JPH11328570A - Method and device for predicting traffic condition and recording medium storing traffic condition predicting program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To predict traffic conditions for optional conditions exactly and in a short time. SOLUTION: Inputted data to be used for prediction is collected and held as time sequential data (S 101). N sub-sets of the inputted data are generated at random (S 102). N sub-sets are inputted in a neutral network (S 103), a predicted value of the number of the sub-sets is outputted (S 104) and the sub- sets are evaluated by comparing the predicted value with an actual value (S 105). When the optimal sub-set exists, the predicted value is calculated by using the sub-set as the inputted data (S 106, S108). When the optimal sub-set does not exist, the sub-set in the next generation is generated by hereditary operation, is inputted in the neutral network again and a processing is repeated (S 106, S 107).

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a traffic condition prediction method and apparatus for predicting a future traffic condition based on a current traffic condition.

[0002]

2. Description of the Related Art Traffic condition prediction methods include (1) a method of accumulating past data, (2) a method of extrapolating time-series data, and (3) a method using correlation between points. and so on.

[0003] The method of (1) accumulating past data is to accumulate data for a specific point over the past several years,
This method analyzes the effects of the month, day of the week, time, and the like based on this data. This method is effective for factors that do not change over time or factors that change over time among the factors that determine traffic conditions, but are highly variable over time, such as sudden accidents and regulations. The factor cannot be reflected.

[0004] The method of (2) extrapolating the time series data is based on the time series data several steps before the present time from the present time.
This is a method in which a curve for extrapolating this time-series data is determined by a method such as an autoregressive model or a neural network, and the future traffic volume is predicted using this curve. This method can reflect factors that fluctuate over time, but cannot reflect factors that do not change over time or that fluctuate over time.

The method using the correlation between the points (3) is as follows.
This is a method of making prediction using a correlation between the current traffic condition of the road to be predicted and the traffic condition of the surrounding road one step before. Unlike this method (1) and (2), this method can reflect spatial factors such as the influence of surrounding roads, but it is difficult to reflect long-term varying time factors.

[0006]

Generally, traffic conditions are
Factors that change over time, such as 24-hour fluctuations due to commuting rushes, fluctuations due to the day of the week, regional (spatial) factors such as highways and general roads, cities and local roads, and accidents and regulations. Factors, the influence of surrounding roads, and the factors that change the traffic situation by providing traffic information services.

However, as described above, in the conventional method, there is no prediction method that considers all the factors, and the factors are limited in advance. In the method of (1) in which past data is accumulated, attention is paid only to a variation factor that changes with a constant period in time. The method of (2) extrapolating the time-series data focuses only on sudden factors and factors that fluctuate in a short period of time. In the method using the correlation between points (3), only spatial factors are narrowed down.

However, which factor is important depends on the situation at the time of prediction, such as time and place, and it is difficult to determine in advance. For example, if there is no sudden event and the road is very general, it is important to capture changes in a certain cycle by accumulating past data. Once an accident or other event occurs, the impact of the accident, including the surrounding roads, must be considered. In other words, limiting the factors in advance can be predicted well in certain special situations, but has a problem in that the prediction accuracy deteriorates in other situations.

On the other hand, if the amount of input data is simply increased, it is necessary to consider unimportant factors, which causes a problem that the prediction speed is reduced.

An object of the present invention is to provide a traffic condition prediction method and apparatus for predicting a traffic condition in an arbitrary condition with an accurate and short processing time.

[0011]

According to the present invention, there is provided a traffic condition estimating method for collecting various input data which is considered to be a factor determining a traffic condition, and determining a method of selecting input data using the input data. And selecting input data based on the selected determination method, and calculating a predicted value using the selected input data.

Further, the traffic condition predicting device of the present invention inputs various data considered to be factors determining the traffic condition,
Input data collecting / holding means to be held, input data selecting method determining means for determining a method of selecting input data using past input data held in the input data collecting / holding means, and the determined input data Input data selecting means for selecting data from the current input data input to the input data collecting / holding means based on the selection method, and a predicted value calculating means for calculating a predicted value using the selected data Having.

By selecting input data before calculating a predicted value, it is possible to calculate a predicted value in consideration of only necessary input data, thereby improving the speed of calculating a predicted value. Also,
Although the method of determining the selection method in advance could deal only with a specific situation, by determining the input data selection method according to the situation, accurate prediction can be performed for any situation .

