Data-driven multi-agent adaptive coordinated control method for connected traffic signals

Abstract

The present invention relates to the field of road network signal light control, and in particular to a data-driven networked traffic signal multi-agent adaptive coordinated control method. The method uses a traffic signal machine multi-agent system to network and exchange data on queue length and green light time, and provides real-time online traffic data. The method uses the length of the car queue as a parameter input and the green light time as a parameter output, and realizes queue balance control of multi-directional traffic flow through signal timing, greatly reducing delays and queues caused by the waste of green time in traditional intersection signal control.

Description

Technical Field

[0001] The present invention relates to the field of road network signal light control, and in particular to a road network signal light control method for data-driven multi-agent adaptive coordinated control of networked traffic signals. Background Art

[0002] With the continuous increase in the number of cars, traffic congestion has occurred at intersections in major cities in my country, which has seriously affected the traffic efficiency, safety and driving experience of vehicles. The problem can be effectively solved by optimizing the control of traffic lights at intersections, which has very important practical application significance.

[0003] Prior Art:

[0004] 1. Timing control method: The timing control method is simple, but it cannot adapt to the random changes in traffic flow. When the traffic conditions change significantly or an emergency occurs, it is difficult to respond in time, and the application effect is not good in actual traffic conditions.

[0005] 2. Induction control method: The limitation of induction control is that there is no operation logic in the signal machine, and the road network conditions of the entire intersection are not taken into account. It lacks the integrity and intelligence of traffic control. The actual traffic environment is complex and changeable, and there are many external interference factors.

[0006] 3. Model-free adaptive coordinated control method: This algorithm belongs to closed-loop control and does not rely on any information from the mathematical model. The green light time adjustment strategy is relatively conservative, but this control method still has room for improvement in reducing the number of vehicles queuing at intersections and reducing vehicle delay time.

[0007] 4. Reinforcement learning signal control: Reinforcement learning is a promising data-driven approach to adaptive traffic signal control in complex urban traffic networks, and deep neural networks further enhance its learning capabilities. However, this approach does not take into account the conditions of neighboring intersections and cannot perform global optimization.

[0008] 5. PID coordinated control: PID is one of the most widely used control methods in the field of industrial control. In the field of intelligent transportation systems, PID is classified as a data-driven method, which can realize the control of traffic lights at intersections. However, it is not very effective in reducing the number of vehicles queuing at intersections, reducing the queuing time of vehicles, and reducing pollutant emissions. Summary of the invention

[0009] Aiming at the problem of balancing vehicles in multi-directional variable-cycle queues in large and complex road networks in urban traffic signal control, a data-driven multi-agent adaptive coordinated control method for networked traffic signals is proposed. Through the multi-agent system of traffic signal machines, the network is connected to exchange data on queue length and green light time, providing real-time online traffic data. This method uses the length of the car queue as a parameter input and the green light time as a parameter output, realizing queue balance control of multi-directional traffic flow through signal timing, greatly reducing the delay and queue caused by the waste of green time in traditional intersection signal control. This new control method has three new features: first, a multi-agent system is constructed to realize coordinated control, which is more suitable for urban road networks with randomly changing traffic flows. Second, the new cycle is controlled in real time using only the queue vehicle data collected by the geomagnetic coil; finally, the traffic flow scenes of urban road intersections with different congestion states can be adapted by adjusting the algorithm parameters; compared with the existing methods, the method of this patent is optimal.

[0010] To achieve the above object, the present invention adopts the following technical solution:

[0011] The data-driven multi-agent adaptive coordinated control method for networked traffic signals is characterized by comprising the following steps:

[0012] Step 1: Set the road network area;

[0013] Step 2: Assign initial values ​​to the first cycle green light time of the traffic lights at each intersection of the road network and execute;

[0014] Step 3: The detectors at each intersection of the road network obtain the number of vehicles queuing at each phase of the current cycle;

[0015] Step 4: Calculate the average queue length at each intersection and the maximum queue length at each phase based on the number of queued vehicles obtained;

[0016] Step 5: Adopt the adaptive coordination control algorithm to dynamically and adaptively adjust the green light time of each stage of the next cycle;

[0017] Step 6: The signal light cycle is updated, and the newly calculated green light time is sent to the signal machine, which sends it to the traffic light for execution;

[0018] Step 7: The detectors at each intersection of the road network obtain the number of vehicles queuing at each phase of the current cycle;

[0019] Step 8: Calculate the average queue length at each intersection and the maximum queue length at each phase based on the number of queued vehicles obtained;

[0020] Step 9: Determine whether the number of vehicles queuing at each intersection in the current cycle is balanced, that is, whether the number of vehicles queuing at each phase is equal. If the number of vehicles queuing is balanced, the signal machine will continue to issue the green light time of each phase of the current cycle in the next cycle. Otherwise, the parameters of the adaptive coordination control algorithm are adjusted, and the algorithm is continued to be used to dynamically and adaptively adjust the green light time of each phase of the next cycle.

[0021] Step 10: The signal light cycle is updated, and the newly calculated green light time is sent to the signal machine, which sends it to the traffic light for execution;

[0022] Step 11: Return to step 7 and repeat steps 7 to 10 to achieve the purpose of real-time control of intersection signals.

