JP3152858B2 - Traffic flow prediction device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a traffic flow prediction device used in, for example, a traffic control system for a highway, and more particularly to a traffic flow prediction device for improving the prediction accuracy of a traffic flow.

[0002]

2. Description of the Related Art In general, in a traffic control system for a highway, a driver of a passing vehicle is provided with a. Area where traffic jam is expected, b. Expected traffic jam length,
C. Whether traffic congestion is likely to spread in the future, or is it going to be resolved? The time it takes to clear the traffic,
E. How fast can you pass?
It provides detailed forecast information such as the time required to travel from one point to another, etc., and based on the predicted traffic volume, such as closing the vehicle just before entering the main highway during a rush hour. It is required to control traffic appropriately.

Here, an outline of a conventional traffic flow prediction model will be described. (1) The traffic density in each section is calculated by dividing the expressway into several sections and dividing the number of vehicles present in each section by each section length. (2) Predict the vehicle speed in each section. (3) The traffic volume flowing into each section and the traffic volume flowing out of each section are obtained. (4) The number of vehicles in each section is updated from the inflow / outflow traffic volume obtained in (3). (5) Repeating the above processes (1) to (4),
Find the predicted value of traffic flow.

Therefore, if a series of processes are performed in accordance with such an algorithm, it is possible to predict a traffic flow, and it is possible to perform a calculation which serves as a base for providing various types of information A to A.

[0005]

However, in the conventional traffic flow prediction model, the means for predicting the vehicle speed in the above-mentioned series of processes (2) has not yet been established, and the prediction error is large. It is a factor. For example, there is a model that empirically calculates a vehicle speed from statistical data, but a learning method of the model has not been established, and sufficient prediction accuracy has not been obtained.

[0006] The present invention has been made in view of the above circumstances, and has as its object to provide a traffic flow prediction device that improves the prediction accuracy of vehicle speed. Another object of the present invention is to provide a traffic flow prediction device that accurately predicts the required traveling time at a plurality of points with the improvement of the vehicle speed prediction accuracy.

Another object of the present invention is to provide a traffic flow predicting device that appropriately predicts the length of congestion with the improvement of the vehicle speed prediction accuracy. Still another object of the present invention is to provide a traffic flow prediction device that accurately grasps a traffic capacity on a road with an improvement in vehicle speed prediction accuracy.

[0008]

In order to solve the above-mentioned problems, the invention corresponding to claim 1 is based on a traffic density actual value of an upstream section obtained from a vehicle detector provided in each section on a road , A neural network learning means for learning a weighting coefficient using a traffic density actual value of a plurality of downstream sections adjacent to the upstream section and a spatial average speed actual value of the upstream section, and learning by the neural network learning means. Traffic density in the upstream section based on
And traffic density of multiple downstream sections adjacent to the upstream section
And a neural network for calculating a spatial average speed predicted value of the upstream section, and using the spatial average speed predicted value calculated by the neural network and the traffic density, a temporal transition of the traffic density and the traffic density. A traffic flow prediction device provided with traffic flow prediction means for predicting and calculating a temporal transition of a spatial average speed.

According to a second aspect of the present invention, in addition to the constituent features described in the first aspect, means for calculating a required travel time on a route from a temporal transition of a spatial average speed predicted by the traffic flow predicting means. Is a traffic flow prediction device to which is added.

According to a third aspect of the present invention, in addition to the constitutional elements described in the first aspect, there is newly provided means for obtaining a congested section on a route from a temporal transition of a spatial average speed predicted by the traffic flow predicting means. This is an added traffic flow prediction device.

[0011] Next, the invention includes a traffic density setting means for setting a traffic density of the downstream multiple sections adjacent to the traffic density of the upstream-side section to the upstream section, the traffic corresponding to claim 4 spatial average velocity of the upstream section based on <br/> a traffic density of the set at a density setting means upstream section <br/> a traffic density of the downstream multiple sections adjacent to the upstream section a neural network for obtaining the prediction value, and multiplying means for obtaining a traffic volume of the upstream section with a spatial average velocity prediction value of the upstream section obtained and the traffic density by the neural network, resulting in the multiplication means It was traffic and traffic density and the traffic volume of the upstream section - traffic creates a characteristic curve of the traffic density - a traffic flow prediction apparatus provided with the traffic density characteristic curve creating means.

