Method, device, system and program product for optimizing ground traffic signals

Abstract

The application provides an optimization method, device, system and program product of ground traffic signal, the optimization method comprises: obtaining the vehicle flow image collected by the low-altitude unmanned aerial vehicle; using an AI image recognition module to identify the number and position of vehicles in the vehicle flow image, and superimposing the identification result on a road segment structure to generate a vehicle flow space-time heat map; based on the vehicle flow space-time heat map, predicting the vehicle flow density after Δt seconds at a prediction time t i In the prediction time t, the vehicle flow density after Δt seconds is predicted, and when the probability that the vehicle flow density is greater than a predetermined congestion critical density threshold is greater than a predetermined probability threshold, the green wave optimization is started, including: determining the optimal green wave passing speed and the phase difference between adjacent signals; the determined optimal green wave passing speed and phase difference are sent to the signal control unit of the corresponding intersection, and the corresponding regulation and control are performed. By using the above technical scheme, through dynamic modeling, trend prediction and formation of multi-intersection linkage optimization and green wave control closed loop of traffic flow, the optimization of traffic signal control is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of traffic signal control, in particular to a method, device, system and program product for optimizing ground traffic signals. BACKGROUND

[0002] Currently, the control of mainstream urban traffic signals is mostly based on ground fixed perception devices and single-point signal timing algorithms, and the structure generally includes: laying ground magnetic coils, cameras, radars and other perception devices at intersections or main roads; the control center collects local traffic data and inputs it into a fixed cycle model or a feedback-based adaptive timing algorithm; by setting a fixed "green wave speed" (such as 40 km / h), the phase difference between several consecutive traffic lights is adjusted to achieve smooth traffic flow. The representative of this kind of technical solution is SCOOT (Split Cycle Offset Optimization Technique) and SCATS (Sydney Coordinated Adaptive Traffic System) system, whose working principle is to gradually adjust the signal cycle, phase and timing according to the traffic flow changes at the detection point, but its essence relies on static point perception + local optimization algorithm, and has the following limitations:

[0003] (1) Insufficient perception coverage

[0004] Ground magnetic and intersection cameras mainly reflect intersection flow, and are difficult to cover the spatial distribution of the entire road section, and are not robust to camera blind areas, obstructions and bad weather.

[0005] (2) Response cycle lag

[0006] Most control systems execute optimization decisions on a minute-level cycle, and cannot respond to sudden traffic fluctuations on a second-level. For example, when a sudden congestion occurs at B intersection, the signal at C intersection lags behind and cannot form a continuous green wave.

[0007] (3) Lack of overall trend prediction ability

[0008] Current systems are mainly based on instantaneous data or moving average, and cannot predict the evolution trend of traffic flow in the next 1-5 minutes, resulting in passive and lagging timing strategies, especially in dynamic traffic scenarios such as morning and evening peak hours, accidents and exhibitions.

[0009] (4) Signal control and vehicle end are not coordinated

[0010] Traditional green wave control only controls the traffic light itself and does not guide the vehicle to actively cooperate, such as controlling the vehicle speed, so even if the green wave timing is optimal, if the traffic flow cannot run at the specified speed, the accessibility rate will still decrease.

[0011] In summary, although the prior art has made some progress in urban traffic intelligentization, there are still significant deficiencies in cross-road section linkage perception, second-level prediction response, and signal-vehicle cooperative control. SUMMARY

[0012] To address one or more of the above technical problems, embodiments of the present application provide a ground traffic signal optimization method, device, system, and program product.

[0013] To achieve the above-mentioned purpose, in one aspect, a ground traffic signal optimization method is provided, comprising: obtaining vehicle flow images collected by a low-altitude unmanned aerial vehicle pre-deployed on a target road section; using a pre-trained AI image recognition module to identify the number and position of vehicles in the vehicle flow images, and superimposing the identification results on the corresponding road section structure to generate a vehicle flow space-time heat map; the vehicle flow space-time heat map is a two-dimensional image updated in time, the horizontal axis is the number of road sections, and the vertical axis is the time sampling point, each element p(x i ,t) in the two-dimensional image represents the vehicle flow density at time t on road section x i ;

[0014] Based on the generated vehicle flow space-time heat map, a GCN-LSTM model that combines a pre-trained graph neural network and a time series model is used to predict the vehicle flow density p(x i ,t+Δt) on road section x i at a future time t+Δt, and when the predicted vehicle flow density is greater than a predetermined congestion critical density threshold with a probability greater than a predetermined probability threshold, green wave optimization is started;

[0015] The green wave optimization includes:

[0016] determining the optimal green wave speed and the phase difference between adjacent signals;

[0017]

[0018] wherein, is a density weight function used to identify the traffic impact size on road section x i at time t, is an adjustable parameter pre-set according to road conditions and traffic flow characteristics; is the vehicle flow density on road section x i at time t; V opt is the optimal green wave speed; d i is the distance from the starting point to the i-th intersection; v is the candidate speed as an independent variable; argmin vrepresents the value of the function v represents the value of v that minimizes the value of the function Ci represents the signal cycle length of the ith intersection; Ci+1 represents the signal cycle length of the i+1th intersection; Pi represents the current green start phase of the ith intersection; Pi+1 represents the phase difference between the ith intersection and the i+1th intersection; Di represents the distance between the ith intersection and the i+1th intersection;

[0019] The determined optimal green wave speed and the phase difference between adjacent signals are sent to the signal control unit of the corresponding intersection, and one or more of the following adjustments are performed: modifying the green start and end time, dynamically adjusting the cycle length of the signal, adjusting the phase difference between adjacent intersection signals, and starting a preset mode; wherein the preset mode includes: green wave priority or main road traffic priority.

