Urban traffic cloud side-end integrated collaborative management and control method, system and platform

Abstract

The invention relates to an urban traffic cloud side-end integrated collaborative management and control method, system and platform, and the method comprises the steps: obtaining a plurality of intersection sets through clustering; for any intersection set, configuring a traffic signal control strategy for each intersection in each intersection set by adopting conditional judgment logic; in response to a signal control task generated at an intersection, constructing a total optimization objective function, and forming a cloud edge-end collaborative scheduling model in combination with an algorithm; and intelligently distributing a signal control task to the marginal small stations for processing based on the model, and forming global traffic situation awareness by analyzing summarized data from the marginal small stations so as to cooperatively control traffic signal control strategies configured at all intersections. According to the method, through accurate intersection clustering, real-time signal control strategy configuration, efficient cloud side-end cooperative scheduling and global traffic situation awareness, the advantages of edge calculation are efficiently utilized, network delay is remarkably reduced, and remarkable benefit improvement is brought to urban traffic management.

Description

Technical Field

[0001] The present invention relates to the field of traffic management technology, and in particular to a method, system and platform for integrated cloud-edge-terminal collaborative management and control of urban traffic. Background Art

[0002] In the field of traffic management, traffic control solutions based on central servers have been widely adopted. Such solutions usually rely on a variety of sensing devices deployed on the road, such as cameras and various sensors, which can collect traffic data in real time. The collected data is then transmitted to the central control system through the network, which will conduct in-depth analysis of the data and adjust the signal timing of each intersection based on the analysis results to achieve the purpose of optimizing traffic flow.

[0003] However, this traffic control solution based on a central server has gradually exposed its limitations in terms of real-time and flexibility in practical applications. First, due to the large number of data transmission and central server processing links, there is a significant network delay in the entire process from the sensing device to the final execution of the control command. This delay makes it difficult for the system to respond quickly to sudden traffic conditions, thereby increasing the risk of traffic congestion.

[0004] Secondly, the centralized control architecture means that all data processing and computing tasks are concentrated on the central server. As the scale of the transportation network continues to expand, the processing pressure on the central server also increases. In extreme cases, the processing capacity of the server may not be able to meet the needs of real-time processing of large amounts of data, resulting in processing delays or even system crashes, which seriously affects the overall efficiency of urban traffic management.

[0005] In addition, existing traffic control systems often use preset control strategies, and although these strategies can be optimized and adjusted regularly, they are relatively slow to respond to rapid changes in local traffic flow, such as those caused by emergencies or traffic accidents. This limits the system's ability to achieve refined, dynamic, and real-time traffic control.

[0006] Finally, in terms of resource scheduling, existing signal control systems often fail to fully consider the specific needs of each intersection. This may lead to excessive allocation of computing resources at some intersections, while resources at other intersections are idle. This imbalance in resource allocation not only causes a waste of system resources, but may also affect the ultimate effect of traffic control. Summary of the invention

[0007] 1. Technical issues to be resolved

[0008] In view of the above-mentioned shortcomings and deficiencies of the prior art, the present invention provides a method, system and platform for integrated cloud-edge-end collaborative management and control of urban traffic, which solves the technical problems of network delay, excessive load of the central server, slow response to local traffic changes and unbalanced resource scheduling in traffic control solutions based on central servers.

[0009] (II) Technical solution

[0010] In order to achieve the above object, the main technical solutions adopted by the present invention include:

[0011] In a first aspect, an embodiment of the present invention provides a method for integrated collaborative management and control of urban traffic cloud, edge and terminal, including:

[0012] Based on the traffic flow data of at least two intersections obtained, the traffic flow similarity is calculated, and the intersection correlation is obtained by combining the distance between intersections and the travel time, and the intersections are clustered with the goal of minimizing the objective function determined by the intersection correlation to obtain a plurality of intersection sets;

[0013] For any intersection set, according to the change rate of the traffic flow and queue length of each intersection, combined with the correlation index of adjacent intersections determined by the traffic flow similarity, the conditional judgment logic is used to configure the corresponding traffic signal control strategy for each intersection in each intersection set;

[0014] In response to the signal control task generated by at least one intersection in each intersection set, a total optimization objective function is constructed according to the task resource requirements, green wave coordination requirements, and edge station status, and a cloud-edge collaborative scheduling model is formed by combining a hybrid evolutionary algorithm model and / or a heuristic algorithm model;

[0015] Based on the cloud-edge collaborative scheduling model, signal control tasks are intelligently assigned to at least one edge station for processing. By analyzing the aggregated data from multiple edge stations, global traffic situation awareness is formed to coordinate the management and control of traffic signal control strategies configured at all intersections.

[0016] Optionally, based on the traffic flow data of at least two intersections obtained, the traffic flow similarity is calculated, and the intersection correlation is obtained by combining the distance between intersections and the travel time, and the intersections are clustered with the goal of minimizing the objective function determined by the intersection correlation, and the obtained intersection sets include:

[0017] Obtaining traffic flow data of at least two intersections and devices, and calculating traffic flow similarity based on the traffic flow data of at least two intersections and devices;

[0018] Obtain the distance and travel time between each pair of intersections, and combine the traffic flow similarity to obtain the intersection correlation;

[0019] Construct a clustering objective function based on intersection correlation;

[0020] With the goal of minimizing the clustering objective function, the improved K-means algorithm is used to cluster the intersections. During the clustering process, each intersection is iteratively assigned to the nearest cluster center, and the position of the cluster center is updated until the convergence condition is met to form multiple intersection sets.

[0021] in,

[0022] The traffic flow similarity is;

[0023]

[0024] In the formula, F ij is the traffic flow similarity, x i is the traffic volume at intersection i, x j is the traffic flow x at intersection j j ;

[0025] The intersection correlation is:

[0026]

[0027] In the formula, R ij is the intersection correlation, T ij is the average travel time between intersections i and j, D ij is the physical distance between intersections i and j, w1, w2, w3 are weight coefficients;

[0028] The clustering objective function is:

[0029]

[0030] In the formula, X k is the clustered intersection set, k=1,2,…K is the sequence number of the intersection set.

[0031] Optionally, for any intersection set, according to the obtained change rates of the flow and queue length of each intersection, combined with the adjacent intersection relevance index determined by the traffic flow similarity, the corresponding traffic signal control strategy is configured for each intersection in each intersection set using conditional judgment logic, including:

[0032] For any set of intersections, the traffic flow and queue length at the intersection are used to calculate the rate of change of the traffic flow and queue length at the intersection;

[0033] According to the similarity of traffic flow, the correlation index of adjacent intersections is calculated;

[0034] According to the traffic volume of each intersection, the change rate of queue length and the correlation index of adjacent intersections, the conditional judgment logic is used to deploy the traffic signal control strategy for each intersection, including single-point center adaptive control, edge single-point second-level adaptive control, center adaptive green wave control and edge second-level adaptive green wave control initialization;

[0035] A deep Q network model in deep reinforcement learning is introduced to determine the action set selected by the initialized traffic signal control strategy, and the effect of the control strategy is evaluated based on the feedback, and the action reward is obtained by combining multiple factors including the reduction of intersection delay, the reduction of queue length and the number of vehicles.

[0036] In the deep Q-network model, the historical and real-time traffic status is continuously tested and learned, the action sequence that maximizes the cumulative action reward is selected by approximating the Q-value function, and the discount factor is used to balance the short-term and long-term rewards;

[0037] According to the action sequence output by the deep Q network model that maximizes the cumulative reward, the traffic signal control strategy of each intersection is adjusted to obtain the optimal traffic signal control strategy at this stage;

[0038] in,

[0039] The rates of change of flow and queue length are:

[0040]

[0041] Among them, x i is the traffic volume at intersection i, V xi is the rate of change of traffic flow xi, q i is the queue length at intersection i, V qi is the rate of change of the queue length at intersection i, std is the standard deviation, mean is the mean, max is the maximum value, and (t→t+τ) represents the time series from time t to time t+τ;

[0042] The correlation index of adjacent intersections is:

[0043]

[0044] The conditional judgment logic is:

[0045]

[0046] Where C1 is the single-point center adaptive control strategy, C2 is the edge single-point second-level adaptive control strategy, C3 is the center adaptive green wave control strategy, C4 is the edge second-level adaptive green wave control strategy, V x is the change rate threshold of vehicle flow, V q is the queue length change rate threshold, VF is the threshold of the correlation index of adjacent intersections;

[0047] The action set is:

[0048] A = {C1, C2, C3, C4};

[0049] The action rewards are:

[0050]

[0051] In the formula, r t is the action reward at time t, Td k is the average delay time, Mq k is the average queue length, maxMx k and Mx k are the maximum average traffic flow and the actual average traffic flow, respectively, α1, α2, α3 are weight parameters;

[0052] The cumulative action reward is:

[0053]

[0054] Where R is the cumulative action reward, γ is the discount factor, and T is the termination time;

[0055] The Q value function is:

[0056]

[0057] Where s is the current state of the agent, a is the action taken by the agent in state s, E[·] is the expected value of the cumulative sum of all possible future reward sequences after the agent takes action a starting from state s, s0=s represents the initial state, and a0=a represents the initial action.

