A Real-Time Microscopic Traffic Simulation Method and System for Highway Networks Based on SUMO
By using SUMO-based data fusion and dynamic model updating methods, the problems of imprecise characterization of vehicle behavior and lack of traffic flow patterns in real-time microscopic traffic simulation of highway networks are solved, achieving high-precision traffic simulation and supporting proactive control and travel services.
Patent Information
- Application Number
- CN202310913874.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-07-25
- Publication Date
- 2026-11-17
- Estimated Expiration
- 2043-07-25
AI Technical Summary
Existing real-time microscopic traffic simulation technologies for highway networks suffer from problems such as imprecise characterization of vehicle behavior, lack of analysis of traffic flow patterns, poor quality of observation data, and low model timeliness, resulting in low simulation accuracy and an inability to accurately reflect real-world traffic conditions.
The SUMO-based approach is adopted to fuse historical and real-time observation data of road network vehicles from multiple sources, analyze traffic flow patterns, establish vehicle behavior models, dynamically update model parameters, construct simulated road networks for simulation and extrapolation, thereby improving data quality and model effectiveness.
It improves the accuracy of microscopic simulation, enabling detailed characterization of vehicle behavior and accurate simulation of traffic flow patterns, thus enhancing the reliability and real-time performance of simulation results and supporting proactive management and travel services.
Smart Images

Figure CN116738761B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of intelligent transportation information technology, and relates to a high-precision real-time microscopic traffic simulation method and system for highway networks based on SUMO. Background Technology
[0002] With the continuous increase in the number of motor vehicles and the growing demand for transportation in my country, the traffic conditions on the expressway network are becoming increasingly complex. To achieve real-time traffic monitoring, expressway management has deployed a large number of roadside units (RSUs), which have significantly enhanced the field perception capabilities of highway operation and management. However, due to the rapid development of my country's highways, the current observation range of RSU equipment is far from covering the entire road network. To achieve traffic observation in blind spots, real-time microscopic traffic simulation is usually required to simulate traffic operations within those blind spots. Real-time microscopic traffic simulation is a crucial technology for real-time perception of traffic conditions in the field of intelligent transportation. Developing a precise simulation system and algorithm based on vehicle characteristics and traffic flow patterns has a profound impact on situation assessment, proactive management, and travel services: 1) In situation assessment, accurate real-time microscopic traffic simulation can help managers efficiently analyze traffic conditions in blind spots and provide traffic flow indicators at any time and location, including flow density, speed, congestion, and queuing, thus using simulation technology to compensate for blind spots; 2) In proactive management, based on comprehensive situation assessment, managers can further leverage decision-making models or expert experience to issue control commands for dynamic traffic intervention, thereby improving overall traffic efficiency and safety; 3) In travel services, the real-time location of each vehicle within the road network can be obtained based on the real-time microscopic simulation system, combined with edge devices to provide accompanying traffic information services, such as push notifications of road conditions ahead, accidents, and maintenance events.
[0003] However, in practical engineering applications, current microscopic traffic simulation faces several problems and challenges:
[0004] 1. Lack of detailed description of the behavioral characteristics of moving vehicles.
[0005] Microscopic traffic simulation should be able to simulate the traffic behavior of each physical vehicle. This relies on accurately defining the behavioral characteristics of each vehicle involved in traffic, including lane selection preferences, vehicle path preferences, and speed preferences. However, current microscopic simulations are based on complete randomness or some kind of distribution to characterize behavioral characteristics, which cannot accurately reflect the true preferences of each vehicle, resulting in low simulation accuracy.
[0006] 2. Lack of analysis of traffic flow patterns
[0007] Considering the difficulty in characterizing all vehicles (especially when a vehicle first appears) in engineering practice, traffic flow patterns can be considered when single-vehicle-level features are lacking. Traffic flow patterns refer to the patterns of overall traffic flow, including lane allocation preferences and travel path allocation preferences.
[0008] However, current microscopic simulations do not fully integrate traffic flow pattern analysis with traffic simulation, but instead use a more random approach to characterize traffic behavior.
[0009] 3. The quality of the observation data is poor.
[0010] There are various types of RSU devices, and the data sources are diverse, including image recognition and radio frequency communication. Currently, microscopic traffic simulations often select one data source as the input to the simulation system. However, based on the current level of data acquisition, all types of data have certain problems. For example, image recognition is susceptible to interference from extreme weather, license plate obstruction, and cloned license plates, while radio frequency communication is susceptible to interference from the opposite lane. In addition, the observation data also suffers from common problems such as duplication, delay, and inaccurate clocking.
[0011] 4. Low model timeliness
[0012] In addition to the precise and detailed model construction mentioned in points 1 and 2 above, the level of microscopic traffic simulation is also affected by the time effectiveness of the model, i.e., whether the vehicle and traffic flow model parameters are updated in a timely manner. However, the current model parameters for microscopic simulation are set with initial values before the system runs. Due to the large amount of computation or imperfect engineering, it is impossible to update the relevant parameters periodically or even in near real-time as real-time data is received.
[0013] In past business applications, to address the above issues, a simple data comparison and processing was typically performed before selecting a single data source for the simulation system. Vehicle path selection and lane selection were often randomly assigned; vehicle speed was typically controlled based on a general model and random parameters. Given this situation, current real-time traffic observation systems based on microscopic simulation technology often suffer from low accuracy and fail to accurately reflect real-world traffic conditions. Summary of the Invention
[0014] This invention aims to overcome the shortcomings of existing real-time microscopic simulation technology for highway networks and provides a real-time microscopic traffic simulation method and system for highway networks based on SUMO, so as to improve the accuracy of existing real-time microscopic simulation from both the model and data aspects.
