Delivery planning device, delivery planning method, and delivery planning program

Abstract

By incorporating driver insights, we generate more appropriate delivery plans. [Solution] The delivery planning device includes a vehicle probability model that has been trained to output a vehicle probability, which is the probability that two delivery destinations will be delivered by the same vehicle, when two delivery destinations are input, and a cost calculation unit that inputs a first delivery destination and a second delivery destination into the vehicle probability model, obtains the vehicle probability output from the vehicle probability model, and calculates the delivery cost between the first delivery destination and the second delivery destination based on the obtained vehicle probability.

Description

Technical Field

[0001] The present disclosure relates to a delivery planning device, a delivery planning method, and a delivery planning program.

Background Art

[0002] Techniques for creating a delivery plan for delivering goods using methods such as combinatorial optimization are known. Patent Document 1 discloses teacher data acquisition means for acquiring, as teacher data, a set consisting of delivery data and a delivery plan corresponding to the delivery data, grouping means for dividing the delivery data into a plurality of groups, and, for each pair of the groups divided by the grouping means, determining whether or not they are included in the same delivery route in the delivery plan included in the teacher data, and learning means for performing learning to create a delivery plan based on the determination result. An information processing apparatus is disclosed.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] However, in Patent Document 1, since the delivery data is grouped in units such as municipalities, even if the delivery destination straddles municipalities or the like, it may be more efficient to make a round trip with the same vehicle in terms of actual delivery efficiency, or even if there are delivery destinations in the same municipality or the like, it may be more efficient to make a round trip with a different vehicle. However, even when the driver knows such facts, such driver knowledge may not be reflected in the delivery plan, resulting in an inappropriate delivery plan.

[0005] Therefore, an object of the present disclosure is to provide a technique that can reflect driver knowledge and generate a more appropriate delivery plan.

Means for Solving the Problems

[0006] One aspect of the present disclosure provides a delivery planning device comprising: a vehicle probability model that, when two delivery destinations are input, outputs a vehicle probability which is the probability that the two delivery destinations will be delivered by the same vehicle; and a cost calculation unit that inputs a first delivery destination and a second delivery destination to the vehicle probability model, obtains the vehicle probability output from the vehicle probability model, and calculates the delivery cost between the first delivery destination and the second delivery destination based on the obtained vehicle probability.

[0007] One aspect of this disclosure provides a delivery planning method that inputs a first delivery destination and a second delivery destination into a vehicle probability model that has been trained to output a vehicle probability, which is the probability that two delivery destinations will be delivered by the same vehicle, when two delivery destinations are input; obtains the vehicle probability output from the vehicle probability model; and calculates the delivery cost between the first delivery destination and the second delivery destination based on the obtained vehicle probability.

[0008] One aspect of this disclosure provides a delivery planning program that causes a computer to input a first delivery destination and a second delivery destination into a vehicle probability model that has been trained to output a vehicle probability, which is the probability that two delivery destinations will be delivered by the same vehicle, when two delivery destinations are input, obtain the vehicle probability output from the vehicle probability model, and calculate the delivery cost between the first delivery destination and the second delivery destination based on the obtained vehicle probability.

[0009] These comprehensive or specific embodiments may be implemented as systems, devices, methods, integrated circuits, computer programs, or recording media, or as any combination of systems, devices, methods, integrated circuits, computer programs, and recording media. [Effects of the Invention]

[0010] According to this disclosure, it is possible to reflect the driver's insights and generate more appropriate delivery plans. [Brief explanation of the drawing]

[0011] [Figure 1] Block diagram showing an example configuration of the delivery planning device according to this embodiment. [Figure 2] This figure shows an example of the configuration of the performance information table according to this embodiment. [Figure 3] This figure shows an example of the configuration of the learning data table according to this embodiment. [Figure 4] A flowchart showing an example of a method for generating the vehicle probability model according to this embodiment. [Figure 5] A flowchart showing an example of a method for generating training data according to this embodiment. [Figure 6] This figure shows an example of the configuration of the vehicle probability model according to this embodiment. [Figure 7] A flowchart showing an example of the delivery plan calculation method according to this embodiment. [Figure 8] A diagram showing an example of the vehicle probability table according to this embodiment. [Figure 9] A diagram showing an example of a delivery cost table according to this embodiment. [Figure 10] A flowchart showing an example of a method for generating a second delivery plan according to this embodiment. [Figure 11] An explanatory diagram showing an example of three improvement methods according to this embodiment. [Figure 12] A diagram showing an example of the delivery plan map according to this embodiment. [Modes for carrying out the invention]

[0012] Hereinafter, embodiments of the present disclosure will be described in detail with appropriate reference to the drawings. However, detailed descriptions that are more than necessary may be omitted. For example, detailed descriptions of well-known matters and duplicate descriptions of substantially the same configurations may be omitted. This is to avoid making the following description unnecessarily redundant and to facilitate understanding by those skilled in the art. Note that the accompanying drawings and the following description are provided to enable those skilled in the art to fully understand the present disclosure, and are not intended to limit the subject matter described in the claims.

[0013] Even if the functions of one configuration shown in this embodiment are realized by two or more physical configurations, or the functions of two or more configurations are realized by, for example, one physical configuration, it does not matter.

[0014] (Background of the Present Disclosure) A method of formulating a delivery plan for goods by a vehicle is known, which determines a combination of delivery routes so that the total delivery cost assigned to roads is reduced. For example, the delivery cost is determined based on the characteristics (distance or time) of the road between two delivery destinations.

[0015] In the delivery plan formulated based on the above-described conventional delivery costs, the judgment of integration or separation of the delivery route by a skilled driver (that is, a veteran driver with considerable experience) is not reflected in the formulation of the delivery plan. For example, even if a skilled driver determines that it is more efficient to deliver a plurality of neighboring delivery destinations together, in the delivery plan based on the conventional delivery costs, it may be separated into a plurality of routes. For example, although the distance between two delivery destinations is short, due to circumstances not reflected in the delivery cost, such as crossing a prefectural border or a national border, or having to pass through a very narrow road, there may be cases where the two delivery destinations are to be delivered by separate vehicles.

[0016] [[ID=十七]] If the delivery plan is manually modified to reflect these judgments, a huge amount of modification work may be required, or the delivery plan may be modified to one with a significant reduction in delivery efficiency. Also, since it is difficult to formulate under what conditions a skilled driver integrates or separates delivery routes, it is also difficult to automatically modify the formulated delivery plan.

[0017] Therefore, in the present embodiment, an apparatus, method, computer program, etc. that can generate a more appropriate delivery plan by estimating the probability that groups of delivery destinations are delivered by the same vehicle based on the performance information of deliveries by a driver (e.g., a skilled driver) and reflecting it in the delivery cost will be described. Thereby, the judgment of the driver who has integrated or separated the delivery routes can be reflected in the delivery plan, and a more appropriate delivery plan can be created. Details will be described below.

[0018] (The present embodiment) <System configuration> FIG. 1 is a block diagram showing a configuration example of a delivery planning apparatus 10 according to the present embodiment.

[0019] The delivery planning apparatus 10 includes, as hardware, a processor 11, a memory 12, a storage 13, an input device 14, a display device 15, and a communication device 16.

