Traffic flow prediction device
By analyzing the characteristics of moving vehicle data and dynamically adjusting the data resolution, the problem of balancing prediction accuracy and processing load in traffic flow forecasting is solved, thus achieving efficient traffic flow forecasting.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- TOYOTA JIDOSHA KK
- Filing Date
- 2025-12-17
- Publication Date
- 2026-07-24
AI Technical Summary
Existing technologies have failed to effectively balance forecast accuracy and processing load in traffic flow prediction, resulting in problems of reduced forecast accuracy and excessive processing load.
By acquiring past movement data, analyzing data characteristics, dynamically adjusting data resolution, and adaptively processing data at the edge or center, the amount of newly acquired data is reduced, achieving low-resolution data to alleviate communication and computational load while maintaining prediction accuracy.
It effectively suppressed the decrease in prediction accuracy, reduced the processing load, and achieved efficient traffic flow prediction.
Smart Images
Figure CN122454745A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to a traffic flow prediction device. Background Technology
[0002] In recent years, technologies for analyzing the flow of moving bodies have been developed. For example, Japanese Patent Application Publication No. 2023-26294 discloses a pedestrian flow analysis system that suppresses the excessive processing load for analyzing pedestrian flow data while simultaneously analyzing the data. Summary of the Invention
[0003] However, the technology described in Japanese Patent Application Publication No. 2023-26294 merely involves a cleaning process based on a unified GPS benchmark (sparse processing is only applied to location information representing the same location). This process does not reflect the trade-offs related to the accuracy of estimations derived from data characteristics inferred from past data.
[0004] In view of this situation, the object of the present invention is to suppress the decrease in prediction accuracy and reduce the processing load by inferring data characteristics from past data.
[0005] An embodiment of the present invention relates to a traffic flow prediction device that performs the following processing: acquiring data of a moving vehicle at a past date and time and at the current time; determining the resolution of newly transmitted data from the moving vehicle using the analysis results of the data or the traffic flow prediction results based on the data; and using the newly transmitted data from the moving vehicle at the resolution to predict future traffic flow.
[0006] According to the present invention, it is possible to suppress the decrease in prediction accuracy and reduce the processing load by inferring data characteristics from past data. Attached Figure Description
[0007] Hereinafter, with reference to the accompanying drawings, the features, advantages, and technical and industrial significance of exemplary embodiments of the present invention will be described, in which the same reference numerals denote the same elements, and wherein:
[0008] Figure 1 This is a diagram illustrating an example of the configuration of a traffic flow prediction system according to one implementation method.
[0009] Figure 2 This is a flowchart illustrating the processing steps of a traffic flow prediction device according to one embodiment.
[0010] Figure 3 This is a diagram illustrating the differences in data resolution. Detailed Implementation
[0011] Figure 1The traffic flow prediction system 1 shown in one embodiment includes one or more mobile bodies (vehicles, people, etc.) 20 and a traffic flow prediction device (data control device) 10. The traffic flow prediction device 10 may be a database server. The traffic flow prediction device 10 is located centrally. The vehicles 20 are communicatively connected to the traffic flow prediction device 10 via a network.
[0012] The mobile unit 20 is equipped with one or more sensors and a communication unit (communication interface) for communicating with the traffic flow prediction device 10. Each mobile unit 20 transmits driving data (hereinafter referred to as "data") detected by the sensors to the traffic flow prediction device 10. The "data" includes, for example, vehicle identification number, GPS location information, time, vehicle speed, acceleration, driving operation (accelerator / brake, etc.), driver / passenger attribute information, etc.
[0013] Traffic flow prediction system 1 acquires past date and time data and current time (real-time) data from mobile bodies 20 in the road traffic network as a mobile data log. Then, it feeds back the prediction results (execution results of the prediction task) obtained from the mobile data log, including the spatiotemporal characteristics of the road traffic network and traffic flow (traffic volume, throughput speed, bias of mobile body 20 attributes, etc.). Traffic flow prediction system 1 considers the balance between the amount of data and prediction accuracy used for traffic flow prediction tasks, adaptively (dynamically) reducing the data resolution. Therefore, traffic flow prediction system 1 can suppress the decrease in prediction accuracy and reduce the amount of new data acquired from mobile bodies 20, thereby reducing communication and computational loads. Traffic flow prediction system 1 can adaptively determine the resolution (sampling period of the acquired data, group of mobile bodies, etc.) based on the target road segment, time period, and other contexts. Furthermore, the low-resolution processing can be performed either distributed processing at the edge (mobile body 20 side) or centralized processing at the center (traffic flow prediction device 10 side).
