Transport vehicle system

Abstract

This transport vehicle system comprises a plurality of transport vehicles and a transport vehicle controller. The transport vehicle controller comprises: a prediction unit that acquires prediction information having a prescribed numerical range regarding future utilization of at least some target links included in a plurality of links; a route determination unit that, on the basis of the link cost of each of the plurality of links, preferentially determines, as a transport route for a transport vehicle, a candidate route which is among a plurality of candidate routes for the transport vehicle to perform a prescribed delivery, and for which the sum of the link costs of the plurality of links included in the candidate route is small; and a correction unit that corrects the link costs of the target links on the basis of the numerical range of the prediction information so that the link costs of the target links change for a plurality of deliveries.

Description

Transport vehicle system

[0001] The present disclosure relates to a guided vehicle system.

[0002] Conventionally, for example, in a semiconductor manufacturing factory or the like, a guided vehicle system has been known that includes a plurality of guided vehicles that transport articles such as FOUPs (Front Opening Unified Pods) that house semiconductor substrates, and a controller that controls the travel of these guided vehicles (see Patent Document 1). Patent Document 1 describes a method for setting a transport route for the guided vehicles based on link costs set for each link on a specified drivable route.

[0003] Patent No. 7059999

[0004] In the above-described methods, the link cost of each link may be statically set based on, for example, the link distance (length), or dynamically set based on past performance (e.g., the average time required for a predetermined number of transport vehicles that recently used the link to pass through the link). In either case, when multiple transports (transportation commands) occur at approximately the same time, the transport routes for each transport are determined under the same conditions (i.e., when the link costs set for each link are approximately the same). As a result, when performing a shortest route search using, for example, the Dijkstra algorithm, the same route is likely to be consistently selected for multiple transports. This leads to the problem that when multiple transports occur simultaneously, the utilization of a specific route increases, making traffic congestion more likely to occur. One possible way to avoid this problem is to perform a route search that restricts the use of the specific route for some of the multiple transports occurring simultaneously. However, this method makes it difficult to detect in advance which routes will cause traffic congestion, and requires complex control.

[0005] The present disclosure aims to provide a guided vehicle system that can suppress, through simple control, the occurrence of congestion caused by the concentration of guided vehicles on a specific route in multiple transports.

[0006] The present disclosure includes the transport vehicle systems [1] to

[10] .

[0007] [1] A transport vehicle system comprising a plurality of transport vehicles that transport items by traveling along a predetermined travel path, and a controller that controls the travel of the plurality of transport vehicles, wherein the travel path is composed of a plurality of links connected to each other, and each of the plurality of links is associated with a link cost, and the controller has: a prediction unit that acquires prediction information having a predetermined numerical range regarding future utilization for at least some target links included in the plurality of links; a route determination unit that, based on the link costs of each of the plurality of links, preferentially determines, as the transport route for the transport vehicle, a candidate route among a plurality of candidate routes for causing the transport vehicle to perform transport, which has a small sum of link costs of a plurality of links included in the candidate route; and a correction unit that corrects the link cost of the target link based on the numerical range of the prediction information so that the link cost of the target link changes for a plurality of transports.

[0008] In the guided vehicle system described above in [1], the link costs of at least some of the target links included in the travel path are corrected so as to vary for multiple transports. This allows for different route search conditions (i.e., the link costs of the target links) for each transport, even when multiple transports occur simultaneously. As a result, the transport routes of multiple transports are more likely to be distributed between routes that use the target links and routes that do not. Furthermore, by using a prediction range (numerical range) for future utilization, the link costs of the target links can be easily corrected and the correction results can be varied. Therefore, the above guided vehicle system can suppress congestion caused by the concentration of guided vehicles on a specific route in multiple transports through simple control.

[0009] [2] The controller controls the travel of the transport vehicle so that the transport vehicle heads toward the destination by assigning a transport command generated within the transport vehicle system and associated with a destination included in the target link to one of the transport vehicles, and the future utilization is information regarding the number of transport commands predicted to occur in the future for the destination included in the target link.

[0010] According to the configuration [2] above, it is possible to obtain rational and highly accurate prediction information regarding the future utilization of the target link based on the predicted number of future transport commands to be generated for the target link.

[0011] [3] A guided vehicle system according to [1] or [2], wherein the prediction unit obtains, as the prediction information, information on probability density in which a probability of occurrence is associated with each of a plurality of numerical values ​​relating to the future utilization, and the correction unit generates a random number according to the probability density and corrects the link cost of the target link based on a numerical value corresponding to the random number included in the numerical range of the prediction information.

[0012] According to the above configuration [3], the link cost of the target link can be corrected by a simple process of generating random numbers. Also, the fluctuation range of the link cost of the target link can be easily adjusted depending on the size of the numerical range of the prediction information.

[0013] [4] The guided vehicle system of [3], wherein the correction unit corrects the link cost of each of the target links based on the ratio of the numerical value of each of the target links to the average value of the numerical values ​​determined by the random numbers for each of the multiple target links.

[0014] According to the configuration of [4] above, in a single correction process for multiple target links, the relative change in link cost between each target link can be increased. This allows the range of variation in the conditions (link cost of each target link) for multiple transports to be increased, making it possible to more effectively distribute the transport routes for multiple transports.

[0015] [5] A transport vehicle system according to any one of [1] to [4], wherein the prediction unit acquires probability density information in which an occurrence probability is associated with each of a plurality of numerical values ​​relating to the future utilization as the prediction information, and sets a range in the probability density defined by a predetermined probability threshold as the numerical range of the prediction information.

[0016] According to the configuration [5] above, it becomes easy to keep the fluctuation range of the link cost of the target link within a desired range.

[0017] [6] The guided vehicle system of [5], wherein the prediction unit generates a plurality of random numbers according to the probability density and determines the probability threshold based on the distribution of the plurality of random numbers.

[0018] According to the configuration [6] above, the probability threshold can be easily and appropriately determined by the process of generating random numbers according to the probability density.

[0019] [7] A guided vehicle system according to [5] or [6], wherein the correction unit corrects the link cost of the target link based on the ratio between a numerical value selected from the numerical range of the prediction information and the most frequent value of the probability density.

