Intelligent traffic network flow prediction and congestion relief scheduling system

Abstract

The application relates to the technical field of intelligent traffic control, and discloses a smart traffic network flow prediction and congestion dredging scheduling system, which comprises a data calibration module, a space occlusion processing module, a dynamic counterpressure calculation module, a topology reconstruction module and a signal optimization execution module. The system first calculates the traffic state evaluation value of each intersection, and generates a passing weight mask to suppress the blocked flow direction weight when internal queuing occlusion is detected; secondly, the first derivative of the downstream traffic state is combined to calculate the corrected pressure difference, and the release weight to the congestion node is cut off; at the same time, when a road network deadlock ring is identified, a virtual connection is added to the peripheral low-congestion node to reconstruct the road network calculation graph; finally, the passing weight is calculated by comprehensively considering the above parameters, and the optimal signal instruction is output under the constraint of the green light time. The application effectively avoids invalid release of the intersection, blocks the spread of congestion, and can realize global evacuation under the deadlock state.

Description

Technical Field

[0001] This invention relates to the field of intelligent traffic control technology, specifically to a smart traffic network traffic prediction and congestion management system. Background Technology

[0002] Most current traffic signal control systems rely on cross-sectional traffic flow and space occupancy for adaptive scheduling. When the road network is highly saturated, the physical space inside intersections is limited and easily occupied by queuing vehicles, causing the main straight-through traffic flow to block other flows such as left turns. Existing control algorithms usually cannot accurately perceive this mutual crowding of physical space and still allocate green light time to obstructed flows that cannot actually enter based on statistical traffic flow, resulting in ineffective consumption of signal cycles and waste of resources.

[0003] When coordinating traffic flow at upstream and downstream intersections, conventional methods for calculating traffic pressure often rely on static observation data such as queue length at the current moment, failing to effectively incorporate the dynamic rate of road condition deterioration over time. This approach leaves the system without a mechanism to intervene in advance when overflow is imminent at downstream intersections. Adjustments are only made passively after congestion has fully formed and overflowed back to upstream sections. This lag in control makes it easy for localized congestion to spread along the main road.

[0004] Under extreme traffic demand, queuing vehicles often wait for and block each other at adjacent intersections, forming a circular traffic topology deadlock. In this state, conventional local weight optimization and intersection coordination algorithms become stagnant because they cannot find exit nodes that can release traffic flow. They are unable to actively guide the stranded traffic flow to the surrounding unobstructed road network areas, ultimately causing a large-scale road network paralysis. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an intelligent traffic network flow prediction and congestion management system. It solves the problems of existing traffic signal control systems in the case of high network saturation, which lead to invalid traffic release at intersections and the inability to guide traffic flow to the periphery when traffic topology deadlock occurs.

[0006] To address the above problems, the present invention provides the following technical solution: A smart traffic network traffic flow prediction and congestion mitigation scheduling system includes: The data calibration module is used to acquire traffic flow and space occupancy data collected by roadside detection equipment, and to calculate the traffic state assessment values ​​for each intersection in the road network. The spatial occlusion processing module is used to generate a passage weight reduction mask to suppress the passage demand weight of the obstructed flow direction when queuing overflow occlusion is detected inside the intersection. The dynamic back pressure calculation module is used to calculate the corrected pressure difference between upstream and downstream intersection nodes. The calculation of the corrected pressure difference incorporates the rate of change of the traffic state assessment value of the downstream intersection node. The topology reconstruction module is used to reconstruct the road network computation graph by adding virtual edges between the nodes constituting the traffic deadlock loop and the surrounding low-congestion nodes when a traffic deadlock loop is detected in the road network topology. The signal optimization execution module is used to integrate the traffic weight reduction mask, the corrected pressure difference, and the reconstructed road network calculation map to calculate the instantaneous integrated traffic weight of each signal phase, select the optimal signal phase, and issue control commands based on the green light running time constraint of the signal controller.

[0007] Preferably, when the data calibration module calculates the traffic state assessment value, if no outbound traffic is detected within a continuously set time period and the space occupancy rate exceeds a preset congestion threshold, the integral gain coefficient for calculating the traffic state assessment value is reset to a preset fixed value to eliminate the integral error caused by queuing stagnation.

[0008] Preferably, the spatial obstruction processing module continuously monitors the queue length of the straight-through traffic flow at the intersection. When the queue length extends to the left-turn widening transition section and exceeds the preset physical obstruction threshold, it triggers a weighted calculation of the traffic demand weight for the obstructed left-turn flow.

[0009] Preferably, when the spatial occlusion processing module triggers the weight reduction calculation, it introduces a preset penalty coefficient and substitutes it into a preset attenuation model to calculate the passage weight reduction mask used to suppress the obstructed left-turn flow passage demand weight.

[0010] Preferably, when calculating the corrected pressure difference, the dynamic back pressure calculation module extracts the first derivative of the traffic state assessment value of the downstream intersection node with respect to time to determine the evolution trend of traffic congestion.

[0011] Preferably, when the dynamic back pressure calculation module determines that the first derivative is continuously greater than zero, it performs low-pass filtering on the first derivative to generate a dynamic penalty term, and then superimposes the dynamic penalty term onto the traffic state assessment value of the downstream intersection node to generate comprehensive traffic pressure.

[0012] Preferably, the dynamic back pressure calculation module calculates the initial pressure difference based on the traffic state assessment value of the upstream intersection node and the comprehensive traffic pressure. When the calculated initial pressure difference is less than zero, the corrected pressure difference is truncated to zero to cut off the release weight towards the upstream intersection node.

[0013] Preferably, the topology reconstruction module extracts a congestion sub-graph based on the traffic state assessment values ​​of each intersection node in the road network and according to a preset phase transition threshold, so as to lock the congested areas in the road network.

[0014] Preferably, the topology reconstruction module uses the Tarjan algorithm to identify strongly connected component nodes that constitute mutual blocking in the congestion subgraph, and determines the identified strongly connected component nodes as nodes that constitute the traffic deadlock loop.

[0015] Preferably, when reconstructing the road network calculation graph, the topology reconstruction module searches outward from the traffic deadlock loop within a preset topology distance hop count and locks the intersection nodes whose traffic state assessment value is lower than the preset smooth flow threshold as the low congestion nodes.

[0016] Preferably, the topology reconstruction module inserts virtual edges connecting the nodes of the traffic deadlock loop and the low-congestion nodes into the directed graph adjacency matrix of the road network.