[0058] Optionally, in response to the signal control task generated by at least one intersection in each intersection set, according to the task resource requirements, green wave coordination requirements and edge station status, a total optimization objective function is constructed, and a cloud-edge-end collaborative scheduling model is formed by combining a hybrid evolutionary algorithm model and / or a heuristic algorithm model, including:

[0059] Monitor and respond to signal control tasks generated by at least one intersection in each intersection set;

[0060] Perform resource demand analysis on the signal control tasks generated at each intersection to determine the resource and bandwidth requirements of the intersection;

[0061] Measure or estimate the edge station's status information including the number of resources, resource upper limit, number of task types, and data transmission delay from each intersection to the station;

[0062] Determine the need for green wave coordination through the results of traffic flow analysis, road network assessment and signal system status survey;

[0063] Based on the task resource requirements, green wave coordination requirements and the status information of edge stations, the overall objective function of the multi-objective optimization problem is constructed with the goal of balancing the load of each edge station, reducing the task types, reducing transmission delay and meeting the green wave coordination requirements. The importance of each objective is adjusted through the benefit coefficient and weight parameter of resource allocation.

[0064] Based on the overall objective function and the improved particle swarm optimization algorithm model, a cloud-edge-device collaborative scheduling model is constructed;

[0065] in,

[0066] The overall objective function is:

[0067]

[0068] Where L is the total objective function, I is the total number of intersections, J is the total number of edge stations, α0 is the profit coefficient of resource allocation, σ i,j is the number of resources allocated to the jth small station at the i-th intersection, α4, α5, α6 are weight parameters, and L R is the load balancing target, L T is the task type target, L D is the transmission delay target;

[0069] The load targets are:

[0070]

[0071] Where Var is the variance;

[0072] The mission types and objectives are:

[0073]

[0074] Where, T j is the number of task types of the j-th station;

[0075] The transmission delay target is:

[0076]

[0077] Where Dij is the delay overhead from the data of the i-th intersection equipment to the j-th small station;

[0078] The constraints in the optimization process are:

[0079]

[0080] In the formula, C jis the resource upper limit of the jth small station, R i For computing resources and bandwidth.

[0081] Optionally, based on the cloud-edge-device collaborative scheduling model, the signal control task is intelligently assigned to at least one edge station for processing, and the aggregated data from multiple edge stations is analyzed to form a global traffic situation awareness to coordinate the control of traffic signals configured at all intersections. The control strategies include:

[0082] A two-dimensional coding matrix is ​​used to represent the task allocation scheme, where the elements of the coding matrix represent the number of resources allocated to the edge station at the intersection;

[0083] Randomly generate an initial particle swarm, each particle represents a possible task allocation scheme;

[0084] Initialize a velocity vector for each particle. The dimension of the velocity vector is the same as the encoding matrix, and the velocity value is randomly initialized within a preset range.

[0085] For each particle in the particle swarm, iteratively update the velocity vector and encoding matrix;

[0086] After each iteration, the optimal fitness value of the current particle swarm is evaluated according to the total objective function and compared with the optimal fitness value of the previous generation to determine whether the termination condition is met;

[0087] If the termination condition is met, the best individual is selected from the current particle swarm as the optimal task allocation scheme, and the signal control tasks generated at each intersection are allocated to at least one edge station for processing based on the optimal task allocation scheme;

[0088] Collect raw traffic data from multiple edge stations, clean the data, unify the format, and synchronize the time to obtain processed traffic data;

[0089] The processed traffic data is classified according to multiple dimensions including intersections, areas and time periods, and the classified data is integrated to form a structured data set;

[0090] Extract basic traffic parameters and dynamic features based on spatiotemporal relationships that can characterize traffic status from structured data sets;

[0091] The extracted basic traffic parameters and dynamic features based on spatiotemporal relationships are fused to form a comprehensive feature set, and the most representative feature vector is extracted from the comprehensive feature set;

[0092] Data mining based on feature vectors is used to form global traffic situation awareness, where traffic situation awareness is multi-dimensional information including traffic flow, congestion conditions and accident risks;

[0093] Based on the results of global traffic situation perception, the variable speed limit on the main line and traffic induction coordinated control are driven by edge stations, and the traffic signal control strategies configured at all intersections are optimized and adjusted to generate a real-time optimization plan that includes the variable speed limit on the main line and the coordinated intelligent control of ramps.

[0094] Optionally, based on the results of global traffic situation awareness, the variable speed limit on the main line and the coordinated control of traffic induction are driven by the edge station to optimize and adjust the traffic signal control strategies configured at all intersections, and generate a real-time optimization solution including the coordinated variable speed limit on the main line and the intelligent control of the ramp, including:

[0095] According to the results of global traffic situation awareness, each main line and ramp is abstracted into multiple control units, and the state information set of each control unit is generated according to the main line number, ramp number and time step;

[0096] According to the traffic signal control strategies, control units and corresponding status information sets currently configured at all intersections, a road traffic efficiency and delay balance model is constructed with road traffic flow and vehicle delay as the joint optimization control objectives;

[0097] Based on the results of global traffic situation awareness, the input parameters of the road traffic efficiency and delay balance model are adjusted and handed over to the edge station for search to find the optimal combination of the main line variable speed limit value and the entrance ramp control rate. During the search process, the goal is to maximize the value of the joint optimization control target, and the weight between road traffic flow and vehicle delay is balanced by adjusting the adjustment coefficient;

[0098] Generate edge real-time optimization solutions based on search and optimization results;

[0099] The generated real-time optimization plan is sent to the terminal execution device through the edge station to adjust the main line speed limit and ramp control rate in real time, and monitor the execution effect of the real-time optimization plan in real time;

[0100] Expand the current real-time optimization solution for a single main line and ramp to other sections of the road network in a random environment to achieve collaborative control of global traffic including shared control solutions, traffic status and experience;

[0101] Among them, the road traffic efficiency and delay balance model is:

[0102]

[0103] In the formula, TTF is the traffic flow efficiency model, TTD is the traffic delay model, is the main line traffic flow density at time k, is the main line traffic flow speed at time k, λ mis the number of main line lanes at time k, is the ramp control rate at time k, is the ramp traffic flow density at time k, is the ramp traffic flow speed at time k, is the number of ramp lanes at time k, Δl m is the length of the road section of control unit m, v f，m is the free flow speed of control unit m, r m is the number of ramps;

[0104] The objective function of the joint optimization control objective is: J = τ·TTF-(1-τ)·TTD, where τ is the adjustment coefficient and its value range is (0,1).

[0105] In a second aspect, an embodiment of the present invention provides an urban traffic cloud-edge-terminal integrated collaborative management and control system, including:

[0106] An intersection clustering module is used to calculate the traffic flow similarity based on the traffic flow data of at least two intersections obtained, and obtain the intersection correlation by combining the distance between intersections and the travel time, and to cluster the intersections with the goal of minimizing the objective function determined by the intersection correlation to obtain a plurality of intersection sets;

[0107] A strategy configuration module is used to configure a corresponding traffic signal control strategy for each intersection in the set of intersections using conditional judgment logic according to the change rate of the traffic flow and queue length of each intersection obtained and the adjacent intersection correlation index determined by the traffic flow similarity for any intersection set;

[0108] A model building module is used to respond to the signal control task generated by at least one intersection in each intersection set, build an overall optimization objective function according to the task resource requirements, green wave coordination requirements and edge station status, and form a cloud-edge collaborative scheduling model in combination with a hybrid evolutionary algorithm model and / or a heuristic algorithm model;

[0109] The collaborative management and control module is used to intelligently assign signal control tasks to at least one edge station for processing based on the cloud-edge collaborative scheduling model, and to form global traffic situation awareness by analyzing the aggregated data from multiple edge stations, so as to collaboratively manage the traffic signal control strategies configured at all intersections.