[0014] refer to Figure 2 The processing steps of the traffic flow prediction device 10 are explained in the following example.
[0015] In S1, the traffic flow prediction device 10 determines the resolution based on past data for balancing. The traffic flow prediction device 10 dynamically determines the resolution specifications of newly transmitted data from the mobile vehicle 20 using analysis results based on past data from the mobile vehicle 20 or traffic flow prediction results. For example, the traffic flow prediction device 10 can analyze past data from the mobile vehicle 20 on the road by each road segment and each time period, and determine the reduced resolution based on the spatiotemporal road links and regularity of traffic flow stability. Furthermore, the traffic flow prediction device 10 can determine the reduced resolution based on traffic flow prediction results, within a range that maintains the desired prediction accuracy obtained empirically based on actual prediction accuracy.
[0016] In S2, the traffic flow prediction device 10 determines whether to perform balancing processing on the edge side.
[0017] In S3, if the traffic flow prediction device 10 performs balancing processing on the edge side ("Yes" in S2), it indicates to the mobile body 20 the resolution of the newly transmitted data from each mobile body 20 (the resolution determined in S1 or S9) (edge side balancing processing).
[0018] In S4, the traffic flow prediction device 10 receives data from the moving body 20 in real time at the resolution determined in S3.
[0019] In S5, if the traffic flow prediction device 10 does not perform balancing processing on the edge side ("No" in S2), it receives data at the default resolution from the moving body 20 in real time.
[0020] In S6, the traffic flow prediction device 10 changes the resolution of the data received in S5 according to the resolution specifications determined in S1, thereby reducing the amount of data (center-side balancing processing). If the edge-side balancing processing in S3 is not performed, the traffic flow prediction device 10 performs the center-side balancing processing in S6.
[0021] In S7, the traffic flow prediction device 10 takes as input the data transmitted by each moving body 20 at a specified resolution, which has been balanced in S3 or S6, and predicts the future traffic flow. In the first embodiment described later, an expanded estimate of the overall traffic volume is performed; in the second embodiment, traffic statistics are predicted.
[0022] In S8, the traffic flow prediction device 10 determines whether to perform repeated processing. If the confidence level exceeds the target value or other conditions for repeated processing are not met (in S8, this is "No"), then the processing ends.
[0023] In S9, if repeated processing is performed ("Yes" in S8), the resolution specification of the balanced processing is updated based on the confidence level of the traffic flow predicted in S7, and the process is returned to S2.
[0024] Implementation Method 1 - Expanded Estimation of Overall Traffic Volume -
[0025] The expanded estimation of total traffic volume is a process of estimating the overall traffic volume of a road network based solely on data obtained from a subset of connected vehicle groups (e.g., specific vehicle type groups, vehicle groups from specific manufacturers, etc.). It is known that the traffic volume of this vehicle group, obtained from location information, time information, etc., contained in the data, is one of the important explanatory variables (e.g., see the reference below). [Reference] Yong, J., Wakabayashi, Y., Okayasu, A., Miki, R., Sasai, T., Inoue, M., & Fukushima, S. (2022). Estimating Total Traffic Volume with Statistical Modeling Approach. In 2022 IEEE 25th International Conference on Intelligent Transportation Systems (ITSC), pp. 304-309, IEEE.
[0026] In S1, the traffic flow prediction device 10 determines a combination of conditions and granularity from past data based on a pre-selected benchmark. That is, the traffic flow prediction device 10 reduces the resolution of the data by coarsening the granularity as much as possible for each condition, thereby making the data easier to process quantitatively.
[0027] Figure 3 This is a schematic diagram illustrating the processing of S1. In links with unstable traffic flow, a resolution close to that of the original data is used; in links with stable traffic flow, the resolution is coarsened from the original data (to increase the reduction rate). The resolution specification is composed of a combination of conditions and granularity, and is determined in a manner consistent with benchmarks. The following is a detailed explanation of the settings for benchmarks, conditions, and granularity.
[0028] The "benchmark" related to resolution and granularity determination is based on the characteristics of past data. Below are examples of benchmarks used to reduce data volume.
[0029] (1) Utilizing the spatiotemporal regularity related to the stability of past data
[0030] For example, traffic volume (locations / time periods with low traffic dispersion, locations with strong periodicity of time periods / weeks, etc.), the tendency of sudden events (locations and times where sudden events such as accidents, congestion, and traffic control are less likely to occur), and the bias of the attributes of the mobile body 20 (locations with high prevalence of the vehicle group of the target) are determined as benchmarks.