[0017] Preferably, after inserting the virtual connection, the topology reconstruction module uses a preset pressure difference gain value as a guiding weight and maps it to the outward flow direction of the shortest topology path leading to the low-congestion node, so as to form a calculation channel for outward evacuation.

[0018] Preferably, when calculating the comprehensive passage weight, the signal optimization execution module performs discrete cumulative integral calculation on the instantaneous passage weight within a set sliding time window to generate a smooth cumulative weight for each signal phase.

[0019] Preferably, the signal optimization execution module locks the signal phase with the globally largest generated smooth cumulative weight value as the optimal signal phase.

[0020] Preferably, the signal optimization execution module performs the green light running time constraint determination before issuing the control command. When the running green light time of the currently executing phase is less than the preset minimum green light time threshold, the phase sequence switching command of the calculation engine is intercepted and the current phase continues to be allowed to pass.

[0021] Preferably, the signal optimization execution module performs the green light running time constraint determination before issuing the control command. When the running green light time of the currently executing phase reaches the preset maximum green light time threshold, the right of passage of the current phase is forcibly terminated. After executing the yellow light and all-red safe transition time sequence, the module issues a control command to the signal controller to switch to the optimal signal phase.

[0022] This invention provides an intelligent traffic network traffic flow prediction and congestion management system. It has the following beneficial effects: 1. This invention continuously monitors the queue length inside the intersection through a spatial obstruction processing module. When the vehicle queue extends and obstructs other traffic flows, the system generates a traffic weight reduction mask to directly suppress the traffic demand weight of the obstructed flow. This weight adjustment mechanism, which combines physical space constraints, can prevent the system from allocating green light time to directions where the actual physical space is already saturated and vehicles cannot enter, effectively improving the actual utilization rate of the intersection's signal resources.

[0023] 2. This invention introduces the time first derivative of the traffic state assessment value of the downstream intersection node through a dynamic backpressure calculation module, and generates a dynamic penalty term by combining it with filtering. When the calculated pressure difference is less than zero, the system cuts off the release weight to the upstream congested node by correcting the pressure difference to zero. This mechanism enables the control system to perceive congestion trends in advance based on the rate of deterioration of downstream road conditions, realizing feedforward blocking control for overflow risk and preventing congestion from spreading between upstream and downstream intersections.

[0024] 3. This invention utilizes a topology reconstruction module combined with the Tarjan algorithm to identify traffic deadlock loops in the road network and adds virtual connections between deadlocked nodes and peripheral low-congestion nodes. By mapping preset pressure difference gain values ​​to the shortest topology path, the system reconstructs the underlying road network computation graph. This feature overcomes the limitation of traditional scheduling algorithms that cannot continue global optimization when local deadlocks occur, enabling the rapid dispersal of stranded traffic to peripheral unobstructed areas and restoring the overall traffic capacity of the road network. Attached Figure Description

[0025] Figure 1 This is a system framework diagram of the present invention; Figure 2 This is a flowchart of the method of the present invention; Figure 3 This is a schematic diagram of the data mapping and state adaptive calibration principle of the present invention; Figure 4 This is a schematic diagram illustrating the principle of constructing a three-dimensional physical potential energy field and generating a spatial dimension reduction mask in this invention. Figure 5 This is a schematic diagram illustrating the calculation principle of dynamic differential damping and corrected back pressure in this invention. Figure 6 This is a schematic diagram of the topology monitoring and asymmetric computation graph reconstruction principle of the present invention; Figure 7 This is a schematic diagram illustrating the principle of tensor momentum optimization and asynchronous safe execution of the present invention. Figure 8 The following are the timing simulation curves of the intersection node N2 phase tensor momentum integral optimization and green light state transition of the present invention. Among them, (a) is the timing diagram of tensor momentum integral optimization, and (b) is the asynchronous state transition and physical boundary constraint diagram.

[0026] Among them, 10 is the data mapping calibration module; 20 is the potential energy mask module; 30 is the back pressure calculation module; 40 is the topology reconstruction module; and 50 is the optimization execution module. Detailed Implementation

[0027] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0028] See attached document Figure 1 This invention provides a traffic network scheduling system based on a physical potential energy field and a modified back pressure model, comprising: The data mapping calibration module 10 is used to receive traffic flow data and spatial state parameters, perform filtering calibration and state integration, and calculate the number of vehicles in each road segment. The potential energy mask module 20 is used to obtain the vehicle ownership output by the data mapping calibration module 10 and convert it into physical potential energy, calculate the occlusion effect based on the lane geometric position dependency and generate dimension reduction mask parameters. The back pressure calculation module 30 is used to read the physical potential energy and extract the first time derivative to generate a dynamic damping tensor, and to calculate the corrected potential energy difference by combining the physical potential energy and the dynamic damping tensor. The topology reconstruction module 40 is used to extract congested subgraphs and search for vacuum nodes when strongly connected components are detected, inserting virtual connecting edges to rewrite the adjacency weight matrix. The optimization execution module 50 is used to calculate the tensor weights based on the dimensionality reduction mask parameters, the corrected potential difference and the adjacency weight matrix, and integrate them within the time window to obtain the optimal decision target. When the hardware release conditions are met, it drives the intersection execution signal to complete the phase sequence switching.

[0029] The data mapping calibration module 10, potential energy mask module 20, back pressure calculation module 30, topology reconstruction module 40, and optimization execution module 50 are all deployed on the central computing platform. The central computing platform is specifically deployed as a cloud server or edge computing node within the city traffic control center. Roadside detection equipment is deployed on roadside poles along intersections and road segments. Intersection enforcement signals are deployed at each intersection. The central computing platform establishes a data connection with the roadside detection equipment and intersection enforcement signals through a communication network. The system maps the traffic network as a directed graph, where the node set represents the intersection site, and the edge set represents the guide lanes or road segments connecting adjacent intersection sites.

