Method for determining whether dynamic dispatching is possible

Abstract

A method for determining whether dynamic dispatching is possible for a route including a plurality of fixed waypoints, according to the present invention, comprises: a first step of receiving, from a terminal of an applicant, a first dispatching request for starting from a first point and arriving at a second point; a second step of aligning, on the basis of a first axis coordinate, a plurality of fixed waypoints, the first point, and the second point; a third step of calculating a first waypoint vector with respect to two fixed waypoints surrounding the first point and the second point; a fourth step of calculating a first demand point vector with respect to the first point and the second point; and a fifth step of calculating a first angle between the first waypoint vector and the first demand point vector, and if the first angle meets a certain first criterion, approving the dynamic dispatching or requesting an additional operation required for the dynamic dispatching.

Description

How to determine whether dynamic dispatching is possible

[0001] The present invention relates to a method for determining whether dynamic dispatch is possible for a transportation device operating on a fixed route to a location other than a fixed stopover.

[0002] Additionally, it relates to a method for accurately inferring destination information included in a dynamic dispatch request using a large language model.

[0003] Furthermore, this paper relates to a method for determining whether dynamic dispatching is necessary based on the cost between stopovers calculated using a multi-level graph. Furthermore, this paper relates to a method for determining which mobile device to assign dynamic dispatching to when multiple dynamic dispatching requests are received simultaneously, using a state machine.

[0004] Additionally, it relates to a traffic prediction method for dynamically dispatched mobile devices.

[0005] Also, it relates to a method of assigning seats for dynamically dispatched moving devices.

[0006] For transportation vehicles that operate on fixed routes from the departure point to the arrival point via pre-designated stopovers, the need for dispatching and operating to places other than fixed stopovers is emerging.

[0007] In order to implement this dynamic dispatching, it is necessary to determine whether to allow dynamic dispatching or not, and calculations must be performed that take into account the additional time required, additional driving distance, and additional travel route due to dynamic dispatching as criteria for the determination.

[0008] As the number of fixed stopovers increases, the number of passengers on board the vehicle increases, and the number of dynamic dispatch requests increases, the amount of computation required to determine whether dynamic dispatch is allowed increases exponentially, which significantly increases resource consumption and also causes a problem in that it takes a lot of time to decide whether to allow dynamic dispatch.

[0009] In addition, many variables must be considered when determining dynamic dispatching, and conventional technologies only provide limited solutions for considering such variables.

[0010] Additionally, there was a limitation that the dispatch server would not recognize the destination if the destination name or address was not entered accurately when requesting dynamic dispatch.

[0011] Meanwhile, as the application of artificial intelligence (AI) has recently expanded, various attempts are being made to derive the optimal route from the starting point to the destination by reflecting the road traffic conditions in relation to autonomous driving of vehicles.

[0012] Accordingly, there have been attempts to develop various algorithms to derive the optimal travel route from the starting point set by the driver to the destination, but there has been a problem in that it is difficult to avoid the decrease in the accuracy of traffic prediction due to traffic congestion caused by unexpected events that occur on the actual road while driving.

[0013] In addition, for various algorithms to derive optimal travel paths, models that take into account numerous environmental factors had to be applied to respond to various environmental changes that occur during driving, so there were problems in that it took a lot of time for learning through deep learning and a lot of resources were required for data processing.

[0014] When the subject of dynamic dispatch is a bus, passengers sit in random seats, and there are frequent cases where passengers sit on the aisle side of the vehicle to move more comfortably, causing inconvenience to other passengers.

[0015] Or, when seats are assigned based on specific criteria at the time of reservation, there is a disadvantage in that passengers who travel relatively short distances between boarding and alighting are assigned to seats furthest from the entrance and exit of the transportation, which causes inconvenience in boarding and alighting and delays in operation.

[0016] [Prior Art Literature]

[0017] Korean Patent Publication No. 10-2023-0006893 (published on January 11, 2023)

[0018] The present invention aims to provide a method that dramatically reduces the resources and time required for dynamic dispatching by resolving the aforementioned problems of the prior art. Furthermore, it aims to provide a method for deriving optimal dynamic dispatch criteria through machine learning. Furthermore, it aims to provide a method that considers various variables to enable significantly more efficient and faster dynamic dispatching than prior art.

[0019] In addition, the present invention aims to provide a real-time road speed-based traffic prediction system and method that simplifies a reinforcement learning model by simplifying link-specific speed changes based on images using real-time road speed information, and provides a traffic situation in an area where a user driving a vehicle or an autonomous driving system will drive as a predicted image derived by reinforcement learning based on the current real-time speed.

[0020] In addition, the present invention aims to provide a seat assignment system and method for a reservation-based transportation means, which can minimize confusion among passengers getting on and off and arrange seats that are convenient to use by assigning seats based on the boarding and disembarking locations of passengers on a MOD (Mobility On Demand) bus or reservation transportation means that can be used by a large number of people.

[0021] A method for determining whether dynamic dispatch is possible for a route including a plurality of fixed waypoints according to the present invention comprises: a first step of receiving a first dispatch request from an applicant's terminal that starts from a first point and arrives at a second point; a second step of aligning a plurality of fixed waypoints, the first point, and the second point based on a first axis coordinate; a third step of calculating a first waypoint vector for two fixed waypoints surrounding the first point and the second point; a fourth step of calculating a first demand point vector for the first point and the second point; and a fifth step of calculating a first angle formed by the first waypoint vector and the first demand point vector, and approving dynamic dispatch or requesting additional operations necessary for dynamic dispatch if the first angle satisfies a predetermined first criterion.

[0022] The dynamic dispatch method according to the present invention further includes a step 2-1 of aligning a plurality of fixed waypoints, a first point, and a second point based on a second axis coordinate; a step 3-1 of calculating a second waypoint vector for two fixed waypoints surrounding the first point and the second point; and a step 4-1 of calculating a second demand point vector for the first point and the second point, wherein the step 5 may be a step of calculating a second angle formed by the second waypoint vector and the second demand point vector, and approving a dispatch request by further considering whether the second angle satisfies a predetermined second criterion.

[0023] The dynamic dispatch method according to the present invention may further include a sixth step of approving dynamic dispatch or requesting additional operations necessary for dynamic dispatch by additionally considering whether the shortest distance from at least one of the first point and the second point to the first waypoint vector or a line extending from the first waypoint vector satisfies a predetermined third criterion.

[0024] The dynamic dispatch method according to the present invention may further include a seventh step of approving dynamic dispatch or requesting additional operations necessary for dynamic dispatch by additionally considering whether the shortest distance from at least one of the first point and the second point to the second waypoint vector or a line extending from the second waypoint vector satisfies a predetermined third criterion.

[0025] The dynamic dispatch method according to the present invention may include: a step 1-1 of transmitting a query based on second dispatch request information including personalized information of a passenger in information included in a first dispatch request received in a first step to a vector database; a step 1-2 of receiving a response regarding third dispatch request information having a relationship with the second dispatch request information of a predetermined degree or higher from the vector database; a step 1-3 of transmitting a final query based on the first dispatch request information to a language model to query the language model; a step 1-4 of receiving a response from the language model; and a step 1-5 of generating a confirmation request message based on the response of step 1-4 and transmitting the confirmation request message to a user terminal.

[0026] The dynamic dispatch method according to the present invention may further include a step 1-6 of extracting destination-related information from a response received in step 1-4 based on a result approved by the user in step 1-5; and a step 1-7 of calling a dispatch API based on the destination information extracted in step 1-6.

[0027] The dynamic dispatch method according to the present invention may further include an 8th step of calculating costs for the first point and the second point and executing dynamic dispatch, and the 8th step may include an 8-1 step of calculating a cost between each fixed waypoint and the first point and a cost between each fixed waypoint and the second point; an 8-2 step of selecting, based on the calculation result of the 8-1 step, points 3-1 and 3-2, which are the two fixed waypoints with the lowest cost to the first point, and points 4-1 and 4-2, which are the two fixed waypoints with the lowest cost to the second point; and an 8-3 step of determining dynamic dispatch for the first point and the second point.

[0028] Step 8-3 is: Step 8-3-1, where the first value, which is the larger value between the 1-1 cost and the 1-2 cost, is compared with the 1-3 cost, and if the first value is less than or equal to the 1-3 cost, the first point is dispatched between the 3-1 point and the 3-2 point; Step 8-3-2, where the first value is greater than or equal to the 1-3 cost and the 1-1 cost is less than or equal to the 1-2 cost, the first point is dispatched before the 3-1 point; Step 8-3-3, where the first value is greater than or equal to the 1-3 cost and the 1-1 cost is greater than or equal to the 1-2 cost, the first point is dispatched after the 3-2 point; Step 8-3-4, where the second value, which is the larger value between the 2-1 cost and the 2-2 cost, is compared with the 2-3 cost, and if the second value is less than or equal to the 2-3 cost, the second point is dispatched between the 4-1 point and the 4-2 point; The step 8-3-5 may include dispatching the second point before the 4-1 point if the second value is greater than or equal to the 2-3 cost and the 2-1 cost is less than or equal to the 2-2 cost; and the step 8-3-6 may include dispatching the second point after the 4-2 point if the second value is greater than or equal to the 2-3 cost and the 2-1 cost is greater than or equal to the 2-2 cost.

[0029] The 1-1 cost is the cost from point 3-1 to point 1, the 1-2 cost is the cost from point 3-2 to point 1, the 1-3 cost is the cost from point 3-1 to point 3-2, the 2-1 cost is the cost from point 4-1 to point 2, the 2-2 cost is the cost from point 4-2 to point 2, and the 2-3 cost may be the cost from point 4-1 to point 4-2.

[0030] Point 3-1 may be a fixed stopover upstream of Point 3-2, and Point 4-1 may be a fixed stopover upstream of Point 4-2.

[0031] The above method may further include a step 8-3-7 for determining that dynamic dispatch is not possible if point 3-1 or point 4-1 is the first fixed stopover point in step 8-3-2 or step 8-3-5.

[0032] The above method may further include an 8-3-8 step, which unconditionally determines dynamic dispatching if the 3-2 point or the 4-2 point is the final fixed stopover point in the 4-3 step or the 4-6 step.

[0033] When it is determined that dynamic dispatch is possible by the present invention, a method for assigning mobile dynamic dispatch includes a ninth step of determining a boarding point of a passenger based on first point information; a tenth step of calculating a travel time from the departure point to the boarding point based on passenger status information and movement reference information; and an eleventh step of instructing dynamic dispatch to a mobile device capable of arriving at the boarding point within a predetermined time range after a time when the applicant can board, taking into account the travel time calculated in the tenth step.

[0034] The passenger status information may include at least one of the passenger's disability type and mobility assistance device information, and the mobility reference information may include at least one of weather information and route information from the departure point to the boarding point.

[0035] Step 10 may be a step of calculating a predicted movement time by applying a first parameter based on passenger status information and a second parameter based on movement reference information to a basic movement speed or basic movement time.

[0036] In the dynamic dispatching method according to the present invention, a machine learning method for predicting a passenger's travel time includes a 12th step of inputting at least one of passenger status information and at least one of movement reference information into an MLP learning engine and a Transformer learning engine; and a 13th step of predicting a travel time by combining the primary result value of the MLP learning engine and the primary result value of the Transformer learning engine in a final MLP layer.

[0037] The dynamic dispatch method according to the present invention comprises: a 14th step of generating a state machine including fixed waypoint information of a mobile device and properties regarding whether dispatch is in progress, and a callback function; a 15th step of receiving a first dynamic dispatch request from a first user terminal; a 16th step of selecting a first mobile device for which dispatch has been decided in response to the first dynamic dispatch request; a 17th step of updating a property indicating whether dispatch is in progress of the first mobile device to "true"; an 18th step of applying departure and destination information of the first dynamic dispatch request to the first mobile device to calculate a cost according to dynamic dispatch, and selecting a route with the minimum cost among them to generate a first response; a 19th step of the state machine receiving the first response and reflecting the changed route of the first mobile device included in the first response to a property of the first mobile device; If a second dynamic dispatch request is received while a first dynamic dispatch request is being processed, steps 16 to 19 are executed in parallel for the second dynamic dispatch request, but if the first mobile device is selected in step 16 for processing the second dynamic dispatch request, the state machine may include: a 20th step of calling a callback function after the first response is received by the state machine; a 21st step of calculating a cost according to dynamic dispatch by applying the departure and destination information of the second dynamic dispatch request to the first mobile device and selecting a route with the minimum cost among them to update the first response; a 22nd step of the state machine receiving the first response and reflecting the changed route of the first mobile device included in the first response to a property of the first mobile device; and a 23rd step of updating a property of whether dispatch is in progress of the first mobile device to "false".

[0038] A traffic prediction method for a mobile device allocated by a dynamic vehicle dispatching method of the present invention comprises: a speed data collection step of collecting speed data of vehicles traveling on each preset link and storing the collected data in a database; an event learning step of converting the speed data stored in the database into differentiated images that enable visual distinction between low and high speeds, and storing changes in the speed data of each link according to events that occur as traffic conditions progress by performing reinforcement learning based on changes in the images in the database; a regional event prediction step of predicting traffic events expected to occur in response to the current location of the vehicle and the real-time operating speed at that location based on the reinforcement learning data stored in the database, and generating a prediction image reflecting the predicted event and storing the predicted image in the database; and a prediction event application step of changing the current travel route to a route that bypasses an area where a congestion event is predicted based on the predicted event or changing an estimated time of arrival (ETA) to change and store an optimal route for reaching a destination in real time.