[0015] To achieve the above objectives, the present invention adopts the following technical solution:
[0016] A real-time microscopic traffic simulation method for highway networks based on SUMO includes:
[0017] The system acquires historical and real-time observation data of road network vehicles from multiple sources and fuses them separately to obtain historical fused data and real-time fused data. A set of observation data includes vehicle identity, location, speed, lane, vehicle type, and time information.
[0018] Based on historical fusion data analysis, traffic flow patterns are analyzed, including lane distribution patterns, path selection patterns, and maximum expected traffic flow speed. Vehicle behavior models are also established, including lane selection probability models, path selection probability models, and maximum expected traffic flow speed models.
[0019] Acquire real-time observation data of vehicles on the road network and perform data fusion, and use the real-time fused data to dynamically update historical fused data and vehicle behavior models;
[0020] Construct a simulated road network and map the observation facilities to the computable nodes in the simulated road network;
[0021] Simulation results are obtained by using real-time fused data and the vehicle behavior model.
[0022] As a preferred implementation, the fused data is transformed into passage order pairs for subsequent processing. A passage order pair refers to any pair of adjacent observation records after sorting the observation records of any vehicle in ascending order of time.
[0023] As a preferred implementation, the traffic flow pattern analysis and vehicle behavior model establishment includes: dividing time into date time ranges in days and hour time ranges in hours;
[0024] For each observation data point, it is divided into different time ranges according to the observation time;
[0025] Using the aforementioned date and time range and hourly time range as time units, traffic flow patterns are analyzed and vehicle behavior models are constructed by vehicle type.
[0026] As a preferred embodiment, the lane distribution pattern is determined based on the lanes selected by traffic flow of a certain type of vehicle under a certain observation facility within a certain time range;
[0027] Lane selection probability is determined based on the lane selected by a vehicle under a certain observation facility within a certain time range.
[0028] As a preferred implementation, the path selection rule is the downstream selection probability of traffic flow of a certain type of vehicle at a certain observation facility, which is determined based on the downstream observation facility selected by traffic flow of a certain type of vehicle at a certain observation facility within a certain time range;
[0029] The path selection probability is determined based on the downstream observation facility selected by a vehicle under a certain observation facility within a certain time range.
[0030] As a preferred embodiment, the maximum expected traffic flow speed is determined as follows:
[0031] If there is no traffic flow of a certain type of vehicle flow under a certain observation facility during a certain period, then the maximum expected speed of that type of vehicle flow is the design speed; otherwise, calculate the type of vehicle flow and determine whether it is in a free flow state.
[0032] When traffic flow is in a completely free-flow state, the time period is divided into several sub-time periods. The average traffic speed of this type of traffic flow in each sub-time period under the observation facility is calculated. The maximum expected traffic speed of this type of traffic flow in the time period is the maximum value among multiple average traffic speeds; otherwise, the maximum expected traffic speed is the design speed.
[0033] The method for determining the maximum expected speed is as follows:
[0034] If there is no observation data for a certain type of vehicle under a certain observation facility during a certain period, then the maximum expected speed of that type of vehicle is the design speed; otherwise, calculate the traffic flow of that type of vehicle and determine whether it is in a free flow state.
[0035] When the traffic flow is in a completely free-flow state, the maximum expected speed is the maximum observed speed of the vehicle under the observation facility during that time period.
[0036] As a preferred implementation, the updated historical fusion data and vehicle behavior model include:
[0037] For any received real-time fused data, it is appended to the historical fused data and the vehicle identity and passage order are updated; and at fixed intervals, the traffic flow pattern and vehicle behavior model are updated using the updated historical fused data.
[0038] As a preferred embodiment, the method of mapping the observation facility includes:
[0039] 1) Extract all node data from the simulated road network, including node number and coordinate information;
[0040] 2) For each observation facility, calculate its distance to each node and obtain the node closest to the observation facility;
[0041] 3) When the distance between the observation facility and the nearest node is less than a preset threshold, the observation facility is mapped to the node; otherwise, the edge closest to the facility is found, and the observation facility is interrupted by its vertical projection point relative to the edge. A new node is then inserted at the vertical projection point.
[0042] 4) Repeat steps 1)-3) until all mapping relationships meet the requirement that the distance is less than the preset threshold.
[0043] As a preferred embodiment, the simulation includes:
[0044] For any real-time fused data, extract the record information it contains;
[0045] Based on this recorded information, the vehicle behavior model is used to obtain its downstream path selection, downstream arrival lane selection, and maximum expected speed; according to the mapping established when constructing the simulated road network, the traffic flow starting point and downstream path observation facilities are converted into computable nodes;
[0046] Simulations are performed based on the recorded information, starting lane, downstream arrival lane, starting node, ending node, and maximum expected speed.
[0047] Another objective of this invention is to provide a real-time microscopic traffic simulation system for highway networks based on SUMO, comprising:
[0048] The data fusion unit integrates historical and real-time observation data of road network vehicles from multiple sources; a set of observation data includes vehicle identity, location, speed, lane, vehicle type, and time information;
[0049] The traffic flow pattern analysis unit analyzes traffic flow patterns based on historical fusion data, including lane distribution patterns, route selection patterns, and maximum expected traffic flow speed, and dynamically updates traffic flow patterns based on real-time fusion data.
[0050] The vehicle behavior model building unit establishes a vehicle behavior model based on the traffic flow pattern analysis results, including a lane selection probability model, a path selection probability model, and a maximum expected traffic flow speed model, and dynamically updates the model based on real-time fused data.
