Traffic congestion bottleneck estimation device and traffic congestion bottleneck estimation method

Abstract

The location of traffic congestion bottlenecks is estimated with high accuracy using only measurement data from roadside sensors. [Solution] The traffic congestion bottleneck estimation device of the embodiment includes: an acquisition unit that acquires measurement data from roadside sensors that measure the traffic conditions on a road on which a vehicle is traveling; a learning preprocessing unit that performs predetermined preprocessing on the measurement data to highlight the characteristics of the traffic conditions and outputs learning preprocessed data; a learning data creation unit that creates learning data, which is training data for machine learning, using the learning preprocessed data; a learning model generation unit that generates a learning model, which is a trained model, by machine learning using the learning data; an estimation preprocessing unit that performs preprocessing on the measurement data that is the target of estimation of the road traffic congestion bottleneck and outputs estimation preprocessed data; and a traffic congestion bottleneck estimation unit that estimates the location of the traffic congestion bottleneck using the estimation preprocessed data and the learning model and outputs the estimation result.

Description

Technical Field

[0001] Embodiments of the present invention relate to a traffic bottleneck estimation device and a traffic bottleneck estimation method.

Background Art

[0002] Conventionally, in freeway traffic control, it is necessary to improve traffic congestion. To that end, it is important to grasp traffic conditions such as traffic congestion and to grasp the location of a traffic bottleneck, which is the cause of the traffic congestion, and the location of the end of the traffic congestion.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Patent Document 2

Patent Document 3

Patent Document 4

Patent Document 5

Non-Patent Documents

[0004]

Non-Patent Document 1

Non-Patent Document 2

[0005] For example, methods for determining the end of a traffic jam have been put into practical use. On the other hand, for traffic bottlenecks, the location of the bottleneck is often determined by experts with the necessary know-how, based on accumulated measurement data from roadside sensors. In such cases, the determination process involves human effort and time.

[0006] In addition, some studies involve creating speed and traffic volume plots from measurement data from vehicle sensors at various points along the target route, and then identifying the location of congestion bottlenecks based on the characteristics of these bottlenecks in the plots. However, although attempts are being made to experiment with statistical methods, practical application has not yet been achieved.

[0007] Furthermore, some methods use roadside sensor (vehicle detector) data and probe car data to predict traffic congestion. However, these methods predict congestion based on speed information and do not directly estimate or predict the location of traffic bottlenecks.

[0008] Furthermore, in addition to regular congestion bottlenecks, there may be cases where congestion bottlenecks occur unexpectedly for various reasons, or where the location of the congestion bottleneck shifts slightly depending on traffic conditions. Moreover, changes in surrounding traffic conditions (traffic environment), changes in traffic conditions over time, and changes in traffic conditions due to transportation policies such as the addition or removal of routes may also cause changes in the location and occurrence of congestion bottlenecks. In such cases, identifying the location of the congestion bottleneck requires analysis and identification that is time-consuming and requires human resources each time.

[0009] Therefore, the objective of this embodiment is to provide a traffic congestion bottleneck estimation device and a traffic congestion bottleneck estimation method that can estimate the location of a traffic congestion bottleneck with high accuracy using only measurement data from roadside sensors. [Means for solving the problem]

[0010] The traffic congestion bottleneck estimation device of the embodiment includes: an acquisition unit that acquires measurement data from roadside sensors that measure traffic conditions on roads on which vehicles travel; an accumulation processing unit that stores the measurement data in a storage unit; a learning preprocessing unit that performs predetermined preprocessing on the measurement data stored in the storage unit to highlight the characteristics of the traffic conditions and outputs learning preprocessed data; a learning data creation unit that creates learning data, which is training data for machine learning, using the learning preprocessed data; a learning model generation unit that generates a learning model, which is a trained model, by machine learning using the learning data; an estimation preprocessing unit that performs the preprocessing on the measurement data that is the target of estimation of traffic congestion bottlenecks on roads and outputs estimation preprocessed data; and a traffic congestion bottleneck estimation unit that estimates the location of the traffic congestion bottleneck using the estimation preprocessed data and the learning model and outputs the estimation result. [Brief explanation of the drawing]

[0011] [Figure 1] Figure 1 is a schematic diagram showing a road in the first embodiment. [Figure 2]FIG. 2 is an overall configuration diagram of the traffic control system according to the first embodiment. [Figure 3] FIG. 3 is a functional configuration diagram of the traffic congestion bottleneck estimation device according to the first embodiment. [Figure 4] FIG. 4 is an explanatory diagram of the processing during learning. [Figure 5] FIG. 5 is a functional configuration diagram of the preprocessing unit and the learning data creation unit during learning. [Figure 6] FIG. 6 is an explanatory diagram of the processing during estimation. [Figure 7] FIG. 7 is a functional configuration diagram of the preprocessing unit during estimation. [Figure 8] FIG. 8 is an explanatory diagram of a processing example according to the first embodiment. [Figure 9] FIG. 9 is a flowchart showing the processing during learning by the traffic congestion bottleneck estimation device according to the first embodiment. [Figure 10] FIG. 10 is a flowchart showing the processing during estimation by the traffic congestion bottleneck estimation device according to the first embodiment. [Figure 11] FIG. 11 is a functional configuration diagram of the traffic congestion bottleneck estimation device according to the second embodiment. [Figure 12] FIG. 12 is an explanatory diagram of the method according to the third embodiment. [Figure 13] FIG. 13 is an explanatory diagram of the method according to the fourth embodiment. [Figure 14] FIG. 14 is a diagram showing an example of a screen according to the fifth embodiment.