[0039] In the above speed data collection step, real-time speed data for each road can be obtained by acquiring public data from a city traffic information center accessed by a communication module equipped in a traffic prediction server.

[0040] The above event learning step may include a speed data visualization step for visualizing the speed data of each road stored in the database as differentiated images that can distinguish speed states such as congestion, slow-moving, and smooth; a unit time image storage step for setting an environment (evn) to be applied to reinforcement learning based on images by storing the speed data of each road visualized as differentiated images at preset unit times; and an image-based reinforcement learning step for learning a path that can maximize a reward in each environment while imposing a preset reward or penalty in response to a road selection of a vehicle driving on a road on which speed data is visualized at preset unit times.

[0041] In the above speed data visualization step, the colors of the roads can be differentiated and stored by expressing the congested area (less than 40 km / h) in the first color, the slow-moving area (40 to 80 km / h) in the second color, and the smooth area (over 80 km / h) in the third color.

[0042] In the above image-based reinforcement learning step, based on the speed and location information of the vehicle transmitted from the vehicle module, when the vehicle proceeds to the next road or link, a penalty is imposed if the vehicle passes through a first-colored area set as a congested area, and a reward is imposed if the vehicle passes through a third-colored area set as a smooth area, thereby learning the movement path of the vehicle driving in the area.

[0043] In the above image-based reinforcement learning step, if the image colors of the speed data are set to be diverse, differential rewards can be set to be applied according to the image colors of the area through which the vehicle passes.

[0044] In the above regional event prediction step, traffic events that may occur while a vehicle is driving in each region or link can be predicted based on real-time road speed data using the results derived through reinforcement learning.

[0045] In the above prediction event application step, when a new route request to reach a destination is made from the vehicle module, the result of reinforcement learning is applied to search for an optimal route based on a predicted image reflecting a traffic prediction for a certain period of time in the future when the vehicle will drive from the current location, and the optimal route can be transmitted to the vehicle module.

[0046] A method for assigning seats to a mobile device assigned by a dynamic dispatching method of the present invention comprises: (a) a step of registering seat information for each vehicle input from an administrator terminal; and (b) a step of assigning seats to each user based on the travel distance of the MOD bus for the departure point and destination included in the boarding request, the boarding time of the user during the travel distance of the bus, and the seat information of the dispatched bus when a bus capable of assigning seats is dispatched in response to a user's boarding request.

[0047] The step of assigning seats to the users may include (b1) calculating a user-specific evaluation index by calculating the boarding time and travel distance for at least one user scheduled to board the dispatched bus, and the remaining travel distance and remaining boarding time of passengers already on board; and (b2) assigning seats closer to the door of the MOD bus in descending order of the calculated user-specific evaluation index.

[0048] The above step (b2) may be a step of assigning seats closer to the entrance / exit door in the order of shortest travel distance, shortest boarding time, and the order of boarding request for users with the same calculated user-specific evaluation index.

[0049] The above step (a) is a step of receiving and storing seat information including the door location, the seat location from the door, and the seat type for each vehicle in operation from the administrator terminal, and the seat type may be one of a socially disadvantaged seat, a short-distance general seat, and a long-distance general seat.

[0050] The seat assignment method according to the present invention further includes a step of receiving a boarding request including the departure point, destination, departure time, and whether the passenger is socially disadvantaged from a user terminal, and when assigning a seat close to the door, the seat type can be determined based on whether the passenger is socially disadvantaged.

[0051] The above step (b) may further include a step of (bo) checking the remaining seats available for dispatch when the user boarding request is received to determine whether seat assignment is possible.

[0052] The above (bo) step is provided with information on seats for the socially disadvantaged. If the user's boarding request includes a request for a seat for the socially disadvantaged, it is determined that seat assignment is possible if there are remaining seats for the socially disadvantaged. If there are no remaining seats for the socially disadvantaged, the remaining seats of other dispatchable candidate vehicles can be checked.

[0053] The seat assignment method of the present invention further includes a step of storing statistical data by travel time based on the boarding and alighting points and boarding and alighting times of each existing bus user; and the step (b) predicts the occurrence probability of other users scheduled to board, the departure point and destination of the other users scheduled to board, based on statistical data corresponding to the travel time from the departure point to the destination of the user requesting boarding, and determines whether the other users scheduled to board will remain on the bus even when the user requesting boarding gets off based on the departure point and destination of the other users scheduled to board, and determines the seat of the user requesting boarding based on whether the other users scheduled to board will remain on the bus.

[0054] The above step (b) can assign a seat to the user requesting boarding to be closer to the door of the dispatched bus than to the other user expected to board, when the other user expected to board maintains boarding even when the user requesting boarding gets off.

[0055] The above step (b) may assign an aisle seat based on the departure point and destination of the other expected boarding user if the occurrence probability of the other expected boarding user is greater than or equal to the standard probability.

[0056] According to the present invention, an effect of efficient and rapid dynamic dispatch is provided by considering various variables and executing them quickly and reliably.

[0057] In addition, according to the present invention, by using real-time speed information of the road, the link-by-link speed change is simplified based on an image, thereby simplifying the reinforcement learning model, and there is an effect of providing the traffic situation of the area in which the user driving the vehicle or the autonomous driving system will drive as a predicted image derived by reinforcement learning based on the current real-time speed.

[0058] In addition, the present invention has the effect of increasing routing efficiency and improving estimated time of arrival (ETA) by predicting traffic events within a certain period of time based on real-time speed by applying reinforcement learning results based on road speed information.

[0059] In addition, the present invention has the effect of being able to adjust or respond to the vehicle's route appropriately to the actual road conditions by predicting traffic events in the relevant region based on the real-time speed of the road in each region.

[0060] In addition, the present invention has the effect of being utilized to enhance cost-based algorithms such as the Djikstra Algorithm by introducing a dynamic link cost so that the cost of a link changes according to an event predicted by reinforcement learning.

[0061] In addition, according to the seat assignment system and method according to the present invention, seats are assigned based on the boarding and disembarking locations of passengers on a MOD (Mobility On Demand) bus or reserved transportation that can be used by a large number of people, thereby minimizing confusion among passengers getting on and off and providing conveniently usable seats.

[0062] In particular, by calculating an evaluation index using the travel distance and boarding time based on the boarding location and location of the passenger, and giving priority to seats closer to the door to passengers with lower evaluation indexes, an efficient boarding and alighting service can be provided.

[0063] Additionally, it is possible to provide quality service by designating seats for the socially disadvantaged, and in the case of a specific infectious disease outbreak, it is possible to operate by setting seat intervals.

[0064] Figure 1 is an example of an environment in which a method for determining whether dynamic dispatching is possible according to the present invention is executed.

[0065] Figure 2 is a drawing showing an example of a route to which the method of the present invention is applied and a waypoint vector and a receiving point vector.

[0066] Figure 3 is a flowchart of a first embodiment of a method for determining whether dynamic dispatching is possible according to the present invention.

[0067] Figure 4 is a flowchart of a second embodiment of a method for determining whether dynamic dispatching is possible according to the present invention.

[0068] Figure 5 is an example of an environment in which a dynamic dispatching method considering passenger movement time according to the present invention is executed.

[0069] Figure 6 is an example of a dynamic dispatch request route to which a dynamic dispatch method considering passenger travel time according to the present invention is applied.

[0070] Figure 7 is a flowchart of the dynamic dispatching method of Figure 5.

[0071] Fig. 8 is a diagram illustrating the structure of machine learning for learning the predicted travel time in the dynamic dispatching method according to the flow chart of Fig. 7.

[0072] Fig. 9 is an example of an environment in which a method for determining dynamic dispatch based on a cost calculated by a multi-stage graph according to the present invention is executed.

[0073] FIG. 10 and FIG. 11 are diagrams showing exemplary dynamic dispatch points and fixed waypoints to explain a method for determining dynamic dispatch based on a cost calculated by a multi-stage graph according to the present invention.

[0074] Fig. 12 is a flowchart of a method for determining dynamic dispatch based on a cost calculated by a multi-stage graph according to the present invention.

[0075] Fig. 13 is a detailed flowchart of step (S4500) among the steps of Fig. 1.

[0076] Figure 14 is an example of an environment in which a dynamic dispatching method using a state machine according to the present invention is executed.

[0077] Fig. 15 is an example of a state machine used in a dynamic dispatching method using a state machine according to the present invention.

[0078] Figure 16 is a flowchart of a dynamic dispatching method using a state machine according to the present invention.

[0079] Figure 17 is an example of an environment in which a destination inference method using a large language model according to the present invention is executed.

[0080] Figure 18 is a flowchart of a destination inference method using a large language model according to the present invention.

[0081] Figure 19 is a configuration diagram of a real-time road speed-based traffic prediction system according to the present invention.

[0082] Figure 20 is a configuration diagram of a real-time road speed-based traffic prediction system according to the present invention.

[0083] Figure 21 is a flowchart of a traffic prediction method according to the present invention.

[0084] Figure 22 is a schematic diagram of a seat assignment system according to the present invention.

[0085] Figure 23 is a block diagram showing a schematic configuration of the seat assignment unit (8130) of Figure 22.

[0086] Figure 24 is a drawing for explaining a seat assignment method according to one embodiment of the present invention.

[0087] Figure 25 is a flowchart of a seat assignment method of a seat assignment system according to one embodiment of the present invention.

[0088] Figure 26 is a flowchart for explaining the step (S8130) of assigning seats to each user in Figure 25.

[0089] Hereinafter, the present invention will be described in detail with reference to the attached drawings.

[0090] The information (data) transmission / reception process performed in this specification may be subject to encryption / decryption as needed, and expressions describing the information (data) transmission process in this specification and patent claims should be interpreted to include cases of encryption / decryption even if not mentioned separately.

[0091] In this specification, expressions such as "transmitting (delivering) from A to B" or "A receiving from B" include cases where another medium is included in the transmission (delivery) or reception, and do not only express direct transmission (delivery) or reception from A to B. In the description of the present invention, the order of each step should be understood as non-limiting, unless a preceding step must be performed logically and temporally before a subsequent step. In other words, except for the exceptional cases above, even if a process described as a subsequent step is performed before a process described as a preceding step, the essence of the invention is not affected, and the scope of rights should be defined regardless of the order of the steps.

[0092] In this specification, “A or B” is defined to mean not only selectively referring to either A or B, but also including both A and B. Furthermore, the term “including” in this specification is intended to encompass the inclusion of additional components in addition to the elements listed as including.

[0093] This specification describes only the minimum components necessary to explain the present invention, and does not mention components irrelevant to the essence of the present invention. Furthermore, the description should not be interpreted as exclusive, encompassing only the mentioned components, but rather as non-exclusive, encompassing other components not mentioned.

[0094] In this specification, “module” or “unit” or “part” or “node” means a logical combination of general-purpose hardware and software that performs its function.

[0095] The expressions “first,” “second,” or similar used in this specification are used to distinguish between the same or similar components or to distinguish the names of components or steps constituting the present invention, and do not imply an order or plurality.

[0096] The method according to the present invention can be executed by an electronic computing device such as a computer, tablet PC, mobile phone, portable computing device, or stationary computing device. It should also be understood that one or more methods or aspects of the present invention can be executed by at least one processor. The processor can be installed in a computer, tablet PC, mobile device, or portable computing device. A memory configured to store computer program instructions can be installed in such a device, and the program can be specifically programmed to cause the processor to execute the stored program instructions, thereby executing one or more processes as described herein. It should also be understood that the information and methods described herein can be executed by a computer, tablet PC, mobile device, or portable computing device that includes one or more additional components and a processor. In addition, the control logic can be implemented as a non-volatile computer-readable medium containing program instructions executable by a processor, a control unit, or the like. Examples of computer-readable media include, but are not limited to, ROM, RAM, CD-ROM, magnetic tape, floppy disk, flash drive, smart card, or optical data storage device. Additionally, the computer-readable recording medium may be distributed across network-connected computers, such that the computer-readable medium may be stored and executed in a distributed manner, for example, by a remote server or cloud.

[0097] The exemplary embodiments described herein provide a general understanding of the structure, function, fabrication, and use of the devices and methods disclosed herein. One or more such embodiments are illustrated in the accompanying drawings. Those skilled in the art will appreciate that the devices and methods specifically described herein and illustrated in the accompanying drawings are non-limiting and exemplary embodiments, and that the scope of the present invention is defined by the claims. Features illustrated and described in connection with one exemplary embodiment may be combined with features of other embodiments. Such modifications or variations are intended to fall within the scope of the present invention.

[0098] The methods of the present invention described in this specification may be implemented and practiced independently, but may also be implemented and practiced by combining multiple methods having different characteristics.

[0099] Figure 1 illustrates an example of an environment in which a method for determining whether dynamic dispatching is possible according to the present invention is applied.

[0100] The above environment includes a user terminal (10), a server (20), a vehicle terminal (30), and a communication network (40).

[0101] A user terminal (10) is a terminal of an applicant who requests dynamic dispatch by setting at least one of the departure and arrival points to a location other than a fixed stopover on a route operated by a mobile device (e.g., a vehicle such as a bus). The user terminal (10) may be an electronic computing device such as a computer, tablet PC, mobile phone, portable computing device, or stationary computing device.

[0102] The server (20) receives a dynamic dispatch request, executes operations according to the method of the present invention, and determines whether to allow or reject the dynamic dispatch request, and notifies the dispatch result to the user terminal (10) or vehicle terminal (30).

[0103] The vehicle terminal (30) may be an electronic computing device provided in a fixed or detachable manner in the vehicle, or may be a terminal of the vehicle driver.

