A Robust Cooperative Optimization Method for Intersection Signal Control and Variable Lanes

Abstract

The present invention proposes a robust collaborative optimization method for intersection signal control and variable lanes, belonging to the technical field of urban traffic management. The specific implementation is as follows: First, select the optimization period and divide the time intervals, and implement a signal control scheme for each time interval on the basis that the variable lane channelization remains unchanged within the same period; Second, extract historical data and collect real-time data, and count the traffic volume by time interval; Third, propose a robust optimization model and algorithm driven by historical data, use the generalized saturation as the optimization index, and solve it using a double-layer nested genetic algorithm; Fourth, propose a robust optimization model and algorithm jointly driven by historical data and real-time data, use the average vehicle delay as the optimization index, and solve it using a multi-mode genetic algorithm; Finally, determine the selection of the joint mode of the two types of data in the next period, which means taking the traffic historical data of the previous 5 days as samples, calculating its standard deviation and determining according to the interval where it is located.

Description

Technical Field

[0001] The present invention belongs to the technical field of urban traffic management, and particularly relates to a traffic control optimization method and device. Background Art

[0002] With the remarkable improvement of the urban motorization level, the phenomenon of urban traffic congestion has become increasingly prominent, seriously affecting the normal commuting of residents, and has become a major problem restricting the normal operation of the urban traffic system. Among them, the contradiction between travel demand and traffic resources supply and demand is the main cause of congestion. However, traffic resources are difficult to be fully utilized, especially restricted by the finiteness of traffic resources, the settings of these traffic facilities in terms of time and space often cannot keep up with the development of traffic demand.

[0003] Traffic flow with tidal characteristics will generate periodic and regular traffic congestion, and this tidal congestion phenomenon reflects the contradiction between dynamic traffic demand and static traffic facilities. For non-tidal congested intersections, it is still difficult to solve the congestion problem only relying on signal control optimization in terms of time, while for tidal congested intersections, in addition to taking effective signal control measures, the lane composition on the road surface should also be optimized to adapt to the tidal demand of certain flow directions, that is, the combined optimization of signal control and lane composition.

[0004] The multi-period control method is a relatively effective method. It not only makes up for the defect that the traditional fixed-time signal control is difficult to adapt to dynamic traffic demand, but also performs better than the induction control method in terms of cost and actual engineering application.

[0005] Traffic flow also has uncertainty. Although the traffic volume shows a similar periodic change trend between days, it shows different traffic demands with the change of time within a day, which brings difficulties to designing a relatively robust signal control scheme. Therefore, when conducting combined optimization, it is necessary to study the modeling under uncertainty. By improving the robustness of signal control to adapt to the fluctuating traffic flow. Summary of the Invention

[0006] To this end, the technical problem to be solved by the present invention is to synergistically optimize the existing traffic control from two dimensions of time and space. According to the multi-period control method, combined with the robust optimization method of traffic signal control, the optimization problem of urban single-point intersections is divided into two stages. In the first stage, a robust optimization model of signal control and variable lanes is established and solved by a double-layer nested genetic algorithm (DN-GA) to achieve the robust optimization of variable lane flow direction decision-making and the initial signal control scheme. In the second stage, a dynamic robust optimization model of signal control is established and solved by a multi-mode genetic algorithm (MM-GA) to achieve the dynamic optimization of the signal control scheme.

[0007] To solve the above technical problem, the technical solution adopted by the present invention is a collaborative optimization method for signal control and variable lanes at single-point intersections. The implementation process of this method is as follows:

[0008] Step 1: Select the optimization period and divide the time interval. The period selection means that with a period length of 2 hours and 7 am as the reference point, one day is divided into 12 periods, and an initial signal control scheme and a variable lane channelization scheme for each intersection are implemented for each period respectively. The time interval division means that with a time interval length of 15 minutes, each period is divided into 8 time intervals, and a signal control scheme is implemented for each time interval respectively on the basis of ensuring that the variable lane channelization scheme remains unchanged within the same period.

[0009] Step 2: Extract the traffic historical data and collect the real-time data for each period, specifically referring to the statistics of the traffic volume in each direction of the intersection according to the time intervals divided in Step 1 for each period. Among them, the historical data for each period needs to extract the traffic volume for 15 days, and the real-time data needs to collect the traffic volume for 1 day.

[0010] According to the time interval length (15 minutes) described in Step 1, a period (2 hours) can be divided into 8 time intervals. A total of 120 (15 * 8) recorded data obtained within the same period of 15 days are stored in the historical database Ω H Among them, a total of 8 recorded data collected within the same period of 1 day are stored in the real-time database Ω R And the historical data and real-time data are numbered in the order of their appearance in the time interval, that is, there are Ω H ={1, 2,..., h,..., 120}, Ω R ={1, 2,..., r,..., 8}. Where h and r are the historical database Ω H And the real-time database Ω RThe label of the time interval corresponding to any data