[0110] In a third aspect, an embodiment of the present invention provides an urban traffic cloud-edge-terminal integrated collaborative management and control platform, including:

[0111] Terminal sensing devices are deployed at various traffic nodes to collect and transmit road information in real time;

[0112] Terminal execution equipment, used to receive and execute control instructions to guide and adjust traffic flow;

[0113] Edge stations are used to receive and process data from terminal sensing devices in real time, and send control instructions to terminal control devices to achieve instant response and control of local traffic, thereby reducing network latency and alleviating the load on the central platform;

[0114] The central platform is used to execute the above-mentioned method by interacting with the edge stations.

[0115] Optionally, the terminal sensing device includes: video surveillance equipment, radar sensing equipment and RSU; the terminal execution equipment includes: signal light, variable lane indicator and induction screen.

[0116] Optionally, the edge station is deployed with an intelligent network connection hardware device that is connected to the edge end downward and an edge application operating system that provides a two-way interconnected data channel for cloud-edge interaction upward. The edge application operating system includes:

[0117] The access layer, as a plug-in device driver framework, is used to provide at least one type of device driver, support unified access and control of different devices at the edge station, and dynamically add new device drivers for new devices;

[0118] The platform layer is configured with multiple data platforms, which are used to meet the general needs of the application layer for storage, query and control by deploying data storage management modules, unified acquisition control modules and device access management modules;

[0119] The application layer is used to run various edge intelligent applications and provide a unified development interface to support device data query, device control distribution, application status monitoring, and remote deployment and maintenance;

[0120] The interface layer is used to achieve two-way collaborative interaction between different cloud-edge applications through set network connections, and can also interact with edge intelligent applications running in adjacent edge stations through messages.

[0121] (III) Beneficial effects

[0122] The beneficial effects of the present invention are:

[0123] First, the present invention calculates traffic flow similarity and intersection correlation, and then clusters intersections. This method can more accurately grasp the traffic conditions of various intersections in the city, providing a solid foundation for subsequent traffic signal control strategies. Secondly, this method combines traffic flow, queue length change rate and adjacent intersection correlation index to configure corresponding traffic signal control strategies for each intersection set, which not only improves the real-time response capability of traffic signals, but also effectively alleviates traffic congestion.

[0124] Furthermore, the present invention intelligently allocates signal control tasks to edge stations for processing by constructing a cloud-edge collaborative scheduling model. This collaborative architecture fully utilizes the advantages of edge computing, significantly reduces network latency, and ensures the timeliness of task processing, which is particularly efficient in complex traffic signal task allocation and resource scheduling scenarios. At the same time, the model also takes into account task resource requirements, green wave coordination requirements, and edge station status, thereby achieving balanced resource allocation and further improving the efficiency of the traffic control system.

[0125] Finally, by analyzing the aggregated data from multiple edge stations, the present invention can form a global traffic situation awareness, providing strong support for optimizing and adjusting the traffic signal control strategy of all intersections. This global optimization not only helps to improve the operating efficiency of the entire urban transportation system, but also brings higher flexibility and intelligence to urban traffic management.

[0126] Therefore, the present invention effectively utilizes the advantages of edge computing through accurate intersection clustering, real-time signal control strategy configuration, efficient cloud-edge-end coordinated scheduling, and global traffic situation awareness, significantly reduces network latency, and thus improves the real-time response capability of traffic signals. It not only optimizes resource allocation and task processing speed, but also ensures the flexibility and scalability of the traffic system, bringing significant benefits to urban traffic management. BRIEF DESCRIPTION OF THE DRAWINGS

[0127] Figure 1 A schematic diagram of a flow chart of a method provided by an embodiment of the present invention;

[0128] Figure 2 A schematic diagram of a specific flow chart of S1 of the method provided in an embodiment of the present invention;

[0129] Figure 3 A schematic diagram of a specific flow chart of S2 of the method provided in an embodiment of the present invention;

[0130] Figure 4 A schematic diagram of a specific flow chart of S3 of the method provided in an embodiment of the present invention;

[0131] Figure 5 A schematic diagram of a specific flow chart of S4 of the method provided in an embodiment of the present invention;

[0132] Figure 6 A specific flow chart of S49 of the method provided in an embodiment of the present invention

[0133] Figure 7 A schematic diagram of the edge application operating system composition of the platform provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0134] 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.

[0135] like Figure 1 As shown, an urban traffic cloud-edge integrated collaborative control method proposed in an embodiment of the present invention includes: calculating traffic flow similarity based on traffic flow data of at least two intersections obtained, and obtaining intersection correlation in combination with the distance between intersections and the travel time, clustering intersections with the goal of minimizing the objective function determined by the intersection correlation to obtain a number of intersection sets; for any intersection set, according to the obtained flow and queue length change rate of each intersection, combined with the adjacent intersection correlation index determined by the traffic flow similarity, using conditional judgment logic to configure a corresponding traffic signal control strategy for each intersection in each intersection set; in response to the signal control task generated by at least one intersection in each intersection set, constructing a total optimization objective function according to the task resource requirements, green wave coordination requirements and edge station status, and forming a cloud-edge-end collaborative scheduling model in combination with a hybrid evolutionary algorithm model and / or a heuristic algorithm model; based on the cloud-edge-end collaborative scheduling model, intelligently assigning the signal control task to at least one edge station for processing, and forming a global traffic situation awareness by analyzing the aggregated data from multiple edge stations, so as to collaboratively control the traffic signal control strategies configured for all intersections.

[0136] First, the present invention calculates traffic flow similarity and intersection correlation, and then clusters intersections. This method can more accurately grasp the traffic conditions of various intersections in the city, providing a solid foundation for subsequent traffic signal control strategies. Secondly, this method combines traffic flow, queue length change rate and adjacent intersection correlation index to configure corresponding traffic signal control strategies for each intersection set, which not only improves the real-time response capability of traffic signals, but also effectively alleviates traffic congestion.

[0137] Furthermore, the present invention intelligently allocates signal control tasks to edge stations for processing by constructing a cloud-edge collaborative scheduling model. This collaborative architecture fully utilizes the advantages of edge computing, significantly reduces network latency, and ensures the timeliness of task processing, which is particularly efficient in complex traffic signal task allocation and resource scheduling scenarios. At the same time, the model also takes into account task resource requirements, green wave coordination requirements, and edge station status, thereby achieving balanced resource allocation and further improving the efficiency of the traffic control system.

[0138] Finally, by analyzing the aggregated data from multiple edge stations, the present invention can form a global traffic situation awareness, providing strong support for optimizing and adjusting the traffic signal control strategy of all intersections. This global optimization not only helps to improve the operating efficiency of the entire urban transportation system, but also brings higher flexibility and intelligence to urban traffic management.

[0139] Therefore, the present invention effectively utilizes the advantages of edge computing through precise intersection clustering, real-time signal control strategy configuration, efficient cloud-edge coordinated scheduling, and global traffic situation awareness, significantly reduces network latency, and thus improves the real-time response capability of traffic signals. The adoption of this architecture not only optimizes resource allocation and task processing speed, but also ensures the flexibility and scalability of the traffic system, bringing significant benefits to urban traffic management.

[0140] In order to better understand the above technical solutions, exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings.

[0141] Specifically, an embodiment of the present invention provides a method for integrated collaborative management and control of urban traffic cloud, edge and terminal, including:

[0142] S1. Calculate traffic flow similarity based on traffic flow data of at least two intersections, and obtain intersection correlation by combining distance between intersections and travel time, perform intersection clustering with the goal of minimizing an objective function determined by intersection correlation, and obtain several intersection sets.

[0143] Furthermore, if Figure 2 As shown, step S1 includes:

[0144] S11. Obtain traffic flow data of at least two intersections and devices, and calculate traffic flow similarity based on the traffic flow data of at least two intersections and devices.

[0145] S12. Obtain the distance and travel time between each pair of intersections, and calculate the intersection correlation by combining the traffic flow similarity.