[0023] In a further optimization of the technical solution, the adaptive coordinated control algorithm includes signal control system modeling,

[0024] To model the road network queue intensity, the expression is defined as (1):

[0025] (1)

[0026] Among them, the subscripts m=1,2,3,4 are the four phases of the intersection. represents the number of queued vehicles in the k+1th period in the mth direction of intersection i, represents the number of queued vehicles in the kth cycle in the mth direction of intersection i, represents the kth variable cycle of intersection i, represents the vehicle arrival rate in the kth period in the mth direction of intersection i, represents the saturated flow in the mth direction of intersection i, represents the green light time of the kth cycle in the mth direction of intersection i. The green light time and signal light cycle of each intersection satisfy the constraint of formula (2):

[0027] (2)

[0028] Among them, the subscripts m=1,2,3,4 are the four phases of the intersection. represents the kth variable cycle of intersection i, represents the green light time of the mth direction at intersection i, It indicates the total lost time at any intersection including the yellow light time state;

[0029] The queue intensity of the i-th intersection in the k-th cycle is defined as (3):

[0030] (3)

[0031] in, represents the queue intensity at the mth entrance line of intersection i in the kth cycle, represents the number of queued vehicles in the kth cycle in the mth direction of intersection i, represents the vehicle storage capacity at the mth entrance line of intersection i;

[0032] (1) Divide both sides of the formula , then the variable period queue intensity model of the road network can be obtained as formula (4):

[0033] (4)

[0034] in, represents the queue intensity at the mth entrance line of intersection i in the k+1th cycle, represents the queue intensity at the mth entrance line of intersection i in the kth cycle, represents the kth variable cycle of intersection i, represents the vehicle arrival rate in the kth period in the mth direction of intersection i, represents the vehicle storage capacity at the mth entrance line of intersection i, represents the saturated flow in the mth direction of intersection i, Indicates the green light time in the mth direction of intersection i.

[0035] In order to avoid wasting green light time, a time-varying control signal cycle is introduced. and the new green time constraint makes the queue intensity satisfy Therefore, the green light time and signal period under the constraint conditions are as follows:

[0036] (5)

[0037] (6)

[0038] in, represents the green light time of the kth cycle in the mth direction of intersection i under the constraint conditions, represents the green light time of the mth direction at intersection i, represents the number of queued vehicles in the kth cycle in the mth direction of intersection i, represents the kth variable cycle of intersection i, represents the vehicle arrival rate in the kth period in the mth direction of intersection i, represents the saturated flow in the mth direction of intersection i, is the maximum green time, is the minimum green light time, Indicates the total lost time at any intersection including the yellow light time status.

[0039] Therefore, equation (4) can be rewritten as equation (7):

[0040] (7)

[0041] in, represents the queue intensity at the mth entrance line of intersection i in the k+1th cycle, represents the queue intensity at the mth entrance line of intersection i in the kth cycle, represents the kth variable cycle of intersection i, represents the vehicle arrival rate in the kth period in the mth direction of intersection i, represents the saturated flow in the mth direction of intersection i, is the green time under the constraint condition, represents the vehicle storage capacity at the mth entrance line of intersection i, To increase the queue intensity, To reduce the queue intensity.

[0042] The technical solution is further optimized, and the coordination control problem is as follows:

[0043] When the queues at each intersection of the road network are balanced, the expected queue intensity can be expressed as formula (8):

[0044] (8)

[0045] in, is the average queue strength in the four directions in the k+1th cycle, is the average queue strength in the four directions in the kth cycle, and is the average increase or decrease of the queue strength in the kth signal cycle; N is the number of intersections, M=4 represents the four directions of the intersection, represents the queue intensity at the mth entrance line of intersection i in the kth cycle, represents the kth variable cycle of intersection i, represents the vehicle storage capacity at the hth entrance line of intersection i, represents the vehicle arrival rate in the kth period in the hth direction of intersection i, represents the saturated flow in the hth direction of intersection i, is the green light time under the constraint.

[0046] Taking the queue intensity in each direction of the intersection as an intelligent agent, which also includes the information of adjacent intersections, the coordination error of the coordination consistency in the mth direction of intersection i is formula (9):

[0047] (9)

[0048] in, represents the mth internal coordination direction of intersection i, represents the elements in the internal adjacency matrix, A = [ a mn ] , m , n ∈ [ 1 , M ] represents the internal coordination adjacency matrix in the single-intersection signal controller network topology, represents the queue intensity at the nth entrance line of intersection i, represents the queue intensity at the mth entrance line of intersection i, represents the mth direction of the external coordination of intersection i to intersection j, represents the elements in the external adjacency matrix, B = [ b mn ] , m , n ∈ [ 1 , M ] represents the external coordination adjacency matrix in the single-intersection signal controller network topology, represents the queue intensity at the nth entrance line of intersection j, represents the connection coefficient between the queue intensity in the m direction of intersection i and the expected queue intensity, It represents the expected queue intensity, that is, the average queue intensity.

[0049] Define the internal coordination in-degree matrix of the network controller agent: , C = diagnosis { c i , j }, j ∈ [ 1 , M ]

[0050] in, represents the mth internal coordination direction of intersection i, and represents the elements in the adjacency matrix, and is the row sum of the adjacency matrix, and is the in-degree matrix, M=4, C is the diagonal matrix consisting of the connection coefficients of the actual queue intensity and the expected queue intensity, is the connection coefficient.

[0051] The Laplace matrix in the signal controller network topology is ;

[0052] The coordinated consistency coordination error can be further expressed as formula (10):

[0053] (10)

[0054] Where A is the internal coordination adjacency matrix in the single-intersection signal controller network topology, and is the in-degree matrix, B represents the external coordination adjacency matrix in the single-intersection signal controller network topology, C is the diagonal matrix consisting of the connection coefficients of the actual queue intensity and the expected queue intensity, is the queue intensity at intersection i, is the queue intensity at intersection j, It represents the expected queue intensity, that is, the average queue intensity.

[0055] Through the Laplace matrix transformation, it can be transformed into formula (11):

[0056] (11)

[0057] in, and is the Laplace matrix, is the in-degree matrix, C is the diagonal matrix consisting of the connection coefficients of the actual queue intensity and the expected queue intensity, is the queue intensity at intersection i, is the queue intensity at intersection j, It represents the expected queue intensity, that is, the average queue intensity.

