City intelligent collaborative decision-making system and method based on large model

Abstract

The invention relates to the technical field of large models, in particular to an urban intelligent collaborative decision-making system and method based on a large model, and the system comprises a multi-source heterogeneous data fusion perception layer, a deep learning cognitive engine, a self-adaptive regulation and control decision-making layer and a system efficiency optimization and safety guarantee system. The method has the advantages that multi-modal data of a traffic camera, a geomagnetic sensor, a vehicle-mounted terminal and the like are integrated, space-time correlation characteristics are extracted by using a multi-layer attention mechanism of a large model, and traffic flow prediction and signal timing optimization linkage regulation and control are realized in combination with an incremental learning algorithm. Compared with the prior art, the platform solves the problems that a traditional method is poor in dynamic adaptability to a complex road network and low in prediction precision, the passing efficiency of the urban road network can be improved by 15%-22%, and the abnormal congestion response time is shortened to be within 3 minutes.

Description

Technical Field

[0001] The present invention relates to the field of large model technology, and specifically to a large model-based urban intelligent collaborative decision-making system and method. Background Art

[0002] With the acceleration of urbanization, traffic congestion has become a core issue affecting urban operational efficiency. Predicting and controlling urban traffic flow is crucial for improving road utilization efficiency and reducing traffic congestion. However, existing traffic flow prediction and control methods have the following problems:

[0003] Complex data processing: Urban traffic data comes from various sources, including sensors and surveillance cameras. The data volume is large and complex. Traditional methods cannot efficiently process and integrate these multi-source data.

[0004] Insufficient prediction accuracy: Existing methods often fail to fully utilize the spatiotemporal and periodic characteristics of traffic flow data, resulting in inaccurate prediction results that are difficult to meet actual needs.

[0005] Poor real-time control: Traditional control methods are mostly based on fixed rules or simple models, which make it difficult to make dynamic adjustments based on real-time traffic conditions and unable to respond to sudden traffic incidents in a timely manner.

[0006] Therefore, there is an urgent need for a large-scale model-based intelligent collaborative decision-making system for cities that leverages advanced technologies such as deep learning and reinforcement learning to accurately predict and control traffic flows in real time. Such a platform can effectively integrate multi-source data, capture the complex spatiotemporal characteristics of traffic flows, and dynamically optimize traffic signal timing and road resource allocation through intelligent algorithms, thereby improving the overall operational efficiency of urban transportation. Summary of the Invention

[0007] The purpose of the present invention is to provide a large-scale model-based urban intelligent collaborative decision-making system and method to solve the problems raised in the above-mentioned background technology.

[0008] To achieve the above objectives, the present invention provides the following technical solutions: a large-scale model-based urban intelligent collaborative decision-making system, comprising a multi-source heterogeneous data fusion perception layer, a deep learning cognitive engine, an adaptive control decision-making layer, and a system performance optimization and security assurance system; wherein:

[0009] The multi-source heterogeneous data fusion perception layer includes a spatiotemporal graph neural network perception network, a traffic event structuring engine, and a traffic digital twin base. The spatiotemporal graph neural network perception network deploys a distributed sensor cluster to collect multi-source data in real time, designs edge computing nodes to pre-process data and construct a spatiotemporal graph structure, and adopts a millisecond-level data synchronization mechanism. The traffic event structuring engine develops a deep learning-based target detection algorithm, builds an event classification system, designs an event correlation algorithm, and establishes an event priority assessment model. The traffic digital twin base uses a spatiotemporal database to store traffic event quadruple data, constructs a road network knowledge graph, and designs a data verification pipeline.

[0010] The deep learning cognitive engine includes a Transformer-based traffic flow prediction model, a reinforcement learning optimizer, and a traffic situation assessment system. The Transformer-based traffic flow prediction model constructs a multi-scale feature extraction network, designs an attention mechanism enhancement module, develops a model adaptive adjustment strategy, and integrates an external factor influencing model. The reinforcement learning optimizer constructs a reward function, designs a multi-agent reinforcement learning framework, develops an online learning mechanism, and implements an exploration-exploitation balance strategy. The traffic situation assessment system establishes a congestion index calculation model, designs a traffic flow stability assessment algorithm, develops a propagation path prediction module, and builds a historical pattern matching engine.

