Vehicle allocation planning device, vehicle allocation planning method, and program

Abstract

A vehicle allocation planning device according to the present invention calculates a vehicle allocation route of one or more electric vehicles that circulate among one or more consumption nodes which consume electric power and one or more charging nodes which recharge electric power, and that supply electric power to the consumption nodes. The vehicle allocation planning device includes a calculation unit that calculates a vehicle allocation route for the one or more electric vehicles so as to minimize an objective function including a penalty related to the number of the consumption nodes in which the remaining electric power is 0 during a certain working time.

Description

Vehicle allocation planning device, vehicle allocation planning method, and program

[0001] The present invention relates to a technology for creating a dispatch route for electric vehicles (EVs) to rescue base stations.

[0002] In mobile communications, communication services are provided by base stations located in various locations communicating with terminals. In the event of a large-scale power outage due to a disaster or other event, it is necessary to maintain the remaining power of base stations for as long as possible in order to continue providing communication services.

[0003] In order to prevent the base station's power from running out, it is being considered to dispatch an EV to the base station and supply power from the EV to the base station.

[0004] The EV dispatching problem can be considered a type of inventory routing problem (IRP). In general, the IRP is a problem of generating optimal vehicle dispatch routes to replenish inventory at delivery destinations while ensuring a certain inventory level at the delivery destinations. In the EV dispatching problem, the delivery destinations correspond to base stations, and electricity corresponds to inventory.

[0005] Wouter Kool, Herke van Hoof, and Max Welling. "Attention, learn to solve routing problems!" In InternationalConference on Learning Representations, 2019.Jingwen Li, Yining Ma, Ruize Gao, Zhiguang Caoy, Andrew Lim, Wen Song, Jie Zhang,"Deep Reinforcement Learning for Solving the Heterogeneous Capacitated Vehicle Routing Problem", IEEE TRANSACTIONS ON CYBERNETICS, 2021Oriol Vinyals, Meire Fortunato, Navdeep Jaitly," Pointer Networks", In Advances in Neural Information Processing Systems, Vol. 28, 2015.Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Lukasz Kaiser, and Illia Polosukhin," Attention is all you need",In Advances in Neural Information Processing Systems, Vol. 30, 2017.

[0006] Regarding IRP, many route generation techniques using deep reinforcement learning have been proposed as conventional techniques (e.g., Non-Patent Documents 1 to 3).

[0007] However, the route generation techniques proposed so far are only applicable to simplified problems and are difficult to apply to complex real-world problems. In other words, it is difficult to generate EV dispatch routes that enable continuity of communication services even during disasters using conventional techniques.

[0008] The present invention has been made in consideration of the above points, and aims to provide a technology for generating EV dispatch routes that enable continuity of communication services even during disasters.

[0009] According to the disclosed technology, there is provided a vehicle dispatch planning device that calculates dispatch routes for one or more electric vehicles that travel around one or more consumption nodes that consume power and one or more charging nodes that charge power and supply power to the consumption nodes, the vehicle dispatch planning device including a calculation unit that calculates the dispatch routes for the one or more electric vehicles so as to minimize an objective function that includes a penalty for the number of consumption nodes whose remaining power reaches zero within a certain working time.

[0010] The disclosed technology provides a technology for generating EV dispatch routes that enable continuation of communication services even during disasters.

[0011] FIG. 1 is a diagram showing how multiple EVs travel around nodes. FIG. 1 is a configuration diagram of a vehicle dispatch planning device 100. FIG. 2 is a configuration diagram of a calculation unit 130. FIG. 2 is a diagram showing an example of time runs of EV1, EV2, and EV3. FIG. 3 is a diagram showing an example of time runs of EV1, EV2, and EV3. FIG. 4 is a diagram showing an image of processing multiple instances at once. FIG. 5 is a diagram for explaining the calculation of the visiting probability of each node for a selected EV. FIG. 6 is a diagram for explaining equation (5). FIG. 7 is a diagram for explaining penalties for base stations. FIG. 8 is a diagram showing an example of the hardware configuration of a vehicle dispatch generation device 100.