[0146] S13. Construct a clustering objective function based on intersection correlation.

[0147] S14. With the goal of minimizing the clustering objective function, an improved K-means algorithm is used to cluster intersections. During the clustering process, each intersection is iteratively assigned to the nearest cluster center, and the position of the cluster center is updated until the convergence condition is met to form multiple intersection sets.

[0148] In one embodiment, first, according to the traffic volume x at intersection i, i and the traffic volume x at intersection j j , calculate the traffic flow similarity F ij ;

[0149]

[0150] Further calculate the intersection correlation R based on the distance between intersections and travel time ij for:

[0151]

[0152] In the formula, R ij Intersection correlation, T ij is the average travel time between intersections i and j, D ij is the physical distance between intersections i and j, w1, w2, w3 are weight coefficients;

[0153] And use the improved K-means algorithm to cluster intersections and minimize the following clustering objective function:

[0154]

[0155] In the formula, X k is the clustered intersection set, k=1,2,…K is the sequence number of the intersection set.

[0156] S2. For any intersection set, according to the change rate of the traffic volume and queue length of each intersection, combined with the correlation index of adjacent intersections determined by the traffic volume similarity, the conditional judgment logic is used to configure the corresponding traffic signal control strategy for each intersection in each intersection set.

[0157] Furthermore, if Figure 3 As shown, step S2 includes:

[0158] S21. For any set of intersections, the traffic volume and queue length at each intersection are used to calculate the rate of change of the traffic volume and queue length at each intersection.

[0159] S22. Calculate the correlation index of adjacent intersections based on the traffic flow similarity.

[0160] S23. According to the flow rate at each intersection, the rate of change of queue length and the correlation index of adjacent intersections, conditional judgment logic is used to deploy an initialized traffic signal control strategy for the traffic status of each intersection, including single-point center adaptive control, edge single-point second-level adaptive control, center adaptive green wave control and edge second-level adaptive green wave control.

[0161] S24. Introduce a deep Q network model in deep reinforcement learning, determine the action set selected by the initialized traffic signal control strategy, and evaluate the effect of the control strategy based on feedback, and obtain the action reward by combining multiple factors including reduction in intersection delay, reduction in queue length, and number of vehicles.

[0162] S25. In the deep Q network model, historical and real-time traffic conditions are continuously tested and learned, the action sequence that maximizes the cumulative action reward is selected by approximating the Q-value function, and the discount factor is used to balance short-term and long-term rewards.

[0163] S26. According to the action sequence output by the deep Q network model that maximizes the cumulative reward, the traffic signal control strategy of each intersection is adjusted to obtain the optimal traffic signal control strategy at this stage.

[0164] In another embodiment, for any intersection set X k , with the traffic volume x at each intersection i and queue length q i As input, calculate the rate of change of flow and queue length:

[0165]

[0166] Among them, V xi is the traffic volume x i The rate of change, V qi is the rate of change of the queue length at intersection i, std is the standard deviation, mean is the mean, and (t→t+τ) represents the time series from time t to time t+τ.

[0167] At the same time, the correlation index of adjacent intersections is:

[0168]

[0169] Next, use conditional judgment to initialize the corresponding control strategy for the traffic state of each intersection. The conditional judgment logic is:

[0170]

[0171] Where C1 is the single-point center adaptive control strategy, C2 is the edge single-point second-level adaptive control strategy, C3 is the center adaptive green wave control strategy, C4 is the edge second-level adaptive green wave control strategy, V x is the change rate threshold of vehicle flow, V q is the queue length change rate threshold, V F is the threshold of the correlation index of adjacent intersections.

[0172] Specifically, the single-point center adaptive control strategy is a signal control strategy that makes real-time adjustments to a single intersection. It dynamically adjusts the timing of traffic lights based on information including the real-time traffic flow at the intersection, the queue length, and the correlation index of adjacent intersections to optimize the traffic efficiency of the intersection. For example, during peak hours in the morning and evening, when the traffic flow in a certain direction increases significantly, the green light time in that direction will be extended to quickly evacuate traffic flow and reduce congestion. During off-peak hours, the green light time is appropriately shortened to balance the traffic demand in each direction and improve the overall traffic efficiency of the intersection. This control strategy relies on the central platform for data processing and decision-making, and can achieve more refined control, but it may have certain requirements for the real-time performance and processing capabilities of the system. It is suitable for intersections with large traffic flow changes in traffic application scenarios and the need to frequently adjust the timing of traffic lights, especially those key intersections with high requirements for traffic efficiency.

[0173] The edge single-point second-level adaptive control strategy is an adaptive signal control strategy with a second-level response implemented on the edge device of the intersection. It quickly adjusts the timing of the signal light based on the information of the real-time traffic flow and queue length of the intersection to cope with the instantaneous changes in traffic flow. The specific adjustment method includes: if the queue length in a certain direction exceeds the preset threshold and lasts for a long time, it is decided to extend the green light time in that direction. At the same time, the green light time in the direction with shorter queues or less traffic flow is shortened, and more travel time is allocated to the required direction to optimize the overall traffic efficiency of the intersection. This control strategy has the characteristics of fast response speed and high flexibility. It can respond to changes in traffic flow within seconds and reduce vehicle waiting time. It is suitable for intersections where traffic flow changes frequently and requires rapid response, especially those intersections where traffic congestion is prone to occur or traffic flow fluctuates greatly.

[0174] The center adaptive green wave control strategy is a green wave signal control strategy that coordinates and controls multiple adjacent intersections. It dynamically adjusts the timing of the signal lights at each intersection based on information including the traffic volume, queue length, and correlation index between intersections to form a continuous green wave belt and improve the speed and efficiency of vehicle traffic. The specific adjustments include: for intersections with large traffic volume, the green light time is extended to ensure that vehicles can pass quickly. For adjacent intersections, a reasonable phase difference is set so that vehicles can just reach the green light period of the next intersection when encountering a red light. The speed and width of the green wave belt are also dynamically adjusted based on information such as real-time traffic volume and queue length to adapt to changes in traffic flow. This control strategy can achieve coordinated control between multiple intersections, form a green wave belt, reduce the waiting time and number of stops of vehicles at intersections, and improve the overall traffic capacity of the road. It is suitable for main roads or important traffic corridors that need to form a continuous green wave belt and increase vehicle traffic speed.

[0175] Furthermore, edge second-level adaptive green wave control is an adaptive green wave signal control strategy with a second-level response implemented on the edge device of the intersection. It quickly adjusts the timing of the traffic lights at each intersection to form a dynamic green wave belt based on information including real-time traffic flow, queue length, and correlation index between adjacent intersections. The specific adjustments include: if the traffic flow in a certain direction is detected to increase significantly and the queue length exceeds the preset threshold, it is decided to extend the green light time in that direction, and consider the correlation of adjacent intersections to adjust the phase difference to form a green wave belt. Monitor the traffic flow in other directions to ensure that the adjusted signal light timing will not have a serious impact on other directions. At the same time, according to the changes in real-time traffic status, the speed and width of the green wave belt are dynamically adjusted to ensure the continuity and effectiveness of the green wave belt. This control strategy combines the fast response capability of edge devices and the optimization effect of green wave control. It can respond to traffic flow changes within seconds and form a dynamic green wave belt, improving the traffic efficiency of the road and the driving experience of drivers. It is suitable for complex traffic networks or important traffic nodes that need to quickly form and adjust green wave belts and respond to instantaneous changes in traffic flow.

[0176] Next, a deep Q network model in deep reinforcement learning is introduced. In this embodiment, a deep Q network (DQN) in deep reinforcement learning (DRL) is used to optimize the control strategy. The model can continuously optimize the control task of the intersection by learning historical and real-time traffic conditions. The action set selected by the strategy is defined as A = {C1, C2, C3, C4}.

[0177] The effect of each control strategy is evaluated, and the action reward is calculated based on factors such as the reduction of intersection delay, the reduction of queue length, and the number of passing vehicles. The action reward is:

[0178]

[0179] In the formula, Td k is the average delay time, Mq k is the average queue length, maxMx k and Mx k are the maximum average traffic flow and the actual average traffic flow, respectively, α1, α2, α3 are weight parameters;

[0180] Through continuous experimentation and learning, choices are made in the long run to accumulate rewards The action sequence that maximizes, where γ is the discount factor, a number between 0 and 1, used to balance short-term and long-term rewards. The discount factor γ determines how the agent weighs immediate rewards and future rewards when making decisions. A larger γ value means that the agent values ​​future rewards more and is willing to sacrifice some immediate rewards in order to obtain greater future rewards. On the contrary, a smaller γ value makes the agent pay more attention to immediate rewards. T is the termination time.

