Adaptive control multi-phase single intersection signal control system and method

Abstract

The application provides a multi-phase single intersection signal control system and method with adaptive control. The application uploads the vehicle queue length data measured by the vehicle detectors arranged at each phase of the intersection to a data processing platform, calculates a pseudo-Jacobi matrix through the input vehicle queue length difference of each phase, and then calculates the green light timing of each phase required to achieve the queue length difference in the period through the pseudo-Jacobi matrix and the theoretical queue length difference. After the calculation is completed, the data is issued to the signal machine for execution. After the execution is completed, the next cycle period is started, so as to realize the model-free adaptive control of the intersection.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent traffic control, and particularly relates to a single intersection signal control system and method based on data-driven adaptive control. BACKGROUND

[0002] In recent years, with the rapid growth of population and urbanization, the traffic demand of domestic cities is steadily rising, and traffic congestion is also increasing. As the hub of regulating urban traffic, the intersection has a great influence on traffic travel and traffic congestion, so the signal control of the intersection is crucial.

[0003] At present, most of the domestic intersection signal control is timing control or main line coordinated control, which is simple and easy to implement, but when the randomness of vehicle arrival is large, the control effect is poor, and the signal light timing cannot be adjusted according to the size of the vehicle arrival rate. Unreasonable phase timing results in a large amount of green loss, reducing the traffic efficiency.

[0004] Many experts and scholars widely apply various models such as Q-learning, neural network and fuzzy logic to traffic control, which can effectively improve the intersection traffic efficiency. However, with the increasing demand for traffic, the modeling difficulty of traffic control gradually increases, the workload increases, the model control effect gradually deteriorates, and the established model is only suitable for traffic with a specific saturation flow, the algorithm is complex and not easy to implement in engineering. SUMMARY

[0005] In view of the above problems, the present application provides an intersection adaptive control method based on data driving, which can apply model-free adaptive control to the field of traffic, effectively alleviate the intersection problem without establishing a large number of models, and greatly reduce the workload. The present application specifically adopts the following technical solutions:

[0006] An adaptive control multi-phase single intersection signal control system, which comprises a data processing platform, an interactive server, a vehicle detector and an actuator.

[0007] The vehicle detector detects the vehicle queue length of each phase at the end of a cycle and sends the data to the data processing platform through the interactive server. The vehicle detector is a radar detector, a video detector or a geomagnetic detector.

[0008] The data processing platform receives the vehicle queue length data of each phase collected by the vehicle detector every cycle, and performs model-free adaptive control with the goal of balancing the queue length of each phase every cycle. The green time of each phase in the next cycle is calculated based on the vehicle queue data of the previous cycle, and is sent to the actuator for operation according to the instruction.

[0009] An interaction server, which connects the vehicle detector and the data processing platform, and connects the data processing platform and the actuator; it transforms data into data that the data processing platform can recognize, and transforms data calculated by the data processing platform into the specified of the actuator;

[0010] An actuator, which receives the instructions issued by the data processing platform through the interaction server, and makes the intersection signal light execute in the next cycle.

[0011] Preferably, the method comprises the following steps:

[0012] First step: obtain the queue length L(k) of each phase by the vehicle detector;

[0013] Second step: calculate the queue length difference y(k) of each phase; wherein y1(k) = L1(k) - L2(k), y2(k) = L2(k) - L3(k), y3(k) = L3(k) - L4(k)

[0014] Third step: calculate the pseudo-block Jacobi matrix Φ(k) of the cycle from the queue length difference y(k);

[0015] Fourth step: calculate the green time g(k) of each phase in the cycle from y(k), Φ(k) and the desired queue length difference, i.e. the theoretical queue length difference yd(k);

[0016] Fifth step: issue the timing of the kth cycle to the actuator and execute it;

[0017] Sixth step: when the actuator finishes executing the kth cycle, k = k + 1, return to the first step and enter the k+1th cycle.