MODE FOR CARRYING OUT THE INVENTION

[0012] Hereinafter, embodiments (first embodiment to sixth embodiment) of the traffic congestion bottleneck estimation device and the traffic congestion bottleneck estimation method of the present invention will be described with reference to the drawings. In the second embodiment and subsequent embodiments, descriptions of matters similar to those already described will be omitted as appropriate.

[0013] (First Embodiment) First, the first embodiment will be described. The first embodiment is a traffic congestion bottleneck estimation device that uses measurement data acquired from roadside sensors (vehicle detectors 2) as training data to create a learning model, and then uses that learning model to estimate the location of the traffic congestion bottleneck.

[0014] Referring to Figure 1, the roads R1 and R2 in the first embodiment will be described. Figure 1 is a schematic diagram showing roads R1 and R2 in the first embodiment. Roads R1 and R2 on which vehicle C travels are, for example, expressways. Vehicle detectors 2 (examples of roadside sensors that measure traffic conditions on the road on which a vehicle travels) are installed on roads R1 and R2 in sections #1, #2, #3, ...

[0015] Figure 1 shows a bidirectional road with two lanes in each direction of vehicle travel. Vehicle detectors 2 are installed in each section and in each lane. Vehicle detectors 2 are, for example, loop coil type traffic counters embedded in the road. However, vehicle detectors 2 are not limited to loop coil type traffic counters; other means such as a combination of a CCTV (Closed Circuit Television System) camera and an image processing device, or a sensor that measures congestion using ultrasonic pulses, may also be used.

[0016] Furthermore, since vehicle detectors 2 are installed in each lane at the same point in the direction of vehicle travel, the road traffic control system 3, which receives detector data from each vehicle detector 2, can create information on the traffic conditions at each location and use it as location information.

[0017] Next, with reference to Figure 2, the overall configuration of the traffic control system S of the embodiment will be described. Figure 2 is an overall configuration diagram of the traffic control system S according to the embodiment. The traffic control system S comprises a congestion bottleneck estimation device 1, a vehicle detector 2, and a road traffic control system 3.

[0018] The Road Traffic Control System 3 is a computer system that comprehensively monitors and manages the actual traffic conditions on the roads under control. The Road Traffic Control System 3 displays traffic conditions based on sensor data (measurement data) received from the vehicle sensors 2, and transmits the sensor data to the congestion bottleneck estimation device 1.

[0019] Vehicle detectors 2 (see Figure 1) are installed along the roadside in each section of the expressway. Vehicle detectors 2 detect vehicles C passing near their installation location and collect detector data based on the detection results. The detector data may include, for example, traffic volume [vehicles / hour], average speed [km / h], vehicle density [vehicles / km], and occupancy rate [%]. By using such vehicle detectors 2, detector data for all vehicles passing through the section in which vehicle detectors 2 are installed can be obtained. Vehicle detectors 2 transmit the collected detector data to the road traffic control system 3.

[0020] The traffic congestion bottleneck estimation device 1 is a device that estimates the location of traffic congestion bottlenecks on a road. The traffic congestion bottleneck estimation device 1 receives sensor data from vehicle sensors 2 and other traffic-related data from the road traffic control system 3.

[0021] Next, with reference to Figure 3, the functional configuration of the traffic congestion bottleneck estimation device 1 of the embodiment will be described. Figure 3 is a functional configuration diagram of the traffic congestion bottleneck estimation device 1 of the first embodiment. The traffic congestion bottleneck estimation device 1 is a computer device and comprises a processing unit 11, a storage unit 12, an input unit 13, a display unit 14, and a communication unit 15.

[0022] In the first embodiment, for the sake of brevity, the congestion bottleneck estimation device 1 is described as being composed of a single computer device, but it is not limited to this. The congestion bottleneck estimation device 1 may be implemented by, for example, multiple computer devices, or by a cloud server.

[0023] The memory unit 12 is a storage device such as an HDD (Hard Disk Drive) or SSD (Solid State Drive) and stores various types of information. For example, the memory unit 12 stores road data 121, sensor data 122, learning data 123, a learning model 124, and estimation results 125.

[0024] Road data 121 is information about the structure of the road, such as section identification information, length (section length), maximum vehicle capacity, number of lanes, interchanges, and parking area locations.

[0025] Sensor data 122 is information obtained from the road traffic control system 3, collected by the vehicle sensor 2, such as traffic volume [vehicles / h], average speed [km / h], vehicle density [vehicles / km], and occupancy rate [%].

[0026] Training data 123 is training data for machine learning (details below).

[0027] The learning model 124 is a model for estimating the location of traffic bottlenecks, etc., and is generated using the learning data 123 by the learning model generation unit 115, which will be described later (details will be described later).

[0028] The estimation result 125 is information that shows the estimation results, such as the location of the congestion bottleneck, generated by the traffic condition estimation unit 118 (congestion bottleneck estimation unit), which will be described later, using the learning model 124.

