A traffic flow state deterioration degree dynamic evaluation method based on space-time adaptive feature fusion

Abstract

This invention relates to the field of intelligent transportation technology, and discloses a dynamic assessment method for traffic flow degradation based on spatiotemporal adaptive feature fusion. The method includes: first, acquiring multi-source micro-trajectory and macro-environmental data; second, constructing a set of macro- and micro-evaluation indicators, including heterogeneous vehicle flow coordination degree and environmental adaptive aggressive driving index, within a sliding time window; then, performing dynamic adaptive dimensionality reduction based on historical sequences to remove redundant features and construct a dynamic evaluation feature matrix; next, introducing a temporal decay mechanism to calculate the dynamic fuzzy grey relational coefficient; and finally, calculating the flow degradation degree through dynamic entropy weight allocation and mapping it to a risk level output. This invention solves the problems of traditional methods being single-dimensional and lacking spatiotemporal adaptability and timeliness. By fusing macro- and micro-features, adaptive dimensionality reduction, and a temporal decay mechanism, it achieves accurate capture of traffic flow degradation patterns, providing high-precision decision support for road network risk early warning.

Description

Technical Field

[0001] This invention relates to the field of intelligent transportation technology, and in particular to a dynamic assessment method for traffic flow degradation based on spatiotemporal adaptive feature fusion. Background Technology

[0002] With the acceleration of urbanization and the continuous growth of motor vehicle ownership, road traffic systems are facing unprecedented pressure. Intelligent traffic management and vehicle safety early warning technologies, as key means to alleviate traffic congestion and improve road safety, have been widely applied to various urban roads, highways, and bridge tunnels. In actual production and daily life, traffic managers typically use fixed detectors deployed on the roadside (such as geomagnetic coils and microwave radar) or mobile sensing sources (such as floating car GPS data and connected vehicle terminals) to collect macro-level traffic flow parameters such as traffic flow, average vehicle speed, and time occupancy in real time. Based on this data, through threshold comparison, statistical analysis, or simple machine learning models, the current traffic operation status is roughly classified, for example, distinguishing between "smooth," "slow," and "congested" modes. These assessment results are directly used for information dissemination on traffic guidance screens, preliminary adjustments to traffic light timings, and public navigation traffic information broadcasts, providing travelers with route selection references to a certain extent and providing auxiliary information for traffic management departments to assess the macro-level situation. Furthermore, some advanced driver assistance systems also incorporate functions such as forward collision warning to make simple judgments about the instantaneous risks surrounding individual vehicles.

[0003] Although the above methods have been applied to some extent in practice, with the increasing complexity of traffic flow (such as the mixing of large trucks, small passenger cars, and non-motorized vehicles) and frequent disturbances from environmental factors such as extreme weather, traditional assessment methods have revealed the following technical problems in specific implementation:

[0004] First, the feature extraction dimension is singular, lacking deep integration of macro and micro features. Most existing assessment techniques rely on single macro cross-sectional data (such as flow rate and average speed) or isolated micro behavioral data (such as instantaneous speed and acceleration of a single vehicle). Such single-dimensional assessment is difficult to accurately depict the evolution mechanism of flow in complex traffic environments. For example, average speed alone cannot reflect the speed dispersion effect and potential frequent blockages caused by mixed traffic of large trucks. Acceleration of a single vehicle alone cannot measure the overall driving aggression of a road segment. Traditional methods fail to effectively integrate the overall situation of macro traffic flow with the volatility of micro driving behavior, resulting in insufficient early identification of the "deterioration" process within traffic flow. Often, it can only be detected after congestion or accidents have already occurred, missing the best time for intervention.

[0005] Secondly, the evaluation model lacks spatiotemporal adaptability and dynamic dimensionality reduction capabilities. With the rapid development of vehicle-road cooperation and IoT technologies, the available traffic data is characterized by its massive volume, multiple sources, and heterogeneity. Traditional methods often lack dynamic processing capabilities when faced with high-dimensional feature sets composed of dozens of macro and micro indicators. On the one hand, there may be high information redundancy and multicollinearity among different indicators (e.g., flow rate and density are highly correlated), and directly incorporating them into the model will trigger the "curse of dimensionality," reducing computational efficiency and model robustness. On the other hand, key indicators representing traffic flow dynamically evolve with changes in traffic conditions (e.g., speed dispersion is more important during congestion, while headway variance is more sensitive during free flow). Traditional methods use a fixed indicator system, which cannot adaptively remove redundant information or highlight key features under the current state, resulting in a significant reduction in the accuracy and real-time performance of the evaluation results.

