An adaptive control method for signalized intersections based on digital twins

Abstract

The present invention relates to a signal intersection adaptive control method based on digital twin, comprising the following steps: before the end of each cycle, a digital twin of a signal intersection is constructed based on an improved cellular transmission model, corresponding virtual grids are delineated in a real environment, and the state data of connected vehicles in each grid are obtained; the parameters of the cellular transmission model in the virtual intersection are actively calibrated according to the state data of the connected vehicles; within a limited margin time, a scheme search algorithm is used to generate multiple alternative signal timing schemes, and the control effect of the scheme is evaluated to select the optimal scheme; after the optimal control scheme is implemented for the actual intersection, the signal intersection adaptive control enters the virtual intersection update, self-calibration and signal control optimization process of the next cycle. Compared with the prior art, the present invention takes into account the virtual-real synchronization problem between the digital twin model and the real traffic environment under non-ideal conditions, and has the advantages of closed-loop control, efficient optimization, strong expansibility and wide applicability.

Description

Technical Field

[0001] The present invention relates to the field of traffic signal control, and in particular to a signal intersection adaptive control method based on digital twins. Background Art

[0002] Time-varying traffic demand is a major challenge for traffic signal control. Signalized intersection adaptive control has emerged as a promising solution to address this problem by dynamically adjusting signal timing based on real-time traffic conditions. In recent years, real-time and reliable communication technology between on-board units and roadside units has been achieved. Connected vehicles within a certain distance from the intersection roadside unit can frequently and accurately update their driving status (speed, position, lane index, and turn direction) through wireless communication. Existing signalized intersection adaptive control methods process data in an aggregate manner and cannot tap the huge potential of fine-grained, real-time connected data.

[0003] In recent years, the development of digital twin technology has opened up new possibilities for adaptive control of signalized intersections. Digital twins are virtual representations of physical objects. By establishing a digital twin model of the actual traffic environment, signalized intersection control can be dynamically simulated, deduced, and selected, thereby improving the control effect. In addition, real-time data is continuously used to update the digital twin model to achieve closed-loop control. However, the current adaptive control method for signalized intersections based on digital twins assumes that the digital twin and the physical twin are tested under ideal synchronization conditions, without considering the synchronization time difference and computing resource limitations of the digital twin and the physical twin. Therefore, once the current adaptive control algorithm for signalized intersections based on digital twins is applied to real-world systems, it cannot achieve the expected performance. Therefore, using real-time networked vehicle data to solve the synchronization problem under limited conditions is of great significance to improving the practical application effect of the adaptive control method for signalized intersections based on digital twins. Summary of the invention

[0004] The purpose of the present invention is to provide a signal intersection adaptive control method based on digital twins, taking into account the virtual-real synchronization problem between the digital twin model and the real traffic environment under non-ideal conditions, so as to optimize the signal control effect.

[0005] The purpose of the present invention can be achieved by the following technical solutions:

[0006] A signalized intersection adaptive control method based on digital twins comprises the following steps:

[0007] S1. Building a virtual intersection: Before the end of each cycle, a digital twin of the signal-controlled intersection, i.e., a virtual intersection, is built based on the improved cellular transmission model. A virtual grid with the same position and size as the cells in the digital twin is delineated in the real environment, and the state data of the connected vehicles in each grid is obtained to simulate the dynamic evolution of lane-level traffic flow at the intersection.

[0008] S2. Self-calibration of virtual intersection parameters: The connected vehicle status data is converted into the internal flow state of the cell through data processing, and the flow state parameters of each cell in the digital twin model are updated by predicting and compensating the external flow input; the parameters of the cellular transmission model in the virtual intersection are actively calibrated according to the connected vehicle status data, so that the digital twin fits the operating status of the real traffic;

[0009] S3, signal control optimization: Considering the synchronization time difference between the digital twin and the real intersection, a scheme search algorithm oriented to optimization effect is used to generate multiple alternative signal timing schemes within a limited margin time. The control effect of the alternative signal timing schemes is deduced by the digital twin, and a score is provided for the scheme search algorithm. After the iteration is completed, the signal timing scheme with the best control effect is selected and applied to the actual signal control;

[0010] S4. Closed-loop control of intersection signals: After implementing optimized signal control on the actual intersection, the signal intersection adaptive control based on the digital twin enters the next cycle of virtual intersection update, self-calibration and signal control optimization process.

[0011] The connected vehicle status data includes real-time location, lane and steering information.