[0104] For the convenience of the city, the communication network (40) is depicted as a single communication network, but multiple communication networks may be used.

[0105] Figure 2 illustrates an example of a route to which the dynamic dispatching method according to the present invention is applied and a stopover added according to the dynamic dispatching.

[0106] The mobile device is assumed to run on a fixed route, and includes multiple fixed waypoints (L1, L2, L3, L4) on the route.

[0107] An applicant requesting dynamic dispatch can request dynamic dispatch through an application program installed on a user terminal (10). The first point (O), which is the starting point of dynamic dispatch, and the second point (D), which is the arrival point of dynamic dispatch, are indicated in Fig. 2 together with the fixed waypoints of the fixed route.

[0108] Figure 3 illustrates a flowchart of a method for determining whether dynamic dispatching is possible according to the present invention. The present invention will be described with reference to Figures 2 and 3.

[0109] In step (S300), the server (20) receives a dynamic dispatch request from a user terminal (10). The dynamic dispatch request includes information on a first point (O), which is a departure point, and location information on a second point (D), which is an arrival point. The dynamic dispatch request is transmitted to the server (20) via a communication network. The first point (O) and the second point (D) may be, for example, bus stops where a bus can stop, or may be arbitrary points.

[0110] The server (20) searches for a mobile device (e.g., a bus) that can be dispatched to the points based on the location information of the first point (O) and the second point (D) included in the dispatch request, and extracts information on the fixed route and fixed stopover (L1, L2, L3, L4) of the searched mobile device.

[0111] And the searched fixed waypoints (L1, L2, L3, L4) and the first point (O) and the second point (D) are aligned based on the first axis. (Step (S310)) When aligned with the x-axis as the first axis, the order becomes [L1, L2, O, D, L3, L4]. It is desirable that the order of the fixed waypoints not be changed during alignment.

[0112] In step (S320), the server (20) calculates a first waypoint vector (Vs). The first waypoint vector (Vs) is a vector of two fixed waypoints (L2, L3) that include the first point (O) and the second point (D) between them in the above-mentioned aligned state. The first waypoint vector (Vs) points from the upstream fixed waypoint to the downstream fixed waypoint.

[0113] When aligned with the first axis, if the first point (O) or the second point (D) is located outside the fixed route, the closest fixed waypoint among the fixed waypoints is selected to calculate the first waypoint vector (Vs).

[0114] In step (S330), the server (20) calculates a first demand vector (Vd). The first demand vector (Vd) points from the first point (O) to the second point (D).

[0115] In step (S340), the server (20) calculates the angle (θ) formed by the first waypoint vector (Vs) and the first demand vector (Vd). The first waypoint vector (Vs), the first demand vector (Vd), and the angle (θ) formed by the two vectors are shown at the bottom of Fig. 2. Since the method for calculating the angle between the two vectors is well known, a detailed description thereof will be omitted.

[0116] In step (S350), the server (20) determines whether the angle (θ) between the two vectors produced satisfies a predetermined first criterion. For example, if the angle (θ) between the two vectors is less than or equal to a predetermined angle, an additional operation for dynamic dispatching may be performed, or it may be determined that dynamic dispatching is possible without an additional operation.

[0117] If the result of the judgment in step (S350) is determined to meet the first criterion, the process proceeds to step (S360), where additional operations described below are performed before a final decision is made on whether to allow dynamic dispatch. Alternatively, dynamic dispatch may be permitted if the first criterion is met without additional operations.

[0118] If it is determined that the first criterion is not met, the process proceeds to step (S370) to reject dynamic dispatch and notify the user terminal (10). If dynamic dispatch is rejected, there is no need to notify the vehicle terminal (30).

[0119] FIG. 4 illustrates a flowchart of a method for determining whether dynamic dispatching is possible according to another embodiment of the present invention.

[0120] In the embodiment of FIG. 4, there is a difference from the embodiment of FIG. 3 in that the fixed waypoints (L1, L2, L3, L4) and the first point (O) and the second point (D) are aligned based on the first axis (x-axis in the example of FIG. 2) coordinate and the second axis (y-axis in the example of FIG. 2) coordinate, respectively (step S410), the second waypoint vector and the second demand point vector are calculated in a state aligned based on the coordinate of the second axis (steps S420 and S430), the second angle formed by the second waypoint vector and the second demand point vector is calculated (step S440), and if the second angle also satisfies the second criterion, an additional operation described later is executed and a final decision on whether or not to perform dynamic dispatch is made (steps S450 and S360). In the embodiment of FIG. 4 as well, dynamic dispatch may be allowed if the first criterion and the second criterion are satisfied without performing an additional operation.

[0121] When aligned based on the y-axis coordinate, which is the second axis, it is aligned as [L1, L2, O, D, L3, L4]. It is preferable not to change the operating order of fixed waypoints for the convenience of existing passengers, but the scope of the present invention is not limited to such embodiments.

[0122] According to another embodiment of the present invention, as an additional condition for determining whether dynamic dispatching is possible, it may be set whether the distance from at least one of the first point (O) and the second point (D) to the waypoint vector (or an extension of the vector) satisfies a predetermined criterion (third criterion). This distance may be the shortest distance from the point to the waypoint vector.

[0123] According to another embodiment of the present invention, a fixed waypoint and a first point and a second point are aligned based on the third axis, that is, the height direction axis coordinate of the route, and the angle between the two vectors considering the waypoint vector and the demand point vector can be used as a criterion for determining whether to perform an additional operation or whether to perform dynamic dispatch. In addition, a distance from at least one of the first point and the second point to the waypoint vector calculated when aligned based on the third axis can be added as a criterion for determining whether to perform an additional operation or whether to perform dynamic dispatch.

[0124] After determining that dynamic dispatching is possible using the above method, additional calculations can be performed to consider additional factors. For example, if the aforementioned method determines that a predetermined condition is met, a final decision on whether to allow dynamic dispatching can be made through a simulation that calculates the additional travel time, additional fuel consumption, and adjustments to the order of stops when driving to the additional stopover. However, dynamic dispatching can also be approved without these additional calculations if at least one of the first through third criteria is met.

[0125] The above-mentioned operation can be performed by calling the Routing API, for example, using OSRM.

[0126] For example, if the dispatchable criterion is set to be an increase of up to 10 minutes in the expected arrival time, and the calculation results show an increase of 11 minutes, a final decision of refusal can be made for the dynamic dispatch even if at least one of the first to third criteria is met.

[0127] To determine the optimal thresholds, which serve as the first to third criteria, learning can be done through simulations and random assignments. For example, dynamic assignments can be set for various situations, and the final dynamic assignment, determined through the aforementioned additional calculations, can be labeled as "success" or "failure" and machine learning can be used to find the optimal threshold.

[0128] According to the conventional dynamic dispatch method, when a dispatch request is received, calculations must be performed on all dispatch requests and a decision must be made on whether to dispatch, which took a long time and required a lot of resources.

[0129] However, according to the present invention, the fixed waypoint, the dynamic dispatch starting point, and the dynamic dispatch arrival point are aligned based on the coordinates of at least one axis among the first to third axes, the waypoint vector and the demand vector are calculated, and whether dynamic dispatch is possible is determined based on the distance between the two vectors, and only when dynamic dispatch is determined to be possible, additional calculations are performed or dynamic dispatch is allowed, so the time required can be drastically reduced and the resources mobilized can also be reduced.

[0130] Referring to FIGS. 5 to 8, a dynamic dispatching method considering passenger travel time according to the present invention is described.

[0131] Figure 5 illustrates an example of an environment in which the above dynamic dispatching method is executed.

[0132] The above environment includes a user terminal (310), a dynamic dispatch server (320), a machine learning engine (330), a mobile device terminal (340), and a communication network (350).

[0133] The user terminal (310) is a terminal of an applicant who requests dynamic dispatch by setting at least one of the departure point and the destination point to a location other than a fixed stopover on a route operated by a mobile device (e.g., a vehicle such as a bus). The user terminal (310) may be an electronic computing device such as a computer, tablet PC, mobile phone, portable computing device, or fixed computing device. An application program capable of requesting dynamic dispatch according to the present invention may be installed on the user terminal (310). The user who requests dynamic dispatch and the passenger who actually boards the vehicle are generally the same person, but may be different people. For example, a family member or acquaintance of the passenger may request dynamic dispatch, and the passenger may move from the departure point (C) to the boarding point (O). If the dynamic dispatch applicant and the passenger are different, the application program may request additional input of passenger status information on the user terminal (310).

[0134] The dynamic dispatch server (320) receives a dynamic dispatch request, executes calculations and the like according to the method of the present invention to determine dynamic dispatch, and notifies the dispatch result to the user terminal (310) or the mobile device terminal (340). In addition, the type of user status information or movement reference information received from the user terminal (310) or stored in advance can be provided to the machine learning engine (330) to enable prediction of movement time.

[0135] The machine learning engine (330) executes machine learning to predict the optimal travel time based on the passenger's physical condition, disability, and movement reference information described below.

[0136] The mobile device terminal (340) may be an electronic computing device provided in a fixed or detachable manner on the mobile device, or may be a terminal of the mobile device driver.

[0137] For convenience of the city, the communication network (350) is shown as a single communication network, but multiple communication networks may be used.

[0138] Figure 6 illustrates an example of a route to which the dynamic dispatching method according to the present invention is applied and a stopover added according to the dynamic dispatching.

[0139] The mobile device is assumed to run on a fixed route, and includes multiple fixed waypoints (L1, L2, L3, L4) on the route.

[0140] An applicant requesting dynamic dispatch can request dynamic dispatch through an application program installed on a user terminal (310). The first point (O), which is the boarding point where the passenger boards through dynamic dispatch, and the second point (D), which is the arrival point, are indicated in FIG. 6 along with the fixed stopovers of the fixed route.

[0141] Additionally, Figure 6 includes a starting point (C) from which passengers depart to board, which is a point where a moving device (e.g., a vehicle) cannot stop. It is assumed that passengers must move from the starting point (C) to the boarding point (O). However, the starting point (C) and the boarding point (O) may coincide. If the starting point (C) and the boarding point (O) coincide, the travel time is considered to be 0.

[0142] The time it takes for a passenger to move from the starting point (C) to the boarding point (O) varies depending on the passenger's physical condition, age, presence or absence of a disability, and if so, the type of disability (e.g., mobility impairment, visual impairment), or whether or not the passenger possesses an assistive device (e.g., wheelchair), and may also vary depending on movement reference information (e.g., temperature, rainfall, snowfall, rainfall amount, snow depth, road conditions along the movement route, the passenger's previous movement history, etc.). The status or condition of the passenger that must be considered in relation to the movement speed or movement time is referred to as "passenger status information," and other status or conditions that must be considered in relation to the movement speed or movement time of the passenger are referred to as "movement reference information."

[0143] The correlation between rider status information and movement reference information can also be considered as a variable in calculating movement speed or time. For example, multiple conditions can be considered, such as age and temperature, temperature and snowfall, or age / temperature / snowfall.

[0144] Figure 7 illustrates a flowchart of a method for determining whether dynamic dispatching is possible according to the present invention. The present invention will be described with reference to Figures 6 and 7.

[0145] In step (S3300), the server (320) receives a dynamic dispatch request from the user terminal (310). The dynamic dispatch request may include passenger status information, information on the departure point (C), and information on the arrival point. If passenger status information is previously stored corresponding to the passenger in the dynamic dispatch server (320) or a separate database, passenger identification information (e.g., ID, mobile phone number, etc.) from which passenger status information can be extracted is transmitted to the dynamic dispatch server (320), and the dynamic dispatch server (320) may also retrieve passenger status information corresponding to the transmitted identification information.

[0146] The dynamic dispatch server (320) can determine the optimal boarding point, point 1 (O), based on passenger status information and departure point (C) information (step S3310), and can determine the optimal disembarkation point, point 2 (D), based on arrival point information (step S3320). The departure point (C) and point 1 (O) may coincide, and the arrival point and point 2 (D) may coincide.

[0147] For example, the optimal first point (O) and second point (D) may be the closest stops from the passenger's starting point (C) and arrival point. Alternatively, the first point (O) and second point (D) may be determined as non-close stops based on the passenger's status information and movement reference information. For example, if the passenger is moving in a wheelchair, the first point (O) or second point (D) may be determined as a stop on a route that is advantageous for moving in a wheelchair, rather than the closest stop. In cases where a stop that is not close can be determined as the first point (O) or second point (D), it is preferable to transmit the information on the nearest stop to the user terminal (10) so that the user can select it.

[0148] According to another embodiment of the present invention, a dynamic dispatch applicant can select a preferred boarding point (O) and transmit it to a dynamic dispatch server (320). If it is determined that dynamic dispatch is not possible at the point transmitted as the preferred boarding point (O), dispatch is notified to the user terminal (310) and the dynamic dispatch server (20) can determine an optimal boarding point (O) through the aforementioned process.

[0149] In step (S3330), the travel time from the starting point (C) to the first point (O), which is the boarding point, is calculated.

[0150] Passenger status information and movement reference information are considered when calculating travel time. The base speed of the normal passenger is set, and a modified speed is calculated by incorporating parameters that take into account the passenger's status information and movement reference information. When setting the base speed, factors such as gender, age, and the time of day (season, nighttime, etc.) can be considered. Alternatively, the base travel time can be used instead of the base speed.

[0151] If the rider has a visual impairment, a walking disability, low temperature, rain, or snow, parameters can be set separately for each condition and reflected in the basic movement speed. For example, if the rider has a visual impairment, the parameter can be set to 0.3, if the rider has a walking disability, 0.5, if the rider has a walking disability, 0.9, if the rider has a low temperature, and 0.6, if the rider has a snow or rain, and the basic movement speed can be multiplied by this parameter. Based on this (slower than the basic movement speed), the travel time from the starting point (C) to the boarding point (O) can be calculated.