[0011] The adaptive control decision-making layer includes a traffic simulation sandbox, a signal timing optimization system, and a traffic event knowledge graph. The traffic simulation sandbox constructs a high-precision traffic flow simulator, designs a virtual experimental platform, develops a solution comparison and analysis tool, and implements a dynamic parameter adjustment interface. The signal timing optimization system establishes a configurable signal control template library, develops a phase jump prediction algorithm, designs a green wave band dynamic adjustment mechanism, and builds an emergency vehicle priority passage module. The traffic event knowledge graph constructs an event type system, designs an event impact propagation model, develops a collaborative disposal recommendation engine, and implements historical case association queries.

[0012] The system performance optimization and security assurance system includes high-performance computing architecture, fine-grained permission management, and full-link monitoring and traceability; the high-performance computing architecture adopts a microservice architecture, deploys a GPU computing cluster, designs a caching strategy, and builds a load balancing system; fine-grained permission management implements a multi-dimensional permission control system, designs a role-based access control model, develops a permission audit system, and builds a data desensitization mechanism; full-link monitoring and traceability builds an operation log system, develops a performance monitoring module, designs an anomaly detection algorithm, and implements a version backtracking function.

[0013] Preferably, in the spatiotemporal graph neural network perception network, the constructed spatiotemporal graph structure abstracts the road network into a graph model, where nodes represent intersections, edges represent road sections, attributes include flow and speed dynamic parameters, and the data update delay of all perception nodes is controlled within 500ms.

[0014] Preferably, in the Transformer-based traffic flow prediction model, a multi-scale feature extraction network is constructed to capture the periodicity, trend and burstiness of traffic flow respectively; the designed attention mechanism enhancement module automatically focuses on the time period and spatial area that are most valuable for prediction; the developed model adaptive adjustment strategy dynamically adjusts the prediction window size and confidence interval according to the real-time data quality; the integrated external factor influencing model includes meteorological and special event external factors to improve the robustness of the prediction.

[0015] Preferably, in the signal timing optimization system, the established configurable signal control template library supports parameterized adjustment according to intersection type and traffic scenario; the developed phase jump prediction algorithm prepares signal phase switching in advance according to real-time traffic flow status, reducing vehicle waiting time; the designed green wave band dynamic adjustment mechanism optimizes the green wave band speed and coverage according to real-time traffic demand; the constructed emergency vehicle priority passage module opens a green channel for emergency and fire-fighting special vehicles by adjusting the signal phase in real time.

[0016] Preferably, actual tests of the system show that it can complete comprehensive situational awareness of the city's main road network within 1 minute, with an accuracy rate of more than 90% in short-term traffic flow prediction. After signal control optimization, the average traffic efficiency is increased by 20%-30%, which is suitable for the intelligent traffic upgrade needs of traffic management departments and intelligent transportation system integrators in large and medium-sized cities.

[0017] A method for a large-scale model-based urban intelligent collaborative decision-making system includes the following steps:

[0018] Multi-source heterogeneous data fusion perception: Deploy distributed sensor clusters to collect multi-source data from road cameras, mobile phone signals, and bus IC cards in real time; use edge computing nodes to preprocess raw data, filter noise data, and extract basic feature vectors; construct a spatiotemporal graph structure, abstract the road network into a graph model, with nodes representing intersections and edges representing road sections, and assign dynamic parameters such as node flow and speed to nodes and edges; adopt a millisecond-level data synchronization mechanism to ensure that the data update delay of all sensing nodes is within 500ms; develop a deep learning-based target detection algorithm to identify vehicles, pedestrians, and traffic signs in video streams in real time; build a hierarchical event classification system, design an event association algorithm to identify event causal relationships and propagation paths, and establish an event priority assessment model to dynamically adjust the processing order; select a spatiotemporal database to store traffic event quadruple, build a road network knowledge graph, and design a data verification pipeline to detect data integrity, consistency, and outliers;

[0019] Deep Learning Cognitive Analysis: Build a Transformer-based traffic flow prediction model, including a multi-scale feature extraction network and an attention mechanism enhancement module. Develop a model adaptive adjustment strategy and integrate external factor influencing models. Construct a reward function, design a multi-agent reinforcement learning framework, develop an online learning mechanism, and implement an exploration-exploitation balance strategy. Establish a congestion index calculation model, design a traffic flow stability assessment algorithm, develop a propagation path prediction module, and build a historical pattern matching engine.