[0058] definition And satisfy , then formula (11) can be further written as formula (12) and formula (13):

[0059] (12)

[0060] (13)

[0061] in, and is the coordinated consistency coordination error, and is the Laplace matrix, is the in-degree matrix, C is the diagonal matrix consisting of the connection coefficients of the actual queue intensity and the expected queue intensity, is the queue intensity at intersection i, is the queue intensity at intersection j, It represents the expected queue intensity, that is, the average queue intensity.

[0062] Unknown nonlinear parameterized function and Satisfying formula (14)

[0063] (14)

[0064] in, represents the kth variable cycle of intersection i, represents the vehicle storage capacity at the mth entrance line of intersection i, represents the vehicle arrival rate in the kth period in the mth direction of intersection i, represents the vehicle arrival rate in the kth period in the hth direction of intersection i, represents the vehicle storage capacity at the hth entrance line of intersection i, Represents the mth coordinated direction of intersection i.

[0065] Combined with parameter separation technology, equation (14) satisfies the following inequality conditions

[0066] (15)

[0067] in, and is an unknown nonlinear parameterized function, Representative and vehicle arrival rate Related unknown functions, represents the coordination error of the coordination formula of the mth direction of intersection i, Representation and signal period Related known functions that satisfy , through formula (14), select ,

[0068] The global road network, the expected queuing intensity, and the consistency coordination error are described as follows:

[0069] (16)

[0070] (17)

[0071] (18)

[0072] in, and are the deviations between the actual queue intensity and the expected queue intensity at time k+1 and time k, respectively. , is the difference between the actual value and the expected value (average value) of the queue strength increase or decrease in the kth signal cycle, To increase the queue intensity, is the average increase in queue strength in the kth signal cycle, To reduce the queue intensity, is the average reduction in queue strength in the kth signal cycle.

[0073] The final form of the coordinated consistency coordination error considering multiple adjacent intersections is:

[0074] (19)

[0075] in, is the coordinated consistency coordination error, and is the Laplace matrix, is the in-degree matrix, C is the diagonal matrix consisting of the connection coefficients of the actual queue intensity and the expected queue intensity, and is the deviation between the actual queue intensity and the expected queue intensity at time k, , It is the difference between the actual value and the expected value of the queue strength increase or decrease in the kth signal cycle.

[0076] The technical solution is further optimized, and the controller is designed as follows:

[0077] Considering the queuing model (7), the multi-agent single intersection data-driven queuing coordination control strategy is constructed as follows:

[0078] u i , m ( k ) = p i , m ( p i , m + m ) [ u i , m ( k − 1 ) − v i , m ( k ) ] (20)

[0079] in, represents an arbitrary control signal designed by the Lyapunov technique, For the set parameters, To coordinate the control gain, is the green light time of the mth direction at intersection i in the k-1th cycle,

[0080] Considering the multi-agent queuing intensity system (7) and (8) in the traffic network, the coordinated adaptive coordination control term Select formula (21):

[0081] v i , m = c [ ∑ n ∈ N m i a mn i ( e i , m − e i , n ) + ∑ n ∈ N m j b mn i ( e i , m − e j , n ) + c i , m e i , m ] + f ¯ i ^ i , m ( k ) e i , m ( k ) (twenty one)

[0082] The data-driven coordinated adaptive coordinated control strategy for regional signal timing is designed as follows:

[0083] (twenty two)

[0084] The parameter learning law is designed as follows:

[0085] i ^ i , m ( k ) = ∑ i = 1 N 1 1 + Mrs. i , m [ i ^ i , m ( k − 1 ) + f ¯ p i , m α i , m e i , m ( ∑ n ∈ N m i a mn i ( e i , m − e i , n ) + ∑ n ∈ N m j b mn i ( e i , m − e j , n ) + c i , m e i , m ) ] (twenty three)

[0086] in, is the parameter learning law of the m-th direction of intersection i in the k-th cycle, is the green light time of the mth direction at intersection i in the kth cycle, To coordinate the control gain, For the set parameters, is the green light time of the mth direction at intersection i in the k-1th cycle, As parameters, represents the mth internal coordination direction of intersection i, represents the elements in the internal adjacency matrix, represents the coordinated consistency coordination error of the queue in direction m at intersection i, represents the coordinated consistency coordination error of the n-direction queue at intersection i, represents the mth external coordination direction of intersection i belonging to intersection j, represents the elements in the external adjacency matrix, represents the coordinated consistency coordination error of the queue in n directions at the j intersection, is the connection coefficient between the m-direction queue at intersection i and the expected queue, is a known parameter related to the signal cycle, N is all the intersections in the road network, As parameters, As parameters, is the parameter learning law for the m-th direction of intersection i in the k-1th cycle.

[0087] Different from the prior art, the above technical solution has the following beneficial effects:

[0088] 1. This method does not require traffic model information and can achieve adaptive control by only using information such as the number of queued vehicles and green light time.

[0089] 2. Compared with other methods, this method can effectively avoid wasting green light time, reduce the number of vehicles in queue, balance the queue length at the intersection, and improve road traffic efficiency.

[0090] 3. The automobile exhaust emissions of this method are less than those of other methods, which reduces environmental pollution and promotes energy conservation and emission reduction.

[0091] 4. This method has less delay time and is superior to other methods, and can alleviate traffic congestion problems to a great extent.

[0092] 5. By adjusting the algorithm parameters, it can adapt to traffic flow scenarios at urban road intersections with different congestion conditions.

[0093] 6. This method can simultaneously take into account the traffic status information of its own intersection and neighboring intersections in the road network structure, and can perform global optimization. BRIEF DESCRIPTION OF THE DRAWINGS

[0094] Figure 1 Flowchart of the data-driven multi-agent adaptive coordinated control method for connected traffic signals;

[0095] Figure 2 The topological structure diagram of the present invention and its corresponding adjacency matrix;

[0096] Figure 3 A scene diagram of an intersection according to an embodiment of the present invention;

[0097] Figure 4 4 is a traffic phase diagram of an embodiment of the present invention. DETAILED DESCRIPTION

[0098] In order to explain the technical content, structural features, achieved objectives and effects of the technical solution in detail, the following is a detailed description in conjunction with specific embodiments and accompanying drawings.