[0051] The simulation environment construction unit builds a simulation road network and maps observation facilities to computable nodes in the simulation road network;
[0052] The simulation unit uses real-time fused data and vehicle behavior models to perform simulations and obtain simulation results.
[0053] This invention fuses historical observation data from multiple sources to obtain a high-quality fused dataset. Traffic flow features are constructed using a traffic flow pattern analysis algorithm, and vehicle features are constructed using a vehicle model construction algorithm. These features are then input as parameters into a simulation system. The system receives real-time traffic observation data from multiple sources and uses a real-time fusion algorithm to improve the quality of the observation data. The fused data is then input into the simulation system. The simulation system receives the observation data, performs simulations based on the simulated road network environment and vehicle and traffic flow features, and outputs simulation results in real-time using virtual observation facilities. Furthermore, the real-time fused data enables dynamic updates to the traffic flow and vehicle behavior models, improving the model's time-sensitive effectiveness.
[0054] The technical solution of the present invention has the following beneficial effects:
[0055] 1) Integrate historical vehicle traffic observation data from multiple sources to improve data quality and form a fused dataset as input for subsequent modeling;
[0056] 2) Based on the fused vehicle traffic observation data, vehicle behavior characteristics are characterized to achieve customized parameter simulation and improve the accuracy of microscopic simulation;
[0057] 3) Traffic flow pattern analysis based on vehicle traffic observation data. When there is a lack of direct vehicle behavior characteristics, the overall traffic flow pattern is used to characterize the behavior of individual vehicles, which greatly improves the reliability of vehicle behavior compared with the purely random model.
[0058] 4) Fusion of real-time data from multiple sources improves the quality of input data for the real-time microscopic simulation system;
[0059] 5) Dynamically update historical observation data and iterate the traffic flow and vehicle behavior models during system operation to improve model effectiveness;
[0060] 6) By setting up virtual observation facilities, the simulation system can output rich simulation result data in real time, including real-time vehicle location, real-time traffic flow indicators, etc., thereby supporting upper-level applications such as active control. Attached Figure Description
[0061] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation
[0062] The method of this invention comprises eight steps: A1. Preparation of historical observation data, A2. Traffic flow pattern mining, A3. Vehicle behavior model construction, A4. Real-time data access and fusion, A5. Dynamic model updating, A6. Simulation environment construction, A7. Simulation deduction, and A8. Real-time result output. The relationships between these steps are shown in [the diagram / document / reference needed]. Figure 1 .
[0063] The observation data generated during vehicle movement mainly includes identity (p), position (s), speed (v), lane (l), vehicle type (c), and time (t) information, which will be referred to below as ordered six-tuples. This refers to vehicle traffic data, used to pinpoint the spatiotemporal location of vehicles during their journey. In the transportation field, vehicle identification is typically represented by license plate number, location by spatial positioning (usually the facility number to which the equipment is attached), vehicle type is generally categorized by toll standards (e.g., passenger class 1, passenger class 2, freight class 1, others, etc.), totaling 17 types, and time is the time the equipment observed the vehicle. Furthermore, assuming there are M observation data sources, the data source set is defined as... .
[0064] Let there be a total of n observation facilities, and let the set of observation facilities be denoted as . In the field of highways, there are three types of observation facilities: entrance, exit, and mainline observation facilities. Let G be the set of entrance observation facilities. en The set of exit observation facilities is G. ex The main observation facility set is G. mid Let any gantry be used. The largest lane number is Then the range of lanes that the vehicle can choose is Where 0 represents the zeroth lane (emergency lane), 1 represents the first lane, and so on. Furthermore, let the set of vehicle types be denoted as C.
[0065] A1. Preparation of historical observation data
[0066] The purpose of this step is to integrate historical observation data from multiple sources to form a more accurate fusion dataset for vehicle traffic observation, which will serve as input for subsequent vehicle model construction and traffic flow pattern mining.
[0067] The prerequisite for this step is the topological structure and distance between facilities. In a highway traffic scenario, vehicles are not allowed to travel in the wrong direction. Therefore, this invention requires that the facility topology be a directed weighted graph, with the weights being the adjacency distances between facilities, specifically defined mathematically as follows:
[0068] G is a set of nodes, where nodes are facilities.
[0069] E is the set of edges, which are the connections between facilities.
[0070] It is an association function that maps each edge to an ordered pair of vertices (i.e., an edge is sequentially associated with two vertices).
[0071] Each edge in set E has a weight, which is the spatial distance along the path between the two nodes connected by the edge.
[0072] In addition, the term length refers to the number of edges between two nodes, and distance refers to the shortest path length between two nodes.
[0073] A1.1 Preparation of Discrete Observation Data
[0074] The purpose of this step is to obtain a precise six-tuple fused dataset. Using the data fusion method disclosed in the applicant's earlier application ZL202211116113.9, the fused dataset can be obtained by inputting M historical observation data sources and the inter-facility topology.
[0075] Let the ordered six-tuple after fusion be denoted as The fused observation dataset contains m records, denoted as m. Let q be the total number of vehicles involved, and let the set of vehicles be denoted as . , .
[0076] A1.2 Vehicle passage sequence preparation
[0077] A vehicle passage order pair refers to the passage order pair obtained by sorting the observation records of any vehicle in ascending order of time. For example, if the time series data of vehicle p is (a,b,c,d), then (a,b), (b,c), and (c,d) are the passage order pairs for that vehicle. The first observation record in each passage order pair is denoted as the source, and the second observation record as the target. This data serves as the input for subsequent path selection calculations. The output of step A1.1 will now be converted into passage order pairs.
[0078] For any vehicle p∈P, the process for extracting its passage order is as follows.