[0012] Hereinafter, an embodiment of the present invention will be described with reference to the drawings. The embodiment described below is merely an example, and the embodiment to which the present invention is applied is not limited to the following embodiment.

[0013] In the following, an embodiment will be described in which an EV travels around base stations to supply power, but the traveling vehicle in the technology according to the present invention is not limited to an EV, and the target to which power is supplied is not limited to a base station. The traveling vehicle may also be called an electric traveling vehicle. Furthermore, the target to which power is supplied may also be called a consumption node.

[0014] (Summary) In this embodiment, an EV travels to a base station and charges at the base station. More specifically, multiple EVs travel asynchronously, receive power (from the EV to the base station), and charge (from the charging station to the EV). This is illustrated in Figure 1. In Figure 1, the EVs, base stations, and charging stations (charging nodes) are as shown in the legend. Furthermore, DEC indicates a decrease in the remaining power, and INC indicates an increase in the remaining power.

[0015] (Device Configuration Example) Fig. 2 shows a configuration diagram of the vehicle dispatch planning device 100 according to this embodiment. As shown in Fig. 2, the vehicle dispatch planning device 100 includes a node information collection unit 110, a vehicle information collection unit 120, a calculation unit 130, a map API unit 140, and a vehicle dispatch unit 150.

[0016] The vehicle dispatch planning device 100 may be implemented by one device (computer) or by multiple devices. For example, the calculation unit 130 may be implemented by one computer, and the other functional units may be implemented by another computer. An overview of the operation of the vehicle dispatch planning device 100 is as follows.

[0017] The node information collection unit 110 acquires information on each base station. Specifically, static information includes the location, power capacity, and power consumption (power consumed per unit time) of each base station. Dynamic information includes the initial remaining power of each base station. This information may be acquired from each base station, or from a server or the like that stores information on each base station. Similarly, the node information collection unit 110 acquires static information on each charging station. This static information can be used to calculate dynamic state quantities that are input to the model.

[0018] The vehicle information collection unit 120 collects information about each EV. The information collected here is static information, such as power capacity, power supply efficiency, and power consumption while driving (power consumption while driving). The information may be collected from each EV itself, or from a server or the like that stores information about each EV.

[0019] The calculation unit 130 generates a dispatch route for each EV by solving the IRP based on information on each node (base station / charging station) and each EV. Details of the calculation unit 130 will be described later. The dispatch route here is information indicating the order of nodes that each EV will visit (the sequence in which the nodes are arranged in order).

[0020] The map API unit 140 performs a route search based on the information on the dispatch route of each EV output from the calculation unit 130, and, for example, draws the dispatch route of each EV on a map. The dispatch unit 150 distributes the information on the dispatch route to each EV (or a terminal at a service center) via a network based on the output result of the map API unit 140. The dispatch unit 150 may also be called an "output unit."

[0021] The map API unit 140 may perform route searches, for example, by accessing an external map server. Alternatively, the map API unit 140 may itself store a map database and perform route searches using the map database.

[0022] (Outline of Operation of Calculation Unit 130) The calculation unit 130 in this embodiment solves the IRP to determine the dispatch routes of each EV. Here, the problem setting, objective function, and constraint conditions for this purpose will be described.

[0023] <Problem Setting> In the EV dispatch problem in this embodiment, one or more EVs, one or more base stations, and one or more charging stations are given. With regard to feature quantities, each EV has a remaining power, a power capacity, a power supply efficiency, and a running power consumption. Each base station has a position, a remaining power, a power capacity, and a power consumption. Each charging station has a position and a charging efficiency. As initial values, the initial position and initial remaining power of the EV, and the initial remaining power of the base station are given.

[0024] <Objective Function> The objective function of the vehicle dispatching problem in this embodiment is "to minimize the number of base stations whose remaining power reaches 0 and the total travel distance of all EVs within a given operation time."

[0025] In the case of IRP in the prior art, the objective function is to minimize the travel distance under the constraint that the remaining power of the base station does not become 0. In contrast, since it is difficult to satisfy this constraint in the EV dispatch planning problem in this embodiment, instead, a penalty according to the number of base stations whose remaining power becomes 0 is incorporated into the objective function, thereby aiming to minimize the number of base stations whose remaining power becomes 0. A specific formulation will be described later.