[0099] See also Figure 1 As shown, a flow chart of a data-driven multi-agent adaptive coordinated control method for networked traffic signals according to a preferred embodiment of the present invention includes the following steps:

[0100] Step 1: Set the road network area. The road network is a road system composed of multiple roads that are interconnected and interwoven into a network within a certain area.

[0101] Step 2: Assign values ​​to the green light time of each stage in the first cycle of the traffic lights at each intersection in the road network, for example =40s, indicating that the green light time in each stage is 40s, and the cycle is executed.

[0102] Step 3: The detectors at each intersection of the road network obtain the number of queued vehicles at each phase of the current cycle (going straight and turning right are considered to be the same phase).

[0103] Step 4: After the first cycle is completed, the average queue length at each intersection and the maximum queue length at each phase are calculated based on the number of queued vehicles obtained.

[0104] Step 5: Take the number of vehicles in the queue in the previous cycle as input, use the adaptive coordinated control algorithm to calculate the parameter learning rate, and bring the parameter learning rate into The green light time of each stage in the next cycle can be calculated and output.

[0105] The adaptive coordinated control algorithm is as follows:

[0106] Step 5.1: Signal control system modeling:

[0107] To model the road network queue intensity, the expression is defined as (1):

[0108] (1)

[0109] Among them, the subscripts m=1,2,3,4 are the four phases of the intersection. represents the number of queued vehicles in the k+1th period in the mth direction of intersection i, represents the number of queued vehicles in the kth cycle in the mth direction of intersection i, represents the kth variable cycle of intersection i, represents the vehicle arrival rate in the kth period in the mth direction of intersection i, represents the saturated flow in the mth direction of intersection i, It represents the green light time of the kth cycle in the mth direction of intersection i.

[0110] The green light time and signal light cycle of each intersection satisfy the constraints of formula (2):

[0111] (2)

[0112] Among them, the subscripts m=1,2,3,4 are the four phases of the intersection. represents the kth variable cycle of intersection i, represents the green light time of the mth direction at intersection i, It indicates the total lost time at any intersection including the yellow light time state;

[0113] The queue intensity of the i-th intersection in the k-th cycle is (3):

[0114] (3)

[0115] in, represents the queue intensity at the mth entrance line of intersection i in the kth cycle, represents the number of queued vehicles in the kth cycle in the mth direction of intersection i, represents the vehicle storage capacity at the mth entrance line of intersection i;

[0116] (1) Divide both sides of the formula , then the variable period queue intensity model of the road network can be obtained as formula (4):

[0117] (4)

[0118] in, represents the queue intensity at the mth entrance line of intersection i in the k+1th cycle, represents the queue intensity at the mth entrance line of intersection i in the kth cycle, represents the kth variable cycle of intersection i, represents the vehicle arrival rate in the kth period in the mth direction of intersection i, represents the vehicle storage capacity at the mth entrance line of intersection i, represents the saturated flow in the mth direction of intersection i, Indicates the green light time in the mth direction of intersection i.

[0119] In order to avoid wasting green light time, a time-varying control signal cycle is introduced. and the new green time constraint makes the queue intensity satisfy Therefore, the green light time and signal period under the constraint conditions are as follows: (5)

[0120] (6)

[0121] in, represents the green light time of the kth cycle in the mth direction of intersection i under the constraint conditions, represents the green light time of the mth direction at intersection i, represents the number of queued vehicles in the kth cycle in the mth direction of intersection i, represents the kth variable cycle of intersection i, represents the vehicle arrival rate in the kth period in the mth direction of intersection i, represents the saturated flow in the mth direction of intersection i, is the maximum green time, is the minimum green light time, Indicates the total lost time at any intersection including the yellow light time status.

[0122] Therefore, equation (4) can be rewritten as equation (7):

[0123] (7)

[0124] in, represents the queue intensity at the mth entrance line of intersection i in the k+1th cycle, represents the queue intensity at the mth entrance line of intersection i in the kth cycle, represents the kth variable cycle of intersection i, represents the vehicle arrival rate in the kth period in the mth direction of intersection i, represents the saturated flow in the mth direction of intersection i, is the green time under the constraint condition, represents the vehicle storage capacity at the mth entrance line of intersection i, To increase the queue intensity, To reduce the queue intensity.

[0125] Step 5.2: Describe the coordination control problem:

[0126] When the queues at each intersection of the road network are balanced, the expected queue intensity can be expressed as formula (8):

[0127] (8)

[0128] in, is the average queue strength in the four directions in the k+1th cycle, is the average queue strength in the four directions in the kth cycle, and is the average increase or decrease of the queue strength in the kth signal cycle; N is the number of intersections, M=4 represents the four directions of the intersection, represents the queue intensity at the mth entrance line of intersection i in the kth cycle, represents the kth variable cycle of intersection i, represents the vehicle storage capacity at the hth entrance line of intersection i, represents the vehicle arrival rate in the kth period in the hth direction of intersection i, represents the saturated flow in the hth direction of intersection i, is the green light time under the constraint.