[0152] The dynamic dispatch server (320) searches for a mobile device that can arrive at the first point (O) within a predetermined time range (e.g., within 5 minutes after the time the passenger arrives at the first point (O)) based on the travel time calculated in step (S3320), and when a mobile device (e.g., a bus) that satisfies the conditions is searched, the dynamic dispatch server can notify the mobile device terminal (340) that the vehicle has been dispatched to stop at the first point (O) and can also notify the user terminal (310).

[0153] The dynamic dispatch server (320) can normalize data by labeling passenger status information (age, presence of disability, presence of disability assistance device, etc.) and movement reference information (weather, temperature, rainfall or snowfall, movement route status, etc.) and provide the data to a machine learning engine (330), thereby enabling prediction of an optimal movement time.

[0154] The machine learning engine (330) may be a learning engine using simple linear regression, MLP (Multilayer Perceptron), or Transformer model. When there is initially insufficient movement time data for each condition depending on whether there is a correlation between passenger status information and movement reference information, parameter values ​​can be calculated using values ​​calculated using linear regression. Conversely, when the correlation between passenger status information and movement reference information is important and a non-linear model is determined to be appropriate (e.g., correlation between temperature and age), each parameter can be calculated using Multilayer Perceptron to obtain the resulting output.

[0155] By applying an attention model to vectorize the correlation between data, we can obtain the result by applying it to the encoding and decoding model (transformer modeling). Then, we can obtain the final result through the final layer that combines the MLP and transformer results.

[0156] According to the machine learning method for predicting travel time of the present invention, a pattern of data including correlations between multiple variables can be identified by combining an existing MLP model and a transform model.

[0157] FIG. 8 illustrates an example of the machine learning structure that can be used to execute the dynamic dispatching method according to the present invention in relation to FIGS. 5 to 7.

[0158] Input values ​​such as temperature, snowfall, precipitation, and age of passengers are input into the MLP learning engine (3100) and the Transformer learning engine (3200), respectively, and the primary results of each learning engine are combined in the final MLP layer (3300) to output the final result value.

[0159] When an input value is input to the MLP learning engine (3100), the gradient and deviation values ​​of the model are learned through a backpropagation process of the error between the output result value and the actual value (label), thereby producing the output value of the MLP learning engine (100). In this process, since the adjustment of hyper parameters or environmental control is very repetitive and difficult, the primary output value can be generated under the basic settings (4 layers, 20 units). At this time, a residual neural network can be considered. The hyper parameters can be readjusted later.

[0160] In order to use the Transformer learning engine (3200), an attention pool must first be created.

[0161] When creating an attention pool, input values ​​such as temperature or snowfall can be indexed by setting them in a specific range. For example, setting the range to 5 for temperatures between -50 and 50 degrees Celsius would create 20 indices.

[0162] For example, if the temperature range is 5, and the snow depth and age are 1, the vector (8, 10, 60) can indicate the conditions of -10 degrees Celsius, 10 cm of snow depth, and 60 years of age.

[0163] The generated attention pool is applied to the Transformer learning engine (200), and the Transformer learning engine (3200) vectorizes the input data using each attention and MLP and generates the movement speed, i.e., the result value, in the decoder.

[0164] Finally, in the final MLP layer (3300), the result value of the MLP learning engine and the result value of the Transformer are combined to obtain the final result value.

[0165] Based on the obtained result values, the hyper parameters of the MLP learning engine (3100) and the Transformer learning engine (20) can be improved.

[0166] Referring to FIGS. 9 to 13, a method for determining dynamic dispatch based on a cost calculated by a multi-stage graph according to the present invention is described.

[0167] Figure 9 illustrates an example of an environment in which the above dynamic dispatching decision method is executed.

[0168] The above environment includes a user terminal (410), a dynamic dispatch server (420), a database (430), a mobile device terminal (440), and a communication network (450).

[0169] The user terminal (410) is a terminal of an applicant who requests dynamic dispatch by setting at least one of the departure and arrival points to a location other than a fixed stopover on a route operated by a mobile device (e.g., a vehicle such as a bus). The user terminal (410) may be an electronic computing device such as a computer, tablet PC, mobile phone, portable computing device, or stationary computing device.

[0170] The dynamic dispatch server (420) receives a dynamic dispatch request, executes calculations according to the method of the present invention, determines the order of dynamic dispatch points in relation to fixed waypoints, and notifies the dispatch result to the user terminal (410) or the mobile device terminal (440).

[0171] The mobile device terminal (440) may be an electronic computing device that is fixedly or detachably provided to the mobile device, or may be a terminal of the mobile device driver. The mobile device may be a bus, passenger vehicle, aircraft, ship, etc. that operates at a fixed route, and includes all mobile devices that operate to pass through a fixed route using a power device. If dynamic dispatch is successful using the method described below, the dynamic dispatch server (420) may transmit the corresponding dynamic dispatch information to the mobile device terminal (440) and the user terminal (410).

[0172] For convenience of the city, the communication network (450) is shown as a single communication network, but multiple communication networks may be used.

[0173] Figures 10 and 11 illustrate examples of routes to which a method for determining dynamic dispatch based on a cost calculated by a multi-stage graph according to the present invention is applied and stopovers added according to dynamic dispatch.

[0174] The mobile device is assumed to operate on a fixed route, with multiple fixed waypoints (L1, L2, L3, L4) along the route. It is desirable that the order of the fixed waypoints remain constant relative to one another.

[0175] An applicant requesting dynamic dispatch can request dynamic dispatch through an application program installed on a user terminal (410), and the first point (O), which is the starting point of dynamic dispatch, and the second point (D), which is the arrival point of dynamic dispatch, are displayed in FIGS. 10 and 11 together with fixed waypoints of a fixed route. The user terminal (410) is typically the terminal of a passenger who actually boards through dynamic dispatch, but may also be the terminal of a person related to the passenger (e.g., a parent, acquaintance, guardian, etc.).

[0176] The dynamic dispatching method of the present invention using a multi-stage graph is explained with reference to FIGS. 12 and 13 for the cases of FIGS. 10 and 11.

[0177] In step (S4400), the dynamic dispatch server (420) receives information on the first point (O) and the second point (D), which are the starting points of the dynamic dispatch request, from the user terminal (410).

[0178] The dynamic dispatch server (420) calculates the cost from each fixed waypoint to the first point and the second point (step S4410).

[0179] Select Point 3-1 and Point 3-2, which are the two fixed waypoints with the lowest cost to Point 1, and select Point 4-1 and Point 4-2, which are the two fixed waypoints with the lowest cost to Point 2 (step S4420).

[0180] Point 3-1 is a fixed stopover upstream of Point 3-2, and Point 4-1 is a fixed stopover upstream of Point 4-2.

[0181] The cost between Point 3-1 and Point 1 is referred to as the 1-1 cost, the cost between Point 3-2 and Point 1 is referred to as the 1-2 cost, the cost between Point 4-1 and Point 2 is referred to as the 2-1 cost, and the cost between Point 4-2 and Point 2 is referred to as the 2-2 cost.

[0182] The cost between Point 3-1 and Point 3-2 is referred to as Cost 1-3, and the cost between Point 4-1 and Point 4-2 is referred to as Cost 2-3. Costs between fixed stopovers may be recorded in the database (430).

[0183] In dynamic dispatching, any known cost calculation method can be used, and an appropriate calculation method can be selected as needed, so a detailed explanation of cost calculation is omitted.

[0184] In the embodiment of Fig. 10, it is assumed that the cost between the first point (O) and the second point (D) and the fixed waypoint is calculated as follows.

[0185] Fixed stopover L1 L2 L3 L4 Cost to point 1 351014 Cost to point 2 10755

[0186] In the embodiment of Fig. 10, the 1st-3rd cost and the 2nd-3rd cost are assumed to be 6. In the embodiment of Fig. 11, the cost between the 1st point and the 2nd point and the fixed waypoint is assumed to be calculated as follows.

[0187] L1L2L3L4 Cost to point 1 431014 Cost to point 2 131074

[0188] In the embodiment of Fig. 11, the 1-3 cost is assumed to be 7, and the 2-3 cost is assumed to be 5. The detailed process of determining dynamic dispatch in step (500) is illustrated in Fig. 13.

[0189] In step (S4505), the 1st-3rd cost and the 2nd-3rd cost are calculated or queried from the database (430). In the embodiments of FIG. 10 and FIG. 11, point 3-1 is L1 and point 3-2 is L2, so the cost between L1 and L2 is calculated or queried from the database.

[0190] In step (S4510), the first value defined as the larger value between the 1-1 cost and the 1-2 cost is compared with the 1-3 cost.

[0191] In this specification, “large,” “big,” “small,” and “small” are defined to mean that two values ​​are the same.

[0192] In step (S4515), it is determined whether the first value is less than or equal to the 1-3 cost, and if the determination result is "Yes", the process proceeds to step (S4520) to place the first point between the 3-1 point and the 3-2 point. In the embodiment of Fig. 10, the first value is 5 and the 1-3 cost is 6, so the first point is placed between L1 and L2. In the embodiment of Fig. 11, the first value is 4 and the 1-3 cost is 7, so the first point is similarly placed between L1 and L2.

[0193] If the result of the judgment in step (S4515) is "No," it is determined whether the 1-2 cost is greater than or equal to the 1-1 cost (Step S4525). If the result of the judgment is "Yes," it proceeds to step (S4530) to determine whether point 3-1 is the first fixed stopover. If point 3-1 is not the first fixed stopover, the 1st point is dispatched before point 3-1 (Step S4535). If point 3-1 is the first fixed stopover, it is determined that dispatch is not possible (Step S4575).

[0194] If the result of the judgment in step (S4525) is "No," then proceed to step (S4540) and dispatch the train from point 1 after point 3-2. Point 3-2 may be a fixed stop located in the middle of multiple fixed stopovers or may be the final fixed stopover. If it is the final fixed stopover, dispatching can be set to occur unconditionally since it occurs after all passengers on the fixed route have disembarked. However, considering cost factors such as additional driving distance, dynamic dispatch may be determined to be impossible even if point 3-2 is the final fixed stopover.

[0195] When the dispatch of the first point (O) is completed in steps (S4520, S4535, S4540), the process proceeds to step (S4545) to determine whether the second value is less than or equal to the 2-3 cost. The second value, which is defined as the larger value between the 2-1 cost and the 2-2 cost, is compared with the 2-3 cost.

[0196] If the judgment result of step (S4545) is “Yes”, the process proceeds to step (S4550) and the second point is dispatched between points 4-1 and 4-2.

[0197] In the embodiment of Fig. 10, since the second value is 7 and the 2-3 cost is 7, the 2nd point (D) can be dispatched between the 4-1 point (L3) and the 4-2 point (L4).

[0198] If the judgment result of step (S4545) is “No,” the process proceeds to step (S4555) to determine whether the 2-2 cost is greater than or equal to the 2-1 cost, and if the judgment result is “No,” the process proceeds to step (S4570) to dispatch the 2nd point after the 4-2 point.

[0199] In the embodiment of Fig. 11, since the second value is 7 and the 2-3 cost is 5, the process proceeds from step (S4545) to step (S4555), and since the 2-2 cost is 4 and the 2-1 cost is 7, the process proceeds from step (S4555) to step (S4570), so that the 2nd point (D) can be dispatched next to the 4-2 point (L4), that is, downstream of L4.

[0200] If the result of the judgment in step (S4555) is "Yes," the process proceeds to step (S4560) to determine whether point 4-1 is the first fixed stopover. If point 4-1 is the first fixed stopover, the process proceeds to step (S4575) to determine that dispatch is not possible. If point 4-1 is the first fixed stopover, dispatch may be determined to be not possible because the mobile device must travel the reverse route before moving to the first fixed stopover. However, dispatch may be set to be possible if necessary.

[0201] According to the present invention, dynamic dispatching can be performed with a simple comparison operation by calculating only the cost from each fixed waypoint to the dynamic dispatch origin / destination, thereby enabling significantly faster, higher-speed dynamic dispatching compared to prior art. For example, if a vehicle is carrying 20 passengers, prior art requires calculating 210 combinations. However, with the present invention, determining whether dynamic dispatching is appropriate requires only one transaction query and up to six comparison statements.

[0202] Referring to FIGS. 14 to 16, a dynamic dispatching method using a state machine according to the present invention is described.

[0203] Figure 14 illustrates an example of an environment in which the above dynamic dispatching method is executed.

[0204] The above environment includes a user terminal (510), a dynamic dispatch server (520), a mobile device terminal (540), and a communication network (550).

[0205] The user terminal (510) is a terminal of an applicant who requests dynamic dispatch by setting at least one of the departure point and the destination point to a location other than a fixed stopover on a route operated by a mobile device (e.g., a vehicle such as a bus). The user terminal (510) may be an electronic computing device such as a computer, tablet PC, mobile phone, portable computing device, or fixed computing device. An application program capable of requesting dynamic dispatch according to the present invention may be installed on the user terminal (510). The user who requests dynamic dispatch and the passenger who actually boards the vehicle are generally the same person, but may be different people. For example, a family member or acquaintance of the passenger may request dynamic dispatch, and the passenger may actually board and travel. If the dynamic dispatch applicant and the passenger are different, the application program may request additional input from the passenger on the user terminal (510).