[0079] S1. Record Extraction. Extract all observation records related to the license plate number p as a set. ,remember Here, |A| represents the number of elements in set A.
[0080] S2. Sort by time. (This refers to the set.) All observation records are arranged in ascending order of time (t'), forming ordered tuples. ,in It is The kth value arranged in ascending order of time.
[0081] S3. Splitting. The ordered tuples in S2 are split into several pairs of sequential pairs: .
[0082] S4. Segmentation. Considering the highway scenario, the entrance should be the first observation of a single passage, and the exit should be the last observation of a single passage. Therefore, the ordered pairs in S3 are segmented. Here, a filtering method is used to complete the logical segmentation, that is, to remove the parts that should be segmented but are retained in the S3 results. The filtering rule is: entrance observation records (i.e., ) must not be used as the target of the passage order pair, and the exit observation record (i.e. (This cannot be used as a pass order pair) Based on this rule, the passing order pair of vehicle p is obtained after filtering.
[0083] S5. Filtering. To improve the usability of the data in this step, it is necessary to further remove travel order pairs that do not conform to the topology. The removal method is as follows: extract the start and end observation facilities of any travel order pair in the S4 result, form an edge, denoted as e'. If e'∈E, then the travel order pair is retained; otherwise, the travel order pair is removed. Traverse all travel order pairs in S4 and perform the aforementioned judgment to obtain the filtered travel order pairs for vehicle p, which do not contain erroneous data.
[0084] Initialize an empty set ST (for ease of understanding, ST stands for source-target). For all vehicles p∈P, execute the above steps S1~S5 to obtain all the passage order pairs of the vehicle and add them to the set ST. Then the observation record set D can be transformed into all the passage order pairs of the vehicles. The set ST is the one that stores this data.
[0085] A2. Traffic Flow Pattern Mining
[0086] Traffic flow patterns refer to the overall trends in traffic operation, including lane distribution, branching ratio, and vehicle speed.
[0087] This step is based on historical observation data to mine traffic flow patterns, including lane distribution patterns and path selection patterns.
[0088] In the practice of intelligent transportation engineering, it is often found that different types of vehicles have different behavior patterns due to different purposes of vehicle use and different road restrictions on vehicle types. Therefore, this invention distinguishes and processes the traffic flow patterns of different vehicle types.
[0089] In addition to vehicle type, vehicle travel patterns also exhibit differences and patterns over time. For example, traffic flow patterns differ significantly between holidays and regular weekdays, including changes in destinations and traffic flow indicators. Similarly, morning and evening rush hours are also factors influencing the modeling.
[0090] Therefore, this invention divides time into different time ranges to ensure that vehicle behavior patterns are similar within the same time range, while behavior patterns differ across different time ranges. The division method is as follows:
[0091] S1. Divide the days, such as dividing the year into holidays, weekdays, weekends, etc. The best division method can be obtained based on data analysis during the specific implementation.
[0092] S2. Divide the day into hours, such as dividing the day into (0,6], (6,12], (12,18], (18,24]. The specific implementation depends on the data analysis.
[0093] After the above time division, any vehicle traffic data can be divided into different time ranges according to its observation time. For example, if t=2022-05-01 10:39:46, then this observation record can be divided into the May Day holiday in 2022 (6,12], and thus the traffic flow pattern of that time period can be applied.
[0094] Next, after overlaying the vehicle type and travel time classifications, we will explore the lane distribution patterns and route selection patterns.
[0095] A2.1 Lane Distribution Pattern
[0096] Lane distribution patterns are used to characterize the lane distribution of traffic flow at each monitoring facility. Within any date and time range T... d Any time range T (in hours) h Within the lane layout, the lane distribution pattern must meet the following requirements:
[0097]
[0098] in Indicates observation facilities The largest lane number below, This indicates the traffic flow of Class C vehicles at the observation facility. The selected lane, where ls represents lane selection, is represented by equation (1), which means that within any time range T... d T h Within the context of any observation facility, for any type of vehicle, the sum of the probabilities of choosing each lane in traffic flow must be 1.
[0099] In any time range T d T h Within, for any observation facility The observation set for Category C vehicles under this observation facility is: Where d(·) represents the date portion of the ·, and h(·) represents the hour portion of the ·. Furthermore, within the time range T...d T h Inside, observation facilities The probability that traffic flow of type c will choose lane i is:
[0100]
[0101] in,
[0102]
[0103] A2.2 Path Selection Patterns
[0104] Path selection patterns are used to characterize how traffic flow chooses downstream observation facilities at each observation facility. Based on the topology between facilities, when an observation facility has only one downstream facility, the probability of choosing that downstream facility as the path is 100%. If there are multiple downstream facilities, probability-based allocation becomes necessary. Within any date and time range T... d Any time range T (in hours) h Within this framework, the path selection rules must meet the following requirements:
[0105]
[0106] in Indicates observation facilities The collection of downstream observation facilities, This indicates the traffic flow of Class C vehicles at the observation facility. Downstream of the selected location, where ts represents target selection, equation (3) represents the selection over any time range T. d T h Within this context, under any observation facility, for any type of vehicle, the sum of the probabilities of choosing each downstream destination must be 1.
[0107] In any time range T d T h Within, for any observation facility The passage order of Class C vehicles starting from this observation facility is set as follows: , where s' (k) Represents record r' (k) The corresponding observation facilities, c' (k) Represents record r' (k) The corresponding vehicle type, t' (k) Represents record r' (k) The corresponding time. Furthermore, within the time range T... d T h Inside, observation facilities Traffic flow selection for Class C vehicles The probability that the facility is downstream is:
[0108]
[0109] in,
[0110]
[0111] A2.3 Maximum Expected Traffic Speed
[0112] Maximum expected traffic flow speed refers to the maximum expected traffic flow speed under free-flow conditions. Free-flow speed typically refers to the speed at which vehicles travel freely on a highway without congestion or blockage.