[0012] The digital twin uses an improved cellular transmission model to simulate lane-level traffic flow propagation. The traffic flow transmission calculation process between cells includes the following steps:

[0013] S11. Calculate the outflow rate S of cell j at time t j (t) and the inflow rate R j (t):

[0014] The outflow rate S of cell j at time t j (t) is calculated as follows:

[0015]

[0016] The inflow rate R of cell j at time t j (t) is calculated as follows:

[0017] R j (t) = R(k j (t)) = min{q c ,w(kjam -k j (t))}

[0018] Among them, v f Indicates free flow speed (m / s); k j (t) represents the density of cell j at time t (veh / m); k jam * represents the blockage density of the improved outflow curve (veh / m); c * Indicates the slope of the improved outflow curve (m / s); q c represents the cell volume (veh / h); w represents the shock wave velocity (m / s); k jam represents the blockage density of the original outflow curve (veh / m);

[0019] S12, the inflow rate R of the cell j calculated according to step S11 j (t) and the outflow rate S of its predecessor cell j-1 j-1 (t), calculate the flow rate Φ from the predecessor cell j-1 into the cell j at time t j (t) is shown in the following formula:

[0020] Φ j (t) = min{S j-1 (t),R j (t)}

[0021] S13. Update the density of cell j after time step Δt as shown below:

[0022]

[0023] S14. To represent the divergence and confluence relationship between lane start and end points and lanes, cells are divided into five types, including source cells, end point cells, ordinary cells, confluence cells, and divergence cells. Let the set of adjacent predecessor cells of cell i be C p (i), the set of adjacent successor cells is C n (i), the set of all cells is C, then the source cell set R, the end cell set E, the confluence cell set M, the divergence cell set D, and the common cell set N are defined in turn as shown in the following formulas:

[0024] R={i∈C|C p (i)=0,C n (i) = 1}

[0025] E={i∈C|C p (i)=1,C n (i) = 0}

[0026] M={i∈C|C p(i)>1,C n (i) = 0}

[0027] D={i∈C|C p (i)=1,C n (i)>0}

[0028] N={i∈C|C p (i)=1,C n (i) = 0}

[0029] S15. Based on the cell classification described in S14, the traffic flow transmission calculation formula of each type of cell is as follows:

[0030]

[0031]

[0032]

[0033]

[0034]

[0035] Among them, D j (t) is the traffic flow demand generated by the source cell at time t, M(j) is the set of predecessor cells that merge into cell j, β j，i represents the flow diversion ratio from j to i, D(j) is the set of downstream cells to which the flow from cell j reaches;

[0036] S16. Further generalize the cell transmission process and define the number of vehicles n owned by cell j at time t according to the cell unit length L. j and the maximum cell capacity N j The calculation formula is as follows:

[0037] n j =k j (t)L

[0038] N j =k jam L

[0039] Q j represents the cell capacity, the number of vehicles y transmitted from the predecessor cell j to the cell j+1 j+1 The calculation formula is as follows:

[0040] Q j =q c / 3600

[0041] y j+1 (t) = min{n j (t),Qj+1 ,w / v f [N j+1 -n j+1 (t)]}

[0042] n j+1 (t+Δt)=n j (t)+y j (t)-y j+1 (t)

[0043]

[0044]

[0045] The step S2 comprises the following steps:

[0046] S21. Define a virtual grid in the real environment that is consistent with the cell position and size in the digital twin;

[0047] S22: collect the position, lane and steering information sent to the roadside unit by all vehicles within a unit time step in real time, and the roadside unit counts the number of vehicles n in each cell. j and the split ratio β of the split cell j,i , calibrate the traffic flow state information of the cellular transmission model;

[0048] S23. Considering that external vehicles are still continuously inputting when testing the traffic signal scheme, assuming that the vehicle arrival will not change dramatically in two consecutive signal control cycles, the traffic volume corresponding to the vehicle arrival traffic volume curve of the previous cycle recorded by the roadside unit is used as the compensation estimate of the external vehicle input to update the traffic flow state parameter n of the cellular transmission model. j ;

[0049] S24, calculating self-correction parameters;

[0050] S25. Correct the parameters of the cellular transmission model according to the self-correction parameters.

[0051] In step S21, the cell size is determined according to the unit time step and free flow speed of the study.

[0052] The step S24 is specifically as follows: maintaining v f and k jam The saturation flow rate SFR of each lane l is calculated based on the time it takes for the connected vehicle to pass through each parking line. l (veh / h / ln) and start loss time SLT l The formula is as follows:

[0053]

[0054]

[0055]

[0056] Among them, SFR l and SLT l is the self-correction parameter, h s is the average saturated headway, T q is the dissipation time of the last queued vehicle in the convoy, T 4 is the time it takes for the fourth vehicle in the queue to dissipate, n q is the total number of observed queued vehicles, h i Defines the headway time of the i-th vehicle.

[0057] The step S25 is specifically as follows: according to the self-correction parameter SFR l and SLT l Correction of the parameter q of the cellular transport model c , w and c * , the calculation formula is as follows:

[0058] q c =SFR

[0059]

[0060]

[0061] The step S3 comprises the following steps:

[0062] S31, determining the margin time of the signal timing scheme;

[0063] S32, determining signal timing scheme parameters;

[0064] S33, initial solution generation;

[0065] S34, scheme generation: based on the initial solution, the selected optimization target, the determined signal timing scheme parameters to be optimized, and the reasonable value range of the parameters inferred from the initial solution, a signal timing scheme is iteratively generated using a variety of parameter search algorithms;

[0066] S35, scheme deduction: iteratively apply the signal timing scheme generated in step S34 to the cellular transmission model to deduce the control effect of the scheme;

[0067] S36, scheme comparison and selection: Compare the control effect evaluation indicators of different schemes, and select the signal timing scheme with the best evaluation indicators as the implementation plan for the next cycle.