[0014]

Next, embodiments of the present invention will be described with reference to the drawings.

FIG. 1 is a flowchart of a traffic condition prediction method according to a first embodiment of the present invention using a genetic algorithm (GA) as a data selection method, and FIG. 2 is a schematic diagram thereof. This embodiment is an example of predicting the average speed of a road that is S minutes ahead of the present.

First, input data used for prediction is collected (step 101). For example, the current measurement data 21 such as the traffic volume, average speed, and vehicle density of the road to be predicted 21
1, past data and its statistics 212, similar data 213 of surrounding roads, data 21 such as day of the week, time, etc.
4, data 215 relating to accidents and regulations, data 216 relating to the current weather, and the like. These data are collected, for example, every 5 minutes, regardless of whether they fluctuate over time.

Assuming that S minutes ahead is predicted, these data are held as time series data for at least S minutes.

In order to determine which type of data should be selected from among these many input data, the data before S minutes (predicted to be the current value, Data to predict)
N subsets (subsets) are created at random (step 102). The subsets 221, 222 and the like represent not only a simple subset of data but also a method of selecting input data such as what kind of data should be selected from input data. In the subset group, for example, a subset that selects a large amount of past traffic data, a subset that selects a large amount of current traffic data, current weather data, and the like are mixed. Hereinafter, this subset is treated as an individual in the genetic algorithm. The subset group of the N input data (221,
, 223) are input to the prediction engines 231, 232,... 233 as input data (step 103). As an example, a neural network is used here.

By inputting the subsets 221 to 223 to the prediction engines 231 to 233, the subsets 221 to 223 are input.
, 243 for each 223 (step 104).

Next, as shown at 251, the prediction values 241 to 243 of each subset are compared with the current actual measurement value, and the prediction accuracy is set so that the closer the predicted value and the actual measurement value, the better the score. Subsets 221-2 according to
23 is given a score (step 105). As an example of a point calculation method, (the number of points in the subset) = {(current measured value) − (predicted value when a subset is input)} ²
and so on. If a situation such as an accident such as an accident is currently occurring on a peripheral road, a subset including the current traffic volume, the degree of congestion, and the like on the peripheral road obtains a good score.

Next, it is determined whether or not the best score given by each subset satisfies a predetermined evaluation criterion (step 106). In the above example, whether the score of the subset is equal to or greater than the threshold value is the evaluation criterion.

If the evaluation criteria are not satisfied, 26
As shown in FIG. 1, genetic operations such as selection, mutation, and crossover are performed on all subsets 221 to 223 to generate a next-generation subset group (step 107). As shown in FIG. 3, for example, “abcde”,
For two subsets Ga and Gb having five input data “12345”, exchanging a certain part (“ab” and “12”) to create two new subsets is called crossover. In addition, as shown in FIG. 4, for example, randomly changing a part b of the subset to different input data B is referred to as mutation. The selection means to increase the number of subsets having a high score and reduce the number of subsets having a low score in the subset set 220. For example, there is a method (roulette strategy) in which each subset is selected as a next-generation subset group with a probability proportional to the score.

After step 107, step 103 is repeated.
And repeat this cycle until the best subset meets the criteria. With such a cycle, the most appropriate subset for the current situation, ie how to select the input data, is automatically obtained without prior knowledge. This cycle is sequentially executed every five minutes, and the method of selecting input data is sequentially updated.

If there is a subset that satisfies the evaluation criteria, it is assumed that, for example, the subset 221 is the subset that obtains the best score. In this case, in step 106, the predicted value 24 output from the prediction engine 231 is output.
1 satisfies the evaluation criterion, and as shown in 270, when selecting the actual data, the subset 22
A predicted value 274 after S minutes from the current time is calculated using the subset 272 of the current input data 271 selected in the same manner as 1 as input data of the actual predicted value calculation unit 273 (step 108).

In this embodiment, data is collected every 5 minutes and data is selected every 5 minutes. However, a method of collecting data every 2 minutes and selecting data every 10 minutes may be used. It is conceivable that the period of data collection and selection is not limited to this embodiment.

In this embodiment, N neural networks are prepared as prediction engines. However, there is a method of using one prediction engine or a method of using autoregressive analysis or the like as the prediction engine. ,
The number is not limited to this embodiment. However, the prediction engine 273 used in step 108 actually used and the prediction engine 231 used in subset evaluation (step 105)
To 233 are the same.

