Intersection signal timing optimization method based on queue length and delay

Abstract

The application provides an intersection signal optimization timing method based on queue length and delay, comprising the following steps: S1, road end equipment collects traffic situation data; S2, historical data aggregation and algorithm data preparation; S3, timing optimization model based on queue length and signal control delay; relying on the traffic data collected by the urban traffic flow monitoring equipment and the phase control data of the signal machine, the fine timing result of the urban intersection signal control scheme is calculated more quickly and conveniently, a set of reasonable timing optimization model for the continuous and real traffic passing demand of the intersection is provided, the more accurate data in different time periods is accurately controlled and optimized, the same set of timing scheme method is used, and the space-time data resources of the intersection equipment and the signal machine are fully utilized.

Description

Technical Field

[0001] This invention relates to the field of urban traffic signal optimization and control, and more specifically, to a method for optimizing intersection signal timing based on queue length and delay. Background Technology

[0002] Existing signal control timing optimization technologies rely heavily on the experience of traffic signal engineers or experts. Based on their familiarity with urban intersections or roads, they simply compile traffic flow data during morning and evening rush hours or conduct long-term observations of intersection videos to determine the approximate traffic flow distribution patterns. They then provide a rough signal timing scheme, which is implemented and tested. If the timing is unsuitable, it is fine-tuned and tested again, and this process is repeated multiple times to arrive at a suitable scheme. In other areas of automatic signal control, there are technologies that optimize intersection delays or saturation through modeling, but these are limited in their applicability and cannot be applied on a large scale online.

[0003] Patent application CN201910739891.5 discloses a traffic signal optimization control method and device. This method acquires traffic flow data from intersection signal systems, historical traffic flow data captured by intersection electronic police cameras, and cross-sectional traffic flow data at the intersection; preprocesses the data and extracts training samples; and performs reinforcement learning based on the training samples to generate a traffic signal timing scheme. This method utilizes reinforcement learning and other techniques for sample training. After obtaining the trained model, it combines it with real-time dynamic signal optimization control based on actual signal control and traffic data. The implementation process is relatively complex, requiring support and optimization testing from specialized fields, making it unsuitable for large-scale deployment. Most signalized intersections still rely on conventional offline schemes for control and long-term operation.

[0004] Patent application CN201610696748.9 discloses a traffic signal optimization control method based on strategy iteration and clustering. This method selects a fixed phase sequence control scheme as the signal control scheme to be optimized, samples and records traffic states, control actions, and the number of vehicles leaving the stop line, and performs K-means discretization and normalization on the traffic states in the samples to obtain the centroids of the traffic states. Based on the pre-set discrete traffic states corresponding to the centroids, it queries the control strategy, obtains the control actions, and sends them to the intersection signal controller for real-time traffic state control. The shortcomings of the prior art are:

[0005] 1. Insufficient automation and low efficiency. Current timing optimization work relies too heavily on manual intervention by signal optimization experts, requiring a large number of domain experts to work on signal control optimization. However, large and medium-sized cities have thousands or even tens of thousands of signal control intersections, making it impossible to rely entirely on signal control experts. Furthermore, even with expert intervention, it is impossible to accurately count traffic flow in each time period. It is even more difficult to comprehensively consider traffic demand over multiple historical days, and traffic aggregation using the average or maximum value of historical traffic flow cannot reflect the true hourly traffic situation.

[0006] 2. High cost, unsuitable for large-scale use. Existing patents mention several signal control optimization methods related to queue length or based on iterative strategies, which are based on real-time control. Since the states of urban road intersections influence each other in real time and change uncontrollably, comprehensive intersection signal control across the entire area is necessary to ensure that real-time control does not affect the control of other intersections. Single-intersection real-time control is not suitable for large-scale deployment without regional control. A more practical approach is to use a general offline signal timing generation method applicable to most intersections that can automatically generate timing schemes based on data. Summary of the Invention

[0007] The purpose of this invention is to provide a method for optimizing the timing of intersection signals based on queue length and delay, so as to solve the problems mentioned in the background art.

[0008] To achieve the above objectives, the present invention provides the following technical solution:

[0009] A method for optimizing intersection signal timing based on queue length and delay includes the following steps:

[0010] S1: Data collection and calculation of traffic situation data by roadside equipment;

[0011] S2: Historical data aggregation and algorithm data preparation;

[0012] S3: Timing optimization model based on queue length and signal control delay.

[0013] Furthermore, step S1 is divided into the following steps:

[0014] S01: Roadside equipment for lane-level data collection at intersections.

[0015] At signalized intersections, appropriate traffic flow detection devices (such as electronic police, traffic flow detectors, radar detectors, etc.) are installed to detect the traffic flow data of each lane on each branch of the intersection in real time and upload it to the corresponding data integration and management platform.

[0016] S02: Historical accumulation and storage of vehicle passage data

[0017] Real-time data received from roadside equipment is processed at a minimum granularity of 5 minutes to calculate traffic flow data for each lane at each intersection branch, and stored in a database for historical data storage. The storage structure is as follows:

[0018]

[0019] S03: Historical Accumulation and Storage of Traffic Situation Data

[0020] (1) Calculation and storage of queue length in 5 minutes:

[0021] The system detects the actual lane queue length data through detectors and stores it in the system in the same format as the traffic flow data in S02. If the queue length sent by the detector is less than 5 minutes (e.g., 1 minute), multiple data points received within 5 minutes are taken, the maximum value is calculated, and stored as the queue length within the 5-minute granularity.

