Lane-wise multi-traffic flow prediction method and device, and recording medium therefor

Abstract

The present invention relates to a technical field of traffic information processing for predicting lane-wise or multi-traffic flow of a road by utilizing an artificial intelligence model such as a deep neural network and a simulation model such as a stochastic agent-based model (SABM), and visually displaying same on a navigation system.

Description

Lane-specific multi-traffic flow prediction method and device, and recording medium thereof

[0001] The present invention relates to the field of traffic information processing technology that utilizes artificial intelligence models such as deep neural networks and simulation models such as probabilistic agent-based models (SABM) to predict lane-specific or multiple traffic flows on a road and visually display them in a navigation system.

[0002] Generally, in-vehicle navigation systems and traffic information systems use a method of representing the traffic conditions of a specific road section as a single average speed value. These systems analyze traffic flow based on data collected from detectors or CCTVs installed on each road section, or GPS data mounted on vehicles.

[0003] A representative prior art is the "Traffic Flow Prediction Method and Device (KR Patent Registration: 1009849620000)." This technology uses traffic statistical information to search for congested sections over time, extracts habitually congested sections where congestion occurs repeatedly, and predicts traffic flow by identifying the speed change patterns of roads affected by the sections. However, this method has the limitation of predicting only a single flow speed for a single road.

[0004] Another prior art, "Method for Predicting Traffic Flow by Lane Using Touring CCTV Footage (KR Patent Registration: 1025866430000)," presents a method for predicting traffic flow information in a CCTV operating environment set to a fixed PTZ. However, this technology is limited to roads captured by CCTV and has the limitation of requiring high-capacity video data.

[0005] Recently, an approach utilizing deep learning has also been proposed, such as the "LSTM-based traffic congestion prediction method and device (KR Patent Registration: 1025465400000)." This method extracts spatiotemporal features from traffic flow data and predicts traffic congestion based on LSTM. However, this also has the limitation of predicting by considering the entire road as a single flow.

[0006] These conventional technologies contain several problems. First, traffic flow displays based on average speed near high-traffic on / off interchanges show significant discrepancies from actual conditions. Furthermore, traffic flow is frequently misinterpreted as smooth because average speeds are displayed even when congestion occurs on entrance / exit ramps or major arterial roads. In particular, excessive waiting times near connections between expressways and general roads are not properly reflected, and users are provided with inaccurate information due to averaged and ignored congestion. Consequently, this undermines driver confidence and causes confusion in route recalculation decisions.

[0007] Therefore, there is a need for the development of new technologies that can resolve these problems and provide more accurate traffic information.

[0008] The present invention was conceived against this technical background and aims to provide navigation users with more accurate traffic information and enable efficient route selection by utilizing artificial intelligence models and simulation models to reflect the complexity of actual traffic flow occurring at road entrances and exits, intersections, ICs / JCs, etc., and by individually predicting and visualizing lane-specific or multiple traffic flows within the same road.

[0009] A lane-by-lane multiple traffic flow prediction method of one embodiment includes, as illustrated in FIG. 3, a first step of collecting single traffic flow data and road type information of adjacent roads; a second step of refining the collected data and classifying the data by road type; a third step of constructing a network by modeling roads and nodes as a graph, setting possible paths and constraints at each edge, and generating a network model for each road type; a fourth step of determining edge-by-edge weights in each network model for each road type generated through the third step using an artificial intelligence model such as a deep neural network; a fifth step of simulating probabilistic interactions between agents and predicting lane-by-lane traffic flow using a simulation model such as a probabilistic agent-based model (SABM); and a sixth step of providing the traffic flow predicted through the fifth step to a user by displaying it in different colors for each lane.

[0010] The above 5th step modifies the prediction result of the above 4th step through a Bayesian neural network if actual data for each lane is available.

[0011] In the above second step, road types include expressways, city roads, and local roads; near the connection between an expressway and a general road, the traffic volume of vehicles exiting the access road is measured in real time, and the geometric information and signal operation information of the access and exit ramps are reflected. Traffic volume is calculated by applying a classified traffic volume model that considers the traffic handling capacity and travel speed characteristics of each road type.

[0012] In the above fourth step, the artificial intelligence model collects constraints such as the maximum hourly vehicle processing capacity of the road, maximum traffic volume and minimum operating speed per lane, and information on intersection signal operation and turn restrictions, and calculates a network model composed of edges capable of actual traffic by considering the collected constraints.

[0013] In the above fifth step, the simulation model calculates travel time information based on predicted traffic volume and speed for each lane, calculates the estimated waiting time for each lane at entrances / exits and junctions, and incorporates the above information as weights into a real-time path search algorithm to improve the accuracy of optimal path search.

[0014] In the first step above, if external factors including traffic accidents, road construction, or weather information occur, the type and severity of the event are quantified, the range of impact is spatially defined, and the duration of the event is estimated as a temporal range to calculate the rate of reduction in traffic flow capacity relative to the normal state and reflect it in the network model.

[0015] The first step above collects one or more of traffic-related public data, including vehicle traffic volume data collected from traffic volume detectors for each road section, video data collected from CCTVs for each road section, GPS data of probe vehicles collected from vehicle navigation terminals, Hi-Pass data, and signal control data.

[0016] The second step described above interpolates missing values ​​of the collected data, removes outliers, synchronizes time units, and standardizes the data format to classify the data by road type.