[0029] The data stored in the memory unit 12 is not limited to the above. For example, the memory unit 12 may store weather data related to the weather near the road being monitored, event data related to events taking place near the road being monitored, and so on.

[0030] The processing unit 11 includes, for example, a CPU (Central Processing Unit), ROM (Read Only Memory), and RAM (Random Access Memory). The CPU comprehensively controls the operation of the congestion bottleneck estimation device 1. ROM is a storage medium for storing various programs and data. RAM is a storage medium for temporarily storing various programs and for rewriting various data. The CPU then uses RAM as a work area to execute programs stored in the ROM, storage unit 12, etc.

[0031] The processing unit 11 includes, as a functional configuration, an acquisition unit 111, a storage processing unit 112, a pre-processing unit for learning 113, a learning data creation unit 114, a learning model generation unit 115, a pre-processing unit for estimation 116, a model selection unit 117, a traffic situation estimation unit 118, and a control unit 119.

[0032] The learning process and the estimation process will be explained separately below. Note that in the explanation of the estimation process, the same matters as those described in the learning process will be omitted as appropriate.

[0033] First, let's explain the processing during training. In addition to Figure 3, please also refer to Figure 4 below. Figure 4 is an explanatory diagram of the learning process.

[0034] The acquisition unit 111 acquires various information from external devices. For example, the acquisition unit 111 acquires sensor data collected by the vehicle sensors 2 from the road traffic control system 3 for each section.

[0035] The storage processing unit 112 stores the sensor data acquired by the acquisition unit 111 in the storage unit 12 as sensor data 122.

[0036] The learning preprocessing unit 113 performs predetermined preprocessing on the sensor data 122 to highlight the characteristics of the traffic situation and outputs the learning preprocessed data.

[0037] Please also refer to Figure 5 below. Figure 5 is a functional configuration diagram of the pre-processing unit 113 and the training data creation unit 114.

[0038] The learning preprocessing unit 113 includes a necessary data extraction unit 1131, a graph creation unit 1132, a feature highlighting unit 1133, a graph resizing unit 1134, and a graph digitization unit 1135.

[0039] The necessary data extraction unit 1131 extracts the necessary data required for machine learning from the sensor data 122. When the necessary data extraction unit 1131 extracts the necessary data required for machine learning from the sensor data 122, for example, it extracts sensor data 122 for a set period.

[0040] The graph creation unit 1132 uses the necessary data to create a plot graph with traffic volume and speed as the two axes.

[0041] The feature enhancement processing unit 1133 performs processing on the plot graph to enhance the characteristics of the traffic situation and outputs a feature enhancement plot graph.

[0042] The graph resizing unit 1134 adjusts the size of the feature highlighting plot graph so that it can be easily quantified by the graph digitization unit 1135, and outputs a resized graph.

[0043] The graph digitization unit 1135 digitizes each part of the resized graph and outputs numerical data.

[0044] The training data creation unit 114 uses numerical data to create training data 123, which is training data for machine learning.

[0045] The learning data creation unit 114 includes a label addition unit 1141 and a learning data storage and extraction unit 1142.

[0046] The label addition unit 1141 adds labels for training data (correct labels) to numerical data (data preprocessed data).

[0047] The learning data storage and extraction unit 1142 stores training data, including numerical data and labels, as learning data 123 in the storage unit 12, and extracts the learning data 123 to be used for learning from the storage unit 12.

[0048] The learning model generation unit 115 generates a trained model 124 using machine learning with the training data 123. This concludes the explanation of the learning process for now.

[0049] Next, we will explain the processing during estimation. In addition to Figure 3, please also refer to Figure 6 below. Figure 6 is an explanatory diagram of the estimation process.

[0050] The acquisition unit 111 acquires sensor data collected by the vehicle sensors 2 from the road traffic control system 3 for each section.

[0051] The estimation preprocessing unit 116 performs predetermined preprocessing on the sensor data 122 to highlight the characteristics of the traffic situation and outputs the estimation preprocessed data.

[0052] Please also refer to Figure 7 below. Figure 7 is a functional configuration diagram of the estimation preprocessing unit 116.

[0053] The estimation preprocessing unit 116 includes a necessary data extraction unit 1161, a graph creation unit 1162, a feature highlighting processing unit 1163, a graph resizing unit 1164, and a graph digitization unit 1165.

[0054] The necessary data extraction unit 1161 extracts the necessary data required for estimation from the sensor data 122.

[0055] The graph creation unit 1162 uses the necessary data to create a plot graph with traffic volume and speed as two axes.

[0056] The feature enhancement processing unit 1163 performs processing on the plot graph to enhance the characteristics of the traffic situation and outputs a feature enhancement plot graph.

[0057] The graph resizing unit 1164 adjusts the size of the feature highlighting plot graph so that it can be easily quantified by the graph digitization unit 1165, and outputs a resized graph.

[0058] The graph digitization unit 1165 digitizes each part of the resized graph and outputs numerical data.

[0059] The model selection unit 117 selects a learning model from among several learning models 124 to be used by the traffic situation estimation unit 118.