[0020] Preferably, the optimization method of the ground traffic signal, the future predetermined time length is a future short time length not greater than 5 minutes.

[0021] Preferably, the optimization method of the ground traffic signal further comprises:

[0022] When starting the green wave optimization, use the visible light communication transmitter pre-deployed at the predetermined intersection to broadcast the optimal green wave speed to the vehicle in a stroboscopic manner.

[0023] Preferably, the optimization method of the ground traffic signal further comprises:

[0024] The vehicle end receiver receives the information broadcast by the visible light communication transmitter, decodes the received information, and displays the information related to the optimal green wave speed on the central control screen.

[0025] Preferably, the optimization method of the ground traffic signal uses a GCN-LSTM model that combines a pre-trained graph neural network and a time series model to predict the road segment x i The step of predicting the vehicle flow density ρ(x i ,t+Δt) at the future time t+Δt seconds includes:

[0026] The vehicle flow space-time thermal map is used as the input of the graph neural network, and the density correlation between adjacent road segments is extracted by the graph neural network, wherein:

[0027]

[0028] represents the adjacency matrix A of the graph plus the identity matrix I, and represents that each node is also connected to itself; For is a degree matrix, whose diagonal elements are the number of connections of each node; is the negative 1 / 2 power of the degree matrix; is the traffic flow density feature of each road segment at the current time t; is a learnable graph convolution weight matrix; is an activation function; is the spatial feature matrix of time t obtained after the graph convolution operation of the graph neural network;

[0029] Input the w spatial feature matrix sequence from time t-w to time t into the LSTM network, and output the future traffic flow density prediction value of the corresponding road segment.

[0030] Preferably, the ground traffic signal optimization method further comprises the following when the probability that the predicted traffic flow density is greater than the predetermined congestion critical density threshold is greater than the predetermined probability threshold:

[0031] Locate the traffic flow density peak point;

[0032] Take the traffic flow density peak point as a potential bottleneck point, and estimate the propagation speed of the potential bottleneck point at the current time t by the following formula and the future time The position of the bottleneck point :

[0033]

[0034] wherein, represents the position of the potential bottleneck point at time t; represents the position of the potential bottleneck point at time .

[0035] Preferably, the ground traffic signal optimization method further comprises one or more of the following:

[0036] According to the propagation speed of the potential bottleneck point at the current time t , adjust the signal phase difference of the downstream intersection in advance so that the green wave window of the downstream intersection is synchronized with the time when the bottleneck point reaches the downstream intersection;

[0037] For the area where the traffic flow density is greater than the predetermined density threshold, shorten the red light signal period length of the corresponding branch to release the main road passing window;

[0038] When the bottleneck point forms and spreads, the red light signal period length of the related intersection is adjusted in advance.

[0039] Preferably, the optimization method of ground traffic signal further comprises: measuring the deviation between the actual traffic speed and the optimal green wave speed after broadcasting the optimal green wave speed to the vehicle , and adjusting the phase strategy of the next round of traffic lights according to the deviation . In a specific implementation, the deviation is measured by introducing a deviation feedback mechanism in the VLC vehicle end module .

[0040] In another aspect, an optimization device of ground traffic signal is also provided, comprising a memory and a processor, the memory storing at least one program, and the at least one program being executed by the processor to implement the steps of the optimization method of ground traffic signal as described in any of the above aspects.

[0041] In yet another aspect, an optimization system of ground traffic signal is also provided, comprising:

[0042] a low-altitude unmanned aerial vehicle, which is pre-deployed above the target road section to collect traffic images of the target road section;

[0043] the optimization device of ground traffic signal as described above; and

[0044] a visible light communication transmitter, which is pre-deployed at a predetermined intersection to broadcast the optimal green wave speed to the vehicle in a stroboscopic manner when starting the green wave optimization.

[0045] In yet another aspect, a computer program product is also provided, comprising a computer program, wherein the computer program is executed by a processor to implement the steps of the optimization method of ground traffic signal as described in any of the above aspects.

[0046] The above technical solutions have the following technical effects:

[0047] The technical scheme of the embodiment of the application uses an AI image recognition module to recognize the number and position of vehicles in the vehicle flow image collected by the low-altitude unmanned aerial vehicle on the target road section, superimposes the recognition result on the corresponding road section structure, generates a vehicle flow space-time heat map, and uses a GCN-LSTM model pre-trained by fusing a graph neural network and a time sequence model and a vehicle flow space-time heat map sequence data to predict the vehicle flow density of each road section in a future predetermined time length, such as T seconds, and based on the vehicle flow density value, starts green wave optimization, determines the optimal green wave passing speed and the phase difference between adjacent signalized intersections, and issues the determined optimization parameters to each intersection signal control unit to perform corresponding regulation and control, thereby realizing dynamic modeling and trend prediction of traffic flow and supporting multi-intersection linkage optimization and green wave control closed loop formation, and further realizing air-ground collaborative control. By constructing a prediction guidance weight function w(x,t), the priority weight of high-density high-risk road sections in green wave speed calculation can be enhanced, and by adjusting the signal cycle length and the advance preprocessing of the passing window of adjacent intersections, prediction-driven signal collaborative scheduling can be realized; the defects of the existing green wave optimization objective function independent operation and not integrated with the prediction module are overcome.