[0020] Adaptive control and decision-making: Build a high-precision traffic flow simulator, design a virtual experimental platform, develop a solution comparison and analysis tool, and implement a dynamic parameter adjustment interface; establish a configurable signal control template library, develop a phase jump prediction algorithm, design a green wave band dynamic adjustment mechanism, and build an emergency vehicle priority passage module; build an event type system, design an event impact propagation model, develop a collaborative disposal recommendation engine, and implement historical case association query;

[0021] System performance optimization and security assurance: Adopt a microservice architecture, deploy a GPU computing cluster, design a caching strategy, and build a load balancing system; implement a multi-dimensional permission control system, design a role-based access control model, develop a permission audit system, and build a data desensitization mechanism; build an operation log system, develop a performance monitoring module, design an anomaly detection algorithm, and implement a version backtracking function.

[0022] Preferably, in the multi-source heterogeneous data fusion perception step, the data verification pipeline sequentially performs integrity verification to ensure key field coverage; consistency verification to eliminate time and space contradictions; and outlier detection to filter sensor false alarms.

[0023] Preferably, in the deep learning cognitive analysis step, the multi-scale feature extraction network of the Transformer-based traffic flow prediction model captures the periodicity, trend and burstiness of traffic flow respectively; the attention mechanism enhancement module automatically focuses on the time period and spatial area that are most valuable for prediction; the model adaptive adjustment strategy dynamically adjusts the prediction window size and confidence interval according to the real-time data quality; the external factor impact model integrates meteorological and special event external factors.

[0024] Preferably, in the adaptive regulation and decision-making step, the configurable signal control template library of the signal timing optimization system supports parameterized adjustment according to the intersection type and traffic scenario; the phase jump prediction algorithm prepares signal phase switching in advance according to the real-time traffic flow status; the green wave band dynamic adjustment mechanism optimizes the green wave band speed and coverage according to real-time traffic demand; the emergency vehicle priority passage module opens a green channel for emergency and fire-fighting special vehicles by adjusting the signal phase in real time.

[0025] Preferably, actual tests of the system constructed using this method show that it can complete comprehensive situational awareness of the city's main road network within 1 minute, the accuracy of short-term traffic flow prediction is over 90%, and the average traffic efficiency is improved by 20%-30% after signal control optimization. It is suitable for the intelligent traffic upgrade needs of traffic management departments and intelligent transportation system integrators in large and medium-sized cities.

[0026] Compared with the prior art, the present invention has the following beneficial effects:

[0027] The urban intelligent collaborative decision-making system and method based on a large model proposed in this invention realizes the real-time integration and analysis of more than 20 types of heterogeneous traffic data such as urban cameras, geomagnetic sensors, and GPS floating vehicles through a multimodal data fusion engine and a spatiotemporal feature extraction algorithm. The data preprocessing efficiency is 5-8 times higher than that of traditional methods, effectively solving the problem of traffic data fragmentation. Based on the spatiotemporal sequence prediction architecture of the large model, it achieves a breakthrough in achieving 15-minute-level accurate prediction of traffic flow, with a prediction accuracy of more than 97%, which is 15%-20% lower than the error rate of traditional models. Combined with the dynamic game optimization algorithm, it can generate global traffic light timing in real time. The scheme and path guidance strategy have increased the traffic efficiency in key areas by more than 30%; through the adaptive optimization mechanism and scenario parameter configuration system, it supports differentiated control needs from megacities to small and medium-sized towns, and the scenario adaptation efficiency is improved by 60%; the construction of data privacy computing modules and multi-level security protection systems achieves 100% desensitization of sensitive location information, meeting the requirements of regulations such as the "Cybersecurity Law"; in actual applications, the platform has reduced the congestion index of urban main roads during peak hours by 40%, increased the emergency response speed by 4 times, and reduced the average carbon emissions of vehicles by 25%, providing a full-stack traffic management solution for smart city construction. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 This is a system architecture diagram of the present invention. DETAILED DESCRIPTION