[0022] (2) Calculation and storage of saturation over 5 minutes:

[0023] The saturation is calculated using traffic data obtained from S02 statistics, combined with the signal control scheme currently in operation at the intersection. The saturation of all lanes and the intersection as a whole is calculated and stored over a 5-minute period.

[0024] Lane green light ratio = Lane green light passage time / Signal cycle

[0025] Lane saturation = 5-minute flow rate converted to hourly flow rate / (lane saturation flow rate * green ratio)

[0026] Intersection saturation = maximum saturation of all lanes

[0027] Furthermore, step S2 is divided into the following steps:

[0028] S04: Historical Situation Data Aggregation

[0029] Optimize the historical situation data used by performing aggregation operations according to date and time period.

[0030] By default, the signal control scheme used at a typical intersection is a one-day scheme covering the entire day. However, depending on actual needs, the day can be divided into multiple time periods, such as morning rush hour, evening rush hour, off-peak, and nighttime. For each time period, historical data can be aggregated and the timing optimized to generate a complete scheme.

[0031] Historical data aggregation scheme for a single intersection within the same time period (e.g., the entire day from 0:00 to 24:00):

[0032] For each lane at the intersection, historical data for 7 days is selected. The aggregation method is to take the average of the 7 days within the same 5-minute period to obtain the averaged aggregated data of 288 time points for each lane, including lane traffic, lane queue length data, lane saturation, and intersection saturation.

[0033] S05: Sliding window aggregates lane hourly traffic data

[0034] All 5-minute time-point traffic data obtained from S04 are converted into hourly traffic data; queue length data is not processed. The conversion method uses a sliding window summation method to calculate and convert the data for each time point into hourly traffic.

[0035] Sliding window summation method:

[0036] (1) For 288 data points over 24 hours, the data from the first half hour (1-6) and the last half hour (283-288) are padded to the end and beginning of the data column, respectively, expanding the 24 hours to 25 hours and 300 data points for calculation. Then, the sliding window is set to a time length of 1 hour (12 time points) for sliding calculation and summation. For example, the first point is the summation of time points (1 to 12), the second point is the summation of time points (2-13), and so on, the i-th time point is the summation of time points (i to i+11).

[0037] (2) For each time point, the sum of the 12 time points before and after it is used, which is converted into hourly traffic data, resulting in a total of 288 time points for each lane. This achieves the effect of smoothing the data and converts it into the hourly traffic volume that needs to be used, which serves as the data foundation for subsequent timing optimization algorithms.

[0038] Furthermore, step S3 is divided into the following steps:

[0039] S07: In the process of optimizing traffic flow based on queue length data, the queue length causes the actual traffic demand to exceed the actual traffic statistics. Therefore, it needs to be converted into actual traffic flow for optimization. A minimum delay optimization model based on queue length improvement is proposed.

[0040] The conversion is performed using the following steps;

[0041] (1) Calculate the average number of vehicles in queue.

[0042] Turning is generally divided into left turn, straight turn, and right turn, and may involve multiple lanes.

[0043] Average vehicle length (including distance between vehicles) = 3m

[0044] Turning queue length = average queue length of multiple turning lanes

[0045] Number of vehicles in queue = Queue length / Average vehicle length

[0046] (2) Queuing traffic flow conversion

[0047] Based on the intersection saturation data calculated by S03, the intersection status is divided into two states:

[0048] A saturation level ≥ 0.7 indicates a saturated state, while a saturation level < 0.7 indicates an unsaturated state.

[0049] Calculate the queuing flow transition for each of the two states, and add the signal control flow:

[0050] Basic turning flow = Sum of flow rates in corresponding lanes

[0051] a. Unsaturated state:

[0052] Signal-controlled turning flow = Number of vehicles in queue * Number of lanes + Basic turning flow

[0053] b. Saturation state

[0054] Signal-controlled turning flow = Number of vehicles in queue * Number of lanes * 1.2 + Basic turning flow

[0055] The signal-controlled turning flow of the intersection is obtained for each turn within a time period, with 288 granularities throughout the day.

[0056] S08: Optimize timing scenario strategies and parameter settings

[0057] In actual optimization of timing, different strategies may be selected depending on the scenario, such as weekday / weekend scenarios, weekday morning / evening peak scenarios, etc. Different strategies should be used to select the traffic for optimization.

[0058] Recommended scenario segmentation, using the signal control traffic in S07:

[0059] (1) All-day single-solution scenario

[0060] All day: Traffic flow controlled by traffic control = maximum traffic flow during the entire day.

[0061] (2) Peak and off-peak scenes throughout the day

[0062] Morning rush hour: Traffic flow controlled by traffic control = maximum traffic flow during the morning rush hour;

[0063] Evening peak hours: Traffic flow controlled by traffic control = maximum traffic flow during the evening peak hours;

[0064] Midday off-peak period: Traffic flow rate controlled by traffic signals = average traffic flow rate during midday off-peak period;

[0065] Nighttime off-peak hours: Traffic flow controlled by traffic signals = average traffic flow throughout the day;

[0066] The specific time periods corresponding to peak and off-peak periods are set according to the actual situation.

[0067] (3) Weekday / Non-week scenarios

[0068] Weekdays: Traffic flow redirection = Maximum traffic volume throughout the weekday. Non-weekdays: Traffic flow redirection = Total traffic volume throughout the non-weekday * 0.8

[0069] S09: Timing optimization model based on queue length and delay

[0070] This model only performs timing adjustments for pre-designed phases, without performing phase design work. Generally, signal timing optimization work involves optimizing and adjusting the timing based on the original operating phases of the intersection signal controllers. This model provides a solution for refined optimal timing based on data, which can improve the efficiency of manual timing work by signal experts when the data is complete.