[0020] The processor 11 realizes the functions of the delivery planning apparatus 10 of the present disclosure by executing a computer program (e.g., a delivery planning program) in cooperation with the memory 12. The processor may be read as a Central Processing Unit (CPU), a controller, an arithmetic circuit, a control circuit, or the like. Also, the processor 11 may include a Graphics Processing Unit (GPU) and / or a Neural network Processing Unit (NPU).

[0021] Memory 12 is composed of a volatile storage medium and / or a non-volatile storage medium, and stores computer programs and data, etc.

[0022] The storage 13 is composed of a non-volatile storage medium and stores computer programs and data. Examples of storage 13 include flash memory, solid state drives (SSDs), or hard disk drives (HDDs).

[0023] The input device 14 is a device that receives input from the user, and is, for example, a touch panel, touchpad, keyboard, mouse, and / or microphone.

[0024] The display device 15 is a device that displays a screen, and is composed of, for example, a liquid crystal display or an organic EL display.

[0025] The communication device 16 is a device that controls wired communication and / or wireless communication. The communication device 16 may support, for example, Ethernet®, Wi-Fi®, cellular communication (e.g., LTE, 4G, 5G), and / or Bluetooth®.

[0026] The delivery planning device 10 has the following functions: a performance information DB 21, a learning model DB 22, a vehicle information DB 23, a cargo information DB 24, a learning data generation unit 31, a learning execution unit 32, a cost calculation unit 33, a delivery plan calculation unit 34, and a delivery plan display unit 35.

[0027] The performance information DB21, the learning model DB22, the vehicle information DB23, and the cargo information DB24 may be configured in the storage 13 (or memory 12). The learning data generation unit 31, the learning execution unit 32, the cost calculation unit 33, the delivery plan calculation unit 34, and the delivery plan display unit 35 may be realized by the processor 11 cooperating with the memory 12 to execute a predetermined computer program.

[0028] The Performance Information DB21 is a database that holds Performance Information 100 (see Figure 2). Performance Information 100 is data that shows which drivers (e.g., experienced drivers) have visited which delivery destinations at what times in the past. Details of Performance Information 100 will be described later (see Figure 2). Performance Information 100 may be generated each time a driver makes a delivery and stored in the Performance Information DB21.

[0029] The learning model DB22 is a database that holds the vehicle probability model 50. The vehicle probability model 50 is a machine learning model that, given two delivery destinations as input, outputs the probability (hereinafter referred to as the vehicle probability) of using the same vehicle to deliver to those two destinations. The vehicle probability model 50 may be machine learning based on actual data 100. In other words, the vehicle probability can be said to represent the probability (predicted value) that a skilled driver would use the same vehicle to deliver to the two input destinations. Details of the vehicle probability model 50 will be described later (see Figure 6).

[0030] Vehicle Information DB23 is a database that holds information (vehicle information) about each vehicle available for delivery. For example, the vehicle information includes information such as the number of vehicles available for delivery at base S and the number of packages each vehicle can carry.

[0031] The package information DB24 is a database that holds information about packages to be delivered (package information). Package information includes information indicating which packages should be delivered, during what period, and to which destination. Package information may be entered from an external device (for example, a device that manages package delivery orders).

[0032] The training data generation unit 31 generates training data 200 to be used for machine learning of the vehicle probability model 50, and a training data table T200 that stores multiple training data sets 200. Details of the training data 200 and the training data table T200 will be described later (see Figure 3).

[0033] The learning execution unit 32 uses the training data 200 contained in the training data table T200 to perform machine learning on the vehicle probability model 50. Details of the learning method will be described later (see Figure 6).

[0034] The cost calculation unit 33 uses the vehicle probability model 50 to calculate the probability of the same vehicle being used at two delivery destinations. The cost calculation unit 33 then uses the calculated vehicle probability to calculate the delivery cost between the two delivery destinations and generates a delivery cost table T400 (see Figure 9). Details of this process will be described later (see Figure 7).

[0035] The delivery plan calculation unit 34 refers to the delivery cost table T400 to calculate how many vehicles and in what manner to make deliveries from base S to multiple delivery destinations Dn (where n is a natural number) in order to minimize delivery costs, and generates a first delivery plan. The delivery plan calculation unit 34 also modifies the first delivery plan to satisfy predetermined constraints and generates a second delivery plan. Details of this process will be described later (see Figures 7 to 11).

[0036] The delivery plan display unit 35 generates and displays a delivery plan map 500 that can distinguish between the first delivery plan and the second delivery plan and / or the differences between them. Details of this process will be described later (see Figure 12).

[0037] <Performance Information Table> Figure 2 shows an example of the configuration of the performance information table T100 according to this embodiment.

[0038] The performance information table T100 is a table that stores at least one performance information item 100. The performance information item has the following fields: performance information ID 101, driver identification information 102, delivery information 103, delivery destination latitude 104A, delivery destination longitude 104B, and delivery completion time 105.

[0039] Performance information ID 101 is an ID used to identify performance information 100. The performance information ID may be omitted. Driver identification information 102 is information used to identify the driver.

[0040] Delivery information 103 is information that indicates a set of delivery destinations corresponding to a single delivery plan. If delivery information 103 is the same, the actual delivery information 100 indicates the delivery destinations that each driver actually visited based on that single delivery plan.

[0041] The destination latitude 104A and destination longitude 104B indicate the location of the delivery destination, as determined by the driver identified in the driver identification information. In this embodiment, the location of the delivery destination is indicated by latitude and longitude. The destination latitude 104A and destination longitude 104B are sometimes collectively referred to as destination Dn (where n is a natural number).

[0042] The delivery completion time 105 indicates the time when the driver identified by the driver identification information completed the delivery to destination 104.

[0043] In the performance information table T100 shown in Figure 2, the performance information for performance information IDs 101 "J1" and "J2" is the same as the driver identification information 102 and the delivery information 103. Therefore, the "AAAA" indicated by the driver identification information 102 indicates that in one delivery (i.e., with one vehicle), the driver visited two locations: delivery destination D1 with delivery destination latitude 104A "35.00" and delivery destination longitude 104B "135.00", and delivery destination D2 with delivery destination latitude 104A "36.00" and delivery destination longitude 104B "136.00".

[0044] <Training Data Table> Figure 3 shows an example of the configuration of the learning data table T200 according to this embodiment.

[0045] The training data table T200 is a table that stores at least one training data 200. The training data 200 has the following items: training data ID 201, first delivery destination latitude 202A, first delivery destination longitude 202B, second delivery destination latitude 203A, second delivery destination longitude 203B, vehicle determination flag 204, and continuous delivery determination flag 205.

[0046] Training data ID 201 is an ID used to identify training data 200. Training data ID 201 may be omitted.

[0047] The first delivery destination latitude 202A and the first delivery destination longitude 202B are information indicating the latitude and longitude of the first delivery destination. Hereafter, the first delivery destination latitude 202A and the first delivery destination longitude 202B may be collectively referred to as the first delivery destination.

[0048] The second delivery destination latitude 203A and second delivery destination longitude 203B are information indicating the latitude and longitude of the second delivery destination. Hereafter, the second delivery destination latitude 203A and second delivery destination longitude 203B may be collectively referred to as the second delivery destination.

