WO2021073526A1

WO2021073526A1 - Trajectory data-based signal control period division method

Info

Publication number: WO2021073526A1
Application number: PCT/CN2020/120838
Authority: WO
Inventors: 马万经; 万丽娟; 俞春辉; 王玲
Original assignee: 同济大学
Priority date: 2019-10-15
Filing date: 2020-10-14
Publication date: 2021-04-22
Also published as: CN110853380A; CN110853380B

Abstract

A trajectory data-based signal control period division method, comprising: obtaining, on the basis of a traffic wave theory, a Greenshields linear model and a fundamental relationship of three parameters, i.e. flow, density and speed, a relationship between a queuing wave speed and the flow (S1); overlapping trajectory data of the same intersection during the same periods in several days, so as to obtain input data (S2); obtaining, on the basis of the input data, a speed threshold value division method and a kinetic equation, the queuing wave speed (S3); and clustering the queuing wave speed, and performing signal control period division on the basis of the relationship between the queuing wave speed and the flow (S4). The signal control period division method not only improves the operating efficiency and safety level of a signal control intersection, but also saves the installation and maintenance cost of a fixed detector.

Description

Method for dividing signal control period based on trajectory data

Technical field

The present invention relates to the research field of signal control at intersections, in particular to a signal control period division method based on trajectory data.

Background technique

Traffic signal control is one of the most effective means to improve traffic efficiency and safety at intersections. Therefore, many scholars optimize traffic control parameters, such as cycle, green letter ratio, phase sequence, etc., as well as traffic software such as TRANSYT, Synchro, PASSER, etc. The signal timing scheme is optimized. Due to the volatility and periodicity of traffic demand, a single timing scheme cannot meet the needs of urban traffic control. Therefore, it is necessary to divide a day into several control periods, namely TOD, which is a commonly used traffic control scheme selection method. TOD mode can significantly improve the efficiency of traffic control, and has low implementation cost and high reliability.

In TOD mode, the most critical part is to determine the time division point, which is mainly determined according to the characteristics of traffic demand, but fixed detectors such as coils, geomagnetism or video cannot be accurate due to easy damage and low coverage. And complete flow data, and the installation and maintenance costs of traditional fixed detector equipment are high.

With the development of science and technology, navigation software such as AutoNavi Maps, Baidu Maps, and taxi-hailing software such as DiDi have derived high-precision vehicle trajectory data, which has low acquisition cost and wide coverage, and can be continuously obtained in time. The data resource contains a wealth of traffic information, such as arrival distribution, queuing, etc. If these data can be used for time division, great benefits will be obtained at a very low cost.

At present, there is a lack of a method for dividing the signal control period by using vehicle trajectory data.

Summary of the invention

The purpose of the present invention is to provide a signal control period division method based on trajectory data in order to overcome the above-mentioned defects in the prior art.

The purpose of the present invention can be achieved through the following technical solutions:

A signal control period division method based on trajectory data, the method includes the following steps:

Step S1: Based on the basic relationship of the three parameters of the traffic wave theory, the Greenhill linear model and the flow density, the relationship between the aggregate wave speed and the flow rate is obtained;

Step S2: superimpose the trajectory data of the same intersection for multiple days and the same time period to obtain input data;

Step S3: Based on the input data, the speed threshold dividing method and the dynamic equation, the wave speed of the build-up wave is obtained;

Step S4: cluster the wave speed of the build-up wave, and divide the signal control period based on the relationship between the wave speed of the build-up wave and the flow rate.

The traffic wave theory is expressed as:

Among them, A and B represent two traffic flows, ω _AB is the traffic wave speed, q _A is the flow of the A traffic flow, q _B is the flow of the B traffic flow, k _A is the density of the A traffic flow, and k _B Is the density of B traffic flow.

The Greenhill linear model is:

Wherein, v _i is the velocity, k _i is the density, v _f is the free stream velocity, k _j to block density, i take A or B.

The basic relationship among the three parameters of flow density is:

q _i ＝k _i v _i

Among them, q _i is the flow rate.

_{The relationship between the mass} wave velocity ω ABstop and the flow rate is:

The step S3 includes:

Step S31: Identify the state of the track point based on the speed threshold division method;

Step S32: Based on the state of the track point and the kinematic equation, identify the track point added to the queue;

Step S33: Fit the trajectory points added to the queue to obtain the wave speed of the build-up wave.

The clustering algorithm is a dichotomous K-means clustering algorithm.

The clustering effect evaluation method of the dichotomous K-means clustering algorithm is:

Among them, SSE is the error sum of squares, K is the preset number of clusters, C _i is the i-th class, c _i is the centroid of the i-th class obtained according to the clustering algorithm, and x is the element of the i-th class.