[0030] See attached document Figure 2This invention provides a traffic network scheduling method based on a physical potential energy field and a modified back pressure model, comprising the following steps: S10, the data mapping calibration module 10 receives traffic flow data and spatial state parameters collected by the roadside detection equipment, performs filtering calibration and state integration, and calculates the number of vehicles in each road segment. S20, the potential energy masking module 20 converts the number of vehicles into physical potential energy, calculates the occlusion effect of the dominant flow queuing on the secondary flow based on the lane geometric position dependency, and generates dimensionality reduction masking parameters. S30, the back pressure calculation module 30 extracts the time first derivative of the physical potential energy of the downstream road section to generate a dynamic damping tensor, and combines the physical potential energy and the dynamic damping tensor to calculate the corrected potential energy difference between adjacent nodes. S40, the topology reconstruction module 40 extracts the congested subgraph in real time, searches for topologically adjacent vacuum nodes when a strongly connected component is detected, and inserts virtual connection edges in the bottom directed graph to rewrite the adjacency weight matrix. S50, the optimization execution module 50 calculates the tensor weights of the legal signal phase combination based on the dimensionality reduction mask parameters, the corrected potential difference and the adjacency weight matrix, integrates within the time window to obtain the optimal decision target, and reads the optimal decision target to complete the phase sequence switching after reaching the minimum running time constraint built into the intersection execution signal controller.

[0031] The working principles and technical details of each module and step in the above system will be explained in detail below with reference to specific embodiments.

[0032] See attached document Figure 3 The data mapping calibration module 10 receives traffic flow data and spatial state parameters, and performs filtering calibration and state integration.

[0033] The data mapping calibration module 10 acquires the real-time inbound and outbound traffic flow of each road segment, performs continuous integration calculations according to a preset sampling period, and calculates the prior vehicle inventory.

[0034] In urban traffic scenarios, the physical accumulation of vehicles on road segments is a continuous spatiotemporal evolution process. From a macroscopic fluid perspective, the total number of vehicles in any given road segment is essentially equal to the sum of the initial accumulation and the net inflow (inflow minus outflow) during that period. Based on this physical law, roadside detection equipment collects data from connected nodes. With nodes Real-time inbound traffic flow at the section cross-section With real-time outbound traffic For the collection of real-time inbound and outbound traffic flow, those skilled in the art can use radar point cloud trajectory tracking or video license plate recognition technology. The specific detection and acquisition principles are well-known technologies in this field and will not be elaborated here.

[0035] Data mapping calibration module 10 is based on the vehicle inventory of the previous time step. Calculate the current time Prior vehicle ownership The calculation formula is: ; In the formula, Indicates the current sampling time. This is the integral variable. As an optional implementation, during system cold start or initial moment, the vehicle inventory at the previous time step is used. The number of vehicles stranded on the current road segment can be statically counted using roadside visual inspection equipment to provide an initial baseline for integration calculations. (Preliminary vehicle inventory) This reflects the physical accumulation state of vehicles derived from the difference in traffic flow between the entry and exit sections. In actual industrial environments, the above integral calculation can be equivalently transformed into the cumulative calculation of traffic flow values ​​within a discrete time step.

[0036] The data mapping calibration module 10 introduces the physical space occupancy rate output by the roadside detection equipment as an independent observation variable and constructs an adaptive gain function to calculate the vehicle inventory after smooth calibration.

[0037] Calculations relying on pure open-loop integration often result in accumulated drift errors due to sensor missed or false detections. To establish closed-loop feedback calibration, the data mapping calibration module 10 acquires the physical space occupancy of the multimodal sensor's calculated output. This is to eliminate the aforementioned cumulative drift error. The physical space occupancy rate represents the proportion of the road segment occupied by physical vehicles, and its value is set between 0 and 1.

[0038] Obtain the physical capacity limit of vehicles on the road segment. In this embodiment, the vehicle's physical capacity limit The data mapping calibration module 10 is calculated by multiplying the total physical length of the target road segment by the number of guiding lanes and dividing by the average effective space occupancy length of a single vehicle in a queue (typically 7 to 8 meters). The data mapping calibration module 10 is configured with adaptive gain coefficients. And combined with physical space occupancy Calculate the vehicle inventory after smoothing and calibration. The calculation formula is: ; Adaptive gain coefficient The adjustment weight of physical space occupancy to the integral prior result is controlled. Under normal operating conditions where traffic flow does not reach congestion or clearing extremes, the adaptive gain coefficient... The value is a preset constant less than 1. Experience shows that this constant is usually set within the range of 0.05 to 0.15. This value setting can achieve a balance between trusting external observations and maintaining the continuity of the evolution of the state variables themselves, avoiding large numerical jumps caused by instantaneous sensor noise, thereby ensuring that the state variables used to extract derivatives maintain numerical continuity.

[0039] When the data mapping calibration module 10 detects that a road segment is in an extreme boundary state, it triggers the gain coefficient forced assignment mechanism and performs a hard reset correction of the extreme boundary.

[0040] Extreme boundary states specifically include the absolutely cleared road segment state and the absolutely deadlocked road segment state. When the physical space occupancy rate output by the roadside detection equipment... When the data mapping calibration module 10 determines that the road segment is in an absolutely cleared state, the real-time outflow traffic volume remains zero within a preset continuous time window (e.g., three consecutive time steps). The introduction of a continuous time window effectively shields against false triggers caused by single-frame detection loss. When the real-time outflow traffic volume of the road segment is detected to be continuously zero within the preset continuous time window, and the physical space occupancy rate is... When the deadlock threshold is exceeded, the data mapping calibration module 10 determines that the road segment is in an absolute deadlock state. This deadlock threshold is typically configured to 0.9.

[0041] When any of the above extreme boundary states are triggered, the data mapping calibration module 10 will adjust the adaptive gain coefficient at the current sampling time. Forced to be assigned a value of 1.

[0042] According to the calculation rules of the aforementioned calibration formula, when the adaptive gain coefficient When the value is 1, it includes the prior vehicle population. The algebraic terms are canceled out. The result of the calculation is then... Numerically equivalent to This assignment operation cuts off the transmission of historical integral data, directly uses the real-time physical space occupancy to reset the system's state variables, clears the accumulated integral drift error, and ensures that the system state variables remain consistent with the actual physical environment under extreme operating conditions.

[0043] See attached document Figure 4 After acquiring the underlying traffic data, the potential energy masking module 20 converts the discrete number of vehicles into continuous physical potential energy representing the spatial congestion state, and generates dimensionality-reduced masking parameters by combining the three-dimensional spatial geometric features of the road network. Specifically, this includes the following sub-steps: The potential energy mask module 20 acquires the vehicle inventory output by the data mapping calibration module 10 and calculates the continuously normalized static physical potential energy by combining it with the preset maximum physical capacity limit of the road section.

[0044] Because the physical length and number of lanes vary across different road segments in the urban traffic network, directly using the absolute number of vehicles as the basis for scheduling makes it difficult to uniformly measure the saturation level of different road segments. To establish comparable state indicators across road segments, the potential energy mask module 20 performs dimensionless mapping processing on the smoothed and calibrated vehicle inventory.

