JP6562431B1 - Server apparatus, system, method and program - Google Patents

Abstract

A cost associated with movement between two points is provided using prediction by machine learning. A server device stores graph information relating to a connected graph composed of a plurality of nodes corresponding to points on a map, edges connecting the nodes, and a cost added to each edge. An information storage unit, a cost calculation unit that calculates a predicted cost between two nodes selected from a plurality of nodes based on graph information, and a point between two points on the map corresponding to the selected two nodes A cost acquisition unit for acquiring the actual measurement cost, a prediction cost calculated by the cost calculation unit, and an actual measurement cost acquired by the cost acquisition unit as a learning data, and a prediction based on the learning data A model generation unit that generates a learning model that learns the relationship between the cost and the actual measurement cost, and a cost between any two points on the map based on the generated learning model. Comprising a cost providing unit for, a. [Selection] Figure 1

Description

  The present disclosure relates to a server device, a system, a method, and a program that provide a cost related to movement between two points on a map.

  In recent years, with the development of computing resource technology, big data-related infrastructure has been developed. As an important component of a big data processing platform that processes a huge amount of data, there is an increasing demand for machine learning technology that derives the laws and knowledge lurking behind the data.

  In addition, IT service development and provision forms and devices are changing in the market, and the trend to more easily develop new services in the short term using the published Web API (Application Programming Interface) has been established. is there. For example, Patent Document 1 discloses a route proposal device that proposes a route that goes around three or more points in a predetermined order using an external route search engine.

JP-A-2018-72150

  However, the above-described Web service has a problem that when a large amount of API is executed from a client service such as a user terminal or other Web services, a load is applied to the server.

  Therefore, the present invention has been made in view of the above problems, and an object of the present invention is to provide a server device, a system, a method, and a program that provide costs related to movement between two points using prediction by machine learning. That is.

In order to achieve the above object, a server device according to the present disclosure sets a plurality of nodes corresponding to points on a map, and adds edges connecting the nodes to which a cost is added to the plurality of nodes. And a graph generation unit that generates a connected graph in which a path exists between any two points of the plurality of nodes , the plurality of nodes, the added edge, and the cost added to each edge. A graph information storage unit that stores graph information related to a connected graph, a cost calculation unit that calculates a predicted cost between two nodes selected from a plurality of nodes based on the graph information, and two selected The cost acquisition unit that acquires the actual measurement cost between two points on the map corresponding to the node, the predicted cost calculated by the cost calculation unit, and the actual measurement cost acquired by the cost acquisition unit A learning data storage unit that stores data, a model generation unit that generates a learning model that learns the relationship between the predicted cost and the actual measurement cost based on the learning data, and an arbitrary map on the map based on the generated learning model The cost provision part which provides the cost between the nodes in the connection graph corresponding to the said 2 points as a cost between the 2 points based on the generated learning model .

In order to achieve the above object, the system according to the present disclosure sets a plurality of nodes corresponding to points on the map, and adds an edge connecting the nodes to each of the plurality of nodes with added cost. And a graph generation unit that generates a connected graph in which a path exists between any two points of the plurality of nodes , the plurality of nodes, the added edge, and the cost added to each edge. A graph information storage unit that stores graph information related to the connected graph, a cost calculation unit that calculates a predicted cost between two nodes selected from a plurality of nodes, based on the graph information, and the selected 2 The cost acquisition unit that acquires the actual measurement cost between two points on the map corresponding to one node, the predicted cost calculated by the cost calculation unit, and the actual measurement cost acquired by the cost acquisition unit As a learning data storage unit, a model generation unit that generates a learning model that learns the relationship between the predicted cost and the actual measurement cost based on the learning data, and an arbitrary map on the map based on the generated learning model The cost provision part which provides the cost between the nodes in the connection graph corresponding to the said 2 points as a cost between the 2 points based on the generated learning model .

In order to achieve the above object, a method according to the present disclosure is a method for generating a learning model for providing a cost between two points on a map, the method being executed in a computer, A control unit and a storage unit are provided, and the control unit sets a plurality of nodes corresponding to points on the map, and adds an edge connecting the nodes to which the cost is added to the plurality of nodes, The step of generating a connected graph in which a path exists between any two points of the plurality of nodes , and the control unit are configured by the plurality of nodes, the added edge, and the cost added to each edge. and storing the graph data in the storage unit about the connection graph that the steps of the control unit, the estimated cost of between two nodes selected from among a plurality of nodes is calculated based on the graph information Control unit, acquiring a measured cost between two points on the map corresponding to the two nodes that are selected, the control unit, and the calculated estimated cost, the actual cost of the acquired storage unit as learning data And a step of generating a learning model in which the control unit learns the relationship between the prediction cost and the actual measurement cost based on the learning data.

  Moreover, in order to achieve the said objective, the program which concerns on this indication is a program which produces | generates the learning model for providing the cost between two points on a map, Comprising: Let the computer perform the said method.

  According to the present disclosure, it is possible to provide a cost related to movement between two points using prediction by machine learning.

It is a block diagram of the travel time provision system 1 between two points. 3 is a functional block diagram illustrating an example of a functional configuration of a user terminal 100. FIG. 3 is a functional block diagram illustrating an example of a functional configuration of a server device 300. FIG. It is a figure which shows the simple example of the graph information in the map. 5 is a flowchart illustrating processing for generating a learning model in a control unit 330. It is a flowchart which shows the process which provides the movement time between two points based on a learning model in the control part. It is a figure which shows the specific example of the node corresponding to the point on a map. FIG. 8 is a diagram in which Delaunay triangulation is applied to the node in FIG. 7. It is a figure which shows the example of a regression equation. 3 is a functional block diagram illustrating an example of a functional configuration of a server device 400. FIG. It is a figure explaining the process which selects two arbitrary points on a map. 3 is a functional block diagram illustrating an example of a functional configuration of a server device 500. FIG.