[0029] In order to clearly and completely describe the objectives and technical solutions of the present invention and make the advantages more clearly understood, the embodiments of the present invention are further described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are part of the embodiments of the present invention, not all of them, and are only used to explain the embodiments of the present invention, not to limit the embodiments of the present invention. All other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0030] For example 1, please refer to Figure 1This invention provides a technical solution: a large-scale model-based urban intelligent collaborative decision-making system. By building a three-layer "perception-cognition-decision-making" technical architecture, it achieves closed-loop management of the entire process, from traffic data collection to intelligent regulation. The specific technical solution includes the following core modules and implementation process:

[0031] 1. Multi-source heterogeneous data fusion perception layer

[0032] Spatiotemporal graph neural network perception network: Deploy distributed sensor clusters to support real-time collection of multi-source data such as road cameras, mobile phone signals, and bus IC cards; design edge computing nodes to preprocess raw data, filter out noise data, and extract basic feature vectors; construct a spatiotemporal graph structure to abstract the road network into a graph model, where nodes represent intersections, edges represent road sections, and attributes include dynamic parameters such as flow and speed; use a millisecond-level data synchronization mechanism to ensure that the data update delay of all perception nodes is controlled within 500ms.

[0033] Traffic event structuring engine: Develop a deep learning-based target detection algorithm to identify vehicles, pedestrians, and traffic signs in real time in video streams; build an event classification system using a hierarchical labeling strategy (first-level classifications such as "congestion" and "accident," and second-level classifications such as "minor scratches" and "multi-vehicle chain collisions"); design an event association algorithm to identify the causal relationships and propagation paths between events through co-occurrence analysis and spatiotemporal proximity analysis; establish an event priority assessment model to dynamically adjust the processing order based on factors such as event type, impact range, and duration.

[0034] Traffic digital twin foundation: A spatiotemporal database is used to store traffic event quadruple data (timestamp, location ID, event type, and severity), supporting sub-second spatiotemporal joint queries. A road network knowledge graph is constructed, with nodes containing intersection attributes (such as number of lanes and signal cycle) and edges containing section attributes (such as speed limit and lane direction). A data verification pipeline is designed, including integrity verification (ensuring coverage of key fields), consistency verification (eliminating spatiotemporal inconsistencies), and outlier detection (filtering sensor false alarms).

[0035] 2. Deep Learning Cognitive Engine

[0036] Transformer-based traffic flow prediction model: Construct a multi-scale feature extraction network to capture the periodicity (daily / weekly patterns), trend (long-term changes) and suddenness (impact of unexpected events) of traffic flow; design an attention mechanism enhancement module to automatically focus on the time periods and spatial regions that are most valuable for prediction; develop a model adaptive adjustment strategy to dynamically adjust the prediction window size and confidence interval based on the quality of real-time data; integrate the impact model of external factors such as weather and special events (such as concerts and exhibitions) to improve prediction robustness.

[0037] Reinforcement Learning Optimizer: Constructs a reward function that comprehensively considers multi-dimensional indicators such as traffic efficiency, energy consumption, and passenger comfort; designs a multi-agent reinforcement learning framework, with each intersection signal light acting as an independent agent, achieving collaborative optimization through shared global rewards; develops an online learning mechanism that enables the model to adapt in real time to the long-term evolution of traffic patterns and the short-term impact of emergencies; implements an exploration-exploitation balance strategy to continuously explore better signal timing solutions while ensuring the current control effect.