[0045] Potential energy mask module 20 extracts the current real vehicle inventory. Divide by the physical capacity limit of the vehicles corresponding to that road section The nodes were calculated. To the node Static physical potential energy in the direction The calculation formula is: ; In this formula, static physical potential energy It is a dimensionless continuous state variable, with values ​​ranging from 0 to 1. The closer the value of this variable is to 1, the more saturated the physical space of the corresponding road segment is. Through this mapping process, the system unifies and abstracts physical road segments of varying sizes into a potential energy field distribution with a consistent scale, providing a standardized data foundation for subsequent higher-order derivative calculations and graph theory analysis.

[0046] The potential energy masking module 20 defines the critical physical threshold for blocking secondary obstructed flows when the main flow overflows due to queuing overflow, based on the physical lane layout structure inside the intersection.

[0047] In actual operation, road network space is constrained by rigid three-dimensional geometric boundaries. Taking a typical intersection containing a main straight-ahead lane and a left-turn widening lane as an example, when the queue length of straight-ahead vehicles extends upstream and covers the starting point of the left-turn widening transition section, vehicles behind that need to turn left will be unable to smoothly enter the left-turn lane due to the physical obstruction of the straight-ahead vehicles in front. The physical lane layout structure and the location of the widening transition section within the intersection can be pre-configured by those skilled in the art through high-precision map data or on-site surveys. The specific methods of acquisition are well-known technologies in this field and will not be elaborated here.

[0048] To incorporate the aforementioned physical constraints into the underlying computational model, the potential energy masking module 20 defines the real-time potential energy of the dominant flow direction as follows: Based on the actual physical distance ratio between the widening transition section and the intersection stop line, a critical physical threshold for occlusion is set. As a preferred approach, if the distance from the starting point of the left-turn widening transition section at an intersection to the stop line accounts for 70% of the total length of that section, then the corresponding critical physical threshold for obstruction for that dominant flow direction is... The value is set to 0.7. This threshold represents the physical starting point where queuing of the dominant flow is about to trigger spatial occlusion.

[0049] When the potential energy of the dominant flow direction crosses the occlusion critical threshold, the potential energy masking module 20 generates a dimension-reduced masking parameter that decays the weight of the secondary flow direction by a non-linear ratio.

[0050] When the system determines that the physical queue length of the dominant flow substantially encroaches on the entry space of the secondary flow, the potential energy mask module 20 triggers the mask generation logic to reduce the weight allocation of the obstructed flow in the global scheduling.

[0051] The dimensionality reduction mask parameters for the secondary flow direction affected by occlusion are set as follows: The formula for calculating the mask parameters by the potential energy mask module 20 is as follows: ; In the formula, This is the penalty coefficient determined by the lane geometry. The value is usually manually calibrated based on the severity of the road segment topology, and typically ranges from 1.0 to 1.5. The formula internally incorporates... Upper bound cutoff function and The lower bound cutoff function is designed to construct strict closed-loop boundary constraints.

[0052] Based on the above logic, when the dominant flow direction is potential energy Less than the critical physical threshold of occlusion This indicates that no physical interference occurred. Although the basic value calculated by the formula is greater than 1, after... After truncation of the function, the dimensionality reduction mask parameters It is forcibly clamped to 1 to ensure that it does not interfere with normal weight calculation. When the potential energy of the dominant flow direction exceeds this critical threshold, the dimensionality reduction mask parameters will decrease non-linearly with the increase of congestion, until they are... The function is truncated to 0. In subsequent calculations, this dimensionality reduction mask parameter is used as a product factor to directly participate in the calculation of the backpressure weights. Through this parallel design of truncation and attenuation, even if secondary flows accumulate huge passage demands, the system will appropriately block their passage permission requests, allowing the underlying pure mathematical optimization model to truly reflect the constraints of the limited three-dimensional physical space.

[0053] See attached document Figure 5 After obtaining the static physical potential energy of each road segment, the back pressure calculation module 30 constructs a damping tensor by combining it with the dynamic change trend over time, and corrects the potential energy difference calculation between adjacent nodes accordingly. Specifically, it includes the following sub-steps: The back pressure calculation module 30 extracts the time first derivative of the physical potential energy of the downstream road section and, in conjunction with low-pass filtering, filters out local high-frequency noise caused by abrupt changes in state.

[0054] Traffic flow evolution has an inherent physical lag, and it is difficult to predict the spread of congestion in advance by relying solely on instantaneous physical space saturation. In order to capture the dynamic evolution characteristics of congestion, in this embodiment, the back pressure calculation module 30 calculates the physical potential energy of the downstream road segment. Discrete differentiation is performed to obtain the time-order first derivative, which characterizes the rate of change in congestion. Specifically, this is based on the system's discrete time step. The back pressure calculation module 30 uses the backward difference method to calculate the approximate value of the first derivative. The calculation formula is as follows: ; In the formula, Indicates the current time node To the node Approximate time first derivative of physical potential energy in the downstream section; Indicates the current sampling time; This indicates the time step in the system's state discretization process; Indicates the current intersection node; Indicates the target downstream intersection node to which the data flows; Indicates the current time , by node Flow to Node The static physical potential energy of the downstream section; Indicates the previous time step , by node Flow to Node The static physical potential energy of the downstream section of the road.

[0055] Because the underlying multi-source data mapping module may trigger a hard reset operation when encountering extreme boundary states, directly using the original derivative calculation will produce a numerical step pulse. To ensure the stability of the computing system, the back-pressure calculation module 30 introduces a low-pass filter function. Smooth the original value of the derivative.

[0056] As a preferred approach, if at the current moment The underlying state update triggers a forced hard reset mechanism. The backpressure calculation module 30 will determine that the derivative is abnormal at that moment and forcibly set the first-order time derivative within the current time period to zero. In the normal state where a hard reset is not triggered, the backpressure calculation module 30 uses a filter based on the exponential moving average algorithm to smooth the continuously sampled derivatives. The smoothing filter coefficient is set to... ,and The range of values ​​is The discrete-time iterative equation of this filter is specifically expressed as: ; In the formula, Indicates the current time The time first derivative is smoothed and output after low-pass filtering; Indicates the current time The original value of the time first derivative obtained from the calculation; This represents the smoothed output result of the low-pass filter from the previous time step. Through proper configuration... With a value of 0.2 (for example, 0.2), the system can effectively filter out transient noise interference and retain the true low-frequency evolution trend.