[0181] The mechanism of selecting the optimal action through the Q value function in strategy. In DQN, the strategy is learned and updated through the neural network, and the optimal strategy is selected by approximating the Q value function. The Q value function represents the expected cumulative return obtained by selecting an action a in a given state s:

[0182]

[0183] In the formula, s is the current state of the agent, a is the action taken by the agent in state s, E[·] is the expected value of the cumulative sum of all possible future reward sequences after the agent takes action a starting from state s, s0=s represents the initial state, and a0=a represents the initial action. The neural network is used to approximate the Q value function, with the input being state s and the output being the Q value Q(s,a) of different actions.

[0184] S3. In response to the signal control task generated by at least one intersection in each intersection set, an overall optimization objective function is constructed according to the task resource requirements, green wave coordination requirements and edge station status, and a cloud-edge collaborative scheduling model is formed by combining the hybrid evolutionary algorithm model and / or the heuristic algorithm model.

[0185] Furthermore, if Figure 4 As shown, step S3 includes:

[0186] S31. Monitor and respond to a signal control task generated by at least one intersection in each intersection set.

[0187] S32: Perform resource demand analysis on the signal control tasks generated at each intersection to determine the resource and bandwidth requirements of the intersection.

[0188] S33, measuring or estimating the status information of the edge station including the number of resources, resource upper limit, number of task types, and data transmission delay from each intersection to the station.

[0189] S34. Determine the green wave coordination requirements through traffic flow analysis, road network assessment and signal system status survey results.

[0190] S35. Based on the task resource requirements, green wave coordination requirements and the status information of the edge stations, with the goals of balancing the load of each edge station, reducing the types of tasks, reducing transmission delays and meeting the green wave coordination requirements, the overall objective function of the multi-objective optimization problem is constructed, and the importance of each objective is adjusted through the benefit coefficient and weight parameters of resource allocation.

[0191] S36. Construct a cloud-edge-end collaborative scheduling model based on the overall objective function, hybrid evolutionary algorithm and heuristic algorithm.

[0192] In another embodiment, a sub-area is defined to include a set of several intersections X. k ={x1, x2, ..., x I} and several edge stations E k ={e1, e2, ..., e J Each intersection will generate a set of signal control tasks, which need to be assigned to edge stations for processing. The task resource requirements of the i-th intersection include computing resources and bandwidth as R i , the number of task types of the i-th station is T j , D ij is the delay overhead from the data of the i-th intersection device to the j-th small station, σ i，j represents the number of resources allocated to the jth small station at the i-th intersection, C j is the resource upper limit of the jth small station. In order to balance the load of each small station, reduce the types of tasks, reduce transmission delay and green wave coordination requirements, the goal is designed as a multi-objective optimization problem, and the overall objective function is:

[0193]

[0194] Where L is the total objective function, I is the total number of intersections, J is the total number of edge stations, α0 is the profit coefficient of resource allocation, σ i,j is the number of resources allocated to the jth small station at the i-th intersection, α4, α5, α6 are weight parameters, and L R is the load balancing target, L T is the task type target, L D is the transmission delay target.

[0195] Next, we will introduce each sub-project and the constraints in the optimization process:

[0196] (1) The load balancing goal is to minimize the variance of the load of each small station so that the resource allocation of different small stations is as balanced as possible. The load target L R for:

[0197]

[0198] Where Var is the variance.

[0199] (2) The penalty term is used to represent the cost of each small station processing multiple task types. The goal is to minimize the number of task types on a small station. The task type target L T for:

[0200]

[0201] (3) Minimize the total delay of data transmission during task allocation and optimize the transmission delay overhead. The transmission delay target L Dfor:

[0202]

[0203] And the constraints in the optimization process are:

[0204]

[0205] More in-depth, in order to solve the above multi-objective optimization problem, a hybrid evolutionary algorithm and heuristic algorithm can be used, such as an improved genetic algorithm or a particle swarm optimization algorithm, which can not only quickly find high-quality scheduling solutions, but also dynamically adjust the allocation strategy according to different objectives.

[0206] S4. Based on the cloud-edge collaborative scheduling model, the signal control task is intelligently assigned to at least one edge station for processing, and by analyzing the aggregated data from multiple edge stations, a global traffic situation awareness is formed to coordinate the management and control of the traffic signal control strategies configured at all intersections.

[0207] Furthermore, if Figure 5 As shown, step S4 includes:

[0208] S41. Use a two-dimensional coding matrix to represent the task allocation scheme, wherein the elements of the coding matrix represent the number of resources allocated to the edge station at the intersection.

[0209] S42. Under preset constraints, randomly generate an initial particle swarm, where each particle represents a possible task allocation scheme; initialize a velocity vector for each particle, where the dimension of the velocity vector is the same as the encoding matrix, and the velocity value is randomly initialized within a preset range.

[0210] S43, for each particle in the particle swarm, iteratively update the velocity vector and the encoding matrix; after each iteration, evaluate the optimal fitness value of the current particle swarm according to the overall objective function, and compare it with the optimal fitness value of the previous generation to determine whether the termination condition is met.

[0211] S44. If the termination condition is met, the best individual is selected from the current particle swarm as the best task allocation scheme, and the signal control tasks generated at each intersection are allocated to at least one edge station for processing based on the best task allocation scheme.

[0212] In a specific embodiment, a suitable encoding method is first used to represent the task allocation scheme, using a two-dimensional matrix {σ i，j},σ i，j represents the number of resources allocated to edge station j at intersection i, and randomly generates the initial particle swarm under the constraint conditions. At the same time, the velocity vector v is initialized for each particle i，j,The dimension of the velocity vector is the same as the encoding matrix, and the velocity value is also randomly initialized within a certain range.

[0213] Next, for each particle in the particle swarm, update its velocity v according to the formula of the particle swarm optimization algorithm. i，j and position σ i，j :

[0214] v ij (t+1)=w·v ij (t)+c1·r1·(p i，j (t)-σ i，j (t))+c2·r2·(g i，j (t)-σ i，j (t))

[0215] σ i，j (t+1)=σ i，j (t)+v ij (t+1); where w is the inertia weight, c1 and c2 are learning factors, r1 and r2 are random numbers between [0,1], and p i，j (t) is the historical optimal position at the tth iteration, g i，j (t) is the global optimal position of the particle swarm at the tth iteration. The first term in the speed update formula represents the particle's dependence on the current speed, the second term represents the particle's tendency to approach its own historical optimal position, and the third term represents the particle's tendency to approach the swarm's global optimal position. The position update formula adjusts the particle's position (task allocation) according to the updated speed.

[0216] If the maximum number of iterations is reached or the optimal fitness value of the population is no longer improved for several consecutive generations, if the termination condition is met, the optimal individual (task allocation scheme) in the population is output as the final result.

[0217] S45. Collect raw traffic data from multiple edge stations, perform data cleaning, format unification, and time synchronization to obtain processed traffic data.

[0218] S46, classifying the processed traffic data according to multiple dimensions including intersections, areas and time periods, and integrating the classified data to form a structured data set.

[0219] S47. Extract basic traffic parameters that can characterize traffic status and dynamic features based on spatiotemporal relationships from structured data sets.

[0220] S48. Perform data fusion on the extracted basic traffic parameters and dynamic features based on time-space relationships to form a comprehensive feature set, and extract the most representative feature vector from the comprehensive feature set; perform data mining based on the feature vector to form a global traffic situation awareness, where the traffic situation awareness is multi-dimensional information including traffic flow, congestion conditions and accident risks.

[0221] Specifically, after receiving a large amount of raw traffic data, the central platform first performs data cleaning. This process aims to remove noise, erroneous data, and outliers in the data to ensure the accuracy and reliability of subsequent analysis. For example, by setting reasonable rules such as vehicle speed range and traffic flow threshold, data points that obviously do not conform to actual traffic conditions are screened out; at the same time, the data from different types of sensing devices are formatted and time synchronized so that various data can be comprehensively analyzed under the same time reference and data format. After cleaning, the multi-source traffic data is classified and integrated according to dimensions such as intersections, regions, and time periods to form a structured data set, providing a solid data foundation for subsequent in-depth analysis and situational awareness.