[0017] The above third step defines roads as edges and intersections and junctions as nodes, sets the number of lanes, speed limit, and maximum processing capacity per hour of each edge as attribute values, sets turning restrictions, signal display, and connecting road geometry at the nodes as constraints, and calculates the final travel cost of the corresponding edge by reflecting the delay degree according to the characteristics of the intersection section of the entrance / exit connecting road for highways, the waiting time according to the delay characteristics of signal intersections for city roads, and driving constraints according to the geometry and road alignment characteristics for local roads as weights in the calculation of the travel cost of each edge.

[0018] In another embodiment, a computing device and a recording medium are provided to implement the lane-by-lane multiple traffic flow prediction method described above.

[0019] The present invention has the effect of accurately predicting lane-specific or multiple traffic flows on a road and providing them to the user by determining edge-specific weights of a network model by road type through a deep neural network, simulating interactions between agents through a probabilistic agent-based model (SABM), and enabling model modification based on actual data through a Bayesian neural network.

[0020] FIG. 1 illustrates a network environment in which the present invention is implemented.

[0021] FIG. 2 illustrates a navigation screen provided to a user according to the present invention.

[0022] Figure 3 is a flowchart of a lane-by-lane multiple traffic flow prediction method of one embodiment.

[0023] Figure 4 is a schematic diagram explaining how to configure a network.

[0024] FIG. 5 is a block diagram of a computing device implementing a lane-by-lane multiple traffic flow prediction method of one embodiment.

[0025] Embodiments of the present invention will be described in detail below with reference to the drawings. However, detailed descriptions of known functions or configurations that may obscure the essence of the present invention in the following description and the attached drawings are omitted. Additionally, throughout the specification, the term 'comprising' a component means that, unless specifically stated otherwise, it does not exclude other components but may include additional components.

[0026] Additionally, terms such as first, second, etc. may be used to describe various components, but said components should not be limited by said terms. said terms may be used for the purpose of distinguishing one component from another component. For example, without departing from the scope of the present invention, the first component may be named the second component, and similarly, the second component may be named the first component.

[0027] The terms used in this invention are used merely to describe specific embodiments and are not intended to limit the invention. The singular expression includes the plural expression unless the context clearly indicates otherwise. In this application, terms such as "comprising" or "comprising" are intended to specify the presence of the described features, numbers, steps, actions, components, parts, or combinations thereof, and should be understood as not precluding the existence or addition of one or more other features, numbers, steps, actions, components, parts, or combinations thereof.

[0028] Unless specifically defined otherwise, all terms used herein, including technical or scientific terms, have the same meaning as generally understood by those skilled in the art to which the present invention pertains. Terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with their meaning in the context of the relevant technology, and should not be interpreted in an ideal or overly formal sense unless explicitly defined in this application.

[0029] Conventional traffic flow prediction technology had limitations in accurately reflecting actual traffic conditions by displaying them based on average road speeds. In particular, it failed to properly identify congestion occurring at on / off ramps or major arterial roads, providing drivers with inaccurate information.

[0030] To solve these problems, the present invention proposes a technology for predicting road traffic flow by subdividing it into lane-wise or multiple flows.

[0031] This invention predicts traffic flow by lane or as multiple flows by utilizing data classified by road type. Here, "multiple flows" refers to several distinct traffic flows occurring on a single road. For example, in cases where different flows occur within lanes of the same direction, the right lane represents the flow of vehicles waiting to turn right, the left lane represents the flow of vehicles waiting to turn left, and the middle lane represents the flow of vehicles going straight; in the case of traffic flows heading toward different destinations, it refers to the flow of vehicles going to Area A, Area B, and Area C. In other words, it means individually identifying and predicting each independent traffic flow occurring by lane or destination, rather than simply the average traffic flow of the entire road.

[0032] The present invention first collects single traffic flow data and road type information of adjacent roads, then refines the collected data and classifies it by road type. Next, it constructs a network by modeling roads and nodes as a graph, sets possible paths and constraints at each edge, and generates a network model after labeling by road type.

[0033] Furthermore, edge weights are determined in the road-type network model generated through artificial intelligence models such as deep neural networks; in this process, constraints such as the road's maximum hourly vehicle handling capacity, maximum traffic volume and minimum operating speed per lane, and information on intersection signal operation and turn restrictions are considered.

[0034] Next, traffic flow by lane is predicted by simulating probabilistic interactions between agents using a Probabilistic Agent-Based Model (SABM). If actual data for each lane is available, the model prediction results can be modified using a Bayesian neural network.

[0035] This invention also considers various external factors. When external factors such as traffic accidents, road construction, or weather information occur, the type and severity of the event are quantified, the scope of influence is spatially defined, and the duration of the event is estimated as a temporal range to calculate the rate of reduction in traffic flow capacity relative to the normal state and reflect it in the network model.

[0036] The final predicted traffic flow is displayed to the user in different colors for each lane. Through this visualization, users can intuitively grasp the traffic conditions of each lane, which can assist in selecting an efficient route.

[0037] As such, the present invention goes beyond simple average speed prediction to approach complex traffic environments in a multi-layered and probabilistic manner. Furthermore, by providing more accurate and reliable traffic information through the integration of advanced artificial intelligence technology and various data sources, the present invention can enhance the efficiency of urban traffic management and significantly improve the mobility convenience of individual drivers.

