalability of the system, but also greatly reduces the interaction cost, making it easier for users to use. This innovative application solves the technical problem that traditional traffic congestion management methods cannot comprehensively and efficiently deal with the complex and changeable causes of congestion, making the cause analysis more accurate and comprehensive, and the collaborative positioning process more time-saving and labor-saving, reducing the impact of data differences and information delays.

[0079] Therefore, the present invention introduces technical means such as Multi-Agent collaboration and large language models, and realizes comprehensive, efficient and intelligent management of urban traffic congestion problems by implementing dynamic interaction and feedback processes, bringing new solutions and ideas to urban traffic congestion management. BRIEF DESCRIPTION OF THE DRAWINGS

[0080] Figure 1 A schematic diagram of the steps of the method provided by an embodiment of the present invention;

[0081] Figure 2 A schematic diagram of the overall process of the method provided by an embodiment of the present invention;

[0082] Figure 3 A specific schematic diagram of step S1 of the method provided in an embodiment of the present invention;

[0083] Figure 4 A specific schematic diagram of step S3 of the method provided in an embodiment of the present invention;

[0084] Figure 5 An overall step implementation diagram of a specific embodiment of the method provided by the embodiment of the present invention;

[0085] Figure 6A user input interface of a human-computer interaction interface in a specific embodiment of the method provided by an embodiment of the present invention;

[0086] Figure 7 A road intersection status display interface of a human-computer interaction interface in a specific embodiment of the method provided in an embodiment of the present invention;

[0087] Figure 8 An overall traffic situation display interface of a human-computer interaction interface in a specific embodiment of the method provided by an embodiment of the present invention;

[0088] Fig. 9 A display interface of congestion statistics data of a target road section of a human-computer interaction interface in a specific embodiment of the method provided by an embodiment of the present invention;

[0089] Fig.10 A display interface of the unobstructed statistical data of the target road section of the human-computer interaction interface in a specific embodiment of the method provided by the embodiment of the present invention. DETAILED DESCRIPTION

[0090] In order to better explain the present invention and facilitate understanding, the present invention is described in detail below through specific implementation modes in conjunction with the accompanying drawings.

[0091] Prior to this, in order to facilitate understanding of the technical solution provided by the present invention, some concepts are first introduced below.

[0092] Agent: refers to intelligent entities with specific functions and autonomy, which can perform corresponding tasks based on preset rules and algorithms. These agents play different roles in the framework and work together to achieve efficient management of traffic congestion.

[0093] Green wave coordinated path: refers to adjusting the timing of traffic lights at each intersection on a designated traffic route so that vehicles traveling at a certain speed can encounter green lights when passing through several adjacent intersections, thereby achieving the purpose of quickly passing through the section without stopping.

[0094] like Figure 1 and Figure 2As shown, an urban traffic congestion management method based on a general large language model proposed in an embodiment of the present invention includes: receiving and parsing the user's question content, dynamically selecting and activating the underlying expert agent most relevant to the question content for a first round of interaction; judging whether the first round of interaction results meet the user's needs based on the user's question content and the feedback information of the underlying expert agent in the first round of interaction; if the first round of interaction results meet the user's needs, entering the information aggregation and feedback stage; if the first round of interaction results do not meet the user's needs, continuing to select other underlying expert agents for at least one round of interaction based on the user's needs and the first round of interaction results, gradually approaching until a solution that meets the user's needs is found; after the management process is completed, the information aggregation and feedback agent in the underlying expert agent obtains and integrates the feedback information of all underlying expert Agens participating in the interaction and the user's question content, performs structured processing according to a predefined data format, and feeds back the final result to the user in at least one form.

[0095] First, the present invention can parse user questions, formulate a plan to respond to user questions, arrange different sub-agents to work in sequence, and prioritize at least one round of interaction by dynamically selecting and activating the underlying multi-expert Agents most relevant to the question content, thereby ensuring the professionalism and accuracy of user questions, thereby improving the efficiency and accuracy of answers.

[0096] If the first round of interaction does not meet the user's needs, the system will continue to select other underlying multi-expert agents for at least one round of interaction based on the user's needs and the results of the first round of interaction, until it gradually approaches and finds a solution that meets the user's needs, and integrates and promptly provides it to the user. The above intelligent overall arrangement makes the traffic congestion management process more orderly and efficient, and can be triggered regularly or manually, so as to respond to and solve traffic congestion problems in a timely manner.

[0097] Secondly, the design of the underlying multi-expert agent is also a highlight of the invention. These agents are responsible for selecting appropriate algorithms for congestion problem identification, cause analysis, suggestions and governance solutions based on data sources of different modalities. This design with clear division of labor and professional expertise enables each agent to play the greatest role in its field, thereby improving the professionalism and accuracy of the entire governance process.

[0098] In addition, the present invention also realizes the scheduling and unlimited expansion of multiple expert agents by introducing a large language model (LLM). This not only enhances the flexibility and scalability of the system, but also greatly reduces the interaction cost, making it easier for users to use. This innovative application solves the technical problem that traditional traffic congestion management methods cannot comprehensively and efficiently deal with the complex and changeable causes of congestion, making the cause analysis more accurate and comprehensive, and the collaborative positioning process more time-saving and labor-saving, reducing the impact of data differences and information delays.

[0099] Therefore, the present invention, by constructing a Multi-Agent collaborative traffic congestion management framework and introducing technical means such as a large language model, realizes a comprehensive, efficient and intelligent management of urban traffic congestion problems by implementing a dynamic interaction and feedback process, bringing new solutions and ideas to urban traffic congestion management.

[0100] In order to better understand the above technical solution, exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided to enable a clearer and more thorough understanding of the present invention and to fully convey the scope of the present invention to those skilled in the art.

[0101] Specifically, an embodiment of the present invention provides a method for managing urban traffic congestion based on a universal large language model, including:

[0102] S1. Receive and parse the user's question content, dynamically select and activate the underlying expert Agent most relevant to the question content for the first round of interaction.

[0103] Furthermore, before step S1, the method further includes: constructing and initializing a traffic congestion management framework, loading a general large language model and a traffic professional knowledge base; creating at least one underlying expert agent in the traffic congestion management framework, and configuring each underlying expert agent to perform a set functional task; establishing a communication mechanism between the collaborative coordination agent and the underlying expert agent, and setting an interface between the agents to achieve at least one operation of dynamic selection, activation and interaction;

[0104] Specifically, at least one underlying expert agent includes: a congestion detection expert role Agent, which is used to identify the real-time signal scheme of the corresponding intersection based on the checkpoint passing data, output the intersection timing signal scheme information, and determine whether the target section is congested based on the intersection timing signal scheme information, and output the congestion detection result of the target section with time information, congestion degree quantitative label and congestion frequency quantitative label; a congestion cause analysis expert role Agent, which is used to analyze the congestion causes including supply and demand matching, section interference, traffic organization, event detection, road network and scheme, and time and space coordination based on the detection results of the target section; a congestion suggestion expert Agent, which is used to provide targeted congestion suggestions according to the causes of congestion; a congestion management expert Agent, which is used to provide management solutions including variable lane switching, reverse variable lane switching and intersection signal scheme optimization according to the congestion situation.

[0105] Furthermore, if Figure 3 As shown, step S1 includes:

[0106] S11. Receive question content input by the user through the user interface.

[0107] S12. Use the pre-built large language model to parse the user's question content, extract keywords and semantic information, and determine the user's needs.

[0108] S13. According to the analyzed user needs, the underlying multi-expert Agent most relevant to the question content is dynamically selected through predefined rules.

