JP2533719B2 - Traffic situation detection method - 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 condition detecting method using a neural network, and particularly to an analog method such as a degree of vehicle congestion in a service area such as an expressway or a parking area and a degree of traffic congestion on a road. The present invention relates to a traffic condition detection method suitable for detecting a traffic condition in a traffic control system that requires condition judgment.

[0002]

2. Description of the Related Art A conventional traffic condition detecting method will be described. For example, it is known that a large parking lot is divided into a plurality of blocks, and the parking rate of a specific block among them is highly correlated with the parking rate of the entire parking lot (see “Highway”). And automobiles "Vol. 32, No. 1, 37-
(See page 49). That is, in view of the existence of a specific block showing a parking rate having a high correlation with the parking rate of the entire parking lot, the congestion degree of the vehicle can be detected by finding the "habit" of this parking lot. Japanese Patent Laid-Open No. 4-15800 proposed by the applicant of the present invention proposes a traffic condition detection method for detecting the congestion degree of a vehicle by learning this "habit" using a neural network.

The traffic condition detecting method by the applicant will be described below with reference to FIG. In FIG.
A parking lot 1 is provided with parking blocks 1 _{1 to} 1 _n . Reference numeral 2 is a vehicle congestion degree detection device that measures the parking condition, and 3 is a representative area parking condition measurement device that measures the representative area of the parking block. Reference numeral 4 is an entrance status measuring device, and 5 is an exit status measuring device.

FIG. 1B shows an outline of the vehicle congestion degree detection device 2, 6 is an information processing unit, and 7 is a display unit. I is an input layer, H is an intermediate layer, and O is an output layer. The input layer I, the intermediate layer H, and the output layer O are neurons (I
_{1 to} I ₄ , H _{1 to} H ₄ , O _{1 to} O ₃ ) constitute a neural network, and each neuron transmits information by a non-linear response. Correct events T _{1 to} T ₃ are supplied to the output layer O, and the back propagation method (back propagation learning method) advances in the opposite direction to the signal propagation as indicated by an arrow 8 to output from the intermediate layer H. The coupling weight coefficient V _jk to the layer O and the coupling weight coefficient W from the input layer I to the intermediate layer H
The _ij, for each teacher signal (correct answer event) T ₁ through T ₃ is input, and the output O _K1 ~ O _K3, the answer event T ₁ corresponding thereto
It is corrected so that the error from ˜T ₃ is minimized. 7a is a full vehicle display, 7b is a congestion degree display, and 7c is an empty vehicle display.

As the response characteristic f (X) of the neuron used in the input layer I, the intermediate layer H, and the output layer O, a nonlinear sigmoid response function as shown in FIG. 6 is used. f (X) = 1 / {1 + exp (−2X / u ₀ )} (1) u ₀ in the equation (1) is a positive parameter that determines the slope of the sigmoid function. Hereinafter, learning of the neural network will be described with reference to FIG. Differentiating the expression (1) with respect to X, the following expression is obtained. f ′ (X) = 2 · f (X) · {1-f (X)} / u ₀ (2) The outputs I _in , H _jn , and 0 _kn of each layer are expressed as follows. .

[0006]

[Equation 1]

(Where θ _j is the offset of neurons in the intermediate layer H and γ _k is the offset of neurons in the output layer O)

If the error between the output O _Kn of the output layer O and the teacher signal T _K is e _K (= T _K −O _Kn ), this mean square error E _K is expressed by the following equation. It E _K = 1 / _2 (T _K −O _Kn ) ² ……………… (5) The state in which the mean square error E _{k of the} neural network is the minimum is the optimal learning state. The mean squared error E _k is minimized with respect to the combination weighting factor V _jk . It is a learning action in the neural network to correct the coupling weight coefficient and the offset so that the mean square error E _k is minimized.

On the other hand, the function of the mean square error is a function of the output O _Kn of the output layer O, and the output O _Kn is a function of the output H _jn of the intermediate layer H. _Therefore , the following relation holds. ∂E _K / ∂V _jk = ∂E _K / ∂O _Kn・ ∂O _Kn / ∂V _jk ……… (6) ∂E _K / ∂O _Kn ＝ － (T _K −O _Kn ) ……………… (7) Here, the equation (4) is written as follows.