[0026] <Constraints> The main constraints during optimization are as follows: The constraints are used when calculating input feature amounts to the vehicle selection unit and deep learning model, which will be described later.

[0027] The remaining power of the EV will not reach 0, whether the EV is moving or supplying power to the base station.

[0028] - The remaining power of an EV decreases linearly with the distance traveled, with the power consumption during driving as a coefficient.

[0029] - The EV arrives at the charging station before the end of work hours.

[0030] At a charging station, the remaining power of the EV increases linearly with the charging time, with the charging efficiency used as a coefficient.

[0031] - Power is supplied to the base station and EVs are charged until their respective power capacities are reached.

[0032] The remaining power of the base station decreases linearly with the elapsed time, with the power consumption as a coefficient.

[0033] - A maximum of one EV can stay at the base station.

[0034] When power is supplied from an EV to a base station, the remaining power of the base station increases linearly over the power supply time, while the remaining power of the EV decreases, using the difference between the EV's power supply efficiency and the base station's power consumption as a coefficient.

[0035] (Configuration and Operation of Calculation Unit 130) The calculation unit 130 in this embodiment corresponds to an extension of the Attention Model disclosed in Non-Patent Document 1 so as to be able to handle a multi-vehicle problem with a time axis.

[0036] The calculation unit 130 first selects one EV from the plurality of EVs, and then determines the destinations for the selected EV. The calculation unit 130 repeats these two processes (selection and destination determination) to finally calculate the travel route for each EV.

[0037] FIG. 3 shows an example configuration of the calculation unit 130 for realizing the above operation. As shown in FIG. 3, the calculation unit 130 includes a vehicle selection unit 11, a vehicle encoder 12, a node encoder 13, a probability calculation unit 14, a destination determination unit 15, a reinforcement learning unit 20, and a state storage unit 40. The vehicle encoder 12, the node encoder 13, and the probability calculation unit 14 are configured as a deep learning model 10. Note that either or both of the vehicle selection unit 11 and the destination determination unit 15 may be included in the deep learning model 10. Furthermore, the model that realizes the function of the "deep learning model 10" is not limited to a deep learning model, and a model other than a deep learning model may also be used. An overview of the operation is as follows.

[0038] The vehicle selection unit 11 selects an EV that is ready to travel to the next destination, starting with the earliest one. The state quantities of each node when the EV is selected and the state quantities of each EV are input from the state holding unit 40 to the deep learning model 10. The deep learning model 10 outputs the probability of each node to be visited next for the selected EV. The destination determination unit 15 determines the node to be visited next for the selected EV based on the probability. The determined node is output to an external device (e.g., the map API unit 140) of the calculation unit 130 and input to the state holding unit 40.

[0039] By repeating the above operations, a dispatch route for each EV is calculated. The state holding unit 40 calculates and outputs the state quantities of each EV and each node at each point in time (e.g., each time an EV is selected), and stores the calculated state quantities in a storage device such as a memory. The state holding unit 40 also stores the determined (sampled) dispatch route for each EV.

[0040] The reinforcement learning unit 20 learns the parameters (which may also be called a policy) of the deep learning model 10 by using, for example, reinforcement learning using a policy gradient method (specifically, for example, REINFORCE with baseline) based on the vehicle dispatch routes stored in the state storage unit 40. The parameters updated by the learning are reflected in the deep learning model 10.

[0041] In addition, reinforcement learning may be performed in parallel with the operation of determining vehicle dispatch routes, or the operation of determining vehicle dispatch routes may be performed separately from reinforcement learning using the deep learning model 10 obtained in the reinforcement learning phase.

[0042] The operation of each part of the calculation part 130 will be described in more detail below.

[0043] <Vehicle Selection Unit 11> First, the vehicle selection unit 11 will be described. In this embodiment, the vehicle selection unit 11 selects an EV that is ready to travel to the next destination in order of earliest arrival. Specifically, this is as follows.

