JPH0729087A - Device for predicting traffic quantity - Google Patents

Abstract

PURPOSE:To obtain a traffic quantity predicting device capable of predicting traffic quantity with high precision even when the trend of traffic quantity changes. CONSTITUTION:A traffic quantity predicting part 12 which predicts future traffic quantity from pre-processed data based on a difference equation indicating the time sequence relation of traffic quantity is provided. The traffic quantity predicting part 12 is formed by a neural network and also the device is provided with a teacher data converting part 15 which converts teacher data taken out from a traffic quantity data base 14 into the data form of the output of the traffic quantity predicting part 12 so as to generate a teacher signal and a prediction traffic quantity evaluating part 16 which transmits an evaluation value based on the comparison of the teacher signal with a prediction output predicted from input data which the traffic quantity predicting part 12 takes out from the traffic quantity data base 14 to the traffic quantity predicting part 12 as data for learning.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a traffic volume predicting device for predicting traffic volume, which is used in a traffic means control device for realizing efficient control of transportation means such as elevators, road traffic and railways. is there.

[0002]

2. Description of the Related Art FIG. 9 shows, for example, August 19-21, 1982.
The conventional traffic volume prediction device shown in the paper entitled "Elevator group management method with traffic pattern learning function" (Preliminary Proceedings page 255-230) presented at the 8th System Symposium held in Hokkaido on the day FIG. In the figure, 1 is a traffic volume database that stores traffic volume data at each time, and 2 is an average traffic volume database that stores the average traffic volume at each time point obtained by learning the traffic volume data. Reference numeral 3 is a traffic volume prediction unit that obtains a predicted traffic volume x (t) at time t using the average traffic volume database 2 and 4 is a traffic volume prediction unit, and 4 is a traffic volume prediction unit in the traffic volume prediction unit 3. It is a diagram illustrating the concept.

Next, the operation will be described. Here, a description will be given by taking an example of the traffic volume prediction of an elevator by the traffic volume prediction apparatus shown in FIG. 2. Description of the Related Art Conventionally, an elevator group management control system that integrally controls a plurality of elevators has been adopted in order to perform a more efficient service when a plurality of elevators are installed in one building. This elevator group management control system monitors the elevator traffic such as the number of passengers getting on and off, the state of the elevator system such as the number of calls in halls and cars, and allocates the elevators so that the elevator system as a whole is most efficient. Is. In this elevator group management control, if not only the current system state but also the future state can be known, more precise group management control becomes possible. Therefore, the traffic volume prediction apparatus shown in FIG. 9 predicts the traffic volume.

That is, in the traffic volume predicting apparatus shown in FIG. 9, the traffic volume is predicted by obtaining the average traffic volume at each time t. That is, as shown in FIG. 10A, the average traffic volume at each time t is obtained in advance from the traffic volume data for several days by learning the actual traffic volume data and stored in the average traffic volume database 2. At the time of prediction, the time t at which the traffic volume prediction unit 3 wants to predict
The data of the predicted traffic volume x (t) at any future time is obtained by extracting the average traffic volume corresponding to the above from the average traffic volume database 2. It should be noted that such an idea can be understood as obtaining a mapping relationship between the time t and the traffic volume x (t), as shown as a prediction concept 4 in FIG.

As described above, in the traffic volume predicting apparatus, the continuity of changes in traffic volume and the time correlation are not so importantly emphasized. Therefore, the time itself and the traffic volume are simply associated with each other, and The predicted traffic volume at any time was obtained by mapping to the predicted traffic volume. Therefore, the prediction depends only on the time, and as shown in FIG. 10B, even if the tendency of the traffic volume is different from the previous day such as the tendency of the traffic volume being shifted with time, the average traffic volume database 2 Based on the above, the traffic volume prediction result was exactly the same as the previous day. ,

In addition to the above, there is, for example, the one in which the feature detection of the traffic demand of the elevator is performed by predicting the start time of the feature vector of the traffic flow.
It is disclosed in Japanese Patent Publication No. However, even in those cases, as in the case described above, the relationship between the feature vector and its change time is still obtained.

[0007]

Since the conventional traffic volume prediction apparatus is configured as described above, the same prediction result as the previous day is output even when the tendency of traffic volume changes, and therefore the number of passengers getting on and off. There is a problem in that it is not possible to predict the traffic volume with high accuracy when there is an increase in traffic flow or when there is a time lag in the tendency of changes in traffic volume.