[0048] In further embodiments, when starting green wave optimization, a visible light communication transmitter pre-deployed at a predetermined intersection is used to broadcast the optimal green wave passing speed to vehicles in a stroboscopic manner, and a vehicle receiver receives the information broadcast by the visible light communication transmitter, decodes the received information, and displays information related to the optimal green wave passing speed on a central control screen, thereby realizing effective closed-loop collaboration between the signal control system and vehicle flow operation through the complete path of unmanned aerial vehicle sensing (air), ground signal light control (ground), and visible light communication vehicle speed guidance (vehicle end), and building an air-ground-vehicle three-in-one collaborative control mechanism.

[0049] In further embodiments, by introducing the bottleneck point propagation speed as a dynamic speed decision reference parameter and predicting the transition trend of vehicle flow density, the phase space is released in advance at the downstream intersection, thereby coordinating the time efficiency difference between green wave scheduling and real-time sensing; the defects of traditional green wave optimization using static speed setting or periodic timing method, which is difficult to adapt to the density changes feedback by real-time monitoring of unmanned aerial vehicles, are overcome.

[0050] In further embodiments, by measuring the offset degree between the actual vehicle flow speed and the recommended speed , adjusting the next round of signal light phase strategy, an integrated closed-loop control system of "prediction-timing-guidance-feedback" is realized; the defects of the traditional signal timing system, which is a one-way issuance structure and cannot sense whether the vehicle end responds to the recommended speed, are overcome.

[0051] The technical scheme of the embodiment of the application solves the mutual exclusion problem caused by independent operation of each module, and through coupling and linkage of density prediction and signal scheduling, a non-obvious systematic invention framework is formed, which provides unprecedented response depth and accuracy for traffic control in a complex dynamic scene. BRIEF DESCRIPTION OF DRAWINGS

[0052] Figure 1 A flowchart of a ground traffic signal optimization method according to an embodiment of the application;

[0053] Figure 2 A structural diagram of a ground traffic signal optimization device according to an embodiment of the application. DETAILED DESCRIPTION

[0054] To further illustrate the embodiments, the application provides accompanying drawings. These drawings are part of the disclosure of the application, mainly used to illustrate the embodiments, and can be used to explain the operating principle of the embodiments in conjunction with the related description of the specification. Those of ordinary skill in the art should be able to understand other possible implementations and advantages of the application by referring to these contents. The components in the drawings are not drawn to scale, and similar component symbols are generally used to represent similar components.

[0055] The application will be further described in conjunction with the accompanying drawings and specific embodiments.

[0056] Embodiment one:

[0057] Figure 1 A flowchart of a ground traffic signal optimization method according to an embodiment of the application. As Figure 1 , the ground traffic signal optimization method of this embodiment includes:

[0058] Obtaining a vehicle flow image collected by a low-altitude unmanned aerial vehicle pre-deployed on a target road section;

[0059] Using a pre-trained AI image recognition module to recognize the number and position of vehicles in the vehicle flow image, and superimposing the recognition result on the corresponding road segment structure to generate a vehicle flow space-time heat map; the vehicle flow space-time heat map is a two-dimensional image that is updated in time rolling, the horizontal axis is the number of road segments, the vertical axis is the time sampling point, and each element ρ(x i ,t) in the two-dimensional image represents the vehicle flow density on the road section x i at time t; the vehicle flow density is the number of vehicles per unit length;

[0060] Based on the generated vehicle flow space-time heat map, using a GCN-LSTM model that combines a pre-trained graph neural network and a time series model to predict the vehicle flow density ρ(x i on the road section x i, and starting the green wave optimization when the predicted traffic density is greater than a predetermined congestion critical density threshold with a probability greater than a predetermined probability threshold; wherein GCN stands for Graph Convolutional Network; and LSTM stands for Long Short-Term Memory;

[0061] The green wave optimization comprises:

[0062] determining the optimal green wave speed and the phase difference between adjacent signals;

[0063]

[0064] wherein w(x i , t) is a function used to identify the road segment x i the density weight function of the traffic impact size at the prediction time t, is an adjustable parameter preset according to road conditions and traffic flow characteristics; is the road segment x i is the growth rate of the traffic density at the prediction time t in the future Δt seconds; V opt is the optimal green wave speed; d i is the distance from the starting point to the ith intersection; v is the candidate speed as an independent variable; argmin v represents the value of v that minimizes the function ; n is the total number of road segments; is the signal cycle length of the ith intersection; is the signal cycle length of the ith+1 intersection; is the current green light starting phase of the ith intersection; is the phase difference between the ith intersection and the ith+1 intersection; is the distance between the ith intersection and the ith+1 intersection;

[0065] sending the determined optimal green wave speed and the phase difference between adjacent signals to the signal control unit of the corresponding intersection, and performing corresponding regulation and control, which comprises one or more of the following regulation and control: modifying the green light start and end time, dynamically adjusting the cycle length of the signal, adjusting the phase difference between adjacent intersection signals, and starting a preset mode; wherein the preset mode comprises: green wave priority or main road priority.

[0066] Preferably, when starting the green wave optimization, a visible light communication transmitter pre-deployed at a predetermined intersection is used to broadcast the optimal green wave speed to vehicles in a stroboscopic manner. In a specific implementation, for example, the front row of vehicles, i.e., the vehicles located at the front, are broadcasted the optimal green wave speed. The vehicle end receiver of the corresponding vehicle receives the information broadcasted by the visible light communication transmitter, decodes the received information, and displays the information related to the optimal green wave speed on the central control screen.

[0067] Embodiment two:

[0068] The ground traffic signal optimization method of the embodiment of the present application realizes real-time green wave optimization based on the dynamicity and time-varying nature of urban arterial traffic flow. The optimization method of the embodiment starts from the five business core stages of "perception → prediction → calculation → execution → feedback", forming a closed-loop control system with real-time response and feedforward control capability.