[0006] Furthermore, the evaluation mechanism does not consider the decay effect of temporal evolution. The evolution of traffic flow status is a typical process of "memory" but "forgetting": historical status has an impact on the current status, but this impact gradually decays over time. Existing algorithms often treat data within the historical time window as equally important when calculating the correlation or weight of evaluation indicators, or simply use unweighted moving averages. This approach ignores the objective evolution law of the decay of traffic flow deterioration characteristics over time, making the evaluation results slow to react to recent drastic changes, resulting in a lag in early warning. For example, a brief disturbance that occurred 5 minutes ago and a disturbance that occurred 1 minute ago obviously have different contributions to the current trend of traffic flow deterioration, but traditional methods fail to effectively reflect this difference.

[0007] Based on the aforementioned technical problems, it is clear that existing traffic operation status assessment methods urgently need to address core issues such as effective fusion of macro and micro features, adaptive dimensionality reduction of high-dimensional features, and temporal evolution assessment under complex mixed traffic flow conditions. Therefore, those skilled in the art require a dynamic assessment method for traffic flow deterioration based on spatiotemporal adaptive feature fusion. Summary of the Invention

[0008] The core technical problem addressed by this invention is how to solve the problems of traditional assessment dimensions being singular, models being rigid, and early warnings being lagging, so as to achieve accurate capture and forward-looking early warning of the evolution process of traffic flow from a stable state to an unstable state.

[0009] To address the aforementioned core technical issues, this invention designs a dynamic assessment method for traffic flow degradation based on spatiotemporal adaptive feature fusion. The purpose is to achieve dynamic assessment of traffic flow degradation, and to deeply integrate macro and micro traffic features during the assessment process, possessing spatiotemporal adaptive dimensionality reduction capabilities and considering temporal decay patterns.

[0010] It includes the following:

[0011] First, microscopic vehicle trajectory data and macroscopic environmental status data of the target road are acquired through multi-source sensing methods;

[0012] Next, within a dynamically sliding observation window, a set of macro and micro traffic characteristic evaluation indicators, including heterogeneous traffic flow speed coordination degree and environmental adaptive aggressive driving behavior variation index, was creatively constructed to comprehensively characterize the intrinsic state of traffic flow.

[0013] Then, based on the historical index sequence, adaptive dimensionality reduction is performed using dynamic correlation coefficient to construct a dynamic evaluation feature matrix that reflects the current key characteristics, effectively eliminating data redundancy.

[0014] Based on this, an exponential time-series decay mechanism is introduced to calculate the dynamic fuzzy grey relational coefficients of each indicator in order to accurately simulate the "memory-forgetting" characteristics of traffic flow evolution.

[0015] Finally, the traffic flow degradation degree is comprehensively calculated through dynamic entropy weight allocation and mapped to an intuitive risk level for output.

[0016] The entire solution forms a complete technical closed loop of "multi-source perception—adaptive assessment—risk classification," providing high-precision decision support for road network risk early warning.

[0017] To achieve the above objectives, the specific technical solution of the present invention is a dynamic assessment method for traffic flow degradation based on spatiotemporal adaptive feature fusion, which includes the following steps:

[0018] Step 1: Obtain microscopic vehicle trajectory data and macroscopic environmental status data for the target road;

[0019] Step 2: Based on microscopic vehicle trajectory data and macroscopic environmental state data, set a sliding observation time window and update step size for dynamic evaluation, and construct a set of macroscopic and microscopic traffic characteristic evaluation indicators based on the trajectory data (i.e., microscopic vehicle trajectory data and macroscopic environmental state data) within the sliding observation time window.

[0020] Step 3: Based on the historical evaluation index sequence, perform dynamic adaptive dimensionality reduction on the macro and micro traffic characteristic evaluation index set to construct a dynamic evaluation feature matrix;

[0021] Step four: Based on the dynamic evaluation feature matrix, a time-series decay mechanism is introduced to calculate the dynamic fuzzy grey relational coefficients of each evaluation index. Here, each evaluation index mainly refers to the dynamic evaluation feature matrix generated after dimensionality reduction in step three, which is the set of macro and micro traffic characteristic evaluation indicators calculated in step two. In other words, each evaluation index is a set of indicators that have been retained after dimensionality reduction;

[0022] Step 5: Based on the dynamic fuzzy grey relational coefficient, perform dynamic entropy weight allocation to calculate the traffic flow degradation degree, and combine it with the pre-calibrated numerical interval threshold to map and classify the calculated flow degradation degree into the corresponding traffic operation risk level for output.