The present invention has been made to solve the above-mentioned problems, and it is possible to predict the traffic volume with high accuracy even if the tendency of the traffic volume is different from that of the previous day. The objective is to obtain a reliable traffic volume prediction device.

[0009]

According to a first aspect of the present invention, there is provided a traffic volume predicting apparatus in which data of a future traffic volume is discretized by a data preprocessing unit and preprocessed such as smoothing and delaying. In addition, a traffic volume predicting unit that predicts traffic volume based on a difference equation indicating a time-series relationship is provided.

In the traffic volume predicting apparatus according to the present invention, the traffic volume predicting unit is formed by a neural network, and the teacher data extracted from the traffic volume database is output in the data format of the traffic volume predicting unit. And a teacher data conversion unit that generates a teacher signal by generating a teacher signal, and the teacher signal and the prediction output predicted based on the input data extracted from the traffic volume database by the traffic volume prediction unit are compared to obtain an evaluation value. Is provided as a data for learning to a traffic volume estimation section.

[0011]

According to the first aspect of the present invention, the traffic volume predicting unit considers that there is a time correlation in the change of the traffic volume, and based on the difference equation showing the time series relationship of the traffic volume, the data preprocessing unit By predicting future traffic volume from data that has been discretized, pre-processed such as smoothing and delay, even if the traffic volume trend is different from the previous day, the traffic volume can be predicted with high accuracy. A traffic volume prediction device that can perform

Further, the predictive traffic volume evaluation unit in the second aspect of the present invention is generated by the teacher data conversion unit converting the teacher data extracted from the traffic volume database into a data format output by the traffic volume prediction unit. The teacher signal is compared with the prediction output predicted by the traffic volume prediction unit based on the input data retrieved from the traffic volume database, the obtained evaluation value is sent to the traffic volume prediction unit, and the neural network of the traffic volume prediction unit is compared. By learning the network, it is possible to automatically acquire the function of the traffic volume prediction unit from the past accumulated data.

[0013]

【Example】

Example 1. Embodiment 1 of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing an embodiment of a traffic volume prediction apparatus according to the present invention when applied to elevator group management control. In the figure, reference numeral 5 denotes an elevator group management device to which the traffic volume prediction device according to the present invention is applied to perform group management control of the operation of all elevators in a building. 6 _{1 to} 6 _p
Is an elevator control device that is connected to the elevator group management device 5 to control a single elevator, detects the state of the sensor and the operation panel, and displays the operation panel. 7 _{1 to} 7 _Q are also the elevator group The hall control device is connected to the management device 6 and controls input / output relating to the hall, such as detecting the state of a call button from the hall and displaying the status of the elevator. Note that these are composed of a plurality of computers and sensors.

Further, in the elevator group management device 5, reference numeral 8 controls the elevator control devices 6 _{1 to} 6 _p and the hall control devices 7 _{1 to} 7 _Q in accordance with the control parameters, and at the same time, manages the operation of fetching actual traffic data. It is a department. Reference numeral 9 is a control rule table in which control rules composed of a plurality of control parameters are tuned in advance and stored, and reference numeral 10 is a search for control rules stored in the control rule table 9 according to predicted traffic data. The control parameter setting changing unit changes the control parameter of the operation management unit 8. Reference numeral 11 is a data pre-processing unit that discretizes the traffic volume actual data received from the operation management unit 8 and executes pre-processing such as smoothing and delay. 12 is data pre-processed by the data pre-processing unit 11. A traffic volume prediction unit for obtaining a prediction output, 13 is the traffic volume prediction unit 1
It is a predicted data conversion unit that converts the predicted output obtained in 2 into predicted traffic data. Reference numeral 14 is a traffic volume database that stores data obtained from the operation management unit 8. Reference numeral 15 is a teacher signal obtained by converting the teacher data extracted from the traffic volume database 14 into a data format output by the traffic volume prediction unit 12. It is a teacher data conversion unit to be generated. Reference numeral 16 is a comparison between the prediction output predicted by the traffic volume prediction unit 12 based on the input data extracted from the traffic volume database 14 and the teacher signal from the teacher data conversion unit 15, and the obtained evaluation value is data for learning. It is a predicted traffic volume evaluation unit that is sent to the traffic volume prediction unit 12.