[0038] Traffic situation assessment system

[0039] Establish a congestion index calculation model that comprehensively considers multi-dimensional indicators such as vehicle speed distribution, queue length, and travel time; design a traffic flow stability assessment algorithm to identify potential congestion risks by analyzing the flow-density relationship curve; develop a propagation path prediction module to predict the direction of congestion spread and the scope of impact based on the current congestion situation; build a historical pattern matching engine to compare the current traffic situation with similar historical scenarios and provide reference control solutions.

[0040] 3. Adaptive Control Decision-Making Layer

[0041] Traffic simulation sandbox: Build a high-precision traffic flow simulator that supports second-level simulation of the effects of different traffic control strategies; design a virtual experimental platform that allows testing of new control algorithms without affecting actual traffic; develop solution comparison and analysis tools to quantify the impact of different control strategies on indicators such as traffic efficiency and energy consumption; implement a dynamic parameter adjustment interface to optimize actual control parameters in real time based on simulation results.

[0042] Signal timing optimization system: Establish a configurable signal control template library to support parameterized adjustments based on intersection type (such as main road intersections and branch road confluences) and traffic scenarios (such as morning rush hour and special events); develop a phase jump prediction algorithm to prepare signal phase switching in advance based on real-time traffic flow status to reduce vehicle waiting time; design a dynamic green wave adjustment mechanism to optimize the green wave speed and coverage based on real-time traffic demand; build an emergency vehicle priority passage module to open a green channel for special vehicles such as emergency and fire vehicles by adjusting signal phases in real time.

[0043] Traffic incident knowledge graph: Build an event type system covering various factors affecting traffic, such as traffic accidents, road construction, and severe weather; design an event impact propagation model to predict the short-term and long-term impact of events on the surrounding road network; develop a collaborative disposal recommendation engine to recommend the optimal disposal plan based on event type and impact scope; implement historical case correlation query to provide reference experience for current event disposal.

[0044] IV. System Performance Optimization and Security Assurance System

[0045] High-performance computing architecture: Adopting a microservices architecture, core functions such as data collection, analysis, prediction, and regulation are modularized to support on-demand expansion; deploying GPU computing clusters to accelerate deep learning model inference and reinforcement learning training; designing caching strategies to locally store high-frequency query results and commonly used model parameters to reduce computing latency; and building a load balancing system to dynamically allocate computing resources based on real-time computing needs.

[0046] Fine-grained permission management: Implement a multi-dimensional permission control system, including data access rights, function operation rights, parameter adjustment rights, etc.; design a role-based access control model to assign differentiated permissions to different user roles (such as ordinary traffic police, traffic engineers, and system administrators); develop a permission audit system to record all sensitive operations and conduct compliance checks; build a data desensitization mechanism to automatically process sensitive information and ensure secure data sharing.

[0047] Full-link monitoring and traceability: Build an operation log system to record the entire process from data collection to control execution. Develop a performance monitoring module to monitor the operating status and response time of each system component in real time. Design an anomaly detection algorithm to automatically identify abnormal patterns in system operation and issue alerts. Implement a version rollback function to support the recovery of system status and data at any historical moment.

[0048] Through this modular design, this technical solution overcomes the three major technical challenges of real-time perception, accurate prediction, and intelligent control of urban traffic flows. Actual testing has demonstrated that the system can achieve comprehensive situational awareness of a city's main road network within one minute, with short-term traffic flow prediction accuracy exceeding 90%. Optimized signal control improves average traffic efficiency by 20%-30%. This provides a next-generation, self-learning, and adaptive solution for intelligent traffic management, suitable for the intelligent transportation upgrade needs of traffic management departments in large and medium-sized cities, intelligent transportation system integrators, and other scenarios.

