Road traffic prediction system based on realtime road velocity data and thereof method

Abstract

The present invention relates to a traffic prediction system and method, and more specifically, to a real-time road speed-based traffic prediction system and method that simplifies a reinforcement learning model by simplifying link-by-link speed changes based on images using real-time road speed information, and enables a user operating a vehicle or an autonomous driving system to provide the traffic conditions of the area to be driven as a prediction image derived by reinforcement learning based on the current real-time speed.

Description

Technology Field

[0001] The present invention relates to a traffic prediction system and method, and more specifically, to a real-time road speed-based traffic prediction system and method that simplifies a reinforcement learning model by simplifying link-by-link speed changes based on images using real-time road speed information, and enables a user operating a vehicle or an autonomous driving system to provide the traffic conditions of the area to be driven as a prediction image derived by reinforcement learning based on the current real-time speed. Background Technology

[0002] With the recent expansion of AI applications, various attempts are underway to derive the optimal route from a starting point to a destination by reflecting road traffic conditions in relation to autonomous driving.

[0003] Accordingly, although there have been attempts to develop various algorithms to derive the optimal travel route from a starting point set by the driver to a destination, there has been a problem in that it is difficult to avoid a decrease in the accuracy of traffic prediction due to traffic congestion caused by unexpected events occurring on the actual road.

[0004] In addition, for various algorithms to derive the optimal movement path, a model considering multiple environmental factors had to be applied to respond to various environmental changes occurring during driving, which resulted in the problem of requiring a lot of resources for data processing and time for training through deep learning. The problem to be solved

[0005] The present invention aims to solve the aforementioned problems by providing a real-time road speed-based traffic prediction system and method that simplifies a reinforcement learning model by simplifying link-by-link speed changes based on images using real-time road speed information, and enables a user operating a vehicle or an autonomous driving system to provide the traffic conditions of the driving area as a prediction image derived by reinforcement learning based on the current real-time speed.

[0006] The above and other objects and advantages of the present invention will become apparent from the following description describing preferred embodiments. means of solving the problem

[0007] A real-time road speed-based traffic prediction system according to an embodiment of the present invention for achieving the above-mentioned purpose comprises: a traffic prediction server that visualizes and stores the average driving speed of vehicles traveling on a road by link in a pre-set zone unit, reinforces learning on traffic events occurring according to changes in the image, predicts traffic events to occur after the image of the current speed for the corresponding zone, and adjusts the vehicle's route or Estimated Time of Arrival (ETA) according to the predicted event and provides it to a vehicle module; and a vehicle module that transmits information on the starting point and destination to be traveled, as well as information on the vehicle's current location and speed, to the traffic prediction server. The traffic prediction server comprises: a speed data collection unit that collects speed data of vehicles traveling on each pre-set link and stores it in a database; and an event learning unit that converts the speed data stored in the database into differentiated images to enable visual distinction between low speed and high speed, and reinforces learning on changes in speed data of each link according to events occurring as traffic conditions progress, based on changes in the image, and stores it in the database. It may include: a regional event prediction unit that predicts traffic events expected to occur in correspondence with the vehicle's current location and real-time driving speed at that location based on 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 that changes the current travel path to a path that bypasses areas where congestion events are predicted, or changes the estimated time of arrival (ETA), and stores the optimal path to reach the destination in real time based on the predicted event.

[0008] At this time, the speed data collection unit can acquire real-time speed data of each road by acquiring public data from the city traffic information center accessed via a communication module provided in the traffic prediction server.

[0009] Additionally, the event learning unit may include: a speed data visualization unit that visualizes speed data of each road stored in the database as differentiated images capable of distinguishing speed states such as congestion, slow speed, and smooth speed; a unit-time image storage unit that stores speed data of each road visualized as differentiated images at preset unit times to set an environment (evn) to be applied to reinforcement learning based on images; and an image-based reinforcement learning unit that learns a path capable of maximizing rewards in each environment while imposing preset rewards or penalties in response to the road selection of a vehicle driving on the road where speed data is visualized at preset unit times.

[0010] In addition, the speed data visualization unit can store the road image colors by differentiating them, such that congested areas (less than 40 km / h) are represented by a first color, slow areas (40 to 80 km / h) are represented by a second color, and smooth areas (80 km / h or more) are represented by a third color.