[0068] The parameter search algorithm adopts grid search, genetic algorithm or Bayesian optimization algorithm.

[0069] The control effect evaluation index is defined as the total delay:

[0070]

[0071] Among them, n j (t) is the traffic flow state parameter, which represents the number of vehicles in cell j at time t, Φ j+1 (t) represents the flow rate from the predecessor cell j to the cell j+1 at time t, and Δt represents the time step.

[0072] Compared with the prior art, the present invention has the following beneficial effects:

[0073] (1) Closed-loop control: The digital twin-based adaptive control method for signalized intersections of the present invention considers the virtual-real synchronization problem between the digital twin model and the real traffic environment under non-ideal conditions, uses real-time networked vehicle data to update the state of the digital twin and self-correct parameters, and provides signal control scheme information to the actual intersection management. The data and information exchange between the digital twin and the actual intersection is continuous, forming a closed-loop process to ensure the adaptability of intersection signal control to dynamic traffic needs.

[0074] (2) Efficient optimization: The digital twin-based adaptive control of signalized intersections of the present invention can take advantage of the fine-grained and high-precision of real-time data, optimize the signal control scheme by using scheme generation, scheme deduction and comparison, and avoid the high computational cost and difficulty of solving the problem of using analytical methods. With minimal resource consumption, high-quality signal control schemes can be quickly provided.

[0075] (3) Strong scalability: When the digital twin-based signal intersection adaptive control method of the present invention is actually applied, the various steps can be assembled in a modular manner. The external input vehicle prediction, signal timing scheme, initial solution generation, scheme generation and other steps can be flexibly expanded and are not limited to the representative methods given as examples.

[0076] (4) Wide applicability: The digital twin constructed based on the cellular transmission model of the present invention simulates the traffic flow propagation within the intersection section through the stacking of cells, and is suitable for diversified intersection signal control simulation scenarios with different shapes and different numbers of lanes. BRIEF DESCRIPTION OF THE DRAWINGS

[0077] Figure 1 is a flow chart of the method of the present invention;

[0078] Figure 2 Schematic diagram of the interaction process between the physical twin and the digital twin of the present invention;

[0079] Figure 3 Traffic volume-density relationship diagram for the improved cellular transport model;

[0080] Figure 4 It is a schematic diagram of cell categories;

[0081] Figure 5 This is a framework diagram of the experimental platform corresponding to a signalized intersection adaptive control method based on digital twins;

[0082] Figure 6 A software architecture diagram of an experimental platform provided by an embodiment of the present invention;

[0083] Figure 7 A schematic diagram of an intersection traffic environment provided by an embodiment of the present invention;

[0084] Figure 8 A signal control phase diagram in a simulation verification provided by an embodiment of the present invention;

[0085] Fig. 9 A schematic diagram of the demand level and demand distribution of a test scenario in a simulation verification provided by an embodiment of the present invention;

[0086] Fig.10 A schematic diagram showing the comparison of the cumulative total waiting time results of three solution search algorithms in the simulation verification provided by the embodiment of the present invention;

[0087] Fig.11 A schematic diagram of the demand level of a group 1 test scenario in the simulation verification provided by an embodiment of the present invention;

[0088] Fig.12 A schematic diagram of demand distribution of a group 2 test scenario in a simulation verification provided by an embodiment of the present invention;

[0089] Fig.13 A schematic diagram showing the comparison of the cumulative total waiting time results of group 1 in the simulation verification provided by the embodiment of the present invention;

[0090] Fig.14 A schematic diagram showing a comparison of the accumulated total waiting time results of group 2 in the simulation verification provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0091] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0092] This embodiment provides a signalized intersection adaptive control method based on digital twins, such as Figure 1 As shown, the following steps are included:

[0093] S1. Building a virtual intersection: Before the end of each cycle, a digital twin of the signal-controlled intersection, i.e., a virtual intersection, is built based on the improved cellular transmission model. A virtual grid with the same position and size as the cells in the digital twin is delineated in the real environment, and the state data of the connected vehicles in each grid is obtained to simulate the dynamic evolution of lane-level traffic flow at the intersection.