Next, the operation will be described. First, prior to the description of the specific operation of the first embodiment, the basic concept of traffic volume prediction in the present invention will be described. In the present invention, attention is paid to the fact that the traffic volume data has a certain degree of correlation with respect to time, and the discretized traffic volume x
It is considered that (n) is represented by the relationship of the difference equation given by the following equation (1).

[0016]

[Equation 1]

Here, n in the above equation (1) is an index indicating the sampling time, and N is the order of the time delay related to changes in traffic usage. When the sampling period is T, the time t is given by the following equation (2).

T = nT (2)

Based on such an idea, when the equation (1) is obtained, x (n + 1) is obtained as an output when the inputs are x (n), ..., X (n−N). Traffic x
(N) can be predicted. Furthermore, when such a prediction is made, even if the state of change in traffic volume changes from the previous day, the prediction is possible as long as the relationship of equation (1) holds, which is a great advantage in the present invention. .

The above equation (1) is a difference equation, which is a kind of non-linear algebraic equation. However,
Generally, it is difficult to determine this equation (1). By the way, it is widely known that a multilayer neural network has an ability to acquire a non-linear mapping between input and output by learning if input and output data is given. Now, since the input and the output can be easily obtained in the equation (1), it is possible to express the equation (1) by a neural network as shown in FIG. Therefore, if equation (1) is acquired by learning using a neural network and is used as a prediction device, as shown in FIG. 3, the tendency of change in traffic volume up to that day is irrespective of time, as shown in FIG. It is possible to predict future traffic volume in a way that takes it into consideration.

Next, the specific operation of the embodiment shown in FIG. 1 will be described. Here, the function of this traffic volume prediction apparatus can be considered as being divided into two: a prediction apparatus learning mode and a traffic volume prediction mode. The predictor learning mode is executed when the elevator is not servicing, such as at midnight. on the other hand,
The traffic volume prediction mode is executed regularly or irregularly while the elevator is in service.

In the following, the procedure of the traffic volume prediction mode will be described first. This traffic volume prediction mode becomes possible after the traffic volume prediction apparatus has sufficiently learned by the prediction apparatus learning mode. The traffic volume prediction mode is executed by each unit from the operation management unit 8 to the prediction data conversion unit 13 surrounded by a broken line in FIG. Here, the data flow in this traffic volume prediction mode is shown in FIG. 4, and the flowchart of the processing procedure in the traffic volume prediction mode is shown in FIG. Note that the following description is for explaining the flowchart of FIG. 5 with reference to FIG.

Since the traffic volume data has a lot of random fluctuations, it is necessary to consider the high frequency components by using a low pass filter or the like. Therefore, first, the actual traffic data from the operation management unit 8 is input to the data preprocessing unit 11 (step ST1). The data preprocessing unit 11 discretizes the traffic volume actual data, performs processing such as smoothing by a low-pass filter and delay, and converts the data into a data format that can be input to the traffic volume prediction unit 12 (step ST2). The converted data x (n) may be, for example, the number of people who have gone up and got off the elevator in 5 minutes on each floor.

Then, at the time t (t = nT), the converted data x (n),
, X (n−N) is input to the traffic volume prediction unit 12, and the prediction output G (n + 1) is calculated (step ST
3). Predicted output G calculated by this traffic volume prediction unit 12
(N + 1) is input to the predicted data conversion unit 13 and converted into predicted traffic volume data x (n + 1) (step ST
4). Here, the reason why the predicted output from the traffic volume prediction unit 12 is converted into the predicted traffic volume data by the predicted data conversion unit 13 is as follows. That is, when the traffic volume predictor 12 is configured by a neural network in which the neuron activation function is a sigmoid function in the output layer, the output has a dynamic range of (1,0). Therefore, the predicted output normalized to (1,0) of the neural network is calculated as the predicted traffic data x (n +
It needs to be converted to 1).

The control parameter setting change unit 10 searches for the control rule stored in the control rule table 9 according to the predicted traffic volume data converted by the predicted data conversion unit 13 (step ST5). Here, this control rule is composed of a plurality of control parameters, and is optimally tuned in advance. The control parameter setting change unit 10 changes the control parameter of the operation management unit 8 according to the search result of the control rule table 9 (step ST6). With the above, the processing of the traffic volume prediction mode is completed and the elevator control is restarted.

Next, the procedure of the prediction device learning mode will be described. This prediction device learning mode needs to be sufficiently performed prior to the traffic volume prediction mode, and is executed by each unit from the data preprocessing unit 11 to the predicted traffic volume evaluation unit 16 surrounded by the one-dot chain line in FIG. Here, the data flow in the prediction device learning mode is shown in FIG. 6, and the flowchart of the processing procedure in the prediction device learning mode is shown in FIG.
Shown in. The following description is for the flowchart of FIG. 7 with reference to FIG.

