JP2019125260A - Data creation device, induction model learning device, induction estimation device, data creation method, induction model learning method, induction estimation method and program - Google Patents

Abstract

To estimate effective induction using road information.SOLUTION: An induction estimation device for, when at least one of road structure indicating the structure of a road and a road state for each road structure is input, estimating information on induction by a model outputting information on the induction comprises induction estimation means for estimating information on the induction by inputting at least one of the road structure and the road state to the model for which a parameter estimated by parameter estimation means in an induction model learning device is set.SELECTED DRAWING: Figure 1

Description

  The present invention relates to a data creation device, a guidance model learning device, a guidance estimation device, a data creation method, a guidance model learning method, a guidance estimation method, and a program.

  By providing appropriate guidance to people passing by the road, it is possible to eliminate congestion, improve safety, and reduce travel time.

  The development of sensor technology has made it possible to easily collect road information such as traffic volume. For this reason, methods have been proposed for obtaining appropriate guidance using the collected road information. For example, there has been proposed a method of estimating an appropriate guidance by estimating a parameter of a simulator from collected road information and performing trial and error with a simulator using the estimated parameter (see, for example, Non-Patent Document 1) .

2015. Real-time and proactive navigation via spatio-temporal prediction. In Adjunct Proceedings of the 2015 ACM International Joint Conference ACM, New York, NY, USA, 1559-1566. Internet <https://doi.org/10.145/2800835.2801624> of the 2015 ACM International Symposium on Wearable Computers (UbiComp / ISWC'15 Adjunct).

  However, estimation of simulator parameters usually requires a large amount of road information. In addition, since it is necessary to perform trial and error by a simulator, it may take a lot of time to obtain an appropriate induction.

  The present invention has been made in view of the above points, and has an object to estimate effective guidance using road information.

  Therefore, in the data creation device according to one embodiment of the present invention, when the road structure representing the structure of the road and the road condition for each road structure are acquired or generated, guidance is performed for each road structure. Road effect data including a guided road structure representing the structure of the road, a guidance effect representing an effect when the guidance is performed for the case where the guidance is not performed, and at least one of the road structure and the road condition Means for creating data.

  In addition, the guidance model learning device according to the embodiment of the present invention is a parameter of a model that outputs information related to guidance when at least one of a road structure representing a road structure and a road condition for each road structure is input. A guidance model learning device for estimating the road effect data, which is output by inputting road effect data created by data creation means in the data creation device as learning data to the model, and road effect data It is characterized by the parameter estimation means which estimates the said parameter so that the difference | error with the guidance effect contained may be minimized.

  Further, in the guidance estimation apparatus according to the embodiment of the present invention, the guidance is performed using a model that outputs information related to guidance when at least one of a road structure representing a road structure and a road condition for each road structure is input. A guide estimation device for estimating information on the road, and at least one of the road structure and the road condition is input to the model in which the parameter estimated by the parameter estimation means in the guide model learning device is set. , And a guidance estimation means for estimating information on the guidance.

  Road information can be used to estimate effective guidance.

It is a figure showing an example of functional composition of a guidance presumption device in an embodiment of the invention. It is a flowchart which shows an example of the process which the guidance estimation apparatus in embodiment of this invention performs. It is a figure which shows an example of the evaluation result of this invention. It is a figure which shows an example of the hardware constitutions of the induction | guidance | derivation estimation apparatus in embodiment of this invention.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings. In the embodiment of the present invention, a guidance estimation device 10 for estimating a guidance from road information will be described. The mobile object to be guided is not limited to the person (pedestrian etc.) moving on the road, and may be any mobile object moving on any graph such as a car or car moving on the road, for example. good. The graph is not limited to the road, and may be, for example, an indoor passage or the like.

<Functional Configuration of Guidance Estimation Device 10>
First, the functional configuration of the guidance estimation device 10 according to the embodiment of the present invention will be described with reference to FIG. FIG. 1 is a diagram showing an example of a functional configuration of a guidance estimation device 10 according to an embodiment of the present invention.

  As shown in FIG. 1, the guidance estimation device 10 according to the embodiment of the present invention includes a data creation unit 110, a parameter estimation unit 120, and a guidance estimation unit 130. Each of these functional units is realized by processing that one or more programs installed in the guidance estimation device 10 cause a CPU (Central Processing Unit) or the like to execute. Each of these functional units may be realized by processing executed by one program, or may be realized by processing executed by different programs.