[0012]

Accordingly, the invention corresponding to claim 1 is characterized in that the above means are taken, and the neural network learning means uses the actual traffic density value of the upstream section, the downstream plurality of sections adjacent to the upstream section. The traffic density actual value is input to the learning neural network for learning and predicting the spatial average speed of the upstream section, and the predicted spatial average speed is compared with the spatial average speed actual value of the upstream section, The weight coefficient is learned so that the error between them is minimized. Thereafter, the traffic density of the upstream section and the traffic densities of a plurality of downstream sections adjacent to the upstream section are input to the neural network using the weighting coefficients after learning, and the spatial average speed prediction value of the upstream section is input. Ask for. Further, the traffic flow prediction means predicts and calculates the temporal transition of the spatial average speed (vehicle speed) using the predicted value of the spatial average speed obtained by the neural network and the traffic density of the upstream section.
A vehicle speed prediction model can be established, and vehicle speed prediction accuracy can be improved.

Next, the invention according to claim 2 has the function of the invention according to claim 1, and if the temporal change of the spatial average velocity can be predicted, the travel time required on the route can be estimated from the predicted value. In addition, the transition of the required travel time between any two points can be accurately obtained by repeating the calculation for determining the required travel time on the route.

Next, according to the third aspect of the present invention, in addition to the effect of the first aspect of the present invention, if the temporal change of the spatial average speed can be predicted, the section below a certain vehicle speed can be congested based on the predicted value. It is possible to determine a section, and further, by repeating the calculation for obtaining the congested section, the transition of the congestion length of the required travel time between any two points can be accurately obtained. After inputting the traffic density of the upstream section set by the means and the traffic densities of a plurality of downstream sections adjacent to the upstream section to the neural network after learning, and calculating the predicted spatial average speed of the upstream section The traffic volume in the upstream section is obtained by multiplying the traffic density by the predicted spatial average speed in the upstream section.
Then, the relationship between the traffic volume and the traffic density, that is, Q to K
Since the characteristic curve is created, the traffic capacity on the road can be accurately grasped.

[0015]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a functional block diagram showing an embodiment of a traffic flow prediction device according to the present invention.

This road traffic flow prediction device is provided at least one in each section on a road, and detects a vehicle passing through each section based on the average vehicle length. Traffic volume (vehicles / mi) based on the vehicle detection signal
n), an occupancy calculation means 3 for obtaining an occupancy from the output of the vehicle detector 1 or the output of the traffic calculation means 2 including the vehicle detector 1, the traffic calculation means 2 and the occupancy A spatial average speed calculating means 4 for obtaining a spatial average speed (corresponding to a vehicle speed) using a signal and other signals output from the calculating means 3 such as an average vehicle length, and an output signal and an average of the occupancy calculating means 3 A traffic density calculating means 5 for obtaining the traffic density using the vehicle length is provided.

Here, the occupancy means a ratio between a predetermined set time and a time occupied by the vehicle detector 1 for vehicle detection within the set time. Then, based on the spatial average speed and the traffic density obtained by the spatial average speed calculating means 4 and the traffic density calculating means 5, the neural network learning means 6, the neural network 7 after learning and the traffic flow predicting means 8 A predicted value of the spatial average speed (vehicle speed) of each section is obtained and stored in the vehicle speed transition prediction storage means 9.

The neural network learning means 6
Is a learning neural network 6A to which an occupancy actual value or a traffic density actual value of a certain section and an occupancy actual value or a traffic density actual value of a plurality of downstream sections of the section are inputted, and a vehicle speed prediction value of a certain section is obtained; A weighting factor learning program 6B for correcting the weighting factor of the network 6A so as to minimize the error between the spatial average speed (equivalent to the vehicle speed) output from the spatial average speed calculating means 4 and the output of the network 6A. ing.