[0020] A target link with a large predicted range (numerical range) of utilization can be said to be a link where the actual link cost (the time it will take for a guided vehicle to pass through the target link in the future) is largely unknown. It is preferable to avoid vehicle concentration on a target link with high uncertainty regarding future utilization. According to the configuration of [7] above, the larger the variance in the probability density distribution (i.e., the larger the predicted range of utilization) of a target link, the larger the link cost correction range can be. As a result, the likelihood of selecting a target link with high uncertainty in future utilization can be made to differ greatly among multiple transports. Ultimately, vehicle concentration on the target link can be appropriately avoided among multiple transports.

[0021] [8] A transport vehicle system according to any one of [5] to [7], wherein the controller controls the travel of the transport vehicle so that the transport vehicle heads toward the destination by assigning to one of the transport vehicles a transport command that is generated within the transport vehicle system and that is associated with a destination included in the target link, and the prediction unit obtains the probability density based on the number of past transport commands in which a point included in the target link was set as the destination.

[0022] According to the configuration [8] above, the probability density of the prediction information can be easily and appropriately acquired based on the history of the number of occurrences of past transport commands.

[0023] [9] A guided vehicle system according to any one of [5] to [7], wherein the prediction unit acquires the probability density based on historical information indicating the number of guided vehicles that have used the target link in the past.

[0024] According to the configuration [9] above, the probability density of the prediction information can be easily and appropriately acquired based on the history of the number of guided vehicles that have used the target link in the past.

[0025]

[10] A guided vehicle system according to any one of [1] to [9], wherein the route determination unit determines whether or not to use the link cost of the target link corrected by the correction unit based on the priority of transportation.

[0026] For example, if a route search is performed using link costs corrected by the correction unit for a transport with a high transport priority, a route different from the actual shortest route may be determined as the transport route for that transport, resulting in a risk of prolonging the time until the transport is completed. According to the configuration of

[10] above, by determining whether or not to use the corrected link costs based on the priority of the transport, for example, by configuring the system to perform a route search using the link costs originally set for each link for transport with a high transport priority, the above risk can be avoided.

[0027] According to the present disclosure, it is possible to provide a transport vehicle system that can suppress, through simple control, the occurrence of congestion caused by transport vehicles concentrating on a specific route in multiple transports.

[0028] FIG. 1 is a diagram illustrating an example of the layout of a guided vehicle system. FIG. 2 is a block diagram illustrating the functional configuration of the guided vehicle system. FIG. 3 is a block diagram illustrating an example of the hardware configuration of a guided vehicle controller. FIG. 4 is a diagram illustrating an example of a transport log. FIG. 5 is a diagram illustrating an example of aggregated data of transport logs for each period. FIG. 6 is a diagram illustrating an example of learning data and verification data used for learning a prediction model. FIG. 7 is a diagram schematically illustrating an example of prediction processing using a prediction model. FIG. 8 is a diagram illustrating an example of an output result of a prediction model for a certain target link. FIG. 9 is a diagram illustrating an example of how to calculate a probability threshold. FIG. 10 is a diagram illustrating a probability density obtained as prediction information. FIG. 11 is a flowchart illustrating an example of processing by a guided vehicle system. FIG. 12 is a diagram illustrating processing by a guided vehicle controller of a first modified example. FIG. 13 is a diagram illustrating processing by a guided vehicle controller of a second modified example. FIG. 14 is a diagram illustrating processing by a guided vehicle controller of a third modified example.

[0029] Hereinafter, an embodiment of the present disclosure will be described with reference to the drawings. In the description of the drawings, the same or equivalent elements are denoted by the same reference numerals, and redundant description may be omitted.

[0030] As shown in FIG. 1 , a guided vehicle system 1 according to this embodiment includes a transport path 4 (a predetermined travel path) and a plurality of guided vehicles 2 traveling along the transport path 4. The transport path 4 is, for example, a rail (track) laid in a factory. The guided vehicles 2 are, for example, automated guided vehicles that transport articles. The guided vehicles 2 are, for example, overhead traveling vehicles, rail-guided trolleys, etc. As an example, the guided vehicles 2 are overhead traveling automated guided vehicles that are capable of traveling along the transport path 4. For example, the guided vehicles 2 are overhead traveling automated guided vehicles (OHTs). As an example, the articles transported by the guided vehicles 2 are cassettes (so-called FOUPs (Front Opening Unified Pods)) that accommodate a plurality of semiconductor wafers.

[0031] The transport path 4 is divided into a plurality of bays (sections) B (three bays B1, B2, and B3 in the example of FIG. 1). The transport path 4 includes an intra-bay route 5, which is a route within a bay B, and an inter-bay route 6, which is a route connecting different bays B. Along the transport path 4, processing equipment 7, a stocker (not shown), and the like are provided. The processing equipment 7 is equipment that performs processing on semiconductor wafers. The stocker is a point where the transport vehicle 2 can temporarily store items and serves as a buffer.

[0032] The transport path 4 includes a plurality of nodes 10 and a plurality of links 11 connecting adjacent nodes 10. That is, the transport path 4 is composed of a plurality of links 11 connected to each other via the nodes 10. The plurality of nodes 10 indicate specific points such as a branching section 8 and a merging section 9. In this embodiment, each link 11 is a unidirectional link. That is, the transport vehicle 2 can travel in a predetermined traveling direction (the direction of the arrow shown in FIG. 1 ) for each link 11. The branching section 8 is a node 10 connected to one upstream link (the link 11 upstream of the branching section 8) and a plurality of downstream links (the links 11 downstream of the branching section 8). The merging section 9 is a node 10 connected to a plurality of upstream links (the links 11 upstream of the merging section 9) and a single downstream link (the link 11 downstream of the merging section 9).

[0033] The processing device 7 is provided with an inlet port for carrying in articles (i.e., a point where the transport vehicle 2 unloads articles) and an outlet port for carrying out articles (i.e., a point where the transport vehicle 2 picks up (loads) articles). The inlet port and outlet port are located below the transport path 4. The inlet port may also serve as an outlet port.