Here, the data required every other week will be described first. The learning data is composed of input data and a corresponding teacher signal which is a desired output. In this embodiment, the learning data is the input data x (n-1), ...,
x (n-N-1) and the corresponding output, that is, the teacher signal R (n). Therefore, this learning data U _n is given in the form of the following equation (3).

[0028]

[Equation 2]

First, an arbitrary traffic volume data is selected from the traffic volume database 14 in which the traffic volume data obtained from the operation management unit 8 is stored in advance and used as the data for learning (step ST11). here,
The traffic volume database 14 is obtained by monitoring actual traffic volume data in the past and accumulating it. The data stored in this traffic volume database 14 is
The data may be pre-processed by the data pre-processing unit 11 such as discretization, smoothing, delay, etc., or may be unprocessed data. If the data is unprocessed,
The preprocessing may be performed in the data preprocessing unit 11 when the prediction device learning mode is executed. In the following description, it is assumed that processed traffic data is stored in the traffic database 14.

Then, the processed traffic data x (n-1), ..., X retrieved from the traffic database 14
(N-N-1) is the traffic volume prediction unit 12 as input data.
The traffic volume prediction unit 12 calculates a prediction output G (n) from the input data (step ST12). On the other hand, the traffic data x (n) extracted from the traffic database 14 is used as teacher data by the teacher data conversion unit 15
Is input to and converted into a data format output from the traffic volume prediction unit 12, and a teacher signal R (n) that is desirable output data for the previous input data is generated (step ST13). The conversion in step ST13 is an inverse conversion of the conversion in step ST4 of FIG.

Predicted output G from the traffic volume prediction unit 12
(N) and the teacher signal R (n) from the teacher data converter 15.
Are input to the predicted traffic volume evaluation unit 16 for comparison and evaluation (step ST14). Next, the learning of the traffic volume prediction unit 12 based on the evaluation result by the predicted traffic volume evaluation unit 16 is performed (step ST15). Here, for example, when a multilayer neural network is selected as the traffic volume prediction unit 12, the learning in step ST15 can be performed by using a back propagation algorithm. In this case, the estimated value E (n) of the predicted traffic volume is the predicted output G (n) and the teacher signal R as shown in the following equation (4).
It is given as the error of (n).

F (n) = R (n) -G (n) (4)

When the learning for one input data is completed, it is judged whether or not the learning for all the input data is completed (step ST16), and if not completed, the process returns to step ST11 and is separated from the traffic volume database 14. Input data and teacher data are extracted, and the learning process similar to the above is repeated. Further, when learning of all data on the traffic volume database 14 is completed, it is determined whether or not the learning process is ended (step ST17). This is determined by whether or not the evaluation in step ST14 has reached a predetermined standard, and steps ST11 to ST1 are performed until the evaluation reaches the standard.
The processes up to 6 are repeated. After sufficient learning is performed, the traffic volume prediction unit 12 comes to represent the mapping relationship of the equation (1).

Next, the control rules and control parameters will be briefly described using specific examples. As mentioned above, the control rule consists of multiple control parameters,
They are tuned in advance to the optimum value for each traffic volume by simulation or the like.
The control parameters include, for example, the number of vehicles dispatched to the lobby floor, the number of doors waiting to open the door, and the permanent call to the lobby floor, taking those used during work hours as an example. These control parameters will be described below.

In the intra-building transportation, the movement from the floor with the lobby to the other floor is significantly increased during the work hours. Therefore, a vehicle allocation method different from other time zones is required.
The above-mentioned three control parameters are control parameters governing the vehicle allocation peculiar to this work time zone. Normally, one elevator is assigned to a call on one floor, but a large number of vehicles are dispatched to the lobby floor during work hours so that waiting on a large number of lobby floors is possible. Improve customer service. The number of waiting doors is the number of elevators waiting on the lobby floor with the door open. This also has the same effect as the number of vehicles dispatched to the lobby floor. Further, a permanent call always considers that a call is being made from the lobby floor, even if the call button is not pressed on the lobby floor. While each of the above control parameters improves the service to the lobby floor, it may reduce the service to other floors. Therefore, the traffic volume predicting unit 12 predicts the traffic volume, and the control parameter setting changing unit 10 switches the control parameters in a timely manner based on the prediction result, thereby improving the service of the entire building. .