[0049] Example 2, based on Example 1, proposes a method for a large-scale model-based urban intelligent collaborative decision-making system, including the following steps:

[0050] Multi-source heterogeneous data fusion perception: Deploy distributed sensor clusters to collect multi-source data from road cameras, mobile phone signals, and bus IC cards in real time; use edge computing nodes to preprocess raw data, filter out noisy data, and extract basic feature vectors; construct a spatiotemporal graph structure, abstract the road network into a graph model, with nodes representing intersections and edges representing road sections, and assign dynamic parameters such as traffic flow and speed to nodes and edges; adopt a millisecond-level data synchronization mechanism to ensure that the data update delay of all sensing nodes is within 500ms; develop a deep learning-based target detection algorithm to identify vehicles, pedestrians, and traffic signs in video streams in real time; build a hierarchical event classification system, design an event association algorithm to identify event causal relationships and propagation paths, and establish an event priority assessment model to dynamically adjust the processing order; select a spatiotemporal database to store traffic event quadruple, build a road network knowledge graph, and design a data verification pipeline to perform data integrity, consistency, and outlier detection; the data verification pipeline sequentially performs integrity verification to ensure coverage of key fields, consistency verification to eliminate spatiotemporal inconsistencies, and outlier detection to filter sensor errors.

[0051] Deep learning cognitive analysis: Construct a Transformer-based traffic flow prediction model, including a multi-scale feature extraction network and an attention mechanism enhancement module, develop a model adaptive adjustment strategy, and integrate the external factor influencing model; construct a reward function, design a multi-agent reinforcement learning framework, develop an online learning mechanism, and implement an exploration-utilization balance strategy; establish a congestion index calculation model, design a traffic flow stability evaluation algorithm, develop a propagation path prediction module, and build a historical pattern matching engine; the multi-scale feature extraction network of the Transformer-based traffic flow prediction model captures the periodicity, trend, and suddenness of traffic flow respectively; the attention mechanism enhancement module automatically focuses on the time period and spatial area that are most valuable for prediction; the model adaptive adjustment strategy dynamically adjusts the prediction window size and confidence interval according to the real-time data quality; the external factor influencing model integrates meteorological and special event external factors.

[0052] Adaptive control and decision-making: Build a high-precision traffic flow simulator, design a virtual experimental platform, develop a solution comparison and analysis tool, and implement a dynamic parameter adjustment interface; establish a configurable signal control template library, develop a phase jump prediction algorithm, design a green wave band dynamic adjustment mechanism, and build an emergency vehicle priority passage module; build an event type system, design an event impact propagation model, develop a collaborative disposal recommendation engine, and implement historical case association query; the configurable signal control template library of the signal timing optimization system supports parameterized adjustment according to intersection type and traffic scenario; the phase jump prediction algorithm prepares signal phase switching in advance according to real-time traffic flow status; the green wave band dynamic adjustment mechanism optimizes the green wave band speed and coverage according to real-time traffic demand; the emergency vehicle priority passage module opens a green channel for emergency and fire vehicles by adjusting the signal phase in real time.

[0053] System performance optimization and security assurance: Adopt a microservice architecture, deploy a GPU computing cluster, design a caching strategy, and build a load balancing system; implement a multi-dimensional permission control system, design a role-based access control model, develop a permission audit system, and build a data desensitization mechanism; build an operation log system, develop a performance monitoring module, design an anomaly detection algorithm, and implement a version backtracking function; actual tests of the system built using this method have shown that it can complete comprehensive situational awareness of the city's main road network within 1 minute, with an accuracy rate of over 90% for short-term traffic flow prediction. After signal control optimization, the average traffic efficiency is increased by 20%-30%, which is suitable for the intelligent transportation upgrade needs of traffic management departments in large and medium-sized cities and intelligent transportation system integrators.