[0060] The traffic condition estimation unit 118 uses the pre-processed estimation data (numerical data output by the graph digitization unit 1165) output by the estimation preprocessing unit 116 and the learning model 124 selected by the model selection unit 117 to estimate the location of the congestion bottleneck and outputs the estimation result 125. This concludes the explanation of the estimation process for now.

[0061] Returning to Figure 3, the control unit 119 performs various controls. For example, the control unit 119 (display control unit) displays various information on the display unit 14.

[0062] The input unit 13 is an input device that accepts user (e.g., system administrator, etc.) operations on the congestion bottleneck estimation device 1, and is, for example, a keyboard, mouse, etc.

[0063] The display unit 14 is a liquid crystal display (LCD), an electro-luminescence (OLED) display, or the like.

[0064] The communication unit 15 is a communication interface for communicating with external devices.

[0065] Next, with reference to Figure 8, an example of the processing in the first embodiment will be described (refer to other figures as appropriate).

[0066] First, let's explain the processing during learning. The necessary data extraction unit 1131 extracts the necessary data required for machine learning from the sensor data 122. When the necessary data extraction unit 1131 extracts the necessary data required for machine learning from the sensor data 122, it extracts, for example, sensor data 122 for a set period. For example, for a target route, the necessary data extraction unit 1131 extracts two days' worth of necessary data from one year's worth of sensor data 122. Hereafter, sensor data 122 may be simply referred to as "data".

[0067] Furthermore, as data to be extracted, for example, as described in Non-Patent Document 2, it is conceivable to extract data from three locations: the immediate downstream (TC3 in Figure 8), the immediate upstream (TC2 in Figure 8), and the immediate upstream (TC1 in Figure 8) of the target location. Non-Patent Document 2 states that there are characteristics immediately downstream, immediately upstream, and upstream of a traffic congestion bottleneck. If the data from these locations shows the same characteristics as when a traffic congestion bottleneck exists, it can be estimated that a traffic congestion bottleneck exists at that location.

[0068] Furthermore, Non-Patent Document 2 states that when a congestion bottleneck exists, traffic volume is limited downstream by the bottleneck, so no congestion occurs downstream. However, upstream, slightly away from the bottleneck, the congestion extends, resulting in a situation where traffic goes from uncongested to congested, and thus both uncongested and congested periods occur. It also states that immediately upstream of the congestion bottleneck, a situation just before congestion occurs (referred to as the critical region), resulting in traffic conditions that are somewhere between congested and uncongested.

[0069] Therefore, when these characteristics are observed, it is conceivable that a congestion bottleneck is present. Non-patent document 2 explains these characteristics from a speed-traffic volume plot. Therefore, it is thought that these characteristics can be determined from the speed-traffic volume plot by human observation. In order to automate these human judgments, this embodiment proposes a method to automatically perform congestion bottleneck estimation equivalent to that performed by a human by creating a learning model 124 that estimates the location of the congestion bottleneck based on the speed-traffic volume plot and training it to perform human judgments.

[0070] The graph creation unit 1132 uses the necessary data to create plot graphs (A1-A3 in Figure 8) with two axes: traffic volume (horizontal axis; higher to the right) and speed (vertical axis; higher to the top). The plot graphs created here are for three locations near the congestion bottleneck (downstream of the bottleneck, immediately upstream, and upstream).

[0071] The feature enhancement processing unit 1133 processes the plot graphs (A1-A3 in Figure 8) to enhance the characteristics of traffic conditions and outputs feature enhancement plot graphs (B1-B3 in Figure 8). For example, the feature enhancement processing unit 1133 sets the maximum traffic volume value of the data being handled as the maximum value on the traffic volume axis, or the maximum speed value of the data being handled as the maximum value on the speed axis, to create a graph that clearly shows the characteristics of the graph being created. In other words, if the maximum and minimum values ​​of each axis of the multiple graphs used are different, it becomes difficult to check the distribution characteristics of those graphs, so the process of matching the maximum and minimum values ​​of each axis is performed here.

[0072] The graph resizing unit 1134 adjusts the size of the feature highlighting plot graphs (B1-B3 in Figure 8) so that they can be easily digitized by the graph digitization unit 1135, and outputs resized graphs (C1-C3 in Figure 8). Furthermore, for example, if it is preferable for processing to have graphs with the same aspect ratio, the aspect ratio can be compressed to the same size. It is also possible to change the resolution of the graphs to match the processing performance of the computer being implemented. For example, in the example in Figure 8, the aspect ratio of Figures B1-B3 is changed and the resolution is reduced to produce Figures C1-C3.

[0073] The graph digitization unit 1135 digitizes each part of the resized graph (C1-C3 in Figure 8) and outputs numerical data (D1-D3 in Figure 8) (pre-processed data from training). Here, for digitization, for example, the digitization from the graph can be performed using Python libraries such as scikit-learn or Pillow.

[0074] D1-D3 in Figure 8 are input data for the machine learning model. The label addition unit 1141 adds labels (correct labels) for training data to this input data D1-D3 in Figure 8, thereby creating training data which is then stored as training data 123. The labels are, for example, "1" for a congestion bottleneck and "0" for no congestion bottleneck. The labels can be created in advance, for example, by experts determining whether or not there is a congestion bottleneck.