[0011] In addition, the image-based reinforcement learning unit can learn the movement path of a vehicle traveling in a corresponding area by imposing a penalty when the vehicle passes through a first-colored area designated as a congested area and imposing a reward when the vehicle passes through a third-colored area designated as a smooth area, based on the vehicle's speed and location information transmitted from the vehicle module.

[0012] In addition, the image-based reinforcement learning unit described above can be configured to impose differential rewards based on the image color of the area through which the vehicle passes, when the image colors of the speed data are set to various values.

[0013] In addition, the image-based reinforcement learning unit can predict traffic events that may occur while a vehicle is driving in each region or link based on real-time road speed data using results derived through reinforcement learning.

[0014] In addition, when there is a new route request from the vehicle module to reach a destination, the prediction image application unit can apply the results of reinforcement learning to search for an optimal route based on a prediction image that reflects traffic predictions for a future period during which the vehicle will drive from its current location, and transmit it to the vehicle module.

[0015] Additionally, a real-time road speed-based traffic prediction method according to another embodiment of the present invention may include: a speed data collection step of collecting speed data of vehicles traveling on each pre-set link and storing it in a database; an event learning step of converting the speed data stored in the database into a differentiated image that enables visual distinction between low speed and high speed, and reinforcing learning the change in speed data of each link according to events occurring as traffic conditions progress based on the change in the image and storing it in the database; a regional event prediction step of predicting traffic events expected to occur corresponding 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 generating a prediction image reflecting the predicted event and storing it in the database; and a prediction event application step of changing the current travel path to a path that bypasses the area where a congestion event is predicted or changing the Estimated Time of Arrival (ETA) to change and store the optimal path to reach the destination in real time based on the predicted event.

[0016] At this time, in the speed data collection step, real-time speed data of each road can be obtained by acquiring public data from the city traffic information center accessed via a communication module equipped in the traffic prediction server.

[0017] Additionally, the event learning step may include: a speed data visualization step that visualizes speed data of each road stored in the database as differentiated images capable of distinguishing speed states such as congestion, slow speed, and smooth speed; a unit-time image saving step that saves the speed data of each road visualized as differentiated images at preset unit-time intervals to set an environment (evn) to be applied to reinforcement learning based on images; and an image-based reinforcement learning step that learns a path capable of maximizing the reward in each environment while imposing preset rewards or penalties in response to the road selection of a vehicle driving on the road where the speed data is visualized at preset unit-time intervals.

[0018] In addition, in the speed data visualization step, the road colors can be differentiated and stored by representing congested areas (less than 40 km / h) with a first color, slow areas (40 to 80 km / h) with a second color, and smooth areas (80 km / h or more) with a third color.

[0019] In addition, in the image-based reinforcement learning step, based on the vehicle speed and location information transmitted from the vehicle module, a penalty is imposed when the vehicle passes through a first-colored area designated as a congested area while proceeding to the next road or link, and a reward is imposed when the vehicle passes through a third-colored area designated as a smooth area, thereby learning the movement path of the vehicle driving in the area.

[0020] In addition, in the above image-based reinforcement learning step, if the image colors of the speed data are set to various values, differential rewards can be applied according to the image colors of the areas where the vehicle passes.

[0021] In addition, in the above-mentioned 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 results derived through reinforcement learning.

[0022] In addition, in the above prediction event application step, if there is a new route request from the vehicle module to reach a destination, the result of reinforcement learning can be applied to search for an optimal route based on a prediction image that reflects traffic predictions for a future period during which the vehicle will drive from its current location, and transmit it to the vehicle module. Effects of the invention

[0023] According to an embodiment of the present invention, by utilizing real-time speed information of the road, the speed change per link is simplified based on images to simplify the reinforcement learning model, and there is an effect of providing the traffic conditions of the area to be driven by a user operating a vehicle or an autonomous driving system as a predicted image derived by reinforcement learning based on the current real-time speed.

[0024] In addition, the present invention has the effect of increasing routing efficiency and improving the Estimated Time of Arrival (ETA) by applying reinforcement learning results based on road speed information to predict traffic events within a certain time period based on real-time speed.

[0025] In addition, the present invention has the effect of being able to adjust or respond to vehicle routes in accordance with actual road conditions by predicting traffic events in a region based on the real-time speed of roads in each region.