[0054] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A large-scale model-based urban intelligent collaborative decision-making system, characterized by: It includes a multi-source heterogeneous data fusion perception layer, a deep learning cognitive engine, an adaptive control decision layer, and a system performance optimization and security assurance system; among which: The multi-source heterogeneous data fusion perception layer includes a spatiotemporal graph neural network perception network, a traffic event structuring engine, and a traffic digital twin base. The spatiotemporal graph neural network perception network deploys a distributed sensor cluster to collect multi-source data in real time, designs edge computing nodes to pre-process data and construct a spatiotemporal graph structure, and adopts a millisecond-level data synchronization mechanism. The traffic event structuring engine develops a deep learning-based target detection algorithm, builds an event classification system, designs an event correlation algorithm, and establishes an event priority assessment model. The traffic digital twin base uses a spatiotemporal database to store traffic event quadruple data, constructs a road network knowledge graph, and designs a data verification pipeline. The deep learning cognitive engine includes a Transformer-based traffic flow prediction model, a reinforcement learning optimizer, and a traffic situation assessment system. The Transformer-based traffic flow prediction model constructs a multi-scale feature extraction network, designs an attention mechanism enhancement module, develops a model adaptive adjustment strategy, and integrates an external factor influencing model. The reinforcement learning optimizer constructs a reward function, designs a multi-agent reinforcement learning framework, develops an online learning mechanism, and implements an exploration-exploitation balance strategy. The traffic situation assessment system establishes a congestion index calculation model, designs a traffic flow stability assessment algorithm, develops a propagation path prediction module, and builds a historical pattern matching engine. The adaptive control decision-making layer includes a traffic simulation sandbox, a signal timing optimization system, and a traffic event knowledge graph. The traffic simulation sandbox constructs a high-precision traffic flow simulator, designs a virtual experimental platform, develops a solution comparison and analysis tool, and implements a dynamic parameter adjustment interface. The signal timing optimization system establishes a configurable signal control template library, develops a phase jump prediction algorithm, designs a green wave band dynamic adjustment mechanism, and builds an emergency vehicle priority passage module. The traffic event knowledge graph constructs an event type system, designs an event impact propagation model, develops a collaborative disposal recommendation engine, and implements historical case association queries. The system performance optimization and security assurance system includes high-performance computing architecture, fine-grained permission management, and full-link monitoring and traceability; the high-performance computing architecture adopts a microservice architecture, deploys a GPU computing cluster, designs a caching strategy, and builds a load balancing system; fine-grained permission management implements a multi-dimensional permission control system, designs a role-based access control model, develops a permission audit system, and builds a data desensitization mechanism; full-link monitoring and traceability builds an operation log system, develops a performance monitoring module, designs an anomaly detection algorithm, and implements a version backtracking function.

2. The large-scale model-based urban intelligent collaborative decision-making system according to claim 1 is characterized by: In the spatiotemporal graph neural network perception network, the constructed spatiotemporal graph structure abstracts the road network into a graph model, where nodes represent intersections, edges represent road sections, and attributes include dynamic parameters of flow and speed. The data update delay of all perception nodes is controlled within 500ms.

3. The large-scale model-based urban intelligent collaborative decision-making system according to claim 2, characterized in that: In the Transformer-based traffic flow prediction model, a multi-scale feature extraction network is constructed to capture the periodicity, trend and suddenness of traffic flow respectively; the designed attention mechanism enhancement module automatically focuses on the time period and spatial area that are most valuable for prediction; the developed model adaptive adjustment strategy dynamically adjusts the prediction window size and confidence interval according to the real-time data quality; the integrated external factor influencing model includes meteorological and special event external factors to improve the robustness of the prediction.

4. The large-scale model-based urban intelligent collaborative decision-making system according to claim 1 is characterized by: In the signal timing optimization system, the established configurable signal control template library supports parameterized adjustment according to intersection type and traffic scenario; the developed phase jump prediction algorithm prepares signal phase switching in advance according to real-time traffic flow status, reducing vehicle waiting time; the designed green wave band dynamic adjustment mechanism optimizes the green wave band speed and coverage according to real-time traffic demand; the constructed emergency vehicle priority passage module opens a green channel for emergency and fire-fighting special vehicles by adjusting the signal phase in real time.

5. The large-scale model-based urban intelligent collaborative decision-making system according to claim 1 is characterized by: Actual tests of the system have shown that it can complete comprehensive situational awareness of the city's main road network within 1 minute, with an accuracy rate of over 90% in short-term traffic flow prediction. After signal control optimization, the average traffic efficiency is increased by 20%-30%. It is suitable for the intelligent traffic upgrade needs of traffic management departments in large and medium-sized cities and intelligent transportation system integrators.