[0057] The back pressure calculation module 30 generates a dynamic damping tensor characterizing the congestion worsening trend based on the set damping response coefficient and the filtered potential energy derivative.

[0058] After obtaining the smoothed first-order time derivative, the back pressure calculation module 30 calculates the dynamic damping tensor corresponding to the downstream road segment. The calculation formula is: ; In the formula, This represents the damping response coefficient, used to quantify the system's sensitivity to the increasing congestion trend in downstream road sections. Typically, the damping response coefficient... The value range is set between 0.5 and 2.0. The larger the value of this parameter, the stronger the system's penalty for worsening congestion.

[0059] Integrated within the formula The function, acting as a lower bound truncation mechanism, constitutes a strict one-way constraint logic. When the filtered derivative is less than or equal to zero, it means that the queue length of the downstream road segment is either dissipating or remaining constant, and the system does not need to apply additional intervention; the dynamic damping tensor is forcibly truncated to zero. Conversely, when the filtered derivative is greater than zero, it indicates that the queue in the downstream road network is showing a worsening trend, and the system generates a positive dynamic damping tensor based on the absolute magnitude of the derivative.

[0060] The back pressure calculation module 30 injects a dynamic damping tensor into the node potential energy difference model to calculate the corrected potential energy difference of cross-node data flow in order to achieve active overflow prevention control.

[0061] Traditional basic potential energy difference calculations only focus on the absolute difference in the amount of energy held by upstream and downstream road segments at the current moment, lacking a forward-looking response to temporal trends. To achieve feedforward-oriented defensive control, the backpressure calculation module 30 considers the direction of cross-node data flow, i.e., from the upstream node... via the current node Heading to the downstream node of the target The path, calculate the corrected potential energy difference The calculation formula is: ; In the formula, This represents the physical potential energy of the upstream road segment at this moment. This is achieved by using the dynamic damping tensor... Physical potential energy of downstream road sections By directly superimposing these values, the system artificially amplifies the equivalent physical saturation of the downstream road segment at the underlying mathematical model level. Based on this logic, when the physical space of the downstream road segment has not yet reached full saturation but the vehicle backlog speed is too fast, its equivalent potential energy, including the dynamic damping term, will rapidly approach or even exceed the potential energy level of the upstream road segment.

[0062] The corrected potential energy difference obtained at this time The value will decrease significantly or even become negative. In a fundamental physical sense, a negative value indicates that the downstream road segment's ability to receive incoming vehicles from upstream has been completely deprived. As a supporting boundary constraint, the system is configured with a bottom-line truncation function in the subsequent global optimization calculation stage, which will uniformly zero out any correction potential difference less than zero. This differential damping correction mechanism effectively overcomes the structural lag defect of traditional models that require waiting for a physical deadlock in the downstream road segment before passively terminating traffic flow, thus constructing an active defense barrier against the spread of road network congestion in the underlying control model.

[0063] See attached document Figure 6 The topology reconstruction module 40 performs real-time monitoring of the underlying topology of the global road network and guides traffic flow evacuation by dynamically modifying the network adjacency calculation structure when extreme congestion and deadlock are detected. Specifically, it includes the following sub-steps: The topology reconstruction module 40 extracts congested subgraphs from the global road network based on the set congestion phase transition threshold, and uses graph theory algorithms to detect closed-loop strongly connected components in the topology network in real time.

[0064] In macro-level traffic systems, when congestion overflows and interweaves at local intersections, there is a risk of forming a circular queuing structure, i.e., a traffic deadlock. To improve the computational efficiency of the underlying algorithm and focus on key congested areas, the topology reconstruction module 40 traverses all edges in the underlying directed graph within each control cycle. A congestion phase transition threshold is set. As a preferred approach, this congestion phase transition threshold The standard configuration is 0.8. The topology reconfiguration module 40 extracts physical potential energy greater than or equal to the congestion phase transition threshold. The road segments and their associated nodes are used to construct a dynamic congestion sub-graph.

[0065] After extracting the congested subgraph, the topology reconstruction module 40 performs closed-loop strongly connected component detection within the congested subgraph. If a set of nodes exists in the subgraph, and there is a bidirectional directed path between any two nodes in the set, then the set of nodes is determined to constitute a closed-loop strongly connected component, i.e., a deadlock loop in the physical road network is identified. For the detection of strongly connected components in a directed graph, those skilled in the art can use the Tarjan algorithm or the Kosaraju algorithm for calculation. The specific algorithm execution flow is well-known in the art and will not be elaborated here.

[0066] After detecting a closed-loop strongly connected component, the topology reconstruction module 40 limits the network topology distance hop count and searches for vacuum nodes with low physical potential energy in the outer region.

[0067] Conventional local backpressure algorithms can only sense the pressure status of adjacent nodes. Once trapped in a deadlock loop, all adjacent nodes are in a high-pressure saturation state, and the system will lose its dredging direction. To break this structural limitation, the topology reconstruction module 40 starts from the node within the detected deadlock loop that has the conditions to connect to the outer region (i.e., the node has a downstream output road segment not included in the deadlock loop) and expands the path search outward along the road network topology.

[0068] During the search process, the system is configured with the maximum number of topology distance hops. This limits the search range and prevents unreasonable routing with excessively large spans. The topological distance hop count... Typically, it's set to 2 or 3 hops. The topology reconfiguration module 40 evaluates the congestion status of each peripheral node within the limited number of hops. A vacuum physical threshold is set. This threshold is typically set between 0.2 and 0.3, representing a large area of ​​unused physical space on the road segment. When the physical potential energy of all downstream outgoing road segments of a certain peripheral node is lower than this vacuum physical threshold... At that time, the topology reconstruction module 40 determines that the node has sufficient vehicle carrying capacity and marks it as a successfully matched vacuum node.

[0069] As part of the algorithm's completeness design, if the maximum topological distance hop count is set... If no vacuum nodes meeting the above threshold conditions are found, the topology reconstruction module 40 will determine that the surrounding area is completely congested and automatically abandon the virtual edge injection operation. The system will then maintain the original physical network topology to avoid the traffic flow blindly spreading to equally saturated distant areas due to forced reconstruction.

[0070] The topology reconstruction module 40 inserts high-order virtual edges from deadlock nodes to vacuum nodes into the underlying directed graph adjacency matrix in memory, and assigns artificial potential energy differences to rewrite the network topology characteristics.