[0026] In addition, the present invention has the effect of being applicable to the advancement of cost-based algorithms, such as the Dijkstra algorithm, by introducing a dynamic link cost such that the cost of a link changes according to events predicted by reinforcement learning. Brief explanation of the drawing

[0027] FIG. 1 is a block diagram of a real-time road speed-based traffic prediction system according to the present invention. FIG. 2 is a configuration diagram of a real-time road speed-based traffic prediction method according to the present invention. FIG. 3 is a flowchart illustrating the process of predicting traffic according to the present invention. Specific details for implementing the invention

[0028] The present invention will be described in detail below with reference to the embodiments and drawings. These embodiments are presented merely as examples to explain the invention more specifically, and it will be obvious to those skilled in the art that the scope of the invention is not limited by these embodiments.

[0029] Furthermore, unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which the present invention pertains, and in the event of a conflict, the description in this specification, including the definitions, shall prevail.

[0030] In order to clearly explain the proposed invention in the drawings, parts unrelated to the explanation have been omitted, and similar parts throughout the specification have been given similar reference numerals. Furthermore, when a part is described as "comprising" a certain component, this means that, unless specifically stated otherwise, it does not exclude other components but may include additional components.

[0031] Hereinafter, embodiments of the present invention will be described in detail with reference to FIG. 1.

[0032] FIG. 1 is a block diagram of a real-time road speed-based traffic prediction system according to the present invention.

[0033] Referring to FIG. 1, a real-time road speed-based traffic prediction system according to one embodiment of the present invention may include a traffic prediction server (10) that images and stores the average driving speed of vehicles driving on a road by link in a preset zone unit, reinforces learning on traffic events that occur according to changes in the image, predicts traffic events that will occur after the image of the current speed for the zone, and adjusts the vehicle's route or Estimated Time of Arrival (ETA) according to the predicted event and provides it to a vehicle module, and a vehicle module (20) that transmits information on the starting point and destination to be driven, as well as information on the vehicle's current location and speed, to the traffic prediction server.

[0034] The above traffic prediction server (10) may include a speed data collection unit (100) that collects speed data of vehicles traveling on each pre-set link and stores it in a database; an event learning unit (200) that converts the speed data stored in the database into differentiated images that enable visual distinction between low speed and high speed, and stores the changes in speed data of each link according to events occurring as traffic conditions progress by performing reinforcement learning based on changes in images and storing them in a database; a regional event prediction unit (300) that predicts traffic events expected to occur corresponding to the vehicle's current location and real-time driving speed at that location based on the reinforcement learning data stored in the database, generates a prediction image reflecting the predicted event, and stores it in a database; and a prediction event application unit (400) that changes the current travel path to a path that bypasses the area where a traffic congestion event is predicted, or changes the estimated time of arrival (ETA), and stores the optimal path to reach the destination in real time based on the predicted event.

[0035] The above speed data collection unit (100) can acquire real-time speed data of each road by acquiring public data from the city traffic information center connected via a communication module provided in the traffic prediction server (10).

[0036] At this time, the speed data collection unit (100) may collect real-time speed data by directly receiving speeds from multiple vehicles operating on the actual road, but it is difficult to secure representativeness for the speed of individual vehicles, so there is the inconvenience of having to calculate the average of the speed data of the vehicles located at each link.

[0037] Accordingly, the speed data collection unit (100) preferably acquires real-time speed data for each road link based on public data provided by a public institution, such as a city traffic information center, so that the collected real-time speed data can be representative.

[0038] Additionally, the event learning unit (200) may include a speed data visualization unit (210) that visualizes speed data of each road stored in the database into differentiated images capable of distinguishing speed states such as congestion, slow speed, and smooth speed; a unit-time image storage unit (220) that stores speed data of each road visualized into differentiated images at preset unit times to set an environment (evn) to be applied to reinforcement training based on images; and an image-based reinforcement learning unit (230) that learns a path capable of maximizing rewards in each environment while imposing preset rewards or penalties in response to the action taken by a vehicle driving on the road where speed data is visualized at preset times, i.e., the selection of a road connected to the driving link.