[0129] The queue strength in each direction of the intersection is regarded as an intelligent entity, which also includes the information of adjacent intersections. Then the coordination error of the coordination consistency in the mth direction of intersection i is formula (9):

[0130] (9)

[0131] in, represents the mth internal coordination direction of intersection i, represents the elements in the internal adjacency matrix, A = [ a mn ] , m , n ∈ [ 1 , M ] represents the internal coordination adjacency matrix in the single-intersection signal controller network topology, represents the queue intensity at the nth entrance line of intersection i, represents the queue intensity at the mth entrance line of intersection i, represents the mth direction of the external coordination of intersection i to intersection j, represents the elements in the external adjacency matrix, B = [ b mn ] , m , n ∈ [ 1 , M ] represents the external coordination adjacency matrix in the single-intersection signal controller network topology, represents the queue intensity at the nth entrance line of intersection j, represents the connection coefficient between the queue intensity in the m direction of intersection i and the expected queue intensity, It represents the expected queue intensity, that is, the average queue intensity.

[0132] Define the internal coordination in-degree matrix of the network controller agent: , C = diagnosis { c i , j }, j ∈ [ 1 , M ]

[0133] in, represents the mth internal coordination direction of intersection i, and represents the elements in the adjacency matrix, and is the row sum of the adjacency matrix, and is the in-degree matrix, M=4, C is the diagonal matrix consisting of the connection coefficients of the actual queue intensity and the expected queue intensity, is the connection coefficient

[0134] The Laplace matrix in the signal controller network topology is .

[0135] The coordinated consistency coordination error can be further expressed as formula (10):

[0136] (10)

[0137] Where A is the internal coordination adjacency matrix in the single-intersection signal controller network topology, and is the in-degree matrix, B represents the external coordination adjacency matrix in the single-intersection signal controller network topology, C is the diagonal matrix consisting of the connection coefficients of the actual queue intensity and the expected queue intensity, is the queue intensity at intersection i, is the queue intensity at intersection j, It represents the expected queue intensity, that is, the average queue intensity.

[0138] Through the Laplace matrix transformation, it can be transformed into formula (11):

[0139] (11)

[0140] in, and is the Laplace matrix, is the in-degree matrix, C is the diagonal matrix consisting of the connection coefficients of the actual queue intensity and the expected queue intensity, is the queue intensity at intersection i, is the queue intensity at intersection j, It represents the expected queue intensity, that is, the average queue intensity.

[0141] definition And satisfy , then formula (11) can be further written as formula (12) and formula (13):

[0142] (12)

[0143] (13)

[0144] in, and is the coordinated consistency coordination error, and is the Laplace matrix, is the in-degree matrix, C is the diagonal matrix consisting of the connection coefficients of the actual queue intensity and the expected queue intensity, is the queue intensity at intersection i, is the queue intensity at intersection j, It represents the expected queue intensity, that is, the average queue intensity.

[0145] Unknown nonlinear parameterized function and Satisfying formula (14)

[0146] (14)

[0147] in, represents the kth variable cycle of intersection i, represents the vehicle storage capacity at the mth entrance line of intersection i, represents the vehicle arrival rate in the kth period in the mth direction of intersection i, represents the vehicle arrival rate in the kth period in the hth direction of intersection i, represents the vehicle storage capacity at the hth entrance line of intersection i, Represents the mth coordinated direction of intersection i.

[0148] Combined with parameter separation technology, equation (14) satisfies the following inequality conditions

[0149] (15)

[0150] in, and is an unknown nonlinear parameterized function, Representative and vehicle arrival rate Related unknown functions, represents the coordination error of the coordination formula of the mth direction of intersection i, Representation and signal period Related known functions that satisfy , through formula (14), select ,

[0151] The global road network, the expected queuing intensity, and the consistency coordination error are described as follows:

[0152] (16)

[0153] (17)

[0154] (18)

[0155] in, and are the deviations between the actual queue intensity and the expected queue intensity at time k+1 and time k, respectively. , is the difference between the actual value and the expected value (average value) of the queue strength increase or decrease in the kth signal cycle, To increase the queue intensity, is the average increase in queue strength in the kth signal cycle, To reduce the queue intensity, is the average reduction in queue strength in the kth signal cycle.

[0156] The final form of the coordinated consistency coordination error considering multiple adjacent intersections is:

[0157] (19)

[0158] in, is the coordinated consistency coordination error, and is the Laplace matrix, is the in-degree matrix, C is the diagonal matrix consisting of the connection coefficients of the actual queue intensity and the expected queue intensity, and is the deviation between the actual queue intensity and the expected queue intensity at time k, , It is the difference between the actual value and the expected value of the queue strength increase or decrease in the kth signal cycle.

[0159] Step 5.3: Controller Design:

[0160] Considering the queuing model (7), the multi-agent single intersection data-driven queuing coordination control strategy is constructed as follows:

[0161] u i , m ( k ) = p i , m ( p i , m + m ) [ u i , m ( k − 1 ) − v i , m ( k ) ] (20)

[0162] in, represents an arbitrary control signal designed by the Lyapunov technique, For the set parameters, To coordinate the control gain, is the green light time of the mth direction at intersection i in the k-1th cycle,

[0163] Considering the multi-agent queuing intensity system (7) and (8) in the traffic network, the coordinated adaptive coordination control term Select formula (21):

[0164] v i , m = c [ ∑ n ∈ N m i a mn i ( e i , m − e i , n ) + ∑ n ∈ N m j b mn i ( e i , m − e j , n ) + c i , m e i , m ] + f ¯ i ^ i , m ( k ) e i , m ( k ) (twenty one)

[0165] The data-driven coordinated adaptive coordinated control strategy for regional signal timing is designed as follows:

[0166] (twenty two)

[0167] The parameter learning law is designed as follows:

[0168] i ^ i , m ( k ) = ∑ i = 1 N 1 1 + Mrs. i , m [ i ^ i , m ( k − 1 ) + f ¯ p i , m α i , m e i , m ( ∑ n ∈ N m i a mn i ( e i , m − e i , n ) + ∑ n ∈ N m j b mn i ( e i , m − e j , n ) + c i , m e i , m ) ] (twenty three)

[0169] in, is the parameter learning law of the m-th direction of intersection i in the k-th cycle, is the green light time of the mth direction at intersection i in the kth cycle, To coordinate the control gain, For the set parameters, is the green light time of the mth direction at intersection i in the k-1th cycle, As parameters, represents the mth internal coordination direction of intersection i, represents the elements in the internal adjacency matrix, represents the coordinated consistency coordination error of the queue in direction m at intersection i, represents the coordinated consistency coordination error of the n-direction queue at intersection i, represents the mth external coordination direction of intersection i belonging to intersection j, represents the elements in the external adjacency matrix, represents the coordinated consistency coordination error of the queue in n directions at the j intersection, is the connection coefficient between the m-direction queue at intersection i and the expected queue, is a known parameter related to the signal cycle, N is all the intersections in the road network, As parameters, As parameters, is the parameter learning law for the m-th direction of intersection i in the k-1th cycle.