In the present embodiment, crossover, mutation, and selection are performed as genetic operations of the genetic algorithm. The crossover method, the mutation method, and the selection method include elite preservation strategies other than those described in the present embodiment. Furthermore, there are various methods in the genetic algorithm, such as a method of creating a subset and a method of calculating a score, and are not limited to the present embodiment.

Further, in the present embodiment, an example using a genetic algorithm as an input data selection method has been described. However, a method of using a neural network in multiple layers or calculating the weight of data from the influence on the predicted value is used to calculate the weight. A method using only high data may be considered, and the present invention is not limited to this embodiment.

Also, as input data, 211 to 216
Although the embodiment has been described with reference to the example, other data may be considered as long as the data is related to traffic condition prediction, and the type of input data is not limited to this embodiment.

Referring to FIG. 5, a traffic condition predicting apparatus according to an embodiment of the present invention includes an input data collecting / holding unit 301, an input data selecting method determining unit 302, and an input data selecting unit 30.
6 and a prediction value calculation unit 307, and the input data selection method determination unit 302 includes a subset holding unit 303, a subset evaluation unit 304, and a subset transformation unit 305.

The input data collecting / holding unit 301 collects and holds various data that are considered to be factors that determine traffic conditions.

The stored past input data is sent to the input data selection method determining unit 302. As an example of the input data selection method determination unit 302, FIG. 5 shows a configuration in the case of input data selection by a genetic algorithm. The subset holding unit 303 generates and holds some subsets (individuals) from past input data.
The subset evaluation unit 304 calculates a prediction value when the subset is used as input data, and evaluates the accuracy of prediction by comparing the prediction value of each subset with an actual measurement value input from the input data collection unit 301. Subset transformation unit 30
5 creates a next-generation subset based on this.

When a subset that outputs a predicted value satisfying the evaluation criterion is determined by the input data selection determining unit 302,
The selection method represented by this subset is the input data selection unit 30
Sent to 6. The current input data is sent from the input data collection / holding unit 301 to the input data selection unit 306, and a subset of data is extracted by the same selection method as the subset determined by the input data selection method determination unit 302, and the predicted value is calculated. The unit 307 outputs a predicted value based on the input data.

By adopting such a configuration, it is possible to automatically narrow down and predict from a very large amount of data for determining the traffic situation to a small number of data having the strongest influence according to the current situation. Becomes

Referring to FIG. 6, a traffic condition prediction device according to another embodiment of the present invention includes an input device 401 and a storage device 402.
, Output device 403, recording medium 404, and data processing device 4
05.

The input device 401 is an input device for inputting various data considered to be factors that determine the traffic situation.
The storage device 402 stores input data, a subset of the input data, and the like. The output device 403 is an output device that outputs the calculated predicted value for use in a predicted value providing service, traffic control, and the like. The recording medium 404 includes a subset evaluation unit 304, a subset transformation unit 305 in FIG.
A traffic situation prediction program including the processing of the input data selection unit 306 and the prediction value calculation unit 307 and the processing of writing and reading data to and from the storage device 402 is recorded.
It is a recording medium such as an FD, a CD-ROM, and a semiconductor memory. The data processing device 405 reads the traffic condition prediction program from the recording medium 404 and executes the program.
It is.

[0037]

As described above, according to the present invention,
Selection of input data according to a situation can be realized, thereby improving the accuracy of prediction in an arbitrary situation and reducing the time required for prediction.

[Brief description of the drawings]

FIG. 1 is a flowchart illustrating a traffic condition prediction method according to an embodiment of the present invention.

FIG. 2 is a schematic diagram of the traffic situation prediction method of FIG.

FIG. 3 is a diagram showing crossover as an example of a subset deformation method in the input data selection method.

FIG. 4 is a diagram illustrating a mutation as an example of a subset transformation method in the input data selection method.

FIG. 5 is a configuration diagram of a traffic situation prediction device according to an embodiment of the present invention.

FIG. 6 is a configuration diagram of a traffic situation prediction device according to another embodiment of the present invention.