[0206] The dynamic dispatch server (520) receives a dynamic dispatch request, executes calculations according to the method of the present invention to determine dynamic dispatch, and notifies the dispatch result to the user terminal (510) or the mobile device terminal (540).

[0207] The mobile device terminal (540) may be an electronic computing device provided in a fixed or detachable manner on the mobile device, or may be a terminal of the mobile device driver.

[0208] For convenience of the city, the communication network (550) is shown as a single communication network, but multiple communication networks may be used.

[0209] Figure 15 illustrates an example of a state machine used in the dynamic dispatching method according to the present invention.

[0210] The state machine (5100) is configured to be a state of a mobile device that is to travel through a fixed waypoint (5110-1 to 5110-m; V1 state, ...., V m The state machine (5100) includes a state) and a callback function (5120). The callback function (5120) executes an operation for dynamic dispatching according to the present invention. In addition to the callback function, the state machine (5100) may also include other functions necessary for executing the dynamic dispatching of the present invention in the form of a list.

[0211] The status (property) of each mobile device can include fixed waypoint information of each mobile device and whether dispatching is in progress ("true" or "false"). For example, the kth mobile device V, which is not undergoing dynamic dispatching. k The properties of can be expressed as follows.

[0212] Vk :: (waypoint: [...], dispatching: false)

[0213] Referring to FIG. 16, a dynamic dispatching method using a state machine according to the present invention is described in detail.

[0214] Figure 16 illustrates a dynamic dispatch process when there is a first dynamic dispatch request that is temporally ahead of the first dynamic dispatch request and a second dynamic dispatch request that is temporally behind the first dynamic dispatch request. The process of Figure 3 can also be extended and applied to three or more preceding and following dynamic dispatch processes.

[0215] As described above in step (S5200), a state machine is created that includes properties (states) of each mobile device, callback functions, and other necessary information.

[0216] When a first dynamic dispatch request is received in step (S5205), a first mobile device to be mobilized for the first dynamic dispatch is selected based on a first criterion (Step S5210). The first criterion may be, for example, a mobile device within a predetermined range from the location of the first dynamic dispatch requester. If multiple mobile devices are found within that range, a mobile device that is not dynamically dispatched may be selected from among them.

[0217] When the first mobile device is selected, the property of the first mobile device included in the state machine, which indicates whether dynamic dispatching is in progress, is updated to “true” (step S5215).

[0218] In step (S5260), it is determined whether a second dynamic dispatch request is received while the first dynamic dispatch request is being processed (step S5260). The steps after step (S5260) are described later.

[0219] In step (S5220), the first dynamic dispatch information is applied to the first mobile device to calculate the cost. The first dynamic dispatch information may include the departure point, destination point, dispatch request time, etc. The calculated cost may include additional travel distance due to dynamic dispatch, additional time required due to dynamic dispatch, additional fuel consumption due to dynamic dispatch, etc.

[0220] In step (S5225), a path with the lowest cost combination is selected and a first response reflecting the changed path is generated. The first response can be transmitted to the user terminal (510) and the mobile device terminal (40). The path information of the first response is reflected in the properties of the first mobile device of the state machine (5100) (step S5230).

[0221] Afterwards, the process proceeds to step (S5250) and changes the dispatch progress property of the first mobile device in the process of dispatching to “false” to put it into a dynamic dispatch reception standby state.

[0222] When a second dynamic dispatch request is received while the first dynamic dispatch is being processed, a second mobile device to be mobilized for the second dynamic dispatch request is selected based on the first criterion (step S5265).

[0223] If the second mobile device selected in step (S5265) is the same as the first mobile device, the process proceeds to step (S5235) and calls a callback function. In this case, the second dynamic dispatch may be delayed until the first response is generated.

[0224] If the second mobile device is not identical to the first mobile device, dynamic dispatching is performed for the second mobile device. In such a case, steps (S5275) to (S5290), which are dynamic dispatching processes for the second mobile device, are identical to steps (S5215) to (S5230) described above, except that they are performed for the second mobile device, and thus description thereof is omitted.

[0225] In step (S5240), the second dynamic dispatch information is applied to the first mobile device to calculate the cost. Since the cost calculation has been described in connection with step (S5220), a detailed explanation is omitted.

[0226] In step (S5245), the route with the minimum cost is selected in relation to the second dynamic dispatch and the first response is updated.

[0227] The updated first response is transmitted to the user terminal (510) and the mobile device terminal (540).

[0228] In step (S5250), the dispatch progress property of the first mobile device in the process of dispatching is changed to “false” to put it into a state of waiting for reception of a dynamic dispatch request.

[0229] Changing the dispatch progress property to “false” after dispatch is completed may also be performed before transmitting the response to the user terminal (510) and the mobile device terminal (540).

[0230] According to the present invention, by applying a state machine, fast and efficient dynamic dispatching is possible even when multiple dynamic dispatching requests are received while multiple mobile devices are in operation.

[0231] Referring to Figures 17 and 18, a destination inference method using a large language model according to the present invention is described.

[0232] Figure 17 illustrates an example of an environment in which the destination inference method is executed.

[0233] As illustrated in FIG. 17, the environment in which the destination inference method using a large language model according to the present invention is executed includes a user terminal (610), an agent module (620), a vector database (630), a language model (640), and a service server (650). In this specification, the service server (650) to which the present invention is applied is described as a transportation dispatch server.

[0234] The user terminal (610) is the terminal of the applicant who requests a ride by setting the departure point and destination. The user terminal (610) may be an electronic computing device such as a computer, tablet PC, mobile phone, portable computing device, or stationary computing device. An application program for requesting rides according to the present invention may be installed on the user terminal (610). The user requesting rides and the passenger actually boarding the vehicle are generally the same person, but may be different people. For example, a family member or acquaintance of the passenger may request rides, and a passenger other than the person requesting the ride may board at the boarding point. If the person requesting the rides and the passenger are different, the application program may request additional input of status information of the actual passenger on the user terminal (10).

[0235] The agent module (620) is a subject that executes the dispatch request recognition method according to the present invention, and performs a role such as receiving dispatch request information received from a user terminal (610), querying a vector database (630) based on the dispatch request information, querying a language model (640) based on a response received from the vector database (630), and calling a dispatch API, etc. based on a response from the language model (640).

[0236] Local knowledge can be vectorized and recorded in the vector database (630). Local knowledge refers to specialized information about a service. For example, in the case of a dispatch service, this may include dispatch type, dispatch service method, regional classification, age- or gender-specific preferred destinations, region-specific preferred destination information, destination location information, etc. Typically, this refers to service-specific knowledge not included in language models. The vector database (630) is included as knowledge in the language model.

[0237] The agent module (620) can collect specialized information by, for example, executing batch jobs on users' usage logs at specific intervals.

[0238] The language model (640) may be a large language model (LLM) or a small language model (sLLM).

[0239] Figure 18 illustrates a flowchart of a destination inference method using a large language model according to the present invention.

[0240] In step (S6200), the user terminal (610) requests dispatch to the agent module (620) based on a dispatch request input by the user. The dispatch request may be input by executing a dispatch service application program installed on the user terminal (610) and transmitted to the agent module (620). The dispatch request may include destination information, which is referred to as first dispatch request information. The destination information included in the first dispatch request information may be a specific destination name, but in an embodiment to which the present invention is applied, it may be destination information that is difficult to identify the destination by itself.

[0241] For example, destination information included in the first dispatch request information may be "hospital," "mart," "son's house," etc. The present invention can be applied in situations where a user does not have a specific destination address or name that can immediately identify the location of the destination when making a dispatch request.

[0242] When a dispatch service application program is executed, a voice or text requesting a destination may be displayed on the user terminal (610). If recognition of the user's voice or text input fails, a list of destinations (e.g., home, supermarket, hospital, son's house, etc.) customized based on the user's identification information or users belonging to a similar category may be displayed for the user to select.

[0243] The agent module (620) generates a query based on the second dispatch request information that adds personalized information of the user (or the passenger, if the user and the passenger are different) to the first dispatch request information, and transmits the query to the vector database (step S6210). The personalized information of the user includes information previously stored about the user (or passenger), such as location information, age information, gender, and the type of disability, if any. Alternatively, the personalized information may be included in the first dispatch request information. For example, when a first-time user or a user who has not agreed to the storage of personalized information makes a dispatch request, the personalized information entered may be requested and used.

[0244] The vector database (630) generates a response regarding third dispatch request information that has a predetermined correlation or higher with the second dispatch request information (step S6230) and transmits the generated response to the agent module (620) (step S6240). The correlation level can be set by the system designer. For example, if the second dispatch request information includes a specific destination name, the location information of the destination can become the third dispatch request information.

[0245] The agent module (620) generates a final query based on the first dispatch request information to the third dispatch request information (step S6245).

[0246] For example, if the first dispatch request information is "to the hospital", the second dispatch request information is "60s, female, current time 10:00 AM, current location Mapo-gu, most recently visited hospital is XX Hospital, three visits history", and the third dispatch request information is "XX Hospital's location information", the content of the final query generated by the agent module (620) based on the first dispatch request information or the third dispatch request information may be "A woman in her 60s wants to go to a nearby hospital at 10:00 AM. The hospital she previously visited is XX Hospital, which is 10 km away, and she has visited it three times. Please find a possible hospital and guide her there."

[0247] The language model (640) that receives the final query generates a response (step S6260) and transmits the generated response to the agent module (620) (step S6270).

[0248] In the above example, the language model (640) can search for available hospitals using specialized knowledge recorded in the vector database and generate a response by judging distance, frequency of visits, etc. by referring to previous visit history.

[0249] In the above example, an example response could be, "The nearest hospital is YY Hospital, and the last hospital you visited was XX Hospital. Where would you like to go?"

[0250] The agent module (620) generates a confirmation request message based on the received response (step S6275). For example, based on the example of the response above, a confirmation request message such as "If you want to be dispatched to the nearest hospital YY, enter 1. If you want to be dispatched to the hospital XX, which is the hospital you visited last time, enter 2" can be generated.

[0251] The generated confirmation request message is transmitted to the user terminal (610) (step S6280), and the user terminal (610) transmits the destination confirmation message selected and entered by the user to the agent module (620) (step S6290).

[0252] The agent module (620) extracts destination information from the received confirmation message (step S6291) and calls the dispatch API so that the vehicle can be dispatched with the corresponding destination information (step S6292).

[0253] According to the present invention, even if a destination that is difficult to specify is entered when requesting a dispatch, the dispatch request is recognized by querying a language model with a query including a response received by adding the passenger's personalized information to a vector database in which specialized knowledge is recorded, and receiving the response from the language model, so that an accurate and personalized dispatch service can be provided even if the destination is not entered accurately.

[0254] Referring to Figures 19 to 21, a method for changing a route through traffic prediction according to the present invention is described.

[0255] Figure 19 is a block diagram of a real-time road speed-based traffic prediction system according to the present invention.

[0256] Referring to FIG. 19, a real-time road speed-based traffic prediction system according to an embodiment of the present invention stores the average driving speed of vehicles driving on the road by link in a preset area unit as an image, and performs reinforcement learning on traffic events that occur according to changes in the image, and then predicts traffic events that will occur after the image of the current speed for the area, and adjusts the vehicle's route or estimated time of arrival (ETA) according to the predicted event and provides the traffic prediction to the vehicle module.

[0257] It may include a server (710) and a vehicle module (20) that transmits information on the starting point and destination to be driven and information on the current location and speed of the vehicle to the traffic prediction server.

[0258] The traffic prediction server (710) may include a speed data collection unit (7100) that collects speed data of vehicles driving on each preset link and stores it in a database, an event learning unit (7200) that converts the speed data stored in the database into differentiated images that enable visual distinction between low and high speeds, and stores changes in the speed data of each link according to events that occur as traffic conditions progress by performing reinforcement learning (Reinforcement Learning basics) based on changes in the images in the database, a regional event prediction unit (7300) that predicts traffic events expected to occur in response to the current location of a vehicle and the real-time driving speed at that location based on the reinforcement learning data stored in the database, and generates a prediction image reflecting the predicted event and stores it in the database, and a prediction event application unit (7400) that changes the current travel route to a route that bypasses an area where a congestion event is predicted based on the predicted event or changes the estimated time of arrival (ETA) to change and store an optimal route for reaching a destination in real time.

[0259] The above speed data collection unit (7100) can obtain real-time speed data for each road by obtaining public data from a city traffic information center connected to a communication module provided in the traffic prediction server (710).

[0260] At this time, the speed data collection unit (7100) may collect real-time speed data by directly receiving speeds from a number of vehicles driving on an actual road, but it is difficult to secure representativeness of the speed of each vehicle, so there is the inconvenience of having to calculate the average of the speed data of vehicles located at each link.

[0261] Accordingly, it is desirable for the speed data collection unit (7100) to obtain real-time speed data for each link of each road based on public data provided by a public institution such as a city traffic information center, so as to secure the representativeness of the collected real-time speed data.

[0262] In addition, the event learning unit (7200) may include a speed data visualization unit (7210) that visualizes the speed data of each road stored in the database as differentiated images that can distinguish speed states such as congestion, slow-moving, and smooth, a unit time image storage unit (7220) that stores the speed data of each road visualized as differentiated images at preset unit times and sets an environment (evn) to be applied to reinforcement learning (Reinforcement Training) based on images, and an image-based reinforcement learning unit (7230) that learns a path that can maximize a reward in each environment while imposing a preset reward or penalty in response to the action taken by a vehicle driving on a road on which speed data is visualized at preset unit times, i.e., the selection of a road connected to the driving link.