[0011] Step 3: Propose a robust collaborative optimization mathematical model and solution algorithm for intersection signal control and variable lanes driven by historical data. The mathematical model is a robust optimization model with the mean-standard deviation (MSD) of the generalized saturation degree of the intersection as the optimization objective, the 15-day traffic historical data as the input, and the variable lane channelization plan and initial signal control plan during the research period as the output. The solution algorithm is a double-layer nested heuristic genetic algorithm (DN-GA) specially designed for the above robust optimization model

[0012] The objective function Z1 of the mathematical model for the robust collaborative optimization of intersection signal control and variable lanes driven by historical data selects the generalized saturation degree χ of the intersection as the optimization index PI1, which consists of its mean value X1 and standard deviation Y1. The model can be written as:

[0013]

[0014] Among them, γ1 is the weight parameter of the optimization model, with a value of γ1 = 0.5, and its range is γ1 ∈ [0, 1]. It can be selected according to the emphasis of traffic management. The larger the value, the more the model emphasizes the stability of signal control

[0015] Optionally, when the model weight parameter γ1 = 0, the average value of the optimization index PI1 is selected for optimization, fully considering the efficiency of signal control

[0016] Optionally, when the model weight parameter γ1 = 1, the standard deviation of the optimization index PI1 is selected for optimization, fully considering the stability of signal control

[0017] The parameter π1(k) of the optimization model is the weight ratio of the historical data corresponding to the kth time interval to the total data, and its value is representing the historical database Ω H ={1, 2,..., h,..., 120}, and each data in it has the same contribution degree to the model; the larger the value of π1(k), the greater the contribution of the historical data corresponding to the kth time interval to the model, and vice versa

[0018] The calculation method of the generalized saturation degree χ(k) of each time interval in the objective function Z1 of the optimization model is as follows:

[0019]

[0020] In the formula, j is the number of the intersection flow direction, and the NEMA coding is used to number each import direction of the intersection; χ j is the saturation degree of flow direction j; q j is the traffic flow of flow direction j; c j is the traffic capacity of flow direction j; s j is the saturated flow rate of flow direction j; λ j is the green signal ratio of flow direction j; N j is the number of lanes of flow direction j; is the effective green light duration of flow direction j; C is the signal cycle duration;

[0021] The decision variable of the model is the variable lane n i and the signal control scheme

[0022] The constraint conditions of the optimization model include the minimum green light time constraint, the phase mode constraint, the cycle duration constraint, the maximum saturation constraint, the integer constraint and the 0-1 constraint;

[0023] Minimum green light time constraint:

[0024] g j ≥g min

[0025] Phase mode constraint:

[0026] In the standard cross intersection where north-south intersects with east-west, taking the standard three-phase as an example to illustrate the constraint relationship of the phase mode. The first phase is the straight-through of the north-south import, the second phase is the left-turn of the north-south import, and the third phase is all turns of the east-west import. The constraint relationship of this phase mode is:

[0027] g1 = g5

[0028] g2 = g6

[0029] g3 = g4 = g7 = g8

[0030] g1 + g2 + g3 + 3(t y +t r ) = C

[0031] Cycle duration constraint:

[0032] C min ≤C≤C max

[0033] Maximum saturation constraint:

[0034] χ j ≤χ max

[0035] Integer constraint:

[0036]

[0037] 0-1 Constraint:

[0038] n i ∈ {0, 1}, i = 1, 2, ..., 4

[0039] where g j is the green light time for flow to j; g min is the minimum green light time, with a value of 10 s; t y is the yellow light time, with a value of 4 s; t r is the all-red time, with a value of 2 s; C min is the minimum cycle length, with a value of 60 s; C max is the maximum cycle length, with a value of 240 s; C is the signal control cycle; χ max is the maximum saturation threshold, with a value of 0.95; χ j is the saturation of flow to j.

[0040] The two sets of decision variables of the model, the variable lane n i and the signal control scheme g j 0 are mutually coupled. A double-layer nested heuristic genetic algorithm (DN-GA) specially designed for the above robust optimization model is used for solution. The optimization problem is divided into a signal control layer (TSC) and a variable lane layer (RL). The main steps are as follows:

[0041] step1: Initialization, calibrate relevant GA parameters and traffic parameters. The GA parameters include population size (100), number of iterations (100), crossover probability (0.8), and mutation probability (0.02); the traffic parameters include: minimum cycle length (60 s), maximum cycle length (240 s), yellow light time (4) s, all-red time (2 s), minimum green light time (10 s), lost time (1 s), and maximum saturation (0.95).

[0042] step2: Let the number of iterations P = 0.

[0043] step3: Generate the initial population of the signal control layer g j . According to the constraint conditions of the model in step three, find the constraint range of the decision variables, and generate the initial population according to the decimal coding method; when some individuals in the population do not meet the constraint conditions, regenerate new individuals according to the above method until all individuals meet the constraint conditions, and use them as the initial population g j .

[0044] Step 4: Generate the initial variable lane layer population n i According to the constraint conditions of the model described in Step 3, solve the constraint range of the decision variables, and generate the initial population n using 0-1 coding i .