[0038]

[0039] The present invention can be implemented under a network environment as exemplified in FIG. 1.

[0040] Figure 1 is an overall system configuration diagram illustrating the network environment of the traffic flow prediction system of the present invention.

[0041] The system's physical infrastructure is broadly divided into a data collection layer, a network layer, a data processing layer, and an artificial intelligence server layer.

[0042] The data collection layer consists of various physical sensors and devices. Fixed traffic volume detectors (110) installed in road sections measure the traffic volume, speed, and occupancy rate of vehicles in real time. These detectors are installed at key points along the road to collect continuous traffic data.

[0043] The CCTV network (120) captures the traffic conditions of the road in video. These CCTVs consist of fixed and rotating (PTZ) cameras and are strategically placed on highways, arterial roads, major intersections, etc. In particular, they are concentrated at points where traffic flow changes significantly, such as entrances and exits, and bottleneck sections.

[0044] The probe data collection device (130) collects GPS location information and movement path data through vehicle navigation and smartphone applications. These devices track the movement patterns of a large number of vehicles in real time.

[0045] The terminal and collection server of the Hi-Pass system (140) collect data of vehicles using the highway. In the present invention, data collection through the Hi-Pass system can significantly improve the accuracy of traffic flow prediction. Since the Hi-Pass system provides origin-destination (O / D) data through the entry and exit records of each vehicle, it enables accurate prediction of traffic demand moving along a specific route. In addition, real-time passing data from Hi-Pass gates allows for tracking accurate traffic volume changes at IC / JCs, making it very useful for understanding real-time traffic conditions. Furthermore, Hi-Pass data enables the analysis of traffic patterns by day of the week and time of day, and can be utilized to learn regularly occurring traffic patterns. In particular, vehicle type information collected through the Hi-Pass toll collection system allows for the reflection of traffic flow characteristics by vehicle type. Such Hi-Pass data can be utilized as part of the "traffic-related public data" specified in claim 7 of the present invention, thereby enabling more accurate and reliable traffic flow prediction.

[0046] At the network layer, information is transmitted from each data collection device to a central data processing center via broadband networks (LTE, 5G). Network switches and routers ensure the stable and fast transmission of large volumes of data.

[0047] The central server (200) of the data processing layer can be configured as a cloud-based distributed computing system. Large-scale storage and high-performance servers refine and store collected data in real time. These servers have a hardware architecture optimized for big data processing.

[0048] The artificial intelligence server layer (300) consists of a GPU cluster dedicated to deep learning. High-performance GPU servers capable of large-scale parallel processing perform training and inference of complex artificial intelligence models. These servers process traffic data coming in in real time and continuously update the prediction model.

[0049] The end user interface server (400) provides predicted traffic flow information to navigation, mobile applications, etc., as shown in FIG. 2. This server processes the prediction results into a user-friendly form and transmits them. In FIG. 2, traffic flow is displayed using different types of lines, but in reality, it is displayed using different colors.

[0050] This complex physical infrastructure constitutes a comprehensive network environment capable of precisely capturing and predicting traffic conditions across the entire city in real time.

[0051]

[0052] Hereinafter, a method for predicting multiple traffic flows by lane according to an embodiment of the present invention, based on the network described above, will be explained.

[0053] According to this, a method for predicting multiple traffic flows by lane according to one embodiment includes, as illustrated in FIG. 3, a first step (S10) of collecting single traffic flow data and road type information of adjacent roads; a second step (S20) of refining the collected data and classifying the data by road type; a third step (S30) of constructing a network by modeling roads and nodes as a graph, setting possible paths and constraints at each edge, and generating a network model for each road type; a fourth step (S40) of determining edge weights in each network model for each road type generated through the third step using an artificial intelligence model such as a deep neural network; a fifth step (S50) of simulating probabilistic interactions between agents and predicting traffic flows by lane using a simulation model such as a probabilistic agent-based model (SABM); and a sixth step (S60) of providing the traffic flow predicted through the fifth step to a user by displaying it in different colors for each lane.

[0054] The above 5th step (S50) modifies the prediction result of the above 4th step through a Bayesian neural network when actual data for each lane is available.

[0055] In the above second step (S20), the road types include expressways, city roads, and local roads, and near the connection point between the expressway and the general road, the length of the vehicle queue on the access road and the volume of traffic exiting are measured in real time, and the geometric information and signal operation information of the access and exit connections are reflected, and a separate queue model is applied that considers the traffic processing capacity and travel speed characteristics of each road type to calculate the waiting time.

[0056] In the above fourth step (S40), the artificial intelligence model collects constraints such as the maximum hourly vehicle processing capacity of the road, maximum traffic volume and minimum operating speed per lane, and information on intersection signal operation and turn limit, and calculates the edge-specific weights by defining a penalty function based on the degree of violation of the collected constraints.

[0057] In the above 5th step (S50), the simulation model calculates travel time information based on predicted traffic volume and speed for each lane, calculates the estimated waiting time for each lane at entrances / exits and junctions, and reflects the information as weights in a real-time path search algorithm to improve the accuracy of optimal path search.

[0058] In the first step (S10) above, if external factors including traffic accidents, road construction, or weather information occur, the type and severity of the event are quantified, the range of influence is spatially set, and the duration of the event is calculated as a temporal range to calculate the reduction rate of traffic flow capacity compared to the normal state and reflect it in the network model.