[0010]

[Equation 2]

When S _K is differentiated with respect to V _jk, it is expressed as ∂S _K / ∂V _jk = H _jn ............ (9).

A change in the output O _Kn with respect to a slight change in the coupling weight coefficient V _jk is expressed as follows. ∂O _Kn / ∂V _jk = ∂O _Kn / ∂S _K・ ∂S _K / ∂V _jk = f '(S _K ) ・ H _jn = 2 / u ₀・ O _Kn・ (1─O _Kn ) ・ H _jn …… (10)

Therefore, the change amount of the mean square error E _K with respect to the minute change of the coupling weight coefficient V _jk is expressed by the following formula from the formulas (7) and (10). ∂E _K / ∂V _jk = ∂E _K / ∂O _Kn・ ∂O _Kn / ∂V _jk = −H _jn・ 2 / u ₀・ (T _K −O _Kn ) ・ O _Kn・ (1─O _Kn ) = -Δ _K · H _jn ………………………… (11) (where δ _K = 2 / u ₀ · (T _K −O _Kn ) · O _Kn · (1
─O _Kn ). )

The mean squared error E _K reaches a minimum value by repeating the learning for modifying the coupling weight coefficient V _jk in the direction in which the mean squared error E _K decreases. The correction amount ΔV _jk
Is expressed by the following equation. ΔV _jk = −η ₁ (∂E _K / ∂V _jk ) = η ₁ · δ _K · H _jn …………………… (12) (where η ₁ is a constant) Next, the same method The correction amount ΔW _ij is calculated by Here, the formula (3) is written as follows.

[0015]

(Equation 3)

The change amount of the mean square error E _K with respect to the minute change of the coupling weight coefficient W _ij is as follows.

[0017]

[Equation 4]

[0018] Therefore, the correction amount ΔW _ij of the coupling weight coefficient W _ij
Is expressed by the following equation.

[0019]

(Equation 5)

When the correction amounts Δγ _k and Δθ _j are _obtained by a similar method, they are expressed by the following equation. ∂E _K / ∂γ _k = ∂E _K / ∂O _Kn · ∂O _Kn / ∂γ _k = 2 / u ₀ · (T _k −O _kn ) · O _Kn · (1-O _Kn ) = − δ _K Therefore, the correction amount Δγ _k is Δγ _K = −η ₃ · (∂E _K / ∂γ _K ) = η ₃ · δ _K …………………… (16) (where η ₃ is a constant)

[0021]

(Equation 6)

The above η ₁ to η ₄ are learning coefficients for controlling the magnitude of the correction amount.

[0023]

In the traffic situation detecting method using the neural network as described above, the sigmoid as shown in the equation (1) is used as the response function of the neurons of the input layer I, the intermediate layer H and the output layer O. Functions are used. The sigmoid function is a response function by an exponential function, and in order to set its output f (X) to 0 or 1, X =
No output unless set to –∞ and + ∞. Therefore, the output f
The number of computations until (X) reaches 0,1 is extremely large, and there is a drawback that it does not easily converge, that is, 0 (empty car) and 1 (full car), which are the extreme values of the output f (X). There is a problem that it is difficult to output the value. However, when measuring the congestion level of a parking lot, the parking rate is 10%.
A value of 0% is a condition that can occur frequently in daily life, and in order to improve the accuracy, it is difficult to output this extreme value, which is a problem.

The present invention has been made in view of the above problems, and an object of the present invention is to provide a traffic condition detecting method capable of detecting a traffic condition with high accuracy.

[0025]

In order to solve the above problems, the present invention provides a traffic condition detecting method for detecting a traffic condition by a multilayer neural network, wherein a linear function is used as a response function of an output layer of the multilayer neural network. It uses a function.