  Hereinafter, embodiments of the present disclosure will be described with reference to the drawings. In all the drawings for explaining the embodiments, common constituent elements are denoted by the same reference numerals, and repeated explanation is omitted. The following embodiments do not unduly limit the contents of the present disclosure described in the claims. In addition, all the components shown in the embodiments are not necessarily essential components of the present disclosure.

<Embodiment 1>
FIG. 1 is a configuration diagram of a travel time providing system 1 between two points. With reference to FIG. 1, the configuration of the system according to the first embodiment will be described.

<Outline of travel time providing system 1>
The system which concerns on this embodiment is a system which provides the movement time regarding the movement between two points on a map.

  In recent years, when planning a delivery plan in a logistics system or the like, a user can acquire a travel time to a delivery destination by a Web API that provides travel time between two points on a map. However, in such a Web service, there is a problem that if there is a large amount of inquiries from clients, a load is applied to the API server, and the use of the API is limited. For example, if the limit number of times per unit time is exceeded, the use of the API for a certain period of time is prohibited or charged according to the number of uses. Therefore, it is not practical to use the API every time a delivery plan is made. There is a case.

  Therefore, the travel time providing system 1 according to the present embodiment provides the travel time between two points using machine learning. Specifically, the travel time providing system 1 includes two travel points calculated based on the combination optimization problem and two actually measured points acquired from an external map information providing device (API server or the like). A learning model that learns the relationship with the travel time between is generated, and the travel time between two points on the map is provided based on the learning model.

  That is, since the travel time providing system 1 uses the API server only when generating the learning model, it is possible to provide the travel time between two points with high accuracy while reducing the load on the API server. In addition, the user can acquire the travel time between two points without worrying about restrictions on the use of services such as APIs, and can easily make a delivery plan and the like.

<Configuration of travel time providing system 1>
In FIG. 1, the travel time providing system 1 includes a user terminal 100, a map information providing device 200, and a server device 300, which are communicably connected via a network NW. The network NW includes a WAN (World Area Network), a LAN (Local Area Network), and the like.

  Although only one user terminal 100 is shown in FIG. 1, in this embodiment, the travel time providing system 1 includes a plurality of user terminals 100 that are held for each user.

  The map information providing apparatus 200 is a server apparatus that provides map information. Here, “map information” is geographical information for presenting a map, and specifically, map data (for example, latitude, longitude, etc.) necessary for specifying various positions including stores, houses, etc. ). Further, the map information providing apparatus 200 can provide the cost between any two points on the map based on the map information. Here, “cost” is a value defined between two points. For example, the distance between two points, the travel time required to move between two points, the amount of fuel consumed, the transportation cost, etc. is there. Hereinafter, in this specification, the cost between two points will be described as the travel time required to move between the two points. The travel time between the two points is preferably provided based on an actual measurement value, but is provided based on a value in which an error from the actual measurement value is guaranteed, such as an approximate value obtained by executing a known route search calculation. May be. The map information providing apparatus 200 is, for example, an external route search engine or an API server. However, the map information providing apparatus 200 is not limited to this as long as the apparatus provides a travel time between any two points on the map.

  The user terminal 100 is an information processing terminal held by the user, and inquires the server device 300 about the travel time for two points on the map received from the user. For example, the user terminal 100 acquires map information from the map information providing apparatus 200 and selects two points on the map. Then, the server apparatus 300 is inquired about the travel time between the two selected locations.

  The server device 300 is a server device that provides the travel time between two points on the map, and provides the travel time between the two points in response to an inquiry from the user terminal 100. Specifically, the server device 300 calculates the relationship between the travel time between two points calculated based on the combination optimization problem and the measured travel time between the two points acquired from the map information providing device 100. Generate a learned learning model. And based on the said learning model, the travel time between two points on a map is provided.

  As described above, in the present embodiment, the server device 300 provides the travel time between two points in response to an inquiry from the user terminal 100. Thereby, the user can acquire the travel time between two points without worrying about the number of times the service such as API is used.

  In the present embodiment, the “server” refers to one information processing apparatus (that is, a server apparatus), and when the server is configured by a plurality of server apparatuses, a server apparatus group (that is, a server system). It means the whole thing. In the present embodiment, the server apparatus 300 is described as an integral configuration. However, the server apparatus 300 may include a plurality of server apparatuses that are divided according to functions and / or roles. For example, the server device 300 includes a data server that stores data acquired from the user terminal 100 and the map information providing device 200, and a service server that provides a travel time between two points based on a learning model. May be. Furthermore, the server apparatus 300 may be configured to include a shop server that performs billing while providing travel time.

  FIG. 2 is a functional block diagram illustrating an example of a functional configuration of the user terminal 100. Note that the user terminal 100 of the present embodiment may have a configuration in which some of the components (each unit) in FIG. 2 are omitted.

  The user terminal 100 is an information processing apparatus, and in this embodiment, for example, a smartphone, a feature phone, a tablet computer, a laptop computer, a desktop computer, a wearable terminal such as a smart glass or a head-mounted display, or It is a multifunction device such as a multifunction television receiver (smart television) having an information processing function.