  Further, the guidance estimation device 10 according to the embodiment of the present invention has a storage unit 140. The storage unit 140 can be realized, for example, using an auxiliary storage device such as an HDD (Hard Disk Drive) or an SSD (Solid State Drive). Note that the storage unit 140 may be realized using a storage device or the like connected to the guidance estimation device 10 via a network.

  The data creation unit 110 creates road information and data on the guidance effect at that time (this data is referred to as “road effect data”) by simulation. The data creation unit 110 includes a road structure creation unit 111, a road condition creation unit 112, a guidance road structure creation unit 113, a guidance effect measurement unit 114, and a road effect data creation unit 115. In addition, arbitrary traffic simulators can be used as a simulator used for simulation.

The road structure creation unit 111 is a set of N road structures G _n

を作成する。以降では、道路構造Ｇ_ｎの集合を「道路構造集合」と表す。

Create Hereinafter, a set of road structures G _n is referred to as a “road structure set”.

The road structure G _n is, for example, information representing the road structure such as a connection between roads, a width, and a slope. The road structure G _n may be information obtained by observing a road in the real world with a sensor or the like, or may be information generated by a simulator or the like. The road structure G _n can be expressed, for example, by a graph, where nodes represent roads and edges represent connections between roads.

The road condition creation unit 112 generates M _n simulation parameters for each road structure G _n . Then, the road condition creation unit 112 performs simulation using these parameters to obtain a set of _Mn road conditions X _nm .

を作成する。以降では、Ｍ_ｎ個の道路状況Ｘ_ｎｍの集合を「道路状況集合」と表す。

Create Hereinafter, a set of _Mn road conditions X _nm will be referred to as a “road condition set”.

The road condition X _nm represents the information of the moving body such as the number of people included in each road in a certain time zone, the number of people passing in a certain place, and the average moving speed of each road. For example, as the road condition X _nm , the number of people included in each road at each time

が利用できる。ここで、ｘ_ｎｍｉｔはｎ番目の道路構造Ｇ_ｎにおけるｍ番目の道路状況Ｘ_ｎｍで時刻ｔにｉ番目の道路に含まれる人数、Ｉ_ｎはｎ番目の道路構造Ｇ_ｎの道路数、Ｔは観測の時刻数を表す。なお、時刻ｔは、シミュレーションの或る時刻から或る時刻までを或る時間幅に分割した場合における各時間幅を表すインデックスである。また、道路状況Ｘ_ｎｍとして全時刻の状況ではなく、観測の最終時刻から或る時間幅の時刻の状況を用いても良い。

Is available. _{Here, x NMIT} the number of people included in the m-th road conditions _{X nm} i-th road at time t in the n-th road structure _{G n, I n} the road number of the n-th road structure _{G n,} T is Represents the number of times of observation. Time t is an index representing each time width in the case where a certain time width is divided from a certain time of simulation to a certain time. Also, as the road condition X _nm , not the situation of all the time but the situation of a certain time width from the last time of observation may be used.

The guidance road structure creation unit 113 is a set of road structures G _nk when K _n pieces of guidance are performed in each road structure G _n.

を作成する。

Create

For example, when each road is closed as a guidance measure, G _nk is a road structure when the k-th road is closed, that is, a node from G _n to the k-th node (road) and an edge corresponding to the node It is expressed by the deleted graph.

  As guidance measures, for example, it is possible to use stoppages, change of road width, restriction of the number of people, guidance to the next road, one way, lanes with different speeds, increase / decrease speed, addition of roads, etc. it can. That is, the guidance measure is not necessarily limited to the guidance, and any measure that can be expressed as a road structure can be used.

The guidance effect measurement unit 114 takes the road condition x _nmT at the final time (final time of observation) created by simulating using the road structure G _n when not guiding as an initial value, and the same parameters as the simulation when not guiding The induction effect s _nmk is measured and created by _performing simulation using the road structure G _nk for each of the induced cases and

In addition, the parameter of the simulation using road structure _Gn when not guiding, and road condition _xnmT use what was produced _| generated and created by the road condition creation part 112. FIG. Also, for example, when the number of people included in each road at each time is used as the road status X _nm , the road status x _{nmT at the} final time is

である。

It is.