[0071] After finding a vacuum node with evacuation capabilities, the topology reconstruction module 40 directly intervenes in the structural properties of the underlying graph. In the original physical road network topology, deadlock nodes and vacuum nodes are not directly adjacent in space. The topology reconstruction module 40 forcibly writes a high-order virtual edge starting from the deadlock node and ending at the vacuum node into the adjacency weight matrix maintained by the system computing engine, and maps and superimposes the artificial potential difference of this high-order virtual edge onto the first outgoing flow direction (i.e., the actual physical exit direction of the intersection) on the shortest physical topology path from the deadlock node to the vacuum node obtained by the graph theory shortest path algorithm.

[0072] To ensure that this higher-order virtual edge receives a higher computational weight in subsequent weight optimization, the topology reconstruction module 40 calculates and assigns it an artificial potential energy difference. The calculation formula is: ; In the formula, Indicates the current time Artificial potential energy difference is assigned to higher-order virtual edges; Indicates deadlocked nodes The combined physical potential energy at the current moment; Indicates vacuum node The combined physical potential energy at the current moment; This represents an artificially set enhancement factor used to increase the weight of virtual edges in the computation matrix. Its value is usually set between 1.5 and 3.0.

[0073] In this embodiment, to achieve the conversion of road segment potential energy into node potential energy, deadlock nodes are used. Comprehensive physical potential energy Defined as the maximum physical potential energy of all upstream input road segments converging into this node, to ensure that the node's potential energy accurately reflects the most severe obstruction state of the intersection; similarly, a vacuum node... Comprehensive physical potential energy Defined as the average (or maximum) physical potential energy of all downstream output road segments originating from this node, to truly reflect the actual physical capacity of this node to continue dispersing traffic downstream.

[0074] By injecting high-order virtual edges and amplifying the artificial potential energy difference, and substantially mapping the abstract virtual edge weights to the real physical flow direction corresponding to the deadlock node, the topology reconstruction module 40 constructs an asymmetric computational graph channel that transcends physical spatial barriers on the underlying mathematical model. This channel allows subsequent routing optimization algorithms to directly perceive the negative pressure attraction of the remote vacuum node, thereby increasing the computational tensor weights of the associated physical phases at the control strategy level, scheduling vehicles in the deadlock region to exit towards the vacuum node, and promoting the dissipation of local closed-loop congestion.

[0075] See attached document Figure 7 The optimization and execution module 50 combines the previously generated dimensionality reduction mask parameters with the corrected potential difference, determines the optimal signal phase through integral optimization calculation, and ensures the safe execution of the underlying hardware by relying on an asynchronous state machine. Specifically, it includes the following sub-steps: The optimization and execution module 50 combines the dimension reduction mask parameters and the corrected potential difference to calculate the instantaneous tensor weights of each legal signal phase combination, and performs integration within a preset time window to output the optimal decision target.

[0076] The core of traffic signal control lies in balancing the traffic demand of each flow direction. In this embodiment, the optimization and execution module 50 is configured with a set of signal phases containing all legal combinations of permitted flow directions at the intersection. For any legal signal phase within this set... The calculation engine extracts all traffic flow directions contained in this phase and fuses the dimensionality-reduced mask parameters output by the potential energy masking module with the corrected potential energy difference output by the backpressure calculation module to calculate the current time. Instantaneous tensor weights The specific calculation formula is as follows: ; In the formula, Indicates signal phase The set of traffic flow directions contained within; This indicates a specific single flow direction within the set (i.e., the specific cross-node data flow path corresponding to the aforementioned upstream node, passing through the current node, and heading towards the target downstream node). Indicates the current time Assigned to flow direction The dimensionality reduction mask parameters (i.e., the dimensionality reduction mask parameters calculated in the previous section for secondary flows affected by physical occlusion) ); Indicates the current time Along the flow direction The corrected potential energy difference obtained from the direction calculation (i.e., the cross-node corrected potential energy difference calculated in conjunction with the dynamic damping tensor mentioned above) The formula internally uses... The function forcibly filters out the flow of negative potential energy differences, preventing them from causing negative computational interference to the overall weight of the phase.

[0077] To overcome decision-making oscillations caused by instantaneous high-frequency disturbances, the computing engine operates within a preset time window. The instantaneous tensor weights are integrated internally to obtain a smooth tensor momentum. In the theoretical continuous domain, the calculation formula is: ; In the formula, For integration time variable, This indicates a preset time window, which is usually set within the range of 3 to 5 seconds; This indicates the lower limit of the integration interval, i.e., the start time of the preset time window; Represents the continuous time variable in the integral calculation process; Indicates the time variable The phase of the signal at that moment The corresponding instantaneous tensor weights; Represents an integral infinitesimal element in continuous time.

[0078] Considering the discretization characteristics of the computing power in the actual system, as a preferred approach, the optimization and execution module 50 uses a discrete accumulation method based on the sampling step size to approximate the above integral result. The system follows the discrete time step size... The instantaneous tensor weights are collected, and their corresponding discretization calculation formulas are expanded as follows: ; In the formula, This represents the discrete time step of the system (e.g., 0.5 seconds). Indicates the preset time window The total number of discrete sampling points contained herein satisfies the constraint ; This is the index variable for the sampling points.

[0079] After obtaining the tensor momentum for each phase, the computation engine iterates through and compares the momentum values ​​of all phases at every time step, and selects the phase with the largest tensor momentum. The signal phase extraction is the optimal decision objective at the current moment. Then it is sent down to the underlying execution communication queue.

[0080] The optimization and execution module 50 configures hardware-level time boundary parameters such as minimum green light running time, maximum green light running time, yellow light clearing time, and all-red safety time in the underlying control firmware of the intersection signal controller.

[0081] Traffic signal release and cutoff at the underlying level are constrained by physical safety principles and drivers' physiological reaction time. The optimization and execution module 50 rigidly sets a set of physical time lock parameters in the underlying hardware of the signal control unit at the front-end intersection. Specifically, these time lock parameters include: the minimum green light duration required to ensure vehicles accelerate and smoothly cross the stop line. The maximum green light duration is designed to prevent a single phase from occupying the right-of-way for an extended period during extreme congestion, thus causing related directions to enter an indefinite waiting state. The yellow light clearing time is used to alert drivers that the signal is about to change and to clear vehicles from inside the intersection. ; and a full-red safety time to provide physical isolation during phase conflict switching. For the calibration and specific value settings of these basic safety times, those skilled in the art can perform initial configuration based on the road traffic safety regulations of their respective country or region, for example... The time is usually configured to be 60 to 90 seconds depending on the size of the intersection. The specific engineering calibration method is a well-known technology in this field and will not be elaborated here.