[0045] Step 5: Solve the RL layer using GA. The input of the RL layer is the initial population n i and a given g j . The GA of the RL layer includes a series of operations such as selection, crossover, mutation, and evolution, and generates a new population. When the maximum number of iterations reaches 100 generations, the optimal variable lane flow setting is obtained

[0046] Step 6: Solve the TSC layer using GA. The input of the TSC layer is the initial population g j and an n solved in Step 5 i . The GA of the TSC layer includes a series of operations such as selection, crossover, mutation, and evolution, and generates a new population. When the maximum number of iterations reaches 100 generations, the optimal signal control timing plan is obtained

[0047] Step 7: Judgment. When the maximum number of iterations reaches 100 generations, the algorithm terminates; otherwise, let the number of iterations P = P + 1, and return to Step 5

[0048] Step 8: Terminate and end. Output the optimal decision variables g j 0 and n i .

[0049] Step 4: Propose a robust collaborative optimization model and solution algorithm for intersection signal control and variable lanes driven by the combination of historical data and real-time data. The mathematical model refers to a robust optimization model with three optional combined modes of historical data and real-time data, with the mean-standard deviation (MSD) of the average vehicle delay at the intersection as the optimization objective, with 15-day traffic historical data and 1-day real-time data as the input, and with the signal control plan for the research period as the output. The selection of the combined mode in the current period is based on the optimization effects of the above three optional combined modes in the previous period, and the optimal combined mode is used as the combined mode in the current period. The solution algorithm is a multi-mode heuristic genetic algorithm (MM-GA) specially designed for the above robust optimization model

[0050] Taking 15 minutes as the time interval, at the end of each time interval, statistically record the real-time traffic flow in each direction (excluding right turns and U-turns) at the intersection, and so on until the statistics for a period (2 hours) are completed. Store the 8 collected data in the real-time database Ω R in, ΩR = {1, 2, ..., r, ..., 8};

[0051] For the r-th time interval, the intersection signal control and variable lane robust collaborative optimization are jointly driven by the historical database Ω H and the real-time database Ω R to jointly drive the intersection signal control and variable lane robust collaborative optimization;

[0052] The objective function Z2 of the intersection signal control and variable lane robust collaborative optimization model jointly driven by the historical data and real-time data selects the average vehicle delay d as the optimization index PI2, which consists of its mean value X2 and standard deviation Y2. The model can be written as:

[0053]

[0054] Among them, γ2 is the weight parameter of the optimization model, with a value of γ2 = 0.5, and its range is γ2 ∈ [0, 1]. It can be flexibly selected according to the emphasis of traffic management. The larger the value, the more the model emphasizes the stability of signal control;

[0055] Optionally, when the model weight parameter γ2 = 0, the average value of the optimization index PI2 is selected for optimization, fully considering the efficiency of signal control;

[0056] Optionally, when the model weight parameter γ2 = 1, the standard deviation of the optimization index PI2 is selected for optimization, fully considering the stability of signal control;

[0057] The parameter π2(k) of the optimization model is the weight ratio of the historical data corresponding to the k-th time interval to the total data, which is divided into historical data π H (h) and real-time data π R (r) two categories, representing the contribution degree of each data in the historical database Ω H = {1, 2, ..., h, ..., 120} and the real-time database Ω R = {1, 2, ..., r, ..., 8} to the model; the larger the value of π2(k), the greater the contribution of the historical data corresponding to the k-th time interval to the model, otherwise vice versa;

[0058] The calculation method of the average vehicle delay d(k) for each time interval in the objective function Z2 of the optimization model is as follows:

[0059]

[0060] In the formula, j is the number of the intersection flow direction, and the NEMA coding is used to number each import direction of the intersection; d j is the delay time of flow direction j; q j is the flow of flow direction j; λ jThe green signal ratio for flow direction j; χ j The saturation degree for flow direction j; C is the signal cycle duration;

[0061] The decision variable of the optimization model is the signal control scheme corresponding to each real-time time interval

[0062] The constraint conditions of the optimization model include the minimum green light time constraint, the phase mode constraint, the cycle duration constraint, the maximum saturation constraint, the integer constraint, and the 0-1 constraint;

[0063] Minimum green light time constraint:

[0064]

[0065] Phase mode constraint:

[0066] In the standard crossroads where north-south and east-west intersect, taking the standard three-phase as an example to illustrate the constraint relationship of the phase mode. The first phase is the straight-ahead of the north-south entrance, the second phase is the left-turn of the north-south entrance, and the third phase is all turns of the east-west entrance. The constraint relationship of this phase mode is:

[0067] g1 = g5

[0068] g2 = g6

[0069] g3 = g4 = g7 = g8

[0070] g1 + g2 + g3 + 3(t y + t r ) = C

[0071] Cycle duration constraint:

[0072] C min ≤ C ≤ C max

[0073] Maximum saturation constraint:

[0074] χ j ≤ χ max

[0075] Integer constraint:

[0076]