[0044] An EV has two states: "moving" and "powered or charging," and the two states alternate.

[0045] Because the power of EVs and the power of base stations both increase or decrease linearly, calculation is easy when no EVs are transitioning states. On the other hand, when an EV state transitions, the manner in which the power of EVs and base stations increases or decreases changes. Therefore, in this embodiment, in order to appropriately manage the remaining power of all EVs and all base stations, the state quantity (e.g., remaining power) of each EV and each base station is calculated when the state of an EV transitions.

[0046] The manner in which the remaining power changes before and after a state transition will be described with reference to Fig. 4. Fig. 4 is a diagram showing an example of a time run of EV1, EV2, and EV3. In the example of Fig. 4, for example, when EV1 first arrives at the base station, the manner in which the remaining power changes before and after a state transition is as follows:

[0047] Before the state transition: The remaining power of EV1 decreases according to "the distance traveled by EV1 x the power consumed by EV1 while traveling," and the remaining power of the base station before EV1 arrives decreases according to "the elapsed time x the power consumed."

[0048] After the state transition: the remaining power of EV1 decreases according to "power supply efficiency of EV1 x elapsed time", and the remaining power of the base station visited by EV1 increases according to "power supply efficiency x elapsed time".

[0049] An EV becomes operational immediately after power supply or charging is completed, so the vehicle selection unit 11 selects one EV in the order in which power supply or charging is completed.

[0050] In the example of FIG. 5 showing the timeline of the actions of three EVs, the vehicle selection unit 11 selects EVs in the order EV2->EV1->EV3->EV1->EV3.

[0051] The state quantities at the time when an EV is selected are input from the state holding unit 40 to the deep learning model 10 (specifically, the vehicle encoder 12 and the node encoder 13). The state quantities input to the deep learning model 10 include the remaining power of each EV, the remaining power of each base station, the current time, and the current location of each EV.

[0052] In the reinforcement learning in this embodiment, batch learning is performed. That is, multiple instances are processed at once. The instances are processed asynchronously. An image of multiple instances being processed at once is shown in FIG. 6.

[0053] If we count movement and "power / charge" as one action each, then in instance 1 in Figure 6, EV2 is first selected at the end of the third action. On the other hand, in instance 2, EV2 is first selected at the end of the fourth action.

[0054] In this case, the processing continues with the number of actions for instance 2 advanced by one step. In other words, the number of actions between instances in the learning batch is also processed asynchronously. This is implemented by looping processing until the EVs selected in all instances in the batch have just finished being powered or charged.

[0055] <Deep Learning Model 10 / Destination Determination Unit 15> After an EV is selected, the next destination of the EV is determined by the "deep learning model 10 and destination determination unit 15" using a Transformer (Non-Patent Document 4)-based method. Note that using a Transformer (Non-Patent Document 4)-based method is just one example. Any method may be used as long as it can output the probability that each node will be the next destination of the selected EV using the deep learning model 10.

[0056] In this embodiment, the vehicle encoder 12 and the node encoder 13 included in the deep learning model 10 each have multiple Transformer encoder layers without position encoding.

[0057] The Transformer encoder layer includes a multi-head attention layer, a feedforward layer, and the like, as described in Non-Patent Document 4.

[0058] The node encoder 13 receives the state quantities (position of each node, remaining power of the base station, etc.) of each node (base station and charging station), and calculates the embedded representation (high-dimensional vector) of each node using the following equation (1).

[0059] Furthermore, the vehicle encoder 12 receives the state quantities of each EV (such as the current position and remaining power of each EV), and calculates the embedded representation of each EV using the following equation (2).

[0060] In the following description, the subscripts of subscripts are represented by "_" in the text of the specification. ｉ indicates the i-th node, and a ｊ indicates the jth EV, N indicates the number of nodes, and M indicates the number of EVs. Enc(l) is the lth Transformer encoder layer. The subscript indicates that the embedding corresponding to the character is extracted.