[0071] The model optimizes the total timing of each phase of the signal scheme (which may actually be the green light time + yellow light time). The model is applicable to many common phases: such as oncoming straight + oncoming left turn, two straight and one left turn, oncoming straight + oncoming left turn + one phase, etc. It only needs to calculate the turning conversion data according to the actual phase. Here, we take the most common set of phases as an example. The scheme has a total of 4 phases: east-west straight + east-west left turn + north-south straight + north-south left turn.

[0072] Furthermore, step S09 is divided into the following steps:

[0073] Step 1: Phase timing range limitation, particle swarm algorithm initialization

[0074] Based on the set 4 phases, set the minimum and maximum ranges, and the period range, and modify them according to actual needs.

[0075] The default minimum value is 15 seconds, and the maximum value is 60 seconds.

[0076] Phase 1: East-West straight line, range [15, 60]

[0077] Phase 2: Turn left (east / west), range [15, 60]

[0078] Phase 3: North-South straight, range [15, 60]

[0079] Phase 4: North-South left turn, range [15, 60]

[0080] Signal period: [60, 200]

[0081] Signal period = sum of the timing times of the four phases

[0082] Particle Swarm Optimization Initialization:

[0083] According to the particle swarm optimization algorithm, particles are the variables to be optimized, namely, phases 1, 2, 3, and 4, a total of 4 dimensions. Following the aforementioned range constraints, initial particles are generated cyclically. Each particle generates integer data within the phase constraint range. If the range is not met or the calculation period does not meet the range, the particles are regenerated. For example, if the generated particles are [22, 18, 25, 46], the signal period is 111, which meets the requirements. A total of 50 initial particles are generated, with the number set according to actual needs, denoted as x. t ,t∈{1,2,3,...,50}

[0084] Step 2: Phase data conversion to steering data

[0085] Based on the actual phase shift, the phase data is converted to channel data, and the green ratio data is calculated.

[0086]

[0087] Where gr 转向 The green light ratio for turning, Cycle is the signal cycle at the intersection, t 转向 The steering timing is shown in the table below, and the conversion relationship between steering and phase timing is as follows:

[0088]

[0089] This converts the phase timing data into green ratio data for each turn at the intersection, where the green ratio is denoted as gr, and there are 8 turns corresponding to the data in the table.

[0090] Step 3: Calculate the delay of the turn signal control system

[0091] Calculating signal control delay is also an objective that the model needs to optimize. The ultimate goal is to minimize signal control delay by optimizing phase timing.

[0092] Calculation of signal saturation:

[0093]

[0094] Where sat is the turn signal saturation, flow is the turn signal flow calculated in S08, and s and gr are the saturation flow and green signal ratio of the corresponding turn, respectively; the turn signal saturation flow can be calculated according to the design value of traffic standard or according to actual needs. Generally, the saturation flow for left turn is taken as 1550, and the saturation flow for straight is taken as 1600.

[0095] Calculation of signal control delays:

[0096]

[0097]

[0098] obj=dt1+dt2

[0099] Where obj represents the unpenalized signal control delay, and dt1 and dt2 are intermediate variables.

[0100] Step 4: Add saturation penalty and calculate the target intersection signal delay with penalty.

[0101] Penalty function:

[0102] F = r × (max(sat-st, 0)) 3 st = 0.85, r = 1000000

[0103] Where r is the penalty coefficient, which is set to a relatively large value, with a default value of 1,000,000; st is the saturation threshold, which considers a saturation of less than 0.85 to be a reasonable range. When the saturation is greater than 0.85, the greater the saturation, the greater the penalty to the optimization objective.

[0104] The overall optimization objective is designed considering two aspects: first, balancing the delays in all directions, and second, controlling the level of saturation.

[0105] Therefore, the basic function form of the overall optimization objective for signal control delay is as follows:

[0106]

[0107] Where OBJ represents the intersection signaling delay with saturation penalty, which is the objective function to be optimized, and i represents each turn at the intersection. Generally, for a four-way intersection, only left turns and straight-ahead turns are considered, totaling 8 turns, so n = 8. Step 5: Particle swarm optimization algorithm updates phase timing.

[0108] The initial OBJ of the first 50 particles is obtained from the previous calculations. The particle corresponding to the minimum value among these 50 particles is selected and compared with the results of the previous iterations to calculate the local optimum and the global optimum.

[0109] (1) Calculate the local optimal solution x under this iteration. t,pbest :

[0110]

[0111] Where t is the current iteration number;

[0112] (2) Recalculate the global optimal solution x from all previous iterations. t,gbest

[0113]

[0114] in, The target value of the local optimum in the current iteration. The objective value of the two is the target value of the global optimal solution in all previous iterations. The minimum of the two values ​​is the global optimal solution of all previous iterations in the current iteration. Both optimal solutions are 4-dimensional vectors.

[0115] (3) Based on the particle swarm optimization algorithm, update the velocity value v using the global optimal solution and the local optimal solution. i :

[0116] c t =rand(0,1)

[0117]

[0118] Where ω is the inertia factor of the particle swarm optimization algorithm, set to 1; c t The random number t between 0 and 1 represents the update rate in this iteration;

[0119] (4) Discretize the values ​​that need to be updated into integers. Since the signal timing is an integer, use the formula to change the amplitude of each update to ±1 or 0:

[0120]

[0121]

[0122] Update the timing of each particle

[0123] x t =x t-1 +r t

[0124] Note: where x t r t u t v t All are 4-dimensional vectors, corresponding to the timing of 4 phases, and are updated through the above calculations.