[0109] It should be understood that the predefined rules include: if the question involves congestion status identification, the congestion detection expert agent is activated; if the question is about the cause of congestion, the congestion cause analysis expert agent is activated; if congestion management suggestions are needed, the congestion suggestion expert agent is activated; if it involves the implementation of congestion management solutions, the congestion management expert agent is activated.

[0110] S14, sending an activation instruction to the selected bottom-level expert Agent to activate the selected bottom-level multi-expert Agent to perform a first round of interaction.

[0111] S2. Based on the user's question content and the feedback information of the underlying expert agent in the first round of interaction, determine whether the results of the first round of interaction meet the user's needs.

[0112] S3. If the results of the first round of interaction meet the user's needs, the information aggregation and feedback stage will be entered; if the results of the first round of interaction do not meet the user's needs, other underlying expert agents will continue to be selected for at least one round of interaction based on the user's needs and the results of the first round of interaction, gradually approaching until a solution that meets the user's needs is found.

[0113] Furthermore, if Figure 4 As shown, step S3 includes:

[0114] S31. If the result of the first round of interaction does not meet the user's needs, identify information points that need to be supplemented based on the result of the first round of interaction and the user's needs.

[0115] S32: Select an underlying expert Agent that is most relevant to the information point to be supplemented for the next round of interaction.

[0116] S33: delivering the user requirements and the results of the previous round of interactions to the underlying expert Agent selected in the current round, and receiving and processing the feedback information of the underlying expert Agent.

[0117] S34. According to the feedback information of the bottom-level expert agent selected in the current round, it is judged again whether the current interaction result meets the user's needs.

[0118] S35. If the current interaction result meets the user's needs, the result summary stage is entered; if the current interaction result does not meet the user's needs, other relevant expert agents are selected to continue the next round of interaction. In each round of interaction, the selection strategy for the underlying expert agent is dynamically adjusted according to the feedback information of the previous round, so as to gradually approach until a solution that meets the user's needs is found.

[0119] S4. After the governance process is completed, the information summary feedback agent in the underlying expert agent obtains and integrates the feedback information of all underlying expert agents participating in the interaction and the user's questions, performs structured processing according to the predefined data format, and feeds back the final results to the user in at least one form.

[0120] On the other hand, an embodiment of the present invention provides a framework for managing urban traffic congestion based on a general large language model, including:

[0121] The collaborative coordination agent is used to execute the above method. The collaborative coordination agent analyzes user questions, formulates a plan to respond to user questions, arranges the sequential work of different sub-agents, and can trigger the traffic congestion management process regularly or manually.

[0122] At least one underlying multi-expert Agent is used to select corresponding algorithms according to data sources of different modes to implement urban traffic congestion control measures including congestion detection, congestion cause analysis, congestion suggestions and control plan formulation.

[0123] Furthermore, at least one underlying expert agent includes: a congestion detection expert role Agent, which is used to identify and analyze the real-time signal scheme of the corresponding intersection based on the acquired checkpoint passing data, output the intersection timing signal scheme information, and determine whether congestion occurs in the target section according to the intersection timing signal scheme information, and output the congestion detection result of the target section with time information, congestion degree quantitative label and congestion frequency quantitative label; a congestion cause analysis expert role Agent, which is used to analyze the congestion causes including supply and demand matching, section interference, traffic organization, event detection, road network and scheme, and time and space coordination; a congestion suggestion expert Agent, which is used to provide at least one targeted congestion suggestion according to the congestion cause; a congestion management expert Agent, which is used to provide a management solution including variable lane switching, reverse variable lane switching and intersection signal scheme optimization according to at least one congestion suggestion.

[0124] Furthermore, the congestion detection expert role Agent is configured as a congestion detection and discovery expert, an intelligent agent that answers questions about congestion identification and discovery, which includes:

[0125] The intersection timing signal scheme information detection sub-agent is used to automatically obtain and process the all-day checkpoint vehicle passing data, generate a vehicle passing data set based on the traffic flow direction classification standard through time division and traffic flow aggregation, use autocorrelation analysis and ordered clustering to identify the signal cycle and divide the control period, and use Fourier series to perform curve fitting to identify the maximum point, combine the time data corresponding to each phase switching point of the fitting curve to compare and analyze, determine the phase sequence and phase duration of each signal phase, and then integrate and output the intersection timing signal scheme information including cycle, period, phase, phase sequence and duration.

[0126] More specifically, the intersection configuration signal scheme information detection sub-agent includes:

[0127] The acquisition and processing module is used to automatically acquire and process the vehicle passing data at the checkpoint throughout the day. By dividing the time dimension and aggregating the traffic flow direction, a vehicle passing data set based on the traffic flow direction is generated. The vehicle passing data includes but is not limited to vehicle information, intersection details, import information, lane information and passing time.

[0128] The cycle identification and time period division module is used to use autocorrelation analysis to process the time series of vehicle passing data in each flow direction in the vehicle passing data set, identify the signal cycle data, and then group the identified signal cycle data through ordered clustering to determine the division scheme of the signal control time period.

[0129] The curve fitting and feature point identification module is used to arrange the vehicle passing data of each flow direction in each signal control period in time sequence, and use Fourier series to perform curve fitting to identify the maximum value points and corresponding time data in each fitting curve.

[0130] The signal phase and timing determination module is used to compare and analyze the time data of the maximum points of the fitting curves of each flow direction in the same time period, classify the traffic flows with different directions but the maximum point times that do not differ by more than a preset deviation threshold into the same signal phase, and then perform graphical superposition operations on the fitting curves of different signal phases in the same signal cycle, and use the maximum point and intersection point information of the superimposed fitting curves to determine the phase sequence and phase duration of each signal phase.

[0131] The output module is used to integrate and output the final intersection timing signal scheme information including signal cycle, signal control period, signal phase, phase sequence and phase duration.

[0132] The congestion detection result output sub-agent is used to perform time correction on the license plate capture data from the electronic police at the checkpoint based on the intersection timing signal scheme information, and then perform correction after interference removal, clustering, and the introduction of at least one correction factor related to the travel speed, to obtain the real-time travel speed of the section without being affected by the red light, and combine the obtained section length information to determine whether the corresponding section is congested within a given time period, and output the congestion detection result of the target section with time information, congestion degree quantitative label, and congestion frequency quantitative label.

[0133] More specifically, the congestion detection result output sub-agent includes:

[0134] The receiving and processing module is used to receive and process the license plate capture data and intersection timing signal scheme information from the electronic police at the checkpoint.

[0135] The time correction module is used to perform time correction on the license plate capture data according to the intersection timing signal scheme information to eliminate the influence of traffic lights on vehicle driving time, thereby obtaining travel time data.

[0136] The interference removal module is used to use the corrected travel time data, combined with the preset free flow travel time and intersection cycle information, to eliminate the interference of periodic traffic flow fluctuations on congestion detection results, and obtain processed travel time data.

[0137] The clustering and re-correction module is used to perform cluster analysis on the processed travel time data, distinguish between samples that do not wait for red lights and samples that wait for red lights, calculate the average travel speed of each type of samples respectively, and introduce correction factors to obtain the travel speed of the real-time road section when it is not affected by the red light.

[0138] The congestion detection result output module is used to determine whether the target road section is congested within a given time period based on the calculated travel speed and the road section length information, and to label the congestion situation with a corresponding quantitative label according to the preset congestion degree quantitative standard, and to determine the congestion frequency according to the statistics of the number of congestion occurrences during the same period of the past week, and to label the target road section with a congestion frequency quantitative label, and to output the congestion detection result containing time information, congestion degree quantitative labels, and congestion frequency quantitative labels. Specifically, the congestion detection result output module determines whether the corresponding road section is congested within a given time period, and outputs quantitative description information such as the target road section with time and smoothness, mild congestion, moderate congestion, severe congestion, speed, etc. According to the congestion period of the road section, the number of congestion occurrences during the same period of the past week of the road section is counted. If the proportion exceeds half, frequent congestion is output, otherwise occasional congestion is output.