[0069] The optimization method of the embodiment includes the following steps:

[0070] Step one: low-altitude unmanned aerial vehicle traffic image acquisition

[0071] One or more low-altitude unmanned aerial vehicles are deployed to conduct periodic cruising flights in the air above the target road section. The unmanned aerial vehicle is equipped with a high-definition camera to collect traffic images of the road below at a pre-set time interval, such as 30 seconds, and transmit them back to the edge computing node or cloud platform in real time.

[0072] In a specific implementation, the requirements for the unmanned aerial vehicle are as follows: the cruising altitude is generally controlled at 50-150 meters; the flight path covers the entire arterial road to avoid monitoring blind spots; GPS+vision combined positioning is supported to ensure positioning accuracy.

[0073] Step two: traffic heat map construction and density estimation

[0074] The collected traffic image data is processed by a pre-trained AI image recognition module to automatically identify the number and position of vehicles in the image. In one specific implementation, the AI image recognition module is a target detection network based on YOLOv8. The image recognition result is superimposed on the road segment structure to generate a two-dimensional traffic space-time heat map. The traffic heat map is a two-dimensional image that is updated in time, which can be represented as a two-dimensional matrix. The horizontal axis is the road segment number, and the vertical axis is the time sampling point. Each element in the two-dimensional image, i.e., each matrix element in the two-dimensional matrix, represents the traffic density per unit length of a certain road segment at time t. The heat map visually represents the congestion level through color intensity, and is the input basis for the AI model to predict traffic trends and the key data basis for the signal optimization algorithm to determine bottleneck positions and movement trends.

[0075] The traffic density estimation formula is as follows:

[0076]

[0077] wherein,

[0078] : the vehicle density per unit length of the road segment at time t, with the unit of vehicle number per meter;

[0079] N(x, t): the number of vehicles detected at time t;

[0080] L(x): the length of the road segment x, with the unit of meter.

[0081] Step three: traffic flow short-time prediction model calculation

[0082] Based on the vehicle density heat map sequence generated by the UAV, the future short-time traffic evolution trend within 30-60 seconds is predicted to provide real-time feedforward basis for green wave bandwidth and signal phase control. This step includes the following three associated prediction targets, described in a progressive manner from basic density prediction to event trend identification, and then to key bottleneck point positioning:

[0083] 1. The GCN-LSTM model structure of "graph neural network and time series model fusion" is adopted to perform graph convolution spatial coding on the input historical heat map sequence, and then realize time prediction through LSTM.

[0084] GCN extracts the density correlation between adjacent road segments:

[0085]

[0086] wherein,

[0087] : the adjacency matrix A of the graph plus the identity matrix I, representing that each node is also connected to itself;

[0088] : the degree matrix of, with the diagonal elements being the number of connections of each node, and the number of connections here including the connection to itself;

[0089] : the negative 1 / 2 power of the degree matrix, used for normalization to avoid the weight of nodes with too many connections being too large;

[0090] : the original input features of each road segment at the current time, such as the vehicle density features of each road segment;

[0091] : the learnable graph convolution parameter matrix, such as the graph convolution weight matrix;

[0092] : activation function;

[0093] : spatial feature matrix of time t obtained after graph convolution operation of the graph neural network;

[0094] The significance of the density correlation is that each road segment, i.e., node, not only refers to its own density value, but also absorbs the density information of adjacent road segments on average; this information fusion is automatically propagated according to the connection structure, not simply added, but normalized with weights; this processing ensures that the model has good numerical stability and propagation consistency when extracting the influence of adjacent road segment density, and is suitable for traffic state modeling of irregular road networks. This simulates that "traffic influence is conductive", for example, A segment congestion, B segment will be affected.

[0095] The LSTM input is : represents a sequence of spatial features of continuous w time slices, such as a sequence of graph convolution features of w 30-second sampling intervals from the current time t to w time slices in the past; these features are structures obtained by spatially encoding the traffic density of each road segment by GCN. The sequence of w spatial feature matrices from time t-w to time t is input into the LSTM, and the output is the predicted value of the future traffic density of road segment x .

[0096] The GCN spatially encodes the original density feature (Xt) of each time t by formula , fusing the density correlation of adjacent road segments (such as the influence of A segment congestion on B segment), to generate the spatial feature Ht of this time, while the LSTM takes the sequence of Ht of continuous w times (i.e., Ht-w as input, to capture the time dependence relationship, and realize the prediction of future traffic density , thereby combining the spatial road segment association with the time evolution trend of traffic, to improve the accuracy of prediction.

[0097] The core role of the graph neural network (GCN) is to spatially encode the original traffic density feature of each time slice, to fuse the density correlation of adjacent road segments by formula , to generate the spatial feature Ht of each time t, which already contains the associated influence of the road segment itself and adjacent road segments (such as the potential influence of A segment congestion on B segment), while the sequence of spatial features of continuous w time slices is a sequence composed of multiple Ht in time order, i.e., Ht-w, Ht-w+1,..., Ht-1, Ht. The LSTM model takes this sequence as input, to further learn the traffic evolution law in the time dimension on the basis of capturing spatial association, and finally outputs the prediction value of future traffic density .

[0098] Summary: The output of GCN is the spatial correlation feature of a single time slice, and the spatial feature sequence of continuous w time slices is the temporal combination of multiple GCN outputs, which provides input data with both spatial correlation and temporal continuity for LSTM, and realizes the collaborative modeling of "spatial feature extraction-temporal trend prediction".