[0170] Step 6: The traffic light cycle is updated, and the newly calculated green light time is sent to the traffic light. The traffic light sends the new green light time to the traffic light and executes it.

[0171] Step 7: The detectors at each intersection of the road network obtain the number of vehicles queuing at each phase of the current cycle;

[0172] Step 8: Calculate the average queue length at each intersection and the maximum queue length at each phase based on the number of queued vehicles obtained;

[0173] Step 9: Determine whether the number of vehicles queuing at each intersection in the current cycle is balanced, that is, whether the number of vehicles queuing at each phase is equal. If the number of vehicles queuing is balanced, the signal machine will continue to issue the green light time of each phase of the current cycle in the next cycle. Otherwise, adjust the algorithm parameters and continue to use the adaptive coordinated control algorithm to calculate the parameter learning rate. Calculate the green light time of each stage in the next cycle;

[0174] Step 10: The traffic light cycle is updated, and the newly calculated green light time is sent to the traffic light. The traffic light sends the new green light time to the traffic light and executes it.

[0175] Step 11: Return to step 7 and repeat steps 7 to 10 to achieve the purpose of real-time control of intersection signals.

[0176] The present invention solves the following problems: 1. It does not rely on traditional traffic flow mechanism modeling, but only designs the controller based on I / O data such as queue length and green light time of the traffic system. 2. It builds a networked traffic multi-agent system, and designs a coordinated sensing control strategy through network interconnection to achieve coordinated control effects among various intersections in the road network. 3. It greatly reduces the delays and queues caused by the waste of green time in traditional intersection signal control.

[0177] The innovative points of the present invention are as follows:

[0178] 1. Traditional control methods cannot adapt to random changes in traffic flow, while this method can adaptively and dynamically adjust the green light time of traffic lights by only using data such as queue length and green light time, so that the queue lengths in each direction are balanced.

[0179] 2. Traditional control methods lack integrity and intelligence. This method builds a multi-agent system. Through the multi-agent system of traffic signals, it connects to the network to exchange data on queue length and green light time, provides real-time online traffic data, and can simultaneously consider the traffic status information of its own intersections and neighboring intersections in the road network structure, thus realizing coordinated control with both integrity and intelligence.

[0180] 3. The signal timing of other methods is usually limited and cannot be fully compatible with under-saturation, critical saturation, and over-saturation conditions. This method avoids wasting green light time by introducing time-varying control signal cycle and green light time constraints.

[0181] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that the process, method, article or terminal device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or terminal device. In the absence of further restrictions, the elements defined by the sentence "include..." or "comprise..." do not exclude the existence of other elements in the process, method, article or terminal device including the elements. In addition, in this article, "greater than", "less than", "exceed" and the like are understood to exclude the number itself; "above", "below", "within" and the like are understood to include the number itself.

[0182] Although the above embodiments have been described, once those skilled in the art know the basic creative concepts, they can make additional changes and modifications to these embodiments. Therefore, the above description is only an embodiment of the present invention and does not limit the patent protection scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the specification and drawings of the present invention, or directly or indirectly used in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A data-driven multi-agent adaptive coordinated control method for networked traffic signals, characterized in that: The following steps are involved: Step 1: Set the road network area; Step 2: Assign initial values ​​to the first cycle green light time of the traffic lights at each intersection in the road network and execute; Step 3: The detectors at each intersection of the road network obtain the number of vehicles queuing at each phase of the current cycle; Step 4: Calculate the average queue length at each intersection and the maximum queue length at each phase based on the number of queued vehicles obtained; Step 5: Adopt the adaptive coordination control algorithm to dynamically and adaptively adjust the green light time of each stage of the next cycle; Step 6: Update the traffic light cycle, send the newly calculated green light time in step 5 to the traffic light, and the traffic light sends it to the traffic light for execution; Step 7: The detectors at each intersection of the road network obtain the number of vehicles queuing at each phase of the current cycle; Step 8: Calculate the average queue length at each intersection and the maximum queue length at each phase based on the number of queued vehicles obtained; Step 9: Determine whether the number of vehicles queuing at each intersection in the current cycle is balanced, that is, whether the number of vehicles queuing at each phase is equal. If the number of vehicles queuing is balanced, the signal machine will continue to issue the green light time of each phase of the current cycle in the next cycle. Otherwise, the parameters of the adaptive coordination control algorithm are adjusted, and the algorithm is continued to be used to dynamically and adaptively adjust the green light time of each phase of the next cycle. Step 10: Update the traffic light cycle, send the newly calculated green light time in step 9 to the traffic light, and the traffic light sends it to the traffic light for execution; Step 11: Return to step 7 and repeat steps 7 to 10 to achieve the purpose of real-time control of intersection signals.