  That is, the user terminal 100 has various functions (for example, an input function, an output (display) function, an information processing function, a network communication function, a sensor function, a call function, a camera function, etc.) that a general multifunction device has. doing.

  The network communication function is a communication function via the Internet or the like and / or a communication function via a mobile communication network. The user terminal 100 may be realized by installing a predetermined function in an off-the-shelf multifunction device. In the present embodiment, the user terminal 100 is used to acquire the travel time between the two points in addition to being used as the multifunction device.

  The user terminal 100 includes a communication unit 110, an input unit 120, an output unit 130, a storage unit 140, a sensor unit 150, and a control unit 160.

  The communication unit 110 performs various controls for communicating with a device connected to the network NW, and the function can be realized by hardware such as various processors or a communication ASIC, a program, or the like.

  The input unit 120 is an interface for receiving input from the user, and sends the user input to the control unit 160. The input unit 120 is, for example, a touch panel, a button, a microphone, or a controller. Sensing data detected by the sensor unit 150 described later may be input from the user. For example, the user can specify two points on the map by confirming the map information displayed on the touch panel and tapping a point on the map.

  The output unit 130 is, for example, a display device such as a display or an audio output device such as a speaker. The output unit 130 displays and outputs various images and sounds generated in the user terminal 100 in response to an input to the input unit 120, or a network. Various images and sounds based on data received by the communication unit 110 via the NW are displayed and output. The output unit 130 includes an artificial intelligence-equipped speaker (smart speaker).

  The storage unit 140 is a storage device for storing a program for causing a computer to function and various data. For example, the storage unit 140 stores map information (not shown) transmitted from the map information providing apparatus 200. Note that the storage unit 140 may include a temporary storage area or storage.

  The sensor unit 150 is various devices that detect various states of the user terminal 100. The sensor unit 150 is, for example, a posture sensor (acceleration sensor or gyro sensor) that detects the posture or inclination of the terminal itself, a gaze sensor that detects the direction of the user's line of sight, an optical sensor that detects ambient brightness, and a user's operation. It is an infrared sensor that detects The sensor unit 150 is a microphone that collects sound around the user terminal 100, a humidity sensor that detects humidity around the user terminal 100, a geomagnetic sensor that detects a magnetic field at the location of the user terminal 100, and the like. May be.

  Further, the sensor unit 150 may detect various information using the sensor function described above. For example, the sensor unit 150 may detect the number of walks of the user who owns the user terminal 100 using the function of the acceleration sensor. In addition, the sensor unit 150 detects the operation information indicating whether the user terminal 100 is operating or stationary using a function of the acceleration sensor at every predetermined time or every timing when the user terminal 100 is operated. May be. The sensor unit 150 sends the sensing data detected as described above to the control unit 160.

  Note that the sensor unit 150 may be an information processing terminal (so-called wearable terminal) that can be worn by a user and connected to the user terminal 100 such as a wristwatch type or a ring type terminal.

  The control unit 160 executes various types of information processing executed on the user terminal 100. The control unit 160 includes a CPU (Central Processing Unit) and a memory. In the user terminal 100, the CPU executes various information processes by executing an information processing program stored in the storage unit 140 using a memory. In the present embodiment, the control unit 160 receives, as the information processing, processing for acquiring map information from the map information providing apparatus 200 or two points on the map from the user, and determines the travel time between the two points as a server device. Processing for inquiring 300 is executed. When the user terminal 100 operates as a multifunction device, the control unit 160 executes information processing for realizing each function.

  FIG. 3 is a functional block diagram illustrating an example of a functional configuration of the server apparatus 300 according to the first embodiment. Note that the server device 300 of the present embodiment may have a configuration in which some of the components (each unit) in FIG. 3 are omitted.

  The server device 300 includes a communication unit 310, a storage unit 320, and a control unit 330.

  The communication unit 310 performs various controls for communicating with a device connected to the network NW, and the function can be realized by hardware such as various processors or a communication ASIC, a program, or the like.

  The storage unit 320 is a storage device for storing a program for causing a computer to function and various data. The storage unit 320 includes map information 321, graph information 322, learning data 323, and a learning model 324. These will be described later.

  The control unit 330 executes various types of information processing executed in the server device 300. The control unit 330 includes a CPU and a memory, and various information processing is executed by the CPU executing an information processing program stored in the storage unit 320 using the memory. In the present embodiment, as the information processing, the control unit 330 generates a learning model related to travel time, and executes processing for calculating travel time between two points based on the learning model. The calculated travel time is transmitted to the user terminal 100.

  The control unit 330 includes an acquisition unit 331, a graph generation unit 332, a time calculation unit 333, a model generation unit 334, and a time provision unit 335.

  The acquisition unit 331 acquires map data from the map information providing apparatus 200 via the communication unit 310. Then, the acquired map data is stored in the map information 321 of the storage unit 320.

  The graph generation unit 332 generates a connected graph based on the map information 321 and stores it in the graph information 322 of the storage unit 320. Specifically, the graph generation unit 332 sets a plurality of nodes corresponding to points on the map. Then, the graph generation unit 332 generates a connected graph in which a path (route) exists between any two points by adding an edge connecting the nodes to the set plurality of nodes. For example, a connected graph may be generated using Delaunay triangulation. The generated graph may be a directed graph or an undirected graph.

  The points on the map corresponding to the nodes may be, for example, houses, facilities, delivery bases, intersections, and the like. Further, the map may be divided into grids, and representative points in the divided areas (points located in the center of the area, population-populated points in the area, etc.) may be used as points on the map.