[0222] Based on the integrated data set, the central platform extracts key features that can characterize traffic status from multiple perspectives, covering not only traditional traffic parameters such as traffic flow, vehicle speed, density, etc., but also dynamic features based on spatiotemporal relationships, such as the spread trend of traffic congestion, the correlation of traffic flow between intersections, etc. By analyzing historical traffic data, a spatiotemporal distribution model of traffic flow is established to identify the peak and trough periods of traffic flow in different time periods and regions and their changing patterns; at the same time, cluster analysis, principal component analysis and other methods are used to extract the most representative feature vectors from a large number of traffic data variables to describe traffic status concisely and effectively.

[0223] S49. Based on the results of global traffic situation perception, the variable speed limit on the main line and traffic induction coordinated control are driven by edge stations, and the traffic signal control strategies configured at all intersections are optimized and adjusted to generate a real-time optimization plan that includes the variable speed limit on the main line and the coordinated intelligent control of ramps.

[0224] Furthermore, if Figure 6 As shown, step S49 specifically includes:

[0225] S491. Based on the result of global traffic situation perception, each main line and ramp is abstracted into multiple control units, and a state information set of each control unit is generated according to the main line number, ramp number and time step. Among them, the control unit state information set: {(ρ m (k), v m (k), n m (k), r r (k),ρr (k), v r (k), n r (k))}where m is the main line number, r is the ramp number, and k is the time step.

[0226] S492. Based on the traffic signal control strategies, control units and corresponding status information sets currently configured at all intersections, a road traffic efficiency and delay balance model is constructed with road traffic flow and vehicle delay as joint optimization control targets.

[0227] S493. Based on the results of global traffic situation perception, the input parameters of the road traffic efficiency and delay balance model are adjusted and handed over to the edge station for search to find the optimal combination of the main line variable speed limit value and the entrance ramp control rate. During the search process, the goal is to maximize the value of the joint optimization control target, and the weight between road traffic flow and vehicle delay is balanced by adjusting the adjustment coefficient.

[0228] S494. Generate an edge real-time optimization solution based on the search and optimization results.

[0229] S495. Send the generated real-time optimization plan to the terminal execution device through the edge station to adjust the main line speed limit and ramp control rate in real time, and monitor the execution effect of the real-time optimization plan in real time.

[0230] S496. Expand the current real-time optimization scheme for the single main line and ramp to other sections of the road network under a random environment to achieve collaborative control of global traffic, including shared control schemes, traffic status, and experience. Specifically, obtain the current real-time optimization scheme for the single main line and ramp; evaluate and predict the traffic status of the road network; design shared control schemes for other sections of the road network based on the real-time optimization scheme and traffic status prediction results; learn from historical traffic data and share successful control schemes and experience knowledge with other sections of the road network; implement the designed shared control scheme in other sections of the road network to achieve collaborative control of global traffic; evaluate the traffic effect after implementation, and adjust the control scheme based on the evaluation results.

[0231] Among them, the road traffic efficiency and delay balance model is:

[0232]

[0233] In the formula, TTF is the traffic flow efficiency model, TTD is the traffic delay model, is the main line traffic flow density at time k, is the main line traffic flow speed at time k, λ m is the number of main line lanes at time k, is the ramp control rate at time k, is the ramp traffic flow density at time k, is the ramp traffic flow speed at time k, is the number of ramp lanes at time k, Δl m is the length of the road section of control unit m, ν f，m is the free flow speed of control unit m, r m is the number of ramps.

[0234] The objective function of the joint optimization control objective is: J = τ·TTF-(1-τ)·TTD, where τ is the adjustment coefficient and its value range is (0,1).

[0235] In addition, an embodiment of the present invention provides an urban traffic cloud-edge-terminal integrated collaborative management and control system, including:

[0236] The intersection clustering module is used to calculate the traffic flow similarity based on the traffic flow data of at least two intersections, and obtain the intersection correlation by combining the distance between intersections and the travel time, and to cluster the intersections with the goal of minimizing the objective function determined by the intersection correlation to obtain several intersection sets.

[0237] The strategy configuration module is used to configure the corresponding traffic signal control strategy for each intersection in each intersection set based on the change rate of the traffic flow and queue length of each intersection obtained, combined with the adjacent intersection correlation index determined by the traffic flow similarity, using conditional judgment logic.

[0238] The model building module is used to respond to the signal control task generated by at least one intersection in each intersection set, construct the overall optimization objective function according to the task resource requirements, green wave coordination requirements and edge station status, and combine the hybrid evolutionary algorithm model and / or heuristic algorithm model to form a cloud-edge collaborative scheduling model.

[0239] The collaborative management and control module is used to intelligently assign signal control tasks to at least one edge station for processing based on the cloud-edge collaborative scheduling model, and to form global traffic situation awareness by analyzing the aggregated data from multiple edge stations, so as to collaboratively manage the traffic signal control strategies configured at all intersections.

[0240] Next, refer to Figure 7, an embodiment of the present invention provides an urban traffic cloud-edge integrated collaborative management and control platform, including: a terminal sensing device, deployed at each traffic node, for collecting and transmitting road information in real time. A terminal execution device, for receiving and executing control instructions to guide and adjust traffic flow. An edge station, for receiving and processing data from the terminal sensing device in real time, and for sending control instructions to the terminal execution device, so as to achieve instant response and effective control of local traffic, thereby reducing network latency and alleviating the load on the central platform. The central platform is used to execute the method described above by interacting with the edge station.

[0241] Further, refer to Figure 7 The terminal sensing equipment includes: video surveillance equipment, radar sensing equipment and RSU (roadside unit); the terminal execution equipment includes: signal machine, variable lane indicator and guidance screen.

[0242] Furthermore, the edge station is deployed with intelligent networked hardware devices that are connected to the edge downward, and an edge application operating system that provides a two-way interconnected data channel for cloud-edge interaction upward. The edge application operating system is an edge basic platform running on the edge computing station hardware devices, providing core functions such as device data access, device operation control, cloud-edge two-way interaction, adjacent station communication, and operating environment support for edge intelligent applications.

[0243] Specifically, edge application operating systems include:

[0244] The access layer, as a plug-in device driver framework, is used to provide various device drivers such as signal machine drivers, video detector drivers, radar drivers, bayonet camera drivers, etc. It supports unified access and control of different devices at edge stations, and can dynamically add new device drivers for new devices to achieve flexible expansion of access layer functions.

[0245] The platform layer is configured with multiple data platforms such as compatibility database, SCADA platform, and transportation IoT access platform, which are used to meet the general needs of the application layer for storage, query, and control by deploying data storage management modules, unified acquisition control modules, and device access management modules;

[0246] The application layer is used to run various edge intelligent applications, such as single-point sensing control algorithms, intersection overflow disposal algorithms, and other edge intelligent applications, and provides a unified JNI API development interface to support device data query, device control distribution, application status monitoring, and remote deployment and maintenance;

[0247] The interface layer is used to achieve two-way collaborative interaction between different cloud-edge applications through specific network connections such as WebSocket or HTTP network connections, and build cloud-edge integrated infrastructure services. At the same time, it also provides corresponding communication interfaces, which can interact with edge intelligent applications running on adjacent edge stations, and support the implementation of key technologies such as multi-agent collaborative management and control.

[0248] It is worth mentioning that the edge application operating system has function extension points at different levels, so that users can dynamically add custom business logic modules during data access, processing, and distribution.

[0249] In summary, the embodiments of the present invention provide a method, system and platform for the integrated collaborative management and control of urban transportation cloud, edge and terminal. First of all, in terms of the architecture of the cloud, edge and terminal integrated collaborative management and control platform, the present invention cleverly integrates the distributed architecture of cloud, edge and terminal. The architecture is mainly composed of four parts: terminal sensing device, terminal execution device, edge computing device and central platform. Through real-time perception and processing of data by edge devices, the present invention successfully reduces the load of the central server and reduces network latency. This design has undoubtedly become the core technical highlight of the system.

[0250] Secondly, the present invention introduces an adaptive traffic signal control strategy. With the help of the deep Q network (DQN) optimization control strategy in deep reinforcement learning, it can intelligently learn from historical and real-time traffic data and optimize the signal control of each intersection. This innovative measure significantly improves the real-time and efficiency of traffic management, demonstrating the outstanding strength of the present invention in intelligence.