[0094] S2. Self-calibration of virtual intersection parameters: The connected vehicle status data is converted into the internal flow state of the cell through data processing, and the flow state parameters of each cell in the digital twin model are updated by predicting and compensating the external flow input; the parameters of the cellular transmission model in the virtual intersection are actively calibrated according to the connected vehicle status data, so that the digital twin fits the operating status of the real traffic;

[0095] S3, signal control optimization: Considering the synchronization time difference between the digital twin and the real intersection, a scheme search algorithm oriented to optimization effect is used to generate multiple alternative signal timing schemes within a limited margin time. The control effect of the alternative signal timing schemes is deduced by the digital twin, and a score is provided for the scheme search algorithm. After the iteration is completed, the signal timing scheme with the best control effect is selected and applied to the actual signal control;

[0096] S4. Closed-loop control of intersection signals: After implementing optimized signal control on the actual intersection, the signal intersection adaptive control based on the digital twin enters the next cycle of virtual intersection update, self-calibration and signal control optimization process.

[0097] According to the above method, the interaction process between physical twin and digital twin is as follows: Figure 2 shown.

[0098] In this embodiment, the connected vehicle status data includes real-time location, lane, and steering information.

[0099] The digital twin uses an improved cellular transmission model to simulate lane-level traffic flow propagation. The traffic flow transmission calculation process between cells includes the following steps:

[0100] S11. Calculate the outflow rate S of cell j at time t j (t) and the inflow rate R j (t):

[0101] In the prior art, the outflow rate S' of cell j at time t is calculated j The formula for (t) is:

[0102] S' j (t) = S'(k j (t)) = min{v f k j (t),q c}

[0103] Due to the above outflow S' j (t) The calculation formula is difficult to accurately model the diversion relationship between lanes, so the improved outflow curve is as follows Figure 3 As shown, the outflow rate S of cell j at time t is j (t) is calculated as follows:

[0104]

[0105] The inflow rate R of cell j at time t j (t) is calculated as follows:

[0106] R j (t) = R(k j (t)) = min{q c ,w(k jam -k j (t))}

[0107] Among them, v f Indicates free flow speed (m / s); k j (t) represents the density of cell j at time t (veh / m); k jam * represents the blockage density of the improved outflow curve (veh / m); c * Indicates the slope of the improved outflow curve (m / s); q c represents the cell volume (veh / h); w represents the shock wave velocity (m / s); k jam Represents the blockage density (veh / m) of the original outflow curve.

[0108] S12, the inflow rate R of the cell j calculated according to step S11 j (t) and the outflow rate S of its predecessor cell j-1 j-1 (t), calculate the flow rate Φ from the predecessor cell j-1 into the cell j at time t j (t) is shown in the following formula:

[0109] Φ j (t) = min{S j-1 (t),R j (t)}

[0110] S13. Update the density of cell j after time step Δt as shown below:

[0111]

[0112] S14. To represent the divergence and confluence relationships between lane start and end points and lanes, cells are divided into five types, including source cells, end point cells, common cells, confluence cells, and divergence cells. Figure 4 As shown, let the set of adjacent predecessor cells of cell i be C p (i), the set of adjacent successor cells is C n (i), the set of all cells is C, then the source cell set R, the end cell set E, the confluence cell set M, the divergence cell set D, and the common cell set N are defined in turn as shown in the following formulas:

[0113] R={i∈C|C p (i)=0,C n (i) = 1}

[0114] E={i∈C|C p (i)=1,C n (i) = 0}

[0115] M={i∈C|C p (i)>1,C n (i) = 0}

[0116] D={i∈C|C p (i)=1,C n (i)>0}

[0117] N={i∈C|C p (i)=1,C n (i) = 0}

[0118] S15. Based on the cell classification described in S14, the traffic flow transmission calculation formula of each type of cell is as follows:

[0119]

[0120]

[0121]

[0122]

[0123]

[0124] Among them, D j (t) is the traffic flow demand generated by the source cell at time t, M(j) is the set of predecessor cells that merge into cell j, β j，i represents the flow diversion ratio from j to i, D(j) is the set of downstream cells to which the flow from cell j reaches;

[0125] S16. Further generalize the cell transmission process and define the number of vehicles n owned by cell j at time t according to the cell unit length L. j and the maximum cell capacity N j The calculation formula is as follows:

[0126] n j =k j (t)L

[0127] N j =k jam L

[0128] Q j represents the cell capacity, the number of vehicles y transmitted from the predecessor cell j to the cell j+1 j+1 The calculation formula is as follows:

[0129] Q j =q c / 3600

[0130] y j+1 (t) = min{n j (t),Q j+1 ,w / v f [N j+1 -n j+1 (t)]}

[0131] n j+1 (t+Δt)=n j (t)+y j (t)-y j+1 (t)

[0132]

[0133]

[0134] In a preferred embodiment, the real-time position and steering information of the connected vehicle is converted into state parameters in the cellular transmission model to achieve traffic flow state update and model parameter self-correction of the improved cellular transmission model, including the following steps:

[0135] S21. In the real environment, define a virtual grid that is consistent with the cell position and size in the digital twin; the cell size can be determined based on the unit time step and free flow speed of the study.

[0136] S22: collect the position, lane and steering information sent to the roadside unit by all vehicles within a unit time step in real time, and the roadside unit counts the number of vehicles n in each cell. j and the split ratio β of the split cell j,i , calibrate the traffic flow state information of the cellular transmission model.