[0018] Preferably, the method is calculated as follows:

[0019] The data model is represented as:

[0020] y(k+1) - y(k) = Φ p,L (k)ΔG L (k) (1)

[0021] Wherein, ΔGL(k) is a vector composed of all control input vectors in the sliding time window [k-L+1, k];

[0022] Estimation criterion function:

[0023] J(Φ p,L (k)) = ||Δy(k) - Φ p,L (k)ΔG L (k-1) || 2 + μ||Φ p,L (k) - Φp,L (k-1)| 2 (2)

[0025] where Δy(k) represents the difference between the queue length difference at time k and the queue length difference at time k-1;

[0026] Minimizing (2) gives the pseudo-block Jacobi matrix Φ of the system p,L (k):

[0027]

[0028] where Φ p,L (k) = [Φ1(k),…,Φ L (k)] ∈ R m×mL ,

[0029] is the corresponding sub-matrix

[0030] φ ii1 (k) = φ ii1 (1) if |φ ii1 (k)| < b2 or |φ ii1 (k)| > αb2 or sign(φ ii1 (k)) ≠ sign(φ ii1 (1)), i = 1,…,m

[0031] φ ij1 (k) = φ ij1 (1) if |φ ij1 (k)| > b1 or sign(φ ij1 (k)) ≠ sign(φ ij1 (1)), i,j = 1,…,m, i ≠ j

[0032] where m represents the dimension of the matrix, α ≥ 1; b2 > b1(2α+1)(m-1), a, b are constants;

[0033] ΔGL(k) = [ΔgT(k),…,ΔgT(k-L+1)]T

[0034] where Δg(k) is the difference between the green time of each phase at time k and at time k-1, Δg(k) = g(k)-g(k-1), y(k) is the control output queue length difference of each phase at the kth sampling period, g(k) is the control input green time of each phase at the kth sampling period, μ > 0 is the weight factor, and η is the step factor;

[0035] The following control input green time of each phase criterion function is considered:

[0036] J(g(k)) = ||yd(k + 1) - y(k + 1) || 2 + λ||g(k) - g(k - 1) || 2 (4)

[0037] Substitute (1) into (4), take the derivative of g(k) and set it equal to zero to obtain the phase green time:

[0038]

[0039] where λ is a penalty factor to avoid too drastic signal changes, ρ1,..., ρ L are step factors, yd(k + 1) is the target queue length difference;

[0040] Phase green time constraints:

[0041] If g i < g min

[0042]

[0043] If g i > g min And

[0044]

[0045] If g i > g min g j < g min g i There are n g j There are (3 - n) g

[0046]

[0047] g min is the minimum green time, C is the cycle length, and Δt is the lost time within the cycle BRIEF DESCRIPTION OF DRAWINGS

[0048] Figure 1 Signal control system joint debugging diagram.

[0049] Figure 2 Signal control system flowchart.

[0050] Figure 3 Signal machine data transmission and interaction diagram.

[0051] Figure 4 Algorithm calculation flowchart. DETAILED DESCRIPTION

[0052] In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the following will clearly and completely describe the technical solutions of the embodiments of the present application with reference to the drawings of the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, rather than all the embodiments of the present application. Based on the described embodiments of the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort belong to the scope of the present application.

[0053] Those skilled in the art can understand that, unless otherwise defined, all terms (including technical terms and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. It should also be understood that terms such as those defined in a general dictionary have meanings consistent with those in the context of the prior art and should not be interpreted in an idealized or overly formal sense unless expressly so defined herein.

[0054] Reference Figure 1 The figure is a signal control system joint debugging figure, which includes main components of the system, i.e. a data processing platform, an interactive server, vehicle detectors and an executor (signal machine). The arrow direction represents the direction of instruction transmission. The data processing platform receives effective data of vehicle queue length of each phase detected by the vehicle detectors through the interactive server, and imports the data into the controller to calculate green light time of each phase required to achieve the balance of vehicle queue length of each phase at the end of the cycle. After the green light time is calculated, the timing scheme of each phase is issued to the executor (signal machine) through the interactive server, so that the executor executes.

[0055] A single intersection signal control system based on adaptive control includes a data processing platform, an interactive server, four vehicle detectors (applied to a four-phase intersection), and an executor. The system includes the following parts:

[0056] The data processing platform functions to input collected data, i.e. vehicle queue length of each phase in each cycle, into an algorithm, and perform model-free adaptive control with the balance of vehicle queue length of each phase in each cycle as the target. The green light time of each phase in the next cycle is calculated through the vehicle queue data of the previous cycle, and is issued to the signal machine to make it operate according to the instruction. The present application uses an edge data processor.

[0057] The interactive server functions to convert data, i.e. convert the data required by the platform into the data calculated by the platform, and convert the data required by the signal machine into the data calculated by the platform. The interactive server is a bridge connecting the vehicle detectors and the data platform, and the data platform and the signal machine, accepts downlink instructions and forwards them to the executor, and accepts uplink vehicle data and forwards them to the data platform.