  As an induction effect, it is possible to use any index that you want to increase by induction, such as shortening the average time to reach the destination, shortening the maximum time to reach the destination, and the number of people arriving at the destination. it can. In addition, although any index to reduce by induction, such as congestion degree and the number of crowded places, can also be used as an induction effect, in this case, the value of the index is taken negative (that is, −1 The multiplied value shall be adopted.

The road effect data creation unit 115 is road effect data in which each road structure G _n , each road condition X _nm , and each guidance effect s _nmk are summarized.

を作成する。ここで、

Create here,

はｎ番目の道路構造Ｇ_ｎにおけるｍ番目の道路状況Ｘ_ｎｍでの誘導効果ｓ_ｎｍｋをまとめたベクトルである。道路状況Ｘ_ｎｍとして、誘導をする直前の時刻のみの道路状況を用いても良いし、誘導をする直前の時刻から或る一定時間の道路状況を用いても良い。

Is a vector summarizing the induction effect s _nmk at the m-th road condition X _nm in the n-th road structure G _n . As the road condition X _nm , the road condition of the time immediately before the guidance may be used, or the road condition of a certain fixed time from the time immediately before the guidance may be used.

Hereinafter, information including the road structure G _n and the road condition X _nm is also referred to as “road information”. Therefore, it can be said that road guidance data is data in which road information and each guidance effect s _nmk are put together.

However, the road information may be information including one of the road structure G _{n and the} road condition X _nm .

  The road effect data created by the road effect data creation unit 115 is stored in the storage unit 140. This road effect data is learning data used for parameter estimation of the guidance model described later.

The data creation unit 110 does not necessarily have to create at least one of each road structure G _n and each road condition X _nm , and for example, data obtained by observing the real world with a sensor or the like is input from the outside. It is good. That is, the road structure creation unit 111 may acquire each road structure G _n input from the outside. Similarly, the road condition creation unit 112 may acquire the road condition X _nm input from the outside.

  The parameter estimation unit 120 estimates the parameters of the guidance model using the road effect data stored in the storage unit 140. The guidance model is a model that takes road information as an input and estimates and outputs guidance (or information corresponding to the guidance).

  The parameter estimation unit 120 estimates the parameters of the guidance model so that a more appropriate guidance (more effective guidance) can be estimated when road information is given. A graph convolution neural network can be used as a guidance model. However, the present invention is not limited to this, and any model that can estimate and output a guidance when road information is given, for example, uses any model such as a neural network, a support vector machine, or a Gaussian process as a guidance model be able to.

  In the case of a graph convolution neural network, for example, a neural net layer as shown in the following Equation 1 can be used.

ここで、

here,

はｌ層目のノードｉにおける隠れ層のベクトルであり、０層目の隠れ層は道路状況、すなわち、

Is the vector of the hidden layer at node i in layer 1, and the hidden layer in layer 0 is the road condition, ie

であるとする。

It is assumed that

  In this case the linear transformation matrix

が推定すべきパラメータである。

Is the parameter to be estimated.

  Assuming that guidance for closing roads is assumed, the index i indicating each road corresponds to the index k indicating each guidance, so the size of the final hidden layer (that is, the hidden layer of the L-th layer) is 1 and Last hidden layer value

を出力とすることができる。なお、簡単のため、最終隠れ層の値を「ｚ_ｎｍｉ」とも表すものとする。

Can be output. Note that, for simplicity, the value of the final hidden layer is also expressed as "z _nmi ".

On the other hand, depending on the guidance measure, there is a possibility that the index i indicating each road and the index k indicating each guidance do not correspond. In this case, the nodes of the graph representing the road structure G _n can be dealt with by constructing the graph so as to correspond to the index indicating each guidance.

In many road effect data, if the output is correlated with the induction effect s _nmk , the value becomes high, otherwise the value becomes low as the objective function, and the value of the objective function is made as small as possible To estimate the parameters. For example, it is assumed that the inductive effect s _nmi is added by the softmax function and normalized to 1 (the sum is 1) to be y _nmi . In addition, since the index i which shows each road and the index k which shows each guidance respond | correspond, the index k is substituted by the index i.