Example 2. In the first embodiment described above, the case where the traffic volume prediction apparatus according to the present invention is applied to the group management control of elevators can be applied to, for example, signal control of intersections in road traffic. In that case, as shown in FIG. 8, the entrances and exits of the intersections 1 to 3 are defined as points 1 to 12, and the number of inflowing vehicles and the number of outflowing vehicles to each of the points 1 to 12 are considered as the traffic volume at that time. The control parameters include the traffic light cycle (time for one cycle from blue to yellow to red), split (the ratio of blinking blue), offset (deviation of the timing of signal change between a series of intersections 1 to 3). ) Etc. In this case, since the number of inflowing vehicles and the number of outflowing vehicles at each of the points 1 to 12 can be measured, the traffic volume prediction device of the present invention can be applied to those predictions. Further, by changing the control parameter according to the prediction result, more efficient signal control becomes possible.

Example 3. The present invention is also applicable to railway control. In this railway control, the number of visitors and the number of participants at each station are considered as the traffic volume, and the stop time and travel time adjustment are considered as the control parameters. In the railway, the stop time is usually determined by the schedule, but during the rush hours in the morning and evening, the stop time and the running time are adjusted in order to facilitate the operation of the train group. Now, if the number of visitors and the number of participants are measured, the traffic volume prediction apparatus of the present invention can be applied to those predictions. Further, by changing the stop time and the operating time which are the control parameters according to the prediction result, it is possible to achieve a remarkable effect on smooth running of the train group.

[0038]

As described above, according to the invention described in claim 1, it is considered that the change in traffic volume has a time correlation, and the difference indicating the time series relationship of the traffic volume from the preprocessed data. Since it is configured to predict future traffic volume by using equations, it is possible to accurately predict traffic volume even when the traffic volume trend changes, and the traffic volume prediction device is robust. There is an effect that can improve the sex.

According to the second aspect of the present invention, the traffic volume predicting unit is formed by a neural network, and the neural network is learned by the evaluation value based on the comparison between the teacher signal and the prediction output of the traffic volume predicting unit. With this configuration, the function of the traffic volume prediction unit can be automatically obtained from the past accumulated data, and the traffic volume prediction apparatus can be easily realized.

[Brief description of drawings]

FIG. 1 is a block diagram showing a traffic volume prediction device according to a first embodiment of the present invention.

FIG. 2 is an explanatory diagram showing the concept of traffic volume prediction in the present invention.

FIG. 3 is a characteristic diagram showing characteristics of the traffic volume prediction.

FIG. 4 is an explanatory diagram showing a data flow in a traffic volume prediction mode in the above embodiment.

FIG. 5 is a flowchart showing a processing procedure in the traffic volume prediction mode.

FIG. 6 is an explanatory diagram showing a data flow in a prediction device learning mode in the above embodiment.

FIG. 7 is a flowchart showing a processing procedure in the prediction device learning mode.

FIG. 8 is an explanatory view showing a model of road traffic to which the second embodiment of the present invention is applied.

FIG. 9 is an explanatory diagram showing a conventional traffic volume prediction device.

FIG. 10 is a characteristic diagram showing characteristics of conventional traffic volume prediction.

[Explanation of symbols]

 11 Data Pre-Processing Section 12 Traffic Volume Prediction Section 14 Traffic Volume Database 15 Teacher Data Conversion Section 16 Predicted Traffic Volume Evaluation Section

Claims (2)

[Claims] 1. A time-series relationship of traffic volume based on a data pre-processing unit that discretizes actual traffic data and executes pre-processing such as smoothing and delay, and data pre-processed by the data pre-processing unit. A traffic volume prediction device including a traffic volume prediction unit that predicts a traffic volume using a difference equation indicating 2. A traffic volume database in which the traffic volume prediction unit is formed by a neural network, and the traffic volume database storing the traffic volume data, and teacher data extracted from the traffic volume database, in a data format output from the traffic volume prediction unit. A teacher data conversion unit that generates a teacher signal by converting to a teacher output from the teacher data conversion unit and a prediction output predicted by the traffic amount prediction unit based on input data extracted from the traffic amount database. The traffic volume prediction apparatus according to claim 1, further comprising: a predicted traffic volume evaluation section that sends the obtained evaluation value as data for learning to the traffic volume prediction section.