[0075] The learning model generation unit 115 then generates a trained model 124 using machine learning with the training data 123. Possible learning models include, for example, decision tree-based random forests or XGBoost. Furthermore, since the learning process only requires that the output becomes training data for the input data, deep learning can also be used.

[0076] More specifically, the process is as follows: A learning model 124 is created by performing machine learning using multiple "combination data of input data (D1-D3 in Figure 8) and correct labels (training data)".

[0077] Furthermore, regarding learning, in the part labeled P in Figure 8, the learning model generation unit 115 calculates the error between the output of the learning model 124 and the correct answer (training data) of the congestion bottleneck, and creates the learning model 124 by learning to reduce this error.

[0078] Next, during estimation, the traffic condition estimation unit 118 uses the learning model 124 created in this way and the pre-processed estimation data output by the estimation preprocessing unit 116 to estimate the location of the congestion bottleneck and output the estimation result. For each point, the estimation result is "1" if it is a congestion bottleneck and "0" if it is not a congestion bottleneck.

[0079] Next, referring to Figure 9, the processing during learning by the traffic congestion bottleneck estimation device 1 of the first embodiment will be described. Figure 9 is a flowchart showing the processing during learning by the traffic congestion bottleneck estimation device 1 of the first embodiment (refer to other figures as appropriate).

[0080] In step S1, the acquisition unit 111 acquires sensor data collected by the vehicle sensors 2 from the road traffic control system 3 for each section.

[0081] Next, in step S2, the storage processing unit 112 stores the sensor data acquired in step S1 in the storage unit 12 as sensor data 122.

[0082] Next, in step S3, the necessary data extraction unit 1131 extracts the necessary data required for machine learning from the sensor data 122.

[0083] Next, in step S4, the graph creation unit 1132 uses the necessary data extracted in step S3 to create a plot graph (A1-A3 in Figure 8) with traffic volume and speed as the two axes.

[0084] Next, in step S5, the feature enhancement processing unit 1133 performs a process on the plot graph (A1 to A3 in Figure 8) to enhance the characteristics of the traffic situation and outputs a feature enhancement plot graph (B1 to B3 in Figure 8).

[0085] Next, in step S6, the graph resizing unit 1134 adjusts the size of the feature highlighting plot graphs (B1 to B3 in Figure 8) and outputs resized graphs (C1 to C3 in Figure 8).

[0086] Next, in step S7, the graph digitization unit 1135 digitizes each part of the resized graph (C1 to C3 in Figure 8) and outputs numerical data (D1 to D3 in Figure 8).

[0087] Next, in step S8, the label addition unit 1141 adds labels for training data to the numerical data (D1 to D3 in Figure 8).

[0088] Next, in step S9, the learning data storage and extraction unit 1142 stores the training data, which includes numerical data (D1 to D3 in Figure 8) and labels ("0", "1"), as learning data 123 in the storage unit 12, and extracts the learning data 123 to be used for learning from the storage unit 12.

[0089] Next, in step S10, the learning model generation unit 115 generates a learning model 124 using machine learning with the learning data 123.

[0090] Next, Figure 10 illustrates the processing during estimation by the traffic congestion bottleneck estimation device 1 of the first embodiment. Figure 10 is a flowchart showing the processing during estimation by the traffic congestion bottleneck estimation device 1 of the first embodiment.

[0091] In step S11, the estimation preprocessing unit 116 acquires the sensor data 122 from the storage unit 12.

[0092] Next, in step S12, the necessary data extraction unit 1161 extracts the necessary data required for machine learning from the sensor data 122 acquired in step S11.

[0093] Next, in step S13, the graph creation unit 1162 uses the necessary data extracted in step S12 to create a plot graph (similar to A1-A3 in Figure 8) with traffic volume and speed as two axes.

[0094] Next, in step S14, the feature highlighting processing unit 1163 performs a process to highlight the characteristics of the traffic situation on the plot graph (a plot graph similar to A1 to A3 in Figure 8) and outputs a feature highlighting plot graph (a feature highlighting plot graph similar to B1 to B3 in Figure 8).

[0095] Next, in step S15, the graph resizing unit 1164 adjusts the size of the feature highlighting plot graph (a feature highlighting plot graph similar to B1 to B3 in Figure 8) and outputs a resized graph (a resized graph similar to C1 to C3 in Figure 8).

[0096] Next, in step S16, the estimation preprocessing unit 116 quantifies each part of the resized graph (a resized graph similar to C1-C3 in Figure 8) and outputs numerical data (numerical data similar to D1-D3 in Figure 8).

[0097] Next, in step S17, the model selection unit 117 selects a learning model from among the multiple learning models 124 to be used by the traffic situation estimation unit 118.

[0098] Next, in step S18, the traffic condition estimation unit 118 uses the numerical data output in step S16 (numerical data similar to D1 to D3 in Figure 8) and the learning model 124 selected in step S17 to estimate the location of the congestion bottleneck and outputs the estimation result. After that, the control unit 119 displays the estimation result on the display unit 14.

[0099] As described above, according to the first embodiment, the above-described process makes it possible to estimate the location of the congestion bottleneck with high accuracy using only the sensor data 122 from the vehicle sensor 2. Therefore, in highway traffic control, when sensor data from the vehicle sensor 2 is available, it becomes possible to automatically estimate the location of the congestion bottleneck, and in traffic management, it is possible to contribute to traffic management and traffic policies by dividing the target route into congestion bottleneck units for management.