[0023] Furthermore, the microscopic vehicle trajectory data obtained in step one includes: timestamp, vehicle ID, vehicle latitude and longitude, instantaneous speed, instantaneous acceleration, heading angle, and vehicle type;

[0024] The macro-environmental data includes: meteorological visibility, rainfall, and road surface friction coefficient.

[0025] Furthermore, in step two, the current evaluation time is set to... The length of the sliding observation time window is The calculation time range for the indicator is... ;

[0026] The set of macro and micro traffic characteristic evaluation indicators includes: macro traffic flow parameters and micro traffic behavior parameters;

[0027] The macroscopic traffic flow parameters include: average flow rate. Average density , road saturation Trip delay Congestion Index and heterogeneous traffic flow speed coordination The formulas for calculating all macroscopic traffic flow parameters are as follows:

[0028] Average traffic flow of the target road segment The calculation formula is:

[0029]

[0030] in, To constitute the first of the target road segments Each section of the road is Average flow rate within, For the first The length of each road segment This represents the total number of sub-segments;

[0031] Average density of the target road segment The calculation formula is:

[0032]

[0033] in, For the first The average vehicle density of each road segment within the sliding observation time window;

[0034] road saturation The calculation formula is:

[0035]

[0036] in, This represents the converted number of vehicles per unit road segment within a sliding observation time window. This represents the maximum traffic capacity of the road.

[0037] Trip delay The calculation formula is:

[0038]

[0039] in, The total length of the target road segment. The average spatial velocity of all vehicles within the target road segment during the sliding observation time window. The free-flow velocity of the target road segment;

[0040] Congestion Index The calculation formula is:

[0041]

[0042] Heterogeneous traffic flow speed coordination The formula used to quantify the speed discrete stagnation effect when different vehicle types are driving together is:

[0043]

[0044] in, for The proportion of large vehicles in total traffic volume; and These represent the average spatial velocity of small cars and large cars within the time window, respectively. This represents the overall average speed of the road segment.

[0045] Furthermore, the microscopic traffic behavior parameters are derived from... Appearing within the target road section Vehicle trajectory data extraction, including: vehicle deviation angle variance. Frequency of lane changing behavior Variance of vehicle headway acceleration variance Root mean square of longitudinal jerk and lateral disturbance velocity volatility The formulas for calculating all micro-level traffic behavior parameters are as follows:

[0046] Lane changing frequency The calculation formula is:

[0047]

[0048] in, for The total number of lane changes for all vehicles within the area. This represents the total traffic volume for the same period.

[0049] Variance of vehicle headway The calculation formula is:

[0050]

[0051] in, For vehicles Average headway within the observation time window This represents the overall average headway.

[0052] acceleration variance The calculation formula is:

[0053]

[0054] in, For vehicles The average acceleration within the observation time window The overall average acceleration;

[0055] Vehicle deviation variance The calculation formula is:

[0056]

[0057] in, For vehicles The average heading angle within the observation time window This represents the overall average heading angle;

[0058] Root mean square of longitudinal jerk The calculation formula is:

[0059]

[0060] in, For vehicles At any moment The degree of agitation, For instantaneous acceleration, For vehicles The total number of valid trajectory points recorded within the observation time window. The time interval for data sampling;

[0061] Lateral disturbance velocity volatility The calculation formula is:

[0062]

[0063] in, For vehicles At any moment instantaneous speed, This is the heading angle.

[0064] Furthermore, the microscopic traffic behavior parameters also include the variance of the environmental adaptive aggressive driving behavior variability index. ;

[0065] The variance of the environmental adaptive aggressive driving behavior variability index The calculation process is as follows:

[0066] First, calculate the dynamic environmental penalty factor affected by macro-environmental state data. :

[0067]

[0068] In the formula, For a moment Meteorological visibility data collected; The preset standard safe visibility threshold; For a moment Rainfall, The threshold for extreme rainfall; For a moment The collected road surface friction coefficient, The ideal friction coefficient for a dry road surface; , , The dynamic environmental sensitivity calibration weights are respectively for visibility, rainfall, and friction coefficient.

