Method for determining whether to open emergency lane of expressway under heavy traffic

Abstract

This invention provides a decision-making method for whether to open emergency lanes on highways under high traffic conditions, relating to the field of traffic management technology. The method includes acquiring vehicle driving data from multiple sections of a target road segment, calculating feature correlation using tensor decomposition, extracting frequency domain features and fluctuation periods using empirical mode decomposition and Fourier transform, and determining whether to open the emergency lane. If opened, a traffic flow control system based on fluid dynamics is established to calculate the optimal opening timing and dynamically correct it, thereby improving highway capacity and alleviating traffic congestion.

Description

Technical Field

[0001] This invention relates to the field of technology, and more particularly to a decision-making method for whether to open emergency lanes on highways under high traffic conditions. Background Technology

[0002] With the rapid growth of car ownership and the rapid expansion of the highway network, highway traffic congestion has become increasingly serious, especially during holidays, inclement weather, and special circumstances such as traffic accidents. Emergency lanes, as dedicated passageways for rescue vehicles and emergencies on highways, can effectively alleviate traffic pressure and improve road capacity when opened appropriately under specific conditions. Currently, highway management departments typically decide whether to temporarily open emergency lanes based on traffic flow and congestion levels, but this decision often relies on human experience and simple threshold standards. Summary of the Invention

[0003] This invention provides a decision-making method for whether to open emergency lanes on highways under high traffic conditions, which can solve the problems in the prior art.

[0004] A first aspect of the present invention provides a method for deciding whether to open emergency lanes on highways under high traffic conditions, comprising:

[0005] Obtain vehicle driving data for the upstream, midstream, and downstream sections of the target road segment during the target time period, and calculate traffic flow indicators for each section;

[0006] Vehicle driving data from the upstream, midstream, and downstream sections are mapped to a high-dimensional feature space. The feature correlation between each section is calculated through tensor decomposition. The feature vectors of each section after mapping are resampled, and the frequency domain features and fluctuation period of traffic flow fluctuations are extracted based on empirical mode decomposition and Fourier transform.

[0007] The feature correlation degree, the frequency domain feature and the fluctuation period are normalized and then weighted and summed. Based on the comparison between the weighted summation result and the preset opening threshold, it is determined whether to open the emergency lane.

[0008] If the emergency lane is opened, a traffic flow intelligent control system based on fluid dynamics is established. The system calculates the traffic flow dynamics value based on the longitudinal spacing of vehicles and traffic flow data, and calculates the traffic pressure based on the traffic flow density. The density propagation speed is calculated based on the traffic pressure derivative. Based on the density propagation speed, the target capacity value of each section is determined by combining the entropy weight method. The optimal opening sequence is calculated by maximizing the target capacity value, and the opening strategy is dynamically corrected based on the principle of traffic flow conservation.

[0009] Based on the output of the intelligent traffic flow control system, an emergency lane opening command is sent to the traffic management system; the traffic flow index of each section of the open road segment is monitored in real time, and when the traffic flow index is detected to drop below the preset closing threshold, an emergency lane closing command is sent.

[0010] Vehicle driving data from the upstream, midstream, and downstream sections are mapped to a high-dimensional feature space. Tensor decomposition is used to calculate the feature correlation between features in each section, including:

[0011] Obtain vehicle speed data for the upstream, midstream, and downstream sections, calculate the standard deviation of vehicle speed for each section, and divide it by the corresponding mean vehicle speed to obtain the traffic flow fluctuation intensity index for each section; when the traffic flow fluctuation intensity index is greater than the fluctuation threshold, multiply the preset time window length by the traffic flow fluctuation intensity index.

[0012] The feature vectors of each interval are extracted within the adjusted time window. The feature vectors are then input into a random projection matrix for local sensitive hashing. The elements of the random projection matrix follow a standard normal distribution. Through multiple mapping iterations, the mapped feature vectors are obtained.

[0013] The mapped feature vectors are organized into third-order tensors in the interval and time dimensions. The tensor reconstruction error is calculated. When the reconstruction error is less than the reconstruction error threshold, the current decomposition rank is determined to be the optimal decomposition rank. The optimal decomposition rank is used to perform kernel tensor decomposition to obtain the interval mode decomposition matrix and the time mode decomposition matrix.

[0014] The static correlation degree between intervals is obtained by performing an inner product operation on the eigenvectors of adjacent intervals in the interval mode decomposition matrix and dividing them by the corresponding modulus product. The time mode decomposition matrix is ​​divided into current feature data and historical feature data according to the time window, and the dynamic correlation degree between the current feature data and the historical feature data is calculated. The static correlation degree and the dynamic correlation degree are weighted and summed to obtain the feature correlation degree between each interval feature.

[0015] The feature vectors of each mapped interval are resampled, and the frequency domain features and fluctuation period of traffic flow fluctuations are extracted based on empirical mode decomposition and Fourier transform, including:

[0016] The feature vectors of each mapped interval are discretized over time according to the sampling interval to obtain the traffic flow feature sampling sequence.

[0017] Empirical mode decomposition (EMD) is performed on the traffic flow feature sampling sequence through iterative screening to obtain multiple intrinsic mode functions (IMFs) representing different frequency characteristics. The IMFs are then converted into analytical signals using Hilbert transform, and instantaneous frequency and instantaneous amplitude are extracted from the analytical signals.

[0018] The discrete Fourier transform of the multiple intrinsic mode functions is performed to obtain the spectral components of each intrinsic mode function. The power spectral density is obtained by weighted summation of the spectral components. The dominant frequency component is extracted from the power spectral density, and the phase accumulation characteristics are calculated by combining the spectral phase of each intrinsic mode function to obtain the frequency domain characteristics of traffic flow fluctuation.

[0019] The frequency deviation between the instantaneous frequency of each intrinsic mode function and the dominant frequency component is calculated. The frequency similarity is obtained by Gaussian weighting. The root mean square of the instantaneous amplitude of each intrinsic mode function is calculated as the energy value. The frequency similarity is multiplied by the energy value to obtain the contribution of each intrinsic mode function to the dominant frequency component. The contribution is used as a weighting coefficient and superimposed with the frequency of the corresponding intrinsic mode function to obtain the traffic flow fluctuation period considering non-stationary characteristics.