[0113] First, define free-flow speed. According to the "Technical Standards for Highway Engineering" (JTG B01-2014), when a highway is at Level 1 service, traffic flow is in a completely free-flow state, drivers can freely choose their desired speed, and vehicles are not affected or are minimally affected by other vehicles in the traffic flow. The conditions for Level 1 service are as follows:
[0114]
[0115] It should be noted that v / C here is a specific notation used in traffic engineering, and is not the speed v and vehicle type set C as used in this invention. The calculation method is as follows:
[0116]
[0117] Where pcu is the standard traffic volume, and pcu / (hr.ln) is the actual hourly standard traffic volume per lane. s These are design values. According to the "Technical Standards for Highway Engineering," the design value is 2200 for a design speed of 120 km / h, 2100 for 100 km / h, and 2000 for 80 km / h. This step assumes knowledge of the set of observation facilities. G The design value corresponding to each observation facility is denoted as .
[0118] Here, it's necessary to clarify the calculation method for PCU. PCU stands for Standard Traffic Flow, which is obtained by multiplying the traffic flow of different vehicle types by the conversion factor for each vehicle type, and then summing the results. The specific calculation method is as follows:
[0119]
[0120] Among them, Q c For the natural traffic flow of vehicle type c, w c The conversion factor for vehicle type c is given in the aforementioned "Highway Engineering Technical Standards".
[0121] Therefore, the maximum expected traffic speed under free-flow conditions is calculated below. Similarly, this traffic flow pattern should be differentiated based on vehicle type, observation location, and observation time range.
[0122] The specific approach is as follows:
[0123] Extract observation location Let be the hourly vehicle-to-standard vehicle flow rate within the observation time range. Let the observation set for vehicle type c be denoted as . ,like If the value is zero, it indicates that there is no traffic flow for vehicle type c at that location and time period, and 120km / h is used instead. Otherwise, according to equation (7), the traffic flow pcu / (hr.ln) can be calculated as follows:
[0124]
[0125] in, This refers to the span of the time range, in hours. If it meets the requirements for an observation facility The number of lanes at that location.
[0126] Substituting equation (8) into equation (6), we further obtain:
[0127]
[0128] like If condition (5) is met, it indicates that the current state is a free flow state. Further, the observation set D... c|g The time range is divided into several 5-minute intervals (this batch of observations can be further divided into...) • Calculate the average traffic speed in the j-th 5-minute interval (12 five-minute intervals). Then, the maximum expected traffic flow speed of vehicle type c at this location within this time range is:
[0129]
[0130] If condition (5) is not met, it means that the flow is not in a free flow state, and therefore the maximum expected traffic speed cannot be calculated. The default speed of 120 km / h is used instead.
[0131] Summarizing the above calculation steps, we can obtain the result for any time range T. d T h Within, for any observation facility The maximum expected traffic flow speed for Class C vehicles under this observation facility is:
[0132]
[0133] Where v / C is given by equation (9), See equation (10).
[0134] A3. Vehicle Behavior Model Construction
[0135] Vehicle behavior models refer to the behavioral patterns or preferences of vehicle travel, including vehicle steering behavior, lane changing behavior, following behavior, route selection behavior, etc.
[0136] This step constructs a refined vehicle behavior model based on historical observation data, aiming to characterize each vehicle's lane selection preferences and route selection preferences. Similar to A2, the vehicle behavior model also considers the impact of different vehicle types and different travel times, and the division of vehicle type and time range is the same as in A2.
[0137] A3.1 Lane Selection Probability
[0138] Similar to the lane distribution pattern described in A2.1, lane selection probability is used to describe a vehicle's preference for choosing different lanes while driving, except that the granularity is no longer at the traffic flow level, but at the single-vehicle level. Lane selection preference should also meet the requirements of equation (1).
[0139] In any time range T d T h Within, for any observation facility The observation set of a certain vehicle p under this observation facility is as follows: Its vehicle model is C, and further, within the time range T... d T h The observation facility is located inside. The lane selection preference of vehicle p is as follows:
[0140]
[0141] in, See equation (2),
[0142]
[0143] Extract all combinations of (vehicle, observation facility, observation date range, observation hour range) from the historical data. Based on equation (12), a mapping between each combination and its corresponding lane selection probability can be established, i.e.:
[0144]
[0145] in, Indicates observation facilities Below it is a collection of observation facilities.
[0146] A3.2 Path Selection Probability
[0147] Similar to the path selection rules described in A2.2, the path selection probability is used to characterize the path selection behavior of vehicles, with the granularity refined from the traffic flow level to the single vehicle level. The path selection probability should also meet the requirements of equation (3).
[0148] In any time range T d T h Within, for any observation facility A vehicle originating from this observation facility The passing order of the set is If its vehicle model is c, then within the time range T d T h Inside, there are observation facilities. The probability that vehicle p's downstream facility is facility i is:
[0149]
[0150] in, See equation (4),
[0151]
[0152] Extract all combinations of (vehicle, observation facility, observation date range, observation hour range) from the historical data. Based on equation (14), a mapping between each combination and its corresponding path selection probability can be established.
[0153]
[0154] in, Indicates observation facilities Below it is a collection of observation facilities.
[0155] A3.3 Maximum Expected Speed
[0156] Maximum expected speed refers to the maximum speed a vehicle is expected to reach in free-flow conditions. Similar to A2.3, the calculation of maximum expected speed assumes that the traffic flow is in a free-flow state.