[0039] At this time, the speed data visualization unit (210) can simplify the environment applied to reinforcement learning by using speed data to differentiate and store the colors of each road image. Accordingly, the speed data visualization unit (210) can differentiate and store the colors of the road by representing congested areas (e.g., less than 40 km / h) with a first color (e.g., red), slow areas (e.g., 40 to 80 km / h) with a second color (e.g., yellow), and smooth areas (e.g., 80 km / h or more) with a third color (e.g., green).

[0040] The above-described distinction of speed and selection of color are merely examples that can be adopted by the speed data visualization unit (210), and it goes without saying that the speed can be further subdivided or the colors can be set more diversely.

[0041] The above unit time-based image storage unit (220) can store the visualized speed data generated by the speed data visualization unit (210) as a new image in a database at preset unit times (e.g., 1 second, 1 minute, 1 hour, etc.).

[0042] Additionally, the unit time-based image storage unit (220) may also store each image to be stored in a database after labeling whether it represents a speed image of which road and time.

[0043] By labeling the images stored in this way, the administrator of the traffic prediction server can later use this information to identify and manage the environment applied to reinforcement learning and past data.

[0044] The image-based reinforcement learning unit (230) can perform reinforcement learning using image-based speed data stored in a database. Accordingly, the image-based reinforcement learning unit (230) can learn to derive a path that maximizes the reward by considering the action chosen by the agent vehicle during reinforcement learning, that is, the effect on the speed data of the road or link the vehicle enters, and by giving a pre-set reward or imposing a penalty as feedback.

[0045] For example, the image-based reinforcement learning unit (230) can learn the movement path of a vehicle traveling in a region by imposing a penalty when the vehicle passes through a region of a first color (e.g., red image) set as a congested area when the vehicle proceeds to the next road or link based on the vehicle speed and location information transmitted from the vehicle module (20), and imposing a reward when the vehicle passes through a region of a third color (e.g., green image) set as a smooth area.

[0046] At this time, the image-based reinforcement learning unit (230) may be configured such that differential rewards are applied according to the image color of the area where the vehicle passes, when the image color of the speed data is set to be varied.

[0047] For example, a penalty of -1 may be imposed when passing through the first color area, 0 when passing through the second color area, and +1 when passing through the third color area. Of course, the points imposed may be increased or decreased depending on the diversity of image colors.

[0048] In addition, as another example, reinforcement learning can be performed by imposing various rewards based on the image color changes of the regions the vehicle enters while driving on the road, such as imposing a +2 reward when entering a 3rd color region from a 1st color region, and imposing a -2 reward when entering a 1st color region from a 3rd color region.

[0049] In addition, the image-based reinforcement learning unit (230) can be used as data for reinforcement learning to derive an optimal travel path by imposing a certain reward when a vehicle newly entering the road or link achieves the maximum speed, i.e., the fastest estimated time of arrival (ETA), while traveling along the registered travel path between the starting point and the destination.

[0050] The above-mentioned 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 on, based on results derived through reinforcement learning.

[0051] For example, if real-time speed data is collected in a first color indicating a congested state, traffic conditions in the area can be predicted, such as by predicting that the congestion will be maintained or worsened due to an increase in vehicles entering the road for a certain period of time.

[0052] In addition, the prediction event application unit (400) can apply a predicted traffic situation event based on the learning result from the image-based reinforcement learning unit to the optimal route for reaching the destination set by the vehicle module, thereby changing and saving the optimal route or adjusting the estimated time of arrival (ETA) and providing it to the vehicle module.

[0053] In order to perform such reinforcement learning, the event learning unit (200) collects node-link data for a target area to be reinforced learning performed, as shown in the flowchart in FIG. 3, and models and stores the environment (evn), state, action, and reward to be reinforced learning performed based on the collected node-link data.

[0054] And, the event learning unit (200) sets an algorithm for finding the shortest path from the current position to the target position (e.g., Dijkstra's algorithm) as the target for reinforcement learning.

[0055] Then, based on images that visualize and store real-time speed data of vehicles traveling on the road of the target area obtained from the speed data collection unit (100) at unit times, the image-based reinforcement learning unit (230) performs reinforcement training.

[0056] The image-based reinforcement learning unit (230) derives a TQS (Trained Quality Set) that reflects the results of reinforcement learning, the regional event prediction unit (300) applies the derived TQS to predict events that will occur continuously thereafter based on speed information in the region, and the predicted event application unit (400) searches for an optimal path to reach a destination based on the current speed of the region and the predicted events.

