CN108648446B - An Iterative Learning Control Method for Road Network Traffic Signals Based on MFD - Google Patents

Abstract

Aiming at the characteristics of large urban traffic volume, huge urban scale and complex structure in my country, the present invention considers a method of iterative learning control of road network traffic signals based on MFD. It includes step 1: 1.1 obtaining the traffic data of the MFD of the sub-area; 1.2 fitting the MFD of the sub-area; 1.3 determining the ideal road occupancy rate based on the MFD. Step 2: 2.1 Open and closed loop iterative learning control strategy; 2.2 Establish state space equation; 2.3 Optimize signal timing at each intersection. The invention can make the overall structure of the road network relatively balanced, increase the outflow vehicles in the sub-districts, thereby increasing the traffic volume of the road network, provide an effective urban road network control means for traffic managers, and improve the traffic flow of the urban road network. Service Level.

Description

An Iterative Learning Control Method for Road Network Traffic Signals Based on MFD

technical field

The invention relates to the traffic signal control problem of the urban road network, and specifically relates to MFD (traffic macro fundamental graph) and an iterative learning control strategy.

Background technique

Due to the limitation of road resources and infrastructure, traffic congestion in modern cities is still one of the major problems of society. As the most important traffic control method, signal control has been greatly developed with the continuous in-depth research of traffic scholars.

Because the urban traffic system is an uncertain and complex system with a large scale, the parameters of the system model are difficult to determine. Through the analysis of traffic data such as Yokohama, Japan, N. Geroliminis et al. found that there is a specific type of cumulative vehicle number and traffic flow in the urban area. On this basis, the MFD (Macro Fundamental Graph) is proposed, which avoids the defects in the traffic flow modeling and analysis method based on the complex network origin-destination matrix (Origin-Destination Matrix). Through the road network traffic data, the iterative learning control strategy is used to control the intersection signal of the traffic road network, including two parts. The first part uses MFD to fit the sub-regions of the road network and obtains the best accumulated vehicle, which can be used as an ideal control target in the iterative learning signal. The urban road network is modeled, and the internal signals of the sub-road network are iteratively controlled.

As a data-driven method, iterative learning control only uses the online and offline I/O data of the controlled system and the knowledge obtained through data processing to design the controller, and has a wide range of applications in traffic signal control, such as ramp control , urban road network control, etc. Therefore, iterative learning control is used combined with the proposed ideal number of road vehicles based on MFD, the green signal ratio of traffic signals is used as the input of iterative learning control, an appropriate learning law is selected, and the green light time of traffic signals is adjusted so that the vehicles in the road network reach Ideally set the target, so that the overall structure of the road network can be relatively balanced, so that the road network is in the best operating state in the MFD characteristics, and the outflow vehicles in the sub-areas can be improved, thereby increasing the traffic volume of the road network.

SUMMARY OF THE INVENTION

In order to overcome the shortcomings of large city scale, complex structure, and difficulty in modeling by traditional methods, the invention proposes an iterative learning signal control based on a layered control structure to balance the vehicles in the road network and make the road network in the macroscopic basic diagram. The best operating state of the road network, thereby increasing the outflow of vehicles on the road network and improving the traffic capacity of the road network.

The present invention is a method for iterative learning of urban signal control based on MFD, comprising the following steps:

1) Obtain the ideal road occupancy rate based on MFD:

1.1 Obtain the traffic data of the sub-district MFD: divide a large-scale urban road network to obtain several sub-districts R _i , where i∈{1,2,3...}, the sub-district division algorithm adopts the Ncuts algorithm Divide the large-scale urban road network into multiple "homogeneous" sub-districts, and obtain the traffic data of each sub-district.

1.2 Sub-area MFD fitting: Through the traffic data of each sub-area, the cumulative number of vehicles at different times and the MFD characteristics of the output flow of the sub-area, a third-order polynomial is used for fitting. The fitting form of R _i for any sub-area is as follows:

Among them, _ni is the cumulative number of vehicles in the sub-region Ri _, and a ₁ to a ₄ are fitting coefficients.

The least squares method is used to determine the fitting coefficients in the empirical formula:

从而得到MFD的拟合结果，根据拟合结果求得拟合曲线的极值点

Among them, y _i is the actual output flow of the sub-region Ri, G(n _{i )} is the approximate fitting curve of the flow of the sub-region Ri _, and the data deviation is minimized according to the above formula

Thereby, the fitting result of MFD is obtained, and the extreme point of the fitting curve is obtained according to the fitting result.