[0100] Furthermore, by using machine learning to train a model that can automatically determine the location of traffic bottlenecks in the same way as experts, it is possible to reduce human workload and time, and to estimate the location of traffic bottlenecks with the same accuracy as experts. If the location of traffic bottlenecks can be automatically estimated in this way, the effort required for conventional traffic bottleneck analysis will be reduced, and the location of traffic bottlenecks can be estimated in a shorter time, enabling more efficient traffic management and application to traffic policies.

[0101] (Second Embodiment) Next, a second embodiment will be described. The second embodiment is an embodiment in which a traffic flow simulation unit is added, and preprocessing is performed on the measurement data and traffic flow simulation data that are the targets for estimating road congestion bottlenecks, and preprocessed data for estimation is output.

[0102] Figure 11 is a functional configuration diagram of the congestion bottleneck estimation device 1 of the second embodiment. Compared to Figure 3, a traffic flow simulation unit 120 has been added to the processing unit 11.

[0103] The traffic flow simulation unit 120 creates traffic flow simulation data, which is simulation data of traffic flow that has not yet been acquired, through traffic flow simulation. In this case, the estimation preprocessing unit 116 preprocesses the measurement data and traffic flow simulation data that are the targets for estimation of road congestion bottlenecks, and outputs estimation preprocessed data. This will be explained below.

[0104] When extracting sensor data 122 for estimation, even if there are missing data points in the sensor data 122 to be used, the method of the first embodiment can be used by supplementing the missing data with the results of the traffic flow simulation performed by the traffic flow simulation unit 120. Furthermore, by performing future simulations in the traffic flow simulation and using the results of future simulations as data, it becomes possible to estimate the location of congestion bottlenecks for future conditions.

[0105] Thus, according to the second embodiment, by supplementing the parts where sensor data is missing with traffic flow simulation data, it becomes possible to estimate the location of congestion bottlenecks in the same way as in the first embodiment, even when there are missing data in the sensor data or regarding future conditions.

[0106] (Third embodiment) Next, a third embodiment will be described. The third embodiment is one in which labels indicating traffic conditions are added to numerical data, and a learning model trained with learning data including numerical data and labels is used to estimate the traffic conditions at each point on the road to be estimated.

[0107] Figure 12 is an explanatory diagram of the method of the third embodiment. TC11 to TC19 are vehicle detectors 2. In the first embodiment, the estimation content was limited to two options: congestion bottleneck (label: 1) and non-congestion bottleneck (label: 2).

[0108] In the third embodiment, the labels for the estimated content are selected from the following four options. • Traffic bottleneck (label: 1) • Traffic congestion extension (Label: 2) • End of traffic jam (label: 3) • No congestion (Label: 4)

[0109] As shown in Figure 12, each of the four choices has a distinctive plotted graph. Therefore, the following process is performed.

[0110] The label addition unit 1141 (Figure 5) adds labels to numerical data for training data, adding labels that indicate one of the following: congestion bottleneck, congestion extension, congestion end, or no congestion.

[0111] Furthermore, the learning model generation unit 115 generates a learning model 124 by machine learning using learning data that includes numerical data and labels (four choices).

[0112] The traffic condition estimation unit 118 then uses the pre-processed data and the learning model 124 to estimate whether each point on the road to be estimated is a traffic bottleneck, traffic congestion extension, traffic congestion end, or no congestion, and outputs the estimation result.

[0113] Thus, according to the third embodiment, it becomes possible to distinguish not only whether or not there is a traffic congestion bottleneck, but also multiple traffic conditions such as the end of a traffic jam, the extension of a traffic jam, and no traffic congestion.

[0114] (Fourth Embodiment) Next, a fourth embodiment will be described. The fourth embodiment is an embodiment in which estimation is performed for a predetermined route with a set starting point and ending point, by repeatedly performing processing by the estimation preprocessing unit and processing by the congestion bottleneck estimation unit while shifting the point to be estimated.

[0115] Figure 13 is an explanatory diagram of the method of the fourth embodiment. In the first embodiment, the location of the congestion bottleneck was estimated for a portion of the route. In the fourth embodiment, we consider estimating the location of the congestion bottleneck for the entire route.

[0116] To achieve this, as shown in Figure 13, sensor data is acquired from each of the multiple vehicle detectors 2 (TC21~TC29) installed along the entire route. Furthermore, since each vehicle detector 2 is assigned identification information, it is possible to identify which vehicle detector 2 each piece of sensor data originates from.

[0117] By acquiring sensor data from the entire route in this way, four possible estimations, including congestion bottlenecks, are made for the installation locations of vehicle sensors 2 throughout the entire route. Specifically, this is as follows:

[0118] For a predetermined route with a set starting and ending point, the system creates and outputs an overall route estimation result by repeatedly processing by the estimation preprocessing unit 116 and the traffic condition estimation unit 118 while shifting the point to be estimated, and organizing the estimation results obtained. The control unit 119 also displays the overall route estimation result on the display unit 14 in an illustrative format (as shown in Figure 13).