2. The data-driven multi-agent adaptive coordinated control method for networked traffic signals according to claim 1, characterized in that: The adaptive coordinated control algorithm includes signal control system modeling, To model the road network queue intensity, the expression is defined as (1): l i,m (k+1)=l i,m (k)+T i (k)q i,m (k)-s i,m g i,m (k) (1) Among them, the subscript m = 1, 2, 3, 4 are the four phases of the intersection, l i,m (k+1) represents the number of vehicles queued in the k+1th period in the mth direction of intersection i, l i,m (k) represents the number of queued vehicles in the kth period in the mth direction of intersection i, T i (k) represents the kth variable period of intersection i, q i,m (k) represents the vehicle arrival rate in the kth period in the mth direction of intersection i, s i,m represents the saturated flow in the mth direction of intersection i, g i,m (k) represents the green light time of the kth cycle in the mth direction of intersection i. The green light time and signal light cycle of each intersection satisfy the constraint of formula (2): Among them, the subscripts m = 1, 2, 3, 4 are the four phases of the intersection, T i (k) represents the kth variable period of intersection i, g i,m (k) represents the green light time of the kth cycle in the mth direction of intersection i, t i,L It indicates the total lost time at any intersection including the yellow light time state; The queue intensity of the i-th intersection in the k-th period is defined as (3): Among them, x i,m (k) represents the queue intensity at the mth entrance line of intersection i in the kth cycle, l i,m (k) represents the number of queued vehicles in the kth period in the mth direction of intersection i, l i,m,max represents the vehicle storage capacity at the mth entrance line of intersection i; (1) Divide both sides by l i,m,max Then we can get the variable period queue intensity model of the road network as (4): Among them, x i,m (k+1) represents the queue intensity at the mth entrance line of intersection i in the k+1th cycle, x i,m (k) represents the queue intensity at the mth entrance line of intersection i in the kth cycle, T i (k) represents the kth variable period of intersection i, q i,m (k) represents the vehicle arrival rate in the kth period in the mth direction of intersection i, l i,m,max represents the vehicle storage capacity at the mth entrance line of intersection i, s i,m represents the saturated flow in the mth direction of intersection i, g i,m (k) represents the green light time in the mth direction of intersection i; in order to avoid wasting green light time, a time-varying control signal period T is introduced i (k) and the new green time constraint make the queue intensity satisfy x i,m (k)≥0, therefore, the green light time and signal period under the constraint condition are as follows: in, represents the green light time of the kth cycle in the mth direction of intersection i under the constraint conditions, g i,m (k) represents the green light time of the mth direction at intersection i, l i,m (k) represents the number of queued vehicles in the kth period in the mth direction of intersection i, T i (k) represents the kth variable period of intersection i, q i,m (k) represents the vehicle arrival rate in the kth period in the mth direction of intersection i, s i,m represents the saturated flow in the mth direction of intersection i, is the maximum green time, g is the minimum green light time, t i,L It indicates the total lost time at any intersection including the yellow light time state; Therefore, equation (4) can be rewritten as equation (7): Among them, x i,m (k+1) represents the queue intensity at the mth entrance line of intersection i in the k+1th cycle, x i,m (k) represents the queue intensity at the mth entrance line of intersection i in the kth cycle, T i (k) represents the kth variable period of intersection i, q i,m (k) represents the vehicle arrival rate in the kth period in the mth direction of intersection i, s i,m represents the saturated flow in the mth direction of intersection i, represents the green light time of the kth cycle under the constraints in the mth direction of intersection i, l i,m,max represents the vehicle storage capacity at the mth entrance line of intersection i, f i,m (k) is the increased queue intensity, To reduce the queue intensity.