[0059] The first step (S10) above collects one or more of traffic-related public data, including vehicle traffic volume data collected from traffic volume detectors for each road section, video data collected from CCTVs for each road section, GPS data of probe vehicles collected from vehicle navigation terminals, Hi-Pass data, and signal control data.

[0060] The second step (S20) above interpolates missing values ​​of the collected data, removes outliers, synchronizes time units, and standardizes the data format to classify the data by road type.

[0061] The above third step (S30) defines roads as edges and intersections and junctions as nodes, sets the number of lanes, speed limit, and maximum processing capacity per hour of each edge as attribute values, sets turning restrictions, signal display, and connecting road geometry at the nodes as constraints, and calculates the final travel cost of the corresponding edge by reflecting the delay degree according to the characteristics of the intersection section of the entrance / exit connecting road for highways, the waiting time according to the delay characteristics of signal intersections for city roads, and driving constraints according to the geometry and road alignment characteristics for local roads as weights.

[0062]

[0063] The following is a detailed explanation of each step.

[0064] Step 1 (S10)

[0065] The first stage is the process of collecting single traffic flow data and road type information for adjacent roads, which is implemented in detail as follows.

[0066] First, physical infrastructure for data collection is established. Various sensors and collection devices are strategically deployed along road sections. Traffic volume detectors are installed at key points along the roads to measure vehicle traffic volume, speed, and occupancy rates in real time. These detectors utilize various technologies, including loop detectors, radar sensors, and image detectors.

[0067] CCTV networks continuously monitor road traffic conditions. Fixed and rotating (PTZ) cameras are installed on expressways, arterial roads, and major intersections to collect video data. They are particularly concentrated at locations with significant changes in traffic flow, such as on / off ramps and bottleneck sections.

[0068] Probe data collection devices via vehicle navigation and smartphone applications collect GPS location information and movement path data in real time. This enables the tracking of movement patterns for a large number of vehicles.

[0069] The terminals and collection servers of the Hi-Pass system collect boarding and alighting information of public transportation users to provide additional traffic flow information.

[0070] The data collection process proceeds in the following detailed steps:

[0071] The first step is the discovery and connection of various data sources. Traffic detectors (110) are like the arteries of the road and detect minute traffic flow at each point. These detectors go beyond simply counting the number of vehicles to precisely measure the speed, movement patterns, and occupancy time of each vehicle.

[0072] The CCTV network (120) acts as the eyes of the road. Fixed cameras continuously monitor traffic conditions at specific points, while rotating cameras continuously monitor traffic flow over a wide area. They capture vehicle movements, lane changes, traffic congestion, etc., in real time.

[0073] Vehicle navigation and smartphone applications (130) provide real-time location and speed information of moving vehicles. Just like tracking the movement of numerous particles, these devices observe the traffic flow of the entire city macroscopically.

[0074] The public transportation high-pass system (140) is another important data source for urban transportation. Passenger boarding and alighting information serves as an important indicator showing the movement patterns and congestion levels of public transportation.

[0075] The collected data is immediately classified and contextualized. The data is subdivided according to the characteristics of highways, city roads, and local roads, and linked to the physical characteristics of the roads—number of lanes, speed limits, road width, etc.

[0076] The collection of external factors becomes more sophisticated. Traffic accidents, road construction, and weather conditions are detected and analyzed in real time. The type, severity, and scope of impact of each factor are measured in detail and integrated into the data.

[0077] The preprocessing process is also important for ensuring data reliability. It guarantees the accuracy and consistency of collected data through time synchronization, removal of duplicate data, and initial error filtering.

[0078] All of this process is transmitted in real time to a central server (200) via a broadband communication network. The data flows ceaselessly like the bloodstream of a city, capturing the vivid present of the road.

[0079]

[0080] Phase 2 (S20)

[0081] Data cleaning is the process of resolving the incompleteness of raw data. First, missing values ​​in the collected data are processed. Data that is temporarily missing from traffic detectors (110) or CCTVs (120) can be interpolated based on data from adjacent time periods. This is filled with the most probable values ​​using statistical interpolation methods or machine learning algorithms.

[0082] The outlier removal process enhances data reliability. It identifies and eliminates extreme values ​​that may occur due to sensor malfunctions or unusual situations. For example, abnormally high or low speed data, unrealistic vehicle densities, and the like are filtered out using statistical methodologies.

[0083] Time unit synchronization resolves time discrepancies in data collected from various sources. It ensures data continuity and consistency by normalizing the timestamps of each data source to a standard time zone. Precise time adjustment is achieved down to the microsecond level.

[0084] Data format standardization is the process of converting data from various sources into a consistent form. It transforms data of different formats (CSV, JSON, XML, etc.) into a unified structure and standardizes data attributes and units. For example, speed data is converted to km / h, and the number of vehicles is converted to a standardized unit.

[0085] Classifying data by road type is one of the most important processes. The collected data is subdivided into expressways, urban roads, local roads, and so on. This process goes beyond simple classification to reflect the unique characteristics of each road type.

[0086] Highway data focuses on on / off ramps, intersections, and long-distance travel patterns. Urban road data focuses on analyzing signalized intersections, short-distance travel, and complex lane changes. Local road data considers road geometric characteristics, limited infrastructure, and irregular traffic patterns.