[0049] The vehicle determination flag 204 indicates whether the first and second delivery destinations on the same line were delivered by the same vehicle (one driver). If the first and second delivery destinations were delivered by the same vehicle (i.e., one driver), the vehicle determination flag 204 will be "1". If the first and second delivery destinations were not delivered by the same vehicle (i.e., one driver), the vehicle determination flag 204 will be "0". Note that the value of this flag is just an example, and any value is acceptable as long as it is distinguishable.

[0050] The consecutive delivery determination flag 205 indicates whether a single vehicle (i.e., a single driver) has made consecutive deliveries to the first delivery destination and then to the second delivery destination. If a single vehicle (i.e., a single driver) has made consecutive deliveries to the first delivery destination and then to the second delivery destination, the consecutive delivery determination flag 205 will be "1". If a single vehicle (i.e., a single driver) has not made consecutive deliveries to the first delivery destination and then to the second delivery destination, the consecutive delivery determination flag will be "0". Note that the value of this flag is just an example, and any value is acceptable as long as it is distinguishable.

[0051] <Method for generating the vehicle's probability model> Figure 4 is a flowchart showing an example of a method for generating the vehicle probability model 50 according to this embodiment.

[0052] The user specifies the locations and time period for which the learning will take place (S101). A location is a point from which delivery vehicles enter and exit. A delivery vehicle departs from a location, makes one or more delivery stops, and returns to the location.

[0053] The learning data generation unit 31 retrieves multiple performance information 100 that match the location and period specified in step S101 from the performance information DB 21 and stores them in the performance information table T100 (S102).

[0054] The learning data generation unit 31 extracts sets of performance information 100 with the same stool information 103 from the performance information table T100 (S103). In the explanation of Figure 4, the set of performance information extracted in step S103 is referred to as a performance information set. For example, in the performance information table T100 of Figure 4, performance information IDs "1", "2", and "3" with the same stool information 103 "2024111401" are extracted as a performance information set.

[0055] The learning data generation unit 31 sorts each piece of performance information from the performance information set extracted in step S103 by driver identification information 102 and delivery completion time 105 (S104).

[0056] The training data generation unit 31 extracts two pieces of performance information 100 from the performance information set sorted in step S104 (S105).

[0057] The training data generation unit 31 performs training data generation processing using the two performance data extracted in step S105 (S106). Details of the training data generation processing will be described later (see Figure 5), but this process generates the training data table T200.

[0058] The training data generation unit 31 determines whether or not it has extracted all two combination patterns from the performance information set (S107).

[0059] If there are two combination patterns that have not yet been extracted in the performance information set (S107: NO), the learning data generation unit 31 returns to step S105 and extracts two performance information items corresponding to the two combination patterns that have not yet been extracted.

[0060] If the training data generation unit 31 extracts all two combination patterns from the performance information set (S107: YES), it determines in the performance information table T100 whether all performance information sets with the same fare information 103 have been extracted (S108).

[0061] If the learning data generation unit 31 finds that there is a set of performance information that is the same as the unextracted travel information 103 in the performance information table T100 (S108: NO), it returns to step S103 and extracts the unextracted travel information 103 that is the same as the unextracted set of performance information.

[0062] If the learning data generation unit 31 extracts all sets of performance information in the performance information table T100 where the fare information 103 is the same (S108: YES), it proceeds to the next step S109.

[0063] The learning execution unit 32 generates the vehicle probability model 50 by performing machine learning on the vehicle probability model 50 using the training data 200 of the training data table T200 generated by the above-described process (S109). The machine learning method will be described later (see Figure 7).

[0064] The learning execution unit 32 stores the generated vehicle probability model 50 in the learning model DB 22 (S110). Then, this process ends.

[0065] Through the above processing, the delivery planning device 10 can generate the vehicle probability model 50.

[0066] <Method for generating training data> Figure 5 is a flowchart illustrating an example of a method for generating training data 200 according to this embodiment. Figure 5 corresponds to a detailed explanation of the training data generation process (S106) in Figure 4.

[0067] The training data generation unit 31 takes the two performance information extracted in step S105 of Figure 4, designates one as performance information A and the other as performance information B, designates the delivery destination of performance information A as delivery destination A, and the delivery destination of performance information B as delivery destination B (S201).

[0068] The learning data generation unit 31 determines whether the driver identification information 102 of performance information A and performance information B are the same (S202).

[0069] First, we will explain the case in step S202 when it is determined that the driver identification information of performance information A and performance information B are the same (S202: YES).

[0070] The training data generation unit 31 determines whether deliveries are made consecutively to destination A and then to destination B (S203). For example, if the delivery completion time 105 of actual data B is later than the delivery completion time 105 of actual data A, and there is no actual data with the same driver identification information as actual data A and actual data B that has a delivery completion time 105 between the delivery completion time 105 of actual data A and the delivery completion time 105 of actual data B, the training data generation unit 31 determines that deliveries are made consecutively to destination A and then to destination B. In other words, if the same driver has a record of delivering from destination A to destination B without any other destinations in between, the training data generation unit 31 determines that deliveries are made consecutively from destination A to destination B.

[0071] If it is determined that deliveries are made consecutively to delivery destination A and then to delivery destination B (S203: YES), the learning data generation unit 31 sets the delivery destination latitude 104A and delivery destination longitude 104B of actual information A as the first delivery destination latitude 202A and the first delivery destination longitude 202B, the delivery destination latitude 104A and delivery destination longitude 104B of actual information B as the second delivery destination latitude 203A and the second delivery destination longitude 203B, sets the vehicle determination flag 204 to "1", and sets the consecutive delivery determination flag 205 to "1", generates learning data 200, and stores it in the learning data table T200 (S204). For example, in the performance information table T100 shown in Figure 2, if performance information A is performance information 100 with performance information ID "J1" and performance information B is performance information 100 with performance information ID "J2", then training data 200 with training data ID 201 "K1" is generated in the training data table T200 shown in Figure 3. Then the process proceeds to step S206.

[0072] If it is determined that deliveries to destination A and destination B are not made consecutively (S203: NO), the learning data generation unit 31 sets the destination latitude 104A and destination longitude 104B of actual information A to the first destination latitude 202A and first destination longitude 202B, the destination latitude 104A and destination longitude 104B of actual information B to the second destination latitude 203A and second destination longitude 203B, sets the vehicle determination flag 204 to "1", and sets the consecutive delivery determination flag 205 to "0", generates learning data 200, and stores it in the learning data table T200 (S205). For example, if, as described later, the swap in step S206 results in the swapped performance information table T100 shown in Figure 2 being performance information 100 with performance information ID 101 "J2" and the swapped performance information B being performance information 100 with performance information ID "J1", then training data 200 with training data ID 201 "K2" is generated in the training data table T200 shown in Figure 3. Then the process proceeds to step S206.

[0073] The training data generation unit 31 swaps performance information A and performance information B (S206).

[0074] The training data generation unit 31 determines whether the training data 200 generated from the combination swapped in step S206 already exists in the training data table T200 (S207).

[0075] If the training data 200 generated from the combination swapped in step S206 does not yet exist in the training data table T200 (S207: NO), the training data generation unit 31 returns to step S203. This allows the continuous delivery determination flag 205 to be set to "1" in the processing from step S203 onward, even if the deliveries are not made consecutively in the order of destination A and destination B, but are made consecutively in the order of destination B and destination A.