Here, as the induction effect s _nmi (also, normalized y _nmi ) is higher, the induction effect is higher (ie, for example, the average time to reach the destination is shorter, to the destination) The number of arrivals of people is low, the degree of congestion is low, the number of places to be crowded is small, etc.). Then, as shown in Equation 2 below, it is possible to use, as an objective function, the cross entropy of y _nmi and the value of the final hidden layer z _nmi obtained by taking the softmax function in all road effect data.

なお、交差エントロピー以外にも、例えば、二乗誤差等の他の誤差を用いても良い。

Besides the cross entropy, for example, another error such as a square error may be used.

  The parameters estimated by the parameter estimation unit 120 are stored in the storage unit 140.

The guidance model is not only a model using two pieces of information of the road structure G _n and the road condition X _nm , but also a model using, for example, any one of the road structure G _{n and the} road condition X _nm Also good.

The guide estimation unit 130 estimates a guide using the guide model in which the parameters stored in the storage unit 140 are set and the road information. That is, for example, the guidance estimation unit 130 obtains the output z _nmi by inputting the road structure G _n and the road condition X _nm included in the road information into the guidance model in which the parameters are set. Of the output z _nmi, the most value corresponds to a high output z _nmi induced optimal induction (i.e., inducing its effect (induction effect) is expected to be high). Also, for example, in the case where it is desired to rank a plurality of leads, they may be rearranged in descending order of output value.

  Also, multiple inductions may be combined. For example, after obtaining the optimum guidance by the guidance model, the road structure is corrected according to the guidance, and using the corrected road structure, it is repeated to obtain the next guidance by the guidance model again. Thereby, a combination of a plurality of inductions can be obtained.

In the case of estimating the parameters of the induction model using one of the information of the road structure G _n or road conditions X _nm, either the road structure G _n or road conditions X _nm included in the road information, the It is sufficient to obtain the induction as an output by inputting the parameter into the induction model set.

  The guidance estimated by the guidance estimation unit 130 is output to a predetermined output destination. The predetermined output destination may be a display device such as a display, the storage unit 140 or an external recording medium, or any other device connected via a network. .

  Although the example illustrated in FIG. 1 describes the case where one device (the guidance estimation device 10) includes the data generation unit 110, the parameter estimation unit 120, and the guidance estimation unit 130, the present invention is not limited thereto. . The data creation unit 110, the parameter estimation unit 120, and the guidance estimation unit 130 may be dispersed in a plurality of devices. For example, even if the embodiment of the present invention is realized by, for example, a data creation device having data creation unit 110, a guidance model learning device having parameter estimation unit 120, and a guidance estimation device having guidance estimation unit 130. good.

<Process Performed by Guide Estimation Device 10>
Next, the flow of processing performed by the guidance estimation device 10 according to the embodiment of the present invention will be described with reference to FIG. FIG. 2 is a flowchart showing an example of processing performed by the guidance estimation device 10 according to the embodiment of this invention.

Step S101: The data creation unit 110 creates road effect data. That is, the data creating unit 110 creates or acquires each road structure G _n by the road structure creating unit 111, creates or acquires each road condition X _nm by the road condition creating unit 112, and each by the guided road structure creating unit 113. The road structure G _nk is created, and each induction effect s _nmk is created by the induction effect measurement unit 114. Then, the data creation unit 110 creates the road effect data by the road effect data creation unit 115, and stores the created road effect data in the storage unit 140.

  Step S102: Next, the parameter estimation unit 120 estimates the parameters of the guidance model using the road effect data stored in the storage unit 140. That is, the parameter estimation unit 120 estimates a parameter that minimizes an error between the output of the guidance model using the road information included in the road effect data and the guidance effect included in the road effect data. Then, the parameter estimation unit 120 stores the estimated parameter in the storage unit 140.

  Step S103: Next, the guidance estimation unit 130 estimates the guidance using the guidance model in which the parameters stored in the storage unit 140 are set and the road information.