根据子区R_i的网路结构对子区内部的车辆加权处理，得到子区R_i中各道路的理想占有率：1.3 Determine the ideal road occupancy rate: According to the MFD fitting results in step 1.2, obtain the optimal cumulative number of vehicles in each sub-area

According to the network structure of the sub-area _Ri , the vehicles inside the sub-area are weighted to obtain the ideal occupancy rate of each road in the sub-area _Ri :

为步骤1.2子区R_i的MFD拟合得到的最佳累计车辆数，D_i表示子区R_i内的各路段长度之和，

为道路j的理想占有率(其中j∈R_i)，作为步骤2)中系统控制设计的参考目标。in

is the optimal cumulative number of vehicles obtained by the MFD fitting of the sub-region Ri in step 1.2, D _{i represents} the sum of the lengths of each road segment in the sub-region Ri _,

is the ideal occupancy rate of road j (where j∈R _i ), as the reference target of system control design in step 2).

2) Based on iterative learning control to optimize intersection signal timing:

2.1 Open and closed loop iterative learning control strategy: The open and closed loop iterative learning control structure can be expressed as the following form:

Among them, u _n (k) is the control vector at the k-th sampling time in the n-th iteration process, e _n (k) is the error at the k-th sampling time in the n-th iteration process, k ^c is the closed-loop learning control rate, k ^o is the open-loop learning control rate.

2.2 Establish the state space equation:

为状态向量，表示路网中各路段包含的车辆数。u(k)＝[g₁(k),...,g_N(k)]^T为控制向量，表示路网中所有相位的绿灯时间。d(k)为状态扰动向量，表示各路段的扰动。y(k)＝[o₁(k),...,o_N(k)]^T为系统输出，反映路网中各路段的占有率。输入矩阵B反映了路网的相位、周期、饱和流量等特征；输出矩阵C表示表示道路容量和车辆长度的特征。in

is the state vector, representing the number of vehicles included in each road segment in the road network. u(k)=[g ₁ (k),...,g _N (k)] ^T is the control vector, representing the green light time of all phases in the road network. d(k) is the state disturbance vector, which represents the disturbance of each road segment. y(k)=[o ₁ (k),...,o _N (k)] ^T is the system output, reflecting the occupancy rate of each road section in the road network. The input matrix B reflects the phase, period, and saturated flow characteristics of the road network; the output matrix C represents the characteristics of road capacity and vehicle length.

道路占有率。2.3 Optimize the signal timing of each intersection: take the green light time u(k) in the traffic model as the control input of the open and closed loop iterative learning, the number of vehicles in the road section x(k) as the control state variable, and the state output of the system is related to the vehicle in the road section. same number. Choose appropriate learning rates k ^c and k ^o , adjust the green light time of the intersection, and control the road occupancy rate inside the sub-area to make it track the ideal

road occupancy.

均衡整个路网内部的车辆数。即实现了算法目标。2.4 Repeat step 2.3 to iteratively adjust the signal timing at each intersection until the number of vehicles on the road network reaches the ideal value set in step 1).

Balance the number of vehicles within the entire road network. That is, the algorithm goal is achieved.

Beneficial effects of the present invention: Aiming at the characteristics that the traffic system is an uncertain and complex system with a large scale, and system model parameters are difficult to determine, the present invention can reduce the calculation amount and dimension of large-scale urban road networks, and achieve a balanced traffic flow distribution in the road network. , to increase the traffic volume of the road network, to achieve the purpose of reducing traffic delays and travel time, and it is of great significance to improve the traffic conditions of the entire city.

Description of drawings

FIG. 1 is a schematic structural diagram of an urban road network in an embodiment of the present invention.

Fig. 2 is an MFD fitting effect diagram in an embodiment of the present invention, wherein Fig. 2a shows the MFD fitting curve of the sub-region R1, _Fig . 2b shows the MFD fitting curve of the sub _- region R2, and Fig. 2c shows the MFD fitting curve of the sub-region _R3 . MFD fitting curve, Figure 2d shows the MFD fitting curve of subregion R4 _.