[0069] Secondly, calculate the observation time window for a single vehicle. Intra-environment adaptive aggressive driving behavior variation index :

[0070]

[0071] in, For a moment The average speed of traffic in the lane in which the vehicle is located; For maximum safe acceleration; This refers to a dynamic environmental penalty factor influenced by macro-environmental status data. , , These are the weighting coefficients for acceleration, relative velocity difference, and deflection angle, respectively.

[0072] Finally, calculate the target road segment Vehicle Environmental Adaptive Aggressive Driving Behavior Variation Index Variance :

[0073]

[0074] in, Within the observation time window The average value of the aggressive driving behavior variation index of a vehicle.

[0075] Furthermore, the process of constructing the dynamic evaluation feature matrix in step three is as follows:

[0076] Extract a historical sequence window Time series of any two evaluation indicators and Its dynamic correlation coefficient The calculation formula is:

[0077]

[0078] in, As the current assessment baseline time, and These are the mean values ​​of the indicators within the historical time series window; when Greater than the dynamic threshold adjusted in real time by the macro congestion index. At that time, redundant indicators with low information entropy are removed, and a dimensionality-reduced dynamic evaluation feature matrix is ​​generated. ,in The number of road sections to be evaluated. The number of valid evaluation indicators to be retained.

[0079] Furthermore, the calculation process of the dynamic fuzzy grey relational coefficient in step four is as follows:

[0080] Dynamic evaluation feature matrix Extreme value standardization is performed to obtain the absolute deviation of membership degree. Then, the dynamic grey relational coefficient with time decay factor. The calculation formula is:

[0081]

[0082] in, The time decay factor, The resolution coefficient.

[0083] Furthermore, the specific calculation process for traffic flow degradation in step five includes:

[0084] The dynamic grey relational coefficients in step four are normalized to obtain several indicators (where the several indicators are derived from the dynamic evaluation feature matrix). A new index is obtained after normalizing the grey relational coefficient. Then, calculate the first... Dynamic information entropy of each indicator and dynamic weights :

[0085]

[0086] The dynamic weights and dynamic grey relational coefficients of the comprehensive indicators are used to calculate the target road segment. At any moment Traffic flow deterioration :

[0087]

[0088] in, This refers to the total number of indicators retained after dimensionality reduction. Number of road segments; It characterizes the deviation of the current traffic flow state from the ideal safe state. And it classifies the degree of traffic flow deterioration according to a preset threshold. This is mapped to the corresponding traffic operation risk level.

[0089] It should be noted that step one is the data foundation for the entire method; the main purpose of step two, establishing a set of macro and micro indicators, is to comprehensively quantify the complex state of traffic flow; steps three and four are the core data processing and evaluation processes, aiming to eliminate feature redundancy and highlight the timeliness of the actual state; step five, by combining weights and correlations, derives the final degree of traffic flow degradation and risk level, completing an accurate "diagnosis" of the road network's operating status; the entire method has a clear logical chain, forming a complete intelligent traffic monitoring closed loop of "multi-source perception - adaptive evaluation - risk classification".

[0090] Compared with the prior art, the technical solution disclosed in this application has the following non-obvious technical features:

[0091] First, this application includes a macro-micro fusion index set comprising "heterogeneous traffic flow speed coordination degree" and "environmentally adaptive aggressive driving behavior variation index." Existing technologies either only focus on macro-level flow / speed or only analyze micro-level individual vehicle behavior; however, this application not only introduces the "heterogeneous traffic flow speed coordination degree" to quantify the blocking effect of mixed traffic of different vehicle types, but also innovatively constructs the "environmentally adaptive aggressive driving behavior variation index." This index does not simply count the number of rapid accelerations, but first dynamically generates an "environmental penalty factor" from macro-level environmental data (visibility, rainfall, friction coefficient), then uses this factor to correct the aggressiveness calculated based on micro-level driving behavior (acceleration, relative speed difference, deflection angle), and finally obtains the overall variance of the road segment. This dynamic composite index, which integrates macro-level environmental tolerance and micro-level relative driving behavior, has not been disclosed in existing literature and patents.