[0137] S23. Considering that external vehicles are still continuously inputting when testing the traffic signal scheme, assuming that the vehicle arrival will not change dramatically in two consecutive signal control cycles, the traffic volume corresponding to the vehicle arrival traffic volume curve of the previous cycle recorded by the roadside unit is used as the compensation estimate of the external vehicle input to update the traffic flow state parameter n of the cellular transmission model. j .

[0138] S24. In order to adapt the improved cellular transmission model to the dynamic traffic flow propagation, the parameter self-correction is used to ensure that the cellular transmission model accurately represents the real intersection traffic flow. f and k jam The saturation flow rate SFR of each lane l is calculated based on the time it takes for the connected vehicle to pass through each parking line. l (veh / h / ln) and start loss time SLT l , the formula is as follows:

[0139]

[0140]

[0141]

[0142] Among them, SFR l and SLT l is the self-correction parameter, h s is the average saturated headway, T q is the dissipation time of the last queued vehicle in the convoy, T 4 is the time it takes for the fourth vehicle in the queue to dissipate, n q is the total number of observed queued vehicles, h i Defines the headway time of the i-th vehicle.

[0143] S25, according to the self-correction parameter SFR l and SLT l Correction of the parameter q of the cellular transport model c , w and c * , the calculation formula is as follows:

[0144] q c =SFR

[0145]

[0146]

[0147] In a preferred embodiment, step S3 includes the following steps:

[0148] S31. Determine the margin time of the signal timing scheme: The deduction of each signal timing scheme requires a certain amount of time. Taking the average computing power of the current RSU as an example, the test time length of each scheme is 0.01s. Obviously, the more schemes tested, the more computing time the system spends, and more margin time should be reserved before the next cycle. However, due to the differences in external demand input and internal traffic flow evolution mechanisms, longer margin times lead to greater differences between real traffic conditions and simulation models. Therefore, the maximum number of schemes to be tested is limited. Taking into account the synchronization time difference between the digital twin and the real intersection, taking a 2s margin time as an example, the maximum number of schemes to be tested for each signal control decision is 200.

[0149] S32, determine the signal timing scheme parameters: Take the four-phase signal timing scheme as an example. Since adaptive signal timing is sensitive to time, the signal control scheme should not be tested in high dimensions under limited conditions. For example, by fixing the phase structure and sequence, adjusting the green light time (g 1 ,g 2 ,g 3 ,g 4 ) can produce different signal control schemes.

[0150] S33. Initial solution generation: Obtaining a better initial solution for the timing scheme through traditional adaptive signal control methods can speed up the efficiency of scheme search. From the basic Webster timing method to various networked adaptive signal control methods, the optimization system can be flexibly connected in a modular form.

[0151] S34, solution generation: Based on the initial solution, the selected optimization target (such as queue length, average vehicle delay, etc.), the determined signal timing solution parameters to be optimized (such as g 1 ,g 2 ,g 3 ,g 4 ), as well as the reasonable range of parameter values ​​inferred from the initial solution, and a variety of parameter search algorithms, such as grid search, genetic algorithm, and Bayesian optimization algorithm, are used to iteratively generate signal timing schemes.

[0152] S35, scheme deduction: iteratively apply the signal timing scheme generated in step S34 to the cellular transmission model to deduce the control effect of the scheme;

[0153] In this embodiment, the control effect evaluation index is defined as the total delay:

[0154]

[0155] Among them, n j (t) is the traffic flow state parameter, which represents the number of vehicles in cell j at time t, Φj+1 (t) represents the flow rate from the predecessor cell j to the cell j+1 at time t, and Δt represents the time step.

[0156] S36, scheme comparison and selection: Compare the control effect evaluation indicators of different schemes, and select the signal timing scheme with the best evaluation indicators as the implementation plan for the next cycle.

[0157] There are differences in the optimization process based on different solution search methods. The characteristic of using grid search is to first generate all the solutions to be deduced, and then uniformly compare the control effects of all solutions; genetic algorithms and Bayesian optimization iteratively generate solutions and deduce solutions. The following uses the genetic algorithm solution search method as an example to illustrate the iterative process of solution generation, deduction and comparison.

[0158] 1) Generate initial solutions. Determine the initial solution population according to the Webster method. The population size can be set as needed. For example, if it is set to 10, the initial population is scaled and expanded by a certain proportion with the Webster initial solution, and 10 initial solutions are obtained as the initial population.

[0159] 2) Define the fitness function. The inverse of the total delay defined in claim 4 is used as the fitness function. The higher the fitness, the better the control effect of the solution.

[0160] 3) Selection: Through a certain selection strategy, the solution with higher fitness is selected from the current solution population as the parent generation to promote the inheritance of excellent genes.

[0161] 4) Crossover. The selected parent individuals exchange genes according to certain crossover operators (such as single-point crossover, multi-point crossover, etc.) to produce new individuals. These new individuals are called offspring.