[0031] (2) The performance of forecast accuracy based on the expansion of past overall traffic volume estimates
[0032] For example, based on actual performance, even with limited data, one can empirically determine the location / time with the highest predictive accuracy for the expanded estimate as the benchmark.
[0033] The "conditions" for setting granularity can be broadly categorized into two types: time and space.
[0034] (i) Time situation
[0035] For example, granularity can be determined by time period / week / season.
[0036] (ii) Spatial conditions
[0037] For example, granularity can be determined by region / road unit.
[0038] The granularity of newly acquired data can be set in two main ways, roughly categorized as time granularity and sample granularity, depending on the specific situation.
[0039] (a) Time granularity (sampling period)
[0040] For example, the sampling period can be changed for each condition.
[0041] (b) Specimen grain size
[0042] For example, the proportion of the vehicle group that is the object of sampling among the vehicles from which data can be obtained can be changed according to each situation. For example, the number of data items (vehicle speed, fuel consumption, etc.) that are the objects of sampling can be changed according to each situation.
[0043] Traffic flow prediction device 10, for example, combines (2)-(i)-(a) above to simulate a low-resolution sampling period for all vehicles acquired over the past 24 hours on highways in Tokyo. Traffic flow prediction device 10 calculates a threshold for the sampling period at which the average error of the expanded estimate is less than a predetermined value (e.g., 10%) for each time point using simulation, using a predetermined time unit (e.g., 1 hour), and determines this threshold as the resolution of newly transmitted data.
[0044] Furthermore, the traffic flow prediction device 10, for example, combines (1)-(ii)-(b) above, and determines the sparsity of the vehicle group for which the data is to be acquired based on the vehicle prevalence (coverage) of each area. When only 25% of all passing vehicles need to be acquired, on roads in areas where the prevalence of the vehicles to be acquired is 50%, indication data is sent to 50% of those vehicles. On roads in areas where the prevalence of the vehicles to be acquired is 25%, indication data is sent to 100% of those vehicles.
[0045] In step S3, the traffic flow prediction device 10 sends the resolution specifications determined in step S1 to the target vehicles on the road traffic network. In subsequent data transmissions, the target vehicles transmit data at the specified resolution, thus reducing the data resolution and lowering the communication load.
[0046] In steps S4 and S5, the traffic flow prediction device 10 acquires data in real time from the target vehicles. In S4, the traffic flow prediction device 10 acquires data at a situation-specific resolution, with a period of 300 milliseconds for road A on weekdays, 400 milliseconds for road A on weekends, a period of 200 milliseconds for road B on weekdays, and a period of 500 milliseconds for road B on weekends. In S5, the traffic flow prediction device 10 acquires data using a default setting, specifying which road and time also acquire data at a 500 millisecond period.
[0047] In S6, the traffic flow prediction device 10 performs low-resolution processing on the data received in S5 according to the resolution specifications determined in S1.
[0048] In S7, the traffic flow prediction device 10 uses the balanced data to perform an expanded estimate of the overall traffic volume. For example, the traffic flow prediction device 10 learns the parameters of a regression model (linear regression model, mixed effects model, etc.) based on past data, then inputs newly acquired real-time data and outputs the prediction results. For example, the target variable is set as the overall traffic volume per hour for each road, and the explanatory variables are set as the traffic volume of the data acquisition target vehicles per hour, average vehicle speed, road specifications, number of lanes, and time period / week category for each road. The feature quantities can also be quantities obtained directly from the data (traffic volume, time period, week, presence or absence of holidays, etc. for each road in the vehicle group). Furthermore, the feature quantities can also be quantities obtained by combining information with external databases (road characteristics of the road where the vehicle is located (road specifications, number of lanes, etc.), weather conditions, etc.).
[0049] In S8, the traffic flow prediction device 10 checks the fit of the traffic engineering model (QV curve, etc.) pre-calculated from past data to the balanced data, and calculates the confidence level of the data according to the resolution. The traffic flow prediction device 10 sets the duplication determination to "yes" only if the confidence level of the acquired data is lower than a predetermined threshold and the acquired data can be upgraded to a higher resolution (the current resolution specification setting is less than the system's upper limit).
[0050] In S9, the traffic flow prediction device 10 updates the granularity of the resolution status that needs further data acquisition based on the confidence level of the repeated determination, using the same method as in S1.