Further, in this device, the required travel time calculating means 10 for obtaining the required travel time from the predicted spatial average speed stored in the vehicle speed transition prediction storage means 9 and the required travel time calculated from the required travel time calculation means 10 The required travel time transition calculating means 11 for calculating the required travel time transition between any two points using the required time, and the traffic congestion section is extracted from the spatial average speed prediction value stored in the vehicle speed transition prediction storage means 9. A congestion section calculating means 12 and a congestion section transition calculating means 13 for calculating a change in the congestion length between any two points based on the transition state of the congestion section extracted by the congestion section calculating means 12 are provided.

FIG. 2 is a diagram showing an example of a configuration for obtaining a Q (traffic volume) to K (traffic density) characteristic curve of a section using the neural network 7 shown in FIG. This figure 2
The traffic density setting means 21 sets and outputs the traffic density K (j) of a section j and the traffic densities K (j + 1), K (j + 2),... Of a plurality of sections downstream thereof. The traffic density K (j) of a certain section j and the traffic densities K (j + 1), K (j + 2),... Of a plurality of sections downstream from the traffic density setting means 21 are input, and the vehicle speed V ( j) neural network 7
Is multiplied by the vehicle speed V (j) obtained by the neural network 7 and the traffic density K (j) of a certain section j set by the traffic density setting means 21, and the traffic volume Q (j) of the certain section j is obtained. And a traffic volume-traffic density characteristic curve creating unit 23 that repeatedly executes a series of processes as described above and creates a traffic volume-traffic density characteristic curve from the obtained operation results. Traffic volume Q (j)-Traffic density K
(j) This is a configuration for creating characteristics.

Next, prior to describing the operation of the apparatus configured as described above, basic operations of the neural network 6A and the learned neural network 7 will be described.

FIG. 3 shows the neural network 6 of FIG.
It is a figure which shows the example of a structure of A and 7. This neural network includes an input layer, an intermediate layer, and an output layer each including a plurality of neuron elements, and a weight coefficient is set between each of these layers.

Now, in the case of the input layer constituting the neural network, the input and the output are the same. Therefore, the example of obtaining the output of the input layer is omitted here, and the example of obtaining the output of the intermediate layer and the output layer is omitted. Is described. The output Y _j (j = 1, 2,..., N) of the intermediate layer is represented by Y _j = F _j (U _Yj ) (1)

[0024]

(Equation 1) Required by In the above equations (1) and (2), Y _j : the output of the j-th neuron element in the intermediate layer in FIG. 3 W _ji : the i-th neuron element in the input layer and the i-th neuron element in the intermediate layer in FIG. X _i : the output of the i-th neuron element in the input layer in FIG. 3 (the input layer is the same as the input and the output) F: a function representing the input / output characteristics of each neuron element. Next, the output Z _k (k = 1, 2,..., N) of the output layer is obtained from the following equation. Z _k = F _k (U _zk ) (3)

[0025]

(Equation 2) Required by In the above equations (3) and (4), Z _k : the output of the k-th neuron element in the output layer W _kj : the weight coefficient between the j-th neuron element in the intermediate layer and the j-th neuron element in the output layer is there. As the function F, for example, a sigmoid function represented by the following equation is adopted.

F (U) = 1 / (1 + e ^−U ) −0.5 (5) Therefore, the neural network shown in FIG. 3 internally has the above-described equations (1) to (5). Is performed.

Next, the neural network learning means 6
Is a part having a function of learning the weighting factors W _kj and W _ji in the neural network of FIG. 3, and is learned by using a back propagation method (BP method) described below. This back propagation method is
Using the teacher signal V _k (spatial average velocity) and the output signal Z _{k of the} network 6A,

[0028]

(Equation 3) After calculating the error function E by performing the following calculation, the weight coefficient is corrected so that the error function E approaches the minimum value. learn. At this time, the weight coefficient correction amount 係数 W _kj ,
_Wji is obtained from the following equation.

[0029]

(Equation 4)

In the above equation, ε is a parameter that determines the magnitude of the correction performed at one time. Therefore, when the partial differential terms of the above equations (7) and (8) are expanded and arranged, the correction amount of the weight coefficient is expressed as follows.