[0077] In the formula, is the green light time of the rth time interval for flow direction j; g min is the minimum green light time, with a value of 10s; t y is the yellow light time, with a value of 4s; t r is the all-red time, with a value of 2s; C min is the minimum cycle length, with a value of 60s; Cmax is the maximum cycle length, with a value of 240 s; C is the signal control cycle; χ max is the maximum saturation threshold, with a value of 0.95; χ j is the saturation degree in the direction of flow j;

[0078] The way of jointly driving by historical data and real-time data is related to the weight ratio π2(k) of the two types of data, and three modes can be designed:

[0079] Optionally, Mode 1: When the standard deviation of the real-time database Ω R ={1, 2,..., r} in the current time interval is between [0.5, 1], for the time interval r (r = 2, 3,..., R; r ∈ Ω R ), it is characterized in that: 1) A total of 120 historical data corresponding to time intervals are obtained, and Ω H ={1, 2,..., h,..., 120}, and each historical time interval has the same weight; 2) A total of r - 1 real-time data are collected, and the weights of each pair of real-time data are the same, and its weight is equal to the weight of the entire historical data in the same period; 3) During the process of collecting real-time data, the weights of each historical data and each real-time data are different;

[0080] The weight of each historical data is:

[0081]

[0082] The weight of each real-time data is:

[0083]

[0084] Optionally, Mode 2: When the standard deviation of the real-time database Ω R ={1, 2,..., r} in the current time interval is less than 0.5, for the time interval r (r = 2, 3,..., R; r ∈ Ω R ), it is characterized in that: 1) A total of 120 historical data corresponding to time intervals are obtained, and each historical time interval has the same weight; 2) A total of r - 1 real-time data are collected, and the weights of each real-time data are the same; 3) The weight of the overall historical data is 1 / 2, and the weight of the overall real-time data is also 1 / 2; 4) During the process of collecting real-time data, the weights of each real-time data change, while the weights of each historical data remain unchanged;

[0085] The weight of each historical data is:

[0086]

[0087] The weight of each real-time data is:

[0088]

[0089] Optionally, Mode 3: the real-time database Ω of the current time interval R When the standard deviation of ={1,2,...,r} is greater than 1, for the time interval r (r = 2, 3,..., R; r ∈ Ω R ), it is characterized in that: 1) A total of 120 historical data corresponding to time intervals are obtained, and each historical time interval has the same weight; 2) A total of r - 1 real-time data are collected, and their weights are not exactly the same; 3) The weight of the (r - 1)th real-time data is 1 / 2, and the sum of the overall historical data and the remaining real-time data (the 1st to the (r - 2)th) has a weight of 1 / 2; 4) During the real-time data collection process, the weight of the latest real-time data remains unchanged at 1 / 2, while the weights of other data change;

[0090] The weight of each historical data is:

[0091]

[0092] The weight of each real-time data is:

[0093]

[0094] The decision variable signal control scheme g of the model j r is dynamically optimized, and the solution algorithm is a multi-mode heuristic genetic algorithm (MM-GA) specially designed for the above robust optimization model. The main steps include:

[0095] step1: Let the time interval r = 1.

[0096] step2: Organically combine multi-mode data. Combine real-time data with historical data.

[0097] step3: Generate the initial signal control layer population g j . According to the constraint conditions of the model, solve the constraint range of the decision variable, and generate the initial population according to the decimal coding method; when some individuals in the population do not meet the constraint conditions, regenerate new individuals according to the method until all individuals meet the constraint conditions, and use it as the initial population g j .

[0098] step4: GA solves the TSC layer. The input of the TSC layer is the initial population g j and the best n solved iThe GA of the TSC layer includes a series of operations such as selection, crossover, mutation, and evolution, and generates a new population. When the maximum number of iterations is reached, the optimal signal control timing plan is obtained by solving.

[0099] Step 5: Let the time interval r = r + 1, and return to Step 2.

[0100] Step 6: Terminate and end when the maximum time interval r is reached.

[0101] Step Five: Determine the selection of the combined mode of historical data and real-time data for the next time period, which means taking the traffic historical data of the previous 5 days as a sample, calculating its standard deviation value to judge the discrete fluctuation degree of the traffic flow, and selecting the optimal combined mode according to the interval where the standard deviation value is located as the combined mode to be adopted in the next time period;

[0102] Optionally, when the standard deviation value is greater than 1, select Mode Three as the combined optimization mode of historical data and real-time data;

[0103] Optionally, when the standard deviation value is between [0.5, 1], select Mode One as the combined optimization mode of historical data and real-time data;

[0104] Optionally, when the standard deviation value is less than 0.5, select Mode Two as the combined optimization mode of historical data and real-time data.