[0057] Subsequently, when there is a new route request from the vehicle module to reach a destination (target location), the prediction event application unit (300) applies TQS reflecting the result of reinforcement learning to search 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) and then transmits it to the vehicle module.

[0058] In this way, the traffic prediction server does not set the estimated time of arrival (ETA) based on the current speed measured at each road or link on the travel path, but generates an image based on real-time speed for the road corresponding to the vehicle's current location; however, for the roads or links on the travel path to be driven thereafter, it provides a travel path based on a predicted image reflecting the learning results of the image-based reinforcement learning unit, thereby increasing the efficiency of the travel path search and improving the accuracy of the estimated time of arrival (ETA).

[0059] In addition, if the speed of the road reached while driving differs from the speed in the aforementioned prediction image, a new prediction image corresponding to the changed speed is generated and provided. This enables the vehicle to quickly bypass the congested section based on the prediction image and search for the optimal travel route, even in cases where there are sections with prolonged delays due to unexpected traffic events.

[0060] In this way, if the speed data of the road the vehicle is traveling on differs from the traffic forecast while driving to the destination along the transmitted optimal route, it goes without saying that route modifications are possible to temporarily change the destination in order to achieve maximum speed.

[0061] As such, unlike conventional reinforcement learning that eliminates uncertainty based on complex algorithms that include diverse and detailed environments (evn) in the modeling for machine learning, the present invention can reduce the computational burden of reinforcement learning by simplifying it to a simplified environment (evn) of real-time speed data images.

[0062] In addition, the present invention stores the entire process of traffic conditions occurring, propagating, spreading, and then being resolved based on real-time speed changes in a specific area as image changes in speed data in a database.

[0063] Accordingly, when deriving the optimal route for a vehicle to reach the same or adjacent destination, it is possible to predict the next conditions, such as congestion, slow speeds, or smooth flow, that will occur on each link based on the vehicle's current speed data, thereby enabling the derivation of a route that avoids or bypasses these conditions.

[0064] Furthermore, the present invention also enables learning of subsequent diffusion patterns based on traffic condition patterns in a specific region, for example, the type in which a first color indicating the onset of congestion spreads along a specific link over time. Accordingly, the present invention can be combined with a system that performs reinforcement learning based on prior algorithms utilizing historical data, as well as reinforcement learning based on images of real-time speed data, for the prediction of optimal routes or traffic events.

[0066] Next, with reference to FIGS. 2 and FIGS. 3, a real-time road speed-based traffic prediction method according to another embodiment of the present invention will be described.

[0067] FIG. 2 is a configuration diagram of a real-time road speed-based traffic prediction method according to the present invention, and FIG. 3 is a flowchart showing the process of predicting traffic according to the present invention.

[0068] Referring to FIGS. 2 and 3, a real-time road speed-based traffic prediction method according to another embodiment of the present invention may include: a speed data collection step (S10) for collecting speed data of vehicles traveling on each pre-set link and storing it in a database; an event learning step (S20) for converting the speed data stored in the database into a differentiated image that enables visual distinction between low speed and high speed, and for performing reinforcement learning on the changes in speed data of each link according to events occurring as traffic conditions progress, based on changes in the image, and storing it in a database; a regional event prediction step (S30) for predicting traffic events expected to occur corresponding 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 for generating a prediction image reflecting the predicted event and storing it in a database; and a prediction event application step (S40) for changing the current travel path to a path that bypasses the area where a congestion event is predicted or changing the estimated time of arrival (ETA) to change and store the optimal path to reach the destination in real time based on the predicted event.

[0069] The above speed data collection step (S10) can obtain real-time speed data of each road by acquiring public data from the city traffic information center accessed by a communication module equipped in the traffic prediction server.

[0070] Additionally, the event learning step (S20) may include a speed data visualization step (S21) that visualizes speed data of each road stored in the database into differentiated images capable of distinguishing speed states such as congestion, slow speed, and smooth speed; a unit-time image saving step (S22) that saves the speed data of each road visualized into differentiated images at preset unit-time intervals to set an environment (evn) to be applied to reinforcement training based on images; and an image-based reinforcement learning step (S23) that learns a path capable of maximizing rewards in each environment while imposing preset rewards or penalties in response to the action taken by a vehicle driving on the road where the speed data is visualized at preset unit-time intervals, i.e., the selection of a road connected to the driving link.