That is, for example, the guidance estimation unit 130 obtains an output by inputting the road structure G _* and the road condition X _* included in the road information into the guidance model in which the parameters are set. Or, when estimating the parameters of the control model road structure G _* or road conditions X _* either with one, induction estimating unit 130, road structure G _* or road conditions X _* either included in the road information Output is obtained by inputting one or the other into the induction model in which the parameter is set. Then, for example, the lead estimation unit 130 sets the lead corresponding to the output of the highest output value as the optimum lead.

  As described above, the guidance estimation device 10 according to the embodiment of the present invention creates road information and data (road effect data) on the guidance effect at that time by simulation. And the guidance estimation apparatus 10 in embodiment of this invention estimates the parameter of a guidance model using this road effect data. Thereby, the guidance estimation device 10 according to the embodiment of the present invention can quickly estimate the optimal guidance (that is, effective guidance) by inputting the road information into the guidance model in which the parameter is set. become able to.

<Evaluation example of the present invention>
Here, evaluation examples of the present invention will be described. The following experiments were performed to evaluate the present invention. That is, first, 700 road structures were created based on the roads around the railway station in Japan. In each road structure, a total of 100,000 people were simulated for 30 minutes, with pedestrians going to the station from some random places in the vicinity and pedestrians going to some random places from the station. With the end time of this 30 minutes simulation as the start time, a simulation was additionally performed for 30 minutes when guiding each road to be closed.

  The induction effect is the average time to reach the destination multiplied by -1 in the simulation for all 1 hour. The parameters of the guidance model were estimated using, as learning data, road effect data including one road condition and one guidance effect in 700 road structures. As road conditions, the number of people included in each road at one time immediately before guidance was used.

  Then, road effect data of 70 road structures different from learning data are used as verification data, and road effect data of 70 road structures different from learning data and verification data are used as test data. As a guidance model, using a graph convolution neural network with three hidden layers and a size of each hidden layer of 10, the initial value was changed and learned five times, and the one with the lowest cross entropy among the verification data was adopted. On the other hand, as a comparison method with the present invention, "random" randomly selecting a road to be closed, "largest number of people" selecting the largest number of roads, and selected roads leading to the largest number of roads We used "before the maximum number of people" and "after maximum number of people" who choose the road connected from the road with the largest number of people.

  For each of the present invention and each comparative method, using test data, the difference between the average arrival time without induction and the average arrival time with induction is shown in FIG. In FIG. 3, a negative value indicates that the average arrival time is shorter in the case where the induction is performed than in the case where the induction is not performed.

  As shown in FIG. 3, it can be seen that when the induction estimated according to the present invention was performed, the arrival time per person was 21.9 seconds shorter than when the induction was not performed. On the other hand, as shown in FIG. 3, in the comparison method, when the induction is performed, the arrival time is longer than when the induction is not performed.

<Hardware Configuration of Guidance Estimation Device 10>
Finally, the hardware configuration of the guidance estimation device 10 according to the embodiment of the present invention will be described with reference to FIG. FIG. 4 is a diagram showing an example of a hardware configuration of the guidance estimation device 10 according to the embodiment of the present invention.

  The guidance estimation device 10 illustrated in FIG. 4 includes an input device 201, a display device 202, an external I / F 203, a random access memory (RAM) 204, a read only memory (ROM) 205, a CPU 206, and a communication I / O. An F 207 and an auxiliary storage device 208 are provided. Each of these pieces of hardware is communicably connected via the bus B.

  The input device 201 is, for example, a keyboard, a mouse, or a touch panel, and is used by the user to input various operations. The display device 202 is, for example, a display, and displays the processing result of the guidance estimation device 10. The guidance estimation device 10 may not have at least one of the input device 201 and the display device 202.

  The external I / F 203 is an interface with an external device. The external device is, for example, a recording medium 203a. The guidance estimation apparatus 10 can read and write the recording medium 203 a and the like via the external I / F 203. The recording medium 203a may store a program or the like for realizing each function of the guidance estimation device 10.

  Examples of the recording medium 203 a include a flexible disk, a CD (Compact Disc), a DVD (Digital Versatile Disk), an SD memory card (Secure Digital memory card), and a USB (Universal Serial Bus) memory card.

  The RAM 204 is a volatile semiconductor memory that temporarily holds programs and data. The ROM 205 is a non-volatile semiconductor memory that can hold programs and data even after the power is turned off. The ROM 205 stores, for example, OS (Operating System) settings, network settings, and the like.