[0263] At this time, the speed data visualization unit (7210) can simplify the environment applied to reinforcement learning by differentiating and storing the colors of each road image using speed data. Accordingly, the speed data visualization unit (7210) can differentiate and store the colors of the roads by expressing a congested area (e.g., less than 40 km / h) with a first color (e.g., red), a slow-moving area (e.g., 40 to 80 km / h) with a second color (e.g., yellow), and a smooth area (e.g., 80 km / h or more) with a third color (e.g., green).

[0264] The above-described speed classification and color selection are only examples of what can be adopted in the speed data visualization unit (7210), and it is of course possible to set the speed in more detail or to set more diverse colors.

[0265] The above unit time image storage unit (7220) can store visualized speed data generated by the speed data visualization unit (7210) as a new image in a database every preset unit time (e.g., 1 second, 1 minute, 1 hour, etc.).

[0266] In addition, the unit time image storage unit (7220) may label each image to be stored with information about which road and which time the speed image is displayed on, and then store the images together in a database.

[0267] By labeling the images stored in this manner, the administrator of the traffic prediction server can use it to manage the environment and past data applied to reinforcement learning.

[0268] The image-based reinforcement learning unit (7230) can perform reinforcement learning using image-based speed data stored in a database. Accordingly, the image-based reinforcement learning unit (7230) can learn to derive a path that maximizes rewards by paying preset rewards or imposing penalties as feedback, taking into account the impact of actions selected by the agent vehicle during reinforcement learning, i.e., the speed data of the road or link the vehicle enters.

[0269] As an example, the image-based reinforcement learning unit (7230) can learn the movement path of a vehicle driving in an area by imposing a penalty when the vehicle passes through an area of ​​a first color (e.g., a red image) set as a congested area when proceeding to the next road or link based on the speed and location information of the vehicle transmitted from the vehicle module (20), and by imposing a reward when the vehicle passes through an area of ​​a third color (e.g., a green image) set as a smooth area.

[0270] At this time, the image-based reinforcement learning unit (7230) may be set to impose differential compensation according to the image color of the area through which the vehicle passes, if the image color of the speed data is set to be diverse.

[0271] For example, passing through the first color area could incur a penalty of -1, passing through the second color area could incur a penalty of 0, and passing through the third color area could incur a penalty of +1. Of course, the points awarded could also be increased or decreased depending on the diversity of the image's colors.

[0272] In addition, as another example, reinforcement learning can be performed by imposing various rewards based on the change in the image color of the region that the vehicle enters while driving on the road, such as imposing a reward of +2 when entering a third color region from a first color region, and imposing a reward of -2 when entering a first color region from a third color region.

[0273] In addition, the image-based reinforcement learning unit (7230) can be used as data for reinforcement learning to derive an optimal travel route by imposing a certain reward when a vehicle newly entering the road or link achieves the highest speed, i.e., the fastest estimated time of arrival (ETA), while driving along a travel route between the registered departure point and destination.

[0274] The above regional event prediction unit (300) can predict traffic events that may occur while a vehicle is driving in each region or link based on real-time speed data of the road currently being driven based on results derived through reinforcement learning.

[0275] For example, if real-time speed data is collected with a first color indicating congestion, it is possible to predict traffic conditions in the area, such as predicting that congestion will persist or worsen for a certain period of time due to an increase in vehicles entering the road.

[0276] And the above prediction event application unit (7400) can apply the predicted traffic situation event based on the learning result of the image-based reinforcement learning unit to the optimal route for reaching the destination set by the vehicle module, change and store the optimal route, or adjust the estimated time of arrival (ETA) and provide it to the vehicle module.

[0277] In order to perform such reinforcement learning, the event learning unit (7200) collects node-link data for a target area where reinforcement learning is to be performed, as shown in the flow chart illustrated in FIG. 21, and models and stores the environment (env), state, action, and reward for performing reinforcement learning based on the collected node-link data.

[0278] And the above event learning unit (7200) sets an algorithm (e.g., Djikstra Algorithm) for searching the shortest path from the current location to the destination (target position) as the target of reinforcement learning.

[0279] And, based on the image that visualizes and stores real-time speed data of vehicles driving on the road of the target area obtained from the speed data collection unit (7100) for each unit of time, reinforcement learning (Reinforcement Training) is executed in the image-based reinforcement learning unit (7230).

[0280] The image-based reinforcement learning unit (7230) derives a TQS (Trained Quality Set) that reflects the results of reinforcement learning, and the regional event prediction unit (7300) applies the derived TQS to predict events that will occur continuously in the future based on speed information in the relevant region, and the predicted event application unit (400) searches for an optimal path to reach the destination based on the current speed in the relevant region and the predicted event.

[0281] Thereafter, when a new route request to reach a destination (target location) is made from the vehicle module, the prediction event application unit (300) searches for an optimal route based on a prediction image reflecting traffic prediction for a certain period of time (e.g., 10 minutes after the current time) by applying TQS reflecting the result of reinforcement learning, and then transmits the search result to the vehicle module.

[0282] In this way, the traffic prediction server does not set an estimated time of arrival (ETA) based on the current speed measured on each road or link along the travel route, but generates an image based on the real-time speed for the road corresponding to the current location of the vehicle, but provides a travel route based on a predicted image that reflects the learning results of the image-based reinforcement learning unit for the roads or links along the travel route to be driven thereafter, thereby increasing the efficiency of travel route search and improving the accuracy of the estimated time of arrival (ETA).

[0283] In addition, when the road speed reached while the vehicle is driving is different from the speed in the predicted image, a new predicted image corresponding to the changed speed is generated and provided, so that even if there is a section where delays are long due to an unexpected traffic event, the optimal travel route can be searched by quickly bypassing the delayed section based on the predicted image.

[0284] While driving to the destination along the optimal route transmitted in this way, if the speed data of the road on which the vehicle is driving differs from the traffic forecast, it is of course possible to modify the route to temporarily change the destination in order to achieve the highest speed.

[0285] As described above, unlike conventional reinforcement learning that eliminates uncertainty based on complex algorithms that include diverse and detailed environments (env) in modeling for machine learning, the present invention can reduce the computational burden of reinforcement learning by simplifying it to a simplified environment (env) as an image of real-time velocity data.

[0286] In addition, the present invention stores all processes in which traffic conditions are generated, propagated, spread, and then organized based on real-time speed changes in a specific area in a database as image changes of speed data.

[0287] Accordingly, when deriving an optimal route for the vehicle to reach the same or adjacent destination, the next image, such as congestion, slow-motion, or smooth driving, that will occur in each link can be predicted based on the current speed data of the vehicle, and thus a route can be derived that avoids or bypasses this.

[0288] Furthermore, the present invention can learn about the type of diffusion that will occur in the future based on traffic conditions in a specific area, for example, the type in which a first color, which has started a traffic jam, spreads along a specific link over time. Accordingly, the present invention can be combined with systems that perform reinforcement learning based on images of real-time speed data, as well as reinforcement learning based on prior algorithms using past data, for optimal route or traffic event prediction.

[0289] Next, with reference to FIGS. 20 and 21, a real-time road speed-based traffic prediction method according to another embodiment of the present invention will be described.

[0290] FIG. 20 is a configuration diagram of a traffic prediction method based on real-time road speed according to the present invention, and FIG. 21 is a flow chart showing a process of predicting traffic according to the present invention.

[0291] Referring to FIGS. 20 and 21, a real-time road speed-based traffic prediction method according to another embodiment of the present invention may include a speed data collection step (S710) of collecting speed data of vehicles driving on each preset link and storing the data in a database, an event learning step (S720) of converting the speed data stored in the database into differentiated images that enable visual distinction between low and high speeds, and storing changes in the speed data of each link according to events that occur as traffic conditions progress by performing reinforcement learning (Reinforcement Learning basics) based on changes in the images in the database, a regional event prediction step (S730) of predicting traffic events expected to occur in response to the current location of a vehicle and the real-time driving speed at that location based on the reinforcement learning data stored in the database, generating a prediction image reflecting the predicted event, and storing the image in the database, and a prediction event application step (S740) of changing the current travel route to a route that bypasses an area where a congestion event is predicted based on the predicted event or changing the estimated time of arrival (ETA) to change and store an optimal route for reaching a destination in real time.

[0292] The above speed data collection step (S710) can obtain real-time speed data for each road by obtaining public data from a city traffic information center accessed by a communication module equipped in a traffic prediction server.

[0293] In addition, the event learning step (S720) may include a speed data visualization step (S721) for visualizing the speed data of each road stored in the database as differentiated images that can distinguish speed states such as congestion, slow-moving, and smooth, a unit time image storage step (S722) for storing the speed data of each road visualized as differentiated images at preset unit times and setting an environment (evn) to be applied to reinforcement learning (Reinforcement Training) based on images, and an image-based reinforcement learning step (S723) for learning a path that can maximize the reward in each environment while imposing a preset reward or penalty in response to the action taken by a vehicle driving on a road on which speed data is visualized at preset unit times, i.e., the selection of a road connected to the driving link.

[0294] At this time, in the speed data visualization step (S721), the environment applied to reinforcement learning can be simplified by differentiating and storing the colors of each road image using the speed data. Accordingly, in the speed data visualization step (S721), the colors of the roads can be differentiated and stored by expressing a congested area (e.g., less than 40 km / h) with a first color (e.g., red), a slow-moving area (e.g., 40 to 80 km / h) with a second color (e.g., yellow), and a smooth area (e.g., 80 km / h or more) with a third color (e.g., green).

[0295] In the above unit time image storage step (S722), the visualized speed data generated in the speed data visualization step can be stored in a database as a new image every preset unit time (e.g., 1 second, 1 minute, 1 hour, etc.).

[0296] Additionally, in the above-mentioned unit time image storage step (S722), each image to be stored can be labeled to indicate which road and which time the speed image is, and then stored together in the database.

[0297] By labeling the images stored in this manner, the administrator of the traffic prediction server can use it to manage the environment and past data applied to reinforcement learning.

[0298] In the image-based reinforcement learning step (S723), reinforcement learning can be performed using image-based speed data stored in the database. Accordingly, in the image-based reinforcement learning step (S723), while reinforcement learning is in progress, the agent vehicle's chosen action, i.e., the impact on the speed data of the road or link the vehicle enters, can be considered, and a preset reward or penalty can be imposed as feedback to learn how to derive a path that maximizes the reward.

[0299] As an example, in the image-based reinforcement learning step (S723), based on the speed and location information of the vehicle transmitted from the vehicle module, when the vehicle proceeds to the next road or link, a penalty is imposed if the vehicle passes through an area of ​​a first color (e.g., a red image) set as a congested area, and a reward is imposed if the vehicle passes through an area of ​​a third color (e.g., a green image) set as a smooth area, thereby learning the movement path of the vehicle driving in the area.

[0300] At this time, in the image-based reinforcement learning step (S723), if the image color of the speed data is set to be diverse, it may be set so that differential compensation is applied according to the image color of the area through which the vehicle passes.

[0301] For example, passing through the first color area could incur a penalty of -1, passing through the second color area could incur a penalty of 0, and passing through the third color area could incur a penalty of +1. Of course, the points awarded could also be increased or decreased depending on the diversity of the image's colors.

[0302] As another example, reinforcement learning can be performed by imposing various rewards based on the change in the image color of the region that the vehicle enters while driving on the road, such as imposing a reward of +2 when entering a third color region from a first color region, and imposing a reward of -2 when entering a first color region from a third color region.

[0303] In addition, in the image-based reinforcement learning step (S723), when a vehicle newly entering the road or link drives along a registered departure and arrival route and achieves the highest speed, i.e. the fastest estimated time of arrival (ETA), a certain reward is imposed, thereby utilizing the data for reinforcement learning to derive an optimal travel route.

[0304] In the above regional event prediction step (S730), based on the results derived through reinforcement learning, events that may occur while the vehicle is driving in each region or link can be predicted based on real-time speed data of the road on which the vehicle is currently driving.

[0305] For example, if real-time speed data is collected with a first color indicating congestion, it is possible to predict traffic conditions in the area, such as predicting that congestion will persist or worsen for a certain period of time due to an increase in vehicles entering the road.

[0306] And, in the predicted event application step (S740), the predicted traffic situation event based on the learning result in the image-based reinforcement learning step is applied to the optimal path for reaching the destination set by the vehicle module to change and store the optimal path or adjust the estimated time of arrival (ETA) and provide it to the vehicle module.

[0307] In this way, for the image-based reinforcement learning, as shown in the flow chart illustrated in Fig. 21, node-link data for a target area where reinforcement learning is to be executed are collected (S110), and the environment (env), state, action, and reward for executing reinforcement learning are modeled (S7120) and stored based on the collected node-link data.

[0308] And, an algorithm (e.g., Djikstra Algorithm) for searching the shortest path from the current location to the destination (target position) is set as the target of reinforcement learning (S7130).

[0309] Then, real-time speed data of vehicles driving on the road of the targeted area is collected (S7210), the collected speed data is visualized at each unit time and an image is saved (S7220), and reinforcement learning is executed based on the image of the visualized speed data (S7230).

[0310] A TQS (Trained Quality Set) reflecting the results of this reinforcement learning is created (S7240), and the derived TQS is applied to search for an optimal path to reach the destination (S7250).

[0311] Thereafter, when a new route request (S7310) to reach a destination (target location) is received from the vehicle module and the vehicle departs (S7320) toward the destination, the TQS reflecting the result of the reinforcement learning is applied based on the real-time road speed at the current location of the vehicle, and the optimal route is searched based on a predicted image reflecting the traffic prediction for a certain period of time in the future (for example, 10 minutes after the current point in time), and then transmitted to the vehicle module (S7330).