6. A method for a large-scale model-based urban intelligent collaborative decision-making system according to claim 5, characterized in that: The following steps are involved: Multi-source heterogeneous data fusion perception: Deploy distributed sensor clusters to collect multi-source data such as road cameras, mobile phone signals, and bus IC cards in real time; Use edge computing nodes to preprocess raw data, filter out noise data, and extract basic feature vectors; Construct a spatiotemporal graph structure, abstracting the road network into a graph model, with nodes representing intersections and edges representing road sections, and assigning dynamic parameters such as traffic flow and speed to nodes and edges. A millisecond-level data synchronization mechanism is used to ensure that data update delays for all sensing nodes are within 500ms. A deep learning-based target detection algorithm is developed to perform real-time identification of vehicles, pedestrians, and traffic signs in video streams. A hierarchical event classification system is constructed, and an event correlation algorithm is designed to identify event causal relationships and propagation paths. An event priority assessment model is established to dynamically adjust the processing order. A spatiotemporal database is used to store traffic event quadruple data, a road network knowledge graph is constructed, and a data verification pipeline is designed to check data integrity, consistency, and outlier detection. Deep Learning Cognitive Analysis: Build a Transformer-based traffic flow prediction model, including a multi-scale feature extraction network and an attention mechanism enhancement module. Develop a model adaptive adjustment strategy and integrate external factor influencing models. Construct a reward function, design a multi-agent reinforcement learning framework, develop an online learning mechanism, and implement an exploration-exploitation balance strategy. Establish a congestion index calculation model, design a traffic flow stability assessment algorithm, develop a propagation path prediction module, and build a historical pattern matching engine; Adaptive control and decision-making: Build a high-precision traffic flow simulator, design a virtual experimental platform, develop a solution comparison and analysis tool, and implement a dynamic parameter adjustment interface; establish a configurable signal control template library, develop a phase jump prediction algorithm, design a green wave band dynamic adjustment mechanism, and build an emergency vehicle priority passage module; build an event type system, design an event impact propagation model, develop a collaborative disposal recommendation engine, and implement historical case association query; System performance optimization and security assurance: Adopting microservice architecture, deploying GPU computing clusters, designing caching strategies, and building a load balancing system; Implement a multi-dimensional permission control system, design a role-based access control model, develop a permission audit system, and build a data desensitization mechanism; build an operation log system, develop a performance monitoring module, design anomaly detection algorithm, and implement version backtracking function.

7. A method according to claim 6, characterized in that: In the multi-source heterogeneous data fusion perception step, the data verification pipeline sequentially performs integrity verification to ensure key field coverage and consistency verification to eliminate time and space contradictions; and outlier detection to filter sensor false positives.

8. A method according to claim 6, characterized in that: In the deep learning cognitive analysis step, the multi-scale feature extraction network of the Transformer-based traffic flow prediction model captures the periodicity, trend, and suddenness of traffic flow respectively; The attention mechanism enhancement module automatically focuses on the time period and spatial area that are most valuable for prediction; the model adaptive adjustment strategy dynamically adjusts the prediction window size and confidence interval according to the real-time data quality; the external factor impact model integrates meteorological and special event external factors.

9. A method according to claim 6, characterized in that: In the adaptive control decision-making step, the signal timing optimization system's configurable signal control template library supports parameterized adjustment according to intersection type and traffic scenario; the phase jump prediction algorithm prepares signal phase switching in advance based on real-time traffic flow status; the green wave dynamic adjustment mechanism optimizes the green wave speed and coverage according to real-time traffic demand; and the emergency vehicle priority passage module opens a green channel for emergency and fire vehicles by adjusting the signal phase in real time.

10. A method according to claim 6, characterized in that: Actual tests of the system built using this method have shown that it can complete comprehensive situational awareness of the city's main road network within 1 minute, with an accuracy rate of over 90% in short-term traffic flow prediction. After signal control optimization, the average traffic efficiency is increased by 20%-30%. It is suitable for the intelligent traffic upgrade needs of traffic management departments in large and medium-sized cities and intelligent transportation system integrators.

Images