[0139] Furthermore, the Congestion Cause Analysis Expert role Agent is configured as a congestion cause analysis expert to answer questions about congestion causes, such as supply and demand matching, section interference, traffic organization, event detection, road network and scheme, time and space coordination, etc., which include:

[0140] The supply-demand matching index sub-agent is used to analyze the peak flow of different road levels based on the congestion detection results, and output the section flow demand results according to the flow sorting and preset flow thresholds. In addition, the peak hour coefficient is calculated according to the maximum traffic volume in the selected peak period, and the duration of the peak is evaluated based on the peak hour coefficient to identify the peak travel distribution results.

[0141] In one embodiment, the supply-demand matching index sub-agent identifies and provides information on the road section flow and whether there is excessive demand and peak travel concentration on the road section. The specific analysis process is as follows:

[0142] (1) Based on the congested sections output by the congestion detection results, analyze the peak flow of different road levels (main roads, secondary roads, and branch roads) and classify them. Sorting them according to the hourly flow during the day (6:00-20:00), averaging the percentile values ​​above 90%, and outputting the conclusion that the demand is large when the average value is reached. Analyze the different road levels and sort them according to the flow, and take the flow of 85% as the basic flow. When the flow of the congested section is greater than the basic flow, output the conclusion that the flow and demand of the section are too large;

[0143] (2) Calculate the peak hour coefficient. The peak hour coefficient reflects the distribution of time periods with large traffic within an hour. The peak hour coefficient is always less than 1. It reflects the characteristics of peak time distribution. The closer it is to 1, the longer the peak duration is, and the closer it is to 0, the shorter the peak duration is. Calculation: First, select the peak time period, which can be divided into 15-minute intervals to obtain the maximum traffic volume in 15 minutes. The peak hour coefficient = peak hour traffic volume / 4 / 15-minute maximum traffic volume. When the peak hour coefficient is >0.8, the output flow and peak travel are more concentrated.

[0144] The cross-section interference sub-agent is used to collect and analyze the vehicle trajectory data of the target section, calculate the difference in the total amount of vehicle data of the upstream and downstream sections, so as to determine the interference of the attraction point; determine the interference of the section opening on the traffic flow according to the road grade of the section and the upstream and downstream intersection information of the adjacent sections; use the GPS location data of the bus station to count the number and frequency of buses arriving at the station within the set time period to determine the bus line and bus gathering interference; and, based on the road grade of the section, the upstream and downstream intersection information of the adjacent sections, the number of bus lines and the location of the bus stations, comprehensively evaluate and output the interference of bus stations and bus lines on the traffic path.

[0145] In one embodiment, the cross-section interference sub-agent is configured to provide information on whether there is an import flow direction and a bidirectional tidal phenomenon in the road section, and a mismatch in the import and export capacity. The specific analysis process is as follows:

[0146] (1) Select the direction side of the congested road section (if a road section is congested from east to west, then select the northern point of interest area), and count the difference between the total amount of vehicle passing data in the upstream section and the total amount of vehicle passing data in the downstream section in the OD pair (starting point-end point pair) of the vehicle passing trajectory of this road section. If the difference accounts for more than half of the total amount of vehicle passing data in the downstream section, then the attraction point interference is output.

[0147] (2) Analyze the intersections upstream and downstream of adjacent sections. According to the road grade of the section, if there is a unit exit within 100 meters of the stop line of the entrance road and 80 meters of the exit road of the intersection, the output section opening impact is considered. Whether there is an entrance and exit within the exit road (main road: 80 meters, secondary road: 50 meters, branch road: 30 meters); whether there is an entrance and exit within the entrance road (main road: 100 meters, secondary road: 70 meters, branch road: 50 meters). The opening spacing is <30 meters.

[0148] (3) According to the GPS location of the bus stop, the number of buses arriving at the station within 5 minutes is counted. If the number is greater than 6, the impact on the bus route is output. When a bus enters the station, the time starts to count whether there is a second bus entering the station within 50 seconds, and the bus route and bus gathering interference are output.

[0149] (4) Analyze the upstream and downstream intersections of adjacent sections. According to the road grade of the section, the distance from the entrance road to the intersection is 80 meters and the distance from the exit road is 40 meters. The data only has the coordinates of the center point of the intersection. The distance is converted according to 3.3 meters / lane. If there are more than 3 bus routes, the distance between the station and the intersection is more than 80 meters, and there is a left-turn line, then the bus station and bus line path interference are output.

[0150] The traffic organization detection sub-agent is used to obtain and analyze the forward and reverse flow data of the target section within a preset time period, calculate the relative flow difference and tidal degree, so as to identify whether there is a tidal phenomenon in the import flow direction of the target point section, and output the description information of the import flow direction tidal phenomenon; according to the changes in the traffic volume in different directions of the section, it is used to determine whether there is a tidal phenomenon in both directions of the target section, and output the description information of the tidal phenomenon in both directions of the section; and by comparing the number of upstream import lanes and downstream export lanes, it outputs the matching information of the upstream import and downstream export capacity of the target section.

[0151] In a specific embodiment, the traffic organization detection sub-agent is configured to provide information on whether there is an import flow and a bidirectional tidal phenomenon on the road section, and the mismatch of import and export road capacity. The specific analysis process is as follows:

[0152] (1) Identify whether there is a tidal phenomenon at the entrance of the road section corresponding to the target point within a given time period, which is the prerequisite for switching the variable lane, and output the description information of the tidal phenomenon of the entrance flow direction.

[0153] (2) Determine whether there is tidal phenomenon in both directions of the road section according to the different directions of the road section, which is the prerequisite for switching the reverse variable lane, and output the description information of the tidal phenomenon in both directions of the road section.

[0154] It should be understood that in modern urban traffic management, the vehicle flow during the morning and evening peak hours shows an obvious "tidal phenomenon", that is, the vehicle flow in the direction of entering the city and leaving the city is significantly different at different times. Every morning, the flow in the direction of entering the city is large and the reverse flow is small, while in the evening, the flow in the direction of leaving the city is large and the flow in the direction of entering the city is small. In an embodiment of the present invention, a tidal phenomenon determination process is provided, as follows:

[0155] The import traffic data and the adjacent relationship table are read, the import traffic data is preprocessed, and the traffic data of different imports of the same intersection are converted from array form to separate row storage to form a data set.

[0156] The input signal machine ID and import identification ID are received, and the associated signal machine identification and associated import identification of the corresponding import are obtained by querying the adjacent relationship table.

[0157] According to the signal machine ID and the import identification ID, the first flow data that meets the conditions is screened out from the data set and recorded as flow_out; at the same time, according to the associated signal machine identification and the associated import identification, the second flow data that meets the conditions is screened out and recorded as flow_in.

[0158] Using time as an index, the first traffic data flow_out and the second traffic data flow_in are merged to form a new data set, in which each row of records represents the forward and reverse traffic flow of a road section within a specific time period.

[0159] In the new data set, the first flow data flow_out is x, the second flow data flow_in is y, according to Calculate the Gap value and add it to the data set as a new column; when the denominator is 0, set the Gap value to 0.

[0160] According to the calculated Gap value, further |gap| -|∑ gap |or 2 ∑ |gap| -|∑ gap |Calculate the tidal degree (tidalDegree) and output it.

[0161] Then, according to the tidal degree, it is determined whether there is tidal phenomenon at the corresponding road section entrance or in both directions of the road section within a given time period.