[0099] 2. Congestion trend change detection

[0100] On the basis of the predicted density, it is determined whether a critical state transition will occur in the future, i.e., from a smooth state to a congested state. The judgment criterion is: , where is the pre-set congestion critical density threshold, i.e., the congestion critical density threshold. That is, when the predicted value of future traffic density is greater than the pre-set congestion critical density threshold, it is concluded that a critical state transition will occur. Correspondingly, there is a certain probability of the occurrence of this conclusion, i.e., the congestion transition probability .

[0101] Congestion transition probability: , where represents the probability of the event in the parentheses, which is used to quantify the probability of the predicted density exceeding the set congestion critical density threshold , reflecting the risk assessment ability of the AI model.

[0102] Vehicle density growth rate, which is used to measure the congestion growth rate, i.e., to measure the critical trend:

[0103] ;

[0104] These detection results can provide key basis for subsequent green wave passing speed and phase optimization, helping the system to identify potential congestion risks and development trends in advance, so as to adjust the signal timing of downstream intersections in a targeted manner, enhance the forward-looking nature of the control system, avoid signal response lag due to congestion outbreaks, and improve the effectiveness of overall traffic scheduling.

[0105] The calculation relies on the uncertainty distribution of the predicted model output, and the GCN-LSTM model can output the probability distribution (such as the mean and variance of the normal distribution) of the predicted value at the output point at the same time. The proportion of the interval where the "density value is greater than " in the distribution can be calculated by integration, which is the congestion transition probability.

[0106] and Both are prediction results, only different symbols are used in different formula contexts to distinguish them, and the core is the prediction expression of future traffic density.

[0107] 3. Bottleneck point evolution path prediction

[0108] If the congestion trend is significant, the density peak point can be further located as a potential bottleneck point, and its propagation path in space is estimated:

[0109] The propagation speed of the current bottleneck point:

[0110] Wherein, represents the location of the potential bottleneck point at time t; represents the location of the potential bottleneck point at time

[0111] The location of the bottleneck point propagated to at future time These prediction results can be used to adjust or reserve phase space for signal timing at downstream intersections in advance, enhancing the forward-looking nature of the control system.

[0112] First, the phase difference of the downstream signal intersection is adjusted in advance to align the green wave window with the predicted bottleneck "front edge". This is based on the bottleneck point propagation speed

[0113] , which can make the signal timing of the downstream intersection adapt to the moving trend of the bottleneck in advance, avoiding signal conflicts when the bottleneck arrives. Second, dynamically adjust the signal cycle compression range. According to the real-time predicted bottleneck pressure function, shorten the branch red light in the super-high density area to release the main road passing window, form a prediction-driven flexible green wave strategy, and improve the passing efficiency of the main road under the influence of the bottleneck.

[0114] Third, support adaptive regulation in abnormal scenarios. When encountering situations such as accident lane closure and temporary surge in traffic flow, combined with bottleneck point evolution path prediction, the system can realize multi-intersection linkage self-adjustment. For example, when an accident causes the formation and spread of a bottleneck, the related intersection signals are adjusted in advance to guide the traffic to avoid the bottleneck spread area, enhancing the system's response capability to complex dynamic scenarios.

[0115] Step 4: Green wave passing speed and phase optimization algorithm

[0116] Based on the predicted traffic density trend, start the green wave optimization to determine the optimal green wave passing speed

[0117] and the phase difference between adjacent signal intersections To achieve deep integration with AI prediction results, this embodiment of the present invention introduces an enhancement mechanism in the green wave optimization based on the density weight function of the heat map prediction result to improve the influence of the bottleneck or high-density section in the optimization target. The optimization objective function is as follows: ​​​

[0118]

[0119] wherein,

[0120] : distance from the origin to the i-th intersection;

[0121] v: candidate speed, which is a variable;

[0122] : is the road segment traffic density influence weight at prediction time t; the density weight function is dynamically generated by the AI prediction module to measure the influence of each road segment in the green wave travel strategy; its calculation method is as follows:

[0123]

[0124] wherein, is the predicted road segment the unit density of the road segment after seconds in the future; represents the growth rate of traffic density; is a tunable parameter, representing the importance of the current density and growth trend, whose value is pre-set; preferably, the values of all are normalized to the interval [0, 1] for function weighting optimization;

[0125] Considering the complexity of the actual application scenario and the initial stability of the system, will be pre-set. For example, according to different urban road types such as trunk roads, secondary trunk roads, etc., and traffic flow characteristics such as peak hours, off-peak hours, etc., traffic engineers or system developers can determine a set of more appropriate values before system deployment through theoretical analysis, historical data simulation, etc. For example, in the off-peak period of the trunk road with relatively stable traffic flow, the value of can be set larger to highlight the influence of the current density on the green wave travel strategy; while in the peak period, to better cope with congestion changes, the value of can be appropriately increased.

[0126] : the i-th intersection signal cycle length; : the i-th intersection current green light starting phase;

[0127] Calculate the signal phase difference between the i-th intersection and its adjacent i+1-th intersection wherein:

[0128]

[0129] wherein, : the distance between the ith intersection and the i+1th intersection; : the signal cycle length of the i+1th intersection, i.e., the intersection.

[0130] The phase difference between the signals is used to dynamically adjust the cycle and phase of the signal lights, so that the traffic can continuously pass through multiple green lights at a speed

[0131] In a specific implementation, preferably, the phase difference of the downstream signal intersection is adjusted in advance on the basis of the bottleneck propagation speed vjam output by the bottleneck point evolution path prediction module, so that the green wave window is aligned with the predicted bottleneck "front edge".