[Explanation of symbols]

101 to 108 Step 200 Input data selection method determination method 211 to 216 Input data 220 Subset group of input data 221 to 223 Subset of input data 231 to 233 Prediction engine 241 to 243 Predicted value of each subset 221 to 223 251 Subset evaluation method 261 Subset deformation method 270 Method of actually calculating predicted value 271 Input data currently collected 272 Input data actually used for prediction 273 Predicted value calculating unit 274 Actual predicted value 301 Input data collecting / holding unit 302 Input data selection Method determination unit 303 Subset holding unit 304 Subset evaluation unit 305 Subset transformation unit 306 Input data selection unit 307 Predicted value calculation unit 401 Input device 402 Storage device 403 Output device 404 Recording medium 405 Data Processing equipment

 ──────────────────────────────────────────────────続 き Continuing on the front page (72) Fumio Adachi, Inventor 3--19-2 Nishi-Shinjuku, Shinjuku-ku, Tokyo Inside Nippon Telegraph and Telephone Corporation (72) Inventor Satoshi Suzuki 3-192-1, Nishi-Shinjuku, Shinjuku-ku, Tokyo No. Japan Telegraph and Telephone Corporation

Claims (12)

[Claims] 1. A traffic condition prediction method for predicting a future traffic condition based on a current traffic condition, comprising: collecting various input data considered to be a factor determining a traffic condition; Determining a selection method of input data using the input method, selecting input data based on the determined selection method, and calculating a predicted value using the determined input data. . 2. The method according to claim 1, wherein the step of determining the input data selection method and the step of selecting the input data generate a subset of the input data, calculate a predicted value based on each subset, and calculate the calculated predicted value. Comparing the measured values with actual measured values and selecting an optimal subset by evaluating the accuracy of the prediction, and the step of calculating the predicted value includes calculating a predicted value using the optimal subset. The method of claim 1. 3. If an optimal subset is not found, a genetic operation is performed on all subsets to generate a next-generation subset, a predicted value is calculated based on these subsets, and the predicted value is measured by an actual measurement. 3. The method of claim 2, wherein comparing is repeated until an optimal subset is found. 4. The method according to claim 2, wherein a predicted value for each subset is obtained by inputting the generated subset to a neural network. 5. A traffic condition prediction device for predicting a future traffic condition based on a current traffic condition, wherein input data collection / retention for inputting and retaining various data considered to be a factor determining the traffic condition is provided. Means, an input data selection method determining means for determining a selection method of input data using past input data held in the input data collection and holding means, based on the determined input data selection method, Traffic condition prediction having input data selection means for selecting data from current input data input to the input data collection / holding means, and prediction value calculation means for calculating a prediction value using the selected data apparatus. 6. A subset holding means for generating and holding a subset of input data from past input data held in said input data collection / holding means, said input data selection method determining means; and inputting said subset. A subset evaluation means for calculating a prediction value in the case of data, comparing the prediction value with an actual measurement value input from the input data collection / holding means to evaluate prediction accuracy, and selecting an optimal subset. The apparatus according to claim 5, wherein the predicted value calculating means calculates a predicted value using the optimal subset. 7. If the optimum subset is not found, the method further includes a subset transforming unit that performs a genetic operation on all subsets to generate a next-generation subset, and outputs the next-generation subset to the subset evaluating unit. The apparatus according to claim 6, wherein the subset evaluation unit performs processing from calculation of a predicted value to selection of an optimal subset for the next-generation subset. 8. The apparatus according to claim 6, wherein the subset evaluation unit obtains a predicted value for each subset by inputting the generated subset to a neural network. 9. An input data collection process for inputting various data which is considered to be a factor for determining a traffic condition and storing the input data in a storage device, and inputting input data using past input data stored in the storage device. Input data selection method determination processing for determining a selection method, based on the determined input data selection method, input data selection processing for selecting data from current input data input in the input data collection processing, A medium in which a traffic condition prediction program for causing a computer to execute a predicted value calculation process of calculating a predicted value using the selected data is recorded. 10. The input data selection method determination processing includes:
Generating a subset of the input data from past input data held in the storage device, performing a subset holding process to hold the storage device, and calculating a predicted value when the subset is used as the input data; And a subset evaluation process of evaluating an accuracy of prediction by comparing the predicted value with an actually measured value input in the input data collection process, and selecting an optimal subset. The recording medium according to claim 9, wherein is calculated. 11. When the optimal subset is not found, the method further includes a subset transformation process for generating a next-generation subset by performing a genetic operation on all subsets and outputting the subset to the subset evaluation process. The recording medium according to claim 10, wherein the subset evaluation process performs a process from calculating a predicted value to selecting an optimal subset for the next-generation subset. 12. The subset evaluation processing according to claim 10, wherein a predicted value for each subset is obtained by inputting the generated subset to a neural network.
2. The recording medium according to 1.