[0087] In-depth characteristic analysis is conducted on the data collected for each road type. Various attributes, such as the number of lanes, speed limits, road width, traffic volume variability, and congestion levels by time of day, are analyzed in detail.

[0088] Data that has undergone these cleaning and classification processes is transformed into a high-quality, refined dataset. This data becomes a key resource for the training and prediction of subsequent artificial intelligence models.

[0089]

[0090] Stage 3 (S30)

[0091] The third step is explained in detail with reference to FIG. 4. In FIG. 4, arrows 1 through 3 indicate at the edge

[0092] First, the entire road network is modeled in the form of a graph. Road sections (solid lines on the drawing) are represented as edges, and intersections or junctions (points on the drawing) are represented as nodes. In this process, the attribute values ​​for each edge are set to the number of lanes, speed limit, and maximum processing capacity per hour, while the attribute values ​​for the nodes include whether turns are possible, the signal system, and the structure of the connecting roads.

[0093] Road types are classified into expressways, city roads, and local roads. Since each road type has different characteristics, the network model specifies these through labeling. For example, expressways reflect the characteristics of staggered sections at on / off ramps, city roads reflect the delay characteristics of signalized intersections, and local roads reflect geometric and road alignment characteristics.

[0094] After constructing the network model, all possible paths are explored at each edge. This refers to all possible routes through which a vehicle can reach its destination from its current location. At this stage, constraints are set for each path, encompassing both physical road constraints (capacity, number of lanes, etc.) and operational constraints (signal systems, turn limits, etc.).

[0095] Of particular importance is the calculation of travel costs at each edge. Travel costs are based on the time taken to pass through the corresponding road section, but delay factors based on the characteristics of each road type are reflected as weights.

[0096] The model's operation process is as follows:

[0097] 1. Select a unidirectional edge and search for and label the nearest reachable edge first, considering the forward direction (arrow 1 in Fig. 4).

[0098] 2. Explore and label the next reachable edge excluding the nearest edge (arrow 2 in Fig. 4).

[0099] 3. This process is sequentially extended to explore and label up to n-th degree nearest edges (in the example in Fig. 4, the case where n is 3 is exemplified).

[0100] 4. In each step, the time cost of traversing the edge is assumed to be the same.

[0101] The network model constructed through this process serves as the basis for traffic flow prediction in subsequent stages. In particular, separate models reflecting the characteristics of different road types enable optimized predictions tailored to their respective properties.

[0102] Meanwhile, this process can be expressed mathematically as follows.

[0103] When inferring traffic flows (left / right turns and branching, etc.) in a specific direction (either forward or reverse) of a target edge from traffic flow data of 1st to nth degree adjacent edges (group) in the forward direction, if the arithmetic distribution for 1st to nth degree adjacent edges is f=(x1,x2,x3,...) (xn is one of the traffic flows of the target edge), and if vectors f1, f2 ... fn = (x11,x12,x13,...), (x21,x22,x23,...), (xn1,xn2,xn3,...), respectively, then the appropriate weighted average of them can be expressed as (w1f1+ w2f2 ... + wnfn) / n.

[0104]

[0105] Step 4 (S40)

[0106] This step is specifically the process of determining weights for each road segment (edge) of the road network. To calculate the weights, the basic attribute values ​​of each edge are considered first. The road's maximum hourly vehicle handling capacity, maximum traffic volume per lane, and minimum operating speed are used as basic input values.

[0107] The learning process of the artificial intelligence model for weight calculation is based on the assumption that each vehicle continuously searches for an efficient path. The short-term result of efficient distribution (assuming the assumption that vehicle traffic volume remains static in the short term (meaning a short time sufficient to proceed to neighboring edges)) is assumed to be the vehicle traffic volume at adjacent edges. Then, the branching flow at the current target edge (f=(x1,x2,x3,...)) can be viewed as the distribution at the target edge through the inverse process of traffic at adjacent edges, or as an appropriate interpolation (weighted average) of the distribution processes for each of the 1st to nth order edges.

[0108] Also, one important assumption is the intervention of traffic originating from edges other than the target edge, which can be resolved through (1) the interpolation above, or (2) appropriately mitigated by the assumption that a road with high traffic volume should be considered an "attractive" section at any adjacent edge.

[0109]

[0110] Artificial intelligence models, such as deep neural networks, are used to predict edge-specific weights, receiving these input values ​​and calculating the weight for each edge. In this process, the edge-specific weights of the network model are calculated through the deep neural network so that the weighted average f=(x1,x2,x3,...) derived from the weights (w1, w2, ... wn) converges to the actual road traffic flow (training data).

[0111]

[0112] In addition, the weight calculation method may differ depending on the road type; for example, weights are learned by distinguishing the characteristics of merging sections at entrance and exit ramps for expressways, delay characteristics at signalized intersections for city roads, and geometric and road alignment characteristics for local roads. These calculated weights are interpreted as the distribution of forward traffic flow at target edges and are subsequently utilized for traffic flow prediction.

[0113] The AI ​​model continuously improves its weight calculation method by learning from actual traffic data. By learning patterns or anomalies that occur repeatedly in specific road sections and incorporating them into the weight calculation, more accurate traffic flow prediction becomes possible. Meanwhile, various types of training data can be used to train the AI ​​model; for example, data obtained through Korean Patent Registration No. 10-2705956, "Camera-based Traffic Display System and Method," can be used as training data. Through learning, the AI ​​model calculates weights for all edges, including edges that lack a ground truth (lane-specific flow observation data).