  In addition, the graph generation unit 332 adds a movement time (cost) to each edge. Specifically, the graph generation unit 332 acquires the travel time between points corresponding to the node to which the edge is connected from the map information providing apparatus 200 and stores it in the graph information 322 in association with the edge. That is, the graph information 322 functions as a graph information storage unit, and includes a plurality of nodes corresponding to points on the map, an edge connecting two nodes among the plurality of nodes, and a node to which the edge connects. Information on the connected graph composed of the travel time, the correspondence between the nodes and points on the map, and the like are stored.

  The time calculation unit 333 functions as a cost calculation unit, and calculates the shortest travel time (predicted cost) between any two points in the connected graph by solving an optimization problem based on the graph information 322. Specifically, for example, the travel time of the shortest path between two points is obtained by the Dijkstra's algorithm, which is a typical technique for solving the shortest path problem. In addition to the Dijkstra method, the travel time may be obtained by solving the shortest path problem by a known method.

  On the other hand, the acquisition unit 331 described above also functions as a cost acquisition unit, and calculates the actual travel time (measurement cost) between two points on the map corresponding to the start point and end point of the shortest path calculated by the time calculation unit 333. Obtained from the map information providing apparatus 200.

  And the acquisition part 331 makes the learning data of the memory | storage part 320 into learning data by making the movement time which the time calculation part 333 calculated | required and the said measured movement time between two arbitrary points of a connection graph as a pair. Stored in H.323. That is, the learning data 323 functions as a learning data storage unit.

  Based on the learning data 323, the model generation unit 334 generates a learning model in which the relationship between the travel time calculated by the time calculation unit 333 and the actually measured travel time acquired by the acquisition unit 331 is learned, and the storage unit 320. Are stored in the learning model 324. Specifically, the model generation unit 334 performs machine learning on the relationship between the calculated travel time and the actually measured travel time by regression analysis. As the regression method, for example, a well-known method such as linear regression, polynomial regression, or logistic regression can be used, and thus detailed description thereof is omitted.

  The time providing unit 335 functions as a cost providing unit, and provides the travel time between two points on the inquired map based on the learning model 324 generated by the model generating unit 334. Specifically, the acquisition unit 331 acquires information on nodes corresponding to two points on the map inquired by the user terminal 100 based on the graph information 322. For example, a node set near the inquired point (within a predetermined distance) is set as a node corresponding to the inquired point. Next, the time calculation unit 333 calculates a travel time by solving the shortest path problem between the acquired nodes. And the time provision part 335 correct | amends the calculated moving time with the regression type stored in the learning model 324, and provides it as the moving time between two points.

  In this embodiment, an area on a map is represented by a connection graph, and a movement time between two points on the map is provided by correcting the movement time calculated based on the connection graph by a learning model. That is, it is only necessary to acquire the travel time between two points for the map information providing apparatus 200 only when generating the learning model. Thereby, it can suppress that the inquiry of the travel time between two points from the user terminal 100 to the map information provision apparatus 200 concentrates.

  Note that the actual measurement time stored as learning data is collected based on the data sensed by the sensor unit 150 of the user terminal 100 after the user has actually moved in response to an inquiry, The learning model may be updated.

  FIG. 4 is a diagram illustrating a simple example of graph information in the map 40. FIG. 5 is a flowchart showing processing for generating a learning model in the control unit 330. FIG. 6 is a flowchart showing a process of providing the travel time between two points based on the learning model in the control unit 330. With reference to FIG. 4 and FIG. 5, the generation of the learning model described above will be specifically described. Moreover, with reference to FIG. 4 and FIG. 6, provision of the travel time between two points based on a learning model is demonstrated concretely.

  In step S <b> 501, the acquisition unit 331 acquires map information for indicating the map 40 from the map information providing apparatus 200.

  In step S502, the graph generation unit 332 generates a graph. Specifically, the graph generation unit 332 first sets a plurality of nodes corresponding to points on the map 40. In the example of FIG. 4, in the map 40, the node v1 corresponding to the house A, the node v2 corresponding to the house B, the node v3 corresponding to the station C, the node v4 corresponding to the facility B, the node v5 corresponding to the facility A, A node v6 corresponding to the house C is set. For example, the graph generation unit 332 sets a node corresponding to a point designated by the user terminal 100 or the system developer. And the graph production | generation part 332 produces | generates the connection graph G (henceforth graph G) using Delaunay triangulation with respect to the nodes v1-v6. A connected graph is a graph in which a path (route) exists between any two nodes. As shown in FIG. 4, there is a path between one arbitrary node and one of which is a start point and the other is an end point. There is at least one.

  Further, the graph generation unit 332 adds time (cost) required for movement between nodes to the edge connecting the nodes. The travel time is acquired from the map information providing apparatus 200. For example, the cost c1 is added to the edge connecting the nodes v1 and v2. Specifically, the cost is the time taken to move from the house A to the house B (or from the house B to the house A), and is a time length of, for example, 10 minutes.

  The graph generation unit 332 stores in the graph information 322 information about the graph G such as nodes, edges, node-to-edge connection relationships, costs added to the edges, and correspondence relationships between nodes and points on the map.