[0119] Thus, according to the fourth embodiment, by applying the four-option estimation including the congestion bottleneck of the third embodiment to the entire target route, estimation for the entire route becomes possible. Then, by visually displaying the estimation results for the entire route on the display unit 14, it becomes possible to visually check the congestion bottleneck situation for the entire route at once. As an example of visualization, for example, as shown in Figure 13, it is possible to simultaneously display the speed-traffic volume plot for each part of the target route, what kind of plot it is, and the estimation results of which of the four options each section was estimated to be.

[0120] Furthermore, by storing the estimation results for the entire route in the memory unit 12, it becomes possible to analyze past congestion bottleneck estimations along the route. For example, by extracting and visualizing the locations estimated to be congestion bottlenecks across the entire route in the past, it is possible to confirm and analyze the parts of the route that were estimated to be congestion bottlenecks in the past.

[0121] (Fifth embodiment) Next, a fifth embodiment will be described. In the fifth embodiment, The fifth embodiment is an embodiment that, in addition to the configuration of the fourth embodiment, displays a list of congestion bottleneck data for the entire route based on the route-wide estimation results stored in the memory unit.

[0122] Figure 14 shows an example screen of the fifth embodiment.

[0123] First, for a predetermined route with a set starting and ending point, the estimation results obtained by repeatedly performing processing by the estimation preprocessing unit 116 and processing by the traffic condition estimation unit 118 while shifting the point to be estimated are organized to create an overall route estimation result, which is then stored in the storage unit 12.

[0124] Then, the control unit 119 extracts data on the estimated location of congestion bottlenecks throughout the entire route based on the route-wide estimation results stored in the memory unit 12, creates route-wide congestion bottleneck data, and displays the route-wide congestion bottleneck data in a list (Figure 14).

[0125] For example, as shown in Figure 14, the estimated location of past traffic congestion bottlenecks could be displayed in a table format, including the estimated date, the estimated location of the congestion bottleneck, and a speed-traffic volume plot at that time.

[0126] Thus, according to the fifth embodiment, when estimation results for the entire route are obtained and historical data has been accumulated, the estimated location of the congestion bottleneck is extracted from the accumulated data and visualized as a list. This allows the user to easily check the location of the congestion bottleneck, etc.

[0127] (Sixth Embodiment) Next, a sixth embodiment will be described. The sixth embodiment is one in which the necessary data extraction unit extracts measurement data for a set period when it extracts the necessary data required for machine learning from the measurement data.

[0128] In the sixth embodiment, by changing the time interval of the data used in the necessary data extraction unit 10 according to the purpose in terms of the time axis, it becomes possible to create graphs for multiple purposes and to estimate the location of congestion bottlenecks for multiple purposes.

[0129] For example, the necessary data extraction unit 10 may extract one day's worth of data. In this case, the congestion bottleneck estimation results will be obtained from a graph of the one day's worth of data, making it possible to check the congestion bottleneck situation on a daily basis. For example, it will be possible to analyze that there is no congestion bottleneck on weekdays, but a bottleneck occurs on weekends.

[0130] Furthermore, by extracting and using data from a period of six months to one year, it becomes possible to obtain congestion bottleneck estimation results in six-month to one-year increments. Using these results, it becomes possible to analyze changes in congestion bottleneck conditions in six-month to one-year increments. This makes it possible to analyze how the location of congestion bottlenecks changes in response to changes over time, road traffic conditions, and changes in route shape such as the addition of new routes.

[0131] The program executed by the CPU of the traffic congestion bottleneck estimation device 1 in this embodiment may be configured to be provided as an installable or executable file recorded on a computer-readable recording medium such as a CD-ROM, flexible disk (FD), CD-R, or DVD (Digital Versatile Disk).

[0132] Furthermore, the program may be configured to be stored on a computer connected to a network such as the Internet and provided by being downloaded via the network. Alternatively, the program executed in this embodiment may be configured to be provided or distributed via a network such as the Internet.

[0133] Although embodiments of the present invention have been described above, these embodiments are merely examples and are not intended to limit the scope of the invention. The above embodiments can be implemented in various forms, and various omissions, substitutions, and modifications can be made without departing from the spirit of the invention. The above embodiments are included in the scope and spirit of the invention, as well as in the claims of the invention and its equivalents. [Explanation of symbols]

[0134] 1...Traffic bottleneck estimation device, 2...Vehicle sensor, 11...Processing unit, 12...Storage unit, 13...Input unit, 14...Display unit, 15...Communication unit, 111...Acquisition unit, 112...Storage processing unit, 113...Pre-processing unit for learning, 114...Learning data creation unit, 115...Learning model generation unit, 116...Pre-processing unit for estimation, 117...Model selection unit, 118...Traffic situation estimation unit, 119...Control unit, 120...Traffic flow simulation unit, 121...Road data, 122...Sensor data, 123...Learning data, 124...Learning model, 125...Estimated result, 1131...Required data extraction unit, 1132...Graph diagram creation unit, 1133...Feature highlighting processing unit, 1134...Graph resizing unit, 1135...Graph digitization unit, 1141...Label addition unit, 1142...Learning data storage / extraction unit, S...Traffic control system

Claims

1. An acquisition unit that acquires measurement data from roadside sensors that measure traffic conditions on the road on which the vehicle is traveling, A storage processing unit that stores the aforementioned measurement data in a storage unit, A learning preprocessing unit performs predetermined preprocessing on the measurement data stored in the memory unit to highlight the characteristics of the traffic situation and outputs learning preprocessed data. A training data creation unit creates training data, which is training data for machine learning, using the aforementioned pre-processed data for training, A learning model generation unit generates a learning model, which is a trained model, by machine learning using the aforementioned training data. An estimation preprocessing unit performs the aforementioned preprocessing on the measurement data that is the target of road congestion bottleneck estimation and outputs preprocessed data for estimation, A traffic bottleneck estimation unit estimates the location of the traffic bottleneck using the pre-processed data and the learning model, and outputs the estimation result. A traffic congestion bottleneck estimation device equipped with the following features.