[0114]

[0115] Step 5 (S50)

[0116] In this stage, interactions between lanes are simulated by modeling each lane as an independent agent. For example, the Stochastic Agent-Based Model (SABM) can be used as a simulation model, but it is not limited to this. The Stochastic Agent-Based Model (SABM) is a simulation model that treats each lane as an independent agent and probabilistically models and simulates their interactions (lane changes, entry / exit, etc.). By individually predicting traffic flow for each lane based on real-time data such as traffic volume, speed, and occupancy rate, it enables more accurate prediction of the overall road traffic situation.

[0117] In addition, the Probabilistic Agent-Based Model (SABM) not only calculates predicted traffic volume and speed for each lane based on real-time data such as traffic volume, speed, and occupancy rate for each lane, but also calculates estimated waiting times for each lane at entrances, exits, and junctions.

[0118] Of particular importance is the calculation of travel time information. The Stochastic Agent-Based Model (SABM) calculates travel times based on the predicted traffic volume and speed of each lane, and this information is reflected as a weight in the real-time pathfinding algorithm. For example, if the travel time for a specific lane is predicted to be longer, the pathfinding algorithm can suggest an alternative route that bypasses that lane.

[0119] When actual data for each sub-selection is available, the model's prediction results can be modified using a Bayesian neural network. This contributes to reducing the difference between the actual observed data and the model's predicted values, thereby improving prediction accuracy. Since the Bayesian neural network also provides the uncertainty regarding the predicted values, it is also helpful in evaluating the reliability of the prediction. Here, the Bayesian neural network is a probabilistic neural network model that expresses the network's weights as probability distributions rather than fixed values, and enables the quantification of the uncertainty of predicted values ​​by utilizing Bayes' theorem to update this distribution whenever new data is received. In this invention, when actual data for each sub-selection is available, it is utilized to modify the model's prediction results and calculate confidence intervals.

[0120] The Probabilistic Agent-Based Model (SABM) also predicts traffic flow by considering lane change possibilities and entry / exit probabilities. For example, since lane changes can occur frequently near interchanges (ICs) and junctions (JCs), the interactions of these agents are reflected in the simulation. This enables realistic traffic flow prediction even in complex traffic situations.

[0121]

[0122] Step 6 (S60)

[0123] First, preprocessing is performed to visualize the predicted traffic flow data. Data such as predicted traffic volume, speed, and occupancy rate for each lane are normalized and transformed into a format suitable for visualization.

[0124] Traffic conditions by lane can be assigned colors based on criteria such as smooth flow (green), slow traffic (yellow), and congestion (red).

[0125] In this case, each state is not simply distinguished by three colors, but can be expressed in more detail through a continuous color spectrum. For example, even a stagnant state can be expressed in different shades of red depending on the degree.

[0126] When displayed on a map, the following elements may also be provided.

[0127] 1. Predicted travel time by lane

[0128] 2. Estimated waiting time at entrances, exits, and junctions

[0129] 3. Detailed information on major intersections or bottleneck sections

[0130] 4. Recommended Detour Information

[0131] In addition, hourly forecast information can be provided in the form of sliders or timelines, allowing users to check future traffic conditions in advance. This can assist with planning departure times or selecting routes.

[0132] These visualizations are updated in real time and provide various viewing modes (such as the entire road network or detailed views of specific sections) depending on user needs. Additionally, they provide notifications to users in the event of rapid changes in traffic conditions, enabling immediate response.

[0133]

[0134] Figure 5 is a reconstruction of the lane-by-lane multi-traffic flow prediction method of the above-described embodiment from a hardware perspective. Therefore, to avoid duplication of explanation, only an overview focusing on the function and operation of each component will be briefly described here.

[0135] A computing device (800) includes a memory (830) that stores a program coded so that a computer can read a multi-lane traffic flow prediction method, and a processor (810) that executes said program. The multi-lane traffic flow prediction method includes a first step of collecting single traffic flow data and road type information of adjacent roads; a second step of refining the collected data and classifying the data by road type; a third step of constructing a network by modeling roads and nodes as a graph, setting possible paths and constraints at each edge, and generating a network model for each road type; a fourth step of determining edge weights in each network model for each road type generated through the third step using an artificial intelligence model such as a deep neural network; a fifth step of simulating probabilistic interactions between agents and predicting traffic flow by lane using a simulation model such as a probabilistic agent-based model (SABM); and a sixth step of providing the traffic flow predicted through the fifth step to a user by displaying it in different colors for each lane.

[0136] In the above fifth step, the processor modifies the model's prediction results through a Bayesian neural network if actual data for each lane is available.

[0137] In the second step above, the road types include expressways, city roads, and local roads, and the processor measures the vehicle exit traffic volume of the access road in real time near the connection between the expressway and the general road, reflects the geometric information and signal operation information of the access and exit ramps, and calculates the traffic volume by applying a classified traffic volume model that considers the traffic processing capacity and travel speed characteristics of each road type.

[0138] In the above fourth step, the artificial intelligence model collects constraints such as the maximum hourly vehicle processing capacity of the road, maximum traffic volume and minimum operating speed per lane, and information on intersection signal operation and turn restrictions, and constructs a network model composed of edges capable of actual traffic by considering the collected constraints.