  In step S503, the time calculation unit 333 selects two arbitrary points in the graph G, and calculates the shortest movement time between the two points. For example, the time calculation unit 333 calculates the travel time by obtaining the shortest path from the node v2 to v6 using the Dijkstra method. There are a plurality of paths from the node v2 to v6, such as path 1 (v2 → v1 → v6), path 2 (v2 → v5 → v6), path 3 (v2 → v3 → v4 → v6), and so on. When the path 2 is obtained as the shortest path by the Dijkstra method, the travel time is (c5 + c10). The Dijkstra method is a well-known algorithm and will not be described in detail. The time calculation unit 333 selects two arbitrary points in the graph G to a degree sufficient as learning data for machine learning, and calculates the movement time between the two points.

  In step S504, the acquisition unit 331 acquires the actually measured travel time between two corresponding points on the map for any two points selected by the time calculation unit 333. Specifically, the acquisition unit 331, for example, for the nodes v2 and v6 selected by the time calculation unit 333, the travel time from the house B, which is a corresponding point on the map, to the house C is obtained from the map information providing apparatus 200. get. Here, the travel time acquired by the acquisition unit 331 is the time (actually measured value) required to actually move from the house B to the house C. On the other hand, the travel time calculated by the time calculation unit 333 is the cost (predicted value) of the shortest path in the graph G from the node v2 corresponding to the house B to the node v6 corresponding to the house C.

  In step S <b> 505, the acquisition unit 331 stores the above measured value and predicted value in the learning data 323.

  In step S506, the model generation unit 334 performs machine learning on the relationship between the predicted value and the actual measurement value between the two points based on the learning data 323 by regression analysis. And it stores in the learning model 324 of the memory | storage part 320 as a learning model.

  Next, with reference to FIG. 6, the process for providing the travel time between two points based on the learning model will be described.

  In step S <b> 601, the communication unit 310 receives an inquiry about travel time between two points from the user terminal 100. Specifically, in the example of FIG. 4, the travel time from the house D to the house E on the map 40 is inquired.

  In step S602, the time providing unit 335 acquires a node v2 corresponding to the house B close to the house D and a node v6 corresponding to the house C close to the house E based on the graph information 322.

  In step S <b> 603, the time providing unit 335 causes the time calculation unit 333 to solve the shortest path problem from the nodes v <b> 2 to v <b> 6 in the graph G. In the example of FIG. 4, the travel time (c5 + c10) is calculated.

  In step S <b> 604, the time providing unit 335 corrects the calculated travel time using a regression equation stored in the learning model 324.

  In step S605, the time providing unit 335 provides the corrected value as the travel time between two points.

  FIG. 7 is a diagram illustrating a specific example of nodes corresponding to points on the map. In FIG. 7, each node corresponds to each address of the area indicated by the map, and each node is associated with position information (latitude, longitude, etc.) of each address.

  FIG. 8 is a diagram in which Delaunay triangulation is applied to the node of FIG. For each edge of the graph in FIG. 8, the time required to move the position (between points) associated with the node to which the edge is connected is added as a cost.

  FIG. 9 is a diagram illustrating an example of a regression equation. The same data is represented by a logarithmic axis (left graph) and a linear axis (right graph). The vertical axis is the actually measured travel time, and the horizontal axis is the travel time calculated by the Dijkstra method. Further, the relationship between the actually measured travel time and the calculated travel time is machine-learned, and regression equations 90 and 91 are derived.

  In FIG. 9, the longer the actually measured travel time, the greater the deviation (prediction error) from the calculated travel time, and this prediction error can be corrected by regressing.

(Explanation of effect)
As described above, the travel time providing system 1 according to the present embodiment provides a travel time between two points using machine learning. That is, the travel time providing system 1 generates a learning model in which the relationship between the travel time between two points calculated based on the combination optimization problem and the measured travel time between the two points is generated, and the learning is performed. Based on the model, it provides the travel time between two points on the map. Since the travel time providing system 1 uses an external map information providing device (for example, an API server) only when generating a learning model, the travel between two points with high accuracy is achieved while reducing the load on the API server. Can provide time. In addition, the user can acquire the travel time between two points without worrying about restrictions on the use of services such as APIs, and can easily make a delivery plan and the like.

<Embodiment 2>
The travel time providing system 2 according to the second embodiment includes a server device 400 in place of the server device 300 in the travel time providing system 1 according to the first embodiment. The server device 400 generates a learning model by machine learning of the relationship between the position of two points on the map and the moving time (cost) between the two points, and calculates the moving time between the two points based on the learning model. provide. That is, the generated learning model is different from the server device 300 according to the first embodiment.

  FIG. 10 is a functional block diagram illustrating an example of a functional configuration of the server apparatus 400 according to the second embodiment. Note that the server device 400 of this embodiment may have a configuration in which some of the components (each unit) in FIG. 10 are omitted.

  The server device 400 includes a communication unit 410, a storage unit 420, and a control unit 430.

  The communication unit 410 performs various controls for communicating with a device connected to the network NW, and the function can be realized by hardware such as various processors or a communication ASIC, a program, or the like.

  The storage unit 420 is a storage device for storing a program for causing a computer to function and various data. The storage unit 420 includes map information 421, learning data 422, and a learning model 423. These will be described later.

  The control unit 430 executes various types of information processing executed in the server device 400. The control unit 430 includes a CPU and a memory, and various information processing is executed by the CPU executing an information processing program stored in the storage unit 420 using the memory. In the present embodiment, as the information processing, the control unit 430 generates a learning model related to travel time, and executes processing for calculating travel time between two points based on the learning model. The calculated travel time is transmitted to the user terminal 100.

  The control unit 430 includes an acquisition unit 431, a model generation unit 432, and a time provision unit 433.

  The acquisition unit 431 acquires map data from the map information providing apparatus 200 via the communication unit 410. Then, the acquired map data is stored in the map information 421 of the storage unit 420.

  The acquisition unit 431 selects any two points on the map, and acquires the travel time between the two points from the map information providing apparatus 200 via the communication unit 410. And the acquisition part 431 pairs the positional information (for example, latitude, longitude, etc.) between two points, and the movement time between two points, and stores them in the learning data 422 of the memory | storage part 420 as learning data. The acquisition unit 431 selects two arbitrary points on the map to a degree sufficient as learning data for machine learning, and acquires the travel time between the two points.

  FIG. 11 is a diagram illustrating a process of selecting two arbitrary points on the map. In FIG. 11, the acquisition unit 431 randomly selects two arbitrary points on the map 40. In addition, in FIG. 11, although the points 41-46 are shown as an example, it is not restricted to this, The arbitrary points on a map are selected.

  For example, the acquisition tool 431 selects the point 41 and the point 42, and acquires the travel time t1 between the point 41 and the point 42 from the map information providing apparatus 200. Then, the acquisition unit 431 stores the position information (x1, y1) such as the latitude and longitude of the point 41 and the movement time t1 in pairs in the learning data 422. Similarly, the acquisition unit 431 randomly selects any two points such as the point 41 and the point 43, the point 43 and the point 44, and acquires the movement time between the two points. Then, the position data of two locations and the travel time are paired and stored in the learning data 422.

  Returning to FIG. 10, the model generation unit 432 generates a learning model in which the relationship between the position information of two points on the map and the travel time between the two points is learned based on the learning data 423, and Store in the learning model 423. For example, based on a known technique such as a Gaussian process or a neural network, the relationship between the position information of two points on the map and the travel time between the two points is learned.

  The time providing unit 433 provides the travel time between two points on the inquired map based on the learning model 423 generated by the model generation unit 432. Specifically, the acquisition unit 431 calculates the travel time based on the learning model with the two points on the map inquired by the user terminal 100 as inputs.

(Explanation of effect)
As described above, the travel time providing system according to the present embodiment provides the travel time between two points using machine learning. Specifically, a learning model is generated by machine learning of the relationship between the position of two points on the map and the movement time (cost) between the two points, and the movement time between the two points is calculated based on the learning model. provide. That is, since the travel time providing system according to the present embodiment uses an external map information providing apparatus (for example, an API server) only when generating a learning model, accuracy can be reduced while reducing the load on the API server. High travel time between two points can be provided. In addition, the user can acquire the travel time between two points without worrying about restrictions on the use of services such as APIs, and can easily make a delivery plan and the like.

<Embodiment 3>
The travel time providing system 3 according to the third embodiment includes a server device 500 instead of the server device 400 in the travel time providing system 2 according to the second embodiment. The server device 500 provides the travel time between two points based on the first learning model generated in the system according to the first embodiment and the second learning model generated in the system according to the second embodiment. Specifically, based on the prediction accuracy of the travel time calculated based on the first learning model and the prediction accuracy of the travel time calculated based on the second learning model, the better travel time is calculated. select. Then, the selected travel time is provided to the user. That is, it differs from the above-described embodiment in that the travel time is calculated based on both the learning model according to the first embodiment and the learning model according to the second embodiment.

  FIG. 12 is a functional block diagram illustrating an example of a functional configuration of the server device 500 according to the third embodiment. Note that the server device 500 of this embodiment may have a configuration in which some of the components (each unit) in FIG. 12 are omitted.

  The server device 500 includes a communication unit 510, a storage unit 520, a first estimation unit 530, a second estimation unit 540, a selection unit 550, and a provision unit 560. In the server device 500, a control unit (not shown) having a CPU and a memory executes various information processing described later by executing an information processing program stored in the storage unit 520.

  The communication unit 510 performs various controls for communicating with a device connected to the network NW, and the function can be realized by hardware such as various processors or a communication ASIC, a program, or the like.

  The storage unit 520 is a storage device for storing a program for causing a computer to function and various data. The storage unit 520 includes map information 521, graph information 522, first learning data 523, a first learning model 524, second learning data 525, and a second learning model 526. These will be described later.

  The first estimation unit 530 has a function of generating the learning model shown in the first embodiment and calculating the travel time between two points on the map based on the learning model. Moreover, the 2nd estimation part 540 has a function which produces | generates the learning model shown in Embodiment 2, and calculates the movement time between two points on a map based on the said learning model.

  The first estimation unit 530 includes a first acquisition unit 531, a graph generation unit 532, a time calculation unit 533, a first model generation unit 534, and a first time estimation unit 535.

  The first acquisition unit 531 has the same function as the acquisition unit 331 of the server apparatus 300 according to the first embodiment, and acquires map data from the map information providing apparatus 200 via the communication unit 510. Then, the acquired map data is stored in the map information 521 of the storage unit 520.

  Further, the first acquisition unit 531 displays the travel time (actual measurement cost) measured between two points on the map corresponding to the start point and the end point of the shortest path calculated by the time calculation unit 533 described later. And stored in the first learning data 523.

  The graph generation unit 532 has the same function as the graph generation unit 332 of the server apparatus 300 according to the first embodiment, generates a connected graph based on the map information 521, and stores it in the graph information 522 of the storage unit 520. .