[0020] Empirical mode decomposition (EMD) is performed on the traffic flow feature sampling sequence through iterative filtering to obtain multiple intrinsic mode functions (EMFs) representing different frequency characteristics. The EMFs are then converted into analytic signals using Hilbert transform, and instantaneous frequency and amplitude are extracted from these analytic signals, including:

[0021] The upper and lower envelopes of the traffic flow sampling feature sequence are calculated by cubic spline interpolation, and the arithmetic mean of the upper and lower envelopes is taken as the mean envelope. The mean envelope is subtracted from the traffic flow sampling feature sequence to obtain the current iteration sequence. The upper envelope calculation, lower envelope calculation, and mean envelope calculation are repeatedly performed on the current iteration sequence to obtain the iterative component sequence.

[0022] Calculate the sum of squares of the differences between the iterative component sequence and the current iterative sequence, divide the sum of squares by the sum of squares of the current iterative sequence to obtain the standard deviation, and when the standard deviation is less than the deviation threshold, determine the iterative component sequence as the first intrinsic mode function;

[0023] Subtracting the first intrinsic mode function from the traffic flow feature time series sequence yields the first residual sequence. When the first residual sequence is a non-monotonic function, the upper envelope calculation, lower envelope calculation, mean envelope calculation, and iterative component sequence calculation are repeatedly performed on the first residual sequence to obtain the second intrinsic mode function.

[0024] Perform Hilbert transforms on the first eigenmode function and the second eigenmode function respectively, and construct an analytic signal in complex form from each eigenmode function and its corresponding Hilbert transform result;

[0025] The phase function is obtained by calculating the arctangent value based on the real and imaginary parts of the analytic signal; the phase derivative is obtained by numerical differentiation of the phase function, and the instantaneous frequency sequence is obtained by dividing the phase derivative by 2π; the instantaneous amplitude sequence is obtained by calculating the square root of the sum of the squares of the real and imaginary parts of the analytic signal.

[0026] Establish a traffic flow intelligent control system based on fluid dynamics. Calculate traffic flow dynamics values ​​based on vehicle longitudinal spacing and traffic flow data, and combine this with traffic flow density to calculate traffic pressure. Calculate density propagation velocity based on the traffic pressure derivative, including:

[0027] Traffic flow density data and traffic flow speed data for three intervals are obtained. The partial derivatives of the traffic flow density data and the traffic flow speed data with respect to time are calculated to obtain the traffic flow density change rate and the traffic flow speed change rate.

[0028] The longitudinal spacing data of adjacent vehicles within each interval is obtained. The difference between the longitudinal spacing and the safe following distance threshold is divided by the square of the longitudinal spacing to obtain the distance influence factor. The traffic flow density change rate, the traffic flow speed change rate and the distance influence factor are weighted and aggregated to obtain the traffic flow dynamics value.

[0029] Calculate the traffic flow density difference between adjacent intervals, substitute the traffic flow density difference into the hyperbolic tangent function to obtain the interval coupling factor, multiply the interval coupling factor, the traffic flow dynamics value and the power function of the current interval traffic flow density to obtain the traffic pressure; calculate the square root of the derivative of the traffic pressure with respect to the traffic flow density, and multiply it by the ratio of the interval length and the road width to obtain the density propagation speed.

[0030] Based on the density propagation speed, the target capacity value for each interval is determined using the entropy weight method. The optimal opening sequence is calculated by maximizing the target capacity value, and the opening strategy is dynamically corrected based on the principle of traffic flow conservation, including:

[0031] The three indicators of traffic flow density, average speed and number of lanes in each interval are normalized to obtain the normalized value of each indicator. The normalized value of each indicator is multiplied by its natural logarithm and then added together to obtain the information entropy of each interval.

[0032] Based on the density propagation speed, the product of traffic flow density and average speed in adjacent intervals is weighted and calculated to obtain the traffic flow density difference at the boundary of each interval. The traffic flow density difference is used as the flow conservation evaluation factor. The information entropy of each interval, the flow conservation evaluation factor, and the real-time flow of the interval are multiplied and accumulated in the time dimension to obtain the traffic capacity target value.

[0033] The spatial gradient is obtained by calculating the rate of change of the interval flow with distance, and the temporal gradient is obtained by calculating the rate of change of the interval flow with time. The spatial gradient and the temporal gradient are added together to obtain the flow conservation constraint value. The flow conservation constraint value, the upper limit of the interval capacity, and the upper limit of the congestion density are used as constraints, and the opening sequence corresponding to maximizing the traffic capacity target value is taken as the optimal opening sequence.

[0034] The difference between the inlet flow and the outlet flow of the interval is calculated to obtain the flow difference value. The flow difference value is multiplied by the current interval opening sequence to obtain the corrected opening sequence. The corrected opening sequence is used as the interval opening control command for the next time period.

[0035] A second aspect of the present invention provides a decision-making system for whether to open emergency lanes on highways under high traffic conditions, comprising:

[0036] The first unit is used to obtain vehicle driving data in the upstream, midstream and downstream of the target road segment during the target time period and to calculate the traffic flow index of each section.

[0037] The second unit is used to map vehicle driving data from the upstream, midstream and downstream sections to a high-dimensional feature space, calculate the feature correlation between the features of each section through tensor decomposition, resample the feature vectors of each section after mapping, and extract the frequency domain features and fluctuation period of traffic flow based on empirical mode decomposition and Fourier transform.

[0038] The third unit is used to normalize the feature correlation degree, the frequency domain feature and the fluctuation period and then perform a weighted summation. Based on the comparison between the weighted summation result and the preset opening threshold, it is determined whether to open the emergency lane.

[0039] The fourth unit is used to establish a traffic flow intelligent control system based on fluid dynamics if the emergency lane is opened. It calculates the traffic flow dynamics value based on the longitudinal spacing of vehicles and traffic flow data, and calculates the traffic pressure based on the traffic flow density. It calculates the density propagation speed based on the traffic pressure derivative. Based on the density propagation speed, it determines the traffic capacity target value of each section by combining the entropy weight method. It calculates the optimal opening sequence by maximizing the traffic capacity target value, and dynamically corrects the opening strategy based on the principle of traffic flow conservation.