[0251] Furthermore, the present invention also introduces a multi-objective optimization scheduling model, which plays a key role in optimizing resource allocation between edge stations. By ensuring load balancing, task type balancing, and minimizing transmission delays, combined with a hybrid evolutionary algorithm or particle swarm optimization algorithm, the scheduling model achieves reasonable and efficient scheduling of resources for each edge station. This feature is not common in the prior art, especially when dealing with complex traffic signal task allocation and resource scheduling scenarios, the scheduling model of the present invention shows higher efficiency.

[0252] In addition, the present invention also integrates intelligent network hardware equipment technology. As the infrastructure of the platform, the edge application operating system not only provides flexible device data access and operation control functions, but also supports unified access and control of different devices in the edge computing station. This design undoubtedly enhances the compatibility and scalability of the system, laying a solid foundation for the future development of intelligent transportation.

[0253] It is worth mentioning that the cloud-edge-end collaborative architecture of the present invention demonstrates high efficiency. Through the collaborative work of cloud, edge and end devices, the present invention gives full play to the advantages of edge computing, reduces network latency and improves the real-time response capability of traffic signals. At the same time, the application of a multi-objective optimization scheduling model also ensures the balanced allocation of resources between edge stations, thereby reducing the delay in task processing. These advantages are not common in the prior art, especially in complex scenarios of traffic signal task allocation and resource scheduling, the scheduling model of the present invention appears to be more efficient and practical.

[0254] Finally, the flexible scalability of intelligent networked devices is also a highlight of this invention. The edge application operating system adopts a modular architecture design, which not only supports the access and collaborative work of multiple traffic sensing and control devices, but also has good scalability. This means that the system can dynamically add new device drivers to adapt to the ever-changing traffic management needs.

[0255] 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.

[0256] 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.

[0257] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments after knowing the basic creative concept. Therefore, the claims should be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present invention.

[0258] 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 present invention and its equivalent technologies, the present invention should also include these modifications and variations.

Claims

1. A method for integrated collaborative management and control of urban traffic cloud, edge and terminal, characterized in that: include: Based on the traffic flow data of at least two intersections obtained, the traffic flow similarity is calculated, and the intersection correlation is obtained by combining the distance between intersections and the travel time, and the intersections are clustered with the goal of minimizing the objective function determined by the intersection correlation to obtain a plurality of intersection sets; For any intersection set, according to the change rate of the traffic flow and queue length of each intersection, combined with the correlation index of adjacent intersections determined by the traffic flow similarity, the conditional judgment logic is used to configure the corresponding traffic signal control strategy for each intersection in each intersection set; In response to the signal control task generated by at least one intersection in each intersection set, a total optimization objective function is constructed according to the task resource requirements, green wave coordination requirements, and edge station status, and a cloud-edge collaborative scheduling model is formed by combining a hybrid evolutionary algorithm model and / or a heuristic algorithm model; Based on the cloud-edge collaborative scheduling model, signal control tasks are intelligently assigned to at least one edge station for processing. By analyzing the aggregated data from multiple edge stations, global traffic situation awareness is formed to coordinate the management and control of traffic signal control strategies configured at all intersections.

2. The urban traffic cloud-edge-terminal integrated collaborative management and control method according to claim 1 is characterized in that: According to the traffic flow data of at least two intersections, the traffic flow similarity is calculated, and the intersection correlation is obtained by combining the distance between intersections and the travel time. The intersections are clustered with the goal of minimizing the objective function determined by the intersection correlation, and several intersection sets are obtained, including: Obtaining traffic flow data of at least two intersections and devices, and calculating traffic flow similarity based on the traffic flow data of at least two intersections and devices; Obtain the distance and travel time between each pair of intersections, and combine the traffic flow similarity to obtain the intersection correlation; Construct a clustering objective function based on intersection correlation; With the goal of minimizing the clustering objective function, the improved K-means algorithm is used to cluster the intersections. During the clustering process, each intersection is iteratively assigned to the nearest cluster center, and the position of the cluster center is updated until the convergence condition is met to form multiple intersection sets. in, The traffic flow similarity is; In the formula, F ij is the traffic flow similarity, x i is the traffic volume at intersection i, x j is the traffic volume x at intersection j j ; The intersection correlation is: In the formula, R ij is the intersection correlation, T ij is the average travel time between intersections i and j, D ij is the physical distance between intersections i and j, w1, w2, w3 are weight coefficients; The clustering objective function is: Where, X k is the clustered intersection set, k=1,2,…K is the sequence number of the intersection set.

3. The urban traffic cloud-edge-terminal integrated collaborative management and control method according to claim 2 is characterized in that: For any intersection set, according to the change rate of the traffic flow and queue length of each intersection, combined with the adjacent intersection correlation index determined by the traffic flow similarity, the corresponding traffic signal control strategy is configured for each intersection in each intersection set using conditional judgment logic, including: For any set of intersections, the traffic flow and queue length at the intersection are used to calculate the rate of change of the traffic flow and queue length at the intersection; According to the similarity of traffic flow, the correlation index of adjacent intersections is calculated; According to the traffic volume of each intersection, the change rate of queue length and the correlation index of adjacent intersections, the conditional judgment logic is used to deploy the traffic signal control strategy for each intersection, including single-point center adaptive control, edge single-point second-level adaptive control, center adaptive green wave control and edge second-level adaptive green wave control initialization; A deep Q network model in deep reinforcement learning is introduced to determine the action set selected by the initialized traffic signal control strategy, and the effect of the control strategy is evaluated based on the feedback, and the action reward is obtained by combining multiple factors including the reduction of intersection delay, the reduction of queue length and the number of vehicles. In the deep Q-network model, the historical and real-time traffic status is continuously tested and learned, the action sequence that maximizes the cumulative action reward is selected by approximating the Q-value function, and the discount factor is used to balance the short-term and long-term rewards; According to the action sequence output by the deep Q network model that maximizes the cumulative reward, the traffic signal control strategy of each intersection is adjusted to obtain the optimal traffic signal control strategy at this stage; in, The rates of change of flow and queue length are: Among them, x i is the traffic volume at intersection i, V xi is the rate of change of traffic flow xi, q i is the queue length at intersection i, V qi is the rate of change of the queue length at intersection i, std is the standard deviation, mean is the mean, max is the maximum value, and (t→t+τ) represents the time series from time t to time t+τ; The correlation index of adjacent intersections is: The conditional judgment logic is: Where C1 is the single-point center adaptive control strategy, C2 is the edge single-point second-level adaptive control strategy, C3 is the center adaptive green wave control strategy, C4 is the edge second-level adaptive green wave control strategy, V x is the change rate threshold of vehicle flow, V q is the queue length change rate threshold, V F is the threshold value of the correlation index of adjacent intersections; The action set is: A = {C1, C2, C3, C4}; The action rewards are: In the formula, r t is the action reward at time t, Td k is the average delay time, Mq k is the average queue length, maxMx k and Mx k are the maximum average traffic flow and the actual average traffic flow, respectively, α1, α2, α3 are weight parameters; The cumulative action reward is: Where R is the cumulative action reward, γ is the discount factor, and T is the termination time; The Q value function is: Where s is the current state of the agent, a is the action taken by the agent in state s, E[·] is the expected value of the cumulative sum of all possible future reward sequences after the agent takes action a starting from state s, s0=s represents the initial state, and a0=a represents the initial action.