[0092] Second, this application proposes an online adaptive dimensionality reduction mechanism based on real-time feedback of dynamic correlation coefficients and information entropy. Existing dimensionality reduction methods are mostly static or offline (such as principal component analysis). This application proposes an online adaptive dimensionality reduction process: calculating the dynamic correlation coefficient of any two indicators within a historical sequence window in real time and comparing it with a "dynamic threshold adjusted by real-time feedback from the macro-congestion index." This means that the dimensionality reduction "threshold" changes dynamically with the macro-traffic state (congestion level). When congestion worsens, the tolerance for collinearity of indicators may be adjusted accordingly, making the retained feature set more adaptable to the current traffic scenario. This mechanism, which couples the dimensionality reduction process with real-time traffic state feedback, is significantly innovative.

[0093] Third, this application employs dynamic fuzzy grey relational analysis with a time-series decay factor for calculating the correlation coefficient. Grey relational analysis is a commonly used evaluation method, but traditional calculations do not consider information decay along the time axis. This scheme innovatively introduces an exponential time decay factor when calculating the correlation coefficient, ensuring that data closer to the current moment contributes more to the correlation coefficient, while data further away contributes less. This cleverly integrates the "memory-forgetting" characteristic (temporal evolution law) of traffic flow into fuzzy grey relational theory, making the evaluation results more timely and sensitive. It is not a simple combination of algorithms, but rather an innovative application of fundamental theory.

[0094] Compared with the prior art, the present invention has the following beneficial effects: 1. The present invention improves the accuracy and scientific nature of the evaluation results. By introducing heterogeneous traffic flow speed coordination degree and multi-dimensional vehicle micro-disturbance behavior (such as jerkiness, deviation angle variance, etc.), and combining environmental parameters to construct an aggressive driving behavior variation index, it reflects the inherent deterioration and evolution mechanism under the coupling effect of complex traffic flow and environment in a delicate and accurate way from both macro-situation and micro-mechanism levels, overcoming the shortcomings of the traditional method's single evaluation dimension;

[0095] 2. This invention significantly enhances the spatiotemporal adaptability and real-time performance of the evaluation model. It employs an adaptive dimensionality reduction algorithm based on dynamic correlation coefficients and real-time feedback thresholds, effectively eliminating multicollinearity among high-dimensional data, avoiding the curse of dimensionality, and improving computational efficiency. Simultaneously, by introducing an exponential temporal decay mechanism, the evaluation results can more sensitively capture recent traffic flow changes, greatly improving the sensitivity and accuracy of the system's early warning system, and solving the problems of rigidity and delayed warnings in traditional models.

[0096] 3. This invention provides high-precision decision support for road network risk management, scientifically mapping the calculated traffic flow degradation degree to intuitive traffic operation risk levels (e.g., stable, slightly fluctuating, significantly unstable, severely unstable). This makes the assessment results not only supported by profound physical mechanisms but also highly readable and operable. This result can directly serve the dynamic adjustment of traffic control strategies, the accurate delivery of safety warning information, and risk decision-making for autonomous vehicles, effectively improving the safety monitoring level and intelligent management capabilities of various types of road networks in actual operation. (See attached figures.)

[0097] Figure 1 This is a flowchart of the dynamic evaluation method described in this invention;

[0098] Figure 2 This is a schematic diagram of the structure of the dynamic evaluation system described in this invention. Detailed Implementation

[0099] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings;

[0100] The technical solution and embodiments of the present invention will now be described in detail with reference to the accompanying drawings:

[0101] In the technical solution of this application, the entire method flow is as follows: Figure 1 As shown; the method includes the following:

[0102] First, the system acquires multi-source microscopic vehicle trajectory data and macroscopic environmental status data for the target road, and sets a sliding observation time window and update step size for dynamic evaluation. The system utilizes sensing devices deployed along the road to collect real-time microscopic trajectory data including timestamps, vehicle IDs, latitude and longitude, instantaneous speed, instantaneous acceleration, heading angle, and vehicle type. Simultaneously, it acquires macroscopic environmental status data including meteorological visibility, rainfall, and road surface friction coefficient. Assume the current evaluation reference time is... The length of the sliding observation time window is The system intercepts Multi-source data within a time frame are used for subsequent calculations.

[0103] Based on trajectory data within this time window, a set of macro and micro traffic characteristic evaluation indicators is constructed. At the macro level, the indicators include average traffic flow. Average density , road saturation Trip delay Congestion Index This invention specifically constructs a heterogeneous vehicle flow speed coordination degree. This formula is used to quantify the speed discrete stagnation effect when different vehicle types are driving together. .