[0040] The fifth unit is used to send an emergency lane opening instruction to the traffic management system based on the output results of the intelligent traffic flow control system; monitor the traffic flow indicators of each section of the open road segment in real time; and send an emergency lane closing instruction when the traffic flow indicators are detected to drop below the preset closing threshold.

[0041] A third aspect of the present invention,

[0042] An electronic device is provided, comprising:

[0043] processor;

[0044] Memory used to store processor-executable instructions;

[0045] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0046] Fourth aspect of the embodiments of the present invention,

[0047] A computer-readable storage medium is provided, having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0048] The beneficial effects of this application are as follows:

[0049] By performing high-dimensional feature mapping and tensor decomposition on vehicle driving data in the upstream, midstream, and downstream sections, and combining empirical mode decomposition and Fourier transform to extract traffic flow fluctuation features, accurate perception and prediction of traffic flow changes are achieved, improving the scientific nature and timeliness of emergency lane opening decisions.

[0050] A traffic flow intelligent control system was constructed based on the principles of fluid dynamics. By calculating the dynamic values ​​of traffic flow, traffic pressure, and density propagation speed, and combining the entropy weight method to determine the optimal opening sequence, the system achieves efficient allocation of traffic resources and significantly improves road capacity and traffic flow management efficiency.

[0051] By adopting a dynamic monitoring and adaptive adjustment mechanism, the opening strategy is corrected and the closing timing is determined based on real-time traffic flow indicators, which ensures the flexibility and safety of emergency lane management, effectively alleviates traffic congestion under high traffic conditions, and improves the utilization rate of road resources. Attached Figure Description

[0052] Figure 1 This is a flowchart illustrating the decision-making method for whether to open the emergency lane on a highway under high traffic conditions, according to an embodiment of the present invention.

[0053] Figure 2 This is a schematic diagram of the system architecture for calculating the correlation degree of interval features. Detailed Implementation

[0054] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0055] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0056] Figure 1 This is a flowchart illustrating the decision-making method for whether to open the emergency lane on a highway under high traffic conditions, as described in an embodiment of the present invention. Figure 1 As shown, the method includes:

[0057] Obtain vehicle driving data for the upstream, midstream, and downstream sections of the target road segment during the target time period, and calculate traffic flow indicators for each section;

[0058] Vehicle driving data from the upstream, midstream, and downstream sections are mapped to a high-dimensional feature space. The feature correlation between each section is calculated through tensor decomposition. The feature vectors of each section after mapping are resampled, and the frequency domain features and fluctuation period of traffic flow fluctuations are extracted based on empirical mode decomposition and Fourier transform.

[0059] The feature correlation degree, the frequency domain feature and the fluctuation period are normalized and then weighted and summed. Based on the comparison between the weighted summation result and the preset opening threshold, it is determined whether to open the emergency lane.

[0060] If the emergency lane is opened, a traffic flow intelligent control system based on fluid dynamics is established. The system calculates the traffic flow dynamics value based on the longitudinal spacing of vehicles and traffic flow data, and calculates the traffic pressure based on the traffic flow density. The density propagation speed is calculated based on the traffic pressure derivative. Based on the density propagation speed, the target capacity value of each section is determined by combining the entropy weight method. The optimal opening sequence is calculated by maximizing the target capacity value, and the opening strategy is dynamically corrected based on the principle of traffic flow conservation.

[0061] Based on the output of the intelligent traffic flow control system, an emergency lane opening command is sent to the traffic management system; the traffic flow index of each section of the open road segment is monitored in real time, and when the traffic flow index is detected to drop below the preset closing threshold, an emergency lane closing command is sent.

[0062] In a specific embodiment of the present invention, when deploying radar-guided surveillance equipment on highways, three key locations (marked as points A, B, and C, respectively) are typically selected for installation. This deployment method can capture the speed changes of vehicles throughout the entire road segment.

[0063] The sampling system of the radar-based equipment consists of multiple hardware components, including a millimeter-wave radar, a high-definition camera, an image processing unit, a communication module, and a power supply system. The millimeter-wave radar operates at a frequency of 77 GHz, with a maximum speed measurement range of 250 km / h and a speed measurement accuracy of ±1 km / h. The high-definition camera uses a 1920×1080 resolution and is equipped with an infrared fill light to ensure all-weather license plate recognition. The image processing unit uses an embedded system with a dedicated license plate recognition algorithm, achieving an accuracy rate of up to 98%. The data acquisition cycle is set to 100 ms, and each acquisition includes a timestamp, vehicle speed, license plate number, and vehicle location information.

[0064] The sampling equipment is installed on a gantry above the highway, maintaining a vertical distance of 6 meters from the ground and a 15-degree downward angle to ensure clear capture of information about the top and front of vehicles. During installation, the radar's speed measurement direction needs to be precisely calibrated to align with the vehicle's direction of travel, minimizing measurement errors. Simultaneously, the camera's focal length needs to be adjusted to a suitable range, typically 8-12mm, to obtain clear license plate images.

[0065] The data processing workflow for radar-based sampling includes steps such as data acquisition, license plate recognition, speed extraction, vehicle matching, and data analysis. First, when a vehicle passes through the detection area, the radar sensor measures the vehicle's instantaneous speed, while the camera captures an image of the vehicle. The image data undergoes preprocessing, including grayscale conversion, contrast enhancement, and edge detection, and then the license plate number is extracted using a deep learning license plate recognition algorithm. The extracted license plate information and speed data are linked to the same timestamp, forming a single-point data record.

[0066] Vehicle matching is a key technology for cross-regional tracking. The system establishes a dynamic matching pool; when a new license plate is detected, a new vehicle record is created. Subsequently, when the same license plate is detected again at points B and C, the system adds the new speed data to the corresponding vehicle record. Considering potential recognition errors, the system uses a fuzzy matching algorithm, allowing for a maximum difference of one character in the license plate. The matching process also verifies the reasonableness of the vehicle's travel time; for example, the time from point A to point B should conform to a reasonable travel time under current road conditions.