[0162] (3) The problem of import and export road capacity matching was found: when the number of straight lanes on the upstream import road is greater than or equal to the number of downstream export roads, it means that there is a mismatch between the upstream import and downstream export roads of the road section, and the output import and export road capacities are mismatched.

[0163] The event detection sub-agent is used to determine whether a traffic safety incident has occurred on the target road section based on the obtained illegal accident information, and output the event description information of the target road section with time and location. The event detection sub-agent is configured to provide illegal parking and accident event detection information. Based on the illegal accident information, it is determined whether a traffic safety incident has occurred on the corresponding road section of the target point within a given time period, and the event description information of illegal parking, accidents, etc. with time and location is output.

[0164] The road network and scheme query sub-agent is used to query and provide road network and signal scheme description information including intersection model, intersection channelization, neighbor relationship and real-time signal scheme from at least one data source. The road network and scheme query sub-agent is configured to provide intersection model, intersection channelization, neighbor relationship and real-time signal scheme information; query intersection model, intersection channelization, neighbor relationship and real-time signal scheme information through multiple data sources, and output road network and signal scheme description information.

[0165] The spatiotemporal coordination sub-agent is used to determine the speed of the upstream and downstream of the target section, the length of the target section, the average speed during the congestion period of the upstream of the target section, the average speed during the congestion period of the downstream of the target section, and to obtain the optimal vehicle trajectory data by marking the path coordination direction, completing the missing trajectory data and connecting the trajectories of the acquired vehicle passing data; then, combining the intersection equipment type and the non-stop travel time, analyzing the parking situation of each intersection, calculating the path parking rate, and comparing the parking rate of the current section with that of the adjacent sections; when the speed of the upstream and downstream of the target section, the length of the target section, the average speed during the congestion period of the upstream of the target section, the average speed during the congestion period of the downstream of the target section, and the parking rate of the current section with that of the adjacent sections meet the set incoherence conditions, outputting the conclusion that the upstream and downstream sections of the target section are spatiotemporally incoherent; and, calculating the speed variation coefficient of multiple adjacent cycles of the target section by the ratio of the speed standard deviation to the speed average value, and if the variation coefficient meets the set unevenness conditions, outputting the conclusion that the target section is uneven.

[0166] The spatiotemporal coordination sub-agent is configured to evaluate the spatiotemporal coordination effect, such as the road section traffic continuity and road section traffic smoothness evaluation. The specific analysis process is as follows:

[0167] (1) Analyze the speed of the upstream and downstream sections (same level, same number of lanes or same section name) of the congested section. 1. The length of the frequently congested section is <500 meters. 2. The average speed of the upstream section during the congestion period is >50% of the current section. 3. The average speed of the downstream section during the congestion period is >50% of the current section. 4. Obtain and mark the vehicle passing data at each intersection to form vehicle trajectory data; secondly, for intersections with missing passing data, use the vehicle speed and adjacent intersection data to repair and complete the trajectory; then, extract and connect the trajectory of the same vehicle from the vehicle trajectory data of adjacent time periods to obtain the optimal vehicle trajectory data; then, combine the intersection equipment type and the non-stop travel time of the section to analyze the parking situation at each intersection, calculate the path parking rate and the intersection continuous traffic rate; finally, by comparing the parking rate of the current section with that of the adjacent section, judge the upstream and downstream spatiotemporal coordination. If the parking rate of the current section is greater than that of the adjacent section or meets specific conditions, output the conclusion that the upstream and downstream sections are not connected. When 1.2.3 are met at the same time or 4 is met, the output shows that the upstream and downstream sections are not connected.

[0168] (2) Analyze the speed variation coefficient of multiple adjacent cycles on this section of road. The coefficient of variation = speed standard deviation / speed average. When the coefficient of variation > 0.8, the output section has low traffic smoothness.

[0169] Furthermore, the congestion suggestion expert agent acts as a knowledge base, and outputs suggestions including:

[0170] (1) For the demand indicated by the traffic demand result of the road section that is greater than the set threshold, a suggestion is output to set up an induction display screen upstream of the target road section for diversion. This is for the case where the demand of the road section is too large. If the current travel demand of (XXX road section) is greater than the travel supply capacity of the road section, an induction display screen can be set up upstream of (XXX road section) to guide the traffic flow arriving at the road section to (XXX road, direction) for diversion.

[0171] (2) For peak travel concentration indicated by the peak travel distribution results, output suggestions for staggered travel or use of public transportation. This is for situations where peak travel is relatively concentrated. If the peak travel concentration phenomenon on the current road section is more serious, it is recommended to stagger travel or use public transportation, which requires reducing the number of cars on the road or increasing the car load rate.

[0172] (3) For attraction point interference, output suggestions on setting up an induction display screen upstream of the target road section to induce non-arrival traffic flow to choose alternative routes, optimizing the access design of the attraction point area opening, and inducing the road section suction and emission flow to concentrate in a fixed area. This is for the case of attraction point interference. If there is a high suction and emission volume around the current (XXX road section), it will affect the traffic efficiency of the road section. The optimization suggestions are as follows:

[0173] 1. An induction display screen can be set up upstream of (XXX road section) to guide non-arrival traffic flow to choose alternative routes. Considering setting up alternative routes or evacuation channels can strengthen the connection with surrounding roads and ensure that the traffic flow at the attraction point can be smoothly integrated into the surrounding road network to cope with emergencies or traffic pressure during peak hours.

[0174] 2. The openings in the attraction area around the road section are designed with a refined access design including the recommended locations and recommended numbers of openings in the attraction area, including:

[0175] Selection of opening location: Select the entrance and exit location of the attraction point based on the traffic flow of the attraction point, the surrounding road conditions and the traffic flow characteristics, and avoid setting the opening directly on the main road with busy traffic to reduce interference with the traffic flow of the main road.

[0176] Control of the number of openings: Determine the number of entrances and exits based on the scale of the attraction and predicted traffic flow, so as to meet the traffic needs of the attraction and avoid traffic congestion and safety hazards caused by too many openings.

[0177] Optimization of opening layout: By optimizing the layout of entrances and exits, such as setting up left-turn or right-turn lanes, setting up traffic lights or signs and markings, etc., we can guide traffic flow in and out of attraction points in an orderly manner, thereby improving the smoothness and safety of traffic flow.

[0178] 3. Induce the suction and emission flow of the road section to concentrate on fixed passenger boarding and alighting areas.

[0179] (4) For the interference of road openings on traffic flow, the output includes suggestions for setting up hard barriers, adding entrance lanes, re-channeling lanes, marking waiting areas, and setting up green wave belts between adjacent intersections. If there is an opening in the entrance lane of (XXX road section) and the entry and exit affect the main line traffic, it is recommended: 1. If there is no central isolation, hard isolation should be set up, such as setting up a central isolation guardrail at the road opening and setting up pedestrian passage isolation facilities at pedestrian crossings, etc., to prevent pedestrians from crossing at will. 2. Optimize the organization of imports and exports, such as increasing the number of import lanes by shifting the center line of the road, compressing the exit lanes, etc., to ease the queue length of vehicles turning left or right; for example, remove the original traffic markings, and re-plan the lane layout according to the direction and data of traffic flow to make the lane division more reasonable; for example, set up straight waiting areas and left turn waiting areas in the import lanes, and use traffic lights to guide vehicles into the waiting areas in advance to improve traffic efficiency; for example, set up "green wave belts" between adjacent intersections, that is, adjust the green light start and closing time of the traffic light, so that vehicles traveling in a specific direction can encounter green lights continuously at one or more intersections, so that vehicles can pass through multiple intersections continuously, reducing the number and time of stopping and waiting.