[0132] The signal cycle compression range will be automatically adjusted according to the real-time predicted bottleneck pressure function, such as shortening the branch red light to release the main road passing window in the super-high-density area, forming a predicted-driven flexible green wave strategy.

[0133] The way of adjusting the phase difference of the downstream signal intersection in advance on the basis of the bottleneck point propagation speed can be combined with the logic deduction of the green wave passing speed and phase optimization: after the system obtains the bottleneck propagation speed vjam through bottleneck point evolution path prediction , it can estimate the time when the bottleneck point reaches each downstream intersection in the future period of time; for example, based on the calculation of the position and time when the bottleneck reaches the downstream intersection . Then, according to this predicted time, the phase difference of the downstream intersection is adjusted in reverse , so that when the critical point when the bottleneck "front edge" or congestion begins to affect the downstream intersection is reached, the green light phase of the intersection is in the open state, thereby passing through the green light window to alleviate congestion.

[0134] Here, "alignment" refers to the synchronization of the green wave window of the downstream intersection, i.e., the time period when the green light is on, with the time when the predicted bottleneck "front edge" reaches the intersection. Specifically, by adjusting the phase difference, the bottleneck "front edge" reaches when the downstream intersection is just in the green light passing period, so that the traffic can quickly pass through the green light window, reduce the queuing backlog caused by the arrival of the bottleneck, realize the active intervention and dredging of congestion, and enhance the forward-looking nature of signal control.

[0135] Step five: ground signal timing linkage execution

[0136] The optimization parameters such as the above-mentioned speed and the above-mentioned phase difference are issued by the central controller or the edge gateway to each intersection signal control unit, and the following regulation and control are performed:

[0137] ​modify the start and end time of the green light; dynamically adjust the signal period; adjust the phase offset value between adjacent signals; start a preset mode, which includes: green wave priority or main road priority.

[0138] The above regulation achieves real-time activation and accessibility control of the green wave band of the main road multi-intersection.

[0139] Step six: vehicle end guidance and traffic regulation

[0140] In a specific implementation, preferably, a visible light communication (VLC) transmitter is deployed at a predetermined intersection, such as the starting intersection of the main traffic flow, and broadcasts the recommended traffic speed in real time to the selected traffic flow, such as the front vehicle , and the specific method includes:

[0141] In a flash mode of the lamp;

[0142] The vehicle end receiver displays the decoded information on the central control screen;

[0143] The recommended traffic speed related information; for example, "recommended speed 55 km / h, enter the green wave band"; the specific words can be flexibly adjusted according to the needs;

[0144] Through this mechanism, the speed stability of the first vehicle can be improved, and the subsequent traffic flow can be guided to enter the optimal traffic rhythm.

[0145] Step seven: closed-loop updating mechanism and optimization feedback

[0146] Maintain a full-process cyclic rolling operation mechanism to repeat the optimization operation in units of a predetermined time, such as 1 minute: new UAV image acquisition; heat map update; re-computation of the prediction model; parameter re-optimization and timing adjustment.

[0147] Preferably, the optimization method of the ground traffic signal of the embodiment further includes: measuring the offset degree between the actual traffic speed and the optimal green wave traffic speed after broadcasting the optimal green wave traffic speed to the vehicle , and adjusting the phase strategy of the next round of signal lights according to the offset degree. In a specific implementation, a deviation feedback mechanism is introduced in the VLC vehicle end module to measure the offset degree .

[0148] The optimization method of the embodiment supports self-adaptation to abnormal scenarios, such as accident lane closure and temporary traffic surge, and has the ability of multi-intersection linkage self-regulation.

[0149] Examples of typical application scenarios of the embodiment

[0150] 1) In a certain city, a certain trunk road is deployed with the method of the embodiment of the application, and in the specific implementation, the method is realized through corresponding program modules:

[0151] The target section includes four road sections A-B-C-D and the corresponding four main intersections, and the early morning peak traffic is concentrated in the B-C section;

[0152] A plurality of unmanned aerial vehicles cruise above the A-D section, and collect road images uploaded every predetermined time length such as 30 seconds;

[0153] Based on image recognition, a heat map is constructed to predict that the density of the C section will break through the congestion threshold within a future predetermined time length such as 60 seconds;

[0154] At the same time, the B section is identified as a bottleneck high point, and its propagation to the C section at a speed of 7 m / s is estimated;

[0155] The green wave optimization module is started, and the recommended green wave passing speed is calculated as 55 km / h;

[0156] Adjust the B→C→D signal phases, for example, adjust them to +3s, +2.5s, and +2s respectively;

[0157] The signal lights are synchronously switched, and the VLC sends the "recommended speed 55 km / h" to the first vehicle;

[0158] The actual measured passing efficiency is improved by 25%, and the green wave penetration rate is increased to 90%.

[0159] 2) In addition to typical urban trunk roads, the optimization method of the embodiment of the application is particularly suitable for the following complex dynamic traffic scenarios:

[0160] High-speed main and auxiliary lane merging section: large speed difference and frequent bottleneck migration, the system can predict the merging pressure change based on the aerial heat map, and actively adjust the signal window to avoid "section out of control";

[0161] Y-shaped intersection or exhibition core area: uneven flow direction and rapid change over time, traditional timing models cannot respond, while the system can quickly switch the timing priority based on short-term prediction;

[0162] Accident or sudden crowd dispersal scene: unmanned aerial vehicles can be quickly deployed to replace the failed ground detection area, and combined with the prediction model, the temporary green wave strategy is switched in real time to improve the dispersal efficiency.