[0105] The above technical solution of the present invention has the following advantages compared with the prior art:

[0106] First, divide a peak / off-peak time period into several equal-length time intervals (15 min), and the division of the time intervals is arbitrary; secondly, extract the traffic data of each lane group at the intersection from the historical database according to the time intervals, and the traffic data includes the cross-sectional traffic volume and saturation flow of 15 days in the same time period; then, with the generalized saturation degree as the optimization objective, solve the robust optimization model based on mean-standard deviation (MSD) to realize the optimization of variable lane flow direction decision-making and the initial signal control plan; then, according to the time intervals, dynamically collect the real-time traffic data of each lane group, and organically combine it with the historical data of the same period, and the combination method of the two types of data is related to the weight tendency; finally, with the average vehicle delay as the optimization objective, solve the historical-real-time data-driven robust optimization model to realize the dynamic robust optimization of intersection signal control. Therefore, the intersection collaborative optimization method described in the present invention can comprehensively optimize both signal control and variable lanes, solve the contradiction between dynamic traffic demand and static traffic facilities from the supply side, improve the utilization efficiency of time and space resources, and effectively relieve the traffic congestion situation. Description of the Drawings

[0107] Figure 1It is a layout schematic diagram of lanes and induction detectors at the case intersection

[0108] Figure 2 It is a schematic diagram of two-stage features of the optimization model

[0109] Figure 3 It is Mode 1 jointly driven by historical data and real-time data

[0110] Figure 4 It is Mode 2 jointly driven by historical data and real-time data

[0111] Figure 5 It is Mode 3 jointly driven by historical data and real-time data

[0112] Figure 6 It is the structural diagram of the DN-GA algorithm

[0113] Figure 7 It is the structural diagram of the MM-GA algorithm

[0114] Figure 8 It is the model iteration process diagram: (a) overall iteration process (b) iteration process of the double-layer nested genetic algorithm (DN-GA) (c) iteration process of the multi-mode genetic algorithm (MM-GA)

[0115] Figure 9 It is the schematic diagram of the variable lane optimization result: (a) current situation of the intersection (b) optimization result during the morning peak (c) optimization result during the flat peak (d) optimization result during the evening peak

[0116] Figure 10 It is the implementation flowchart of the present invention Detailed implementation manners

[0117] Taking Figure 1 the shown intersection as the research object, a case study is carried out

[0118] Step 1: Select the optimization period and divide the time interval. The period selection means that taking 2 hours as a period length and 7:00 am as the reference point, one day is divided into 12 periods, and an initial signal control scheme and a variable lane channelization scheme for each intersection are implemented for each period respectively; the time interval division means that taking 15 minutes as a time interval length and each period is divided into 8 time intervals, and a signal control scheme is implemented for each time interval respectively on the basis of ensuring that the variable lane channelization scheme remains unchanged within the same period. Taking the morning peak as an example for time division, as shown in Table 1

[0119] Table 1 Morning peak time division

[0120]

[0121]

[0122] Step 2: Extract traffic historical data and collect real-time data for each time period, specifically referring to the statistics of traffic volumes in each direction of the intersection according to the time intervals divided in Step 1 for each time period. Among them, the historical data for each time period needs to extract the traffic volume for 15 days, and the real-time data needs to collect the traffic volume for 1 day;

[0123] According to the time interval length (15 minutes) described in Step 1, a time period (2 hours) can be divided into 8 time intervals, and the total 120 (15 * 8) recorded data obtained within the same time period of 15 days are stored in the historical database Ω H The total 8 recorded data collected within the same time period of 1 day are stored in the real-time database Ω R And label the historical data and real-time data in the order of the time intervals that appear, that is, there is Ω H ={1, 2,..., h,..., 120}, Ω R ={1, 2,..., r,..., 8}. Among them, h and r are the labels of the time intervals corresponding to any data in the historical database Ω H and the real-time database Ω R respectively.

[0124] Step 3: Propose a robust collaborative optimization mathematical model and solution algorithm for intersection signal control and variable lanes driven by historical data. The mathematical model is a robust optimization model with the mean-standard deviation (MSD) of the generalized saturation degree of the intersection as the optimization objective, the traffic historical data for 15 days as the input, and the variable lane channelization plan and initial signal control plan for the research time period as the output. The solution algorithm is a double-layer nested heuristic genetic algorithm (DN-GA) specially designed for the above robust optimization model. Calculate the variable lane channelization plan and initial signal control plan for the morning peak. The variable lane optimization results are as Figure 9 shown; the initial signal control optimization plan is shown in Table 2.

[0125] Table 2 Initial signal control optimization plan for the morning peak

[0126]

[0127] Step 4: Propose a robust collaborative optimization model and solution algorithm for intersection signal control and variable lanes driven by the combination of historical data and real-time data. The mathematical model refers to a robust optimization model with three optional combined modes of historical data and real-time data, aiming at the mean-standard deviation (MSD) of the average vehicle delay at intersections, taking 15-day traffic historical data and 1-day real-time data as inputs, and the signal control plan for the research period as the output. The selection of the combined mode in the current period is based on the optimization effects of the above three optional combined modes in the previous period, and the optimal combined mode is used as the combined mode in the current period. The solution algorithm is a multi-mode heuristic genetic algorithm (MM-GA) specially designed for the above robust optimization model. Calculate the dynamic optimization solutions of signal control under 3 modes, and the results of Mode 1 are shown in Table 3.