[0104] In terms of assessing micro-level traffic behavior characteristics, the system is based on the data generated within the observation time window. Vehicle trajectory data is used to extract and calculate the frequency of lane-changing behavior. Variance of vehicle headway acceleration variance and vehicle deviation variance To deeply characterize the intensity of vehicle handling and lane-keeping stability, this invention systematically calculates the root mean square of longitudinal jerk. With lateral disturbance velocity volatility The root mean square of the longitudinal jerk is calculated by first determining the jerk. Substitute The transverse disturbance velocity fluctuation rate is obtained through... To obtain.

[0105] Meanwhile, to overcome the blind spots of absolute threshold evaluation, this invention innovatively and systematically integrates relative vehicle speed deviation and environmental parameters, first constructing a dynamic environmental penalty factor, the calculation formula of which is: Based on the calculated structure, a single-vehicle environment adaptive aggressive driving behavior variation index was further constructed. Then, the variance of the overall aggressive driving behavior variation index of the road segment is obtained. The aforementioned macro and micro indicators together constitute the initial set of traffic characteristic evaluation indicators.

[0106] Secondly, dynamic adaptive dimensionality reduction is performed based on historical evaluation index sequences to construct a dynamic evaluation feature matrix and extract past historical sequence windows. Time series of any two evaluation indicators and The formula for the dynamic correlation coefficient is:

[0107]

[0108] when Greater than the dynamic threshold adjusted in real time by the macro congestion index. At the same time, redundant indicators with low information entropy are removed, thereby generating a dynamic evaluation feature matrix that retains the core features. .

[0109] Then, a time-series decay mechanism is introduced to calculate the dynamic fuzzy grey relational coefficient, and dynamic entropy weight allocation is performed to calculate the traffic flow degradation degree. The absolute deviation is obtained by standardizing the extreme values ​​of the dynamic evaluation feature matrix. Then, calculate the time decay factor. The grey relational coefficient is calculated using the following formula: Then normalization yields Calculate dynamic information entropy With dynamic weights .

[0110] Finally, the target road segment was determined. Traffic flow deterioration Based on this, traffic operation risk levels are mapped to stable, slightly fluctuating, significantly unstable, or severely unstable and on the verge of collapse.

[0111] In the technical solution of this application, the basic mathematical theories and algorithms involved (such as dynamic information entropy, correlation analysis, fuzzy grey relational decision-making, etc.) are all mature algorithms with abundant open-source engineering implementations; the data sources (integrated radar-visual sensing equipment, meteorological environmental sensors, etc.) have been widely deployed in modern intelligent transportation and vehicle-road cooperative networks, with low acquisition cost and high accuracy; the dynamic adaptive mechanism and time series model (such as sliding time window update, time decay memory function) effectively filter out the transient noise of sensors and occasional disturbances in traffic flow, significantly enhancing the robustness of system evaluation; the core operating parameters (such as time decay factor, dynamic environment penalty factor, risk classification threshold, etc.) can be accurately calibrated according to the actual road network type (such as urban expressways, main roads, intersections, etc.) and local driving characteristics, possessing strong engineering feasibility and universality.

[0112] Example 1;

[0113] A dynamic assessment method for traffic flow degradation based on spatiotemporal adaptive feature fusion, such as Figure 1 As shown, the implementation process is as follows:

[0114] 1. Acquisition of multi-source microscopic vehicle trajectory data and macroscopic environmental status data for the target road;

[0115] First, taking a frequently congested road section in a certain city as the target road, the millimeter-wave radar and high-definition video integrated perception equipment deployed along the route are used to acquire microscopic vehicle continuous trajectory data in real time at a sampling frequency of 10Hz. This data includes timestamp, vehicle ID, vehicle latitude and longitude, instantaneous speed, instantaneous acceleration, heading angle and vehicle type.

[0116] Secondly, by connecting to the meteorological department's data interface and roadside environmental detectors, macroscopic environmental data of the road section, including meteorological visibility, rainfall, and road surface friction coefficient, are collected simultaneously.