[0162] 5) Mutation. Gene mutation is performed on offspring individuals according to a certain mutation probability to introduce some randomness and diversity. Mutation helps avoid falling into a local optimal solution.

[0163] 6) Generate the next generation population. The offspring individuals and some of the unchanged parent individuals form a new population, replacing the previous generation population.

[0164] 7) The schemes within the population are deduced in parallel to obtain the fitness evaluation of the scheme deduced, which is compared with the recorded historical optimal fitness scheme. If a higher fitness appears, the optimal fitness is updated.

[0165] 8) When the number of searches reaches the requirement, stop searching and output the solution corresponding to the highest fitness.

[0166] This example verifies and analyzes the adaptive control method for signalized intersections based on an empirical verification scenario. Figure 5 )Using Urban Traffic Simulation (SUMO), C++ and Python integration to build:

[0167] 1) SUMO is used as a data source and operation object. At each time step, the information of all connected vehicles is obtained from SUMO and relayed to Python through the traffic control interface function (TraCI);

[0168] 2) The cellular transport model is written in C++ and encapsulated as a Python extension package using the Pybind11 library;

[0169] 3) Python is the middle layer for implementing connected vehicle data conversion, preprocessing and solution search algorithms.

[0170] At each time step, the adaptive control scheme of the signalized intersection will be optimized and calculated and sent back to SUMO through TraCI. Based on the above experimental platform, the adaptive control method of the signalized intersection based on digital twin technology is compared and analyzed for traffic demand level and traffic demand distribution. The proposed software architecture is as follows: Figure 6 shown.

[0171] Based on the experimental platform, this embodiment creates a research site of an isolated intersection (such as Figure 7 The traffic volume and vehicle parameters are set to the SUMO default values. The signal timing scheme is configured as follows Figure 8 shown.

[0172] The performance of the signalized intersection adaptive control based on digital twin (DT-ATSC-GS, DT-ATSC-GA, DT-ATSC-BO) using three solution search algorithms, grid search (GS), genetic algorithm (GA), and Bayesian optimization (BO), was tested in two different demand scenarios. For both cases, it is assumed that the traffic volume in the opposite directions is at the same level, i.e., V N =V S and V W =V E .like Fig. 9 As shown in (a) and (b), scenario 1 is designed to test the control algorithms’ ability to adapt to changes in demand levels, while scenario 2 is created to compare their performance under changing demand splits. In both scenarios, the ratio of left turn: straight: right turn is always equal to 1:7:2.

[0173] The present invention introduces the cumulative total waiting time (CTWT) to measure the control performance of the scheme, as shown in the following formula:

[0174]

[0175] Where n is the total number of vehicles in the intersection at time t, WT ij is a Boolean value indicating whether vehicle i is waiting at time j. If vehicle V ij The speed is less than the threshold, then WT ij is equal to 1, otherwise it is equal to 0.

[0176] Fig.10 The signal control results within 30 minutes (including 24 cycles) are shown. From the experimental results, it can be seen that among the three scheme search algorithms, the trend line of DT-ATSC-BO is lower than that of DT-ATSC-GA and DT-ATSS-GS, which shows that when the demand level of the intersection changes, the control performance of DT-ATSS-BO is better than the other two. When the timestamp is 850s, the CTWT difference between DT-ATSC-BO and DT-ATSC-GS even reaches 7000s. Fig.10 The same trend can be found in (b), which shows that DT-ATSC-BO is also better than DT-ATSC-GA and DT-ATSC_GS in adapting to changes in demand splitting. As the simulation time increases, the difference between DT-ATSC-BO and DT-ATSC-GS becomes more significant. When the simulation time is 1700 seconds, the difference between them becomes 10000 seconds. Therefore, under limited search schemes (this embodiment assumes a maximum of 200 schemes), applying Bayesian optimization can find a better signal control scheme.

[0177] The performance of DT-ATSC-BO and DT-ATSC-GS is further tested in more scenarios compared with the adaptive Webster method proposed by Genders W, Razavi S et al. in "An Open-Source Framework for Adaptive Traffic Signal Control". The adaptive Webster signal control mode generates the signal timing scheme for the next cycle through the Webster theory at the end of each cycle based on the aggregated traffic demand collected in the current cycle. In addition, 6 scenarios are designed based on the above 2 scenarios, such as Fig.11 and Fig.12 The first group represents scenarios where the demand level changes (4 different peak flows), and the second group represents scenarios where the demand distribution changes (4 rates of change).

[0178] Fig.13It shows that the trend line of the proposed DT-ATSC-BO is lower than that of DT-ATSC-GS and Adaptive Webster. The results show that the two DT-ATSC methods maintain good control performance compared with the adaptive Webster method. The above results further show that by using the solution solved by the Webster method as the initial solution, the controller does find a better solution after searching and stimulating through grid search or Bayesian optimization. At the same time, it is further noted that when the peak volume of the group becomes V N =V S =350veh / h and V W =V E =700veh / h, the trend lines of DT-ATSC-BO and DT-ATSC-GS intertwine at 850s. By examining the simulation records, we found that in some methods, once traffic congestion (such as secondary queuing) occurs, DT-ATSC-BO may generate inappropriate signal control schemes.