[0180] (5) For the interference of bus routes and bus clusters, output suggestions for adjusting bus routes and arrival times; this is for the interference of bus routes and bus clusters, which can adjust the route arrival time, optimize the routes that arrive at the bus station during peak hours, and avoid concentrated arrival of multiple routes.

[0181] (6) For the interference of bus stops and bus lines on the traffic path, output the suggestion of moving the bus stop location upstream. This is for the influence of bus stops and bus lines on the path. If the current bus stop is close to the downstream intersection, the bus stops and passengers get on and off the bus and the bus starts and stops, which will affect the main line traffic. It is recommended to move the bus stop location upstream to reduce the impact on the traffic flow in the entrance road.

[0182] (7) For the tidal phenomenon described in the information on the tidal phenomenon of the import flow direction, if there is a tidal phenomenon, output suggestions to increase and switch to a variable lane or optimize the intersection signal. This is for the tidal phenomenon at the section import: when the current section structure cannot be changed, it is possible to appropriately consider adding a variable lane. When it is unconditional to switch the variable lane at the section import, consider optimizing the intersection signal, such as adjusting the intersection signal cycle duration and, within the adjusted signal cycle duration, redistributing the green light durations of each flow direction according to the traffic flow ratio of each flow direction. The following provides a specific adjustment plan:

[0183] The current signal cycle duration is X seconds. According to the recent traffic flow monitoring data, during the morning peak period (7:00 - 9:00), the traffic flow in the south-north direction increases significantly, and the traffic flow in the north-south direction decreases relatively. Therefore, it is decided to adjust the signal cycle duration during the morning peak period to Y seconds (Y > X) to increase the passing time in the south-north direction.

[0184] Next, within the adjusted signal cycle duration of Y seconds, redistribute the green light durations of each flow direction. During the morning peak period, the green light duration in the south-north direction is increased to A seconds to ensure that vehicles in this direction can pass smoothly. The green light duration in the north-south direction is adjusted to B seconds (B < A, but ensuring a reasonable passing time) to ensure that the traffic flows in other directions can also pass through the intersection within a reasonable time. The green light duration in the east-west direction is appropriately adjusted to C seconds according to the actual situation to ensure the coordination and balance of traffic flows in all directions at the intersection. During non-morning peak periods, according to the traffic flow monitoring data, adjust the signal cycle duration and green light duration in a timely manner to meet the traffic demands of different periods. For example, during the evening peak period (17:00 - 19:00), if the traffic flow in the north-south direction increases significantly, then adjust the signal cycle duration and green light duration accordingly.

[0185] (8) For the tidal phenomenon indicated by the information on the two-way tidal phenomenon of the section, if there is a tidal phenomenon, output suggestions including adding and switching tidal lanes, adjusting the intersection signal cycle duration, and within the adjusted signal cycle duration, redistributing the green light durations of each flow direction according to the traffic flow ratio of each flow direction. This is for the two-way tidal phenomenon of the section. If the current section structure has a certain number of lanes and no green belt, it is possible to consider adding tidal lanes. When it is unconditional to switch the two-way tidal variable lanes on the section, consider optimizing the intersection signal, such as adjusting the intersection signal cycle duration and, within the adjusted signal cycle duration, redistributing the green light durations of each flow direction according to the traffic flow ratio of each flow direction. In addition, for the intersection signal, if the current section structure has a certain number of lanes and no green belt, it is possible to consider adding tidal lanes.

[0186] (9) If the matching information of the upstream entrance and downstream exit of the target road segment indicates that the capacity does not match, a suggestion is output to increase the number of lane merges in the widening section of the exit road. This is to increase the number of lane merges in the widening section of the exit road (XX Road, XX Direction) in response to the mismatch in the capacity of the entrance and exit roads.

[0187] (10) For the conclusion that the upstream and downstream sections of the target section are not connected in time and space, the output includes suggestions for strengthening queue length monitoring, optimizing signal linkage release, setting up motor vehicle and non-motor vehicle separation strips in the middle of adjacent intersection sections, and adjusting the number of upstream and downstream lanes to be consistent. This is to address the traffic discontinuity of the upstream and downstream sections. We can strengthen the monitoring of queue length data and optimize the upstream and downstream signal linkage release strategy, and conduct refined traffic design for adjacent intersections. Specifically, this includes setting up motor vehicle and non-motor vehicle separation strips in the middle of adjacent intersection sections to ensure the smoothness and safety of different types of traffic flows; at the same time, adjusting the number of upstream and downstream lanes to keep them consistent to reduce traffic bottlenecks and congestion. In addition, the design may also involve optimization of signal light linkage release strategy, adjustment of intersection geometry, improvement of traffic signs and markings, improvement of pedestrian crossing facilities, etc., aiming to comprehensively improve the traffic efficiency and safety of adjacent intersections and achieve the spatiotemporal coherence and smoothness of traffic flow.

[0188] (11) For the conclusion of the uneven traffic on the target road section, the output includes suggestions for optimizing queue lanes, monitoring queue length data, setting common cycles at adjacent intersections, and setting green wave belts between adjacent intersections. This is for the road section with low traffic smoothness: focus on optimizing lanes with long queue lengths and conduct refined traffic design, strengthen queue length data monitoring and optimize upstream and downstream signal linkage release strategies. Specifically, this includes setting common cycles locally at adjacent intersections, that is, allowing several adjacent intersections to use the same or similar cycle lengths in signal light control to ensure that vehicles can encounter relatively consistent signal light states when passing through these intersections, reducing waiting time and the number of stops. At the same time, implement a green wave direction strategy, that is, by adjusting the green light on and off time of the signal light, vehicles traveling in a specific direction can continuously encounter green lights at one or more intersections, thereby forming a "green wave belt" and improving the speed and efficiency of vehicles in that direction. These measures are intended to improve the traffic conditions of the target road section, reduce traffic congestion, and enhance road capacity and driving experience.

[0189] (12) For illegal parking events indicated by the event description information, output suggestions for using video technology to automatically identify and set up smart parking spaces. This is for illegal parking events where vehicles illegally occupy lanes. It is recommended that traffic police strengthen law enforcement or use video technology to automatically identify illegal parking behaviors and send reminders to car owners. At the same time, setting up smart parking spaces can enable parking and going.

[0190] (13) For the accident events indicated by the event description information, output suggestions including remote video evaluation and electronic document submission to handle minor accidents that meet the set conditions, and establish an efficient linkage mechanism for non-minor accidents. This is for accident events, it is recommended to use remote video evaluation and electronic document submission to handle minor traffic accidents to reduce on-site processing time; for serious accidents, strengthen the coordination and cooperation between traffic police, fire, medical and other departments, and establish an efficient linkage mechanism.

[0191] Furthermore, the congestion management expert Agent includes:

[0192] The variable lane switching sub-agent is used to determine the positive deviation of the left-turn flow and the straight-through flow at the same entrance and the reverse deviation after switching a lane based on the intersection model provided by the traffic organization detection sub-agent and the lane-level flow data produced by the supply-demand matching index sub-agent. When the positive deviation and the reverse deviation meet the preset variable lane switching conditions, the recommended lane flow direction is output and automatically sent to the signal control system with communication connection.

[0193] The tidal lane switching sub-agent is used to determine the positive deviation of the flow in different directions in both directions of the road section and the reverse deviation after switching a lane to the target lane based on the intersection model provided by the traffic organization detection sub-agent and the lane-level flow data produced by the supply-demand matching index sub-agent. When the positive deviation and the reverse deviation meet the preset tidal lane switching conditions, the recommended switching section is output and automatically sent to the signal control system with communication connection.