[0076] If the training data 200 generated from the combination swapped in step S206 already exists in the training data table T200 (S207: YES), this process is completed and the process proceeds to step S107 in Figure 4.

[0077] Next, we will explain the case in step S202 when it is determined that the driver identification information 102 of performance information A and performance information B are different (S202: NO).

[0078] The learning data generation unit 31 sets the destination latitude 104A and destination longitude 104B of actual information A to the first destination latitude 202A and first destination longitude 202B, the destination latitude 104A and destination longitude 104B of actual information B to the second destination latitude 203A and second destination longitude 203B, sets the vehicle determination flag 204 to "0", and sets the continuous delivery determination flag 205 to "0", generates learning data 200, and stores it in the learning data table T200 (S211). For example, in the actual information table T100 shown in Figure 2, if actual information A is actual information 100 with actual information ID 101 "J1", and actual information B is actual information 100 with actual information ID 101 "J3", then learning data 200 with learning data ID "K3" is generated in the learning data table T200 shown in Figure 3.

[0079] The training data generation unit 31 swaps performance information A and performance information B (S212).

[0080] The learning data generation unit 31 sets the destination latitude 104A and destination longitude 104B of the swapped actual information A as the first destination latitude 202A and first destination longitude 202B, the destination latitude 104A and destination longitude 104B of the swapped actual information B as the second destination latitude 203A and second destination longitude 203B, sets the vehicle determination flag 204 to "0", and sets the continuous delivery determination flag 205 to "0", generates learning data 200, and stores it in the learning data table T200 (S212). For example, in the performance information table T100 shown in Figure 2, if the swapped performance information A is performance information 100 with performance information ID 101 "J3", and the swapped performance information B is performance information 100 with performance information ID 101 "J1", then training data 200 with training data ID 201 "K4" will be generated in the training data table T200 shown in Figure 3. This process is then completed, and the process proceeds to step S107 in Figure 4.

[0081] Through the above process, the delivery planning device 10 can generate a training data table T200 containing multiple training data 200.

[0082] <Probability model for the same vehicle> Figure 6 shows an example of the configuration of the vehicle probability model 50 according to this embodiment.

[0083] The vehicle probability model 50 may be configured as a neural network including an input layer 51, a hidden layer 52, and an output layer 53.

[0084] The input layer 51 has four neurons (nodes) that receive input from the first delivery destination latitude 202A, the first delivery destination longitude 202B, the second delivery destination latitude 203A, and the second delivery destination longitude 203B, respectively, as shown in Figure 6. However, these are just examples, and in addition to the four neurons mentioned above, the input layer 51 may also have neurons that receive input from the number of delivery vehicles, the number of deliveries, the distance between delivery destinations, the time of delivery, and at least one of the delivery destinations to be delivered simultaneously.

[0085] The hidden layer 52 consists of two layers, each containing, for example, 16 neurons. The ReLU function is used as the activation function. However, these are just examples, and the number of neurons per layer and the number of layers in the hidden layer 52 can be any number.

[0086] The output layer 53 consists of one neuron. The sigmoid function is used as the activation function. The output layer 53 outputs values ​​between 0 and 1. However, this is just an example, and the number of neurons in the output layer 53 and the range of values ​​it outputs can be any number.

[0087] The vehicle probability model 50 takes the latitude 202A, longitude 202B, latitude 203A, and longitude 203B of the first delivery destination, which are stored in the training data table T200, as input data, and performs machine learning (e.g., deep learning) using the vehicle determination flag 204 (0 or 1) of the training data 200 as training data. As a result, when the vehicle probability model 50 takes the latitude and longitude of the first delivery destination and the second delivery destination (latitude and longitude) as input, it can output the probability that the first and second delivery destinations will be delivered by the same vehicle (or the same driver) (i.e., the same vehicle probability). It can also be said that by calculating (1 - same vehicle probability), the vehicle probability model 50 can output the probability that the first and second delivery destinations will be delivered by different vehicles (or different drivers) (i.e., different vehicle probability).

[0088] Furthermore, the vehicle probability model 50 may be trained using the first delivery latitude 202A, the first delivery longitude 202B, the second delivery latitude 203A, and the second delivery longitude 203B from the training data 200 stored in the training data table T200 as input data, and the continuous delivery determination flag 205 (0 or 1) from the training data 200 as training data. As a result, the vehicle probability model 50, upon input of the first delivery destination (latitude and longitude) and the second delivery destination (latitude and longitude), can output the probability (vehicle probability) that the first and second delivery destinations are delivered consecutively by the same vehicle (or one driver) in that order.

[0089] Furthermore, the vehicle probability model 50 may be generated by a method different from the machine learning described above. For example, the vehicle probability model 50 may be a classification algorithm based on logistic regression.

[0090] <Delivery Plan Calculation Method> Figure 7 is a flowchart showing an example of the delivery plan calculation method according to this embodiment. Figure 8 is a diagram showing an example of the vehicle probability table T300 according to this embodiment. Figure 9 is a diagram showing an example of the delivery cost table T400 according to this embodiment. Next, the delivery plan calculation method will be explained with reference to Figures 7 to 9.

[0091] The delivery plan calculation unit 34 obtains the vehicle probability model 50 from the learning model DB22 (S301).

[0092] The delivery plan calculation unit 34 acquires delivery destination information for which a delivery plan is to be created (S302). The delivery destination information includes at least one base S and multiple delivery destinations Dn.

[0093] The delivery plan calculation unit 34 extracts two delivery destinations from the delivery destination information (S303).

[0094] The delivery plan calculation unit 34 determines the delivery cost between the two delivery destinations extracted in step S303 (S304). For example, the delivery plan calculation unit 34 generates a route between the two delivery destinations using a road network and determines the delivery cost according to the distance of that route. Note that the delivery cost may be determined not only according to the distance of the route, but also according to the predicted travel time of the route, etc. Furthermore, this is just one example, and the delivery cost may be determined by various known methods.

[0095] The delivery plan calculation unit 34 inputs the two delivery destinations extracted in step S303 into the vehicle probability model 50, obtains the vehicle probability output from the vehicle probability model 50, and stores it in the vehicle probability table T300 (S305).

[0096] For example, suppose the two delivery destinations extracted in step S303 are D1 and D2, and the vehicle probability between these delivery destinations D1 and D2 output from the vehicle probability model 50 is 0.97. In this case, the delivery plan calculation unit 34 enters 0.97 in the intersection column of delivery destinations D1 and D2, as shown in the vehicle probability table T300 in Figure 8.

[0097] For example, suppose the two delivery destinations extracted in step S303 are D1 and D3, and the vehicle probability between these delivery destinations D1 and D3 output from the vehicle probability model 50 is 0.01. In this case, the delivery plan calculation unit 34 enters 0.01 in the intersection column of delivery destinations D1 and D3, as shown in the vehicle probability table T300 in Figure 8. The same applies to the other intersection columns.

[0098] The delivery plan calculation unit 34 determines whether the probability of the vehicle is above a predetermined first threshold (S306).

[0099] If the probability of the vehicle being the same is greater than or equal to the first threshold (S306: YES), the delivery plan calculation unit 34 corrects the delivery cost determined in step S304 to the minimum (S307).