[0179] In addition, Table 1 shows the average delay and control effect comparison of the three ATSC methods.

[0180] Table 1

[0181]

[0182] The results show that DT-ATSC-BO has the lowest average delay in the first three cases, which is 16.6%, 53.7% and 22.8% lower than Adaptive Webster, respectively. In the second case, DT-ATSS-GO shows the most significant improvement of 30.4% compared with DT-ATSC-GS. In addition, in the fourth case, DT-based ATSC still leads Adaptive Webster by 23.4%.

[0183] The results of group 2 are as follows Fig.14 (a), (b), (c) and (d). DT-ATSC-BO and DT-ATSC-GS perform significantly better than DT-ATSC. Interestingly, as V N (V S ) and V W (V E ) becomes larger, the performance gap between DT-ATSC-BO, DT-ATSC-GS, and Adaptive Webster becomes smaller. However, this trend is negligible, which indicates that the change rate of demand segmentation is not an important factor affecting the performance of the adaptive control method.

[0184] The present invention relates to a signal intersection adaptive control optimization method based on digital twins, which constructs a cellular transmission model based on the vehicle position, steering and other information recorded by the networked vehicle data, updates the cellular state, corrects the model parameters, and provides an optimized signal control solution for intersection signal management. The present invention takes into account the synchronization problem under limited conditions and has the advantages of closed-loop control, efficient optimization, strong scalability and wide applicability.

[0185] The preferred specific embodiments of the present invention are described in detail above. It should be understood that a person skilled in the art can make many modifications and changes based on the concept of the present invention without creative work. Therefore, any technical solution that can be obtained by a person skilled in the art through logical analysis, reasoning, or limited experiments based on the concept of the present invention on the basis of the prior art should be within the scope of protection determined by the claims.

Claims

1. A signalized intersection adaptive control method based on digital twin, characterized in that: The following steps are involved: S1. Building a virtual intersection: Before the end of each cycle, a digital twin of the signal-controlled intersection, i.e., a virtual intersection, is built based on the improved cellular transmission model. A virtual grid with the same position and size as the cells in the digital twin is delineated in the real environment, and the state data of the connected vehicles in each grid is obtained to simulate the dynamic evolution of lane-level traffic flow at the intersection. S2. Self-calibration of virtual intersection parameters: The connected vehicle status data is converted into the internal flow state of the cell through data processing, and the flow state parameters of each cell in the digital twin model are updated by predicting and compensating the external flow input; the parameters of the cellular transmission model in the virtual intersection are actively calibrated according to the connected vehicle status data, so that the digital twin fits the operating status of the real traffic; S3, signal control optimization: Considering the synchronization time difference between the digital twin and the real intersection, a scheme search algorithm oriented to optimization effect is used to generate multiple alternative signal timing schemes within a limited margin time. The control effect of the alternative signal timing schemes is deduced by the digital twin, and a score is provided for the scheme search algorithm. After the iteration is completed, the signal timing scheme with the best control effect is selected and applied to the actual signal control; S4, closed-loop control of intersection signals: After implementing optimized signal control on the actual intersection, the adaptive control of the signal intersection based on the digital twin enters the next cycle of virtual intersection update, self-calibration and signal control optimization process; The digital twin uses an improved cellular transmission model to simulate lane-level traffic flow propagation. The traffic flow transmission calculation process between cells includes the following steps: S11. Calculate the outflow rate S of cell j at time t j (t) and the inflow rate R j (t): The outflow rate S of cell j at time t j (t) is calculated as follows: The inflow rate R of cell j at time t j (t) is calculated as follows: Rj(t)=R(kj(t))=min{qc,w(kjam-kj(t))} Among them, v f represents the free flow speed; k j (t) represents the density of cell j at time t; k jam * represents the blockage density of the improved outflow curve; c * represents the slope of the improved outflow curve; q c represents the cell capacity; w represents the shock wave velocity; k jam represents the blocking density of the original outflow curve; S12, the inflow rate R of the cell j calculated according to step S11 j (t) and the outflow rate S of its predecessor cell j-1 j-1 (t), calculate the flow rate Φ from the predecessor cell j-1 into the cell j at time t j (t) is shown in the following formula: Φ j (t)=min{S j-1 (t),R j (t)} S13. Update the density of cell j after time step Δt as shown below: S14. To represent the divergence and confluence relationship between lane start and end points and lanes, cells are divided into five types, including source cells, end point cells, ordinary cells, confluence cells, and divergence cells. Let the set of adjacent predecessor cells of cell i be C p (i), the set of adjacent successor cells is C n (i), the set of all cells is C, then the source cell set R, the end cell set E, the confluence cell set M, the divergence cell set D, and the common cell set N are defined in turn as shown in the following formulas: R={i∈C|C p (i)=0,C n (i)=1} E={i∈C|C p (i)=1,C n (i)=0} M={i∈C|C p (i)>1,C n (i)=0} D={i∈C|C p (i)=1,C n (i)>0} N={i∈C|C p (i)=1,C n (i)=0} S15. Based on the cell classification described in S14, the traffic flow transmission calculation formula of each type of cell is as follows: Among them, D j (t) is the traffic flow demand generated by the source cell at time t, M(j) is the set of predecessor cells that merge into cell j, β j，i represents the flow diversion ratio from j to i, D(j) is the set of downstream cells to which the flow from cell j reaches; S16. Further generalize the cell transmission process and define the number of vehicles n owned by cell j at time t according to the cell unit length L. j and the maximum cell capacity N j The calculation formula is as follows: n j =k j (t)L N j =k jam L Q j represents the cell capacity, the number of vehicles y transmitted from the predecessor cell j to the cell j+1 j+1 The calculation formula is as follows: Q j =q c / 3600 yj +1 (t)=min{n(t),Qj +1 ,w / vf[Nj +1 -nj +1 (t)]} n j+1 (t+Δt)=n j (t)+y j (t)-y j+1 (t) yj +1 (t)=min{n(t),Qj +1 ,w / vf[Nj +1 -nj +1 (t)], 2. The signalized intersection adaptive control method based on digital twin according to claim 1 is characterized in that: The connected vehicle status data includes real-time location, lane and steering information.