[0067] In practical applications, the matching success rate may be affected when traffic volume is high. To address this, the system is designed with an adaptive matching window, which automatically adjusts the matching time window size based on the current average vehicle speed. For example, when the average vehicle speed is 100 km / h, if the distance between points A and B is 2 km, the matching window is set to 72-144 seconds (considering speed variation). This improves matching accuracy while reducing computational resource consumption.

[0068] Data storage employs a distributed architecture, with each sampling point having a local cache storing the raw data from the most recent 4 hours. Complete vehicle trajectory data after matching is uploaded to a cloud data center and stored using a time-series database, supporting fast time-range queries. The data structure is designed with fields for: license plate number, vehicle type, time of passage through point A, speed at point A, time of passage through point B, speed at point B, time of passage through point C, and speed at point C.

[0069] The precise vehicle speed change data obtained from radar-based sampling provides a solid foundation for traffic flow characteristic correlation analysis. By analyzing the speed changes of the same vehicle at points A, B, and C, the acceleration trend of the vehicle can be calculated. If the average acceleration is negative and the absolute value is greater than 0.5 m / s², it indicates that there may be traffic congestion on that road segment. The system will count the acceleration changes of vehicles passing through this road segment within 20 minutes. If more than 80% of the vehicles show a deceleration trend, a warning signal for opening the emergency lane will be triggered.

[0070] Based on precise single-vehicle tracking data, the system can calculate more accurate interval feature correlation. Compared with traditional methods, directly tracking the behavioral changes of the same batch of vehicles in different intervals can eliminate interference caused by differences in vehicle composition. For example, in a real-world test, the system tracked complete data of 500 vehicles passing through points A, B, and C. Analysis showed that the average vehicle speed in the upstream interval was 98 km / h, dropping to 75 km / h in the midstream interval, and further decreasing to 62 km / h in the downstream interval, clearly demonstrating the propagation process of traffic congestion.

[0071] The radar-based sampling technology also significantly improves the accuracy of traffic flow fluctuation analysis. By recording the speed changes of the same vehicle, the system can directly calculate the acceleration distribution of the traffic flow without relying on traditional average speed difference estimation. In one application scenario, the system analyzed complete trajectory data of 2,000 vehicles during peak hours, identifying a traffic flow fluctuation cycle of 28 minutes, which is far more accurate than the 35-minute estimate by traditional methods, providing a precise basis for emergency lane opening decisions.

[0072] Vehicle trajectory data is also used to calculate more accurate inter-vehicle distances, serving as input for traffic pressure calculations. By continuously tracking vehicle positions, the system can calculate real-time vehicle distribution density and longitudinal spacing changes. Real-world data shows that during periods of sudden traffic congestion, the average longitudinal spacing between vehicles rapidly decreases from 78 meters to 32 meters, with a density propagation speed reaching 13.5 meters per second, significantly higher than the normal 8.2 meters per second.

[0073] Figure 2 This is a schematic diagram of the system architecture for calculating the correlation degree of features between intervals. In one optional implementation, vehicle driving data from the upstream, midstream, and downstream intervals are mapped to a high-dimensional feature space, and the feature correlation degree between the features of each interval is calculated through tensor decomposition, including:

[0074] Obtain vehicle speed data for the upstream, midstream, and downstream sections, calculate the standard deviation of vehicle speed for each section, and divide it by the corresponding mean vehicle speed to obtain the traffic flow fluctuation intensity index for each section; when the traffic flow fluctuation intensity index is greater than the fluctuation threshold, multiply the preset time window length by the traffic flow fluctuation intensity index.

[0075] The feature vectors of each interval are extracted within the adjusted time window. The feature vectors are then input into a random projection matrix for local sensitive hashing. The elements of the random projection matrix follow a standard normal distribution. Through multiple mapping iterations, the mapped feature vectors are obtained.

[0076] The mapped feature vectors are organized into third-order tensors in the interval and time dimensions. The tensor reconstruction error is calculated. When the reconstruction error is less than the reconstruction error threshold, the current decomposition rank is determined to be the optimal decomposition rank. The optimal decomposition rank is used to perform kernel tensor decomposition to obtain the interval mode decomposition matrix and the time mode decomposition matrix.

[0077] The static correlation degree between intervals is obtained by performing an inner product operation on the eigenvectors of adjacent intervals in the interval mode decomposition matrix and dividing them by the corresponding modulus product. The time mode decomposition matrix is ​​divided into current feature data and historical feature data according to the time window, and the dynamic correlation degree between the current feature data and the historical feature data is calculated. The static correlation degree and the dynamic correlation degree are weighted and summed to obtain the feature correlation degree between each interval feature.

[0078] This invention provides a method for calculating the correlation degree of interval features through tensor decomposition, which realizes high-dimensional mapping and feature correlation analysis of vehicle driving data in the upstream, midstream and downstream intervals.

[0079] In a specific embodiment of the invention, vehicle speed data for the upstream, midstream, and downstream sections are acquired. This vehicle speed data can be collected by vehicle detectors installed on the road, recorded once per minute for 24 consecutive hours, resulting in 1440 vehicle speed data points for each section. For the vehicle speed data of each section, the standard deviation is calculated and divided by the corresponding mean vehicle speed to obtain a traffic flow fluctuation intensity index. For example, if the mean vehicle speed in the upstream section is 60 km / h and the standard deviation is 15 km / h during a certain period, then the traffic flow fluctuation intensity index is 0.25. When this index is greater than a preset fluctuation threshold (e.g., 0.2), the basic time window length (e.g., 10 minutes) is multiplied by the traffic flow fluctuation intensity index for adjustment. In this example, the adjusted time window is 10 × 0.25 = 2.5 minutes, which is rounded up to 3 minutes.