[0163] The optimization method of the embodiment of the application has the following beneficial technical effects:

[0164] (1) Wide sensing range and strong deployment flexibility

[0165] By means of the low-altitude unmanned aerial vehicle cluster, a main road or a multi-intersection area can be covered, so that the blind area problem of the ground fixed monitoring equipment can be effectively made up, and the unmanned deployment and rapid response capability are possessed, and the system is especially suitable for dynamic traffic scenes such as sudden traffic flow, traffic accident and major activity guarantee.

[0166] (2) Dynamic modeling and trend prediction of traffic flow are realized

[0167] The system constructs a traffic flow heat map based on an image recognition technology, and realizes 1-5 minute traffic flow trend prediction by means of a time sequence AI model, so that the system has a feedforward decision capability, the signal response lag is avoided, and the foresight of green wave scheduling is improved.

[0168] (3) Multi-intersection linkage optimization and green wave control closed loop formation are supported

[0169] By means of a green wave speed calculation and phase difference optimization algorithm, the parameters of the multi-intersections are adjusted in linkage, the continuous traffic capacity is significantly improved, and the number of stop times and the intersection queue length are reduced.

[0170] (4) A three-in-one air-ground-vehicle collaborative control mechanism is constructed

[0171] Through the complete path of aerial vehicle sensing in the air, ground signal lamp control on the ground, and visible light communication vehicle speed guidance at the vehicle end, effective closed loop cooperation between the signal control system and the traffic flow is realized, and the'vehicle with signal' mechanism is first realized.

[0172] (5) The system as a whole has real-time, intelligent and self-adaptive capabilities

[0173] The overall optimization process supports short-time, for example, 1 minute level, rolling update, forms a continuous closed loop, and has intelligent scheduling capability, supports dynamic response and self-adjustment to sudden situations, and significantly improves the system robustness and traffic operation efficiency. In the test, it is predicted that the scheme of deploying the embodiment of the application on the main road in the peak period can realize an average traffic speed increase of 12%-25%, and the number of signal stop times is reduced by more than 30%, and has broad popularization and application prospect and industrial transformation value.

[0174] (6) The scheduling robustness is strong in high dynamic complex scenes

[0175] The embodiment of the application is especially suitable for main and auxiliary road interlacing, rapid intersection, traffic disturbance sudden scene, and in a typical 'unbalanced dynamic flow' environment, by means of air-ground collaborative prediction and linkage control, the traffic efficiency can be improved by about 30%, the response time delay is compressed to less than 1 second, and compared with the traditional fixed green wave scheme, the scheduling advantage is obvious.

[0176] Embodiment three

[0177] The present invention also provides an optimization device for ground traffic signals, such as... Figure 2 As shown, the device includes a processor 201, a memory 202, a bus 203, and a computer program stored in the memory 202 and executable on the processor 201. The processor 201 includes one or more processing cores. The memory 202 is connected to the processor 201 via the bus 203. The memory 202 is used to store program instructions. When the processor executes the computer program, it implements the steps in the method embodiment of the above-described method for optimizing ground traffic signals according to Embodiment 1 of the present invention.

[0178] Furthermore, as an executable solution, the optimization device can be a computer unit, which can be implemented as an edge computing device or a cloud device. The computer unit can be a desktop computer, laptop, handheld computer, or cloud server, etc. The computer unit may include, but is not limited to, a processor and memory. Those skilled in the art will understand that the above-described computer unit structure is merely an example and does not constitute a limitation on the computer unit. It may include more or fewer components, or combine certain components, or use different components. For example, the computer unit may also include input / output devices, network access devices, buses, etc., and this embodiment of the invention does not limit this.

[0179] Furthermore, as an executable solution, the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor, etc. The processor is the control center of the computer unit, connecting various parts of the entire computer unit via various interfaces and lines.

[0180] The memory can be used to store the computer program and / or modules, and the processor realizes various functions of the computer unit by running or executing the computer program and / or modules stored in the memory, and calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system and at least one application required by a function; and the data storage area can store data created according to the use of the mobile phone and the like. In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, for example, a hard disk, a memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, at least one disk storage device, a flash memory device, or other volatile solid-state memory devices.

[0181] Embodiment four:

[0182] The application further provides a ground traffic signal optimization system, comprising: a low-altitude unmanned aerial vehicle, which is pre-deployed above a target road section and used to collect traffic flow images of the target road section; the ground traffic signal optimization device as described above; and a visible light communication transmitter, which is pre-deployed at a predetermined intersection and used to broadcast the optimal green wave passing speed to vehicles in a stroboscopic manner when starting green wave optimization.

[0183] Preferably, the ground traffic signal optimization system further comprises a visible light communication receiver arranged at the vehicle end, which is used to receive and decode the signal reflected by the visible light communication transmitter, and then further display the corresponding optimal green wave passing speed information on the center control screen of the vehicle. In a specific implementation, for example, the optimal green wave passing speed is displayed in the form of a text prompt.

[0184] Embodiment five:

[0185] The application further provides a computer program product, comprising a computer program, which realizes the steps in the method embodiments of the ground traffic signal optimization method as described above when executed by a processor.

[0186] Although the application is specifically shown and described in connection with preferred embodiments, those skilled in the art should understand that various changes can be made in form and details without departing from the spirit and scope of the application as defined in the appended claims, and all such changes are intended to be within the protection scope of the application.