[0125] Step 6: Repeat steps 2-5 to obtain the optimal solution.

[0126] Advantages of this invention:

[0127] (1) This method is applicable to timing optimization in different time periods (S08). It is suitable for signal timing tasks in different scenarios and at different times, such as weekdays, non-working days, morning and evening peak hours, off-peak hours, and nighttime. This model method can be used for timing optimization in different scenarios.

[0128] (2) The data aggregation method uses the moving average method (S05). The moving window method is based on the simple summation method. It calculates the moving sum by sequentially adding and subtracting new and old data in each period to eliminate random fluctuations, identify trends, and make predictions accordingly. To ensure the accuracy and reliability of hourly flow statistics, the moving average method is more reliable. (The number of signal cycles in each granularity of the flow data varies, causing fluctuations in the statistical flow data.)

[0129] (3) Optimize the delay calculation method in the timing model (S09-Step3). The sliding window method is based on the simple summation method. It calculates the moving sum by sequentially adding and subtracting new and old data in each period to eliminate random fluctuations, find the trend of development, and make predictions accordingly. To ensure the data fluctuation of the smallest granularity of traffic flow data (the different number of signal cycles contained in each granular time period causes fluctuations in statistical traffic flow data), the moving average method can better ensure the authenticity and reliability of hourly traffic flow statistics.

[0130] (4) Use of improved signal control flow calculation based on queue length (S07). Traditional signal timing models for intersection delay or intersection saturation optimization do not take into account that the actual traffic demand at the intersection is greater than the actual traffic flow detected by the equipment. This model uses the queue length detected by the detector to convert it into traffic flow demand with missing traffic flow according to different saturation states of the intersection, and adds it to the detected traffic flow to obtain the signal control flow demand. Using this data to calculate the intersection signal control delay can more accurately complete the intersection timing task and meet the signal control needs of real road travel.

[0131] (5) Discretization method used in particle swarm timing update (S09-Step5). Traditional intersection delay or intersection saturation optimization signal timing models do not take into account that the actual traffic demand at the intersection is greater than the actual traffic flow detected by the equipment. This model uses the queue length detected by the detector to convert it into traffic flow demand with missing traffic flow according to different saturation states of the intersection, and adds it to the detected traffic flow to obtain the signal control traffic demand. Using this data as the calculation of intersection delay can more accurately complete the intersection timing task and meet the signal control needs of real road travel. Attached Figure Description

[0132] Figure 1 This is a flowchart illustrating the specific implementation steps of the present invention.

[0133] Figure 2 This is a flowchart of the particle swarm optimization algorithm for solving the problem in this invention. Detailed Implementation

[0134] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0135] The invention will be further described below with reference to the accompanying drawings, such as... Figure 1 As shown, an intersection signal timing optimization method based on queue length and delay is implemented in three parts:

[0136] (1) Data collection by roadside equipment and calculation of traffic situation data

[0137] S01: Roadside equipment for lane-level data collection at intersections.

[0138] At signalized intersections, appropriate traffic flow detection devices (such as electronic police, traffic flow detectors, radar detectors, etc.) are installed to detect the traffic flow data of each lane on each branch of the intersection in real time and upload it to the corresponding data integration and management platform.

[0139] S02: Historical accumulation and storage of vehicle passage data

[0140] Real-time data received from roadside equipment is processed at a minimum granularity of 5 minutes to calculate traffic flow data for each lane at each intersection branch, and stored in a database for historical data storage. The storage structure is as follows:

[0141]

[0142] S03: Historical Accumulation and Storage of Traffic Situation Data

[0143] (1) Calculation and storage of queue length in 5 minutes:

[0144] The system detects the actual lane queue length data through detectors and stores it in the system in the same format as the traffic flow data in S02. If the queue length sent by the detector is less than 5 minutes (e.g., 1 minute), multiple data points received within 5 minutes are taken, the maximum value is calculated, and stored as the queue length within the 5-minute granularity.

[0145] (2) Calculation and storage of saturation over 5 minutes:

[0146] The saturation is calculated using traffic data obtained from S02 statistics, combined with the signal control scheme currently in operation at the intersection. The saturation of all lanes and the intersection as a whole is calculated and stored over a 5-minute period.

[0147] Lane green light ratio = Lane green light passage time / Signal cycle

[0148] Lane saturation = 5-minute flow rate converted to hourly flow rate / (lane saturation flow rate * green ratio)

[0149] Intersection saturation = maximum saturation of all lanes

[0150] The lane saturation flow rate has a traffic design standard value depending on the direction of traffic, which is generally around 1600. It can also be adjusted according to the actual situation, but this is not the focus of this patent.

[0151] (2) Historical data aggregation and algorithm data preparation

[0152] S04: Historical Situation Data Aggregation

[0153] Optimize the historical situation data used by performing aggregation operations according to date and time period.

[0154] By default, the signal control scheme used at a typical intersection is a one-day scheme covering the entire day. However, depending on actual needs, the day can be divided into multiple time periods, such as morning rush hour, evening rush hour, off-peak, and nighttime. For each time period, historical data can be aggregated and the timing optimized to generate a complete scheme.

[0155] Historical data aggregation scheme for a single intersection within the same time period (e.g., the entire day from 0:00 to 24:00):

[0156] For each lane at the intersection, historical data for 7 days is selected. The aggregation method is to take the average of the 7 days within the same 5-minute period to obtain the averaged aggregated data of 288 time points for each lane, including lane traffic, lane queue length data, lane saturation, and intersection saturation.