[0080] Feature vectors are extracted for each interval within the adjusted time window. Each feature vector contains five dimensions: average vehicle speed, standard deviation of vehicle speed, traffic flow, vehicle density, and occupancy rate. For example, the feature vector for the upstream interval within a 3-minute window is [65 km / h, 12 km / h, 150 vehicles / hour, 25 vehicles / km, 0.15], where the occupancy rate of 0.15 indicates that the detector is occupied by passing vehicles for 15% of the total detection time. These original feature vectors are then input into a random projection matrix for local sensitive hashing. The elements of this random projection matrix follow a standard normal distribution N(0,1), and the matrix size is 10×5, mapping the 5-dimensional features to a 10-dimensional space. For example, by generating a random projection matrix R, the feature vector v is mapped to R×v to obtain a new 10-dimensional feature vector. To improve the stability of the mapping, five mapping iterations are performed, each using a different random matrix, and the average value is taken as the final mapping result.

[0081] The feature vectors of the three mapped intervals are organized into a third-order tensor X in the interval and time dimensions, with dimensions of [3 intervals × 10 feature dimensions × 480 time windows] (assuming there are 480 3-minute time windows within 24 hours). Before decomposing this tensor, the optimal decomposition rank needs to be determined. Starting with a rank value r=1, the rank value is gradually increased while calculating the reconstruction error. The reconstruction error is defined as the Frobenius norm difference between the original tensor X and the reconstructed tensor X'. When the rank value increases from r to r+1 and the reconstruction error decreases by less than 1% (as the reconstruction error threshold), then r is determined to be the optimal decomposition rank. In practical applications, the optimal decomposition rank usually falls between 3 and 5; for example, r=4.

[0082] Using the determined optimal decomposition rank r=4, the third-order tensor X is decomposed using Tucker decomposition, yielding a kernel tensor G (dimension [4×4×4]) and three factor matrices: interval mode decomposition matrix A (dimension [3×4]), eigenmode decomposition matrix B (dimension [10×4]), and time mode decomposition matrix C (dimension [480×4]). The row vectors in interval mode decomposition matrix A represent the representation of each interval in the low-dimensional latent space, with each row corresponding to one interval.

[0083] The static correlation degree between intervals is obtained by performing an inner product operation on the eigenvectors of adjacent intervals in the interval mode decomposition matrix A and dividing by the corresponding product of magnitudes. For example, if the upstream interval vector is [0.5, 0.3, 0.4, 0.6] and the midstream interval vector is [0.4, 0.5, 0.3, 0.5], then their inner product is 0.5×0.4+0.3×0.5+0.4×0.3+0.6×0.5=0.69, and the magnitudes of the two vectors are 0.92 and 0.87 respectively, so the static correlation degree is 0.69 / (0.92×0.87)=0.86. Similarly, the static correlation degree between the midstream and downstream intervals is calculated.

[0084] The temporal modality decomposition matrix C is divided into current feature data (the most recent 30 minutes, i.e., 10 time windows) and historical feature data (data from the first 30 minutes of the same time period within the same day). The mean cosine similarity between the current feature data and the historical feature data is calculated to obtain the dynamic correlation. For example, if the cosine similarity between the decomposed features of the current time period (17:00-17:30) and the decomposed features of the current time period (16:30-17:00) is 0.78, then the dynamic correlation of this time period is 0.78.

[0085] The static and dynamic correlation degrees are weighted and summed to obtain the feature correlation degree between each interval. The weights can be set to 0.7 for static correlation degree and 0.3 for dynamic correlation degree. For example, the feature correlation degree between the upstream and midstream intervals is 0.86 × 0.7 + 0.78 × 0.3 = 0.836. This feature correlation degree can be used to quantitatively assess the traffic condition propagation relationship between different road intervals, providing a basis for traffic management and prediction.

[0086] This method effectively captures the traffic state association patterns in different sections of the road network through high-dimensional feature mapping and tensor decomposition techniques. It can adapt to the feature extraction needs under fluctuating traffic flow conditions and improve the accuracy and reliability of feature association analysis.

[0087] In one optional implementation, the mapped feature vectors of each interval are resampled, and the frequency domain features and fluctuation period of traffic flow fluctuations are extracted based on empirical mode decomposition and Fourier transform, including:

[0088] The feature vectors of each mapped interval are discretized over time according to the sampling interval to obtain the traffic flow feature sampling sequence.

[0089] Empirical mode decomposition (EMD) is performed on the traffic flow feature sampling sequence through iterative screening to obtain multiple intrinsic mode functions (IMFs) representing different frequency characteristics. The IMFs are then converted into analytical signals using Hilbert transform, and instantaneous frequency and instantaneous amplitude are extracted from the analytical signals.

[0090] The discrete Fourier transform of the multiple intrinsic mode functions is performed to obtain the spectral components of each intrinsic mode function. The power spectral density is obtained by weighted summation of the spectral components. The dominant frequency component is extracted from the power spectral density, and the phase accumulation characteristics are calculated by combining the spectral phase of each intrinsic mode function to obtain the frequency domain characteristics of traffic flow fluctuation.

[0091] The frequency deviation between the instantaneous frequency of each intrinsic mode function and the dominant frequency component is calculated. The frequency similarity is obtained by Gaussian weighting. The root mean square of the instantaneous amplitude of each intrinsic mode function is calculated as the energy value. The frequency similarity is multiplied by the energy value to obtain the contribution of each intrinsic mode function to the dominant frequency component. The contribution is used as a weighting coefficient and superimposed with the frequency of the corresponding intrinsic mode function to obtain the traffic flow fluctuation period considering non-stationary characteristics.

[0092] In practical applications, this invention extracts traffic flow fluctuation characteristics by resampling and analyzing the mapped interval feature vectors. Specifically, the mapped interval feature vectors are discretized at fixed sampling intervals. For example, when collecting traffic flow data on a highway, the original data collection interval is 5 minutes. This can be used as the basic sampling interval to discretize the feature vectors of each interval, forming a traffic flow feature sampling sequence. Assume that traffic flow data at 96 time points are collected over 8 hours for a certain road segment; these data constitute the traffic flow feature sampling sequence to be analyzed.