[0312] In addition, when the road speed reached while the vehicle is driving is different from the speed in the predicted image, a new predicted image corresponding to the changed speed is generated and provided, and a new optimal route based on the new predicted image is suggested (S7340), so that even if there is a section where delay is long due to an unexpected traffic event, an optimal travel route that quickly bypasses the delayed section can be searched for based on the predicted image.

[0313] The seat assignment system and method according to the present invention will be described with reference to FIGS. 22 to 26.

[0314] The seat assignment system for a reservation-based transportation vehicle according to various embodiments of the present invention minimizes confusion among passengers boarding and alighting and provides convenient seating by assigning seats based on the boarding and alighting locations of passengers on a Mobility On Demand (MOD) bus or reservation-based transportation vehicle available to multiple users. While the following description uses an MOD bus as an example for clarity, the system can be applied to any transportation service that allows multiple users to use the vehicle through reservation.

[0315] Referring to FIG. 22, the seat assignment system (hereinafter referred to as the “system”) of the reservation-based transportation means of the present invention may include a seat information registration unit (8110), an interface unit (8120), and a seat assignment unit (8130), and may be configured with a server (8100). Alternatively, the system may be configured with a server (8100), an administrator terminal (8200), a user terminal (8300), and a database (8400).

[0316] The seat information registration unit (8110) can receive seat information including the door location, seat location from the door, and seat type for each vehicle in operation from the administrator terminal (8200) and store the information in the DB (8400).

[0317] Here, the seat type can be one of the following: socially disadvantaged seats, general seats for short distances, or general seats for long distances. Furthermore, seat types by seat location can be assigned in order of proximity to the door by a preset number of seats. Alternatively, the administrator can assign seat types by seat location. The preset number of seats by seat type and the seat type by seat location can be varied.

[0318] The interface unit (8120) can receive a boarding request from a user terminal (8300) and provide an interface for receiving boarding request information including the departure point, destination, departure time, and whether the user is a socially disadvantaged person from the user terminal (8300).

[0319] Additionally, you can input more target time.

[0320] At this time, MOD buses can be dispatched based on boarding request information. For example, MOD buses can be dispatched in real-time / dynamically based on the detour time of existing passengers, the likelihood of the dispatch requester meeting the arrival time, and the likelihood of existing passengers meeting the arrival time when the dispatch requester boards, among the available dispatchable vehicles searched based on boarding request information.

[0321] When a MOD (Mobility On Demand) bus capable of assigning seats is dispatched in response to a user's boarding request, the seat assignment unit (8130) can assign seats to each user who requested boarding based on the distance traveled by the MOD bus from the departure point to the destination included in the boarding request, the boarding time the user boards during the distance, and the seat information of the dispatched MOD bus.

[0322] Figure 23 is a block diagram showing a schematic configuration of the seat assignment unit (8130) of Figure 22. Referring to Figure 23, the seat assignment unit (8130) may include a seat management unit (8131), an evaluation index calculation unit (8132), and a seat determination unit (8133).

[0323] The seat management unit (8131) can manage and store seat management information (whether a seat is occupied, whether a seat is reserved, and user information regarding use / reservation) for each vehicle based on the user's boarding and alighting in the DB (8400). In addition, the seat management unit (8131) can manage and store seat management information for each vehicle registered by the administrator terminal (8200).

[0324] When a user boarding request is received, the seat management unit (8131) can determine whether seat assignment is possible by checking the remaining seats in the candidate vehicles available for dispatch. In other words, the MOD bus dispatched in response to the user boarding request may be a vehicle determined by the seat management unit (8131) to be eligible for seat assignment among the candidate vehicles available for dispatch based on the departure point and destination.

[0325] At this time, the seat management department (8131) determines whether seat assignment is possible if there is a remaining seat type corresponding to the socially disadvantaged person included in the user's boarding request, and if there is no such seat type, it can determine whether seat assignment is possible by checking the remaining seats of other available candidate vehicles.

[0326] The evaluation index calculation unit (8132) can determine available seats based on the seat management information of the seat management unit, and can calculate an evaluation index for each user based on the boarding time and travel distance of at least one user scheduled to board a MOD bus dispatched according to a boarding request, and the remaining travel distance and remaining boarding time of passengers who have already boarded the bus.

[0327] More specifically, the evaluation index calculation unit (8132) may include a travel distance and boarding time calculation engine and a seat assignment calculation engine.

[0328] The distance traveled and boarding time calculation engine can calculate the boarding time and distance traveled for users who have requested boarding for each vehicle (dispatched MOD bus) based on the departure point and destination included in the boarding request information.

[0329] Additionally, the seat assignment calculation engine can calculate user-specific evaluation metrics to determine seat assignment priorities. These user-specific evaluation metrics can be calculated based on the user-specific boarding time and travel distance calculated by the travel distance and boarding time calculation engine.

[0330] For example, the user-specific evaluation index can be calculated as follows. Here, for passengers who have already boarded, the travel distance can be applied as the remaining travel distance, and the boarding time can be applied as the remaining boarding time.

[0331] (Equation 1)

[0332] Evaluation index = (boarding time (minutes) * α) + (travel distance (km) * β)

[0333] Here, α is the boarding time weight, and β is the moving distance weight. For example, α can be set to 0.7 and β can be set to 0.3, and can be changed by the administrator.

[0334] The seat determination unit (8133) sorts the user-specific evaluation indices calculated by the evaluation indices calculation unit (8132) in descending order, and assigns a seat closest to the door of the corresponding MOD bus among the available seats to a user with a lower evaluation indices. For example, if there are no reserved passengers on the assigned MOD bus, the user's seat may be assigned to the location closest to the door, and if there are already reserved passengers or users, the seat may be assigned to the location closest to the next door.

[0335] At this time, the seat determination unit (8133) determines the seat type based on the boarding information, whether the passenger is socially disadvantaged and whether the distance traveled is short or long, and then determines the seats that can be assigned in the corresponding seat type, and may assign seats of the seat type that are available for assignment and are close to the door.

[0336] In addition, if there are users with the same evaluation criteria, the seat decision unit (8133) can sequentially compare them in the order of shortest travel distance, shortest boarding time, and boarding request, and assign seats closer to the door.

[0337] As an example, seat assignment can be explained with reference to Fig. 24. Fig. 24 is a drawing for explaining a seat assignment method of a seat assignment unit according to one embodiment of the present invention.

[0338] Referring to Figure 24, seat information registered in the seat registration section (8110) can be confirmed. Each seat can be assigned a seat number, etc. based on its location from the door, and seat types (810, 820, 830) can be set based on convenience of use, such as location from the door, window location, and aisle location. The settings in Figure 24 are merely examples for illustrative purposes and can be modified and designed in various ways by the administrator depending on convenience of use.

[0339] For example, during times of epidemics, seating arrangements may be adjusted to accommodate different intervals.

[0340] Here, the seat type can be set as a socially disadvantaged seat (810), a general seat for short distances (820), a general seat for long distances (830), etc., and the user seats that have already been assigned are marked with an X. In addition, the judgment of long distances and short distances can be determined by comparing them with a preset standard distance.

[0341] At this time, the seat assignment unit (8133) arranges seats in descending order based on the evaluation index. If an aisle seat is assigned before a window seat, or if the user is socially disadvantaged, seat number 6 may be assigned to the user. Alternatively, if the user is a general short-distance user, seat number 7 of the short-distance general seats (820) may be assigned. Meanwhile, if all short-distance seats have already been assigned, even a short-distance user may be assigned to a long-distance general seat (830).

[0342] Meanwhile, for users with the same evaluation criteria, priority may be given to the user with the shorter travel distance. If the travel distance is the same, priority may be given to the user with the shorter boarding time. Furthermore, if the boarding time is the same, priority may be given to the user who requested boarding first.

[0343] Additionally, the seat determination unit (8133) can assign seats to users scheduled to board based on statistical data by travel time period, based on the boarding and alighting points and times of each existing bus user, stored in the database (8400). At this time, seat assignment is based on an evaluation index, and the probability of other expected boarding users, as well as the departure and destination points of the other expected boarding users, can be predicted based on statistical data corresponding to the travel time from the departure point to the destination of the user requesting boarding.

[0344] At this time, the seat determination unit can determine whether the other expected boarding user will maintain boarding even when the boarding request user gets off based on the departure point and destination of the other expected boarding user, and determine the seat of the boarding requesting user based on whether the other expected boarding user will maintain boarding.

[0345] In addition, the seat determination unit may assign a seat to the user requesting boarding that is closer to the door of the dispatched bus than to the other user expected to board, when the other user expected to board continues to board even when the user requesting boarding gets off.

[0346] In other words, even if there are many empty seats, users who are expected to have high demand during their travels and who statistically travel long distances can be assigned window seats. Meanwhile, even if demand is expected to be low during the same time period and users who statistically travel long distances are expected to have long distances, aisle seats can be assigned based on demand forecasts.

[0347] Figure 25 is a flowchart illustrating a seat assignment method of a seat assignment system according to one embodiment of the present invention. Figure 26 is a flowchart illustrating the step (S8130) of assigning seats to users in Figure 25. The seat assignment methods of Figures 25 and 26 can be explained based on the seat assignment systems of Figures 22 to 24.

[0348] First, the seat information registration unit (8110) can receive vehicle-specific seat information from the management terminal (8200) (S8100). At this time, the vehicle-specific seat information may include door location, seat location relative to the door, and seat type.

[0349] Next, when a user boarding request is received (S8105), the seat management information for the dispatchable candidate vehicles based on the departure and destination included in the boarding request is checked (S8110), thereby determining whether the vehicle is eligible for seat assignment (S8115). In other words, the availability of available seats in the dispatchable candidate vehicles can be determined by checking the remaining seats.

[0350] Specifically, when there is a remaining seat type corresponding to the socially disadvantaged person included in the user's boarding request, it is determined that seat assignment is possible (S8115, Y) and a candidate vehicle for which seat assignment is possible is dispatched (S8120). If the corresponding seat type is not available (S8115, N), another dispatcher can check the seat management information of the candidate vehicle to confirm the remaining seats (S8125).

[0351] Next, when a MOD (Mobility On Demand) bus with assigned seats is dispatched (S8120), the seat assignment unit (8130) can assign seats to each user based on the distance traveled by the MOD bus from the departure point and destination included in the boarding request, the boarding time the user boards during the distance, and the seat information of the dispatched MOD bus (S8130). In addition, seat assignments can be made based on information from the seat management unit (8131).

[0352] Specifically, referring to FIG. 26, the evaluation index calculation unit (8132) can calculate the boarding time and travel distance for at least one user scheduled to board the MOD bus and the remaining boarding time and travel distance for a user who has already boarded the bus (S8131), and can calculate the user-specific evaluation index according to the mathematical expression 1 above based on the calculated boarding time and travel distance for each user (S8132).

[0353] Next, the seat determination unit (8133) can sort the user-specific evaluation indices calculated in step S8132 in descending order, and assign seats closer to the MOD bus doors in descending order of user-specific evaluation indices (S8133). Furthermore, seat assignments can be made by differentiating by seat type, as described in the embodiment of FIG. 24.

[0354] Meanwhile, if there are users with the same usage-specific evaluation index calculated in step S8132 (S8134, Y), seats can be assigned in order of shortest travel distance by comparing travel distances (S8135). If travel distances are also the same (S8136, Y), seats can be assigned in order of shortest travel time by comparing boarding times (S8137).

[0355] Additionally, if the boarding time is the same (S8138, Y), the order of boarding requests can be compared to determine seat assignment in the order of boarding requests made first (S8139).

[0356] In addition, the seat assignment steps S8133 to S8139 of the user-specific seat assignment step (S8130) may assign available seats close to the door by considering the seat type based on whether the user included in the boarding request is socially disadvantaged and whether the departure and destination are short-distance or long-distance.

[0357] As a result, the seat assignment system of the reservation-based transportation means of the present invention and the seat assignment method thereof calculate an evaluation index using the travel distance and boarding time based on the boarding location and location of the passenger (user), and give priority to assigning seats closer to the door to passengers with lower evaluation indexes, thereby providing efficient boarding and alighting services, enabling provision of quality services by designating seats for the socially disadvantaged, and operating by setting seat assignment intervals in the case of a period when a specific infectious disease is prevalent.

[0358] While the present invention has been described with reference to the attached drawings, the scope of the present invention is determined by the claims described below and should not be construed as being limited to the aforementioned embodiments and / or drawings. It should also be clearly understood that improvements, modifications, and variations apparent to those skilled in the art, as defined in the claims, are also included within the scope of the present invention.

Claims

1. A method for determining whether dynamic dispatching is possible for a route including multiple fixed waypoints executed by a computer, A first step of receiving a first dispatch request from the applicant's terminal to depart from a first point and arrive at a second point, A second step of aligning multiple fixed waypoints, the first point, and the second point based on the first axis coordinate, A third step of calculating the first waypoint vector for two fixed waypoints surrounding the first and second points, Step 4: Calculating the first demand vector for the first and second points; A fifth step comprising calculating a first angle formed by a first waypoint vector and a first demand point vector, and approving dynamic dispatch or requesting additional operations required for dynamic dispatch if the first angle meets a predetermined first criterion. How to determine whether dynamic dispatching is possible.

2. In claim 1, Step 2-1 of aligning multiple fixed waypoints, the first point, and the second point based on the second axis coordinate, Step 3-1 for calculating the second waypoint vector for two fixed waypoints surrounding the first and second points, Includes step 4-1 of calculating the second demand vector for the first and second points, Step 5 is a step of calculating the second angle formed by the second waypoint vector and the second demand point vector, and approving the dispatch request by additionally considering whether the second angle meets the second criterion. How to determine whether dynamic dispatching is possible.