  The time calculation unit 533 has the same function as the time calculation unit 333 of the server apparatus 300 according to the first embodiment, and solves an optimization problem based on the graph information 522, thereby making it possible to connect any two points of the connected graph. Calculate the shortest travel time (predicted cost). The travel time is stored in the first learning data 523 as a pair with the measured travel time acquired from the map information providing apparatus 200.

  The first model generation unit 534 has the same function as that of the model generation unit 334 of the server device according to the first embodiment. The first model generation unit 534 includes the travel time calculated by the time calculation unit 533 based on the first learning data 523, and the first A learning model that learns the relationship with the actually measured travel time acquired by the acquisition unit 531 is generated and stored in the first learning model 524 of the storage unit 520.

  The first time estimation unit 535 has the same function as the time provision unit 335 of the server device 300 according to the first embodiment, and is based on the first learning model 524 generated by the first model generation unit 534. Calculate (estimate) the travel time between two points. At this time, an index for evaluating the prediction accuracy is calculated. About evaluation of prediction accuracy, since well-known techniques, such as an average error and an average square-square error, are employ | adopted, detailed description is abbreviate | omitted.

  The second estimation unit 540 includes a second acquisition unit 541, a second model generation unit 542, and a second time estimation unit 543.

  The second acquisition unit 541 has the same function as the acquisition unit 431 of the server device 400 according to the second embodiment, selects any two points on the map based on the map information 521, and selects between the two points. The travel time is acquired from the map information providing apparatus 200 via the communication unit 510. And the 2nd acquisition part 541 makes the 2nd learning data 525 of the memory | storage part 520 as learning data by making the positional information (for example, latitude, longitude, etc.) between 2 points | pieces and the movement time between 2 points | pieces into a pair. Store.

  The second model generation unit 542 has the same function as the model generation unit 432 of the server apparatus 400 according to the second embodiment. Based on the second learning data 525, the position information of two points on the map and the distance between the two points A learning model that learns the relationship with the travel time is generated and stored in the second learning model 526 of the storage unit 520.

  The second time estimation unit 543 has the same function as the time provision unit 433 of the server apparatus 400 according to the second embodiment, and is based on the second learning model 526 generated by the second model generation unit 542. Calculate (estimate) the travel time between two points. At this time, an index for evaluating the prediction accuracy is calculated. About evaluation of prediction accuracy, since well-known techniques, such as an average error and an average square-square error, are employ | adopted, detailed description is abbreviate | omitted.

  The selection unit 550 receives an inquiry about the travel time between two points on the map from the user terminal 100, the travel time calculated by the first estimation unit 530, and the travel time calculated by the second estimation unit 540, Select the one with better prediction accuracy. Specifically, the selection unit 550 selects the travel time by comparing the indexes for evaluating the prediction accuracy described above.

  The providing unit 560 provides the travel time selected by the selecting unit 550. Specifically, the data is transmitted to the user terminal 100 via the communication unit 510.

(Explanation of effect)
As described above, the travel time providing system according to the present embodiment calculates a travel time between two points based on a learning model generated by a different method, and provides a better prediction accuracy. Thereby, the travel time providing system according to the present embodiment can provide a more accurate travel time between two points while reducing the load on an external map information providing device (for example, an API server). . In addition, the user can acquire a highly accurate travel time between two points without worrying about usage restrictions on services such as APIs, and can easily plan a delivery plan and the like.

<Modification>
The travel time providing system according to the present invention described above can be used in a system for determining the order of visiting three or more points on a map. The order of visiting three or more points on the map can be determined by solving the Traveling Salesman Problem (TSP), which is one of the typical problems of the combinatorial optimization problem.

  The TSP is a graph that defines a plurality of nodes (corresponding to points on the map), edges connecting the nodes, and costs added to each edge (for example, travel time). This is a problem of obtaining a route that minimizes the sum of the costs required to go through the node once and go round. In TSP, the cost for each edge is required as input data. However, since the number of edges is the number of combinations (nC2) for selecting any two points among a plurality of nodes, the number of edges increases as the number of traveling points increases. Will become a huge number. That is, if the travel time between two points is acquired as the cost of an edge from an external map information providing device (API server or the like), an enormous number of inquiries are required. End up.

  Therefore, the travel time between two points provided by the travel time providing system according to the present invention is acquired as the cost of the edge in the graph input in the TSP. As a result, the traveling order can be determined by solving the TSP without imposing a load on services such as API.

  The above-described embodiment can be implemented in various other forms, and various omissions, replacements, and changes can be made without departing from the scope of the invention. These embodiments and modifications thereof are included in the invention described in the claims and equivalents thereof, as long as they are included in the scope and gist of the invention.

  In addition, the method described in the above embodiment can be executed by a computer (software means) such as a magnetic disk (flexible disk, hard disk, etc.), an optical disk (CD-ROM, DVD, MO, etc.), It can be stored in a recording medium such as a semiconductor memory (ROM, RAM, flash memory, etc.), or transmitted and distributed via a communication medium. The program stored on the medium side also includes a setting program that configures in the computer software means (including not only the execution program but also a table and data structure) that is executed by the computer. A computer that realizes the server reads a program recorded on a recording medium, and constructs software means by a setting program according to circumstances, and executes the above-described processing by controlling the operation by the software means. The recording medium referred to in this specification is not limited to distribution, but includes a storage medium such as a magnetic disk or a semiconductor memory provided in a computer or a device connected via a network.