[0100] For example, if the first threshold is 0.9, and the probability of the same vehicle being delivered between delivery destinations D1 and D2 obtained in step S305 is 0.97, then the probability of the same vehicle being delivered (0.97) is greater than or equal to the first threshold (0.9). In this case, if the delivery plan calculation unit 34 determines that the delivery cost between delivery destinations D1 and D2 determined in step S304 is 1.2, it multiplies 1.2 by a predetermined minimum coefficient of 0.0001 to calculate the corrected delivery cost. Then the process proceeds to step S310.

[0101] If the probability of the vehicle being the same is less than the first threshold (S306: NO), the process proceeds to step S308.

[0102] The delivery plan calculation unit 34 determines whether the vehicle probability is less than or equal to a predetermined second threshold which is smaller than the first threshold (S308).

[0103] If the probability of the vehicle being the same is below the second threshold (S308: YES), the delivery plan calculation unit 34 corrects the delivery cost determined in step S304 to its maximum (S309).

[0104] For example, if the second threshold is set to 0.1, and the probability of the same vehicle being used between delivery destinations D1 and D3 obtained in step S305 is 0.01, then the probability of the same vehicle being used (0.01) is less than or equal to the second threshold (0.1). In this case, if the delivery plan calculation unit 34 assumes that the delivery cost between delivery destinations D1 and D3 determined in step S304 was 2.0, it multiplies 2.0 by a predetermined maximum coefficient of 10000 to calculate the corrected delivery cost. Then the process proceeds to step S310.

[0105] If the probability of the vehicle being the same is greater than the second threshold (S308: NO), the process proceeds to step S310.

[0106] The delivery plan calculation unit 34 inputs the delivery cost between the two delivery destinations, calculated or corrected by the above process, into the intersection column of the delivery cost table T400 (S310).

[0107] For example, if the minimum adjustment for delivery costs between delivery destinations D1 and D2 is "1.2 × 0.0001", the delivery plan calculation unit 34 inputs the calculation result "1.2 × 0.0001" into the intersection column of delivery destinations D1 and D2 in the delivery cost table T400 in Figure 9.

[0108] For example, if the maximum adjusted delivery cost between delivery destinations D1 and D3 is "2.0 × 10000", the delivery plan calculation unit 34 inputs the calculation result "2.0 × 10000" into the intersection column of delivery destinations D1 and D3 in the delivery cost table T400 in Figure 9.

[0109] For example, if the delivery cost between delivery destinations D2 and D3 is 2.3, and no correction is needed for delivery destinations D2 and D3 in step S307 or S309, the delivery plan calculation unit 34 enters "2.3" in the intersection column of delivery destinations D2 and D3 in the delivery cost table T400 in Figure 9.

[0110] The delivery plan calculation unit 34 determines whether or not it has extracted all possible combination patterns of two delivery destinations included in the delivery destination information (S311).

[0111] If the delivery plan calculation unit 34 finds that there are still two delivery destination combination patterns that have not yet been extracted in the delivery destination information (S311: NO), it returns to step S303 and extracts the two delivery destination combinations that have not yet been extracted from the delivery destination information.

[0112] If the delivery plan calculation unit 34 extracts all possible combinations of two delivery destinations from the delivery destination information (S311:YES), it generates a first delivery plan based on the generated delivery cost table T400 (S312). For example, the delivery plan calculation unit 34 calculates the number of vehicles and the delivery destinations that each vehicle will visit to minimize the overall delivery cost, based on the delivery cost table T400, and generates a first delivery plan.

[0113] The delivery plan calculation unit 34 sets the constraints that the delivery plan must adhere to (S313). These constraints may be entered by the user or may be set in advance.

[0114] The delivery plan calculation unit 34, while referring to the delivery cost table T400, modifies the first delivery plan to minimize the overall delivery cost while satisfying the constraints set in step S313, and generates a second delivery plan (S314). Then this process ends.

[0115] Through the above process, a second delivery plan can be generated that attempts to use the same vehicle as much as possible for deliveries to destinations that the driver has previously visited with the same vehicle, while also satisfying the constraints as much as possible.

[0116] <Details on how to generate the second delivery plan> Figure 10 is a flowchart illustrating an example of a method for generating a second delivery plan according to this embodiment. Figure 10 corresponds to a detailed explanation of step S311 in Figure 9.

[0117] The delivery plan calculation unit 34 determines the first delivery plan generated in step S312 of Figure 9 as the initial solution (S401).

[0118] The delivery plan calculation unit 34, based on the current solution, the delivery plan, sequentially applies the three improvement methods shown in Figure 11 to all combinations of delivery destinations (S402). Here, the current solution is either the initial solution (first delivery plan) determined in step S401, or the improved solution obtained in step S406, which will be described later. Details of the three improvement methods will be described later (see Figure 11).

[0119] The delivery plan calculation unit 34 determines whether the constraint violation has worsened as a result of applying the three types of improvement methods to all combinations of delivery destinations (S403).

[0120] If the delivery plan calculation unit 34 determines that the constraint violation has worsened (S403: YES), it outputs the current solution as the second delivery plan (S407) and terminates this process.

[0121] If the delivery plan calculation unit 34 determines that the constraint violation has not worsened (S403: NO), it determines whether the constraint violation has improved (S404).

[0122] If the delivery plan calculation unit 34 determines that the constraint violation has been resolved (S404: YES), it sets the resolved solution as the current solution (S406) and returns the process to step S402.

[0123] If the delivery plan calculation unit 34 determines that the constraint violation has not been improved (S404: NO), it determines whether the delivery cost has been improved (S405).

[0124] If the delivery plan calculation unit 34 determines that the delivery cost has improved (S405: YES), it sets the improved solution as the current solution (S406) and returns the process to step S402.

[0125] If the delivery plan calculation unit 34 determines that the delivery cost has not improved (S405: NO), it outputs the current solution as the second delivery plan (S407) and terminates this process.

[0126] For example, in order to reduce the sense of unfairness among drivers, the following constraints (A1) to (A3) may be set. Note that it is not necessary for all of the following constraints (A1) to (A3) to be set; at least one of them may be set, or none of them may be set. (A1) The difference in the number of delivery destinations assigned to each vehicle should be less than or equal to n (where n is a non-negative integer). (A2) To equalize delivery costs, the delivery cost of each vehicle should be kept within x% of the average (x is a real number greater than or equal to 0). (A3) Ensure that each vehicle does not exceed the pre-set maximum and / or minimum operating time.

[0127] The second delivery plan calculated by the process shown in Figure 10 is a solution that does not violate the constraints described above. However, if such a solution cannot be found, the second delivery plan that minimizes the constraint violations may be calculated.

[0128] Specifically, in step S403, if the constraint violation in the candidate solution has worsened, it is considered that no improvement has been made regardless of the delivery cost; however, if the constraint violation has been resolved, it is considered that no improvement has been made regardless of the delivery cost.

[0129] Furthermore, if the constraint violation has neither been improved nor worsened, in step S405, if the delivery cost has been improved, the solution is considered to have been improved. This "constraint violation has neither been improved nor worsened" means that there is no room for improvement because the constraint violation has already been eliminated, or that although the constraint violation remains, there is no room for further improvement. Therefore, after this judgment, a solution with no constraint violations, or a solution with the smallest constraint violation, is output as the second delivery plan.