3. The signalized intersection adaptive control method based on digital twin according to claim 1 is characterized in that: The step S2 comprises the following steps: S21. Define a virtual grid in the real environment that is consistent with the cell position and size in the digital twin; S22: collect the position, lane and steering information sent by all vehicles to the roadside unit in real time within a unit time step, and the roadside unit counts the number of vehicles n in each cell. j and the split ratio β of the split cell j,i , calibrate the traffic flow state information of the cellular transmission model; S23. Considering that external vehicles are still continuously inputting when testing the traffic signal scheme, assuming that the vehicle arrival will not change dramatically in two consecutive signal control cycles, the traffic volume corresponding to the vehicle arrival traffic volume curve of the previous cycle recorded by the roadside unit is used as the compensation estimate of the external vehicle input to update the traffic flow state parameter n of the cellular transmission model. j ; S24, calculating self-correction parameters; S25. Correct the parameters of the cellular transmission model according to the self-correction parameters.

4. The signalized intersection adaptive control method based on digital twin according to claim 3 is characterized in that: In step S21, the cell size is determined according to the unit time step and free flow speed of the study.

5. The signalized intersection adaptive control method based on digital twin according to claim 3 is characterized in that: The step S24 is specifically as follows: maintaining v f and k jam The saturation flow rate SFR of each lane l is calculated based on the time it takes for the connected vehicle to pass through each parking line. l and Start-up Lost Time SLT l The formula is as follows: Among them, SFR l and SLT l is the self-correction parameter, h s is the average saturated headway, T q is the dissipation time of the last queued vehicle in the convoy, T4 is the dissipation time of the fourth queued vehicle, and n q is the total number of observed queued vehicles, h i Defines the headway time of the i-th vehicle.

6. The method for adaptive control of signalized intersections based on digital twins according to claim 5 is characterized in that: The step S25 is specifically as follows: according to the self-correction parameter SFR l and SLT l Correction of the parameter q of the cellular transport model c , w and c * , the calculation formula is as follows: q c =SFR Among them, SFR represents the saturation flow rate of the cell, and SLT represents the startup loss time of the cell.

7. The signalized intersection adaptive control method based on digital twin according to claim 1 is characterized in that: The step S3 comprises the following steps: S31, determining the margin time of the signal timing scheme; S32, determining signal timing scheme parameters; S33, initial solution generation; S34, scheme generation: based on the initial solution, the selected optimization target, the determined signal timing scheme parameters to be optimized, and the reasonable value range of the parameters inferred from the initial solution, a signal timing scheme is iteratively generated using a variety of parameter search algorithms; S35, scheme deduction: iteratively apply the signal timing scheme generated in step S34 to the cellular transmission model to deduce the control effect of the scheme; S36, scheme comparison and selection: Compare the control effect evaluation indicators of different schemes, and select the signal timing scheme with the best evaluation indicators as the implementation plan for the next cycle.

8. The signalized intersection adaptive control method based on digital twin according to claim 7 is characterized in that: The parameter search algorithm adopts grid search, genetic algorithm or Bayesian optimization algorithm.

9. The signalized intersection adaptive control method based on digital twin according to claim 7 is characterized in that: The control effect evaluation index is defined as the total delay: Among them, n j (t) is the traffic flow state parameter, which represents the number of vehicles in cell j at time t, Φ j+1 (t) represents the flow rate from the predecessor cell j to the cell j+1 at time t, and Δt represents the time step.

Images