[0194] The intersection signal scheme optimization sub-agent is used to analyze the congestion status of all entrance sections according to the intersection model and real-time signal scheme provided by the traffic organization detection sub-agent and the lane-level flow data produced by the supply-demand matching index sub-agent, and determine whether the intersection is congested. When the intersection is congested, according to the phase-level phase sequence S1, signal cycle and green wave coordination path status in the real-time signal scheme, the cycle length and the green light duration of the phase-level phase sequence S1 are adjusted to obtain the optimized signal scheme, and automatically send it to the signal control system connected by communication; wherein, when optimizing the signal scheme, if the current signal cycle does not reach the maximum signal cycle and the intersection is not on the green wave coordination path, the cycle length is increased and the green light duration of the phase-level phase sequence S1 is gradually increased; when optimizing the signal scheme, if the current signal cycle does not reach the maximum signal cycle but the intersection is on the green wave coordination path, the cycle length is kept unchanged, the green light duration of the phase-level phase sequence S1 is gradually increased, and the green light duration of the non-phase sequence phase S1 is reduced.

[0195] In another specific embodiment, the congestion management expert agent is configured as a congestion management expert, an intelligent agent that is good at answering congestion management solution questions. If it is necessary to solve the variable and reverse variable lane switching and intersection signal solution optimization problems, it specifically includes the following sub-agents:

[0196] (1) Variable lane switching agent:

[0197] It is good at solving the single-point control space - variable lane switching problem. Before calling, it needs to rely on the intersection model of the traffic organization detection agent, the variable lane problem discovery model algorithm and the flow data produced by the supply and demand matching index agent.

[0198] Input the intersection model and lane-level flow data, including intersection ID, entrance ID, lane ID, flow direction, lane type, and number of vehicles passing data, and determine whether the positive deviation of the left-turn flow and the straight-through flow at the same entrance is too large. At the same time, determine if a lane is switched (a left-turn lane is switched to a right-turn lane or a right-turn lane is switched to a left-turn lane), and calculate the reverse deviation. If the positive deviation is too large and the reverse deviation is not too large, the variable lane switching conditions are met, and the recommended lane flow direction is output, for example: left, left, straight, right, and then the plan is automatically sent to the signal control system.

[0199] (2) Reverse variable lane switching, i.e. tidal lane switching agent:

[0200] It is good at solving the single-point control space - the problem of reverse lane switching. Before calling, it needs to rely on the intersection model of the traffic organization detection agent, the reverse or tidal variable lane problem discovery model algorithm and the flow data produced by the supply and demand matching index agent.

[0201] Input the intersection model and import-level flow data, including intersection ID, import ID, and vehicle number data, to determine whether the forward deviation of the flow in different directions in both directions of the road section is too large. At the same time, if a lane is switched to the target lane, calculate the reverse deviation. If the forward deviation is too large and the reverse deviation is not too large, it meets the tidal lane switching conditions, output the recommended switching section, and then automatically send the plan to the signal control system.

[0202] (3) Intersection signal scheme optimization agent:

[0203] It is good at solving the single-point control time-intersection signal scheme optimization problem. Before calling, it needs to rely on the intersection model and real-time signal scheme of the traffic organization detection agent and the flow data produced by the supply and demand matching index agent.

[0204] Input the model of each intersection, the flow direction corresponding to the phase, and the current signal scheme of the intersection (including the intersection cycle, phase sequence phase and the traffic light duration corresponding to the phase), analyze the congestion status of all import sections, and judge whether the intersection is congested according to the section status (the intersection is congested if only one import is congested). When the intersection is congested, identify the phase level phase sequence S1 in the current signal scheme where the congested intersection import is located, and judge whether the current intersection signal cycle reaches the maximum signal cycle. If the maximum signal cycle is not reached and the current intersection is not on the green wave coordination path, the cycle duration is increased and the S1 green light duration is steadily increased; if the maximum signal cycle is not reached, but the current intersection is on the green wave coordination path, the cycle duration remains unchanged, and the S1 green light duration is steadily increased, and the green light duration of the phase sequence phase other than S1 is reduced. Output the optimized signal scheme, including the intersection cycle, phase sequence phase and the traffic light duration corresponding to the phase, and then automatically send it to the signal control system.

[0205] refer to Figure 5In a specific embodiment, the collaborative coordination agent selects to interact with multiple sub-agents in multiple rounds according to user questions. For example, when a user asks a question, "Which intersection is the most congested in the current area? What suggestions and solutions are there?", the collaborative coordination agent first selects the promoter closest to the question according to the user's question, that is, it will choose to interact with the congestion detection expert agent, and feed back the speed, congestion status and type information of the most congested section output by the congestion detection expert agent to the collaborative coordination agent. The collaborative coordination agent will then combine the user's question and the returned result of the congestion detection expert agent to determine whether the user's needs are met. If so, it will directly enter the information summary feedback agent. If not, it will continue to decide to interact with the congestion cause analysis expert agent in the sub-agent. The congestion cause analysis expert role agent will execute all algorithms of its sub-agents and return the cause of the problem to the collaborative coordination agent. The collaborative coordination agent will determine whether the user's needs are met according to the returned congestion cause results. If so, it will directly enter the information summary feedback agent. If not, it will continue to decide to interact with the congestion suggestion expert agent in the sub-agent. The collaborative coordination agent will select the closest suggestion plan from the congestion suggestion expert agent (knowledge base) according to the congestion cause results and feedback it to the collaborative coordination agent. The collaborative coordination agent will then combine the user's question and the return result of the congestion detection expert agent to determine whether the user's needs are met. If yes, it will directly enter the information summary feedback agent. If not, it will continue to decide to interact with the congestion management expert agent in the sub-agent. The congestion management expert agent selects the sub-agent corresponding to the closest promote according to the suggestion results. For example, when it comes to the import variable lane switching suggestion, the collaborative coordination agent can choose to interact with the variable lane switching agent with a single point control space, and give the information summary feedback agent according to the variable lane switching plan results, each step of the feedback results and the user's question content. The information summary feedback agent will organize and summarize the analysis results of each step and return the final result to the user. The return result can be data, a summary in one sentence, or the execution status of the plan. The returned data can be in the agreed data format with the front end and displayed in different charts to finally complete the congestion management closed-loop operation and analysis process report.

[0206] At the same time, reference Figure 6-Figure 10, Based on the above methods and frameworks, user input: When the user enters the query "Binjiang District traffic congestion", the big model will analyze and respond: There are currently 941 road sections in Binjiang District, among which only 3 sections are congested, while the rest remain unobstructed. In order to more intuitively display the congestion situation, the big model will further provide information on the top 5 congested sections and display them in the form of a list.

[0207] In summary, the present invention provides a method and framework for urban traffic congestion management based on a general large language model. The present invention is aimed at the comprehensive detection, cause analysis and management implementation of urban traffic congestion problems. The present invention is based on a general large language model + traffic professional knowledge base to build a Multi-Agent collaborative traffic congestion management framework, which realizes the comprehensive detection, cause analysis and management implementation of urban traffic congestion problems. The establishment of this framework not only improves the comprehensiveness and efficiency of traffic congestion management, but also effectively responds to the complex and changeable causes of congestion.

[0208] This framework consists of multiple components, including collaborative coordination agents, underlying multi-expert agents (including congestion detection expert agents, congestion cause analysis expert agents, congestion suggestion expert agents, and congestion management expert agents), and information summary feedback agents. Among them, the collaborative coordination agent is responsible for parsing user questions, formulating response plans, and scheduling each sub-agent to work in an orderly manner. This intelligent coordination mechanism makes the governance process more efficient and orderly, and can be triggered regularly or manually, thereby ensuring timely response and handling of traffic congestion problems.

[0209] In addition, the design of the underlying multi-expert Agent fully utilizes the advantages of specialization. They can accurately select appropriate algorithms to identify congestion problems and analyze causes based on data sources of different modalities, and put forward targeted suggestions and governance solutions. This specialized division of labor has significantly improved the professionalism and accuracy of the governance process.