[0031]

(Equation 5)

From the viewpoint of stabilizing the convergence and improving the convergence speed at the time of learning, it is also effective to employ the following expression instead of the expressions (9) and (10). _{ΔW kj} (m) = − ε · λ _k · Y _j + α · _{ΔW kj} (m-1) (13) ΔW _ji (m) = − ε · μ _j · Y _i + α · ΔW _ji (m-1) (14) Here, there is a relation of 0 <α <1, where α is a parameter for stabilizing learning, and m represents the number of times of learning.

Next, the operation of the traffic flow prediction device when the above-described neural network learning means 6 and neural network 7 are used will be described. The vehicle detection signal detected by the vehicle detector 1 in each section is sent to the traffic volume calculation means 2 and the occupancy calculation means 3, respectively.
Here, the traffic volume calculation means 2 counts the number of vehicles from the vehicle detection signal, and calculates the traffic volume [vehicle / mi] of each section (section number j = 1, 2,...).
n] is sent to the spatial average speed calculating means 4 for calculating the spatial average speed corresponding to the vehicle speed.

On the other hand, the occupancy calculating means 3 obtains a so-called occupancy [%], which is a ratio of the time during which the vehicle detector 1 is occupied by a passing vehicle in each section, and calculates a spatial average speed calculating means 4 and a traffic density calculating means 5. Respectively.

Here, the spatial average speed calculating means 4 calculates the spatial average speed [km / h] = {(average vehicle length [m / vehicle]) · (traffic volume [vehicle / min])} / {(occupancy [ %]]) / 100 [%])｝ (15) to calculate the spatial average velocity of each section.

In the traffic density calculating means 5, the traffic density [vehicles / m] = {(occupancy [%]]) / 100 [%])} / (average vehicle length [m / vehicle]) 16) The traffic density [vehicles / m 2] of each section is determined based on the following arithmetic expression.

From equation (16), since the average vehicle length is fixed, the occupancy and the traffic density are in a proportional relationship, so that the occupancy may be used in the neural network 6A instead of the traffic density.

After the traffic densities and the spatial average velocities of the sections and the plurality of sections downstream of the sections are determined in this way, the traffic density groups of the respective sections are introduced into the learning neural network 6A. , After calculating the predicted vehicle speed of a certain section, the weighting coefficient learning program 6B
By comparing the spatial average speed (measured vehicle speed value) with the predicted vehicle speed value, a weighting factor that minimizes the error between the two, that is, a weighting factor with high accuracy representing the predicted vehicle speed value relative to the measured vehicle speed value, is learned and the neural network is trained. It is stored in the network 7. Such a learning process is repeatedly performed for each section.

FIG. 4 is a diagram showing a calculation procedure for actually obtaining a predicted value of the spatial average speed (vehicle speed) using the traffic flow predicting means 8 after performing the learning process as described above.
The description will be made below in the order of the numbers of step STNO shown in FIG.

Step 1: The time t managed in the traffic flow prediction means 6 is set to the current time t0. Step 2: The current time t0 is obtained by multiplying the traffic density of each section obtained from the traffic density calculation means 5 by the section length.
In each section j (j = 1, 2,...) E (j, t
0) is calculated.

Step 3: The traffic density K (j, t) of each section is obtained by dividing the number E (j, t) of vehicles in each section at time t by the length L (j) of each section. It is introduced into the learned neural network 7 shown in FIG.

K (j, t) = E (j, t) / L (j) (j = 1,2,3,...) (17) where K (j, t): time of section j traffic density K at t
(j, t) E (j, t): Number of vehicles present at time t in section j L (j): length of section j Step 4: Set section number j to 0.

Step 5: Set the section number j to j + 1. j = j + 1 Step 6: Thereafter, the section j and a plurality of downstream sections of the section, that is, a plurality of adjacent sections j, j + 1, j +
, Based on the traffic density at time t, the vehicle speed V (j,
t) is obtained and introduced into the traffic flow prediction means 8 as shown in FIG.

* The following steps 5 to 6
Is repeated by the required number of sections. Step 7: Set section number j to 0. Step 8: Set the section number j to j + 1.