[0093] Empirical Mode Decomposition (EMD) is performed on the generated traffic flow feature sampling sequence. This decomposition process employs an iterative selection method: first, all extreme points in the sequence are identified; then, cubic spline interpolation is used to connect all maxima to form an upper envelope, and similarly, all minima are connected to form a lower envelope. The average of the upper and lower envelopes is calculated, and this average is subtracted from the original sequence. The resulting difference sequence is then checked to see if it satisfies the conditions for an intrinsic mode function (EMF). If not, the above steps are repeated until the conditions are met, yielding the first EMF. This EMF is subtracted from the original sequence to obtain the residual. The residual is then used as a new sequence to be processed, and the above steps are repeated until the residual becomes a monotonic function or its amplitude is sufficiently small. In this way, for example, traffic flow data at the aforementioned 96 time points can be decomposed to obtain 5-8 different EMFs, each representing traffic flow fluctuations at different frequencies.

[0094] Each obtained intrinsic mode function (IMF) is subjected to a Hilbert transform to convert it into an analytic signal. Specifically, the Hilbert transform is performed on each IMF to obtain its corresponding imaginary component. The original IMF is used as the real part, and the imaginary component together constitute the analytic signal. From this analytic signal, instantaneous frequency and instantaneous amplitude information can be extracted. The instantaneous frequency is calculated as the time derivative of the phase of the analytic signal, and the instantaneous amplitude is the magnitude of the analytic signal. For example, after performing a Hilbert transform on a certain IMF, 96 instantaneous frequency values ​​corresponding to different time points can be obtained, ranging from 0.002 to 0.01 Hz. The instantaneous amplitude reflects the fluctuation intensity of traffic flow at that frequency.

[0095] Next, the multiple intrinsic mode functions (IMFs) are subjected to Discrete Fourier Transform (DFT) to obtain frequency domain features. DFT is applied to each IMF to obtain frequency components and their corresponding amplitude and phase information. In practical applications, DFT is performed on each of the five IMFs, revealing that the main frequency components are distributed in the range of 0.001-0.05 Hz. These spectral components are weighted and summed to obtain the power spectral density, where the weights are determined based on the energy proportions of each IMF. The dominant frequency component, i.e., the frequency corresponding to the peak value of the power spectral density, is extracted from the power spectral density. Combined with the phase information of each IMF at this frequency, the phase accumulation feature is calculated as the frequency domain feature of traffic flow fluctuations. For example, the dominant frequency component of a certain road segment is 0.0033 Hz, corresponding to a period of approximately 5 hours.

[0096] To account for the non-stationary characteristics of traffic flow, the contribution of each intrinsic mode function (IMF) to the dominant frequency component needs to be calculated. First, the frequency deviation between the instantaneous frequency of each IMF and the dominant frequency component is calculated. A smaller frequency deviation indicates a stronger correlation between the IMF and the dominant frequency component. Frequency similarity is calculated using a Gaussian weighted average. The standard deviation of the Gaussian function can be set to 10% of the dominant frequency component value, ensuring that IMFs with frequency deviations within this range receive higher similarity values. Simultaneously, the root mean square (RMS) value of the instantaneous amplitude of each IMF is calculated as its energy value. Multiplying the frequency similarity by the energy value yields the contribution of each IMF to the dominant frequency component. For example, for a dominant frequency component of 0.0033 Hz, the average instantaneous frequency of the first IMF is 0.0035 Hz, with a small frequency deviation. Its RMS instantaneous amplitude is 25, and the calculated frequency similarity is 0.9, resulting in a contribution of 22.5.

[0097] The contribution of each intrinsic mode function (IMF) is used as a weighting coefficient, and then weighted and superimposed with the corresponding IMF frequency to obtain the traffic flow fluctuation period considering non-stationary characteristics. In practical applications, if the contribution of the five IMFs are 22.5, 15.8, 8.2, 4.5, and 1.2, respectively, and the corresponding principal frequencies are 0.0035, 0.0031, 0.0042, 0.0027, and 0.0050 Hz, then the weighted average frequency is 0.0034 Hz, corresponding to a traffic flow fluctuation period of approximately 4.9 hours. This more accurately reflects the actual fluctuation characteristics of traffic flow than the 5-hour period calculated solely based on the principal frequency component.

[0098] In one optional implementation, empirical mode decomposition is performed on the traffic flow feature sampling sequence through iterative filtering to obtain multiple intrinsic mode functions (IMFs) representing different frequency characteristics; the IMFs are then converted into analytic signals using Hilbert transform, and instantaneous frequency and instantaneous amplitude are extracted from the analytic signals, including:

[0099] The upper and lower envelopes of the traffic flow sampling feature sequence are calculated by cubic spline interpolation, and the arithmetic mean of the upper and lower envelopes is taken as the mean envelope. The mean envelope is subtracted from the traffic flow sampling feature sequence to obtain the current iteration sequence. The upper envelope calculation, lower envelope calculation, and mean envelope calculation are repeatedly performed on the current iteration sequence to obtain the iterative component sequence.

[0100] Calculate the sum of squares of the differences between the iterative component sequence and the current iterative sequence, divide the sum of squares by the sum of squares of the current iterative sequence to obtain the standard deviation, and when the standard deviation is less than the deviation threshold, determine the iterative component sequence as the first intrinsic mode function;

[0101] Subtracting the first intrinsic mode function from the traffic flow feature time series sequence yields the first residual sequence. When the first residual sequence is a non-monotonic function, the upper envelope calculation, lower envelope calculation, mean envelope calculation, and iterative component sequence calculation are repeatedly performed on the first residual sequence to obtain the second intrinsic mode function.

[0102] Perform Hilbert transforms on the first eigenmode function and the second eigenmode function respectively, and construct an analytic signal in complex form from each eigenmode function and its corresponding Hilbert transform result;

[0103] The phase function is obtained by calculating the arctangent value based on the real and imaginary parts of the analytic signal; the phase derivative is obtained by numerical differentiation of the phase function, and the instantaneous frequency sequence is obtained by dividing the phase derivative by 2π; the instantaneous amplitude sequence is obtained by calculating the square root of the sum of the squares of the real and imaginary parts of the analytic signal.