[0082] The optimization and execution module 50 uses an asynchronous state machine architecture to decouple the operation of the computing engine and the execution engine. The execution engine independently determines the underlying state transition by monitoring the comparison status of the current phase runtime and the time lock parameters.

[0083] To avoid the risk of underlying hardware malfunctions due to data congestion or algorithm lag in the computing layer, which is common in traditional centralized control architectures, the execution engine in this embodiment is deployed as an independent asynchronous state machine.

[0084] When the signal is in the green light clear state, the execution engine accumulates the running time of the current signal phase in real time.

[0085] If the running time is less than the minimum green light running time The execution engine triggers a time-locking mechanism, refusing to respond to any new decision targets issued by the computing engine, and maintaining the current light color unchanged.

[0086] When the running time reaches or exceeds the minimum green light running time At that time, the execution engine actively reads the latest optimal decision target from the communication queue. .

[0087] If the decision objective aligns with the current phase that is being released, and the current running time has not exceeded the maximum green light running time. If so, the current green light status will be extended; If it is found that the decision objective has changed, or the current phase's running time has reached the maximum green light running time. Upon encountering a rigid boundary, the execution engine immediately suspends subsequent read requests from the computing engine and autonomously triggers a forced clearing of the phase sequence.

[0088] During the execution cycle of the forced phase sequence clearing, the execution engine strictly follows the yellow light clearing time. Turn on the yellow light, then proceed with the full red safety time. Turn on all traffic lights at the intersection to red. Only after this safe transition period has elapsed will the execution engine proceed according to the decision objective. The green light signal illuminates the new phase. This interactive logic, which decouples high-frequency optimization at the computational layer from low-frequency asynchronous execution at the hardware layer, preserves the algorithm's ability to flexibly adjust weights at the top level while ensuring the physical safety of traffic operations at the intersection through state transition determination at the bottom level.

[0089] To further illustrate the practical engineering application effects and data flow details of the technical solution of this invention, the following provides a specific application example based on a typical urban core area road network morning rush hour congestion evolution and automatic evacuation scenario.

[0090] Specific application examples: In the urban physical road network set in this embodiment, there exists a closed-loop topology consisting of four core intersections (defined as nodes N1, N2, N3, and N4), which face continuous saturation traffic flow input during the morning rush hour. The data mapping calibration module 10 continuously receives cross-sectional traffic flow data transmitted back from the roadside millimeter-wave radar and vision integrated machine with a sampling period of 0.5 seconds.

[0091] Taking the road segment from node N1 to node N2 as an example, the total physical length of this road segment, combined with three guide lanes, determines its physical vehicle capacity limit. The initial estimate is 60 vehicles. During the peak morning traffic flow period, the data mapping calibration module 10 calculates the prior vehicle count as 52 vehicles by discretely integrating the real-time inbound and outbound traffic flows. During this period, if the roadside sensors detect that the real-time outbound traffic flow for this road segment is zero for three consecutive time steps, and the physical space occupancy exceeds the deadlock threshold of 0.9, the data mapping calibration module 10 immediately triggers a hard reset mechanism at the extreme boundary, forcibly assigning the adaptive gain coefficient to 1. This eliminates the sensor's drift caused by queuing and static movement, outputting accurate and normalized static physical potential energy. It is 0.92.

[0092] As the potential energy of local road sections approaches its extreme value, the three-dimensional spatial constraint effect of the intersection becomes apparent. The potential energy masking module 20 detects that the physical queue of straight-through traffic at intersection N1 has overflowed into the left-turn widening transition section. Based on the preset obstruction critical physical threshold of 0.7 and the current straight-through potential energy of 0.85, a penalty coefficient of 1.2 is introduced into the nonlinear attenuation model to calculate the dimensionality-reduced masking parameter of the obstructed left-turn flow to be approximately 0.4. This masking parameter reduces the demand weight of the left-turn phase in the underlying weighting calculation, preventing the system from allocating invalid green light time to left-turn traffic that is actually unable to pass due to physical obstruction.

[0093] Simultaneously, the congestion potential energy propagates upstream along the graph topology. The backpressure calculation module 30 extracts the time first derivative of the physical potential energy downstream of N2 and determines that it is consistently greater than zero. After smoothing by a low-pass filter with a coefficient of 0.2, the system generates a positive dynamic damping tensor and superimposes it on the comprehensive potential energy term of the downstream node. This differential damping mechanism causes the corrected potential energy difference between N1 and N2 to converge rapidly to a negative value and be truncated to zero by the lower bound, thereby triggering feedforward blocking and actively cutting off the flow to the congested node N2.

[0094] Local concurrency in the anti-overflow mechanism leads to a traffic deadlock loop forming between regions N1 and N4. The topology reconstruction module 40 extracts the congested subgraph with a phase transition threshold of 0.8 and identifies it as a strongly connected component using the Tarjan algorithm. To establish evacuation routes, the system initiates an external search within a 2-hop topology constraint, finding an external node N5 with a physical potential energy of 0.25 (below the vacuum threshold of 0.3). Based on this, the computation engine forcibly injects a high-order virtual edge from the deadlock node N2 directly to the vacuum node N5 into the adjacency weight matrix and assigns it an artificial potential energy difference with a magnification factor of 2.5. This potential energy difference gain is directly mapped to the first-hop outflow direction of the shortest physical topology path from N2 to N5, giving node N2 a clear external evacuation optimization weight.

[0095] See attached document Figure 8 , Figure 8 This is a timing simulation curve of the N2 phase tensor momentum integral optimization and green light state transition at an intersection node according to an embodiment of the present invention. It demonstrates the asynchronous interaction characteristics of continuous top-level computation and discrete bottom-level execution.

[0096] After receiving the dimensionality reduction mask and corrected potential difference, the optimization and execution module 50 performs discrete accumulation approximation of instantaneous tensor weights within a 4-second sliding window with a step size of 0.5 seconds. The timing curves show that the phase tensor momentum, which receives the N5 evacuation weight mapping, rapidly climbs to the global optimum. However, at the underlying hardware level, if the green light running time of the currently executing phase is less than the minimum threshold of 15 seconds, the asynchronous state machine will forcibly intercept the phase sequence switching command of the computing engine; after exceeding the lower limit, the state machine autonomously executes the yellow light and all-red safety transition, and then illuminates the green light of the evacuation phase. During the evacuation release cycle, if the phase continues to run and reaches the rigid boundary of the maximum green light time of 80 seconds, the anti-deadlock mechanism of the execution engine will be activated, suspending the hold command of the top-level algorithm and forcibly truncating the right of way. This architecture guides the traffic flow to break the topological deadlock through tensor integral optimization, while relying on the underlying hardware time boundary to ensure the physical safety of the intersection operation.