[0157] S05: Sliding window aggregates lane hourly traffic data

[0158] All 5-minute time-point traffic data obtained from S04 are converted into hourly traffic data; queue length data is not processed. The conversion method uses a sliding window summation method to calculate and convert the data for each time point into hourly traffic.

[0159] Sliding window summation method:

[0160] (1) For 288 points of 24-hour data, first, the data of the first half hour (1-6) and the last half hour (283-288) are added to the end and the beginning of the data column respectively, expanding the 24 hours into 25 hours and 300 data points for calculation.

[0161] Then, the sliding window duration is set to 1 hour (12 time points) for sliding calculation and summation. For example, the first point is the summation of time points (1 to 12), the second point is the summation of time points (2-13), and so on, the i-th time point is the summation of time points (i to i+11).

[0162] (2) For each time point, the sum of the 12 time points before and after it is used, which is converted into hourly traffic data, resulting in a total of 288 time points for each lane. This achieves the effect of smoothing the data and converts it into the hourly traffic volume that needs to be used, which serves as the data foundation for subsequent timing optimization algorithms.

[0163] (3) Timing optimization model based on queue length and signal control delay

[0164] This model converts lane queue length data into incomplete lane traffic demand, calculates traffic delay based on traffic demand, analyzes signal delay under different signal control timings, and uses particle swarm optimization to optimize the average signal delay at the target intersection, minimizing the overall intersection delay and obtaining a good timing optimization solution.

[0165] The optimization algorithm steps are as follows:

[0166] S07: Calculate the signal control flow based on queue length data transformation

[0167] During the optimization process, queue length causes the actual traffic flow demand to exceed the actual traffic volume. This needs to be factored into the actual traffic flow during optimization. A minimum delay optimization model based on queue length improvement is proposed.

[0168] The conversion is performed using the following steps;

[0169] (1) Calculate the average number of vehicles in queue.

[0170] Turning is generally divided into left turn, straight turn, and right turn, and may involve multiple lanes.

[0171] Average vehicle length (including distance between vehicles) = 3m

[0172] Turning queue length = average queue length of multiple turning lanes

[0173] Number of vehicles in queue = Queue length / Average vehicle length

[0174] (2) Queuing traffic flow conversion

[0175] Based on the intersection saturation data calculated by S03, the intersection status is divided into two states:

[0176] A saturation level ≥ 0.7 indicates a saturated state, while a saturation level < 0.7 indicates an unsaturated state.

[0177] Calculate the queuing flow transition for each of the two states, and add the signal control flow:

[0178] Basic turning flow = Sum of flow rates in corresponding lanes

[0179] a. Unsaturated state:

[0180] Signal-controlled turning flow = Number of vehicles in queue * Number of lanes + Basic turning flow

[0181] b. Saturation state

[0182] Signal-controlled turning flow = Number of vehicles in queue * Number of lanes * 1.2 + Basic turning flow

[0183] The signal-controlled turning flow of the intersection is obtained for each turn within a time period, with 288 granularities throughout the day.

[0184] S08: Optimize timing scenario strategies and parameter settings

[0185] In actual timing optimization, different strategies may be selected based on different scenarios, such as weekday / weekend scenarios and weekday morning / evening peak scenarios. Different strategies should be used to select the traffic for optimization, and it is recommended to segment the scenarios accordingly. The traffic control signals used in S07 are as follows:

[0186] (1) All-day single-solution scenario

[0187] All day: Traffic flow controlled by traffic control = maximum traffic flow during the entire day.

[0188] (2) Peak and off-peak scenes throughout the day

[0189] Morning rush hour: Traffic flow controlled by traffic control = maximum traffic flow during the morning rush hour;

[0190] Evening peak hours: Traffic flow controlled by traffic control = maximum traffic flow during the evening peak hours;

[0191] Midday off-peak period: Traffic flow rate controlled by traffic signals = average traffic flow rate during midday off-peak period;

[0192] Nighttime off-peak hours: Traffic flow controlled by traffic signals = average traffic flow throughout the day;

[0193] The specific time periods corresponding to peak and off-peak periods are set according to the actual situation.

[0194] (3) Weekday / Non-weekday scenarios

[0195] Weekdays: Traffic flow controlled by the signal system = maximum traffic flow during the entire weekday period.

[0196] Non-working days: Traffic flow controlled by the traffic control system = Traffic flow during the entire non-working day * 0.8

[0197] S09: Timing optimization model based on queue length and delay

[0198] This model only performs timing adjustments for pre-designed phases, without performing phase design work. Generally, signal timing optimization work involves optimizing and adjusting the timing based on the original operating phases of the intersection signal controllers. This model provides a solution for refined optimal timing based on data, which can improve the efficiency of manual timing work by signal experts when the data is complete.

[0199] The model optimizes the total timing of each phase of the signal scheme (which may actually be the green light time + yellow light time). The model is applicable to many common phases: such as oncoming straight + oncoming left turn, two straight and one left turn, oncoming straight + oncoming left turn + one phase, etc. It only needs to calculate the turning conversion data according to the actual phase. Here, we take the most common set of phases as an example. The scheme has a total of 4 phases: east-west straight + east-west left turn + north-south straight + north-south left turn.

[0200] Step 1: Phase timing range limitation, particle swarm algorithm initialization

[0201] Based on the set 4 phases, define the minimum and maximum value ranges, and the period range, which can be modified according to actual needs. The default minimum value is 15 seconds, and the default maximum value is 60 seconds.