[0130] <Details of the 3 types of improvement methods> Figure 11 is an explanatory diagram showing an example of three types of improvement methods used in step S402 of Figure 10. The three types of improvement methods are, for example, substitution, exchange, and transfer. Substitution means changing the order of delivery destinations. Exchange means changing delivery destinations between multiple delivery routes. Transfer means moving a delivery destination to a different delivery route. Specific examples of substitution, exchange, and transfer are shown below in (A), (B), and (C).

[0131] (A) Substitution In the pre-training delivery plan, before replacement, one vehicle delivers in the order of base S → destination D1 → D2 → D3 → D4. In this pre-training and pre-replacement delivery plan, the delivery cost (in other words, distance) is "10 + 10 + 10 = 30" corresponding to the same first vehicle.

[0132] On the other hand, before learning and after replacement, the same vehicle is delivered in the order of base S → delivery destination D1 → D3 → D2 → D4. In this delivery plan before learning and after replacement, the delivery cost is "15 + 10 + 15 = 40" corresponding to the same vehicle. Therefore, if replacement is performed before learning, the delivery cost increases and the delivery plan does not improve. Note that "before learning" can mean before learning begins or immediately after learning begins. This is also true in (B) and (C) below.

[0133] Furthermore, in the post-training delivery plan, before replacement, one vehicle delivers in the order of base S → destination D1 → D2 → D3 → D4. In this post-training and pre-replacement delivery plan, the delivery cost (in other words, distance) is the same as before training, "10 + 10 + 10 = 30" corresponding to the same first vehicle.

[0134] On the other hand, after learning and replacement, the same vehicle will be delivered in the order of base S → delivery destination D1 → D3 → D2 → D4. In this post-learning and replacement delivery plan, the delivery cost is "5 + 10 + 5 = 20" corresponding to the same first vehicle. Therefore, replacing after learning reduces the delivery cost and improves the delivery plan. Note that "after learning" means that a considerable amount of learning has been completed. This is also true in (B) and (C) below.

[0135] (B) Exchange In the pre-learning delivery plan, before the exchange, the first vehicle makes deliveries in the order of base S → destinations D1 → D2 → D3 → D4, and the second vehicle makes deliveries in the order of base S → destinations D5 → D6 → D7 → D8. In this pre-learning and pre-exchange delivery plan, the delivery cost (in other words, distance) is 40, which is the sum of "10 + 10 + 0 = 20" corresponding to the same first vehicle and "10 + 10 + 0 = 20" corresponding to the same second delivery vehicle.

[0136] On the other hand, before learning and after the exchange, the same first vehicle is used for deliveries in the order of base S → destinations D1 → D6 → D3 → D4, and the same second vehicle is used for deliveries in the order of base S → destinations D5 → D2 → D7 → D8. In this delivery plan before learning and after the exchange, the delivery cost (in other words, distance) is 45, which is the sum of "15 + 10 + 0 = 25" corresponding to the first vehicle and "10 + 10 + 0 = 20" corresponding to the second vehicle. Therefore, if the exchange occurs before learning, the overall delivery cost will increase, and the delivery plan will not improve.

[0137] Furthermore, in the post-learning delivery plan, before the exchange, the first vehicle is scheduled to deliver in the order of base S → destinations D1 → D2 → D3 → D4, and the second vehicle is scheduled to deliver in the order of base S → destinations D5 → D6 → D7 → D8. In this post-learning and pre-exchange delivery plan, the delivery cost is the same as before learning, which is the sum of "10+10+0=20" for the first vehicle and "10+10+0=20" for the second vehicle, resulting in "40".

[0138] On the other hand, after learning and exchange, the same first vehicle will be delivered in the order of base S → destination D1 → D6 → D3 → D4, and the same second vehicle will be delivered in the order of base S → destination D5 → D2 → D7 → D8. In this post-learning and post-exchange delivery plan, the delivery cost is 35, which is the sum of "5 + 10 + 0 = 15" corresponding to the same first vehicle and "10 + 10 + 0 = 20" corresponding to the same second vehicle. Therefore, if an exchange occurs after learning, the overall delivery cost after learning decreases, and the delivery plan is improved.

[0139] (C) Transfer In the pre-learning delivery plan, before the transfer, the first vehicle will deliver in the order of base S → destinations D1 → D2 → D3 → D4, and the second vehicle will deliver in the order of base S → destinations D5 → D6 → D7 → D8, similar to the exchange case. In this pre-learning and pre-transfer delivery plan, the delivery cost (in other words, distance) is 40, which is the sum of "10 + 10 + 0 = 20" for the first vehicle and "10 + 10 + 0 = 20" for the second vehicle.

[0140] On the other hand, before learning and after the transfer, the same first vehicle makes deliveries in the order of base S → destinations D1 → D3 → D4, and the same second vehicle makes deliveries in the order of base S → destinations D5 → D6 → D2 → D7 → D8. In this delivery plan before learning and after the transfer, the delivery cost (in other words, distance) is 55, which is the sum of "25 + 0 = 25" corresponding to the first vehicle and "10 + 10 + 10 + 0 = 30" corresponding to the second vehicle. Therefore, if the transfer occurs before learning, the overall delivery cost will increase, and the delivery plan will not improve.

[0141] Furthermore, in the post-learning delivery plan, before the transfer, the first vehicle is scheduled to deliver in the order of base S → destinations D1 → D2 → D3 → D4, and the second vehicle is scheduled to deliver in the order of base S → destinations D5 → D6 → D7 → D8. In this post-learning and pre-transfer delivery plan, the delivery cost is "40", which is the same as before the learning, obtained by adding "10 + 10 + 0 = 20" corresponding to the first vehicle and "10 + 10 + 0 = 20" corresponding to the second vehicle.

[0142] On the other hand, after learning and transfer, the same first vehicle will be delivered in the order of base S → destination D1 → D3 → D4, and the same second vehicle will be delivered in the order of base S → destination D5 → D6 → D2 → D7 → D8, just as before learning. In this post-learning and post-transfer delivery plan, the delivery cost is "35", which is the sum of "5+0=5" corresponding to the same first vehicle and "10+10+10+0=30" corresponding to the same second vehicle. Therefore, if the transfer occurs after learning, the overall delivery cost after learning will decrease, and the delivery plan will be improved.

[0143] <Delivery Plan Map> Figure 12 shows an example of the display of the delivery plan map 500 according to this embodiment.

[0144] For example, suppose the probability of the same vehicle being delivered between delivery destinations D1 and D2, between delivery destinations D2 and D3, between delivery destinations D3 and D6, and between delivery destinations D4 and D5 is above the first threshold (solid line in Figure 12), and the probability of the same vehicle being delivered between delivery destinations D1 and D4, between delivery destinations D3 and D4, and between delivery destinations D4 and D6 is below the second threshold (dotted line in Figure 12).

[0145] The delivery plan calculation unit 34 then generates a first delivery plan that minimizes the overall delivery cost by delivering destinations D1, D2, D3, and D6 with one vehicle and delivering destinations D4 and D5 with another vehicle.

[0146] Furthermore, suppose the delivery plan calculation unit 34 modifies the first delivery plan in step S314 to satisfy the constraints set in step S313 of Figure 7, and generates a second delivery plan. For example, suppose a constraint is set that the number of delivery destinations for each vehicle should be equal, and the delivery plan calculation unit 34 modifies the first delivery plan to satisfy that constraint by having one vehicle deliver to destinations D1, D2, and D3, and another vehicle deliver to destinations D4, D5, and D6, and generates a second delivery plan.