Step 9: Interval j between time t and t + Δt
The traffic volume QIN (j, t to t + △ t) flowing into the vehicle is calculated as follows: QIN (j, t to t + △ t) = V (j-1, t) · K (j-1, t) 18), and the traffic volume QOUT (j, t to t + △ t) flowing out of the section j is calculated as follows: QOUT (j, t to t + △ t) = V (j, t) · K (J, t)... (19)

Step 10: The number of vehicles existing in the section j at the time t + Δt is obtained from the following equation. “Vehicle presence number E (j, t + △ t) in section j” = “vehicle presence number E in section j
(J, t)] + [traffic flow QIN (j, t-t + Δt) flowing into section j between time t-t + Δt] − [time t-t + Δt]
During this time, the traffic volume QOUT flowing out of section j (j, t to t + △
t)] * The calculations in steps 8 to 10 are repeated for the required number of sections.

Step 11: Advance the time t by Δt, t
Set to + Δt. * Note that the calculations in steps 3 to 11 are repeatedly terminated for a required time.

As described above, the vehicle speed V (j, t
) (J = 1,2,3,…) is predicted.
It is stored in the vehicle speed transition prediction storage means 9. Therefore, the transition of the vehicle speed in each section can be extracted from the vehicle speed transition prediction storage means 9 with high accuracy.

FIG. 5 shows a temporal transition of the measured value and the predicted value of the vehicle speed in a certain section obtained as described above, and the measured value and the predicted value of the vehicle speed accurately match each other. I can understand.

Further, the required travel time calculating means 10 calculates V (j, t) (j = 1,2) which is a predicted value of a temporal change of the vehicle speed in each section j stored in the vehicle speed transition prediction storing means 9. ,
3, ...) and a predetermined section length L (j) (j = 1,
2, 3,...), The required travel time between any two points can be obtained, and the required travel time changes with the departure time. The required time, that is, the transition predicted value of the required travel time between any two points can be calculated and accurately obtained. The calculation of the required travel time here is performed at the point A
It is to predict the travel time required from the vehicle to the point B when the vehicle departs from the point A.

Further, in the traffic congestion section calculating means 12, V (j, t) (j = 1,2,3) which is a time change of the vehicle speed in each section j stored in the vehicle speed change prediction storing means 9. ,
…) And a predetermined section length L (j) (j = 1,2,
3,...) And a certain speed, for example, 20 km / h
If it is the following section, the section is extracted as a congestion section. Then, below the extracted congested section, the congested section transition calculating means 13 can accurately obtain a temporal transition predicted value of the total length of the congested section included between any two points.

Next, an example of obtaining a Q (traffic volume) to K (traffic density) characteristic curve from the configuration of FIG. 2 using the neural network 7 will be described with reference to the flowchart shown in FIG. Incidentally, the Q-K characteristic curve is a very important curve for knowing the traffic capacity of the route, for example, how much traffic density at which congestion occurs or how much traffic can be passed. Used to figure out.

Step 21: First, the traffic density setting means 2
1 For example, from traffic density calculation means 5 or the like,
The traffic densities K (j), K (j + 1), K (j + 2
), (In FIG. 1, the traffic density obtained in step 3 for explaining the operation of the traffic flow predicting means 8) is set.

Step 22: Traffic densities K (j), K (j + 1), K (j + 2) set by the traffic density setting means 21
), Are input to the neural network 7, and the vehicle speed V (j) is obtained based on the weight coefficient learned here.

Step 23: The traffic density K (j) of the section j is multiplied by the vehicle speed V (j) output from the neural network 7 by the multiplying means 22 to obtain the section j shown in the following equation.
Traffic volume Q (j) is calculated.

Q (j) = K (j) · Vehicle speed V (j) (20) Step 24: Note that the calculations in steps 21 to 23 are performed by various K (j), K (j + 1), K (j + 2),... Is repeated to determine the relationship (characteristic) between the traffic volume and the traffic density in the section j, thereby obtaining the downstream section j, j + 1,
You can understand how the traffic density situation of j + 2, ... changes.