[0061] h （０） ｖ＿ｉis a feature obtained by linearly transforming the state quantity of the node calculated based on the constraints described above. （０） ａ＿ｊ is a feature obtained by linearly transforming the state quantity of the EV calculated based on the constraint conditions described above. Note that the calculation of the linear transformation may be performed by the deep learning model 10 or may be performed outside the deep learning model 10.

[0062] In each of the node encoder 13 and the vehicle encoder 12, the embedded representation obtained in the (l-1)th layer is used as input, and the embedded representation in the lth layer is calculated repeatedly, thereby performing calculations in the Lth layer.

[0063] The node encoder 13 outputs an embedded representation of each node, and the vehicle encoder 12 outputs an embedded representation of each EV. The embedded representation of each node and the embedded representation of the selected EV are input to the probability calculation unit 14.

[0064] The probability calculation unit 14 calculates the inner product of the embedded representation of each node and the embedded representation of the selected EV, and obtains the visiting probability (probability distribution) of each node for the selected EV. An image of this calculation is shown in Figure 7. Specifically, the probability calculation unit 14 calculates the following equations (3) and (4).

[0065]

[0066] p(a ｓ , v ｉ ) is the selected EV a ｓ Node v of (s∈{1,...,M}) ｉ is the visit probability for q ａ＿ｓ =W Ｑ h （Ｌ） ａ＿ｓ and k ｖ＿ｉ =W Ｋ h （Ｌ） ｖ＿ｉ and d ｋ Ha K ｖ＿ｉ is the number of dimensions of W Ｑ , W Ｋ are trainable transformation matrices.

[0067] Node v that cannot be visited due to constraints ｘFor u(a ｓ , v ｘ ) = -∞ to forcibly set the visit probability to 0.

[0068] The visiting destination determination unit 15 determines the actual visiting destinations based on the obtained visiting probability p(a ｓ , v ｉ The vehicle dispatch route is determined by randomly sampling one node according to the distribution of all nodes in the vehicle dispatch route table (D). The vehicle dispatch route information determined by sampling is stored in the state holding unit 40.

[0069] During inference after learning, the destination determination unit 15 may output approximately 1,000 to 10,000 different vehicle allocation routes through this random sampling, and select the optimal vehicle allocation route from among them as the final vehicle allocation route. The optimal vehicle allocation route is, for example, the vehicle allocation route that minimizes the objective function described below.

[0070] <Reinforcement Learning> Next, a method of reinforcement learning executed by the reinforcement learning unit 20 will be described. The learning parameter θ in this embodiment is a parameter included in the formulas (1) and (2) and a parameter W Ｑ and W Ｋ is.

[0071] The reinforcement learning unit 20 optimizes the above parameters so as to minimize the sum of the penalty related to the number of base stations whose remaining power reaches 0 and the total travel distance of all EVs. More specifically, the reinforcement learning unit 20 performs reinforcement learning so as to minimize the following objective function (penalty):

[0072] Here, π is the set of all EV dispatch routes. Furthermore, as shown in Figure 8, the first term on the right side of equation (5) is the total travel distance of all EVs, and the second term is the time average of the ratio of base stations whose remaining power is 0 to all base stations. T is the operation time, and r(t) is the ratio of base stations whose remaining power is 0 to all base stations at time t.

[0073] The second term on the right side of equation (5) is a general description, and in this case, r(t) changes discontinuously, so it is possible to perform the calculation using discrete addition as in the following equations (6) and (7).

[0074]

[0075] In equation (7), T is the time interval t ｅ (e.g., the intervals between the vertical dotted lines in Figure 4).

[0076] Consump in Equation (6) ｉ is the power consumption of base station i, and ｉ (t ｅ ) is the time t ｅ is the remaining power after t in Equation (6) ｄｏｗｎ，ｉ is the base station i's batt ｉ (t ｅ ) is negative. ｄｏｗｎ，Ｂ , t ｄｏｗｎ，Ｃ The penalty in equation (7) is ｉ (t ｅ ) for a base station where t ｄｏｗｎ，ｉ The sum of t ｅ The second term of equation (5) can be calculated by dividing the result of equation (7) by the work time. ｅ After the penalty is calculated, ｉ are all clipped to 0.