[0210] It is worth mentioning that the present invention also innovatively introduces the large language model (LLM), which realizes the flexible scheduling and unlimited expansion capabilities of multiple expert agents. This measure not only greatly enhances the flexibility and scalability of the system, but also effectively reduces the interaction cost and improves the user experience. Through this innovative application, we have successfully solved the limitations of traditional traffic congestion management methods in dealing with complex and changeable congestion causes, making the cause analysis more accurate and comprehensive, and collaborative positioning more efficient and time-saving, thereby significantly reducing the adverse effects of data differences and information delays.

[0211] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0212] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems) and computer program products according to embodiments of the present invention. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions.

[0213] Although the preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments after knowing the basic creative concept. Therefore, the technical solution should be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present invention.

[0214] Obviously, those skilled in the art can make various modifications and variations to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention belong to the scope of the technical solution of the present invention and its equivalent technology, the present invention should also include these modifications and variations.

Claims

1. A method for managing urban traffic congestion based on a general large language model, characterized in that: include: Receive and analyze the user's questions, dynamically select and activate the underlying expert agent most relevant to the question content for the first round of interaction; Based on the user's question content and the feedback information from the underlying expert agent in the first round of interaction, determine whether the results of the first round of interaction meet the user's needs; If the results of the first round of interaction meet the user's needs, the information aggregation and feedback phase will begin; If the results of the first round of interaction do not meet the user's needs, then based on the user's needs and the results of the first round of interaction, continue to select other underlying expert agents for at least one round of interaction, gradually approaching until a solution that meets the user's needs is found; After the governance process is completed, the information summary feedback agent in the underlying expert agent obtains and integrates the feedback information of all underlying expert agents participating in the interaction and the user's questions, structures them according to the predefined data format, and feeds back the final results to the user in at least one form.

2. The urban traffic congestion management method based on a general large language model as claimed in claim 1, characterized in that: Before receiving and parsing the user's question content and dynamically selecting and activating the underlying expert agent most relevant to the question content for the first round of interaction, it also includes: Build and initialize the traffic congestion management framework, load the general large language model and traffic professional knowledge library; Create at least one underlying expert agent in the traffic congestion management framework, and configure each underlying expert agent to perform the set functional tasks; Establish a communication mechanism between the collaborative coordinating agent and the underlying expert agent, and set up interfaces between the agents to achieve dynamic selection, activation and interaction of at least one operation; Among them, at least one underlying expert agent includes: a congestion detection expert role Agent, which is used to identify the real-time signal scheme of the corresponding intersection based on the checkpoint passing data, output the intersection timing signal scheme information, and determine whether the target section is congested according to the intersection timing signal scheme information, and output the congestion detection result of the target section with time information, congestion degree quantitative label and congestion frequency quantitative label; a congestion cause analysis expert role Agent, which is used to analyze the congestion causes including supply and demand matching, section interference, traffic organization, event detection, road network and scheme, and time and space coordination based on the detection results of the target section; a congestion suggestion expert Agent, which is used to provide targeted congestion suggestions according to the congestion causes; a congestion management expert Agent, which is used to provide management solutions including variable lane switching, reverse variable lane switching and intersection signal scheme optimization according to the congestion situation.

3. The urban traffic congestion management method based on a universal large language model as claimed in claim 2 is characterized in that: Receiving and parsing the user's question content, dynamically selecting and activating the underlying expert agent most relevant to the question content for the first round of interaction includes: Receiving question content input by the user through the user interface; Use the pre-built large language model to parse the user's questions, extract keywords and semantic information, and determine the user's needs; According to the analyzed user needs, the underlying multi-expert agent most relevant to the question content is dynamically selected through predefined rules; Send activation instructions to the selected underlying expert agent to activate the selected underlying multi-expert agent for the first round of interaction; The predefined rules include: If the question involves congestion identification, the congestion detection expert Agent is activated; If the question is about the cause of congestion, the congestion cause analysis expert Agent will be activated; If congestion management suggestions are needed, the congestion suggestion expert agent is activated; If it involves the implementation of the congestion management plan, the congestion management expert Agent will be activated.

4. The urban traffic congestion management method based on a universal large language model according to any one of claims 1 to 3, characterized in that: If the results of the first round of interaction do not meet the user's needs, then based on the user's needs and the results of the first round of interaction, continue to select other underlying expert agents for at least one round of interaction, gradually approaching until a solution that meets the user's needs is found, including: If the results of the first round of interaction do not meet the user's needs, identify the information points that need to be supplemented based on the results of the first round of interaction and the user's needs; Select a bottom-level expert agent that is most relevant to the information point to be supplemented for the next round of interaction; Deliver user requirements and the results of the previous round of interactions to the underlying expert agent selected in the current round, and receive and process the feedback information from the underlying expert agent; According to the feedback information of the bottom-level expert agent selected in the current round, it is judged again whether the current interaction result meets the user's needs; If the current interaction result meets the user's needs, the result summary phase will begin; If the current interaction result does not meet the user's needs, other relevant expert agents are selected to continue the next round of interaction. In each round of interaction, the selection strategy for the underlying expert agent is dynamically adjusted according to the feedback information of the previous round, so as to gradually approach until a solution that meets the user's needs is found.

5. A framework for managing urban traffic congestion based on a general large language model, characterized in that: include: A collaborative coordination Agent, used to execute the method according to any one of claims 1 to 4; At least one underlying multi-expert Agent is used to select corresponding algorithms according to data sources of different modes to implement urban traffic congestion control measures including congestion detection, congestion cause analysis, congestion suggestions and control plan formulation.

6. The urban traffic congestion management framework based on a general large language model as claimed in claim 5 is characterized in that: At least one underlying expert agent includes: The congestion detection expert role Agent is used to identify and analyze the real-time signal scheme of the corresponding intersection based on the acquired checkpoint passing data, output the intersection timing signal scheme information, and determine whether the target section is congested based on the intersection timing signal scheme information, and output the congestion detection result of the target section with time information, congestion degree quantitative label and congestion frequency quantitative label; Congestion cause analysis expert role Agent, used to analyze the causes of congestion including supply and demand matching, section interference, traffic organization, event detection, road network and scheme, and time and space coordination based on congestion detection results; A congestion suggestion expert Agent, used to provide at least one congestion suggestion according to the cause of congestion; The congestion management expert Agent is used to provide a management solution including variable lane switching, reverse variable lane switching and intersection signal solution optimization according to at least one congestion suggestion.

7. The urban traffic congestion management framework based on a general large language model as claimed in claim 6 is characterized in that: Congestion detection expert role Agent includes: The intersection timing signal scheme information detection sub-agent is used to automatically obtain and process the all-day checkpoint vehicle passing data, generate a vehicle passing data set based on the vehicle flow direction classification standard through time division and vehicle flow direction aggregation, use autocorrelation analysis and ordered clustering to identify the signal cycle and divide the control period, and use Fourier series to perform curve fitting to identify the maximum point, combine the time data corresponding to each phase switching point of the fitting curve to compare and analyze, determine the phase sequence and phase duration of each signal phase, and then integrate and output the intersection timing signal scheme information including cycle, period, phase, phase sequence and duration; The congestion detection result output sub-agent is used to perform time correction on the license plate capture data from the electronic police at the checkpoint based on the intersection timing signal plan information, and then perform correction after interference removal, clustering, and introduction of at least one correction factor related to the travel speed, to obtain the real-time travel speed of the section without being affected by the red light, and combine the obtained section length information to determine whether the corresponding section is congested within a given time period, and output the congestion detection result of the target section with time information, congestion degree quantitative label, and congestion frequency quantitative label.