FIG. 7 is a graph showing the relationship between the traffic density and the traffic volume in a certain section on the route. The horizontal axis indicates the occupancy proportional to the traffic density, and the vertical axis indicates the traffic volume. It is proportional (occupancy) and takes the vehicle speed. It can be understood that the traffic volume that can pass through decreases as the occupancy on the downstream side increases. Although the traffic density is input to the neural networks 6A and 7 in the above embodiment, an occupancy proportional to the traffic density may be input.

[0058]

As described above, according to the present invention, the following various effects can be obtained. According to the first aspect of the present invention, a learning model for predicting the vehicle speed using the neural network can be established, and the vehicle speed can be predicted with high accuracy.

According to the second aspect of the present invention, since the vehicle speed can be predicted with high accuracy, it is possible to accurately predict the required traveling time at a plurality of points from the predicted value of the vehicle speed. Next, the invention of claim 3 is:
Since the vehicle speed can be predicted with high accuracy, the traffic congestion section and the traffic congestion length can be appropriately predicted from the predicted value of the vehicle speed. Furthermore, according to the invention of claim 4, since the vehicle speed can be predicted with high accuracy, the traffic capacity on the road can be accurately grasped.

[Brief description of the drawings]

FIG. 1 is an overall configuration diagram showing an embodiment of a traffic flow prediction device according to the present invention.

FIG. 2 is a diagram showing a configuration for obtaining characteristics of traffic volume-traffic density in each section.

FIG. 3 is a diagram showing a basic configuration of a neural network used in the device of the present invention.

FIG. 4 is a flowchart for explaining the operation of the traffic flow prediction means shown in FIG. 1;

FIG. 5 is a diagram comparing a temporal transition between an actually measured value and a predicted value of the vehicle speed in a certain section.

FIG. 6 is a flowchart for explaining the operation of the configuration shown in FIG. 2;

FIG. 7 is a graph showing a relationship between traffic density and traffic volume in a section on a route using the apparatus shown in FIG. 2;

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 ... Vehicle detector, 3 ... Occupancy calculation means, 4 ... Spatial average speed calculation means, 5 ... Traffic density calculation means, 6 ... Neural network learning means, 7 ... Neural network after learning, 8 ... Traffic flow prediction means, 9 ... vehicle speed transition prediction storage means, 10 ... required travel time calculation means, 11 ... required travel time transition calculation means, 12 ... congestion section calculation means, 13 ... congestion section transition calculation means, 21 ... traffic density setting means, 22 ...
Multiplying means, 23: means for creating traffic volume-traffic density characteristic curve.

Claims (4)

(57) [Claims] 1. A traffic density actual value of an upstream section obtained from a vehicle detector provided in each section on a road, a traffic density actual value of a plurality of downstream sections adjacent to the upstream section, and the upstream section. spatial average and neural network learning means for learning the weighting factor using the actual speed, based on the weight coefficient learned by the neural network learning device, the traffic density and the upstream section of the upstream section of the enter the traffic density downstream plurality of sections adjacent to a neural network for determining the spatial average velocity estimated value of the upstream section, the spatial average speed predicted value obtained in the neural network and the traffic density using Traffic flow prediction means for predicting and calculating a temporal change in traffic density and a temporal change in the spatial average speed. Place. 2. The traffic flow prediction device according to claim 1, further comprising means for calculating a required travel time on a route from a temporal change of a spatial average speed predicted by said traffic flow prediction means. Traffic flow prediction device. 3. The traffic flow prediction device according to claim 1, further comprising means for obtaining a traffic congestion section on a route from a temporal change of a spatial average speed predicted by said traffic flow prediction means. Flow prediction device. Wherein the upstream section Transport density and traffic density setting means and, the traffic density setting means set upstream section to set the traffic density downstream plurality of sections which is adjacent to the upstream section Traffic
And traffic density of multiple downstream sections adjacent to this upstream section
A neural network on the basis of the degree determine the spatial average velocity estimated value of the upstream section, the upstream section with a spatial average velocity prediction value of the upstream section obtained and the traffic density by the neural network multiplication means for determining the traffic volume, traffic from a traffic <br/> density of the obtained traffic volume and the upstream section in the multiplying means - traffic creates a characteristic curve of the traffic density - traffic density characteristic curve A traffic flow prediction device comprising: a creation unit;