[0097] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A smart traffic network traffic flow prediction and congestion mitigation scheduling system, characterized in that, include: The data calibration module is used to acquire real-time inbound traffic, real-time outbound traffic, and physical space occupancy rate collected by roadside detection equipment. Based on the real-time inbound traffic and real-time outbound traffic, it calculates the prior vehicle inventory of each road segment and uses the physical space occupancy rate to calibrate the prior vehicle inventory to obtain the vehicle inventory of each road segment. The spatial occlusion processing module is used to calculate the static physical potential energy of each road segment based on the number of vehicles in each segment and the corresponding physical capacity limit of vehicles. When queuing overflow occlusion is detected inside the intersection, a passage weight reduction mask is generated to suppress the passage demand weight of the obstructed flow direction. The dynamic back pressure calculation module is used to generate a dynamic damping tensor based on the rate of change of the static physical potential energy of the downstream road segment with respect to time, use the dynamic damping tensor to correct the static physical potential energy of the downstream road segment, generate comprehensive traffic pressure, and calculate the corrected potential energy difference based on the static physical potential energy of the upstream road segment and the comprehensive traffic pressure. The topology reconstruction module is used to extract congestion sub-graphs based on the static physical potential energy of each road segment. When a strongly connected component constituting a traffic deadlock loop is detected in the congestion sub-graph, a virtual connection is added between the deadlock node in the strongly connected component and the peripheral low-congestion node. An artificial potential energy difference is assigned to the virtual connection, and the artificial potential energy difference is mapped to the traffic flow direction corresponding to the real physical exit direction of the deadlock node in the shortest physical topology path from the deadlock node to the low-congestion node, so as to increase the weight of the signal phase associated with the traffic flow direction. The signal optimization execution module is used to judge the corrected potential energy difference corresponding to each traffic flow direction. When the corrected potential energy difference is less than zero, it is set to zero. The module multiplies the traffic weight reduction mask corresponding to each traffic flow direction with the processed corrected potential energy difference to obtain the flow weight of each traffic flow direction. The module also sums the flow weights of each traffic flow direction included in each legal signal phase to obtain the instantaneous comprehensive traffic weight of each signal phase. The module selects the optimal signal phase and issues control commands based on the green light running time constraint of the signal controller.

2. The intelligent transportation network traffic flow prediction and congestion mitigation scheduling system according to claim 1, characterized in that, The spatial occlusion processing module continuously monitors the queue length of the straight-through traffic flow at the intersection. When the queue length extends to the left-turn widening transition section and exceeds the preset physical occlusion threshold, the module uses a preset attenuation model to calculate the traffic weight reduction mask for the obstructed left-turn flow.

3. The intelligent transportation network traffic flow prediction and congestion mitigation scheduling system according to claim 1, characterized in that, When calculating the corrected potential energy difference, the dynamic back pressure calculation module uses the backward difference method to calculate the first derivative of the static physical potential energy of the downstream road segment with respect to time, based on the discrete time step of the system. The first derivative is then low-pass filtered to obtain a filtered output result. When the filtered output result is greater than zero, a dynamic damping tensor is generated based on the damping response coefficient and the filtered output result to characterize the congestion worsening trend of the downstream road segment. When the filtered output result is less than or equal to zero, the dynamic damping tensor is truncated to zero.

4. The intelligent transportation network traffic flow prediction and congestion mitigation scheduling system according to claim 3, characterized in that, When the corrected potential energy difference is less than zero, the corrected potential energy difference is set to zero to cut off the release weight of traffic flow from the upstream road segment to the downstream road segment via the current intersection node.

5. The intelligent transportation network traffic flow prediction and congestion mitigation scheduling system according to claim 1, characterized in that, The topology reconstruction module extracts road segments and their associated nodes whose static physical potential energy is greater than or equal to a preset congestion phase transition threshold to construct the congestion subgraph, and uses the Tarjan algorithm to identify the strongly connected component nodes that constitute the traffic deadlock loop in the congestion subgraph.

6. The intelligent transportation network traffic flow prediction and congestion mitigation scheduling system according to claim 5, characterized in that, When reconstructing the road network calculation map, the topology reconstruction module searches for peripheral nodes whose static physical potential energy of each downstream outbound road segment is lower than a preset vacuum physical threshold within a preset topology distance hop count, and identifies the peripheral nodes as the low-congestion nodes. A virtual edge connecting the strongly connected component node and the low-congestion node is inserted into the adjacency weight matrix of the directed graph of the road network. The maximum static physical potential energy of all upstream input road segments flowing into the deadlock node is determined as the comprehensive physical potential energy of the deadlock node, and the average static physical potential energy of all downstream output road segments leaving the low-congestion node is determined as the comprehensive physical potential energy of the low-congestion node. The artificial potential energy difference of the virtual edge is determined based on the difference between the comprehensive physical potential energy of the deadlock node and the comprehensive physical potential energy of the low-congestion node and a preset enhancement amplification coefficient.

7. The intelligent transportation network traffic flow prediction and congestion mitigation scheduling system according to claim 1, characterized in that, After calculating the instantaneous integrated passage weight of each signal phase, the signal optimization execution module performs discrete cumulative integration calculation on the instantaneous integrated passage weight within a set sliding time window to generate a smooth cumulative weight for each signal phase, and determines the signal phase with the largest smooth cumulative weight value as the optimal signal phase.

8. The intelligent transportation network traffic flow prediction and congestion mitigation scheduling system according to claim 1, characterized in that, Before issuing control commands, the signal optimization execution module performs the green light running time constraint determination. When the running green light time of the currently executing phase is less than the preset minimum green light time threshold, the phase switching command is intercepted and the current phase continues to allow passage.

9. The intelligent transportation network traffic flow prediction and congestion mitigation scheduling system according to claim 8, characterized in that, Before issuing control commands, the signal optimization execution module performs the green light running time constraint determination. When the running green light time of the currently executing phase reaches the preset maximum green light time threshold, the right of passage of the current phase is forcibly terminated. After executing the yellow light and all-red safe transition time sequence, the module issues a control command to the signal controller to switch to the optimal signal phase.

Images