Detailed ways

The present invention will be further described below through the accompanying drawings and embodiments.

The present invention is directed to an urban road network with 34 intersections as shown in FIG. 1 , and each intersection and road section is equipped with real-time detection equipment for detecting required traffic parameters. Two adjacent intersections are two-way lanes, each road has 2 lanes, the length of each road segment has been determined, and the road network has 21 input nodes.

The present invention is a method for iterative learning of urban signal control based on MFD, comprising the following steps:

1) Obtain the ideal road occupancy rate based on MFD:

1.1 Obtaining the traffic data of the sub-district MFD: Divide the urban road network in Figure 1, and use the Ncuts algorithm to divide to obtain 4 "homogeneous" sub-districts. Different colors represent different sub-districts. Among them, R ₁ contains 8 intersections, and R ₂ contains 7 intersections, R ₃ contains 7 intersections, and R ₄ contains 12 intersections. Based on the traffic data of each sub-area, the fitting curve of the MFD is obtained, and the optimal operating state of the sub-area is calculated.

1.2 Sub-area MFD fitting: Through the traffic data of each sub-area, the cumulative number of vehicles at different times and the MFD characteristics of the output flow of the sub-area, a third-order polynomial is used for fitting. The fitting form of R _i for any sub-area is as follows:

Among them, _ni is the cumulative number of vehicles in the sub-region Ri _, and a ₁ to a ₄ are fitting coefficients.

The least squares method is used to determine the fitting coefficients in the empirical formula:

从而得到MFD的拟合结果，根据拟合结果求得拟合曲线的极值点

Among them, y _i is the actual output flow of the sub-region Ri, G(n _{i )} is the approximate fitting curve of the flow of the sub-region Ri _, and the data deviation is minimized according to the above formula

Thereby, the fitting result of MFD is obtained, and the extreme point of the fitting curve is obtained according to the fitting result.

根据子区R_i的网路结构对子区内部的车辆加权处理，得到子区R_i中各道路的理想占有率：1.3 Determine the ideal road occupancy rate: According to the MFD fitting results in step 1.2, obtain the optimal cumulative number of vehicles in each sub-area

According to the network structure of the sub-area _Ri , the vehicles inside the sub-area are weighted to obtain the ideal occupancy rate of each road in the sub-area _Ri :

为步骤1.2子区R_i的MFD拟合得到的最佳累计车辆数，D_i表示子区R_i内的各路段长度之和，

为道路j的理想占有率(其中j∈R_i)，作为步骤2)中系统控制设计的参考目标。in

is the optimal cumulative number of vehicles obtained by the MFD fitting of the sub-region Ri in step 1.2, D _{i represents} the sum of the lengths of each road segment in the sub-region Ri _,

is the ideal occupancy rate of road j (where j∈R _i ), as the reference target of system control design in step 2).

2) Based on iterative learning control to optimize intersection signal timing:

2.1 Open and closed loop iterative learning control strategy: The open and closed loop iterative learning control structure can be expressed as the following form:

Among them, u _n (k) is the control vector at the k-th sampling time in the n-th iteration process, e _n (k) is the error at the k-th sampling time in the n-th iteration process, k ^c is the closed-loop learning control rate, k ^o is the open-loop learning control rate.

2.2 Establish the state space equation:

为状态向量，表示路网中各路段包含的车辆数。u(k)＝[g₁(k),...,g_N(k)]^T为控制向量，表示路网中所有相位的绿灯时间。d(k)为状态扰动向量，表示各路段的扰动。y(k)＝[o₁(k),...,o_N(k)]^T为系统输出，反映路网中各路段的占有率。输入矩阵B反映了路网的相位、周期、饱和流量等特征；输出矩阵C表示表示道路容量和车辆长度的特征。in

is the state vector, representing the number of vehicles included in each road segment in the road network. u(k)=[g ₁ (k),...,g _N (k)] ^T is the control vector, representing the green light time of all phases in the road network. d(k) is the state disturbance vector, which represents the disturbance of each road segment. y(k)=[o ₁ (k),...,o _N (k)] ^T is the system output, reflecting the occupancy rate of each road section in the road network. The input matrix B reflects the phase, period, and saturated flow characteristics of the road network; the output matrix C represents the characteristics of road capacity and vehicle length.