[0058] Vehicle detector, which is used to detect the vehicle queue length of each phase at the end of a cycle, and send the data to the data processing platform through the interaction server. The commonly used vehicle detectors on the market include radar detectors, video detectors and geomagnetic detectors, etc. Through signal analysis, the relevant information of the detection target can be obtained, and through processing, the corresponding data of the vehicle can be obtained.

[0059] Actuator, which is the signal machine and signal lamp in the present application, receives the command issued by the data platform through the interaction server, and makes the intersection signal lamp execute in the next cycle.

[0060] Reference Figure 2 The signal control system flow chart, wherein the specific operation is as follows:

[0061] First step: the queue length L(k) of each phase can be obtained by the vehicle detector.

[0062] Second step: calculate the queue length difference y(k) of each phase. (wherein y1(k) = L1(k)-L2(k), y2(k) = L2(k)-L3(k), y3(k) = L3(k)-L4(k))

[0063] Third step: calculate the pseudo-block Jacobi matrix Φ(k) of the cycle from the queue length difference y(k).

[0064] Fourth step: combine y(k) with Φ(k) and the desired queue length difference, i.e. the theoretical queue length difference yd(k), to calculate the green time g(k) of each phase in the cycle.

[0065] Fifth step: and the timing of the kth cycle is issued to the actuator, i.e. the signal machine, and executed.

[0066] Sixth step: when the actuator executes the kth cycle, k = k + 1, return to the first step, and the controller enters the k+1 cycle.

[0067] Reference Figure 3 Signal machine data transmission and interaction:

[0068] The data detected by the detector is converted into data that can be used by the edge computing processor through the data interaction; the control algorithm is input in the edge computing processor, the data of the detector is connected to the edge computing processor through the communication interface, and the processor transmits the calculated timing of each phase to the signal machine.

[0069] Reference Figure 4 Algorithm calculation process:

[0070] The PFDL data model can be represented as:

[0071] y(k+1)-y(k) = Φ p,L(k) ΔG L (k) (1)

[0072] ΔGL(k) is a vector composed of all control input vectors in the sliding time window [k-L+1, k].

[0073] Estimation criterion function:

[0074] J(Φ p,L (k)) = || Δy(k) - Φ p,L (k) ΔG L (k-1) || 2 + μ || Φ p,L (k) - Φ p,L (k-1) || 2 (2)

[0076] Δy(k) represents the difference between the queue length difference at time k and the queue length difference at time k-1.

[0077] Minimizing (2) can obtain the pseudo-block Jacobian matrix Φ p,L (k) of the system:

[0078]

[0079] wherein Φ p,L (k) = [Φ1(k), …, Φ L (k)] ∈ R m×mL ,

[0080] is the corresponding sub-matrix

[0081] φ ii1 (k) = φ ii1 (1), if |φ ii1 (k)| < b2 or |φ ii1 (k)| > αb2 or sign(φ ii1 (k)) ≠ sign(φ ii1 (1)), i = 1, …, m

[0082] φ ij1 (k) = φ ij1 (1), if |φ ij1 (k)| > b1 or sign(φ ij1 (k)) ≠ sign(φ ij1 (1)), i, j = 1, …, m, i ≠ j

[0083] wherein m represents the dimension of the matrix, α ≥ 1; b2 > b1 (2α+1)(m-1), a and b are constants.

[0084] AGL(k) = [AGT(k),..., AGT(k-L+1)]T, AG(k) is the difference of green time of each phase between k period and k-1 period, AG(k) = g(k) - g(k-1). y(k) is the difference of queue length of each phase in the kth sampling period. g(k) is the green time of each phase in the kth sampling period. μ > 0 is the weight factor. η is the step factor.

[0085] The control input green time of each phase is considered as follows:

[0086] J(g(k)) = ||yd(k+1) - y(k+1) || 2 + λ||g(k) - g(k-1) || 2 (4)

[0087] Substitute (1) into (4), take the derivative of g(k) and set it equal to zero to obtain the green time of each phase:

[0088]

[0089] where λ is the penalty factor to avoid too intense signal changes, ρ1,..., ρ L are step factors. yd(k+1) is the target output function, i.e., the target difference of queue length.

[0090] Green time constraints of each phase:

[0091] If g i < g min

[0092]

[0093] If g i > g min and

[0094]

[0095] If g i > g min , g j < g min , g i has n, g j has (3-n)

[0096]

[0097] g min is the minimum green time, C is the cycle length, and Δt is the lost time in the cycle.