[0034] 2, the guided vehicle system 1 includes a plurality of guided vehicles 2 and a guided vehicle controller 20 (controller). The guided vehicle controller 20 receives a transport command from a host controller 12.

[0035] The transport vehicle controller 20 controls the travel of the transport vehicle 2 so that the transport vehicle 2 to which the transport command is assigned heads toward the destination. Specifically, the transport vehicle controller 20 receives a transport command from the upper controller 12 instructing the transport vehicle 2 to head toward the destination (in this embodiment, a From point and a To point, which will be described later) on the transport path 4. The transport vehicle controller 20 controls the travel of the transport vehicle 2 by assigning the transport command received from the upper controller 12 to a transport vehicle 2 selected from the plurality of transport vehicles 2.

[0036] A transport command is associated with a destination included in any one of the multiple links 11. Examples of destinations include the above-mentioned processing device 7 (incoming port, outgoing port), a stocker, etc. In this embodiment, a transport command is associated with a "From point" indicating a point where the item to be transported is to be picked up (i.e., the origin of the item to be transported) and a "To point" indicating a point where the item to be transported is to be unloaded (i.e., the destination of the item to be transported). That is, one transport command is associated with the From point as a first destination and the To point as a second destination. A transport vehicle 2 to which such a transport command is assigned first travels toward the From point, which is the first destination. Next, the transport vehicle 2 picks up the item to be transported at the From point, transports the item to the To point, which is the second destination, and unloads the item at the To point.

[0037] The upper controller 12 has, for example, an MES (Manufacturing Execution System) and an MCS (Material Control System). The MES is managed by a manufacturer or the like. The MES is capable of communicating with the processing device 7. For example, the processing device 7 transmits a transport request (item grab request, item unloading request) for an item for which processing has been completed to the MES. The MES transmits the transport request received from the processing device 7 to the MCS. When the MCS receives a transport request from the MES, it converts the transport request into a transport command as described above and transmits the transport command to the guided vehicle controller 20. As a result, the transport command is assigned to a specific guided vehicle 2 via the guided vehicle controller 20.

[0038] The transport vehicle controller 20 determines a transport vehicle 2 to which the transport command is to be assigned based on predetermined selection criteria. For example, the transport vehicle controller 20 determines the transport vehicle 2 closest to the From point and to which no other transport commands are assigned (an available transport vehicle) as the transport command destination. The transport vehicle controller 20 also executes a predetermined route search algorithm (e.g., a shortest path search algorithm such as Dijkstra's algorithm) based on the link cost associated with each link 11 to determine a transport route for executing the transport command (i.e., a travel route from the current position of the transport vehicle 2 to the To point via the From point), and notifies the transport vehicle 2 of the transport route. As a result, the transport vehicle 2 travels on the transport path 4 based on the transport route.

[0039] The guided vehicle controller 20 may be configured to distribute the above-described processing among multiple sub-controllers. For example, the transport path 4 may be divided into multiple areas (modules), with one sub-controller provided for each area. In this case, the transport route within each area may be determined by the sub-controller corresponding to each area. For example, if the transport route for executing a transport command passes through three areas a, b, and c in this order, the sub-controller managing area a may determine the sub-transport route within area a, the sub-controller managing area b may determine the sub-transport route within area b, and the sub-controller managing area c may determine the sub-transport route within area c. Furthermore, when the transport route is divided into multiple sub-transport routes for each area in this manner, the timing for determining the sub-transport route for each area (i.e., the timing for calculating the optimal sub-transport route using the above-described path search algorithm) may differ for each area. For example, the sub-controller managing area c may determine the sub-transport route within area c immediately before the guided vehicle 2 enters area c and notify the guided vehicle 2 of the determined sub-transport route. Furthermore, the transport vehicle controller 20 may once determine a transport route for a transport vehicle 2, and then update the route by recalculating the transport route for the transport vehicle 2 while the transport vehicle 2 is traveling along the determined transport route. In either case, the transport route for the transport vehicle 2 is determined based on the link cost of each link 11.

[0040] The link cost associated with a link 11 is, for example, a value related to the time required for a guided vehicle 2 to pass through the link 11 (hereinafter referred to as the "passing time"). A first example of the link cost is the length of the link 11 (the distance from the upstream node to the downstream node of the link 11). A second example of the link cost is the value obtained by dividing the length of the link 11 by the average speed of the guided vehicles 2 (i.e., the theoretical average passing time). A third example of the link cost is the average passing time of the link 11 (e.g., the average passing time of a predetermined number of guided vehicles 2 that have most recently passed through the link 11). However, the link cost may be in a form other than the first to third examples described above. In this embodiment, the link cost of each link 11 is the third example described above. In this case, the guided vehicle controller 20 may update the link cost of a link 11 each time a guided vehicle 2 passes through the link 11, or may update the link costs of each link 11 collectively at a predetermined time step interval (15-minute intervals in this embodiment).

[0041] The transport vehicle controller 20 (the route determination unit 22 described later) preferentially determines, as the transport route of the transport vehicle 2, a candidate route having a small sum of link costs of the multiple links 11 included in the candidate route among multiple candidate routes for the transport vehicle 2 to perform transport (transport corresponding to a predetermined transport command). For example, when a shortest path search algorithm such as the Dijkstra algorithm described above is used, the candidate route having the smallest sum of link costs of the multiple links 11 included in the candidate route is determined as the transport route. Note that the transport vehicle controller 20 does not necessarily determine the shortest path in the strict sense as the transport route. For example, when the transport path 4 is divided into multiple areas as described above and a route search is performed for each area, the route formed by connecting the shortest paths (sub-transport paths) determined for each area may not match the shortest path when viewed as the entire transport path 4. Furthermore, even when the transport vehicle controller 20 performs a route search within a limited calculation time in order to improve the response speed of notifying the transport route to the transport vehicle 2, the determined transport route may not match the shortest path when viewed as the entire transport path 4.