  The CPU 206 is an arithmetic device that reads a program or data from the ROM 205, the auxiliary storage device 208, or the like onto the RAM 204 and executes processing.

  The communication I / F 207 is an interface for connecting the guidance estimation device 10 to a communication network. A program for realizing each function of the guidance estimation device 10 may be acquired (downloaded) from a predetermined server device or the like via the communication I / F 207.

  The auxiliary storage device 208 is, for example, a HDD, an SSD, or the like, and is a non-volatile storage device storing programs and data. The programs and data stored in the auxiliary storage device 208 include, for example, an OS, an application program for realizing various functions on the OS, and a program for realizing each function of the guidance estimation device 10.

  The guidance estimation apparatus 10 according to the embodiment of the present invention can realize the various processes described above by having the hardware configuration shown in FIG. 4.

  The present invention is not limited to the above specifically disclosed embodiments, and various modifications and changes are possible without departing from the scope of the claims.

DESCRIPTION OF SYMBOLS 10 Guidance estimation apparatus 110 Data preparation part 111 Road structure preparation part 112 Road condition preparation part 113 Guided road structure preparation part 114 Guidance effect measurement part 115 Road effect data preparation part 120 Parameter estimation part 130 Guidance estimation part 140 Storage part

Claims (9)

When the road structure representing the structure of the road and the road condition for each road structure are acquired or created, the guidance road structure representing the structure of the road in the case where guidance is performed for each road structure, and the guidance is performed Data creation means for creating road effect data including a guidance effect representing an effect when the guidance is performed and a road structure and / or at least one of the road conditions when no guidance is given;
A data creation apparatus characterized by having: A guidance model learning device that estimates parameters of a model that outputs information related to guidance when at least one of a road structure representing a road structure and a road condition for each road structure is input,
Information about the said guidance output by inputting into the said model the road effect data created by the data creation means in the data creation device according to claim 1 as learning data, and the guidance effect included in the road effect data Parameter estimation means for estimating the parameters to minimize errors;
A guidance model learning device characterized by having. The parameter estimation means
The guidance model learning device according to claim 2, wherein the parameter is estimated so as to minimize an error between information on the guidance and a value obtained by normalizing the guidance effect using a predetermined function. . A guidance estimation device that estimates information related to guidance by a model that outputs information related to guidance when at least one of a road structure representing a structure of a road and road conditions for each road structure is input,
4. The guidance according to claim 2, wherein at least one of the road structure and the road condition is input to the model in which the parameter estimated by the parameter estimation unit in the guidance model learning device according to claim 2 or 3 is set. Guidance estimation means for estimating information,
A guidance estimation device characterized by having. When the road structure representing the structure of the road and the road condition for each road structure are acquired or created, the guidance road structure representing the structure of the road in the case where guidance is performed for each road structure, and the guidance is performed A data creating procedure for creating road effect data including a guiding effect representing an effect when the guiding is performed for no case and at least one of the road structure and the road condition;
A computer executes the data creation method. When at least one of a road structure representing a road structure and a road condition for each of the road structures is input, a computer that estimates parameters of a model that outputs information related to guidance,
Information regarding the said guidance output by inputting into the said model the road effect data created by the data creation procedure in the data creation method according to claim 5 as learning data, and the guidance effect included in the road effect data A parameter estimation procedure to estimate the parameters to minimize errors;
A guided model learning method characterized in that: The parameter estimation procedure is
The method according to claim 6, wherein the parameter is estimated so as to minimize an error between information on the guidance and a value obtained by normalizing the guidance effect using a predetermined function. . When at least one of a road structure representing a road structure and a road condition for each of the road structures is input, a computer for estimating the information on the guidance with a model that outputs information on the guidance,
8. The guidance according to claim 6, wherein at least one of the road structure and the road condition is input to the model in which the parameter estimated by the parameter estimation procedure in the guidance model learning method according to claim 6 or 7 is set. Guidance estimation procedure to estimate information,
A method of estimating a lead according to claim 1.   A program for causing a computer to function as each means in the data generation device according to claim 1, each means in the parameter estimation device according to claim 2 or 3, or each means in the guidance estimation device according to claim 4.

Images