Claims

1. A method of optimizing ground transportation signals, characterized by, The method comprises: obtaining vehicle flow images collected by a low-altitude unmanned aerial vehicle pre-deployed on a target road section; The pre-trained AI image recognition module is used to identify the number and position of vehicles in the traffic flow image, and the identification result is superimposed on the corresponding road segment structure to generate a traffic flow space-time heat map; the traffic flow space-time heat map is a two-dimensional image updated in time rolling, the horizontal axis is the number of road segments, the vertical axis is the time sampling point, and each element p(x i ,t) in the two-dimensional image represents the traffic flow density of time t on the road segment x i ; the traffic flow density is the number of vehicles per unit length; Based on the generated traffic space-time thermal map, a GCN-LSTM model fused with a pre-trained graph neural network and a time series model is used to predict the road section x i The traffic density ρ(x i , t+Δt) after Δt seconds at the prediction time t is predicted, and when the probability that the predicted traffic density is greater than the predetermined congestion critical density threshold is greater than the predetermined probability threshold, the green wave optimization is started. wherein the green wave optimization comprises: determining an optimal green wave passing speed and a phase difference between adjacent signals; wherein, is the length of the road segment x i is the density weight function of the traffic impact size at the prediction time t, is an adjustable parameter preset according to road conditions and traffic flow characteristics; is the length of the road segment x i is the growth rate of the traffic flow density after Δt seconds at the prediction time t relative to the traffic flow density at the prediction time t; V opt is the optimal green wave passing speed; d i is the distance from the starting point to the ith intersection; v is the candidate passing speed as the independent variable; argmin v represents the value of v that minimizes the function ; n is the total number of road segments; is the signal light cycle length of the ith intersection; is the signal light cycle length of the ith+1 intersection; is the current green light starting phase of the ith intersection; is the phase difference between the ith intersection and the ith+1 intersection signal; is the distance between the ith intersection and the ith+1 intersection; sending the determined optimal green wave passing speed and the phase difference between adjacent signals to a signal control unit of a corresponding intersection, and performing one or more of the following adjustments: modifying the start and end time of the green light, dynamically adjusting the cycle length of the signal light, adjusting the phase difference between adjacent intersection signals, and starting a preset mode; wherein the preset mode includes: green wave priority or main road passing priority.

2. The method for optimizing ground traffic signals according to claim 1, wherein, The method further comprises: when starting the green wave optimization, using a visible light communication transmitter pre-deployed at a predetermined intersection to broadcast the optimal green wave passing speed to vehicles in a stroboscopic manner.

3. The method for optimizing ground traffic signals according to claim 2, wherein, The method further comprises: a vehicle-side receiver receives information broadcast by the visible light communication transmitter, decodes the received information, and displays information related to the optimal green wave passing speed on a central control screen.

4. The method for optimizing ground transportation signals according to claim 1, wherein, The GCN-LSTM model using a pre-trained graph neural network and a time series model is used to predict the road segment x i The step of predicting the traffic density p(x i ,t+Δt) Δt seconds after the prediction time t includes: The method further comprises: , which represents the adjacency matrix A of the graph plus the identity matrix I, indicating that each node is also connected to itself; is the degree matrix of , the diagonal elements of which are the number of connections of each node; is the negative 1 / 2 power of the degree matrix; is the traffic density feature of each road segment at the current time t; is the learnable graph convolution weight matrix; is the activation function; is the spatial feature matrix of time t obtained after the graph convolution operation of the graph neural network. inputting the w spatial feature matrix sequences from time t-w to time t into an LSTM network to output a future vehicle flow density prediction value for the corresponding road section.

5. The method for optimizing ground transportation signals according to claim 1, wherein, When the predicted vehicle flow density is greater than a predetermined congestion critical density threshold with a probability greater than a predetermined probability threshold, the method further comprises: locating the vehicle flow density peak point. The vehicle flow density peak point is taken as a potential bottleneck point, and the propagation speed of the potential bottleneck point at the current time t is estimated by the following formula and the future time The position of the bottleneck point : wherein, represents the position of the potential bottleneck point at time t; represents the position of the potential bottleneck point at time t.

6. The method for optimizing ground transportation signals according to claim 5, wherein, The method further comprises one or more of the following: According to the propagation speed of the potential bottleneck point at the current time t , the signal phase difference of the downstream intersection is adjusted in advance, so that the green wave window of the downstream intersection is synchronized with the time when the bottleneck point reaches the downstream intersection; for areas where the vehicle flow density is greater than a predetermined density threshold, shortening the red light signal cycle length of the corresponding branch to release the main road passing window; when a bottleneck point is formed and spreads, the red light signal cycle length of the related intersection is adjusted in advance.

7. The method for optimizing ground transportation signals according to claim 2, wherein, After broadcasting the optimal green wave passing speed to vehicles, the method further comprises: by measuring the degree of deviation between the actual traffic speed and the optimal green wave speed and adjusting the phase strategy of the next round of traffic lights according to the degree of deviation ​ 8. An optimization device for ground traffic signals, characterized by a memory and a processor, the memory storing at least one program, and the processor executing the at least one program to implement the steps of the ground traffic signal optimization method according to any one of claims 1 to 7.

9. An optimization system for ground traffic signals, characterized by, The method comprises: a low-altitude unmanned aerial vehicle pre-deployed above a target road section for collecting vehicle flow images of the target road section; the ground traffic signal optimization device according to claim 8; and a visible light communication transmitter pre-deployed at a predetermined intersection for broadcasting the optimal green wave passing speed to vehicles in a stroboscopic manner when starting the green wave optimization.

10. A computer program product comprising a computer program, characterized in that, The computer program is executed by a processor to implement the steps of the ground traffic signal optimization method according to any one of claims 1 to 7.

Images