[0071] At this time, in the speed data visualization step (S21), the environment applied to reinforcement learning can be simplified by using speed data to differentiate and store the colors of each road image. Accordingly, in the speed data visualization step (S21), congested areas (e.g., less than 40 km / h) can be represented by a first color (e.g., red), slow areas (e.g., 40 to 80 km / h) can be represented by a second color (e.g., yellow), and smooth areas (e.g., 80 km / h or more) can be stored by differentiating the colors of the roads by a third color (e.g., green).

[0072] In the above unit time-based image storage step (S22), the visualized speed data generated in the speed data visualization step can be stored in a database as a new image at preset unit times (e.g., 1 second, 1 minute, 1 hour, etc.).

[0073] In addition, in the above unit time-based image storage step (S22), each image to be stored may be labeled to indicate which road and at which time the speed image represents, and then stored together in the database.

[0074] By labeling the images stored in this way, the administrator of the traffic prediction server can later use this information to identify and manage the environment applied to reinforcement learning and past data.

[0075] In the above image-based reinforcement learning step (S23), reinforcement learning can be performed using image-based speed data stored in a database. Accordingly, in the above image-based reinforcement learning step (S23), while reinforcement learning is in progress, the vehicle acting as an agent can learn to derive a path that maximizes the reward by considering the action selected by the agent—that is, the impact on the speed data of the road or link the vehicle enters—and by providing a pre-set reward or imposing a penalty as feedback.

[0076] For example, in the image-based reinforcement learning step (S23), based on the vehicle speed and location information transmitted from the vehicle module, a penalty is imposed when the vehicle passes through a first color area (e.g., red image) set as a congested area when proceeding to the next road or link, and a reward is imposed when the vehicle passes through a third color area (e.g., green image) set as a smooth area, thereby learning the movement path of the vehicle driving in the area.

[0077] At this time, in the image-based reinforcement learning step (S23), if the image colors of the speed data are set in various ways, differential rewards may be imposed according to the image colors of the areas where the vehicle passes.

[0078] For example, a penalty of -1 may be imposed when passing through the first color area, 0 when passing through the second color area, and +1 when passing through the third color area. Of course, the points imposed may be increased or decreased depending on the diversity of image colors.

[0079] In addition, as another example, reinforcement learning can be performed by imposing various rewards based on the image color changes of the regions the vehicle enters while driving on the road, such as imposing a +2 reward when entering a 3rd color region from a 1st color region, and imposing a -2 reward when entering a 1st color region from a 3rd color region.

[0080] In addition, in the image-based reinforcement learning step (S23), when a vehicle newly entering the road or link achieves the maximum speed, i.e., the fastest estimated time of arrival (ETA), while driving along the registered route between the starting point and the destination, a certain reward is imposed, and this can be used as data for reinforcement learning to derive the optimal route.

[0081] In the above-mentioned regional event prediction step (S30), based on the results derived through reinforcement learning, events that may occur while a vehicle is driving in each region or link can be predicted based on real-time speed data of the road currently being driven.

[0082] For example, if real-time speed data is collected in a first color indicating a congested state, traffic conditions in the area can be predicted, such as by predicting that the congestion will be maintained or worsened due to an increase in vehicles entering the road for a certain period of time.

[0083] In addition, in the prediction event application step (S40), the predicted traffic situation event based on the learning result from the image-based reinforcement learning step is applied to the optimal route for reaching the destination set by the vehicle module to change and save the optimal route or adjust the estimated time of arrival (ETA) and provide it to the vehicle module.

[0084] In this way, for the image-based reinforcement learning described above, as shown in the flowchart in FIG. 3, node-link data for a target area to be reinforced is collected (S110), and based on the collected node-link data, an environment (evn), state, action, and reward to be reinforced is modeled (S120) and stored.

[0085] Then, an algorithm for finding the shortest path from the current position to the target position (e.g., Dijkstra's algorithm) is set as the target for reinforcement learning (S130).

[0086] Then, real-time speed data of vehicles traveling on the road of the target area is collected (S210), the collected speed data is visualized at unit times and an image is saved (S220), and reinforcement training is performed based on the image of the visualized speed data (S230).