4. The urban traffic cloud-edge-terminal integrated collaborative management and control method according to claim 1 is characterized in that: In response to the signal control task generated by at least one intersection in each intersection set, the overall optimization objective function is constructed according to the task resource requirements, green wave coordination requirements and edge station status. The cloud-edge collaborative scheduling model is formed by combining the hybrid evolutionary algorithm model and / or the heuristic algorithm model, including: Monitor and respond to signal control tasks generated by at least one intersection in each intersection set; Perform resource demand analysis on the signal control tasks generated at each intersection to determine the resource and bandwidth requirements of the intersection; Measure or estimate the edge station's status information including the number of resources, resource upper limit, number of task types, and data transmission delay from each intersection to the station; Determine the need for green wave coordination through the results of traffic flow analysis, road network assessment and signal system status survey; Based on the task resource requirements, green wave coordination requirements and the status information of edge stations, the overall objective function of the multi-objective optimization problem is constructed with the goal of balancing the load of each edge station, reducing the task types, reducing transmission delay and meeting the green wave coordination requirements. The importance of each objective is adjusted through the benefit coefficient and weight parameter of resource allocation. Based on the overall objective function and the improved particle swarm optimization algorithm model, a cloud-edge-device collaborative scheduling model is constructed; in, The overall objective function is: Where L is the total objective function, I is the total number of intersections, J is the total number of edge stations, α0 is the profit coefficient of resource allocation, σ i,j is the number of resources allocated to the jth small station at the i-th intersection, α4, α5, α6 are weight parameters, L R is the load balancing target, L T is the task type target, L D is the transmission delay target; The load targets are: Where Var is the variance; The mission types and objectives are: Where, T j is the number of task types of the j-th station; The transmission delay target is: Where Dij is the delay overhead from the data of the i-th intersection equipment to the j-th small station; The constraints in the optimization process are: In the formula, C j is the resource upper limit of the jth small station, R i For computing resources and bandwidth.

5. The urban traffic cloud-edge-terminal integrated collaborative management and control method according to claim 1 is characterized in that: Based on the cloud-edge-device collaborative scheduling model, the signal control task is intelligently assigned to at least one edge station for processing, and the aggregated data from multiple edge stations is analyzed to form a global traffic situation awareness to coordinate the control of traffic signals at all intersections. The control strategies include: A two-dimensional coding matrix is ​​used to represent the task allocation scheme, where the elements of the coding matrix represent the number of resources allocated to the edge station at the intersection; Randomly generate an initial particle swarm, each particle represents a possible task allocation scheme; Initialize a velocity vector for each particle. The dimension of the velocity vector is the same as the encoding matrix, and the velocity value is randomly initialized within a preset range. For each particle in the particle swarm, iteratively update the velocity vector and encoding matrix; After each iteration, the optimal fitness value of the current particle swarm is evaluated according to the total objective function and compared with the optimal fitness value of the previous generation to determine whether the termination condition is met; If the termination condition is met, the best individual is selected from the current particle swarm as the optimal task allocation scheme, and the signal control tasks generated at each intersection are allocated to at least one edge station for processing based on the optimal task allocation scheme; Collect raw traffic data from multiple edge stations, clean the data, unify the format, and synchronize the time to obtain processed traffic data; The processed traffic data is classified according to multiple dimensions including intersections, areas and time periods, and the classified data is integrated to form a structured data set; Extract basic traffic parameters and dynamic features based on spatiotemporal relationships that can characterize traffic status from structured data sets; The extracted basic traffic parameters and dynamic features based on spatiotemporal relationships are fused to form a comprehensive feature set, and the most representative feature vector is extracted from the comprehensive feature set; Data mining based on feature vectors is used to form global traffic situation awareness, where traffic situation awareness is multi-dimensional information including traffic flow, congestion conditions and accident risks; Based on the results of global traffic situation perception, the variable speed limit on the main line and traffic induction coordinated control are driven by edge stations, and the traffic signal control strategies configured at all intersections are optimized and adjusted to generate a real-time optimization plan that includes the variable speed limit on the main line and the coordinated intelligent control of ramps.

6. The urban traffic cloud-edge-terminal integrated collaborative management and control method according to claim 5 is characterized in that: Based on the results of global traffic situation awareness, the variable speed limit on the main line and the coordinated control of traffic induction are driven by the edge station, and the traffic signal control strategies configured at all intersections are optimized and adjusted to generate a real-time optimization solution including the coordinated variable speed limit on the main line and the intelligent control of the ramp, including: According to the results of global traffic situation awareness, each main line and ramp is abstracted into multiple control units, and the state information set of each control unit is generated according to the main line number, ramp number and time step; According to the traffic signal control strategies, control units and corresponding status information sets currently configured at all intersections, a road traffic efficiency and delay balance model is constructed with road traffic flow and vehicle delay as the joint optimization control objectives; Based on the results of global traffic situation awareness, the input parameters of the road traffic efficiency and delay balance model are adjusted and handed over to the edge station for search to find the optimal combination of the main line variable speed limit value and the entrance ramp control rate. During the search process, the goal is to maximize the value of the joint optimization control target, and the weight between road traffic flow and vehicle delay is balanced by adjusting the adjustment coefficient; Generate edge real-time optimization solutions based on search and optimization results; The generated real-time optimization plan is sent to the terminal execution device through the edge station to adjust the main line speed limit and ramp control rate in real time, and monitor the execution effect of the real-time optimization plan in real time; Expand the current real-time optimization solution for a single main line and ramp to other sections of the road network in a random environment to achieve collaborative control of global traffic including shared control solutions, traffic status and experience; Among them, the road traffic efficiency and delay balance model is: In the formula, TTF is the traffic flow efficiency model, TTD is the traffic delay model, is the main line traffic flow density at time k, is the main line traffic flow speed at time k, λ m is the number of main line lanes at time k, is the ramp control rate at time k, is the ramp traffic flow density at time k, is the ramp traffic flow speed at time k, is the number of ramp lanes at time k, Δl m is the length of the road section of control unit m, v f，m is the free flow speed of control unit m, r m is the number of ramps; The objective function of the joint optimization control objective is: J = τ·TTF-(1-τ)·TTD, where τ is the adjustment coefficient and its value range is (0,1).

7. An urban traffic cloud-edge-end integrated collaborative management and control system, characterized in that: include: An intersection clustering module is used to calculate the traffic flow similarity based on the traffic flow data of at least two intersections obtained, and obtain the intersection correlation by combining the distance between intersections and the travel time, and to cluster the intersections with the goal of minimizing the objective function determined by the intersection correlation to obtain a plurality of intersection sets; A strategy configuration module is used to configure a corresponding traffic signal control strategy for each intersection in the set of intersections using conditional judgment logic according to the change rate of the traffic flow and queue length of each intersection obtained and the adjacent intersection correlation index determined by the traffic flow similarity for any intersection set; A model building module is used to respond to the signal control task generated by at least one intersection in each intersection set, build an overall optimization objective function according to the task resource requirements, green wave coordination requirements and edge station status, and form a cloud-edge collaborative scheduling model in combination with a hybrid evolutionary algorithm model and / or a heuristic algorithm model; The collaborative management and control module is used to intelligently assign signal control tasks to at least one edge station for processing based on the cloud-edge collaborative scheduling model, and to form global traffic situation awareness by analyzing the aggregated data from multiple edge stations, so as to collaboratively manage the traffic signal control strategies configured at all intersections.

8. An urban traffic cloud-edge-device integrated collaborative management and control platform, characterized in that: include: Terminal sensing devices are deployed at various traffic nodes to collect and transmit road information in real time; Terminal execution equipment, used to receive and execute control instructions to guide and adjust traffic flow; Edge stations are used to receive and process data from terminal sensing devices in real time, and send control instructions to terminal control devices to achieve instant response and control of local traffic, thereby reducing network latency and alleviating the load on the central platform; The central platform is used to execute the method as described in any one of claims 1-6 by interacting with the edge station.

9. The urban traffic cloud-edge-terminal integrated collaborative management and control platform as claimed in claim 8 is characterized in that: Terminal sensing devices include: video surveillance equipment, radar sensing equipment and RSU; Terminal execution equipment includes: signal lights, variable lane indicators and guidance screens.

10. The urban traffic cloud-edge-terminal integrated collaborative management and control platform according to claim 8, characterized in that: The edge station is deployed with intelligent network hardware devices that are connected to the edge end downward, and an edge application operating system that provides a two-way interconnected data channel for cloud-edge interaction upward. The edge application operating system includes: The access layer, as a plug-in device driver framework, is used to provide at least one type of device driver, support unified access and control of different devices at the edge station, and dynamically add new device drivers for new devices; The platform layer is configured with multiple data platforms, which are used to meet the general needs of the application layer for storage, query and control by deploying data storage management modules, unified acquisition control modules and device access management modules; The application layer is used to run various edge intelligent applications and provide a unified development interface to support device data query, device control distribution, application status monitoring, and remote deployment and maintenance; The interface layer is used to achieve two-way collaborative interaction between different cloud-edge applications through set network connections, and can also interact with edge intelligent applications running in adjacent edge stations.

Images