[0157] In any time range T d T h Within, for any observation facility Let the observation set of a certain vehicle p under this observation facility be denoted as . Its model is .
[0158] like If no data is available, the default speed of 120 km / h will be used. Furthermore, based on equations (8) and (9), the v / C value within this time range can be calculated.
[0159] If v / C satisfies condition (5), it indicates that the vehicle is in a free-flow state. Therefore, the maximum expected speed of vehicle p is:
[0160]
[0161] in For set The Middle data The corresponding observation speed.
[0162] If v / C does not satisfy condition (5), it means that the vehicle is not in a free-flow state at that time. Then the maximum expected speed of the vehicle p is the default 120km / h.
[0163] Summarizing the above calculation steps, we can obtain the result for any time range T. d T h Within, for any observation facility The maximum expected speed of a certain vehicle p is
[0164]
[0165] Where v / C is given by equation (9), See equation (16).
[0166] Similarly, by extracting all combinations of (vehicle, observation facility, observation date range, observation hour range) from the historical data, a mapping between each combination and its corresponding maximum expected speed can be established according to equation (17), i.e.
[0167]
[0168] A4. Real-time data access and fusion
[0169] The goal of this step is to access M real-time observation data sources. Similarly, using the data fusion method disclosed in ZL 202211116113.9, the M real-time observation data sources and the inter-facility topology are input into the method to obtain real-time fused data. Let the ordered six-tuple in the real-time fused data be denoted as... .
[0170] A5. Dynamically Updated Model
[0171] The purpose of this step is to add the real-time data obtained in step A4 to the historical observation data, and based on this, dynamically update the traffic flow pattern model and vehicle behavior model described in A2 and A3 to ensure their time validity.
[0172] This process is divided into two steps: updating historical data and updating the model.
[0173] A5.1 Update Historical Data
[0174] For any received real-time fused data The specific steps are as follows:
[0175] S1. Update the observation dataset. Add the data to set D.
[0176] S2. Update the license plate set. Extract license plate p* from the data. If p*∉P, add it to set P.
[0177] S3. Update the passage order pair set. For this vehicle p*, perform steps S1 to S5 of A1.2 separately, and add the resulting passage order pairs to the set ST.
[0178] Executing S1 to S3 above means using this real-time fused data to update the observation data.
[0179] A5.2 Update Model
[0180] Every Over time (e.g., 1 hour), the vehicle's lane selection model (i.e., updating the mapping) can be dynamically updated according to equations (13), (15), and (17). The downstream selection model of the vehicle (i.e., the update mapping ω) and the maximum expected speed model of the vehicle (i.e., the update mapping ω) ).
[0181] A6. Simulation Environment Construction
[0182] This step is used to build the simulation environment, including the road network model and the location mapping of the observation facilities in the road network model (to accurately map the observed vehicle data into the road network model).
[0183] A6.1 Simulated Road Network Construction
[0184] The simulated road network is the foundation for the operation of the microscopic traffic simulation system. SUMO road network files are generally composed of nodes, edges, lanes, junctions, connections, etc. The construction of the simulated road network is not the core work of this invention, but a prerequisite. It is only necessary to meet the general requirements of SUMO for road network files, which will not be elaborated here.
[0185] A6.2 Mapping of Observation Facilities
[0186] To achieve real-time simulation, the data sensed by real-world observation facilities must be correctly projected onto the simulated road network. Therefore, a mapping between observation facilities and simulated road network nodes needs to be constructed. For small-scale road networks, manual comparison can be used to complete the mapping construction. However, for large-scale road networks, a construction algorithm is required, which will be described in detail below.
[0187] S1. Node Extraction. Extract all nodes from the simulated road network described in step A6.1, including node numbers and coordinates.
[0188] S2. Nearest node search. For each observation facility, calculate its distance to each node; the closest one is the mappable node.
[0189] S3. Inspection. To ensure the accuracy of the simulation, the distance between the observation facility and the node must be controlled to be less than a certain threshold, denoted as... If the conditions are not met, it indicates that there is no node near the observation facility in the initial simulated road network, and the road network needs to be adjusted accordingly. The adjustment method is to find the edge closest to the facility, use the vertical projection point of the observation facility to interrupt the edge, and insert a new node at the vertical projection point (i.e., the interruption point).
[0190] S4. After adjusting the simulated road network according to the S3 check, re-execute S1~S3 until all mapping relationships satisfy the condition that the distance is less than 1 / 3. Requirements.
[0191] A7. Simulation and Deduction
[0192] This step, based on the aforementioned preparatory work and real-time fused data, involves simulation and deduction. For any set of real-time data from the microscopic simulation system, the following steps are performed.
[0193] S1. Information Extraction. Based on the ordered six-tuple... Let the observation facility be s*, the license plate be p*, the vehicle type be c*, the lane be l*, and the time be t*. Let the date and time range to which t* belongs be . The hourly time range is .
[0194] S2. Determine whether the license plate p* already exists in the current simulation system. If it does, remove the vehicle from the system (because the new observation data is real, the spatiotemporal information of the corresponding vehicle in the simulation system can be directly reset).
[0195] S3. Downstream Path Selection. Let the set of downstream observation facilities be s * . License plate (p*), vehicle type (c*), facility (s*), time range , Substituting into equation (15), we can obtain the probabilities of vehicle p* going to each downstream destination as follows:
[0196]
[0197] The elements in the set are denoted as follows: .
[0198] By summing up the probabilities from smallest to largest, we get:
[0199]
[0200] In equation (18), the terms from largest to smallest are respectively: According to equation (3), the last term is 1.