[0026]

The traffic condition detecting method of the present invention is based on a multi-layer neural network including an input layer, at least one intermediate layer, and an output layer. The sigmoid response function neurons are used for the neurons of the input layer and the intermediate layer. The learning efficiency is maintained by using, and by using a neuron having a linear function as a response function of the output layer, the neuron is easily converged so that the extreme value of 1 can be output quickly.

[0027]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings. In FIG. 1A, 1 is a specific block 1 ₁ to 1 _n.
A relatively large-scale parking lot divided into 4 is shown, 4 is an entry situation measuring device, 5 is a departure situation measuring device, 11 to 14 are representative area parking situation measuring device, and 15 is a vehicle congestion degree detecting device. .

FIG. 1B shows a vehicle congestion degree detection device 15, which comprises an information processing unit 16 and a parking rate display unit 17. The vehicle congestion degree detection device 15 receives signals indicating vehicle conditions from the entry status measurement device 4, the departure status measurement device 5, and the representative area parking status measurement devices 11 to 14, respectively. The information processing unit 16 constitutes a neural network by an input layer I composed of neurons I _{1 to} I ₅ , an intermediate layer H composed of neurons H _{1 to} H ₄ and an output layer L composed of neurons L ₁ . The teacher signal T based on the measured value of the vehicle congestion situation is input to the output layer L, and as shown by an arrow 8, the output layer L to the intermediate layer H in the direction opposite to the signal direction.
Then, it is propagated to the input layer I and is repeatedly learned, and the connection weight coefficient is modified so that the output error is minimized. The parking rate from the empty state to the full state is output to the parking rate display unit 17 as an output signal L _k . The parking rate signal is
Parking status display board (not shown) installed at the entrance of the parking lot
Is also output to

For the neurons I _{1 to} I ₅ of the input layer I and the neurons H _{1 to} H ₄ of the intermediate layer H, the sigmoid response function shown in the equation (1) is used. Neuron L of output layer L
₁ uses a linear function having a linear response characteristic in the range of g (X) = 1,0 as shown in FIG. g (X) = aX + b (18) A in the equation (18) is a parameter that determines the slope of the linear function, and changes as shown in the figure depending on the magnitude of a.
In addition, b takes a value of 0.5.

The operation of the embodiment shown in FIG. 1 will be described below separately for the learning operation and the measurement of the parking condition. (A) Learning Operation The entry status measuring device 4 measures the entry rate by counting the number of vehicles entering the parking lot 1 per unit time. The departure condition measuring device 5 measures the departure ratio by counting the number of vehicles leaving the parking lot 1 per unit time. The representative area parking status measuring devices 11 to 14 detect the parking status in each representative parking block of the parking lot 1, measure the parking rate of each parking block, and input layer of the neural network configuring the information processing unit 16. Input to the neurons I _{1 to} I ₆ of I, respectively. On the other hand, information about the actual parking situation is input to the output layer L of the information processing unit 16 as a teacher signal (correct answer event) T.

The neural network of the information processing unit 16 uses the entry rate from the entry status measuring device 4, the departure rate from the departure status measuring device 5, and the representative area parking status measuring devices 11 to 11.
Based on each measurement value regarding the parking rate from 14 and the teacher signal (correct answer event), learning is performed many times based on the back propagation method. The neural network has a learning function and a self-organizing function so that the coupling weight coefficients W _ij of the input layer I, the intermediate layer H, and the output layer L are reduced so that the error between the output L _kn and the corresponding correct event T becomes small. Automatically correct the value of U _jk . W _ij is a coupling weight coefficient from the input layer I to the intermediate layer H, and U _jk is a coupling weight coefficient to the intermediate layer H and the output layer L.

Next, the learning performed by the information processing section 16 will be described. As shown in FIG. 1B, the coupling weight coefficient of the input layer I and the intermediate layer H is W _ij , the coupling weight coefficient of the intermediate layer H and the output layer L is U _jk , and the neurons of the intermediate layer H and the output layer are The correction amounts will be described _assuming that the offsets are θ _j0 and γ _k0 , respectively. The outputs H _jn and L _kn of each layer are expressed by the following equation.