道路占有率。2.3 Optimize the signal timing of each intersection: take the green light time u(k) in the traffic model as the control input of the open and closed loop iterative learning, the number of vehicles in the road section x(k) as the control state variable, and the state output of the system is related to the vehicle in the road section. same number. Choose appropriate learning rates k ^c and k ^o , adjust the green light time of the intersection, and control the road occupancy rate inside the sub-area to make it track the ideal

road occupancy.

均衡整个路网内部的车辆数。即实现了算法目标。2.4 Repeat step 2.3 to iteratively adjust the signal timing at each intersection until the number of vehicles on the road network reaches the ideal value set in step 1).

Balance the number of vehicles within the entire road network. That is, the algorithm goal is achieved.

The specific implementation examples described herein are only for the specific illustration of the present invention, and are not intended to limit the right scope of the present invention.

Claims (1)

1. A road network traffic iterative learning control method based on MFD is characterized in that the MFD is combined with the iterative learning control idea to control signals of a large-scale urban road network, and the steps are as follows:

1) acquiring an ideal road occupancy rate based on the MFD:

1.1 obtaining traffic data of the sub-region MFD: dividing a large-scale city road network to obtain a plurality of subareas R_iWherein i ∈ {1,2,3. }, the algorithm for sub-area division adopts the Ncuts algorithm for division, and a large-scale urban road network is decomposed into a plurality of 'homogeneous' sub-areas to obtain traffic data of each sub-area;

1.2 sub-region MFD fitting: fitting by using a 3-order polynomial according to the traffic data of each subarea, the accumulated vehicle number at different moments and the MFD (MFD) characteristic of the output flow of the subarea, and aiming at any subarea R_iThe fit form is as follows:

wherein n is_iIs a sub-region R_iAccumulated number of vehicles of a₁～a₄Is a fitting coefficient;

determining a fitting coefficient in an empirical formula by using a least square method:

wherein, y_iIs a sub-region R_iActual output flow of G (n)_i) Is a sub-region R_iApproximate fitting curve of flow, minimizing data deviation according to the above equation_i ²So as to obtain the fitting result of MFD and obtain the extreme point of the fitting curve according to the fitting result

1.3, determining the ideal road occupancy rate: obtaining the optimal cumulative vehicle number of each subarea according to the MFD fitting result of the step 1.2

According to sub-region R_iThe network structure of (2) weighting the vehicles in the sub-area to obtain the sub-area R_iIdeal occupancy of each road:

wherein

In the step 1.2 sub-region R_iThe optimal cumulative vehicle number, D, obtained by MFD fitting of_iDenotes the sub-region R_iThe sum of the lengths of the various paths within,

ideal occupancy for road j (where j ∈ R_i) As a reference target for the system control design in step 2);

2) controlling and optimizing intersection signal timing based on iterative learning:

2.1 open-close loop iterative learning control strategy: the open-closed loop iterative learning control structure can be represented in the form:

wherein u is_n(k) For the control vector at the kth sampling instant of the nth iteration, e_n(k) For the error at the kth sampling instant, k, of the nth iteration^cFor closed loop learning control rate, k^oThe open loop learning control rate;

2.2 establish the state space equation:

wherein

A state vector representing the number of vehicles included in each road segment in the road network; u (k) ═ g₁(k),...,g_N(k)]^TRepresenting the green light time of all phases in the road network as a control vector; d (k) is a state disturbance vector which represents the disturbance of each road section; y (k) ═ o₁(k),...,o_N(k)]^TReflecting the occupancy of each road section in the road network for system output; the input matrix B reflects the phase, the period and the saturation flow of the road network; the output matrix C represents characteristics representing road capacity and vehicle length;

2.3 optimizing the signal timing of each intersection: taking the green light time u (k) in the traffic model as the control input of the iterative learning of the open-close loop, taking the number x (k) of vehicles on the road section as a control state variable, and taking the state output of the system to be the same as the number of vehicles on the road section;selecting a suitable learning rate k^cAnd k^oAdjusting the green time of the intersection, controlling the road occupancy in the sub-area to make it track the ideal

Road occupancy;

2.4 repeating the step 2.3, iteratively adjusting the signal timing of each intersection until the number of vehicles in the road network reaches the ideal value set in the step 1)

And balancing the number of vehicles in the whole road network.

Images