[0201] Given a uniform region U(0,1) between 0 and 1, a random number is generated, denoted as a. Then, the vehicle p* selects P. (i) The corresponding downstream observation facilities, if:
[0202]
[0203] Note that when i=1, Cum0 is meaningless, so we stipulate that Cum0=0 to ensure logical integrity. Finally, let the selected downstream be... .
[0204] S4. Downstream arrival lane selection. Select the downstream lane chosen in S3. As input, let its maximum lane number be . License plate (p*), vehicle model (c*), facilities Time range , Substituting into equation (13), we can obtain the probabilities of vehicle p* choosing each lane when it reaches the downstream area as follows:
[0205]
[0206] Similar to S3, by summing up the probabilities one by one, we get:
[0207]
[0208] The terms in equation (19) are respectively According to equation (1), the last term is 1.
[0209] Given a uniform region U(0,1) between 0 and 1, a random number, denoted as b, is generated. Then, the vehicle p* is downstream. If the i-th lane is selected, then:
[0210]
[0211] Note that when i=0, Meaningless, therefore agreed upon Ensure logical integrity. Finally, record the selected lane as... .
[0212] Furthermore, in this step (S4), when calculating the probability of a vehicle choosing each lane upon reaching the downstream destination, the time range substituted is... , For the sake of rigor, the time range when the vehicle arrives downstream should be substituted. However, this approach will undoubtedly increase the amount of calculation and may result in inaccurate estimation. In addition, since the normal time interval between the vehicle's arrival at the upstream and downstream observation facilities in the highway scenario is within an acceptable range (10km distance, 100km / h, 6 minutes), the impact on the choice of time period is negligible. Therefore, this error is ignored here.
[0213] S5. Maximum expected speed setting. Based on license plate p*, observation location, and s* time range. The maximum expected speed is obtained as follows:
[0214]
[0215] S6. Based on the mapping relationship described in step A6.2, set the starting point... ,end The observation facility is converted into a computable node in SUMO.
[0216] S7. Record the license plate p*, speed v*, time t*, vehicle type c*, starting lane l*, and ending lane. The starting and ending nodes, and the maximum expected speed *mes* are input to SUMO, and its API is called to perform simulation. It is worth noting that since only the starting and ending points are given, the specific vehicle path is determined by SUMO using the shortest path algorithm. This approach is suitable for highway scenarios because highway networks are closed routes with relatively simple structures. Furthermore, due to their dual functions such as toll collection and auditing, the paths between observation facilities are generally unambiguous and are all shortest paths. For complex scenarios such as urban road networks, a more sophisticated path allocation method is required. Since this scenario is not covered by this invention, it will not be considered.
[0217] A8. Real-time result output
[0218] This step outputs the simulation results, which are divided into two parts: single-vehicle level and traffic flow indicators.
[0219] 1) Single-vehicle level data
[0220] Vehicle-level data includes the real-time location, speed, acceleration / deceleration, and azimuth of any vehicle. This data can be used for vehicle location visualization or for custom aggregation or complex calculations to obtain higher-level statistical results. This vehicle-level data can be obtained through the API provided by SUMO.
[0221] 2) Traffic flow indicators
[0222] Traffic flow metrics are based on traffic flow rather than individual vehicles, and typically include traffic volume, vehicle speed, traffic density, queue length, etc. In SUMO, these metrics can be obtained via API or persistent files by setting up various detectors in the simulated road network.
[0223] Since the real-time results are output using the SUMO normalization function, we will not go into further detail here.
Claims
1. A SUMO-based high-speed network real-time microscopic traffic simulation method, characterized in that, include: Acquire historical and real-time observation data of road network vehicles from multiple sources and fuse them separately to obtain historical fused data and real-time fused data; A set of observation data includes vehicle identity, location, speed, lane, vehicle type, and time information; historical fusion data is transformed into traffic sequence pairs for analyzing traffic flow patterns and establishing vehicle behavior models. The traffic sequence pair refers to any pair of adjacent observation records after sorting the observation records of any vehicle in ascending order of time. The time is divided into date time ranges in days and hour time ranges in hours. For each observation data point, it is divided into different time ranges according to the observation time; Using the aforementioned date and hour time ranges as time units, the vehicle type and travel time are superimposed, and traffic flow patterns are analyzed based on historical fusion data, including lane distribution patterns, path selection patterns, and maximum expected traffic flow speed. A single-vehicle-level vehicle behavior model is established, including a lane selection probability model, a path selection probability model, and a maximum expected speed model. The path selection pattern is as follows: within a certain time range, a set of travel order pairs for a certain type of vehicle starting from a certain observation facility is obtained; when the set of travel order pairs is not empty, the probability of traffic flow for that type of vehicle choosing each downstream observation facility is determined based on the proportion of travel order pairs whose endpoints are downstream observation facilities to the total number of travel order pairs in the set; when the set of travel order pairs is empty, the selection probability of each downstream observation facility is set to be equal. The path selection probability model is as follows: Within a certain time range, obtain the set of passage order pairs for a certain vehicle starting from a certain observation facility; when the set of passage order pairs for the vehicle is not empty, determine the probability of the vehicle choosing each downstream observation facility based on the proportion of the number of vehicle passage order pairs whose destinations are each downstream observation facility to the total number of vehicle passage order pairs for the vehicle; when the set of vehicle passage order pairs for the vehicle is empty, use the probability of the traffic flow of the vehicle's type choosing each downstream observation facility as the probability of the vehicle choosing each downstream observation facility. Dynamically update historical fused data and vehicle behavior models using real-time fused data; Construct a simulated road network and map the observation facilities to the computable nodes in the simulated road network; Simulation results are obtained by using real-time fused data and the vehicle behavior model.