[0033]

(Equation 7)

If the error between the output L _Kn of the output layer L and the teacher signal T is δ _K0 (= T−L _K ), this average 2
The multiplication error E _K is expressed by the following equation. E _K = 1/2 ・ (T-L _Kn ) ² ……………… (21)

On the other hand, the function of the mean square error E _k is the output layer L
Is a function of the output L _K, the output L _K output H of the intermediate layer H
_Since it is a function with _jn , the following relation holds. ∂E _K / ∂U _jk = ∂E _K / ∂L _Kn · ∂L _Kn / ∂U _jk ……… (22) ∂E _K / ∂L _Kn ＝ － (T−L _Kn ) ……………… (23) Here, the equation (20) is written as follows.

[0036]

(Equation 8)

The influence on the output L _K with respect to a slight change in the coupling weight coefficient U _jk is expressed as follows. ∂L _Kn / ∂U _jk = ∂L _Kn / ∂S _K・ ∂S _K / ∂U _jk = g '(S _K ) ・ H _jn = a ・ H _jn …………………… (26)

Therefore, the change amount of the mean square error E _K with respect to the minute change of the coupling weight coefficient U _jk is expressed by the equation (23) and (2)
From the equation (6), it is expressed as the following equation. _{∂E K / ∂U jk = ∂E K} / ∂L Kn · ∂L Kn / ∂U jk = -a · (T-L Kn) · H jn = -δ k '· H jn .................. (27) (where δ _k ′ = a · (T−L _Kn )) Mean square error is repeated by repeating learning to correct the coupling weight coefficient U _jk in the direction of decreasing the mean square error E _K. Error E
_K reaches a local minimum. The correction amount ΔU _jk is expressed by the following equation. ΔU _jk = −η ₁ ′ (∂E _K / ∂U _jk ) = η ₁ ′ · δ _k ′ · H _jn …………………… (28) (However, η ₁ ′ is a constant)

Next, the correction amount Δ
When W _ij , _{Δγ k0} , and Δθ _j0 are obtained, they are expressed by the following equation. Note that γ _k0 and θ _j0 indicate the offsets of the neurons in the output layer L and the intermediate layer H. Hereinafter, the correction amount ΔW _ij will be obtained.

[0040]

[Equation 9]

Next, the correction amount Δγ _k0 is obtained. ∂E _K / ∂γ _K0 = ∂E _K / ∂L _Kn・ ∂L _Kn / ∂γ _K0 = −a (T−L _Kn ) = − δ _K ′ Δγ _K0 = −η ₃ ′ (∂E _K / ∂ γ _K0 ) = η ₃ ′ · δ _K ′ ……………… (30) (However, η ₃ ′ is a constant.)

[0042]

[Equation 10]

However, η ₁ ′ to η ₄ ′ are learning coefficients that control the magnitude of the correction amount.

(B) Measurement of parking condition In the traffic condition detecting method of the present invention, when the parking condition is measured, the number of vehicles in the parking lot 1 is counted by the entrance condition measuring device 4 per unit time. The vehicle rate, the vehicle departure rate obtained by counting the number of vehicles leaving the parking lot 1 per unit time from the vehicle departure situation measuring device 5, and the parking rate from the representative area parking situation measuring devices 11 to 14 are respectively processed as information. Device 16
Are input to the neurons I _{1 to} I ₆ of the input layer I of FIG. The information processing unit 16 performs arithmetic processing based on the parking information, and outputs the parking status of the entire parking lot as the parking rate. As the output L _k, the parking rate from the empty state to the full state is output. Based on the parking rate obtained from the information processing unit 16, the full rate (parking rate 0.7 to 1.0), the congestion rate (parking rate 0.5 to 0.7), and the vacancy rate (parking rate 0.0 to). 0.
If it is 5), the parking state can be displayed accordingly.

FIG. 2 is a table showing each data from the entry status measuring device 4, exit status measuring device 5 and representative area parking status measuring devices 11-14. FIG. 3 is a comparison between the actual measurement value of the parking rate and the parking condition measuring device, and is a table showing the measurement results (parking rate) obtained by using the sigmoid function and the linear function in the output layer. is there. The leftmost columns in FIGS. 2 and 3 are sample numbers (1 to 40), which are shown by selecting odd numbered data from the measured data.
The sample numbers in FIGS. 2 and 3 correspond to each other. This data is obtained by measuring the parking situation at a parking lot for a predetermined period of time and rearranging the data by the actually measured parking rate.