Compared with the prior art, the present invention has the following advantages:

(1) Using vehicle trajectory data instead of traffic data obtained by fixed detectors to divide the signal control time period at intersections not only improves the operating efficiency and safety level of signal controlled intersections, but also saves the installation and maintenance costs of fixed detectors.

(2) Based on the theory of traffic waves, we deduced the relationship between mass wave speed and flow, and found that mass wave speed can reflect changes in traffic demand; flow data is usually obtained through coils, and coil coverage is low, not easy to obtain, and installation and maintenance costs are high. The speed of the build-up wave is obtained through trajectory data, and the trajectory data is easy to obtain, which has the advantages of wide coverage and low cost.

(3) The problem of low data penetration and low sampling frequency is overcome by superimposing the trajectory data of the same intersection for multiple days and the same time period.

Description of the drawings

Figure 1 is a flow chart of the present invention;

Figure 2 is a schematic diagram of the formation of traffic waves according to the present invention;

Figure 3 is a schematic diagram of trajectory data superimposition of the present invention;

Figure 4 is a schematic diagram of the identification of track points added to the queue according to the present invention;

5 is a schematic diagram of a VISSIM simulated signalized intersection according to an embodiment of the present invention;

Fig. 6 is a schematic diagram of traffic flow changes at each entrance of the embodiment of the present invention;

FIG. 7 is a schematic diagram of the wave speed change of the assembled wave of different phases according to the embodiment of the present invention;

FIG. 8 is a schematic diagram of the comparison between the total flow and the average wave speed of the build-up wave at the intersection according to the embodiment of the present invention;

FIG. 9 is a diagram of the division result of signal control period at an intersection according to an embodiment of the present invention;

FIG. 10(a) is the result of dividing the signal control period with a sampling interval of 5s and a low penetration rate according to an embodiment of the present invention;

FIG. 10(b) is the result of dividing the signal control period with a sampling interval of 10s and a low penetration rate according to an embodiment of the present invention;

FIG. 10(c) is the result of dividing the signal control period with a sampling interval of 15s and a low penetration rate according to an embodiment of the present invention;

Fig. 10(d) is the result of dividing the signal control period with a sampling interval of 20s and a low permeability according to an embodiment of the present invention;

FIG. 10(e) is the result of dividing the signal control period with a sampling interval of 30s and a low penetration rate according to an embodiment of the present invention;

Fig. 10(f) is the result of dividing the signal control period with a sampling interval of 60s and a low permeability according to an embodiment of the present invention.

Detailed ways

The present invention will be described in detail below with reference to the drawings and specific embodiments. This embodiment is implemented on the premise of the technical solution of the present invention, and provides detailed implementation and specific operation procedures, but the protection scope of the present invention is not limited to the following embodiments.

Example

This embodiment provides a signal control period division method based on trajectory data, including the following steps:

further,

In step S1, the process of solving the relationship between mass wave velocity and flow rate is:

Suppose there are two traffic flows A and B. The flow, density and speed of A traffic flow are q _A , k _A and v _A , and the flow, density and speed of B traffic flow are q _B , k _B and v respectively. _A. When the traffic flow state changes, a traffic wave S will be generated.

According to the traffic wave theory, the formula of traffic wave velocity can be obtained:

In order to further simplify the formula of traffic wave velocity, this paper introduces the Greenhill linear model, which describes the relationship between flow and density, and the expression is as follows:

In order to facilitate the derivation, let:

Combine the basic relationship of flow density:

q _i ＝k _i v _i

Among them, q _i is the flow rate.

The formula for traffic wave velocity can be derived as follows:

ω _AB ＝v _f [1-(η _A +η _B )]

As shown in Figure 2, a build-up wave is _{a traffic wave formed by a vehicle fleet running at the average speed v A of the} interval and stopping at a red light at an intersection. ω _AB represents the wave speed of the build-up wave. When stopping, k _B is equal to k _j , That is, η _B is 1, so the condensing wave velocity can be calculated by the following formula:

Combining the Greenhill linear model and the basic relationship of flow density, the relationship between v _A and q _A can be obtained:

That is, under unsaturated conditions (unsaturated state means that the saturation of each phase of the intersection is less than 1, and phase saturation refers to the ratio of the flow and capacity of a key lane controlled by a phase), v _A is a function of q _A , That is, v _A =f(q _A ), and the wave speed of the build-up wave can be further obtained:

Since v _f and k _j are known constants, and v _A is a function of q _A , there is a one-to-one correspondence between the wave speed of the build-up wave and the flow rate, and the wave speed of the build-up wave can reflect the traffic demand.