[0202] Phase 1: East-West straight line, range [15, 60]

[0203] Phase 2: Turn left (east / west), range [15, 60]

[0204] Phase 3: North-South straight, range [15, 60]

[0205] Phase 4: North-South left turn, range [15, 60]

[0206] Signal period: [60, 200]

[0207] Signal period = sum of the timing times of the four phases

[0208] Particle Swarm Optimization Initialization:

[0209] According to the particle swarm optimization algorithm, particles are the variables to be optimized, namely, four-dimensional indicators: phase 1, 2, 3, and 4. Following the aforementioned range constraints, initial particles are generated cyclically. Each particle generates integer data within the phase constraint range. If the range is not met or the calculation period does not meet the range, the particles are regenerated. For example, if the generated particles are...

[0210] The signal period is [22, 18, 25, 46], with a period of 111, which meets the requirements. A total of 50 initial particles are generated, the number of which is set according to actual needs and denoted as x.t ,t∈{1,2,3,...,50}

[0211] Step 2: Phase data conversion to steering data

[0212] Based on the actual phase shift, the phase data is converted to channel data, and the green ratio data is calculated.

[0213]

[0214] Where gr 转向 The green light ratio for turning, Cycle is the signal cycle at the intersection, t 转向 The steering timing is shown in the table below, and the conversion relationship between steering and phase timing is as follows:

[0215]

[0216]

[0217] This converts the phase timing data into green ratio data for each turn at the intersection, where the green ratio is denoted as gr, and there are 8 turns corresponding to the data in the table.

[0218] Step 3: Calculate the delay of the turn signal control system

[0219] Calculating signal control delay is also an objective that the model needs to optimize. The ultimate goal is to minimize signal control delay by optimizing phase timing.

[0220] Calculation of signal saturation:

[0221]

[0222] Where sat is the turn signal saturation, flow is the turn signal flow calculated in S08, and s and gr are the saturation flow and green signal ratio of the corresponding turn, respectively; the turn signal saturation flow can be calculated according to the design value of traffic standard or according to actual needs. Generally, the saturation flow for left turn is taken as 1550, and the saturation flow for straight is taken as 1600.

[0223] Calculation of signal control delays:

[0224]

[0225]

[0226] obj=dt1+dt2

[0227] Where obj represents the unpenalized signal control delay, and dt1 and dt2 are intermediate variables.

[0228] Step 4: Add saturation penalty and calculate the target intersection signal delay with penalty.

[0229] Penalty function:

[0230] F = r × (max(sat-st, 0)) 3 st = 0.85, r = 1000000

[0231] Where r is the penalty coefficient, which is set to a relatively large value, with a default value of 1,000,000; st is the saturation threshold, which considers a saturation of less than 0.85 to be a reasonable range. When the saturation is greater than 0.85, the greater the saturation, the greater the penalty to the optimization objective.

[0232] The overall optimization objective considers two aspects: first, balancing delays in all directions; and second, controlling the saturation level. Therefore, the basic functional form of the overall optimization objective for signal control delays is as follows:

[0233]

[0234] Where OBJ is the intersection signal delay with saturation penalty, which is the objective function to be optimized, and i is each turn at the intersection. Generally, for a four-way intersection, only left turns and straight turns are considered, totaling 8 turns, so n=8.

[0235] Step 5: Particle Swarm Optimization Algorithm Updates Phase Timing

[0236] The initial OBJ of the first 50 particles is obtained from the previous calculations. The particle corresponding to the minimum value among these 50 particles is selected and compared with the results of the previous iterations to calculate the local optimum and the global optimum.

[0237] (1) Calculate the local optimal solution x under this iteration. t,pbest :

[0238]

[0239] Where t is the current iteration number;

[0240] (2) Recalculate the global optimal solution x from all previous iterations. t,gbest

[0241]

[0242] in, The target value of the local optimum in the current iteration. The objective value of the two is the target value of the global optimal solution in all previous iterations. The minimum of the two values ​​is the global optimal solution of all previous iterations in the current iteration. Both optimal solutions are 4-dimensional vectors.

[0243] (3) Based on the particle swarm optimization algorithm, update the velocity value v using the global optimal solution and the local optimal solution. i :

[0244] c t =rand(0,1)

[0245]

[0246] Where ω is the inertia factor of the particle swarm optimization algorithm, set to 1; c t The random number between 0 and 1 represents the update speed in this iteration;

[0247] (4) Discretize the values ​​that need to be updated into integers. Since the signal timing is an integer, use the formula to change the amplitude of each update to ±1 or 0:

[0248]

[0249]

[0250] Update the timing of each particle

[0251] x t =x t-1 +r t

[0252] Where x t r t u t v t All are 4-dimensional vectors, corresponding to the timing of 4 phases, and are updated through the above calculations.

[0253] Step 6: Repeat steps 2-5 to obtain the optimal solution.

[0254] Set a large number of iterations or stop iterating after the intersection delay remains essentially unchanged after a certain number of iterations to obtain the optimal solution. Using the same method for calculating the target delay, calculate the signal control delay for the intersection before optimization and after optimization using the phase timing, and compare the results to improve the performance.

[0255] This invention discloses a method and apparatus for optimizing intersection signal timing based on queue length and delay, aiming to improve the shortcomings of traditional traffic signal control timing optimization processes, such as the subjective nature of manual division, the inability to accurately count intersection traffic flow data, and the poor adaptability of timing results for different intersections. These problems result in the inability of refined timing control of urban intersection signal control optimization schemes to achieve the expected results, the inability to accurately meet actual traffic demand at different times, and the inability to alleviate traffic pressure caused by urban road planning problems through signal control.