The parking lot of the parking lot based on this data has a capacity of 138 cars, which is equivalent to a small car.
The "number of vehicles entering, rate of entry" column in FIG. 2 is the result measured by the vehicle entry status measuring device 4, and shows the number of vehicles entering the parking lot and the rate of entry. "Number of departures, departure rate"
The column shows the results measured by the departure status measuring device 5, and shows the number of vehicles leaving the parking lot and the departure rate thereof at a fixed time. The "sensor 1" to "sensor 4" columns show the number of vehicles in each specific block from the representative area parking status measuring devices 11 to 14 and the parking rate thereof.

The "actually measured value" column in FIG. 3 shows the number of parked vehicles and the parking rate in the entire parking lot by the actual measurement, and the "determination result" column shows the measurement result by the vehicle congestion degree detecting device. . The parking rate when the sigmoid function and the linear function are used as the response function of the output layer of the neural network which comprises an information processing part is shown. As is clear from the measurement result of FIG. 3, the measured parking rate and the parking rate determined by the vehicle congestion degree detection device are similar to each other.
Also, as a response characteristic of the neuron in the output layer, comparing the parking rate when using the sigmoid function and the linear function,
The approximate values are obtained, and it can be demonstrated that sufficient learning is done.

Further, the sample number 40 is data when 138 cars are parked with respect to the 138 small-sized storage units. In the sigmoid function, the parking rate is 0.95, while the linear function In this case, the parking rate is 1.00. That is, by using a linear function for the neuron L ₁ of the output layer L, the number of calculations can be reduced and a characteristic that can be easily converged, that is, a full state (parking rate of about 1.0) can be easily output. It shows the characteristics to obtain.

FIG. 4 is a graph of a part of the data shown in FIGS. 2 and 3. The solid line in FIG. 4 is the actual value of the parking rate, and the dotted line is the measured value when the sigmoid function is used. The alternate long and short dash line is the measured value when the linear function is used. When the sigmoid function was used, it did not converge easily, but the measurement result using the linear function for the neurons in the output layer shows a full state (parking rate 1.0), indicating that the traffic situation of the present invention is high. It has been demonstrated that the detection method can faithfully output the full state.

Of course, in the traffic situation detecting method of the present invention, only the parking situations of a plurality of specific blocks in the parking lot may be input to the vehicle congestion degree detecting device to detect the parking rate of the entire parking lot. Data on rate and departure rate are not always required. Therefore, the entire parking rate can be detected only by the data of the parking situation from a plurality of specific blocks. It is obvious that the intermediate layer in the neural network of the information processing unit may be a multilayer instead of one layer as in the embodiment.

Furthermore, as an application example of the traffic situation detecting method of the present invention, not only the detection of the parking rate as in the above embodiment,
It can be applied to various kinds of traffic condition judgment, and is obviously effective when analog judgment in the range of (0.0 to 1.0) in the detection of traffic condition is required. For example, when detecting the degree of traffic congestion on the road from the speed of vehicles traveling on the road, the number of vehicles passing, etc., by giving the input layer the vehicle speed and the number of vehicles passing per unit time as input parameters, the congestion level from the output layer ( 0.0 to 1.0) can be output. Also in this case, as in the above-mentioned embodiment, “no congestion” (congestion level 0.0), “complete congestion”
It is possible to reasonably judge the extreme value state of (congestion degree 1.0).