[0128] Table 3 Dynamic Optimization Solutions of Morning Peak Signal Control

[0129]

[0130] Step 5: Determine the selection of the combined mode of historical data and real-time data for the next period, which means taking the traffic historical data of the previous 5 days as a sample, calculating its standard deviation value to judge the discrete fluctuation degree of the traffic flow, and selecting the optimal combined mode according to the interval where the standard deviation value is located as the combined mode to be adopted in the next period. The optimization effects and recommended modes for each time interval are shown in Table 4.

[0131] Table 4 Optimization Effects of Average Vehicle Delay Time under Each Mode in Morning Peak

[0132]

[0133]

Claims

1. A robust collaborative optimization method for intersection signal control and variable lanes, characterized in that, It includes the following steps: Step 1: Select the optimization time period and divide the time intervals. The time period selection means that taking 2 hours as a time period length and 7:00 am as a reference point, one day is divided into 12 time periods, and an initial signal control plan and a variable lane channelization plan for each intersection are implemented for each time period respectively. The time interval division means that taking 15 minutes as a time interval length, each time period is divided into 8 time intervals, and a signal control plan is implemented for each time interval respectively on the basis of ensuring that the variable lane channelization plan remains unchanged within the same time period; Step 2: Extract traffic historical data and collect real-time data for each time period, specifically referring to the statistics of the traffic volume in each direction of the intersection according to the time intervals divided in Step 1 for each time period. Among them, the traffic volume of 15 days needs to be extracted for the historical data of each time period, and the traffic volume of 1 day needs to be collected for the real-time data; According to the time interval length of 15 minutes described in Step 1, a 2-hour period can be divided into 8 time intervals, and the 120 record data obtained in the same period on the 15th are stored in the historical database Ω H Second, the 8 record data collected in the same period on the 1st are stored in the real-time database Ω R And the historical data and real-time data are numbered in the order of the time intervals, that is, there is Ω H ={1, 2,..., h,..., 120}, Ω R ={1, 2,..., r,..., 8}; where h and r are the labels of the time intervals corresponding to any data in the historical database Ω H and the real-time database Ω R respectively; Step 3: Propose a robust collaborative optimization mathematical model and solution algorithm for intersection signal control and variable lanes driven by historical data. The mathematical model is a robust optimization model with the mean-standard deviation MSD of the generalized saturation degree of the intersection as the optimization objective, the traffic historical data of 15 days as the input, and the variable lane channelization plan and the initial signal control plan of the research time period as the output. The solution algorithm is a double-layer nested heuristic genetic algorithm specially designed for the above robust optimization model; The objective function Z1 of the robust collaborative optimization mathematical model for intersection signal control and variable lanes driven by historical data selects the generalized saturation degree χ of the intersection as the optimization index PI1, which consists of its mean value X1 and standard deviation Y1. The model is: Among them, γ1 is the weight parameter of the optimization model, and its value range is γ1 ∈ [0,1], which is selected according to the emphasis of traffic management. The larger the value, the more the model emphasizes the stability of signal control; When the model weight parameter γ1 = 0, the average value of the optimization index PI1 is selected for optimization, fully considering the efficiency of signal control; when the model weight parameter γ1 = 1, the standard deviation of the optimization index PI1 is selected for optimization, fully considering the stability of signal control; The parameter π1(k) of the optimization model is the weight ratio of the historical data corresponding to the k-th time interval to the overall data, and its value is representing the historical database Ω H = {1, 2,..., h,..., 120}, where each data has the same contribution degree to the model; the larger the value of π1(k), the greater the contribution of the historical data corresponding to the k-th time interval to the model, and vice versa; The calculation method of the generalized saturation degree χ(k) of each time interval in the objective function Z1 of the optimization model is as follows: In the formula, j is the number of the intersection flow direction, and the NEMA coding is used to number each import direction of the intersection; χ j is the saturation degree of flow direction j; q j is the traffic flow of flow direction j; c j is the traffic capacity of flow direction j; s j is the saturated flow rate of flow direction j; λ j is the green signal ratio of flow direction j; N j is the number of lanes of flow direction j; is the effective green light duration of flow direction j; C is the signal cycle duration; The decision variables of the model are the variable lanes n i and the signal control plan The constraint conditions of the optimization model include the minimum green light time constraint, the phase mode constraint, the cycle length constraint, the maximum saturation constraint, the integer constraint, and the 0-1 constraint; Minimum green time constraint: g j ≥ g min Phase mode constraint: In a standard cross intersection where the north-south and east-west intersect, taking the standard three-phase as an example to illustrate the constraint relationship of the phase mode. The first phase is the straight movement of the north-south entrance, the second phase is the left turn of the north-south entrance, and the third phase is all turns of the east-west entrance. The constraint relationship of this phase mode is: g1 = g5 g2 = g6 g3 = g4 = g7 = g8 g1 + g2 + g3 + 3(t y + t r ) = C Cycle length constraint: C min C ≤ C ≤ C max Maximum saturation constraint: χ j ≤ χ max Integer constraint: 0-1 constraint: n i ∈ {0, 1}, i = 1, 2, ..., 4 where g j is the green light time for flow direction j; g min is the minimum green light time, with a value of 10 s; t y is the yellow light time, with a value of 4 s; t r is the all-red time, with a value of 2 s; C min is the minimum cycle length, with a value of 60 s; C max is the maximum cycle length, with a value of 240 s; C is the signal control cycle; χ max is the maximum saturation threshold, with a value of 0.95; χ j is the saturation of flow direction j; The two sets of decision variables of the model, the variable lane n i and the signal control scheme are mutually coupled. A double-layer nested heuristic genetic algorithm specially designed for the above robust optimization model is used to solve the problem. The optimization problem is divided into a signal control layer TSC and a variable lane layer RL, including: Step 1: Initialization, calibrate relevant GA parameters and traffic parameters; the GA parameters include population size 100, number of iterations 100, crossover probability 0.8, and mutation probability 0.02; the traffic parameters include: minimum cycle length 60s, maximum cycle length 240s, yellow light time 4s, all-red time 2s, minimum green light time 10s, lost time 1s, and maximum saturation 0.95; Step 2: Let the number of iterations P = 0; Step 3: Generate the initial population g of the signal control layer j ; According to the constraint conditions of the model described in step three, solve the constraint range of the decision variables, and generate the initial population according to the decimal coding method; when some individuals in the population do not meet the constraint conditions, regenerate new individuals according to the above method until all individuals meet the constraint conditions, and use them as the initial population g j ; step4: Generate the initial variable lane layer population n i ; According to the constraint conditions of the model described in step three, solve the constraint range of the decision variables, and generate the initial population n using the 0-1 coding method i ; Step 5: Solve the RL layer using GA; the input of the RL layer is the initial population n i and a given g j ; The GA of the RL layer includes a series of operations such as selection, crossover, mutation, and evolution, and generates a new population; when the maximum number of iterations reaches 100 generations, the optimal variable lane flow setting is obtained by solving Step 6: Solve the TSC layer using GA; the input of the TSC layer is the initial population g j and an n i solved in step 5; the GA of the TSC layer includes a series of operations such as selection, crossover, mutation, and evolution, and generates a new population; when the maximum number of iterations reaches 100 generations, the optimal signal control timing plan is obtained by solving Step 7: Judgment; when reaching the maximum number of iterations of 100 generations, the algorithm terminates, otherwise let the number of iterations P = P + 1, and return to Step 5; Step 8: Terminate and end; output the optimal decision variables g j 0 and n i obtained by the optimization solution Step 4: Propose an intersection signal control and variable lane robust collaborative optimization model and solution algorithm jointly driven by historical data and real-time data. The mathematical model refers to a robust optimization model with three optional historical data and real-time data joint modes, taking the mean-standard deviation MSD of the average vehicle delay at intersections as the optimization objective, using 15-day traffic historical data and 1-day real-time data as inputs, and the signal control scheme during the research period as the output. The selection of the joint mode in the current period is based on the optimization effects of the above three optional joint modes in the previous period, and the optimal joint mode is used as the joint mode in the current period. The solution algorithm is a multi-mode heuristic genetic algorithm specially designed for the above robust optimization model; At 15-minute intervals, at the end of each time interval, count and record the real-time traffic flow in each direction at the intersection, excluding right turns and U-turns, and so on until the statistics for a 2-hour period are completed; store the 8 collected data in the real-time database Ω R where Ω R ={1, 2,..., r,..., 8}; For the r-th time interval, jointly driven by the historical database Ω H and the real-time database Ω R carry out robust collaborative optimization of intersection signal control and variable lanes; The objective function Z2 of the intersection signal control and variable lane robust collaborative optimization model jointly driven by historical data and real-time data selects the average vehicle delay d as the optimization index PI2, which consists of its mean X2 and standard deviation Y2. The model can be written as: Among them, γ2 is the weight parameter of the optimization model, and the value range is γ2 ∈ [0, 1], which can be flexibly selected according to the emphasis of traffic management. The larger the value, the more the model emphasizes the stability of signal control; When the model weight parameter γ2 = 0, the average value of the optimization index PI2 is selected for optimization, fully considering the efficiency of signal control; when the model weight parameter γ2 = 1, the standard deviation of the optimization index PI2 is selected for optimization, fully considering the stability of signal control; The parameter π2(k) of the optimization model is the weight ratio of the historical data corresponding to the k-th time interval to the overall data, which is divided into historical data π H (h) and real-time data π R (r), representing the contribution degrees of the data in the historical database Ω H ={1, 2,..., h,..., 120} and the real-time database Ω R ={1, 2,..., r,..., 8} to the model; the larger the value of π2(k), the greater the contribution of the historical data corresponding to the k-th time interval to the model, otherwise vice versa; The calculation method of the average vehicle delay d(k) in each time interval in the objective function Z2 of the optimization model is as follows: In the formula, j is the number of the intersection flow direction, and the NEMA coding is used to number each approach direction of