[0098] The above merely describes the embodiments of the present application, which are specific and detailed, but should not be understood as a limitation to the patent scope of the present application. It should be noted that, for those skilled in the art, without departing from the concept of the present application, several modifications and improvements can be made, which all belong to the protection scope of the present application.

Claims

1. An adaptive control multi-phase single intersection signal control system, comprising a data processing platform, an interactive server, a vehicle detector, an executor, characterized in that, a vehicle detector, which detects the vehicle queue length of each phase at the end of a cycle and sends data to the data processing platform through the interactive server, said vehicle detector being a radar detector, a video detector or a geomagnetic detector; a data processing platform, which receives the vehicle queue length data of each phase collected by the vehicle detector every cycle, and performs model-free adaptive control with the goal of balancing the queue length of each phase every cycle, calculates the green light time of each phase in the next cycle through the vehicle queue data of the previous cycle, and issues instructions to the executor to operate according to the instructions; an interactive server for connecting the vehicle detector and the data processing platform, and connecting the data processing platform and the executor; it converts data into data that the data processing platform can recognize, and converts data calculated by the data processing platform into the specified of the executor; an executor that receives instructions issued by the data processing platform through the interactive server, and makes the intersection signal light execute in the next cycle; The system adopts the following method steps: First step: obtain the queue length L(k) of each phase by the vehicle detector; Second step: calculate the queue length difference y(k) of each phase; where y1(k) = L1(k) - L2(k), y2(k) = L2(k) - L3(k), y3(k) = L3(k) - L4(k) Third step: calculate the pseudo-block Jacobi matrix Φ(k) of this cycle from the queue length difference y(k); Fourth step: calculate the green light time g(k) of each phase in this cycle from y(k), Φ(k) and the desired queue length difference, i.e. the theoretical queue length difference yd(k); Fifth step: and issue the timing of the kth cycle to the executor and execute; Sixth step: when the executor executes the kth cycle, k = k + 1, return to the first step and enter the k+1 cycle; The specific calculation method is as follows: The data model is represented as: y(k + 1) - y(k) = Φ p,L (k) ΔG L (k) (1) Where ΔGL(k) is a vector composed of all control input vectors within the sliding time window [k-L+1, k]; the estimation criterion function is: J(Φ p,L (k) = ||Δy(k) - Φ p,L (k)ΔG L (k-1)| 2 + μ||Φ p,L (k) - Φ p,L (k-1)| 2 (2) Where Δy(k) represents the difference between the queue length difference at time k and the queue length difference at time k-1; Minimizing (2) gives the pseudo-block-Jacobi matrix Φ of the system p,L (k): where Φ p,L (k) = [Φ1(k),...,Φ L (k)] ∈ R m×mL , for the corresponding sub-matrix φ ii1 (k) = φ ii1 (1), if |φ ii1 (k)| < b2or |φ ii1 (k)| > αb2or sign(φ ii1 (k)) ≠ sign(φ ii1 (1)), i = 1,..., m φ ij1 (k)=φ ij1 (1), if |φ ij1 (k)|>b1 or sign(φ) ij1 (k))≠sign(φ ij1 (1)), i,j=1,…,m,i≠j Where m represents the dimension of the matrix, α ≥ 1; b2 > b1(2α+1)(m-1), a, b are constants; ΔGL(k) = [ΔgT(k), …, ΔgT(k-L+1)]T Where Δg(k) is the difference between the green light time of each phase in the kth cycle and the k-1th cycle, Δg(k) = g(k) - g(k-1), y(k) is the control output queue length difference of each phase of the system in the kth sampling cycle, g(k) is the control input green light time of each phase of the system in the kth sampling cycle, μ > 0 is the weight factor, and η is the step factor; Consider the following control input criterion function for each phase green light time: J(g(k)) = ||yd(k + 1) - y(k + 1) || 2 + λ||g(k) - g(k - 1) || 2 (4) Substitute (1) into (4), take the derivative of g(k), and set it equal to zero to get the green light time of each phase: where λ is a penalty factor to avoid too drastic signal changes, ρ1,..., ρ L is a step size factor and yd(k+1) is the target queue length difference; Phase green time constraints: If g i < g min if g i > g min and If g i >g min g j g min g i n g j 3-n g g min C is the minimum green time, C is the cycle length, and Δt is the lost time within the cycle.

Images