Further, the traffic condition detecting method of the present invention can be applied to a traffic control system for controlling lighting time of red and green signals for controlling traffic flow. For example, the average vehicle speed from the various vehicle detection sensors installed at a plurality of points, the number of passing vehicles, etc. may be given to the input layer, and the output layer may output the signal lighting time at a normalized value of 0.0 to 1.0 ( For example, 0.0
In case of, lights for 1 minute, and in case of 1.0, lights for 5 minutes). In this case, if the vehicle speed is equal to or higher than a certain value and the number of passing vehicles is reduced by the learning process, the green signal lighting time may be the shortest, and the lowest value of 0.0 is output. Also, if the vehicle speed is below a certain value and the number of vehicles passing through is increasing, it is expected that there will be considerable congestion, so the traffic signal lighting time will be 1.0
To be output. In the intermediate state, the intermediate value between 0.0 and 1.0 is continuously output according to the situation. By applying it to the traffic control system in this way,
It is possible to control the lighting time of the traffic signal by automatically determining the degree of congestion according to the input from the sensor.

[0053]

As described above, according to the traffic condition detection method of the present invention, the linear function is used as the response function of the neurons in the output layer of the neural network, so that the congestion degree of the vehicle and the road It is extremely effective in being able to detect traffic conditions such as the degree of traffic congestion, and has the advantage that the extreme values of 0.0 and 1.0 can be output with high accuracy without difficulty as the output of the neural network. are doing. In addition, the neural network used in the traffic condition detection method of the present invention has an advantage that the number of calculations can be reduced because the output result can be obtained from one output layer, and the calculation efficiency can be improved. It is possible to obtain a processing result at high speed. Further, as the output of the neural network, the extreme values of 0.0 and 1.0 are output with high accuracy without any difficulty.
Can be used for traffic control systems.

Further, according to the traffic condition detection method of the present invention, a continuous output from 0.0 to 1.0 can be output with high accuracy in detecting the traffic condition. It is an extremely effective method for detecting various traffic conditions in a traffic control system that requires judgment.

Of course, by using the linear function as the response function of the neuron in the output layer of the neural network, there is an advantage that the processing time for correcting the coupling weight coefficient between the respective layers can be shortened also in the learning process, and the extreme value 0 It is extremely effective in that .0 and 1.0 can be learned relatively quickly.

[Brief description of drawings]

1A is a diagram showing an arrangement of a vehicle congestion degree detection device in a parking lot, FIG. 1B is a block diagram showing an embodiment of a traffic situation detection method, and FIG. 1C is a neuron response characteristic. FIG.

FIG. 2 is a table showing measured values of each measuring device.

FIG. 3 is a table showing determination results when a sigmoid function and a linear function are used for an output layer with respect to an actually measured parking rate.

FIG. 4 is a graph showing a measurement result of a vehicle congestion degree when a measured value and a sigmoid function and a linear function are used in an output layer of a neural network.

5A is a diagram showing an arrangement of a vehicle congestion degree detection device in a parking lot, and FIG. 5B is a block diagram showing an example of a conventional traffic condition detection method.

FIG. 6 is a diagram showing response characteristics of neurons.

FIG. 7 is a diagram for explaining a neural network.

[Explanation of symbols]

1 Parking lot 1 ₁ to 1 _n Specific block 4 Vehicle entry status measuring device 5 Vehicle departure status measuring device 11-14 Representative area parking status measuring device 15 Vehicle congestion degree detection device 16 Information processing unit 17 Parking rate display device I Input layer I _{1 to} I ₆ input layer neuron H intermediate layer H _{1 to} H ₄ intermediate layer neuron L output layer L ₁ output layer neuron

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Muneo Yamada, Miwa-cho, Kaifu-gun, Aichi Oda Shinoda, Ota 29-1 Miwa Plant, Nagoya Denki Kogyo Co., Ltd. (56) Reference JP-A-4-15800 (JP, A) )

Claims (1)

(57) [Claims] 1. A traffic situation detecting method using a neural network, wherein the neural network receives an input layer to which traffic information such as a vehicle speed, a parking rate, or the number of passing vehicles is input, and an output from the input layer. It is composed of at least one intermediate layer and an output layer for outputting the degree of congestion and the degree of vehicle congestion, and uses a sigmoid function as the response characteristic of each neuron of the input layer and the intermediate layer, and also the response function of the neuron of the output layer. A traffic condition detection method characterized by detecting the degree of traffic congestion and the degree of vehicle congestion by using a linear function as.