3. In claim 1, A sixth step further comprising approving dynamic dispatching or requesting additional operations required for dynamic dispatching by additionally considering whether the shortest distance from at least one of the first point and the second point to the first waypoint vector or a line extending the first waypoint vector satisfies a predetermined third criterion. How to determine whether dynamic dispatching is possible.

4. In claim 3, A seventh step further comprising approving dynamic dispatching or requesting additional operations required for dynamic dispatching by further considering whether the shortest distance from at least one of the first point and the second point to the second waypoint vector or a line extending the second waypoint vector satisfies a predetermined third criterion. How to determine whether dynamic dispatching is possible.

5. In claim 1, Step 1-1 of transmitting a query based on second dispatch request information including passenger personalized information in the information included in the first dispatch request received in the first step to the vector database; Step 1-2 of receiving a response regarding third dispatch request information having a relationship with second dispatch request information of a predetermined degree or higher from a vector database; Steps 1-3 of sending a final query based on the first dispatch request information to the third dispatch request information to the language model and making a query; Steps 1-4 of receiving responses from the language model, Step 1-5 comprising generating a confirmation request message based on the response of Step 1-4 and transmitting the confirmation request message to the user terminal. How to determine whether dynamic dispatching is possible.

6. In claim 5, Steps 1-6 extract information related to the destination from the response received in steps 1-4 based on the results approved by the user in steps 1-5; Further comprising steps 1-7 of calling a dispatch API based on the destination information extracted in steps 1-6. How to determine whether dynamic dispatching is possible.

7. In claim 1, It further includes an 8th step of calculating costs for the first and second points and executing dynamic dispatching. Step 8 is, Step 8-1, which calculates the cost between each fixed waypoint and the first point and the cost between each fixed waypoint and the second point; Based on the output result of step 8-1, step 8-2 includes selecting point 3-1 and point 3-2, which are the two fixed waypoints with the lowest cost to point 1, and point 4-1 and point 4-2, which are the two fixed waypoints with the lowest cost to point 2, and step 8-3 includes determining dynamic dispatch for point 1 and point 2. Step 8-3 is, Step 8-3-1 of comparing the first value, which is the larger value among the 1-1 cost and the 1-2 cost, with the 1-3 cost, and if the first value is less than or equal to the 1-3 cost, distributing the first point between the 3-1 point and the 3-2 point; Step 8-3-2, which dispatches the first point before the third point if the first value is greater than or equal to the first-third cost and the first-first cost is less than or equal to the first-second cost; Step 8-3-3, where the first value is greater than or equal to the 1-3 cost and the 1-1 cost is greater than or equal to the 1-2 cost, dispatches the 1st point after the 3-2 point; Step 8-3-4 of comparing the second value, which is the larger value between the 2-1 cost and the 2-2 cost, with the 2-3 cost, and if the second value is less than or equal to the 2-3 cost, then distributing the second point between the 4-1 point and the 4-2 point; Step 8-3-5, where the second value is greater than or equal to the 2-3 cost and the 2-1 cost is less than or equal to the 2-2 cost, dispatches the 2nd point before the 4-1 point; If the second value is greater than or equal to the 2-3 cost and the 2-1 cost is greater than or equal to the 2-2 cost, step 8-3-6 includes dispatching the 2nd point after the 4-2 point. The 1-1 cost is the cost from point 3-1 to point 1. The 1-2 cost is the cost from point 3-2 to point 1. The 1-3rd cost is the cost from point 3-1 to point 3-2. The 2-1 cost is the cost from point 4-1 to point 2. The 2-2 cost is the cost from point 4-2 to point 2. The 2nd-3rd cost is the cost from point 4-1 to point 4-2. How to determine whether dynamic dispatching is possible.

8. In claim 7, Point 3-1 is a fixed stopover upstream of Point 3-2, and Point 4-1 is a fixed stopover upstream of Point 4-2. How to determine whether dynamic dispatching is possible.

9. In claim 7 or 8, In step 8-3-2 or step 8-3-5, if point 3-1 or point 4-1 is the first fixed stopover, step 8-3-7 is additionally included, which determines that dynamic dispatch is not possible. How to determine whether dynamic dispatching is possible.

10. In claim 7 or 8, In step 4-3 or step 4-6, if point 3-2 or point 4-2 is the final fixed stopover, step 8-3-8 is additionally included, which unconditionally determines dynamic dispatch. How to determine whether dynamic dispatching is possible.

11. In a method for assigning dynamic dispatch to a mobile device when it is determined that dynamic dispatch is possible by the method of claim 1, Step 9: determining the boarding point of the passenger based on the first point information; Step 10 of calculating the travel time from the departure point to the boarding point based on passenger status information and movement reference information; Including an 11th step of dynamically dispatching a mobile device that can arrive at the boarding point within a predetermined time range after the applicant's boarding time, taking into account the travel time calculated in the 10th step. Dynamic dispatching method.

12. In claim 11, The passenger status information includes at least one of the passenger's disability type and mobility assistance device information, Travel reference information includes either weather information or route information from the departure point to the boarding point. Dynamic dispatching method.

13. In claim 11 or 12, Step 10 is a step of calculating a predicted movement time by applying a first parameter based on passenger status information and a second parameter based on movement reference information to the basic movement speed or basic movement time. Dynamic dispatching method.

14. In a machine learning method for predicting the travel time of the dynamic dispatching method of claim 13, A 12th step of inputting at least one of the passenger state information and at least one of the movement reference information into the MLP learning engine and the Transformer learning engine, Including the 13th step of predicting the movement time by combining the primary output value of the MLP learning engine and the primary output value of the Transformer learning engine in the final MLP layer. A machine learning method for predicting travel times in dynamic dispatching methods.

15. In claim 11, Step 14: Creating a state machine including properties regarding the fixed waypoint information of the mobile device and whether dispatching is in progress, and a callback function; Step 15 of receiving a first dynamic dispatch request from a first user terminal; Step 16 of selecting the first mobile device for which dispatch has been decided for the first dynamic dispatch request; Step 17: updating the dispatch progress property of the first mobile device to “true”; Step 18: applying the origin and destination information of the first dynamic dispatch request to the first mobile device to calculate the cost according to dynamic dispatch and selecting the route with the minimum cost among them to generate the first response; The state machine receives the first response and reflects the changed path of the first mobile device contained in the first response in step 19 to the properties of the first mobile device; If a second dynamic dispatch request is received while a first dynamic dispatch request is being processed, steps 16 to 19 are executed in parallel for the second dynamic dispatch request, but if the first mobile device is selected in step 16 for processing the second dynamic dispatch request, Step 20, where the state machine calls a callback function after the first response is received by the state machine; Step 21: calculating the cost according to dynamic dispatch by applying the origin and destination information of the second dynamic dispatch request to the first mobile device, and updating the first response by selecting the route with the minimum cost; The state machine receives the first response and reflects the changed path of the first mobile device contained in the first response in step 22 to the properties of the first mobile device. Including step 23 of updating the dispatch progress property of the first mobile device to “false”. Dynamic dispatching method.

16. In a traffic prediction method for a mobile device allocated by the dynamic dispatching method of claim 11, A speed data collection step that collects speed data of vehicles running on each preset link and stores it in a database; An event learning step in which the speed data stored in the above database is converted into differentiated images to enable visual distinction between low and high speeds, and changes in the speed data of each link according to events occurring as traffic conditions progress are reinforced based on changes in the images and stored in the database; A regional event prediction step that predicts a traffic event expected to occur based on the current location of the vehicle and the real-time driving speed at that location based on reinforcement learning data stored in the above database, and creates a predicted image reflecting the predicted event and stores it in the database; and A prediction event application step for changing the current travel route to a route that bypasses an area where a congestion event is predicted based on the predicted event or changing the estimated time of arrival (ETA) to change and save the optimal route to reach the destination in real time; A traffic prediction method based on real-time road speed.

17. In claim 16, In the above speed data collection step, real-time speed data of each road is obtained by obtaining public data from the city traffic information center accessed by the communication module equipped in the traffic prediction server. A traffic prediction method based on real-time road speed.

18. In claim 16, The above event learning step is a speed data visualization step that visualizes the speed data of each road stored in the database into differentiated images that can distinguish speed states such as congestion, slow-moving, and smooth; A unit time image storage step for setting an environment (evn) to be applied to reinforcement learning based on images by storing speed data of each road visualized as a differentiated image at preset unit time intervals; and An image-based reinforcement learning step for learning a path that maximizes rewards in each environment while imposing preset rewards or penalties in response to road choices of a vehicle driving on a road where speed data is visualized by unit time; A traffic prediction method based on real-time road speed.

19. In claim 18, In the above speed data visualization step, The color of the road is differentiated and stored by expressing the congested area (speed less than 40 km / h) in the first color, the slow-moving area (speed 40 to 80 km / h) in the second color, and the smooth area (speed over 80 km / h) in the third color. A traffic prediction method based on real-time road speed.

20. In claim 19, In the above image-based reinforcement learning step, Based on the speed and location information of the vehicle transmitted from the vehicle module, when the vehicle proceeds to the next road or link, a penalty is imposed if the vehicle passes through the first color area set as a congested area, and a reward is imposed if the vehicle passes through the third color area set as a smooth area, while learning the movement path of the vehicle driving in the area. A traffic prediction method based on real-time road speed.

21. In claim 19, In the above image-based reinforcement learning step, When the image color of the speed data is set to be diverse, differential compensation is applied according to the image color of the area where the vehicle passes. A traffic prediction method based on real-time road speed.

22. In claim 19, In the above regional event prediction stage, Using the results derived through reinforcement learning, we predict traffic events that may occur while a vehicle is driving in each region or link based on real-time speed data on the road. A traffic prediction method based on real-time road speed.

23. In claim 22, In the above predicted event application step, When a new route request to reach a destination is received from the above vehicle module, the result of reinforcement learning is applied to search for an optimal route based on a predicted image reflecting traffic prediction for a certain period of time in the future when the vehicle will drive from the current location, and the optimal route is transmitted to the vehicle module. A traffic prediction method based on real-time road speed.

24. A method for assigning seats to a mobile device assigned by the dynamic dispatching method of claim 11, (a) a step of registering vehicle-specific seat information input from the administrator terminal; and (b) a step of assigning seats to each user based on the travel distance of the MOD bus for the departure point and destination included in the travel request, the boarding time of the user during the travel distance of the bus, and the seat information of the dispatched bus when a bus with a seat assignment is dispatched according to the user's boarding request; including; Seat assignment method for reservation-based transportation.

25. In claim 24, The step of assigning seats to each user is as follows: (b1) Calculate and use the boarding time and travel distance for at least one user scheduled to board the above-deployed bus, and the remaining travel distance and remaining boarding time for passengers who have already boarded. A step of calculating the self-evaluation index; and (b2) In descending order of the user-specific evaluation index calculated above, the entry and exit of the MOD bus including the step of assigning a seat close to the door; Seat assignment method for reservation-based transportation.

26. In claim 25, The above step (b2) is, For users with the same user-specific evaluation index calculated above, seats closer to the entrance are assigned in the order of shortest travel distance, shortest boarding time, and the order of boarding requests. Seat assignment method for reservation-based transportation.

27. In claim 24, Step (a) above, A step of receiving and storing seat information including the door location, the seat location from the door, and the seat type for each vehicle in operation from the above administrator terminal, and the seat type is one of a socially disadvantaged seat, a short-distance general seat, and a long-distance general seat. Seat assignment method for reservation-based transportation.

28. In claim 25 or 26, It further includes a step of receiving a boarding request including the departure point, destination, departure time, and social disadvantage from the user terminal, and when assigning a seat close to the entrance, determining the seat type based on the social disadvantage. Seat assignment method for reservation-based transportation.

29. In claim 25, Step (b) above, (bo) When the above user boarding request is received, a step of checking the remaining seats available for dispatch to determine whether seat assignment is possible; further comprising; Seat assignment method for reservation-based transportation.

30. In claim 29, The above (bo) step is to determine that seat assignment is possible if there are remaining seats for the socially disadvantaged, if the user's boarding request includes a request for a seat for the socially disadvantaged, by providing information on seats for the socially disadvantaged, and if there are no remaining seats for the socially disadvantaged, to check the remaining seats of other available candidate vehicles. Seat assignment method for reservation-based transportation.

31. In claim 27, Further comprising a step of storing statistical data by travel time based on the boarding and alighting points and boarding and alighting times of each existing bus user; The above step (b) predicts the probability of occurrence of other users scheduled to board, the departure point and destination of the other users scheduled to board, based on statistical data corresponding to the travel time from the departure point to the destination of the user requesting boarding, Based on the departure point and destination of the other expected boarding user, it is determined whether the other expected boarding user will maintain boarding even when the boarding request user gets off. Determine the seat of the user requesting boarding based on whether the other expected boarding users will remain on board. Seat assignment method for reservation-based transportation.

32. In claim 31, The above step (b) is to assign a seat to the user requesting boarding to be closer to the door of the dispatched bus than to the other user expected to board, when the other user expected to board is still on board when the user requesting boarding gets off. Seat assignment method for reservation-based transportation.

33. In claim 32, The above step (b) assigns an aisle seat based on the departure point and destination of the other expected boarding user when the occurrence probability of the other expected boarding user is greater than or equal to the standard probability. Seat assignment method for reservation-based transportation.

Images