8. The urban traffic congestion management framework based on a general large language model as claimed in claim 6, characterized in that: Congestion cause analysis expert role Agent includes: The supply-demand matching index sub-agent is used to analyze the peak traffic of different road levels based on the congestion detection results, and output the road section traffic demand results according to the traffic sorting and the preset traffic threshold, and calculate the peak hour coefficient according to the maximum traffic volume in the selected peak period, evaluate the duration of the peak based on the peak hour coefficient, and identify the peak travel distribution results; The cross-section interference sub-agent is used to collect and analyze the vehicle trajectory data of the target section, calculate the difference in the total amount of vehicle data of the upstream and downstream sections, and determine the interference of the attraction point; determine the interference of the section opening on the traffic flow according to the road grade of the section and the upstream and downstream intersection information of the adjacent sections; use the GPS location data of the bus station to count the number and frequency of buses arriving at the station within the set time period to determine the bus line and bus gathering interference; and comprehensively evaluate and output the interference of bus stops and bus lines on the traffic path according to the road grade of the section, the upstream and downstream intersection information of the adjacent sections, the number of bus lines and the location of the bus stations; Traffic organization detection sub-agent is used to obtain and analyze the forward flow and reverse flow data of the target section within a preset time period, calculate the relative flow difference and tidal degree, so as to identify whether there is a tidal phenomenon in the import flow direction of the target point section, and output the description information of the import flow direction tidal phenomenon; according to the changes in the traffic volume in different directions of the section, determine whether there is a tidal phenomenon in both directions of the target section, and output the description information of the tidal phenomenon in both directions of the section; and by comparing the number of upstream import lanes and downstream export lanes, output the matching information of the upstream import and downstream export traffic capacity of the target section; The event detection sub-agent is used to determine whether a traffic safety incident occurs on the target road section based on the obtained illegal accident information, and output the event description information of the target road section with time and location; A road network and solution query sub-agent, used to query and provide road network and signal solution description information including intersection model, intersection channelization, neighbor relationship and real-time signal solution from at least one data source; The spatiotemporal coordination sub-agent is used to determine the speed of the upstream and downstream of the target section, the length of the target section, the average speed during the congestion period of the upstream of the target section, the average speed during the congestion period of the downstream of the target section, and to obtain the optimal vehicle trajectory data by marking the path coordination direction, completing the missing trajectory data and connecting the trajectories of the acquired vehicle passing data; then, combining the intersection equipment type and the non-stop travel time, analyzing the parking situation of each intersection, calculating the path parking rate, and comparing the parking rate of the current section with that of the adjacent sections; when the speed of the upstream and downstream of the target section, the length of the target section, the average speed during the congestion period of the upstream of the target section, the average speed during the congestion period of the downstream of the target section, and the parking rate of the current section with that of the adjacent sections meet the set incoherence conditions, outputting the conclusion that the upstream and downstream sections of the target section are spatiotemporally incoherent; and, calculating the speed variation coefficient of multiple adjacent cycles of the target section by the ratio of the speed standard deviation to the speed average value, and if the variation coefficient meets the set unevenness conditions, outputting the conclusion that the target section is uneven.

9. The urban traffic congestion management framework based on a general large language model as claimed in claim 8, characterized in that: The congestion suggestion expert agent returns the results after executing the corresponding processes of each sub-agent in the congestion cause analysis expert role agent. The output suggestions include: If the demand indicated by the traffic demand result of a road section is greater than the set demand threshold, a suggestion is outputted to set up an induction display screen upstream of the target road section for traffic diversion; For peak travel concentration indicated by the peak travel distribution results, output suggestions for staggered travel or use of public transportation; For attraction point interference, the output includes setting up an induction display screen upstream of the target road section to induce non-arrival traffic flow to choose an alternative path, the recommended location and number of openings in the attraction point area, and suggestions for concentrating the suction flow of the inducing road section to a fixed area; For the disruption of road openings to traffic flow, the output includes suggestions for installing hard barriers, adding entry lanes, re-channeling lanes, marking waiting areas, and setting up green wave belts between adjacent intersections; For bus route and bus cluster interference, the output includes suggestions for adjusting bus routes and arrival times; For the interference of bus stops and bus lines on traffic routes, output suggestions for moving the location of bus stops upstream; For the presence of tidal phenomena in the import flow direction description information, the output increases and switches to variable lanes, adjusts the intersection signal cycle duration, and reallocates the green light duration for each flow direction; For the presence of tidal phenomena indicated by the bidirectional tidal phenomenon description information of the road section, the output includes suggestions for adding and switching tidal lanes, adjusting the signal cycle duration of the intersection, and reallocating the green light duration of each flow direction; If the matching information of the traffic capacity of the upstream entrance and the downstream exit of the target road section indicates a mismatch, a suggestion is output to increase the number of lanes merging into the widening section of the exit road; For the conclusion that the upstream and downstream sections of the target section are not spatiotemporally coherent, the output includes suggestions for strengthening queue length monitoring, optimizing signal linkage release, setting up motor vehicle and non-motor vehicle isolation belts in the middle of adjacent intersection sections, and adjusting the number of upstream and downstream lanes to be consistent; The conclusion of the traffic irregularity of the target road section is output through suggestions including optimizing queue lanes, monitoring queue length data, setting common cycles at adjacent intersections, and setting green wave belts between adjacent intersections; For illegal parking events indicated by the event description information, the output includes suggestions for automatically identifying and setting up smart parking spaces using video technology; For accident events indicated by the event description information, the output includes remote video evaluation, electronic document submission for handling minor accidents that meet the set conditions, and recommendations for establishing an efficient linkage mechanism for non-minor accidents.

10. The urban traffic congestion management framework based on a general large language model as claimed in claim 9, characterized in that: Congestion management expert Agent includes: The variable lane switching sub-agent is used to determine the positive deviation of the left-turn flow and the straight-through flow at the same entrance and the reverse deviation after switching a lane based on the intersection model provided by the traffic organization detection sub-agent and the lane-level flow data produced by the supply-demand matching index sub-agent. When the positive deviation and the reverse deviation meet the preset variable lane switching conditions, the recommended lane flow direction is output and automatically sent to the signal control system connected by the communication; The tidal lane switching sub-agent is used to determine the positive deviation of the flow in different directions in both directions of the road section and the reverse deviation after switching a lane to the target lane based on the intersection model provided by the traffic organization detection sub-agent and the lane-level flow data produced by the supply-demand matching index sub-agent. When the positive deviation and the reverse deviation meet the preset tidal lane switching conditions, the recommended switching section is output and automatically sent to the signal control system connected by the communication; The intersection signal scheme optimization sub-agent is used to analyze the congestion status of all entrance sections according to the intersection model and real-time signal scheme provided by the traffic organization detection sub-agent and the lane-level flow data produced by the supply-demand matching index sub-agent, and determine whether the intersection is congested. When the intersection is congested, according to the phase-level phase sequence S1, signal cycle and green wave coordination path status in the real-time signal scheme, the cycle length and the green light duration of the phase-level phase sequence S1 are adjusted to obtain the optimized signal scheme, and automatically send it to the signal control system connected by communication; wherein, when optimizing the signal scheme, if the current signal cycle does not reach the maximum signal cycle and the intersection is not on the green wave coordination path, the cycle length is increased and the green light duration of the phase-level phase sequence S1 is gradually increased; when optimizing the signal scheme, if the current signal cycle does not reach the maximum signal cycle but the intersection is on the green wave coordination path, the cycle length is kept unchanged, the green light duration of the phase-level phase sequence S1 is gradually increased, and the green light duration of the non-phase sequence phase S1 is reduced.

Images