1, 2, 3 Travel time providing system, 100 user terminal, 110, 310, 410, 510 communication unit, 120 input unit, 130 output unit, 140, 320, 420, 520 storage unit, 150 sensor unit, 160, 330, 430 control unit, 200 map information providing device, 300, 400, 500 server device, 530 first estimation unit, 540 second estimation unit, 550 selection unit, 560 provision unit

Claims (10)

Set a plurality of nodes corresponding to points on the map, add an edge connecting the nodes to each of the plurality of nodes, and add a cost between any two of the plurality of nodes. A graph generation unit for generating a connected graph in which a route exists,
Wherein a plurality of nodes, and edges that the additional, and graph information storage unit for storing graphical information relating to the said connection graph composed of a cost which is added to each edge,
A cost calculator that calculates a predicted cost between two nodes selected from the plurality of nodes based on the graph information;
A cost acquisition unit for acquiring an actual measurement cost between two points on the map corresponding to the two selected nodes;
A learning data storage unit that stores the predicted cost calculated by the cost calculation unit and the actual measurement cost acquired by the cost acquisition unit as learning data;
Based on the learning data, a model generation unit that generates a learning model that learns the relationship between the predicted cost and the actual measurement cost;
The cost between any two points on the previous SL map, and cost providing unit to which the cost between nodes in the connection graph corresponding to the two points, provided on the basis of a learning model that the generated,
A server device comprising:   The server device according to claim 1, wherein the cost added to the edge is a movement time required to move between two points on the map corresponding to a node to which the edge is connected.   The server apparatus according to claim 1, wherein the model generation unit performs machine learning on the relationship by regression analysis.   The server device according to any one of claims 1 to 3, wherein the cost acquisition unit acquires the actual measurement cost between two points on the map from an external map information providing device.   The server device according to any one of claims 1 to 4, wherein the cost calculation unit calculates a cost between the two nodes by solving an optimization problem based on the graph information.   The server device according to any one of claims 1 to 5, wherein the connection graph is generated by applying a Delaunay triangulation method to the plurality of nodes. A server device that provides a cost between two points on a map,
A first estimation unit for estimating a cost between two points on the map based on a first learning model;
Based on a second learning model, a second estimation unit for estimating a cost between two points on the map;
A selection unit that selects a cost estimated by the first estimation unit or a cost estimated by the second estimation unit based on a prediction accuracy of the cost estimated by the first estimation unit and the second estimation unit;
A cost providing unit that provides the cost selected by the selection unit,
The first estimation unit includes:
A graph information storage unit that stores graph information about a connected graph composed of a plurality of nodes corresponding to points on the map, edges connecting the nodes, and costs added to the edges;
A cost calculator that calculates a predicted cost between two nodes selected from the plurality of nodes based on the graph information;
A first cost acquisition unit for acquiring an actual measurement cost between two points on the map corresponding to the two selected nodes;
A first learning data storage unit that stores the predicted cost calculated by the cost calculation unit and the actual measurement cost acquired by the first cost acquisition unit as learning data;
A first model generation unit that generates a first learning model that learns a relationship between the predicted cost and the actual measurement cost based on the learning data stored in the first learning data storage unit;
A first cost estimation unit that estimates a cost between any two points on the map based on the generated first learning model;
The second estimation unit is
A second cost acquisition unit for acquiring an actual measurement cost between two points on the map;
A second learning data storage unit that stores, as learning data, position information of two points on the map and the acquired actual measurement cost between the two points;
Based on the learning data stored in the second learning data storage unit, a second model generation unit that generates a second learning model that has learned the relationship between the position information of the two points on the map and the cost between the two points When,
A second cost estimation unit that estimates a cost between any two points on the map based on the generated second learning model;
Including a server device. Set a plurality of nodes corresponding to points on the map, add an edge connecting the nodes to each of the plurality of nodes, and add a cost between any two of the plurality of nodes. A graph generation unit for generating a connected graph in which a route exists,
Wherein a plurality of nodes, and edges that the additional, and graph information storage unit for storing graphical information relating to the said connection graph composed of a cost which is added to each edge,
A cost calculator that calculates a predicted cost between two nodes selected from the plurality of nodes based on the graph information;
A cost acquisition unit for acquiring an actual measurement cost between two points on the map corresponding to the two selected nodes;
A learning data storage unit that stores the predicted cost calculated by the cost calculation unit and the actual measurement cost acquired by the cost acquisition unit as learning data;
Based on the learning data, a model generation unit that generates a learning model that learns the relationship between the predicted cost and the actual measurement cost;
The cost between any two points on the previous SL map, and cost providing unit to which the cost between nodes in the connection graph corresponding to the two points, provided on the basis of a learning model that the generated,
A system comprising: A method for generating a learning model for providing a cost between two points on a map, the method being executed by a computer, the computer comprising a control unit and a storage unit,
The control unit sets a plurality of nodes corresponding to the points on the map, adds an edge connecting the nodes to each of the plurality of nodes, and adds a cost to the plurality of nodes. Generating a connected graph in which a path exists between any two points of
A step in which the control unit stores the plurality of nodes, and edges that the additional, and costs the added to each edge, the graph information on the connection graph consists in the storage unit,
The control unit calculating a predicted cost between two nodes selected from the plurality of nodes based on the graph information;
The control unit obtaining an actual measurement cost between two points on the map corresponding to the selected two nodes;
The control unit stores the calculated predicted cost and the acquired actual measurement cost as learning data in the storage unit;
Generating a learning model in which the control unit learns a relationship between the predicted cost and the actual measurement cost based on the learning data;
A method comprising: A program for causing a computer to execute the method according to claim 9 .