[0051] Implementation Method 2 - Prediction of Traffic Statistics Information -
[0052] For traffic statistics (statistics related to the smoothness of traffic flow) of each road in the road network, the data of the moving body 20, which is used for prediction tasks related to statistical quantities such as average speed and average transit time (time required), is balanced. Hereinafter, the parts that differ from the first embodiment will be indicated in brackets <>.
[0053] In S1, the resolution specification used for balancing processing is determined from past data based on a pre-selected <determined> benchmark, which determines a combination of conditions and granularity. For example, the traffic flow prediction device 10 simulates a low-resolution sampling period using data of all vehicles acquired over the past 24 hours on highways within Tokyo. The traffic flow prediction device 10 calculates a threshold for the sampling period at which the average error of the <average speed estimate or average transit time> is less than a predetermined value (e.g., 10%) for each time point, using a predetermined time unit (e.g., 1 hour), and determines this threshold as the resolution for newly transmitted data.
[0054] In S7, the traffic flow prediction device 10 uses the balanced data to perform <estimate traffic statistics>. For example, the traffic flow prediction device 10 learns the parameters of a regression model based on past data, then inputs newly acquired real-time data and outputs prediction results. For example, the target variable is set as the <average vehicle speed> per hour for each road, and the explanatory variables are the <average vehicle speed>, traffic volume>, road specifications, number of lanes, and time period / week category for each road per hour. The features can be quantities obtained directly from the data (<average vehicle speed>, time period, week, presence or absence of holidays, etc. for each road in the vehicle group), or quantities obtained by combining information with external databases.
[0055] Furthermore, in the above embodiment, when the moving body 20 is, in particular, a "vehicle," in S1, the traffic flow prediction device 10 can determine the resolution by considering the characteristics of vehicle data, including driving characteristics and vehicle characteristics, and the characteristics of road traffic, including road characteristics and traffic characteristics. "Driving characteristics" are characteristics estimated from past data using driving operation biases (driving skill level, number of detours / rests, etc.) as a driver model. "Vehicle characteristics" are the vehicle's driving performance estimated from past data such as vehicle model, total driving distance, and fuel consumption. "Road characteristics" are information about road segments related to traffic flow smoothness (number of lanes, width, curvature, etc.). "Traffic characteristics" are traffic control information related to road traffic laws (speed limits, control signs, etc.), traffic information (congestion / accident information provided by operators, etc.). Furthermore, the traffic flow prediction device 10 can use the fit obtained by utilizing the above characteristics to determine whether the repeated processing in S8 is feasible.
[0056] Since the data from the mobile vehicle 20 is continuous in time and space, prioritizing sampling the data resolution with small time intervals or large samples would result in a massive data volume, making the communication load and computing load for task processing impractical. To address this, the traffic flow prediction device 10 uses the analysis results of the data acquired from the mobile vehicle 20 or the traffic flow prediction results to determine the resolution of newly transmitted data from the mobile vehicle 20, and uses the newly transmitted data at that resolution to predict future traffic flows. Therefore, the traffic flow prediction device 10 can suppress the decrease in prediction accuracy and reduce the processing load.
[0057] For the traffic flow prediction device 10 described above to function, a computer capable of executing program commands can also be used. The program can be recorded on a computer-readable, non-transitory recording medium.
[0058] The above embodiments have been described as representative examples, but changes and substitutions can be made within the spirit and scope of the present invention, as will be clear to those skilled in the art. For example, multiple steps described in the flowchart of the embodiments can be integrated into one, or one step can be divided.
Claims
1. A traffic flow prediction device, characterized in that, Perform the following processing: Retrieve data on past dates and times, as well as the current moving object's data; The resolution of newly transmitted data from the mobile vehicle is determined using the analysis results of the data or the traffic flow prediction results based on the data. and Using the data newly transmitted from the moving body at the stated resolution, future traffic flows are predicted.
2. The traffic flow prediction device according to claim 1, characterized in that, The traffic flow forecast is an expanded estimate of the overall traffic volume. The acquired data is simulated at low resolution, and a threshold is calculated for the average error of the traffic flow prediction result to be less than a specified value within a specified time unit. This threshold is then determined as the resolution of the newly transmitted data.
3. The traffic flow prediction device according to claim 1, characterized in that, The traffic flow prediction is a prediction based on traffic statistics. The acquired data is simulated and reduced in resolution. A threshold is calculated in a specified time unit for the average error of the average velocity estimate or average transit time to be less than a specified value. This threshold is then determined as the resolution of the newly transmitted data.