2. The aforementioned measurement data includes information on vehicle traffic volume and speed. The aforementioned pre-processing unit for learning is: A data extraction unit that extracts the necessary data required for machine learning from the aforementioned measurement data, A graph creation unit creates a plot graph with traffic volume and speed as two axes using the aforementioned necessary data. A feature enhancement processing unit performs a process on the aforementioned plot graph to highlight the characteristics of the traffic situation and outputs a feature enhancement plot graph. A graph resizing unit that adjusts the size of the feature highlighting plot graph and outputs a resized graph, The system includes a graph digitization unit that digitizes each part of the resized graph and outputs numerical data, The aforementioned training data creation unit, A label addition unit adds labels for training data to the aforementioned numerical data, A learning data storage and extraction unit stores the aforementioned numerical data and the aforementioned label-containing training data as learning data in the storage unit, and extracts the learning data to be used for learning from the storage unit. The traffic congestion bottleneck estimation device according to claim 1, comprising:

3. The roadside sensor is a vehicle detector installed on the road, The acquisition unit acquires the measurement data from the vehicle detector. The traffic congestion bottleneck estimation device according to claim 1.

4. The aforementioned congestion bottleneck estimation device further includes a traffic flow simulation unit that creates traffic flow simulation data, which is traffic flow simulation data that has not yet been acquired, through traffic flow simulation. The estimation preprocessing unit performs the preprocessing on the measurement data and traffic flow simulation data that are the targets for estimating road congestion bottlenecks, and outputs the estimation preprocessed data. The traffic congestion bottleneck estimation device according to claim 1.

5. The label addition unit adds labels to the numerical data for training data, adding labels that indicate either a congestion bottleneck, congestion extension, congestion end, or no congestion. The learning model generation unit generates the learning model by machine learning using the learning data which includes the numerical data and the labels. The aforementioned congestion bottleneck estimation unit uses the pre-processed data and the learning model to estimate whether each point on the road to be estimated is a congestion bottleneck, congestion extension, congestion end, or no congestion, and outputs the estimation result. The traffic congestion bottleneck estimation device according to claim 2.

6. The aforementioned necessary data extraction unit extracts the necessary data required for machine learning from the measurement data, and extracts the measurement data for a set period. The traffic congestion bottleneck estimation device according to claim 2.

7. The aforementioned traffic congestion bottleneck estimation device creates and outputs an overall route estimation result by organizing the estimation results obtained by repeatedly performing the processing by the estimation preprocessing unit and the processing by the traffic congestion bottleneck estimation unit for a predetermined route with set start and end points, while shifting the points to be estimated. The traffic congestion bottleneck estimation device according to claim 1.

8. The aforementioned traffic congestion bottleneck estimation device further comprises a display control unit that displays the estimation results for the entire route graphically. The traffic congestion bottleneck estimation device according to claim 7.

9. The aforementioned traffic congestion bottleneck estimation device, for a predetermined route with set starting and ending points, creates an overall route estimation result by organizing the estimation results obtained by repeatedly performing the processing by the estimation preprocessing unit and the processing by the traffic congestion bottleneck estimation unit while shifting the points to be estimated for that predetermined route, and stores it in the storage unit. The aforementioned traffic congestion bottleneck estimation device is The system further comprises a display control unit that extracts data on the estimated location of congestion bottlenecks throughout the entire route based on the route-wide estimation results stored in the memory unit, creates route-wide congestion bottleneck data, and displays the route-wide congestion bottleneck data in a list. The traffic congestion bottleneck estimation device according to claim 1.

10. The acquisition unit performs an acquisition step of acquiring measurement data from roadside sensors that measure traffic conditions on the road on which the vehicle is traveling, The storage processing unit performs a storage processing step of storing the measurement data in the storage unit, A learning preprocessing step in which the learning preprocessing unit performs predetermined preprocessing on the measurement data stored in the memory unit to highlight the characteristics of the traffic situation and outputs learning preprocessed data, The training data creation unit creates training data, which is training data for machine learning, using the pre-processed training data, in the training data creation step. The learning model generation unit performs a learning model generation step in which it generates a learning model, which is a trained model, by machine learning using the training data, The estimation preprocessing unit performs the preprocessing on the measurement data that is the target of the estimation of road congestion bottlenecks and outputs the estimation preprocessed data, The traffic congestion bottleneck estimation unit estimates the location of the traffic congestion bottleneck using the pre-processed data and the learning model, and outputs the estimation result in a traffic congestion bottleneck estimation step. A method for estimating congestion bottlenecks, including the following:

Images