[0147] In this case, the first delivery plan, which was generated to minimize the overall delivery cost, differs from the second delivery plan, which also takes constraints into account.

[0148] As shown in Figure 12, the delivery plan display unit 35 generates a delivery plan map 500 that allows for visual distinction between the points of agreement and / or differences between the first delivery plan and the second delivery plan, and displays it on the screen of the display device 15.

[0149] For example, the delivery plan display unit 35 displays a delivery plan map 500 that distinguishes between the following (B1) to (B4). (B1) Two delivery destinations where the same vehicle was predicted in the first delivery plan (i.e., the probability of the same vehicle being above the first threshold), and the same vehicle was also predicted in the second delivery plan. (B2) Two delivery destinations where the first delivery plan predicted the same vehicle (i.e., the probability of it being the same vehicle was above the first threshold), but the second delivery plan predicted different vehicles. (B3) Two delivery destinations where the first delivery plan predicted different vehicles (i.e., the probability of the same vehicle being below the second threshold), and where the second delivery plan also predicted different vehicles. (B4) Two delivery destinations where the first delivery plan predicted different vehicles (i.e., the probability of the same vehicle being below the second threshold), but the second delivery plan predicted the same vehicle.

[0150] For example, in Figure 12, the spaces between delivery destinations D1, D2, and D3, and between delivery destinations D4 and D5, corresponding to (B1), are shown with thick solid lines; the space between delivery destinations D3 and D6, corresponding to (B2), is shown with thick double lines; the spaces between delivery destinations D1 and D4, and between delivery destinations D3 and D4, corresponding to (B3), are shown with dotted lines; and the space between delivery destinations D4 and D6, corresponding to (B4), is shown with double dotted lines.

[0151] Note that the above display method is just an example, and (B1) to (B4) may be distinguished by the color of the lines, the pattern of the lines, and / or the thickness of the lines. Furthermore, the thickness or shading of the lines connecting the two delivery destinations may be changed according to the probability of the same vehicle being present between the two delivery destinations. For example, the closer the probability of the same vehicle is to 0 or 1, the thicker the line may be, and the further it is from 0 or 1, the thinner the line may be.

[0152] The delivery plan display unit 35 may also display the likelihood of the same vehicle being realized, calculated by (B1) / ((B1) + (B2)), together with the delivery plan map 500. The likelihood of the same vehicle being realized indicates the percentage of two delivery destinations that were predicted to have the same vehicle in the first delivery plan and that also had the same vehicle in the second delivery plan.

[0153] The delivery plan display unit 35 may also display the degree of different vehicle realization, calculated by (B3) / ((B3) + (B4)), together with the delivery plan map 500. The degree of different vehicle realization indicates the percentage of two delivery destinations that were predicted to have different vehicles in the first delivery plan, and for which the second delivery plan also has different vehicles.

[0154] This allows users to see how well the predictions in the first delivery plan were realized in the second delivery plan.

[0155] (Other variations) In the embodiment described above, delivery costs were determined based on the distance or time between the two delivery destinations. However, delivery costs may also be determined based on other characteristics of the road between the two delivery destinations. Such characteristics could include, for example, the presence or absence of slopes or curves, the width of the road, and the presence or absence of traffic restrictions such as one-way streets. Even when delivery costs are set based on these characteristics, the same principles as in the embodiment described above can be applied to reflect the driver's judgment in the delivery plan and create a more appropriate delivery plan.

[0156] In the above-described embodiment, the driver's judgment was reflected in the delivery plan by adjusting the delivery cost to be either extremely high or extremely low. However, if the selection of a road based on the vehicle probability is to be set more strictly, for example, a constraint may be imposed such that if the vehicle probability is above a first threshold, that road is always selected, and if the vehicle probability is below a second threshold, that road is always not selected. However, when formulating a delivery plan, there are cases where it is necessary to select a road with a high delivery cost or not be able to select a road with a low delivery cost in order to satisfy other constraints such as one-way streets. Therefore, in the above-described embodiment, instead of restricting the selection of roads, the value of the delivery cost is adjusted to accommodate such constraints.

[0157] In the above-described embodiment, the driver's judgment was reflected in the delivery plan by correcting the delivery cost to a maximum or minimum value. As a result, in the delivery plan, roads corresponding to the maximum corrected delivery cost are very unlikely to be selected, and roads corresponding to the minimum corrected delivery cost are very likely to be selected. However, if it is only necessary to slightly change the likelihood of these roads being selected or selected, the delivery cost may be corrected with a coefficient closer to 1 than the extreme values ​​in the above-described embodiment. In addition, in the above-described embodiment, multiple thresholds may be set to compare with the vehicle probability, and the value used to correct the delivery cost may be increased or decreased in stages. For example, if the vehicle probability is very high (e.g., 0.9 or higher), the delivery cost may be corrected to a minimum value, and if the vehicle probability is moderately high (e.g., 0.6 or higher and less than 0.9), the delivery cost may be corrected with a coefficient closer to 1 than the minimum value (e.g., 0.1). Similarly, if the probability of the vehicle being present is very low (e.g., 0.1 or less), the delivery cost may be corrected to its maximum value, and if the probability of the vehicle being present is moderately low (e.g., greater than 0.1 and 0.4 or less), the delivery cost may be corrected to a coefficient closer to 1 than the maximum value (e.g., 10). This makes it possible to create a delivery plan that more precisely reflects the probability of the vehicle being present in the correction of the delivery cost. In addition, in the above embodiment, instead of using multiple thresholds such as a first threshold and a second threshold as the threshold for comparison with the probability of the vehicle being present, one threshold may be used. In this case, for example, if the probability of the vehicle being present is above the threshold, the delivery cost may be corrected to its minimum, and if the probability of the vehicle being present is below the threshold, the delivery cost may be corrected to its maximum.

[0158] In the embodiment described above, the vehicle probability was compared with a threshold, and a predetermined coefficient was multiplied by the delivery cost according to the result. However, the coefficient could also be calculated from the vehicle probability and multiplied by that coefficient. For example, if the coefficient is configured to reflect the reciprocal of the vehicle probability, the delivery cost will be adjusted to a smaller value when the vehicle probability is high, and to a larger value when the vehicle probability is low. This makes it possible to create a delivery plan that is more likely to include roads with a high vehicle probability and less likely to include roads with a low vehicle probability, similar to the embodiment described above.

[0159] In the embodiment described above, the vehicle probability estimated by the vehicle probability model was used to create a delivery plan. However, the vehicle probability may be used for purposes other than creating a delivery plan. For example, it could be used to estimate the number of vehicles required for delivery.

[0160] (Summary of this embodiment) The following technologies are disclosed based on the above description of this embodiment.

[0161] <Technology 1> The delivery planning device (10) according to this embodiment includes a vehicle probability model (50) that has been trained to output a vehicle probability, which is the probability that two delivery destinations will be delivered by the same vehicle, when two delivery destinations are input, and a cost calculation unit (33) that inputs a first delivery destination and a second delivery destination to the vehicle probability model, obtains the vehicle probability output from the vehicle probability model, and calculates the delivery cost between the first delivery destination and the second delivery destination based on the obtained vehicle probability. This allows us to obtain the vehicle probability between the first and second delivery destinations using the vehicle probability model, and to calculate the delivery cost between the first and second delivery destinations.