[0077] More specifically, with regard to reinforcement learning, the reinforcement learning unit 20 may optimize learning parameters using REINFORCE when Greedy Rollout is used as a baseline, similar to the reinforcement learning in Non-Patent Document 1. In this case, the reinforcement learning unit 20 calculates the advantage by subtracting the penalty (Equation (5)) of a vehicle allocation route generated by greedily selecting destinations from the penalty (Equation (5)) of a vehicle allocation route generated by sampling. Greedily selecting destinations means selecting the node with the highest destination probability as the destination.

[0078] In normal reinforcement learning, learning is performed to maximize the reward, but in this embodiment, the opposite is true, and learning is performed to minimize the penalty.

[0079] (Hardware Configuration Example) The vehicle dispatch planning device 100 can be realized, for example, by causing a computer to execute a program. This computer may be a physical computer or a virtual machine on the cloud.

[0080] That is, the vehicle dispatch planning device 100 can be realized by executing a program corresponding to the processing performed by the vehicle dispatch planning device 100 using hardware resources such as a CPU and memory built into a computer. The program can be recorded on a computer-readable recording medium (such as a portable memory) and can be saved or distributed. The program can also be provided via a network such as the Internet or email.

[0081] Fig. 10 is a diagram showing an example of the hardware configuration of the computer. The computer in Fig. 10 includes a drive device 1000, an auxiliary storage device 1002, a memory device 1003, a CPU 1004, an interface device 1005, a display device 1006, an input device 1007, an output device 1008, and the like, all of which are interconnected by a bus BS.

[0082] The program that realizes the processing on the computer is provided by a recording medium 1001, such as a CD-ROM or a memory card. When the recording medium 1001 storing the program is set in the drive device 1000, the program is installed from the recording medium 1001 to the auxiliary storage device 1002 via the drive device 1000. However, the program does not necessarily have to be installed from the recording medium 1001, but may be downloaded from another computer via a network. The auxiliary storage device 1002 stores the installed program as well as necessary files, data, etc.

[0083] When an instruction to start a program is received, the memory device 1003 reads and stores the program from the auxiliary storage device 1002. The CPU 1004 realizes functions related to the vehicle dispatch planning device 100 in accordance with the program stored in the memory device 1003. Specifically, for example, the CPU 1004 realizes the functions of the calculation unit 130.

[0084] The interface device 1005 is used as an interface for connecting to a network, etc. The display device 1006 displays a GUI (Graphical User Interface) or the like according to a program. The input device 1007 is composed of a keyboard, mouse, buttons, a touch panel, etc., and is used to input various operation instructions. The output device 1008 outputs the results of calculations.

[0085] (Effects of the embodiment) As described above, the technology according to the present embodiment makes it possible to realize the vehicle dispatch planning device 100 that derives a dispatch route for multiple EVs by alternately repeating "selection of an EV" and "determining destinations for the selected EV."

[0086] The vehicle dispatch planning device 100 makes it possible to deal with constraints in accordance with actual problem settings, etc. As a result, it is possible to minimize the impact of base station power outages during disasters and generate EV dispatch routes that enable continuation of communication services even during disasters.