[0256] The signal timing optimization method and device based on queue length and delay provided by this invention, relying on traffic flow data and signal phase control data collected by urban traffic flow monitoring equipment, can more quickly and conveniently calculate the refined timing results of urban intersection signal control schemes. It provides a set of reasonable timing optimization models for continuous and real traffic demand at intersections, and performs precise signal control timing optimization on more accurate data in different time periods. It uses the same timing scheme method to make full use of the spatiotemporal data resources of intersection equipment and signals.

Claims

1. A method for optimizing intersection signal timing based on queue length and delay, characterized in that, Includes the following steps: S1: Data collection and calculation of traffic situation data by roadside equipment; S2: Historical data aggregation and algorithm data preparation; S3: Timing optimization model based on queue length and signaling delay; Specifically, S3 includes the following steps: S07: Calculate traffic flow based on queue length data conversion: During the optimization process, the queue length causes the actual traffic flow demand to be greater than the actual traffic flow statistics. It needs to be converted into actual traffic flow during the optimization process and participate in the optimization. The minimum delay optimization model is improved based on queue length. The conversion is performed using the following steps: (1) Calculate the average number of vehicles in the queue Turning is generally divided into left turn, straight, and right turn, and may involve multiple lanes; Average vehicle length = 3m; Turning queue length = Average of the queue lengths across multiple turning lanes; Number of vehicles in queue = Queue length / Average vehicle length; (2) Conversion of queuing traffic flow Based on the calculated intersection saturation data, the intersection status is divided into two states: A saturation level ≥ 0.7 indicates a saturated state, while a saturation level < 0.7 indicates an unsaturated state. Calculate the queuing flow transition for each of the two states, and add the signal control flow: Basic turning flow = Sum of flow rates in corresponding lanes; a. Unsaturated state: Traffic flow control = Number of vehicles in queue Number of lanes + base turning flow; b. Saturation state Traffic flow control = Number of vehicles in queue Number of lanes 1.2 + base steering flow; The signalized turning flow at the intersection is obtained for each turn within a time period, with 288 granularities throughout the day; S08: Optimize timing scenario strategies and parameter settings When optimizing timing, different strategies should be selected based on different scenarios, such as weekdays and non-weekdays, and weekday morning and evening peak hours. Different strategies should be used to select the traffic to be optimized. Recommended scenario segmentation, using the signal control traffic in S07: (1) All-day single-solution scenario All-day period: Traffic flow controlled by the signal system = Maximum traffic flow during the entire day; (2) Scenes during morning and evening rush hours and off-peak hours throughout the day Morning rush hour: Traffic flow controlled by traffic control = Maximum traffic flow during the morning rush hour; Evening Peak Hour: Traffic Control Turnaround Flow = Maximum Traffic Flow During Evening Peak Hour; Midday off-peak period: Traffic flow controlled by traffic signals = Average traffic flow during midday off-peak period; Nighttime off-peak hours: Traffic flow controlled by traffic signals = Average traffic flow throughout the day; The specific time periods corresponding to peak and off-peak periods will be set according to the actual situation. (3) Weekday / Non-weekday scenarios Weekdays: Traffic flow for traffic control = Maximum traffic volume throughout the entire weekday; Non-working days: Traffic flow controlled by the traffic control system = Traffic flow throughout the entire non-working day 0.8; S09: A time-matching optimization method model based on queue length and delay; This model only performs timing for pre-designed phases and does not perform phase design work. Generally, signal timing optimization work is based on the original operating phases of the intersection signal controllers to optimize and adjust the timing. This model provides a solution for refined optimal timing based on data. Under the condition of complete data, it can improve the work efficiency of signal experts in manual timing. The model optimizes the total timing of each phase of the signal scheme. The model is applicable to many common phases: such as opposing straight + opposing left turn, two straight and left turns in one direction, opposing straight + opposing left turn + one-way phase, etc. It only needs to calculate the turning conversion data according to the actual phase. Here, we take the most common set of phases as an example. The scheme has a total of 4 phases: east-west straight + east-west left turn + north-south straight + north-south left turn.

2. The intersection signal timing optimization method based on queue length and delay according to claim 1, characterized in that, Step S1 includes the following steps: S01: Roadside equipment for lane-level data acquisition at intersections: Install corresponding traffic flow detection equipment at signal-controlled intersections to detect the traffic flow data of each lane of each branch of the intersection in real time and upload it to the corresponding data integration and management platform; S02: Historical accumulation and storage of vehicle traffic data: Real-time data received from roadside equipment is calculated at a minimum granularity of 5 minutes to statistically analyze the traffic flow data of each branch and lane at the intersection and store it in the database for historical data storage; S03: Historical accumulation and storage of traffic situation data: (1) Calculation and storage of 5-minute queue length: The queue length data of the real lane is detected by the detector and connected to the system for data storage. The storage format is consistent with the traffic data in S02. If the queue length sent by the detector is less than 5 minutes, multiple data received within 5 minutes are taken and the maximum value is calculated and stored as the queue length within 5 minutes. (2) Calculation and storage of saturation in 5 minutes: The saturation is calculated using the traffic data obtained from S02 statistics, combined with the signal control scheme of the current real-time operation of the intersection, to calculate and store the saturation of all lanes and the intersection in 5 minutes. Lane green light ratio = Lane green light duration / Signal cycle; Lane saturation = 5-minute flow rate converted to hourly flow rate / (lane saturation flow rate) (Green light ratio) Intersection saturation = maximum saturation of all lanes.