the intersection; d j is the delay time of flow direction j; q j is the flow of flow direction j; λ j is the green signal ratio of flow direction j; χ j is the saturation of flow direction j; C is the signal cycle length; The decision variable of the optimization model is the signal control plan corresponding to each real-time time interval The constraint conditions of the optimization model include minimum green light time constraint, phase mode constraint, cycle length constraint, maximum saturation constraint, integer constraint, and 0-1 constraint; Minimum green light time constraint: Phase mode constraint: In a standard crossroads where north-south and east-west intersect, taking the standard three-phase as an example to illustrate the constraint relationship of the phase mode. The first phase is for straight traffic at the north-south entrance, the second phase is for left-turn traffic at the north-south entrance, and the third phase is for all turns at the east-west entrance. The constraint relationship of this phase mode is: g1 = g5 g2 = g6 g3 = g4 = g7 = g8 g1 + g2 + g3 + 3(t y + t r ) = C Cycle length constraint: C min C ≤ C ≤ C max Maximum saturation constraint: χ j ≤ χ max Integer constraint: wherein, is the green light time for the j-th flow in the r-th time interval; g min is the minimum green light time, with a value of 10 s; t y is the yellow light time, with a value of 4 s; t r is the all-red time, with a value of 2 s; C min is the minimum cycle length, with a value of 60 s; C max is the maximum cycle length, with a value of 240 s; C is the signal control cycle; χ max is the maximum saturation threshold, with a value of 0.95; χ j is the saturation of the j-th flow; The way of jointly driving by historical data and real-time data is related to the weight ratio π2(k) of the two types of data, and three modes are designed: Mode 1: Real-time database Ω of the current time interval R When the standard deviation of {1, 2,..., r} is between [0.5, 1], for the time interval r (r = 2, 3,..., R; r ∈ Ω R ): 1) A total of 120 historical data corresponding to the time intervals are obtained, Ω H ={1, 2,..., h,..., 120}, and each historical time interval has the same weight; 2) A total of r - 1 real-time data are collected, and the weights between each real-time data are the same, and its weight is equal to the weight of the entire historical data of the same period; 3) During the process of collecting real-time data, the weights of each historical data and each real-time data are different; The weight of each historical data is: The weight of each real-time data is: Mode 2: Real-time database Ω of the current time interval R When the standard deviation of {1, 2,..., r} is less than 0.5, for the time interval r (r = 2, 3,..., R; r ∈ Ω R ): 1) A total of 120 historical data corresponding to time intervals are obtained, and each historical time interval has the same weight; 2) A total of r - 1 real-time data are collected, and each real-time data has the same weight; 3) The weight of the overall historical data is 1 / 2, and the weight of the overall real-time data is also 1 / 2; 4) During the real-time data collection process, the weight of each real-time data changes, while the weight of each historical data remains unchanged; The weight of each historical data is: The weight of each real-time data is: Mode 3: Real-time database Ω of the current time interval R When the standard deviation of {1, 2,..., r} is greater than 1, for the time interval r (r = 2, 3,..., R; r ∈ Ω R ): 1) A total of 120 historical data corresponding to time intervals are obtained, and each historical time interval has the same weight; 2) A total of r - 1 real-time data are collected, and their weights are not exactly the same; 3) The weight of the (r - 1)-th real-time data is 1 / 2, and the sum of the overall historical data and the remaining real-time data, that is, the weights of the 1st to the (r - 2)-th are 1 / 2; 4) During the real-time data collection process, the weight of the latest real-time data remains constant at 1 / 2, while the weights of other data change; The weight of each historical data is as follows: The weight of each real-time data is as follows: Decision variable signal control scheme of the model is dynamically optimized, and the solution algorithm is a multi-mode heuristic genetic algorithm MM-GA specially designed for the above robust optimization model. The steps include: Step 1: Let the time interval r = 1; Step 2: Combine multi-mode data organically; Combine real-time data with historical data; Step 3: Generate the initial signal control layer population g j ; According to the constraint conditions of the model, solve the constraint range of the decision variables, and generate the initial population according to the decimal coding method; when some individuals in the population do not meet the constraint conditions, regenerate new individuals according to the above method until all individuals meet the constraint conditions, and use them as the initial population g j ; Step 4: Solve the TSC layer using GA; the input of the TSC layer is the initial population g j and the best n obtained from the solution i ; The GA of the TSC layer includes a series of operations such as selection, crossover, mutation, and evolution, and generates a new population; when the maximum number of iterations is reached, the optimal signal control timing plan is obtained Step 5: Let the time interval r = r + 1, and return to Step 2; Step 6: Terminate and end when the maximum time interval r is reached; Step 5: Determine the selection of the combined mode of historical data and real-time data for the next period, which means taking the traffic historical data of the previous 5 days as a sample, calculating its standard deviation to judge the discrete fluctuation degree of the traffic flow, and selecting the optimal combined mode according to the interval where the standard deviation is located as the combined mode to be adopted in the next period; When the standard deviation is greater than 1, select Mode 3 as the combined optimization mode of historical data and real-time data; When the standard deviation is between [0.5, 1], select Mode 1 as the combined optimization mode of historical data and real-time data; When the standard deviation is less than 0.5, select Mode 2 as the combined optimization mode of historical data and real-time data.

Images