[0087] <Additional Notes> (Additional Item 1) A vehicle dispatch planning device that calculates dispatch routes for one or more electric vehicles that travel around one or more consumption nodes that consume power and one or more charging nodes that charge the power and supply power to the consumption nodes, comprising: a calculation unit that calculates the dispatch routes for the one or more electric vehicles so as to minimize an objective function that includes a penalty for the number of consumption nodes whose remaining power becomes zero within a certain working time. (Additional Item 2) The vehicle dispatch planning device according to Additional Item 1, wherein the calculation unit includes: a vehicle selection unit that sequentially selects an electric vehicle that is ready to move to the next destination, and a model that calculates a probability distribution of destination nodes for the electric vehicle selected by the vehicle selection unit. (Additional Item 3) The vehicle dispatch planning device according to Additional Item 2, wherein the calculation unit calculates a state quantity to be input to the model based on constraints that include a condition that the remaining power of the electric vehicle decreases linearly in proportion to the traveling distance, with power consumption during traveling as a coefficient. (Supplementary Item 4) The vehicle dispatch planning device according to Supplementary Item 2 or 3, further comprising: a reinforcement learning unit that learns parameters of the model so as to minimize the objective function, based on samples of vehicle dispatch routes determined on the basis of a probability distribution obtained by the model. (Supplementary Item 5) The vehicle dispatch planning device according to any one of Supplementary Items 1 to 4, wherein the objective function is the sum of the penalty for the number of consumption nodes whose remaining power is zero and a total travel distance of all electric vehicles, and the penalty is a time average of the proportion of consumption nodes whose remaining power is zero to all consumption nodes. (Supplementary Item 6) A vehicle dispatch planning method executed by a vehicle dispatch planning device that calculates vehicle dispatch routes for one or more electric vehicles that travel around one or more consumption nodes that consume power and one or more charging nodes that charge power, and that supply power to the consumption nodes, wherein the vehicle dispatch planning method calculates the vehicle dispatch routes of the one or more electric vehicles so as to minimize an objective function that includes a penalty for the number of consumption nodes whose remaining power is zero within a certain working time. (Supplementary Item 7) A non-transitory storage medium storing a program for causing a computer to function as a calculation unit in the vehicle allocation planning device according to any one of Supplementary Items 1 to 5.

[0088] Although the present embodiment has been described above, the present invention is not limited to such a specific embodiment, and various modifications and changes are possible within the scope of the gist of the present invention described in the claims.

[0089] 10 Deep learning model 11 Vehicle selection unit 12 Vehicle encoder 13 Node encoder 14 Probability calculation unit 15 Visit destination determination unit 20 Reinforcement learning unit 40 State retention unit 100 Vehicle allocation planning device 110 User information collection unit 120 Vehicle information collection unit 130 Calculation unit 140 Map API unit 150 Vehicle allocation unit 1000 Drive device 1001 Recording medium 1002 Auxiliary storage device 1003 Memory device 1004 CPU 1005 Interface device 1006 Display device 1007 Input device 1008 Output device

Claims

1. A vehicle dispatch planning device that calculates dispatch routes for one or more electric vehicles that travel around one or more consumption nodes that consume electric power and one or more charging nodes that charge the power and supply power to the consumption nodes, the vehicle dispatch planning device having a calculation unit that calculates the dispatch routes for the one or more electric vehicles so as to minimize an objective function that includes a penalty for the number of consumption nodes whose remaining power becomes zero within a certain working time.

2. The vehicle allocation planning device according to claim 1, wherein the calculation unit includes: a vehicle selection unit that sequentially selects an electric vehicle that is able to move to the next destination; and a model that calculates a probability distribution of destination nodes for the electric vehicle selected by the vehicle selection unit.

3. The vehicle dispatch planning device according to claim 2, wherein the calculation unit calculates state quantities to be input to the model based on constraints including a condition that the remaining power of the electric vehicle decreases linearly in proportion to the distance traveled, with the power consumption during travelling as a coefficient.

4. The vehicle allocation planning device according to claim 2, further comprising a reinforcement learning unit that learns parameters of the model so as to minimize the objective function based on samples of vehicle allocation routes determined based on the probability distribution obtained by the model.

5. The vehicle dispatch planning device according to claim 1, wherein the objective function is the sum of the penalty for the number of consumption nodes whose remaining power is zero and the total travel distance of all electric vehicles, and the penalty is the time average of the ratio of consumption nodes whose remaining power is zero to all consumption nodes.

6. A vehicle dispatch planning method executed by a vehicle dispatch planning device that calculates dispatch routes for one or more electric vehicles that travel around one or more consumption nodes that consume electricity and one or more charging nodes that charge electricity and supply power to the consumption nodes, wherein the vehicle dispatch planning method calculates the dispatch routes for the one or more electric vehicles so as to minimize an objective function that includes a penalty for the number of consumption nodes whose remaining power becomes zero within a certain working time.

7. A program for causing a computer to function as a calculation unit in the vehicle allocation planning device according to any one of claims 1 to 5.

Images