In step S2, the biggest problem in the application of vehicle trajectory data is the low permeability of floating vehicles and the long sampling interval. However, this problem is overcome in calculating the build-up wave velocity by superimposing the trajectory data of the same intersection for multiple days and the same time period. Assumptions: 1) Applicable to timing signal control (timing signal control refers to the control of a fixed signal timing plan, that is, the intersection signal control machine operates according to the preset timing plan); 2) The arrival of vehicles in different cycles The model is the same; 3) does not consider the instability of traffic flow; 4) applies to the state of unsaturated traffic flow.

Due to the low penetration rate of floating vehicles (for example: 2%) and the long sampling interval (for example: 20s), the number of vehicles detected in each cycle is very small, so there are very few trajectory points detected to join the queue. However, due to the above assumptions, the measure of superimposing the trajectory data of multiple days at the same time can be adopted to obtain a large number of trajectory points added to the queue during the period, so as to fully fit the wave speed of the build-up wave, as shown in Figure 3.

In step S3, due to the uncertainty of the traffic flow and the long sampling interval of the trajectory data, it is difficult to obtain a build-up wave in practice. The difficulty lies in extracting the trajectory points that are added to the queue from the trajectory data. In order to estimate the speed of the build-up wave, the following three main steps will be carried out:

1. Identify the status of the track point (moving or stopping)

According to a simple speed threshold division method, all trajectory points are divided into two categories: stop and motion:

among them:

It represents the speed of vehicle n at time m, and v _th is a preset speed threshold, which is close to 0. If the speed of a track point is less than this threshold, it means that the point is in a parking state.

2. Identify the track points added to the queue

Vehicle trajectory data is continuous in both space and time dimensions, based on one-dimensional kinematics equations and two consecutive different state trajectory points of the same vehicle trajectory (the first point is movement, the second point is stop), Can calculate the time when the vehicle joins the queue

among them:

with

Respectively represent the time and position of vehicle n in step m;

Represents the position of vehicle n at step m+1; a _d is vehicle deceleration; γ∈(0,1) is used to indicate whether the vehicle is at a free flow speed or is accelerating or decelerating. The first formula shows that the vehicle n is at a speed

After driving for a period of time, it decelerates to a stop; the second formula shows that the vehicle n starts from the speed

Decelerate until it stops. Figure 4 is a schematic diagram of the identification of track points added to the queue.

3. Estimate the speed of the build-up wave

Linear fitting is performed on the extracted trajectory points added to the queue, and the build-up wave velocity can be obtained.

In step S4, K-means clustering is one of the most common clustering methods in data mining and unsupervised learning. Traditionally, the K-means clustering method is used to cluster traffic data to obtain the time period division points for signal control. Similarly, the aggregation wave speed can also be clustered. The principle is that the time periods with the same traffic flow pattern should use the same signal configuration. Time plan (signal timing plan is to arrange several control states in a signal cycle, each control state assigns right of passage to vehicles or pedestrians in certain directions, and reasonably arranges the display order of these control states) .

Although the K-means algorithm is easy to implement, it may converge to a local minimum, and the clustering result largely depends on the initial partition and centroid. Therefore, the extended dichotomous K-means clustering algorithm is used, and the sum of squares of errors (SSE) is used to express the effect of clustering:

Among them: K is the preset number of clusters; C _i is the i-th class; c _i is the centroid of the i-th class obtained according to the clustering algorithm; x is the element of the i-th class.

In this embodiment, a certain period of time (for example: 8 hours) is divided into smaller periods (for example: 15 minutes), and multiple days of simultaneous trajectory data are superimposed, and the above-mentioned method is used to obtain the build-up wave in each period. Wave speed, and then divide the signal control period by the binary K-means clustering algorithm. The basic process of the binary K-means clustering algorithm is as follows:

Step 1: Treat all time periods as one cluster.

Step 2: Perform 2-mean clustering on the cluster.

Step 3: If the number of clusters is less than the preset K value, go to the next step, otherwise, end the algorithm.

Step 4: Calculate the SSE value of the newly generated cluster, and go to step 2.

The following is a specific example to illustrate:

The specific example uses simulation to obtain trajectory data for method verification. VISSIM is used to establish signal-controlled intersections. As shown in Figure 5, the main road is east-west and north-south is the secondary road; the speed limit is 50km/h; the fixed six is adopted. The phase signal timing scheme, the signal period is 104s. 20 different random seeds are used in the simulation software. Each simulation time is 30600s, including the warm-up time of 1800s, then the analysis time period is 1800s to 30600s (ie 8 hours), and then the analysis time period is 15 minutes apart Divided into several periods. The change in traffic flow entered every 15 minutes is shown in Figure 6.