[0042] 3, the transport vehicle controller 20 may be configured as a computer system including one or more processors 201 such as CPUs (Central Processing Units), one or more RAMs (Random Access Memories) 202 and one or more ROMs (Read Only Memories) 203 as main storage devices, an input device 204 such as a keyboard for an operator to input operations, an output device 205 such as a display that presents information to the operator, a communication module 206 for communicating with the transport vehicle 2, and an auxiliary storage device 207 such as an HDD and SSD. The transport vehicle controller 20 may be configured as a single computer device or may be configured as a plurality of computer devices (for example, a plurality of sub-controllers as described above).

[0043] Each function of the transport vehicle controller 20 is realized, for example, by loading a predetermined program into a memory such as RAM 202, operating the input device 204 and output device 205 under the control of the processor 201, operating the communication module 206, and reading and writing data in RAM 202 and the auxiliary storage device 207.

[0044] Next, the functions of the guided vehicle controller 20 will be described with reference to Fig. 2. The guided vehicle controller 20 has a storage unit 21, a route determination unit 22, a prediction unit 23, and a correction unit 24.

[0045] The storage unit 21 is a database that stores various information related to the guided vehicle system 1. The storage unit 21 may be configured with a single database device or multiple database devices. In this embodiment, the storage unit 21 stores the layout of the transport path 4 as shown in FIG. 1 and link costs associated with each of the multiple links 11. In this embodiment, the storage unit 21 also stores a prediction model M used for processing by the prediction unit 23, which will be described later.

[0046] The route determination unit 22 determines a transport route for the transport vehicle 2 based on the link costs of each of the multiple links 11 stored in the storage unit 21. For example, when a transport command is assigned to a transport vehicle 2, the route determination unit 22 determines a transport route for the transport vehicle 2 to execute the transport command (e.g., a route from the current location of the transport vehicle 2 to the To location of the transport command via the From location of the transport command). As described above, the route determination unit 22 preferentially determines, as the transport route for the transport vehicle 2, a candidate route having a small sum of link costs of the multiple links 11 included in the candidate route, from among multiple candidate routes for the transport vehicle 2 capable of executing the transport command. For example, the route determination unit 22 determines the transport route for the transport vehicle 2 to which the transport command is assigned by using a shortest path search algorithm such as Dijkstra's algorithm. The transport vehicle controller 20 notifies the corresponding transport vehicle 2 of the transport route determined by the route determination unit 22. As a result, the transport vehicle 2 can travel on the transport path 4 according to the transport route determined by the route determination unit 22.

[0047] The route determination unit 22 is basically configured to perform a route search using the link costs corrected by the correction unit 24 for target links described below. However, in the present embodiment, the route determination unit 22 is configured to be able to determine whether to use the link costs of the target links corrected by the correction unit 24 based on the transportation priority. Information indicating the transportation priority is associated with, for example, a transportation command. The information indicating the transportation priority is, for example, a numerical value indicating the level of priority. As an example, in the present embodiment, the route determination unit 22 is configured to perform a route search without using the link costs corrected by the correction unit 24 (i.e., using the link costs originally set for each link) for transportation commands in which the transportation priority is equal to or higher than a predetermined threshold.

[0048] The prediction unit 23 acquires prediction information having a predetermined numerical range regarding future utilization for at least some target links included in the plurality of links 11. In this embodiment, the prediction unit 23 acquires, as prediction information P (see FIG. 8 ), probability density information in which an occurrence probability is associated with each of a plurality of numerical values ​​regarding utilization. The prediction unit 23 then sets, as the numerical range of the prediction information P, a range R (see FIGS. 9 and 10 ) that is defined by predetermined probability thresholds (in this embodiment, the first and third quartiles) in the probability density and includes the mode. In this embodiment, the future utilization of a certain target link is information regarding the number of transport commands predicted to be generated in the future for destinations included in the target link (i.e., transport commands that include a point included in the target link as a From point or a To point). The prediction unit 23 acquires the probability density of the target link based on the number of past transport commands in which a point included in the target link was set as a destination (in this embodiment, a From point or a To point).

[0049] The target link is a link 11 that is the target of processing to correct the link cost by the correction unit 24, which will be described later. In this embodiment, the target link is a link 11 that includes a point that can be the destination (From point or To point) of the transport command (for example, a point that can be the target of grabbing or unloading FOUPs such as the above-mentioned processing device 7 or a stocker). The number of target links may be one or more. In the layout example shown in FIG. 1 , at least link La that includes one processing device 7 and link Lb that includes two processing devices 7 correspond to the target links.

[0050] The prediction unit 23 periodically executes the above-described process at a predetermined time step interval (15-minute intervals in this embodiment). More specifically, at a certain processing time point (the present time point), the prediction unit 23 acquires prediction information (probability density and numerical range) regarding the utilization of each target link for a future period from the present time point to the next processing time point (15 minutes after the present time point). In this embodiment, the prediction unit 23 acquires the prediction information (probability density) by using a prediction model M.

[0051] An example of the prediction model M will be described with reference to FIGS. 4 to 7. FIG. 4 shows an example of the transport log for each of a plurality of transports that arrived at the From point during a certain period (two time steps TS1 and TS2). Information in one record (row) in FIG. 4 shows a transport log corresponding to one transport command. As an example, the transport log includes information on the From arrival date and time, the transport completion date and time, the From link, and the To link. The "From arrival date and time" is information indicating the date and time when the transport vehicle 2 arrived at the From point. The "Transport completion date and time" is information indicating the date and time when the transport vehicle 2 arrived at the To point and completed unloading of the items (FOUP). The "From link" is information (link ID) that identifies the link 11 (target link) that includes the From point. The "To link" is information (link ID) that identifies the link 11 (target link) that includes the To point.

[0052] The aggregated data for each period shown in Fig. 5 is data obtained by aggregating the multiple transport logs shown in Fig. 4 for each period (time step) and for each target link. More specifically, by aggregating for each target link the number of transport commands in which the target link is the From link (number of From transport commands) and the number of transport commands in which the target link is the To link (number of To transport commands), data is obtained that correlates a predetermined past period, the target link (link ID), and the total number of transport commands (number of From transport commands + number of To transport commands; hereinafter simply referred to as the "number of transport commands").