[0117] 2. Construction of macro- and micro-level traffic characteristic evaluation index sets;

[0118] First, set a sliding observation window for dynamic evaluation. The interval is 5 minutes, the update step is 1 minute, and the evaluation time is extracted. of Cross-sectional and trajectory data within the window;

[0119] Secondly, calculate the macroscopic parameters within this time window, including average flow rate. Average density , road saturation Trip delay and congestion index ; and combined with the proportion of large and medium-sized passenger and freight vehicles Based on speed differences, the speed coordination degree of heterogeneous traffic flows is calculated. ;

[0120] Then, based on the micro-trajectory data, the frequency of lane-changing behavior is calculated. Variance of vehicle headway acceleration variance and vehicle deviation variance Extracting the degree of agitation Calculate the root mean square of longitudinal jerk. ; Calculate the lateral disturbance velocity fluctuation rate using the formula ;

[0121] Finally, the environmental penalty factor is dynamically adjusted based on the current moderate rain weather conditions. Calculate the variation index of adaptive aggressive driving behavior in a single vehicle environment. Then, the variance of the overall variation index of the road segment is obtained. The construction of a set of evaluation indicators for macro and micro traffic characteristics was completed.

[0122] 3. Dynamic adaptive dimensionality reduction based on historical evaluation index sequences;

[0123] First, extract the historical sequence window of the past 30 minutes. Time series of any two evaluation indicators and The dynamic correlation coefficient between the two is calculated using a formula. ;

[0124] Secondly, the calculated dynamic correlation coefficient With dynamic thresholds adjusted in real time by macro congestion index Perform a comparison;

[0125] Finally, when > At the same time, the dynamic information entropy of the two is compared, and redundant indicators with low information entropy and little effective information are removed (for example, road saturation indicators that are highly collinear with the average density under the current traffic flow), and a dynamic evaluation feature matrix that retains the core effective features is generated. .

[0126] 4. Calculation of dynamic fuzzy grey relational coefficient;

[0127] First, the dynamic evaluation feature matrix after dimensionality reduction is subjected to extreme value standardization to obtain the absolute deviation of the membership degree of each index. ;

[0128] Then, the ideal safe and smooth flow state is used as the reference sequence, and an exponential time decay memory function characterizing the traffic flow memory effect is introduced. Calculate the time decay factor Dynamic fuzzy grey relational coefficient This reflects the exponential decay of historical minor congestion or disturbances on the current state.

[0129] 5. Calculation of traffic flow deterioration and output of risk level;

[0130] First, the dynamic grey relational coefficients calculated in step four are normalized to obtain... ;

[0131] Secondly, combining information entropy theory, the dynamic information entropy of each retained indicator is calculated. and dynamic weights Then, by combining the dynamic weights of each dimension with the grey relational coefficient, the current traffic flow degradation degree of the target road segment is calculated. ;

[0132] Finally, based on the preset numerical range threshold, the calculated traffic flow degradation degree is... The risk level of traffic operation is mapped to a stable state, a slightly fluctuating state, a significantly unstable state, or a severely unstable state on the verge of collapse, and then output.

[0133] Example 2:

[0134] A dynamic assessment system for traffic flow degradation based on spatiotemporal adaptive feature fusion, the system structure is as follows: Figure 2 As shown, the system is used to implement the steps of the method described in Embodiment 1, and the system includes:

[0135] The multi-source data acquisition and alignment module is used to acquire and synchronize microscopic vehicle trajectory data and macroscopic environmental status data;

[0136] The macro- and micro-feature extraction module is used to construct a set of macro- and micro-traffic characteristic evaluation indicators within a sliding observation time window;

[0137] The adaptive dimensionality reduction module is used to perform dynamic adaptive dimensionality reduction based on historical index sequences and construct a dynamic evaluation feature matrix.

[0138] The temporal fuzzy grey relational calculation module is used to introduce a temporal decay mechanism to calculate the dynamic fuzzy grey relational coefficients of each evaluation index.

[0139] The flow deterioration assessment module is used to calculate the traffic flow deterioration degree by dynamically allocating entropy weights and output the traffic operation risk level.

[0140] Example 3:

[0141] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in Embodiment 1.

[0142] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, the phrase "comprising an element defined as..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0143] The above technical solutions only embody the preferred technical solutions of the present invention. Any modifications that may be made by those skilled in the art to certain parts thereof embody the principles of the present invention and fall within the protection scope of the present invention.