In the above method, the free-flow vehicle speed v _{f is} set to 50 km/h, the deceleration a _{d is} set to -3.5 m/s ² , and the speed threshold value is v _th =1 m/s. The binary K-means clustering algorithm presets the number of clusters to be 4, which usually includes peak traffic in the morning and afternoon.

The basic idea is that the aggregate wave speed can reflect changes in traffic demand, so it can be used to identify the points of signal control time period. For this reason, the embodiment adopts 100% high permeability and short sampling interval of 1s trajectory data, the number of superimposed days is 3 days, according to the above method, the gathering wave velocity of each phase every 15 minutes is obtained, as shown in FIG. 7.

Comparing Figure 7 and Figure 6, it can be seen that the trend of mass wave velocity is very close to the trend of traffic flow. In order to better illustrate the relationship between the two, Figure 8 plots the sum of traffic at intersections and intersections in every 15 minutes. The comparison chart of the average build-up wave speed change shows that the average build-up wave speed and the total traffic flow have almost the same changing trend, so it can be explained that the build-up wave speed can effectively reflect the traffic demand.

Then, the bipartite K-means clustering algorithm is used to cluster the total flow and average build-up wave velocity respectively, and it is found that the time division points are the same, as shown in Figure 9, the time division results are 1h, 4h, and 6.5h.

Since the actual trajectory data is low sampling rate and low precision, in order to verify the performance of the proposed method under different sampling rates and sampling intervals, this example uses low precision (5s, 10s, 15s, 20s, 30s) , 60s) Low sampling rate (2%, 5%, 10%, 15%, 20%) trajectory data for sensitivity analysis. The number of days for trajectory data superimposition ranges from 1 day to 17 days. At the same time, it is based on the time period identified by the trajectory data of 100% permeability and 1s sampling interval. If they are the same, the result is marked as 1, which is the black grid in Figure 10; otherwise, it is marked as 0.

When the trajectory data penetration rate is low and the sampling interval is long, it usually takes more days to superimpose the trajectory data to get the same result as the dividing point of the reference period. For example, when the sampling interval is 5s, the penetration rate is 10%, 15% and respectively. In 20% of cases, it takes 8 days of trajectory data to be superimposed. In contrast, under the same sampling interval, a low permeability of 2% requires 13 days of trajectory data overlay.

Claims

A signal control period division method based on trajectory data is characterized in that the method includes the following steps:

Step S1: Based on the basic relationship of the three parameters of the traffic wave theory, the Greenhill linear model and the flow density, the relationship between the aggregate wave speed and the flow rate is obtained;

Step S2: superimpose the trajectory data of the same intersection for multiple days and the same time period to obtain input data;

Step S3: Based on the input data, the speed threshold dividing method and the dynamic equation, the wave speed of the build-up wave is obtained;

Step S4: cluster the wave speed of the build-up wave, and divide the signal control period based on the relationship between the wave speed of the build-up wave and the flow rate.
The method for dividing signal control time periods based on trajectory data according to claim 1, wherein the traffic wave theory is expressed as:

Among them, A and B represent two traffic flows, ω AB is the traffic wave speed, q A is the flow of the A traffic flow, q B is the flow of the B traffic flow, k A is the density of the A traffic flow, and k B Is the density of B traffic flow.
The method for dividing a signal control period based on trajectory data according to claim 2, wherein the Greenhill linear model is:

Wherein, v i is the velocity, k i is the density, v f is the free stream velocity, k j to block density, i take A or B.
The method for dividing the signal control period based on trajectory data according to claim 3, wherein the basic relationship between the three parameters of the flow density is:

q i ＝k i v i

Among them, q i is the flow rate.
A signal control period division method based on trajectory data according to claim 4, wherein the relationship between the build-up wave velocity ω ABstop and the flow rate is:
A signal control period division method based on trajectory data according to claim 1, wherein the step S3 comprises:

Step S31: Identify the state of the track point based on the speed threshold division method;

Step S32: Based on the state of the track point and the kinematic equation, identify the track point added to the queue;

Step S33: Fit the trajectory points added to the queue to obtain the wave speed of the build-up wave.
The method for dividing a signal control period based on trajectory data according to claim 1, wherein the clustering algorithm is a dichotomous K-means clustering algorithm.
A signal control period division method based on trajectory data according to claim 7, wherein the evaluation method of the clustering effect of the dichotomous K-means clustering algorithm is:

Among them, SSE is the error sum of squares, K is the preset number of clusters, C i is the i-th class, c i is the centroid of the i-th class obtained according to the clustering algorithm, and x is the element of the i-th class.