[0139] In the above fifth step, the simulation model calculates travel time information based on predicted traffic volume and speed for each lane, calculates the estimated waiting time for each lane at entrances / exits and junctions, and incorporates the above information as weights into a real-time path search algorithm to improve the accuracy of optimal path search.

[0140] In the first step above, when external factors including traffic accidents, road construction, and weather information occur, the processor quantifies the type and severity of the event, spatially sets the range of influence, and calculates the duration of the event as a temporal range to calculate the rate of reduction in traffic flow capacity relative to the normal state and reflects it in the network model.

[0141] In the first step above, the processor collects one or more of traffic-related public data, including vehicle traffic volume data collected from traffic volume detectors for each road section, video data collected from CCTVs for each road section, GPS data of probe vehicles collected from vehicle navigation terminals, Hi-Pass data, and signal control data.

[0142] In the second step above, the processor interpolates missing values ​​of the collected data, removes outliers, synchronizes time units, and standardizes the data format.

[0143] In the third step above, the processor defines roads as edges and intersections and junctions as nodes, sets the number of lanes, speed limit, and maximum processing capacity per hour of each edge as attribute values, sets turning restrictions, signal display, and road geometry at the nodes as constraints, and calculates the final travel cost of the corresponding edge by reflecting the delay based on the characteristics of the intersection section of the entrance / exit ramps for highways, the waiting time based on the delay characteristics of signal intersections for city roads, and driving constraints based on the geometry and road alignment characteristics for local roads as weights.

[0144]

[0145] Embodiments according to the present disclosure may be implemented by various means, e.g., hardware, firmware, software, or a combination thereof. In the case of implementation by hardware, an embodiment of the present disclosure may be implemented by one or more ASICs (application specific integrated circuits), DSPs (digital signal processors), DSPDs (digital signal processing devices), PLDs (programmable logic devices), FPGAs (field programmable gate arrays), processors, controllers, microcontrollers, microprocessors, etc. In the case of implementation by firmware or software, an embodiment of the present disclosure may be implemented in the form of modules, procedures, functions, etc., that perform the capabilities or operations described above. Software code may be stored in memory and executed by a processor. The memory may be located inside or outside the processor and may exchange data with the processor by various means already known.

[0146] Meanwhile, the lane-specific multi-traffic flow prediction method of the above-described embodiment can be implemented as computer-readable code on a computer-readable recording medium. A computer-readable recording medium includes all types of recording devices in which data that can be read by a computer system is stored.

[0147] Examples of computer-readable recording media include ROM, RAM, CD-ROM, magnetic tape, floppy disk, optical data storage device, etc. Additionally, computer-readable recording media may be distributed across networked computer systems, allowing computer-readable code to be stored and executed in a distributed manner. Furthermore, functional programs, codes, and code segments for implementing the present invention can be easily inferred by programmers in the art to which the present invention belongs.

[0148] The present invention has been described above with reference to various embodiments. Those skilled in the art will understand that the present invention may be implemented in modified forms without departing from the essential characteristics of the invention. Therefore, the disclosed embodiments should be considered in an illustrative rather than a restrictive sense. The scope of the invention is defined by the claims, not by the foregoing description, and all variations within the scope of the claims should be interpreted as being included in the invention.

Claims

1. In a traffic flow prediction method that predicts traffic flow by lane or multiple flows using data classified by road type, Step 1: Collecting single traffic flow data and road type information of adjacent roads; Step 2: Refining the collected data and classifying the data by road type; A third step of constructing a network by modeling roads and nodes as a graph, setting possible paths and constraints at each edge, and generating a network model for each road type; A fourth step of determining edge weights in each network model for each road type generated through the third step using an artificial intelligence model such as a deep neural network; Step 5, simulating probabilistic interactions between agents and predicting lane-specific traffic flow through a simulation model such as a Probabilistic Agent-Based Model (SABM); Step 6, which provides the user with the traffic flow predicted through Step 5 by displaying it in different colors for each lane; A traffic flow prediction method including 2. In Paragraph 1, The above fifth step is, A traffic flow prediction method that modifies the model's prediction results through a Bayesian neural network when actual data for each lane is available.

3. In Paragraph 1, In the second step above, Road types include expressways, city roads, and local roads, and A traffic flow prediction method characterized by measuring the volume of vehicles exiting the access road in real time near the connection point between the expressway and the general road, reflecting geometric information and signal operation information of the access and exit connections, and calculating the traffic volume by applying a classified traffic volume model that considers the traffic processing capacity and travel speed characteristics of each road type.

4. In Paragraph 1, In the above fourth step, The above artificial intelligence model is a traffic flow prediction method that collects constraints such as the maximum hourly vehicle processing capacity of the road, maximum traffic volume and minimum operating speed per lane, and information on intersection signal operation and turn limit, and constructs a network model composed of edges capable of actual traffic by considering the collected constraints.

5. In Paragraph 1, In the above fifth step, A traffic flow prediction method characterized by the above simulation model calculating travel time information based on predicted traffic volume and speed by lane, calculating expected waiting time by lane at entrance / exit ramps and junctions, and reflecting the above information as weights in a real-time path search algorithm to improve the accuracy of optimal path search.

6. In Paragraph 1, In the first step above, A traffic flow prediction method characterized by quantifying the type and severity of the event, spatially setting the range of influence, and calculating the duration of the event as a temporal range, and calculating the rate of reduction in traffic flow capacity relative to the normal state and reflecting it in the network model when external factors including traffic accidents, road construction, and weather information occur.