[0104] In a specific embodiment, a traffic flow feature sampling sequence is obtained. This sequence can be traffic flow data collected from intersections, such as the number of vehicles passing through every 5 minutes, forming a daily traffic flow sequence X(t)={x(1), x(2), ..., x(288)} of length 288. Empirical mode decomposition is performed on this sampling sequence, and intrinsic mode functions (IMFs) are identified through an iterative screening process.

[0105] During the iterative selection process, cubic spline interpolation is performed on the traffic flow feature sampling sequence X(t) to determine the set of local maxima {(t)}. max1 , x max1 ), (t max2 , x max2 ), ..., (t maxp , x maxp Construct the upper envelope E up (t). The set of local minimum points {(t)} is determined in the same way. min1 , x min1 ), (t min2 , x min2 ), ..., (t minq , x minq Construct the lower envelope E down (t). Calculate the mean envelope m(t) of the upper and lower envelopes = (E up (t)+E down (t)) / 2, subtract the mean envelope from the original sequence X(t) to get h(t)=X(t)-m(t) as the current iteration sequence.

[0106] Repeat the above envelope calculation process for h(t) to obtain a new mean envelope m1(t), and calculate h1(t) = h(t) - m1(t). Continue this iterative process, and after the k-th iteration, obtain h. k (t). To determine whether the iteration is complete, calculate the standard deviation SD = Σ[(h k (t)-h k-1 (t)) 2 ] / Σ[h k-1 (t) 2 The iteration terminates when SD is less than a preset threshold (e.g., 0.2), and h is set to... k (t) is determined as the first IMF component c1(t).

[0107] Taking actual data as an example, the traffic flow data collected at a certain intersection in one day was filtered through 10 iterations to obtain the first IMF component c1(t), whose SD value is 0.19, which is less than the set threshold of 0.2. This IMF component represents the high-frequency change characteristics of traffic flow and reflects the rapid fluctuations during morning and evening peak hours.

[0108] Subtracting the first IMF from the original sequence yields the residual r1(t) = X(t) - c1(t). Checking if r1(t) is monotonic, if not, repeating the filtering process with r1(t) as the new input sequence yields the second IMF component c2(t). In the example data, r1(t) exhibits a non-monotonic change; after 8 iterations, c2(t) is obtained with an SD value of 0.18. This process continues until the residual r1(t) is obtained. n (t) continues until it becomes a monotonic function or has only one extreme point.

[0109] In this example, five IMF components {c1(t), c2(t), ..., c5(t)} and a residual r5(t) are obtained. c1(t) reflects rapid changes over a 5-10 minute period, c2(t) reflects changes over a 20-30 minute period, c3(t) reflects changes over a 1-2 hour period, c4(t) and c5(t) reflect changes over semi-diurnal and full-diurnal scales, respectively, and r5(t) represents the overall trend.

[0110] Perform a Hilbert transform on each IMF component to extract instantaneous features. Taking c1(t) as an example, its Hilbert transform H[c1(t)] is calculated as follows: Perform a discrete Fourier transform on c1(t) to obtain the frequency domain representation, multiply the negative frequency components by -i, multiply the positive frequency components by i, and keep the zero frequency component unchanged, and then perform an inverse Fourier transform to obtain H[c1(t)].

[0111] Construct the analytic signal z1(t) = c1(t) + iH[c1(t)] by combining c1(t) with its Hilbert transform H[c1(t)]. Calculate the phase function θ1(t) = arctan(H[c1(t)] / c1(t)) based on the analytic signal, noting the need to handle quadrant issues to obtain continuous phase. Perform a central difference numerical differentiation on θ1(t): dθ1(t) / dt ≈ [θ1(t+1) - θ1(t-1)] / 2, and divide the result by 2π to obtain the instantaneous frequency f1(t). Calculate the amplitude of the analytic signal a1(t) = sqrt(c1(t)). 2 +H[c1(t)] 2 The instantaneous amplitude is obtained.

[0112] In the example, the average instantaneous frequency of c1(t) during the morning peak from 7:30 to 9:00 is 0.1 times / minute, and the average instantaneous amplitude is 32 vehicles / 5 minutes; during the evening peak from 17:00 to 19:00, the instantaneous frequency is 0.12 times / minute, and the instantaneous amplitude is 38 vehicles / 5 minutes, indicating that the traffic flow fluctuations are more intense during the evening peak.

[0113] The same processing was performed on c2(t) to c5(t) to obtain their respective instantaneous frequencies and instantaneous amplitudes. By comparing the instantaneous characteristics of different IMF components, it was found that c3(t) exhibits obvious bimodal characteristics on weekdays, with instantaneous amplitudes reaching 25 and 30 vehicles / 5 minutes during the morning and evening peak hours, respectively, while exhibiting unimodal characteristics on weekends.

[0114] This method, through multi-scale decomposition and instantaneous characteristic analysis of traffic flow features, can accurately capture the time-varying characteristics of traffic flow, providing data support for traffic management and signal control. Especially in responding to emergencies, by monitoring abnormal changes in the instantaneous amplitude of high-frequency IMF components, rapid assessment and early warning of traffic conditions can be achieved.

[0115] In one optional implementation, a traffic flow intelligent control system based on fluid dynamics is established. This system calculates traffic flow dynamics values ​​based on vehicle longitudinal spacing and traffic flow data, and calculates traffic pressure in conjunction with traffic flow density. It also calculates density propagation velocity based on the traffic pressure derivative, including:

[0116] Traffic flow density data and traffic flow speed data for three intervals are obtained. The partial derivatives of the traffic flow density data and the traffic flow speed data with respect to time are calculated to obtain the traffic flow density change rate and the traffic flow speed change rate.

[0117] The longitudinal spacing data of adjacent vehicles within each interval is obtained. The difference between the longitudinal spacing and the safe following distance threshold is divided by the square of the longitudinal spacing to obtain the distance influence factor. The traffic flow density change rate, the traffic flow speed change rate and the distance influence factor are weighted and aggregated to obtain the traffic flow dynamics value.

[0118] Calculate the traffic flow density difference between adjacent intervals, substitute the traffic flow density difference into the hyperbolic tangent function to obtain the interval coupling factor, multiply the interval coupling factor, the traffic flow dynamics value and the power function of the current interval traffic flow density to obtain the traffic pressure; calculate the square root of the derivative of the traffic pressure with respect to the traffic flow density, and multiply it by the ratio of the interval length and the road width to obtain the density propagation speed.