[0053] The number of From transport commands for the target link with link ID "1001" in time step TS1 can be obtained by counting the number of transport logs (see FIG. 4) whose From arrival dates and times are included in the period of time step TS1 (here, the period from "January 1st, 00:00 to 00:14:59") in which the From link is "1001". Similarly, the number of To transport commands for the target link with link ID "1001" in time step TS1 can be obtained by counting the number of transport logs whose To link is "1001" in which the transport completion dates and times are included in the period of time step TS1.

[0054] Here, it is considered that there is a certain relationship (pattern) between the number of transport commands for each target link in the most recent past period and the number of transport commands for each target link in the future period (next time step). Therefore, in this embodiment, a prediction model M is used that inputs the number of transport commands for each target link in the past period and outputs a prediction result of the number of transport commands for each target link in the future period corresponding to the next time step.

[0055] In this embodiment, a vector (data V1, V2 in the example of FIG. 5) containing as elements the number of transport commands for each of a plurality of target links (links 11 with link IDs "1001" to "9990" in the example of FIG. 5) at each time step (time steps TS1, TS2 in the example of FIG. 5) is used as data for training the prediction model M (training data or verification data, which will be described later).

[0056] The prediction model M is generated by machine learning so as to output the probability density of the above-described prediction information (see FIG. 8 ) as a prediction result of the number of transport commands. Such a prediction model M is generated, for example, by a gradient boosting decision tree (GBDT), which is a type of supervised learning. However, the learning method for generating the prediction model M is not limited to the above-described GBDT, as long as it can be trained to output a prediction result in the form of a probability density (see FIG. 8 ).

[0057] FIG. 6 illustrates an example of training data and verification data used in training the prediction model M. In the example of FIG. 6, the prediction model M is configured to output a prediction result of the number of transport commands for each target link in a future period (15 minutes, or one time step) immediately following a past period (three time steps, or 45 minutes) based on data (data V1 and V2 in FIG. 5) on the number of transport commands for each target link. In the example of FIG. 6, past periods corresponding to time steps with time step IDs "350" to "364" are displayed. As shown in FIG. 6, data on the number of transport commands for multiple past time steps is divided into training data and verification data. It is preferable that the training data period (time steps prior to ID "355" in the example of FIG. 6) and the verification data period (time steps after ID "358" in the example of FIG. 6) are not contiguous in time. In the example of FIG. 6, a 30-minute interval is provided between the training data period and the verification data period.

[0058] In FIG. 6 , one row of data corresponds to one piece of training data or validation data. The data for the hatched period (a period of three time steps) corresponds to the input data for the prediction model M, and the data for the black period (a period of one time step) corresponds to the output (correct label) of the prediction model M. In this way, by acquiring data for four consecutive time steps from data for a past period including multiple time steps while sliding the time steps one by one, multiple training data and validation data can be efficiently obtained. The prediction model M is trained using multiple training data, for example, so that the difference between the output result (a value obtained probabilistically according to the output probability density) obtained by inputting the first three time steps of data for each validation data into the prediction model M is minimized from the last one time step of data for each validation data. Note that the length of the input data period for the prediction model M (three time steps in this embodiment), the number of training data, and the number of validation data are appropriately set based on the results of accuracy evaluation of the prediction model M, etc.

[0059] FIG. 7 is a diagram schematically illustrating an example of prediction processing using a prediction model M (trained model) created by machine learning as described above. As shown in FIG. 7 , when data (V1, V2, V3) from a past period (t-2, t-1, t) corresponding to the most recent three time steps is input to the prediction model M, the prediction model M outputs data X from the future period (t+1) corresponding to the next time step. The data X represents the number of transport commands (the number of transport commands predicted to be issued) for each target link in the future period using a probability density. That is, the data X includes probability density information for each target link. Thus, in this embodiment, the prediction unit 23 uses the prediction model M to obtain probability density information regarding the future utilization of each target link (in this embodiment, data X output from the prediction model M) based on the number of past transport commands in which a location included in each target link is set as the destination (in this embodiment, data corresponding to the most recent three time steps input to the prediction model M).

[0060] 8 is a diagram showing an example of the output result of the prediction model M for a certain target link La (a part of the data X in FIG. 7 ). As shown in FIG. 8 , in this embodiment, the prediction process of the prediction model M described above obtains probability density information (e.g., a probability density function) as shown in FIG. 8 for each target link as prediction information P. That is, information (prediction information P) indicating the predicted occurrence number in the future period t+1 of transport commands in which a point included in the target link La is set as the destination (From point or To point) is obtained in the form of probability density.

[0061] The prediction unit 23 sets a range defined by a predetermined probability threshold in the probability density obtained by the prediction model M as the numerical range of the prediction information P. In this embodiment, the predetermined probability threshold is a lower limit LL corresponding to the first quartile (25th percentile) of the probability density and an upper limit UL corresponding to the third quartile (75th percentile). As an example, the prediction unit 23 generates multiple random numbers according to the probability density obtained by the prediction model M, and determines the probability thresholds (lower limit LL and upper limit UL) based on the distribution of the multiple random numbers.

[0062] An example of the above processing by the prediction unit 23 will be described with reference to FIG. 9 . First, the prediction unit 23 generates multiple (e.g., 3,000) random numbers according to the probability density (prediction information P) of the number of transport commands shown in FIG. 8 . The histogram H in FIG. 9 shows an example of the results of generating such random numbers. Next, the prediction unit 23 calculates a lower limit LL corresponding to the first quartile and an upper limit UL corresponding to the third quartile based on the distribution of the multiple random numbers. The prediction unit 23 determines the range R between the lower limit LL and the upper limit UL thus determined as the numerical range of the prediction information P (i.e., the numerical range effective in the processing by the correction unit 24, described later). The range R determined in this manner includes the mode MO of the multiple random numbers.

[0063] The prediction unit 23 executes the above process for each of the plurality of target links, thereby obtaining prediction information P (information including the probability density and range R) for each target link.

[0064] The correction unit 24 corrects the link cost of the target link based on the numerical range (range R from lower limit LL to upper limit UL) of the prediction information P of the target link so that the link cost of the target link changes between multiple transports. In other words, the correction unit 24 executes a correction process for the link cost of the target link for each transport command so that the link costs of the target link used when the route determination unit 22 searches for a route differ from one another between multiple transport commands.