7. In Paragraph 1, The above first step is, Vehicle traffic volume data collected from traffic volume detectors for each road section; Video data collected from CCTVs for each road section; GPS data of the probe vehicle collected from the vehicle navigation terminal; and Traffic-related public data including Hi-Pass data and signal control data A traffic flow prediction method characterized by collecting one or more of the following.

8. In Paragraph 1, The above second step is, Interpolate missing values ​​of the collected data; Removing outliers; Synchronize time units; Standardizing data formats; A traffic flow prediction method characterized by classifying data by road type.

9. In Paragraph 1, The above third step is, Define roads as edges, and intersections and junctions as nodes; The number of lanes, speed limit, and maximum processing capacity per hour of each edge are set as attribute values; Set rotation limits, signal display, and connection path geometry at the nodes as constraints; A traffic flow prediction method characterized by calculating the final travel cost of each edge by reflecting as weights the delay degree according to the characteristics of the intersection section of the entrance / exit connecting road for highways, the waiting time according to the delay characteristics of signal intersections for city roads, and driving constraints according to geometric structure and road alignment characteristics for local roads.

10. Memory for storing a program coded so that a multi-lane traffic flow prediction method can be read by a computer; It includes a processor that executes the above program, The above lane-specific multi-traffic flow prediction method is, Step 1: Collecting single traffic flow data and road type information of adjacent roads; Step 2: Refining the collected data and classifying the data by road type; A third step of constructing a network by modeling roads and nodes as a graph, setting possible paths and constraints at each edge, and generating a network model for each road type; A fourth step of determining edge weights in each network model for each road type generated through the third step using an artificial intelligence model such as a deep neural network; Step 5, simulating probabilistic interactions between agents and predicting lane-specific traffic flow through a simulation model such as a Probabilistic Agent-Based Model (SABM); Step 6, which provides the user with the traffic flow predicted through Step 5 by displaying it in different colors for each lane; An arithmetic unit including 11. In Paragraph 10, In the above fifth step, the processor is a computing device that modifies the prediction result of the model through a Bayesian neural network when actual data for each lane is available.

12. In Paragraph 10, In the second step above, the processor is a computing device that calculates a waiting time by applying a separate traffic volume model that considers traffic processing capacity and travel speed characteristics by road type, and in the vicinity of the connection between the expressway and the general road, measures the traffic volume of vehicles exiting the access road in real time, reflects geometric information and signal operation information of the access and exit connections, and includes road types such as expressways, city roads, and local roads.

13. In Paragraph 10, In the above fourth step, the artificial intelligence model collects constraints such as the maximum hourly vehicle processing capacity of the road, maximum traffic volume and minimum operating speed per lane, and information on intersection signal operation and turn limit, and constructs a network model composed of edges capable of actual traffic by considering the collected constraints, a computing device.

14. In Paragraph 10, In the above fifth step, the simulation model calculates travel time information based on predicted traffic volume and speed by lane, calculates the expected waiting time by lane at entrance / exit ramps and junctions, and incorporates the information as weights into a real-time path search algorithm to improve the accuracy of optimal path search, a computing device.

15. In Paragraph 10, In the first step above, the processor is a computing device that, when external factors including traffic accidents, road construction, and weather information occur, quantifies the type and severity of the event, spatially sets the range of influence, calculates the duration of the event as a temporal range, calculates the rate of reduction in traffic flow capacity relative to the normal state, and reflects it in the network model.

16. In Paragraph 10, In the first step above, the processor is a computing device that collects one or more of traffic-related public data, including vehicle traffic volume data collected from a traffic volume detector for each road section, video data collected from a CCTV for each road section, GPS data of a probe vehicle collected from a vehicle navigation terminal, Hi-Pass data, and signal control data.

17. In Paragraph 10, In the second step above, the processor is a computing device that interpolates missing values ​​of collected data, removes outliers, synchronizes time units, and standardizes data formats.

18. In Paragraph 10, In the third step above, the processor defines roads as edges and intersections and junctions as nodes, sets the number of lanes, speed limit, and maximum processing capacity per hour of each edge as attribute values, sets turning restrictions, signal display, and connecting road geometry at the nodes as constraints, and calculates the final travel cost of the corresponding edge by reflecting as weights the delay degree according to the characteristics of the intersection section of the entrance / exit connecting road for highways, the waiting time according to the delay characteristics of signal intersections for city roads, and driving constraints according to the geometry and road alignment characteristics for local roads.

19. In one or more non-transitory computer-readable media storing one or more instructions, The one or more instructions executable by one or more processors are, Step 1: Collecting single traffic flow data and road type information of adjacent roads; Step 2: Refining the collected data and classifying the data by road type; A third step of constructing a network by modeling roads and nodes as a graph, setting possible paths and constraints at each edge, and generating a network model for each road type; A fourth step of determining edge weights in each network model for each road type generated through the third step using an artificial intelligence model such as a deep neural network; Step 5, simulating probabilistic interactions between agents and predicting lane-specific traffic flow through a simulation model such as a Probabilistic Agent-Based Model (SABM); Step 6, which provides the user with the traffic flow predicted through Step 5 by displaying it in different colors for each lane; By including, A computer-readable medium that implements a method for predicting traffic flow by lane or multiple flows.

Images