2. The method of claim 1, wherein, The lane distribution pattern is determined based on the lanes selected by traffic flow of a certain type of vehicle under a certain observation facility within a certain time range; Lane selection probability is determined based on the lane selected by a vehicle under a certain observation facility within a certain time range.
3. The method of claim 1, wherein, The method for determining the maximum expected traffic flow speed is as follows: If there is no traffic flow of a certain type of vehicle flow under a certain observation facility during a certain period, then the maximum expected speed of that type of vehicle flow is the design speed; otherwise, it is determined whether that type of vehicle flow is in a free flow state. When traffic flow is in a completely free-flow state, the time period is divided into several sub-time periods. The average traffic speed of this type of traffic flow in each sub-time period under the observation facility is calculated. The maximum expected traffic speed of this type of traffic flow in the time period is the maximum value among multiple average traffic speeds; otherwise, the maximum expected traffic speed is the design speed. The method for determining the maximum expected speed is as follows: If there is no observation data for a certain type of vehicle under a certain observation facility during a certain period, then the maximum expected speed of that type of vehicle is the design speed; otherwise, calculate the traffic flow of that type of vehicle and determine whether it is in a free flow state. When the traffic flow is in a completely free-flow state, the maximum expected speed is the maximum observed speed of the vehicle under the observation facility during that time period.
4. The method of claim 1, wherein, The updated historical fusion data and vehicle behavior model include: For any received real-time fused data, it is appended to the historical fused data and the vehicle identity set and passage order pair set are updated; and at fixed intervals, the traffic flow pattern and vehicle behavior model are updated using the updated historical fused data.
5. The method of claim 1, wherein, The methods for mapping the observation facilities include: 1) Extract all node data from the simulated road network, including node number and coordinate information; 2) For each observation facility, calculate its distance to each node and obtain the node closest to the observation facility; 3) When the distance between the observation facility and the nearest node is less than a preset threshold, the observation facility is mapped to the node; otherwise, the edge closest to the facility is found, and the observation facility is interrupted by its vertical projection point relative to the edge. A new node is then inserted at the vertical projection point. 4) Repeat steps 1)-3) until all mapping relationships meet the requirement that the distance is less than the preset threshold.
6. The method of claim 1, wherein, The simulation includes: For any real-time fused data, extract the record information it contains; Based on this recorded information, the vehicle behavior model is used to obtain its downstream path selection, downstream arrival lane selection, and maximum expected speed; according to the mapping established when constructing the simulated road network, the traffic flow starting point and downstream path observation facilities are converted into computable nodes; Simulations are performed based on the recorded information, starting lane, downstream arrival lane, starting node, ending node, and maximum expected speed.
7. A SUMO-based high-speed network real-time microscopic traffic simulation system, characterized in that, include: The data fusion unit acquires historical and real-time observation data of road network vehicles from multiple sources and fuses them separately to obtain historical fused data and real-time fused data. A set of observational data includes vehicle identification, location, speed, lane, vehicle type, and time information; The traffic flow pattern analysis unit transforms historical fusion data into traffic sequence pairs for analyzing traffic flow patterns and establishing vehicle behavior models. The traffic sequence pair refers to any pair of adjacent observation records after sorting the observation records of any vehicle in ascending order of time. The time is divided into a date time range in days and an hour time range in hours. For each observation data point, it is divided into different time ranges according to the observation time. Using the date and hour time ranges as time units, the vehicle type and travel time divisions are superimposed, and traffic flow patterns are analyzed based on historical fusion data, including lane distribution patterns, path selection patterns, and maximum expected traffic flow speed. The traffic flow patterns are dynamically updated based on real-time fusion data. The path selection pattern is as follows: within a certain time range, a set of travel order pairs for a certain type of vehicle starting from a certain observation facility is obtained. When the set of travel order pairs is not empty, the probability of traffic flow for that type of vehicle choosing each downstream observation facility is determined based on the proportion of the number of travel order pairs in the set whose endpoints are each downstream observation facility to the total number of travel order pairs in the set. When the set of travel order pairs is empty, the selection probability of each downstream observation facility is set to be equal. The vehicle behavior model construction unit establishes a single-vehicle-level vehicle behavior model based on historical fusion data, including a lane selection probability model, a path selection probability model, and a maximum expected traffic flow speed model, and dynamically updates the model based on real-time fusion data. The path selection probability model is as follows: within a certain time range, a set of passage order pairs for a vehicle originating from a certain observation facility is obtained; when the set of passage order pairs for the vehicle is not empty, the probability of the vehicle choosing each downstream observation facility is determined based on the proportion of the number of vehicle passage order pairs ending at each downstream observation facility to the total number of vehicle passage order pairs for the vehicle; when the set of vehicle passage order pairs for the vehicle is empty, the probability of the vehicle choosing each downstream observation facility is taken as the probability of the vehicle choosing each downstream observation facility based on the traffic flow probability of the vehicle's vehicle type. The simulation environment construction unit builds a simulated road network and maps observation facilities to computable nodes in the simulated road network. The simulation unit uses real-time fused data and vehicle behavior models to perform simulations and obtain simulation results.
Citation Information
Patent Citations
Real-time fusion method for multi-source roadside traffic observation data
CN115186781A
Expressway congestion prediction method based on vehicle behavior analysis
CN114023073A
Microcosmic simulation multi-source data fusion analysis method, electronic equipment and storage medium
CN115080638A
Simulation method and verification method for online microscopic traffic of large-scale road network
CN115310278A