[0065] An example of a process for correcting the link cost of a target link La will be described below. The correction unit 24 corrects the link cost of the target link La based on the ratio between a numerical value n selected from the range R of the prediction information P of the target link La and the most frequent value MO of the probability density. For example, the correction unit 24 generates random numbers according to the probability density of the prediction information P until a numerical value included in the range R is obtained, and when a numerical value included in the range R is obtained, selects that numerical value as the numerical value n. The correction unit 24 also calculates a correction coefficient K (= n / MO) for the target link La based on the numerical value n and the most frequent value MO. The correction unit 24 then corrects the link cost of the target link La by multiplying the link cost of the target link La by the correction coefficient K. If there are multiple target links, the correction unit 24 may perform the above-described correction process for each target link.

[0066] 10A shows an example in which a probability density with a relatively large variance is obtained as the prediction information P. In this example, the mode MO is "40," the lower limit LL (first quartile) is "10," and the upper limit UL (third quartile) is "70." In this case, the minimum value of the correction coefficient K is 0.25 (=10 / 40) when the lower limit LL is selected as the value n, and the maximum value of the correction coefficient K is 1.75 (=70 / 40) when the upper limit UL is selected as the value n.

[0067] 10B shows an example in which a probability density with relatively small variation is obtained as the prediction information P. In this example, the mode MO is "40," the lower limit LL (first quartile) is "35," and the upper limit UL (third quartile) is "45." In this case, the minimum value of the correction coefficient K is 0.875 (=35 / 40) when the lower limit LL is selected as the value n, and the maximum value of the correction coefficient K is 1.125 (=45 / 40) when the upper limit UL is selected as the value n.

[0068] Next, an example of the processing of the guided vehicle system 1 (guided vehicle controller 20) will be described with reference to Fig. 11. As shown in Fig. 11, the processing of the path determination unit 22 and the processing of the prediction unit 23 are executed independently of each other.

[0069] First, the processing flow of the prediction unit 23 will be described. As described above, the prediction unit 23 acquires (or updates) prediction information P for each target link every time a predetermined time step (a period of 15 minutes in this embodiment) elapses (step S11: YES → step S12). The prediction information P for each target link is acquired using the prediction model M stored in the storage unit 21. The prediction information P for each target link acquired or updated in step S12 is stored in the storage unit 21.

[0070] Next, the process flow of the route determination unit 22 and the correction unit 24 will be described. When the route determination unit 22 receives a transport command from the upper controller 12, it determines whether the priority of the transport command is equal to or higher than a predetermined threshold (step S1: YES → step S2). If the priority is equal to or higher than the threshold (step S1: YES), the processes of steps S3 to S5 are skipped. If the priority is not equal to or higher than the threshold (step S2: NO), the route determination unit 22 requests the correction unit 24 to perform a correction process (step S3). When the correction unit 24 receives the correction request from the route determination unit 22, it corrects the link cost of each target link based on the prediction information P of each target link stored in the storage unit 21 (step S4). The correction unit 24 notifies the route determination unit 22 of correction information indicating the corrected link cost of each target link (step S5). The route determination unit 22 also obtains link cost information of links 11 other than the target link from the storage unit 21 (step S6). Next, the route determination unit 22 assigns the transportation command received in step S1 to one of the transportation vehicles 2, and performs a route search for a transportation route of the transportation vehicle 2 (step S7). If the route determination unit 22 has acquired correction information in step S5, the route determination unit 22 performs a route search by using the link costs corrected by the correction unit 24 as the link costs of each target link.

[0071] [Effects] In the above-described guided vehicle system 1, the link costs of at least some of the target links included in the transport path 4 (e.g., links La and Lb in FIG. 1 ) are corrected so as to vary between multiple transports (transport commands). More specifically, the correction process (step S4 in FIG. 11 ) by the correction unit 24 is performed for each transport command. In this embodiment, the correction process is performed using a correction coefficient K based on a random number. However, because a different random number (which may coincidentally match, but is highly unlikely) is generated for each correction process, a correction process using a different correction coefficient K is performed for each transport command. This allows different route search conditions (i.e., link costs of target links La and Lb) to be set for each transport command, even if multiple transport commands are generated simultaneously. As a result, the transport routes of multiple transport commands are more likely to be distributed between routes that use target link La and routes that do not use target link La. The same applies to target link Lb. Furthermore, by using a prediction range (range R in this embodiment) for future utilization, the link cost of the target link can be easily corrected and the correction results can be varied. Therefore, with the guided vehicle system 1, congestion caused by the concentration of guided vehicles on a specific route in multiple transports can be suppressed through simple control.

[0072] For example, in FIG. 1 , a transport vehicle 2A located near the exit from bay B1 to the upper interbay route 6 must pass through bay B2 or bay B3 to reach link Lc of the lower interbay route 6. Consider a situation where, before the link costs of the target links La and Lb are corrected, the shortest route for the transport vehicle 2A to reach link Lc is a route passing through the target link La. Under such circumstances, if the correction process for the target links La and Lb is not performed, when multiple transport commands that require the transport command to pass near the exit of bay B1 to reach link Lc are generated intensively, each of the transport routes for the multiple transport commands will use the target link La. As a result, congestion may occur in the target link La or the link 11 upstream of the target link La. In contrast, by correcting the link costs of the target links La and Lb so that they vary for each transport command, as in the present embodiment, it is possible to create a situation in which the transport routes for the multiple transport commands can be distributed so that the transport routes corresponding to some of the transport commands use the target link Lb. In other words, a situation can be created in which the magnitude relationship between the route cost of the route to link Lc via target link La (the sum of the link costs of all links 11 included in the route) and the route cost of the route to link Lc via target link Lb is reversed. More specifically, when the correction coefficient Ka of target link La is greater than 1 and the correction coefficient Kb of target link Lb is less than 1, the above-described route cost reversal phenomenon can be caused, and the route determination unit 22 can determine a transportation route that passes through target link Lb. Therefore, according to this embodiment, when a plurality of transportation commands that require the transportation vehicles to pass near the exit of bay B1 to reach link Lc are generated in a concentrated manner, it is possible to prevent transportation vehicles from concentrating on a specific route (the route that passes through target link La in the above example).