3. The data-driven multi-agent adaptive coordinated control method for networked traffic signals according to claim 2, characterized in that: It also includes the following coordination control issues: When the queues at each intersection of the road network are balanced, the expected queue intensity can be expressed as formula (8): Among them, x r (k+1) is the average queue strength in the four directions in the k+1th cycle, x r (k)∈R is the average queue strength in the four directions in the kth cycle, f r (k)∈R is the average increase in queue strength in the kth signal cycle, u r (k)∈R is the average reduced queue intensity in the kth signal cycle; N is the number of intersections, M=4 represents the four directions of the intersection, x i,m (k) represents the queue intensity at the mth entrance line of intersection i in the kth cycle, T i (k) represents the kth variable period of intersection i, l i,h,max represents the vehicle storage capacity at the hth entrance line of intersection i, q i,h (k) represents the vehicle arrival rate in the kth period in the hth direction of intersection i, s i,h represents the saturated flow in the hth direction of intersection i, is the green light time under the constraint; Taking the queue intensity in each direction of the intersection as an intelligent agent, which also includes the information of adjacent intersections, the coordination error of the coordination consistency in the mth direction of intersection i is formula (9): in, represents the mth internal coordination direction of intersection i, Represents the elements in the internal adjacency matrix, A=[a mn ], m, n∈[1,M] represents the internal coordination adjacency matrix in the single-intersection signal controller network topology, x i,n represents the queue intensity at the nth entrance line of intersection i, x i,m represents the queue intensity at the mth entrance line of intersection i, represents the mth direction of the external coordination of intersection i to intersection j, Represents the elements in the external adjacency matrix, B = [b mn ], m, n∈[1,M] represents the external coordination adjacency matrix in the single-intersection signal controller network topology, x j,n represents the queue intensity at the nth entrance line of intersection j, c i,m represents the connection coefficient between the queue intensity in the m direction of intersection i and the expected queue intensity, x r It indicates the expected queue intensity, i.e. the average queue intensity; Define the internal coordination in-degree matrix of the network controller agent: in, represents the mth internal coordination direction of intersection i, a mn and b in represents the elements in the adjacency matrix, d a and d bi is the row sum of the adjacency matrix, D a and D b is the in-degree matrix, M = 4, C is the diagonal matrix consisting of the connection coefficients of the actual queue strength and the expected queue strength, c i,j is the connection coefficient; The Laplace matrix in the signal controller network topology is L a =D a -A,L b =D b -B; The coordinated consistency coordination error can be further expressed as formula (10): e i =Ax i -D a x i +Bx j -D b x i +Cx r -Cx i (10) Where A is the internal coordination adjacency matrix in the single-intersection signal controller network topology, D a and D b is the in-degree matrix, B represents the external coordination adjacency matrix in the single-intersection signal controller network topology, C is the diagonal matrix consisting of the connection coefficients of the actual queue intensity and the expected queue intensity, x i is the queue intensity at intersection i, x j is the queue intensity at intersection j, x r It represents the expected queue intensity, i.e., the average queue intensity; Through the Laplace matrix transformation, it can be transformed into formula (11): e i =-(L a +C+D b )(x i -x r )-(L b -D b )(x j -x r ) (11) Among them, L a and L b is the Laplace matrix, D b is the in-degree matrix, C is the diagonal matrix consisting of the connection coefficients of the actual queue intensity and the expected queue intensity, x i is the queue intensity at intersection i, x j is the queue intensity at intersection j, x r It indicates the expected queue intensity, i.e. the average queue intensity; definition And satisfy (D a -A)x r =0,(D b -B)x r =0, then equation (11) can be further written as equation (12) and equation (13): Among them, e i (k) and e i (k+1) is the coordinated consistency coordination error, L a and L b is the Laplace matrix, D b is the in-degree matrix, C is the diagonal matrix consisting of the connection coefficients of the actual queue intensity and the expected queue intensity, x i is the queue intensity at intersection i, x j is the queue intensity at intersection j, x r represents the expected queue intensity, that is, the average queue intensity. is the difference between the queue intensity at intersection i and the expected queue intensity, is the difference between the queue intensity at intersection j and the expected queue intensity, is the difference between the queue intensity at intersection i in the k+1th period and the expected queue intensity, is the difference between the queue intensity at intersection j in the k+1th period and the expected queue intensity; Unknown nonlinear parameterized function f i (k) and f r (k) satisfies formula (14) Among them, T i (k) represents the kth variable period of intersection i, l i,m,max represents the vehicle storage capacity at the mth entrance line of intersection i, q i,m (k) represents the vehicle arrival rate in the kth period in the mth direction of intersection i, q i,h (k) represents the vehicle arrival rate in the kth period in the hth direction of intersection i, l i,h,max represents the vehicle storage capacity at the hth entrance line of intersection i, Represents the mth coordinated direction of intersection i. Combined with parameter separation technology, equation (14) satisfies the following inequality conditions: Among them, f i (k) and f r (k) is the unknown nonlinear parameterized function, θ i,m (k) represents the vehicle arrival rate q i,m (k) Related unknown function, e i,m (k) represents the coordination error of the coordination formula in the mth direction of intersection i, Represents the signal period T i (k) related known functions, and satisfy According to formula (14), we can select The global road network, the expected queuing intensity, and the consistency coordination error are described as follows: in, and are the deviations between the actual queue intensity and the expected queue intensity at time k+1 and time k, respectively. is the difference between the actual value and the expected value of the queue strength increase or decrease in the kth signal cycle, f i,M (k) is the increased queue intensity, f r (k) is the average increase in queue strength in the kth signal cycle, To reduce the queue intensity, u r (k)∈R is the average reduced queue strength in the kth signal cycle, T i (k) represents the kth variable period of intersection i, is a known parameter related to the signal period, The final form of the coordinated consistency coordination error considering multiple adjacent intersections is: Among them, e i (k+1) is the coordinated consistency coordination error, L a and L b is the Laplace matrix, D b is the in-degree matrix, C is the diagonal matrix consisting of the connection coefficients of the actual queue intensity and the expected queue intensity, and is the deviation between the actual queue intensity and the expected queue intensity at time k, is the difference between the actual value and the expected value of the queue intensity increase in the kth signal cycle of intersection i, is the difference between the actual value and the expected value of the queue intensity increase in the kth signal cycle of intersection j, is the difference between the actual value and the expected value of the queue intensity reduction in the kth signal cycle of intersection i, is the difference between the actual value and the expected value of the queue intensity reduction in the kth signal cycle of intersection j.

4. The data-driven multi-agent adaptive coordinated control method for networked traffic signals according to claim 3, characterized in that: Also includes controller design: Considering the queuing model formula (7), the multi-agent single intersection data-driven queuing coordination control strategy is constructed as follows: Among them, v i,m (k) represents an arbitrary control signal designed by Lyapunov technique, p i,m >0 is the set parameter, μ>0 is the coordinated control gain, u i,m (k-1) is the green light time of the mth direction at intersection i in the k-1th cycle, Considering the multi-agent based queuing intensity system formula (7) and formula (8) in the traffic network, the coordinated adaptive coordination control term v i,m (k) is selected as formula (21): The data-driven coordinated adaptive coordinated control strategy for regional signal timing is designed as follows: The parameter learning law is designed as follows: in, is the parameter learning law of the mth direction of intersection i in the kth cycle, u i,m (k) is the green light time of the mth direction of intersection i in the kth cycle, μ>0 is the coordination control gain, p i,m >0 is the set parameter, u i,m (k-1) is the green light time of the mth direction of intersection i in the k-1th cycle, c is a parameter, represents the mth internal coordination direction of intersection i, represents the elements in the internal adjacency matrix, e i,m represents the coordinated consistency coordination error of the queue in direction m at intersection i, e i,n represents the coordinated consistency coordination error of the n-direction queue at intersection i, represents the mth external coordination direction of intersection i belonging to intersection j, represents the elements in the external adjacency matrix, e j,n represents the coordinated consistency coordination error of the queue in direction n at intersection j, c i,m is the connection coefficient between the m-direction queue at intersection i and the expected queue, is a known parameter related to the signal cycle, N is all the intersections in the road network, κ is a parameter, α i,m As parameters, is the parameter learning law for the m-th direction of intersection i in the k-1th cycle.

Images