[0087] A Trained Quality Set (TQS) reflecting the results of this reinforcement learning is generated (S240), and the optimal path to reach the destination is searched by applying the derived TQS (S250).

[0088] Subsequently, when there is a new route request (S310) from the vehicle module to reach a destination (target location) and the vehicle departs (S320) toward the destination, the TQS reflecting the result of the reinforcement learning is applied based on the real-time road speed at the vehicle's current location, and the optimal route is searched based on a prediction image reflecting traffic prediction for a future period of time (e.g., 10 minutes after the current time) and then transmitted to the vehicle module (S330).

[0089] In addition, if the speed of the road reached while the vehicle is driving differs 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 presented (S340), thereby enabling the search for an optimal route that quickly bypasses the delayed section based on the predicted image, even if there is a section where delays are prolonged due to unexpected traffic events.

[0090] In this specification, only a few examples among the various embodiments carried out by the inventors are described; however, the technical concept of the present invention is not limited or restricted thereto, and it is obvious that it can be modified and implemented in various ways by those skilled in the art. Explanation of the symbols

[0091] 10: Traffic prediction server 20: Vehicle module 100: Speed ​​data collection unit 200: Event Learning Section 210: Speed ​​Data Visualization Section 220: Image storage unit by time unit 230: Image-based reinforcement learning unit 300: Regional Event Forecast Section 400: Prediction Event Application Section

Claims

Claim 1 A real-time road speed-based traffic prediction system comprising: a traffic prediction server; and a vehicle module that transmits information on the starting point and destination to be traveled, as well as information on the vehicle's current location and speed, to the traffic prediction server; wherein the traffic prediction server comprises: a speed data visualization unit that classifies speed data for each link stored in a database into multiple speed states including congestion, slow speed, and smooth speed, and generates a road image represented by a color corresponding to each speed state; an image-based reinforcement learning unit that performs reinforcement learning while imposing a reward or penalty on the vehicle's road selection based on a road image labeled with information on road link information and the time of creation, and generates a prediction image; a regional event prediction unit that predicts traffic events based on a road image corresponding to the vehicle's current location using the results derived through the reinforcement learning; and a prediction event application unit that changes the current travel route or adjusts the estimated time of arrival according to the predicted traffic event. Claim 2 A real-time road speed-based traffic prediction system according to claim 1, wherein the traffic prediction server further comprises a speed data collection unit that acquires real-time speed data of each road by acquiring public data of an urban traffic information center accessed by a communication module provided in the traffic prediction server. Claim 3 In claim 1, the traffic prediction server further comprises a unit-time image storage unit that stores speed data of each road visualized as a differentiated image at preset unit-time intervals and sets an environment (evn) to be applied to reinforcement learning based on the image; and the image-based reinforcement learning unit learns a path that can maximize the reward in each environment while imposing preset rewards or penalties in response to the road selection of a vehicle driving on a road where speed data is visualized at preset unit-time intervals. Claim 4 A real-time road speed-based traffic prediction system according to claim 1, wherein the speed data visualization unit stores the colors of a road image by differentiating them as follows: congestion (less than 40 km / h) is represented by a first color, slow traffic (40 to 80 km / h) is represented by a second color, and smooth traffic (80 km / h or more) is represented by a third color. Claim 5 In claim 4, the image-based reinforcement learning unit learns the movement path of a vehicle traveling in a said area by imposing a penalty when the vehicle passes through a first-colored area set as congested and imposing a reward when the vehicle passes through a third-colored area set as smooth, based on the vehicle speed and location information transmitted from the vehicle module. Claim 6 In claim 5, the image-based reinforcement learning unit is configured to impose differential compensation according to the image color of the area through which the vehicle passes when the image color of the speed data is set to be varied, in a real-time road speed-based traffic prediction system. Claim 7 In paragraph 5, the above-mentioned regional event prediction unit is a real-time road speed-based traffic prediction system that predicts traffic events that may occur while a vehicle is driving on each link based on results derived through learning and real-time speed data of the road. Claim 8 In claim 7, the real-time road speed-based traffic prediction system further includes a prediction image application unit, and when there is a new route request from the vehicle module to reach a destination, the prediction event application unit searches for an optimal route based on a prediction image that reflects a traffic prediction for a future period of time during which the vehicle will drive from its current location by applying the results of learning, and transmits it to the vehicle module.

Images