3. The intersection signal timing optimization method based on queue length and delay according to claim 1, characterized in that, Step S2 includes the following steps: S04: Historical Situation Data Aggregation Optimize the historical situation data used, and perform aggregation operations according to date and time period; by default, the signal control scheme required for general intersections is a one-day scheme with a time period of the whole day. It can also be divided into multiple time periods according to actual needs, perform historical data aggregation and timing optimization, and generate a complete scheme. For a given intersection within the same time period, the historical data aggregation scheme is as follows: for each lane at the intersection, the historical data is selected for 7 days, and the aggregation method is to take the average of the 7 days within the same 5-minute period. This yields the averaged aggregated data for each lane at 288 time points, including lane traffic flow, lane queue length data, lane saturation, and intersection saturation. S05: Sliding window aggregates lane hourly traffic data All 5-minute time point traffic data obtained in S04 are converted into hourly traffic data, while queue length data is not processed. The conversion method uses a sliding window summation method to calculate the conversion, converting the data of each time point into hourly traffic. Sliding window summation method: (1) For 288 data points in 24 hours, the data of the first half hour and the last half hour are added to the end and the beginning of the data column respectively, expanding the 24 hours into 25 hours and 300 data points for calculation. Then, the sliding window time length is set to 1 hour for sliding calculation and summation. (2) Each time point is summed using the 12 time points before and after it, which is converted into hourly traffic data, resulting in a total of 288 time points for each lane; this achieves the effect of smoothing the data and converts it into the hourly traffic volume that needs to be used, which serves as the data basis for subsequent timing optimization algorithms.

4. The intersection signal timing optimization method based on queue length and delay according to claim 1, characterized in that, Step S09 includes the following steps: Step 1: Phase timing range limitation, particle swarm algorithm initialization; Based on the set 4 phases, set the minimum and maximum value ranges and the period range, and modify them according to actual needs; The default minimum value is 15 seconds, and the maximum value is 60 seconds. Phase 1: East-West straight line, range [15, 60]; Phase 2: Turn left (east / west), range [15, 60]; Phase 3: North-South straight, range [15, 60]; Phase 4: North-South left turn, range [15, 60]; Signal period: [60, 200]; The signal period = the sum of the timing times of the four phases; Particle Swarm Optimization Initialization: According to the particle swarm optimization algorithm, the particles are the variables to be optimized, namely the four-dimensional indicators of phase 1, 2, 3, and 4. The initial particles are generated cyclically in accordance with the above range constraints. Each particle generates integer data within the range of the phase constraints. If the range is not met or the calculation period does not meet the range, the particles are regenerated. Step 2: Phase data conversion to steering data Based on the actual phase direction, the phase data is converted to channel data, and the green ratio data is calculated: in For the green credit ratio of the turn, For the signal cycle at the intersection, For the turning timing, the phase timing data is converted into green ratio data for each turn at the intersection, where the green ratio is denoted as gr, and there are 8 turns corresponding to the data in the table. Step 3: Turning to Signalling Delay Calculation Calculating signal control delay is also an objective that the model needs to optimize. The ultimate goal is to minimize signal control delay by optimizing phase timing. Calculation of signal saturation: in To shift to signal control saturation, S is the signal-controlled turning flow calculated in S08, where s and gr are the saturation flow and green ratio of the corresponding turn, respectively. The saturation flow of the turn can be calculated according to the design value of the traffic standard or according to actual needs. Generally, the saturation flow of left turn is taken as 1550 and the saturation flow of straight turn is taken as 1600. Calculation of signal control delays: in For unpunished delays in credit control, and As an intermediate variable; Step 4: Add saturation penalty and calculate the target intersection signal delay with penalty; Penalty function: Where r is the penalty coefficient, which can be set to a relatively large value; the default is 1,000,000. The saturation threshold is defined as a value less than 0.85, which is considered a reasonable range. When the saturation value is greater than 0.85, the greater the saturation value, the greater the penalty to the optimization objective. The overall optimization objective considers two aspects: first, balancing delays in all directions; and second, controlling the saturation level. Therefore, the basic functional form of the overall optimization objective for signal control delays is as follows: in The intersection signal delay with saturation penalty is the objective function to be optimized. i represents each turn at the intersection. For a four-way intersection, only left turns and straight turns are considered, totaling 8 turns, so n=8. Step 5: Particle swarm optimization algorithm updates phase timing The initial 50 particles were determined based on the previous calculations for this iteration. Take the particle corresponding to the minimum value among the 50 particles, compare it with the previous iteration results, and calculate the local optimum and the global optimum: (1) Calculate the local optimal solution under this iteration. : Where t is the current iteration number; (2) Recalculate the global optimal solution from all previous iterations. in, The target value of the local optimum in the current iteration. The objective value of the two is the target value of the global optimal solution in all previous iterations. The minimum of the two values ​​is the global optimal solution of all previous iterations in the current iteration. Both optimal solutions are 4-dimensional vectors. (3) Based on the particle swarm optimization algorithm, update the velocity value using the global optimum and local optima. : in, The inertia factor for the particle swarm optimization algorithm is set to 1. The random number t between 0 and 1 represents the update rate in this iteration; (4) Discretize the values ​​that need to be updated into integers. Since the signal timing is an integer, use the formula to change the amplitude of each update to ±1 or 0: Update the timing for each particle: in , , , All are 4-dimensional vectors, corresponding to the timing of 4 phases, and are updated through the above calculations; Step 6: Repeat steps 2-5 to obtain the optimal solution.

Images