Claims

1. A traffic flow state deterioration degree dynamic evaluation method based on spatio-temporal adaptive feature fusion, characterized in that, Includes the following steps: Step 1: Obtain microscopic vehicle trajectory data and macroscopic environmental status data for the target road; Step 2: Based on microscopic vehicle trajectory data and macroscopic environmental status data, set a sliding observation time window and update step size for dynamic evaluation, and construct a set of macroscopic and microscopic traffic characteristic evaluation indicators based on the trajectory data within the sliding observation time window; Step 3: Perform dynamic adaptive dimensionality reduction based on the historical evaluation index sequence to construct a dynamic evaluation feature matrix; Step 4: Introduce a time-series decay mechanism to calculate the dynamic fuzzy grey relational coefficients of each evaluation index; Step 5: Perform dynamic entropy weight allocation, calculate the traffic flow deterioration degree, and combine it with the pre-calibrated numerical range threshold to map the calculated flow deterioration degree to the corresponding traffic operation risk level for output.

2. The method of claim 1, wherein, The microscopic vehicle trajectory data obtained in step one includes: timestamp, vehicle ID, vehicle latitude and longitude, instantaneous speed, instantaneous acceleration, heading angle, and vehicle type; The macro-environmental data includes: meteorological visibility, rainfall, and road surface friction coefficient.

3. The method of claim 1, wherein, The set of macro- and micro-traffic characteristic evaluation indicators in step two includes: macro-traffic flow parameters and micro-traffic behavior parameters.

4. The dynamic evaluation method according to claim 3, characterized in that, The macroscopic traffic flow parameters include: average flow , average density , road saturation , travel delay , congestion index , and heterogeneous vehicle flow speed coordination , wherein the heterogeneous vehicle flow speed coordination is used to quantify the speed dispersion blocking effect when different vehicle types are mixed.

5. The dynamic evaluation method according to claim 3, characterized in that, The microscopic traffic behavior parameters include: vehicle deflection variance. Frequency of lane changing behavior Variance of vehicle headway acceleration variance Root mean square of longitudinal jerk and lateral disturbance velocity volatility ; The micro-level traffic behavior parameters also include: the variance of the environmental adaptive aggressive driving behavior variability index. The calculation process includes: First, calculate the dynamic environmental penalty factor based on macro-environmental status data. ; Then, the variation index of the vehicle's environmental adaptive aggressive driving behavior within the observation time window is calculated. ; Finally, the variance of the environmental adaptive aggressive driving behavior variability index for all vehicles within the target road segment was calculated. .

6. The dynamic evaluation method according to claim 1, characterized in that, The process of constructing the dynamic evaluation feature matrix in step three is as follows: Extract the time series of any two evaluation indicators within a historical sequence window and calculate their dynamic correlation coefficient. ; When the dynamic correlation coefficient is greater than the dynamic threshold adjusted in real time by the macro congestion index, At that time, redundant indicators with low information entropy are removed to generate a dimensionality-reduced dynamic evaluation feature matrix.

7. The dynamic evaluation method according to claim 1, characterized in that, The calculation process of the dynamic fuzzy grey relational coefficient in step four is as follows: After obtaining the absolute deviation of membership degrees by extreme value standardization of the dynamic evaluation feature matrix, a time decay factor is introduced to calculate the dynamic grey relational coefficient with the time decay factor. .

8. The dynamic evaluation method according to claim 1, characterized in that, The specific calculation process for traffic flow degradation in step five includes: Several indicators are obtained by normalizing the dynamic grey relational coefficients. Then, the dynamic information entropy and dynamic weight of each indicator are calculated; By combining the dynamic weights of each indicator with the grey relational coefficient, the target road segment at time [time value missing] is calculated. Traffic flow deterioration .

9. A dynamic assessment system for traffic flow degradation based on spatiotemporal adaptive feature fusion, the system being used to implement the steps of the method according to any one of claims 1 to 8, characterized in that, The system includes: The multi-source data acquisition and alignment module is used to acquire and synchronize microscopic vehicle trajectory data and macroscopic environmental status data; The macro- and micro-feature extraction module is used to construct a set of macro- and micro-traffic characteristic evaluation indicators within a sliding observation time window; The adaptive dimensionality reduction module is used to perform dynamic adaptive dimensionality reduction based on historical index sequences and construct a dynamic evaluation feature matrix. The temporal fuzzy grey relational calculation module is used to introduce a temporal decay mechanism to calculate the dynamic fuzzy grey relational coefficients of each evaluation index. The flow deterioration assessment module is used to calculate the traffic flow deterioration degree by dynamically allocating entropy weights and output the traffic operation risk level.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 8.

Images