[0073] The future utilization is information relating to the number of transport commands predicted to be issued in the future for destinations included in the target link. According to the above configuration, reasonable and highly accurate prediction information P relating to the future utilization of the target link can be obtained based on the predicted number of transport commands to be issued in the future for the target link.

[0074] As shown in FIG. 8 , the prediction unit 23 acquires probability density information as prediction information P, in which a probability of occurrence is associated with each of a plurality of numerical values ​​related to utilization. The correction unit 24 generates random numbers according to the probability density and corrects the link cost of the target link based on a numerical value corresponding to the random number included in a numerical range (range R) of the prediction information P. With the above configuration, the link cost of the target link can be corrected by a simple process of generating random numbers. Furthermore, the fluctuation range of the link cost of the target link can be easily adjusted depending on the size of range R.

[0075] 9 and 10 , the prediction unit 23 sets a range R defined by predetermined probability thresholds (in this embodiment, a lower limit LL corresponding to the first quartile and an upper limit UL corresponding to the third quartile) in the probability density as the numerical range of the prediction information P. This configuration makes it easy to keep the fluctuation range of the link cost of the target link within a desired range. For example, if it is desired to more actively diversify the transport routes between transport commands, the probability density can be divided into percentiles, and the lower limit LL can be set to a value (e.g., 15th percentile) lower than the 25th percentile (first quartile) and the upper limit UL can be set to a value (e.g., 85th percentile) higher than the 75th percentile (third quartile). This increases the range R and the fluctuation range of the correction coefficient K. On the other hand, if you want to keep the degree of dispersion of transport routes between transport commands below a certain level (to maintain to some extent the situation where the shortest route based on the link cost before correction is likely to be used), you can narrow the range R by increasing the lower limit LL (or decreasing the upper limit UL), and the fluctuation range of the correction coefficient K can be reduced.

[0076] 9, the prediction unit 23 generates a plurality of random numbers according to a probability density and determines probability thresholds (in this embodiment, a lower limit LL corresponding to the first quartile and an upper limit UL corresponding to the third quartile) based on the distribution of the plurality of random numbers. With the above configuration, the process of generating random numbers according to a probability density makes it possible to easily and appropriately determine the probability thresholds (the lower limit LL and the upper limit UL).

[0077] The correction unit 24 corrects the link cost of the target link based on the ratio (in this embodiment, the correction coefficient K (= n / MO)) between a value n selected from the numerical range (range R) of the prediction information P and the most frequent value MO of the probability density. A target link with a large utilization prediction range (range R) is a link 11 for which the future (next time step) utilization (number of transportation command issues) is highly uncertain (in other words, there is a risk that the number of transportation command issues will exceed the expected value). It is preferable to avoid concentrating vehicle traffic on target links with such high uncertainty regarding future utilization. More specifically, if such a specific link is located on the shortest route between two predetermined points that are some distance apart, there is a possibility that vehicle traffic will concentrate on the specific link. Furthermore, if the future utilization of the specific link (i.e., the number of transportation command issues with a point included in the specific link as a destination) exceeds the expected value, congestion may occur on the specific link (or on links upstream of the specific link). According to the above-described configuration for correcting the target link cost, the link cost correction range (variation range of the correction coefficient K) can be made larger for target links with larger variations in probability density distribution (wider range R). As a result, the likelihood of selecting a target link with high uncertainty in future utilization can be made to differ greatly among multiple transportation commands. This in turn makes it possible to appropriately prevent transportation vehicles from concentrating on the target link in multiple transportation commands.

[0078] The prediction unit 23 obtains a probability density based on the number of past transport commands in which a point included in the target link was set as a destination. In this embodiment, as shown in FIG. 7 , the prediction unit 23 inputs data V1, V2, and V3 on the number of past transport commands for each target link into a prediction model M, and thereby obtains the probability density (see FIG. 8 ) for each target link output from the prediction model M. With the above configuration, the probability density of the prediction information P can be easily and appropriately obtained based on the history of the number of past transport commands.

[0079] The route determination unit 22 determines whether to use the link costs of the target links corrected by the correction unit 24 based on the priority of the transportation. For example, if a route search is performed using the link costs corrected by the correction unit 24 for a transportation command with a high transportation priority, a route different from the actual shortest route may be determined as the transportation route for the transportation, resulting in a risk of extending the time required to complete the transportation. According to the above configuration, the risk can be avoided by determining whether to use the corrected link costs based on the priority of the transportation. For example, as in the example of processing shown in FIG. 11 , the risk can be avoided by configuring the route search to be performed using the link costs originally set for each link for transportation with a high transportation priority.

[0080] [Modifications] Although the embodiments of the present disclosure have been described above, the present disclosure is not limited to the above-described embodiments, and various modifications are possible without departing from the spirit of the present disclosure.

[0081] 12, the prediction unit 23 may create a histogram H1 of the number of past transport commands based on the number of past transport commands in which a point included in the target link was set as the destination (the number of transport commands tallied for each past period as shown in FIG. 5). The prediction unit 23 may then acquire an approximation curve C obtained by fitting the histogram H1 as prediction information P representing the probability density. Subsequent processing by the transport vehicle controller 20 (determination of the probability threshold, correction processing, etc.) is the same as in the above embodiment.

[0082] According to the first modification, it is possible to omit the learning process of the prediction model M. However, when the prediction model M is used as in the above embodiment, it is possible to obtain a probability density that reflects the correlation between the number of transport commands in the most recent past period and the number of transport commands in the next period, and therefore it is possible to improve the prediction accuracy compared to the first modification.