[0162] <Technology 2> In the delivery planning device described in Technology 1, the cost calculation unit corrects the delivery cost between the first and second delivery destinations to be smaller if the acquired vehicle probability is greater than or equal to a predetermined first threshold, and corrects the delivery cost between the first and second delivery destinations to be larger if the acquired vehicle probability is less than or equal to a second threshold which is smaller than the first threshold. This makes it possible to reduce delivery costs between the first and second delivery destinations where the probability of the same vehicle being present is relatively high, and to increase delivery costs between the first and second delivery destinations where the probability of the same vehicle being present is relatively low.

[0163] <Technology 3> In the delivery planning device described in Technology 1 or 2, the cost calculation unit further comprises a delivery planning calculation unit (34) that calculates the delivery costs between two different delivery destinations to generate a delivery cost table (T400), and based on the delivery cost table, calculates one or more delivery destinations to be delivered by the same vehicle and one or more delivery destinations to be delivered by different vehicles, and generates a first delivery plan. Based on delivery costs that take into account the probability of using the same vehicle, it is possible to calculate one or more delivery destinations to be delivered by the same vehicle and one or more delivery destinations to be delivered by different vehicles, thereby generating a first delivery plan.

[0164] <Technology 4> In the delivery planning device described in Technical 3, the delivery plan calculation unit modifies the first delivery plan based on predetermined constraints and generates a second delivery plan. This allows for the generation of a second delivery plan that takes into account the constraints in addition to the vehicle probability.

[0165] <Technology 5> The delivery planning device described in Technical 4 further includes a delivery planning display unit (35) that displays a delivery planning map (500) capable of distinguishing points of agreement and / or differences between the first delivery planning and the second delivery planning. This allows users viewing the delivery plan map to recognize the similarities and / or differences between the first and second delivery plans.

[0166] <Technology 6> The delivery planning device described in any one of technologies 1 to 5 further comprises a learning execution unit (32) that performs machine learning of the vehicle probability model using learning data (200) which includes at least information indicating whether the two delivery destinations have been delivered by the same vehicle in the past. This allows the vehicle probability model to be trained based on past delivery performance data (100).

[0167] <Technology 7> The delivery planning method according to this embodiment involves inputting a first delivery destination and a second delivery destination into a vehicle probability model that has been trained to output a vehicle probability, which is the probability that two delivery destinations will be delivered by the same vehicle, when two delivery destinations are input. The vehicle probability output from the vehicle probability model is then obtained, and the delivery cost between the first and second delivery destinations is calculated based on the obtained vehicle probability. This allows us to obtain the vehicle probability between the first and second delivery destinations using the vehicle probability model, and to calculate the delivery cost between the first and second delivery destinations.

[0168] <Technology 8> The delivery planning program according to this embodiment causes a computer (e.g., delivery planning device 10) to input a first delivery destination and a second delivery destination into a vehicle probability model that has been trained to output a vehicle probability, which is the probability that the two delivery destinations will be delivered by the same vehicle, when two delivery destinations are input, to obtain the vehicle probability output from the vehicle probability model, and to calculate the delivery cost between the first delivery destination and the second delivery destination based on the obtained vehicle probability. This allows us to obtain the vehicle probability between the first and second delivery destinations using the vehicle probability model, and to calculate the delivery cost between the first and second delivery destinations.

[0169] While embodiments have been described above with reference to the attached drawings, this disclosure is not limited to such examples. It is clear to those skilled in the art that various modifications, alterations, substitutions, additions, deletions, and equivalents can be conceived within the scope of the claims, and these are also understood to fall within the technical scope of this disclosure. Furthermore, the components of the embodiments described above can be combined in any way without departing from the spirit of the invention. [Industrial applicability]

[0170] The technology disclosed herein is useful for developing more optimal delivery plans. [Explanation of Symbols]

[0171] 10 Delivery planning device 11 processors 12 memory 13 Storage 14 Input devices 15 Display device 16. Communication equipment 31. Training Data Generation Unit 32 Learning Execution Unit 33. Cost Calculation Section 34 Delivery Planning Department 35 Delivery Plan Display Section 50 The same vehicle probability model 51 Input Layer 52 Middle Class 53 Output Layer 100 Achievement Information 101 Performance Information ID 102 Driver Identification Information 103 Flight information 104 Delivery address 104A Delivery latitude 104B Delivery destination longitude 105 Delivery completion time 200 training data points 201 Training Data ID 202A First delivery destination latitude 202B First delivery destination longitude 203A Second delivery latitude 203B Second delivery destination longitude 204 Same vehicle identification flag 205 Continuous delivery determination flag 500 Delivery Plan Map T100 Performance Information Table T200 Training Data Table T300 Vehicle Probability Table T400 Shipping Cost Table

Claims

1. A vehicle probability model that has been trained to output the probability that two delivery destinations will be delivered by the same vehicle when two delivery destinations are entered, The system includes a cost calculation unit that inputs a first delivery destination and a second delivery destination into the vehicle probability model, obtains the vehicle probability output from the vehicle probability model, and calculates the delivery cost between the first delivery destination and the second delivery destination based on the obtained vehicle probability. Delivery planning device.

2. The aforementioned cost calculation unit, If the acquired probability of the vehicle is equal to or greater than a predetermined first threshold, the delivery cost between the first delivery destination and the second delivery destination is corrected to reduce it. If the acquired probability of the vehicle is less than or equal to a second threshold which is smaller than the first threshold, the delivery cost between the first and second delivery destinations is corrected to be larger. The delivery planning device according to claim 1.

3. The cost calculation unit calculates the delivery costs between two different delivery destinations and generates a delivery cost table. The system further includes a delivery plan calculation unit that calculates one or more delivery destinations to be delivered by the same vehicle and one or more delivery destinations to be delivered by different vehicles, based on the aforementioned delivery cost table, and generates a first delivery plan. A delivery planning device according to claim 1 or 2.

4. The delivery plan calculation unit modifies the first delivery plan based on predetermined constraints and generates a second delivery plan. The delivery planning device according to claim 3.

5. The system further includes a delivery plan display unit that displays a delivery plan map capable of distinguishing the points of agreement and / or differences between the first delivery plan and the second delivery plan. The delivery planning device according to claim 4.

6. The system further comprises a learning execution unit that performs machine learning on the vehicle probability model using training data that includes at least information indicating whether the two delivery destinations were previously delivered by the same vehicle. The delivery planning device according to claim 1.

7. A vehicle probability model, trained to output the probability that two delivery destinations will be served by the same vehicle when two delivery destinations are entered, is used to input the first and second delivery destinations. The vehicle probability output from the aforementioned vehicle probability model is obtained, Based on the acquired vehicle probability, the delivery cost between the first delivery destination and the second delivery destination is calculated. Delivery planning method.

8. A vehicle probability model, trained to output the probability that two delivery destinations will be served by the same vehicle when two delivery destinations are entered, is used to input the first and second delivery destinations. The vehicle probability output from the aforementioned vehicle probability model is obtained, Based on the acquired vehicle probability, the delivery cost between the first delivery destination and the second delivery destination is calculated. A delivery planning program that has a computer perform a task.

Images