[0083] (Second Modification) The correction unit 24 calculates the average value n of the numerical values ​​n determined by random numbers generated according to the probability density for each of the plurality of target links. ＡＶＥ (That is, the sum of the numerical values ​​n of the multiple target links n ＴＯＴＡＬ The ratio of the value n of each target link to the value obtained by dividing by the number of target links (n / n ＡＶＥ ) the link cost of each target link may be corrected based on the above-mentioned rule. For simplicity of explanation, consider the case where there are two target links La and Lb. FIG. 13 shows an example in which the correction process for the target links La and Lb according to the above rule is performed four times (i.e., for four transport commands). In the example of FIG. 13, in the first trial, "15" is determined as the numerical value n of the target link La, and "21" is determined as the numerical value n of the target link Lb. In this case, the average value n ＡＶＥ is "18", the correction coefficient K of the target link La is 0.83 (=15 / 18), and the correction coefficient K of the target link Lb is 1.17 (=21 / 18). The same applies to the other trials.

[0084] According to the second modification, a single correction process for multiple target links La, Lb can increase the relative change in the link cost between each of the target links La, Lb. This increases the range of variation in the conditions (the link cost of each of the target links La, Lb) between multiple transport commands, making it possible to more effectively distribute transport routes between multiple transport commands.

[0085] (Third Modification) The prediction unit 23 may acquire prediction information P (probability density) for each target link based on history information indicating the number of guided vehicles 2 that have used the target link in the past. As shown in FIG. 14 , the prediction unit 23 may acquire the prediction information P (probability density) for each target link by performing learning and prediction using a prediction model M using data V1 and V2 consisting of the number of guided vehicles that have used (passed) the target link in the past, instead of the number of transport commands shown in FIG. 5 . Alternatively, without using the prediction model M, a histogram H1 as in the first modification may be generated and an approximation curve C obtained by fitting the histogram H1 may be acquired as the prediction information P representing the probability density. Note that the number of guided vehicles that use each link shown in FIG. 14 indicates the number of guided vehicles that have passed through the link in the target period. Such information on the number of guided vehicles that use each link can be calculated, for example, by using a vehicle position information log that records the position coordinates of all guided vehicles 2 on the transport path 4 at regular intervals.

[0086] According to the third modification, the probability density of the prediction information P can be easily and appropriately acquired based on the history of the number of transportation vehicles that have used the target link in the past. Furthermore, according to the third modification, even links 11 that do not include a point that can be the destination (From point or To point) of the transportation command can be treated as target links. In other words, according to the third modification, all links 11 included in the transportation path 4 can be treated as target links.

[0087] In the above embodiment, a FOUP containing multiple semiconductor wafers is exemplified as an item (carried object) transported by the transport vehicle 2, but the item is not limited to this and may be, for example, other containers containing glass wafers, reticles, etc., or other items. Furthermore, the location where the transport vehicle system 1 is installed is not limited to a semiconductor manufacturing factory, and the transport vehicle system 1 may be installed in other facilities.

[0088] 1...Transport vehicle system, 2, 2A...Transport vehicle, 4...Transport path (travel path), 11...Link, 20...Transport vehicle controller (controller), 22...Route determination unit, 23...Prediction unit, 24...Correction unit, La, Lb...Target link, LL...Lower limit (probability threshold), MO...Modulus, n...Numerical value, n ＡＶＥ ...average value, P...prediction information, R...range (numerical range), UL...upper limit (probability threshold).

Claims

1. A transport vehicle system comprising a plurality of transport vehicles that transport items by traveling along a predetermined travel path, and a controller that controls the travel of the plurality of transport vehicles, wherein the travel path is composed of a plurality of links connected to each other, and each of the plurality of links is associated with a link cost, and the controller has: a prediction unit that acquires prediction information having a predetermined numerical range regarding future utilization for at least some target links included in the plurality of links; a route determination unit that, based on the link costs of each of the plurality of links, preferentially determines, as a transport route for the transport vehicle, a candidate route having a small sum of link costs of a plurality of links included in the candidate route, from among a plurality of candidate routes for allowing the transport vehicle to perform transport; and a correction unit that corrects the link cost of the target link based on the numerical range of the prediction information, so that the link cost of the target link changes in a plurality of transports.

2. The transport vehicle system of claim 1, wherein the controller controls the travel of the transport vehicle so that the transport vehicle heads toward a destination by assigning a transport command generated within the transport vehicle system and associated with a destination included in the target link to one of the transport vehicles, and the future utilization is information regarding the number of transport commands predicted to occur in the future for a destination included in the target link.

3. The transport vehicle system of claim 1, wherein the prediction unit obtains, as the prediction information, probability density information in which a probability of occurrence is associated with each of a plurality of numerical values ​​relating to the future utilization, and the correction unit generates a random number according to the probability density and corrects the link cost of the target link based on a numerical value corresponding to the random number included in the numerical range of the prediction information.

4. The transport vehicle system according to claim 3, wherein the correction unit corrects the link cost of each of the target links based on the ratio of the numerical value of each of the target links to the average value of the numerical values ​​determined by the random numbers for each of the plurality of target links.

5. The transport vehicle system of claim 1, wherein the prediction unit obtains probability density information as the prediction information, in which a probability of occurrence is associated with each of a plurality of numerical values ​​relating to the future utilization, and sets a range in the probability density defined by a predetermined probability threshold as the numerical range of the prediction information.

6. The transport vehicle system according to claim 5, wherein the prediction unit generates a plurality of random numbers according to the probability density, and determines the probability threshold value based on a distribution of the plurality of random numbers.

7. A guided vehicle system as described in claim 5, wherein the correction unit corrects the link cost of the target link based on the ratio between a numerical value selected from the numerical range of the prediction information and the most frequent value of the probability density.

8. The transport vehicle system described in claim 5, wherein the controller controls the travel of the transport vehicle so that the transport vehicle heads toward the destination by assigning to one of the transport vehicles a transport command generated within the transport vehicle system and associated with a destination included in the target link, and the prediction unit obtains the probability density based on the number of past transport commands in which a point included in the target link was set as the destination.

9. The transport vehicle system according to claim 5, wherein the prediction unit obtains the probability density based on history information indicating the number of transport vehicles that have used the target link in the past.

10. The transport vehicle system according to claim 1, wherein the route determination unit determines whether or not to use the link cost of the target link corrected by the correction unit based on the priority of transportation.