[0119] This embodiment provides a traffic flow intelligent control system based on fluid dynamics. The system analyzes vehicle longitudinal spacing and traffic flow data to calculate traffic flow dynamics values ​​and combines this with traffic flow density to calculate traffic pressure. Finally, it calculates the density propagation velocity based on the traffic pressure derivative, thereby achieving intelligent control of traffic flow.

[0120] In practical implementation, the system acquires traffic density and speed data for three sections using sensors installed on the road. Taking the first section as an example, assuming the traffic density at time t1 is 85 vehicles / km and at time t2 is 92 vehicles / km, with a time interval of 10 minutes, the traffic density change rate for this section is calculated to be (92-85) / (10×60) seconds = 0.0117 vehicles / km / second. Similarly, if the traffic speed in this section is 45 km / h at time t1 and 42 km / h at time t2, the traffic speed change rate is (42-45) / (10×60) seconds = -0.005 km / h / second. The system performs similar calculations for each of the three sections to obtain the traffic density and speed change rates for each section.

[0121] Next, the system acquires the longitudinal distance data of adjacent vehicles within each interval through in-vehicle equipment or road monitoring cameras. Taking two adjacent vehicles in the first interval as an example, if the measured longitudinal distance between them is 25 meters, and the safe following distance threshold calculated based on the current vehicle speed is 35 meters, then the difference between the longitudinal distance and the safe distance threshold is 25 - 35 = -10 meters. Dividing this difference by the square of the longitudinal distance yields the distance influence factor, i.e., -10 / (25 × 25) = -0.016 meters. -1 Calculate the distance influence factor for all adjacent vehicle pairs within the interval, and take the average value as the distance influence factor for that interval.

[0122] The system weights and aggregates the traffic density change rate, traffic speed change rate, and distance influence factor to calculate the traffic flow dynamics value. The weights are set to 0.4, 0.35, and 0.25. Taking the first interval as an example, the traffic flow dynamics value = 0.4 × 0.0117 - 0.35 × 0.005 + 0.25 × (-0.016) = -0.00107. Here, a negative value indicates a slight deterioration in traffic conditions within that interval.

[0123] After calculating the traffic flow dynamics values, the system further calculates the traffic density difference between adjacent intervals. Assuming the traffic density in the first interval is 92 vehicles / km and in the second interval is 78 vehicles / km, the density difference is 92 - 78 = 14 vehicles / km. Substituting this difference into the hyperbolic tangent function yields the interval coupling factor. Specifically, the density difference is first divided by a standardization factor (e.g., 20), resulting in 0.7, and then tanh(0.7)≈0.604 is calculated. This value indicates a moderate degree of correlation between the traffic conditions of the two intervals.

[0124] The system multiplies the interval coupling factor, the traffic flow dynamics value, and the power function of the current interval traffic flow density to obtain the traffic pressure. Taking the first interval as an example, assuming the power exponent of the traffic flow density is 1.2, then the traffic pressure = 0.604 × (-0.0006) × (92^1.2) ≈ -0.0823. A negative value indicates that there is traffic pressure in this interval, causing the traffic flow to slow down.

[0125] The system calculates the square root of the derivative of traffic pressure with respect to traffic density and multiplies it by the ratio of the interval length to the road width to obtain the density propagation velocity. Assuming the derivative of traffic pressure with respect to traffic density is -0.25, the interval length is 500 meters, and the road width is 12 meters, then the density propagation velocity = (|-0.25|×(500 / 12)). 0.5 =3.227 m / s. This value represents the propagation speed of the traffic density wave in this interval, which the system can use to predict the propagation of traffic congestion.

[0126] Based on the calculated density propagation velocity, the system can predict the speed and direction of traffic congestion spread. When a sudden increase in traffic pressure and a high density propagation velocity are detected in a certain section, the system determines that traffic congestion has occurred and is rapidly propagating to adjacent sections. At this time, the system will issue a warning to the upstream section and advise vehicles to slow down or choose alternative routes through variable message signs, navigation systems, and other channels. Simultaneously, the system can automatically adjust traffic light timings based on the changing trends of traffic pressure and density propagation velocity, increasing the green light time in the congested direction to alleviate traffic pressure.

[0127] Through practical application verification, after implementing the system on a three-lane highway, the average vehicle speed during peak hours increased by 15%, travel time was reduced by 12%, and vehicle queue length was reduced by 20%. In addition, the system's response time to emergencies (such as traffic accidents) was shortened from the traditional 10-15 minutes to 3-5 minutes, significantly improving traffic management efficiency and road utilization.

[0128] The implementation of this system has not only improved road traffic efficiency and reduced traffic congestion, but also reduced vehicle idling time, fuel consumption, and exhaust emissions, resulting in significant social and environmental benefits. Furthermore, the system's algorithm can adjust parameters according to different road types, weather conditions, and traffic compositions, demonstrating strong adaptability and broad application prospects.

[0129] In one optional implementation, based on the density propagation speed, the target capacity value for each interval is determined using the entropy weight method. The optimal opening sequence is calculated by maximizing the target capacity value, and the opening strategy is dynamically corrected based on the principle of traffic flow conservation, including:

[0130] The three indicators of traffic flow density, average speed and number of lanes in each interval are normalized to obtain the normalized value of each indicator. The normalized value of each indicator is multiplied by its natural logarithm and then added together to obtain the information entropy of each interval.

[0131] Based on the density propagation speed, the product of traffic flow density and average speed in adjacent intervals is weighted and calculated to obtain the traffic flow density difference at the boundary of each interval. The traffic flow density difference is used as the flow conservation evaluation factor. The information entropy of each interval, the flow conservation evaluation factor, and the real-time flow of the interval are multiplied and accumulated in the time dimension to obtain the traffic capacity target value.

[0132] The spatial gradient is obtained by calculating the rate of change of the interval flow with distance, and the temporal gradient is obtained by calculating the rate of change of the interval flow with time. The spatial gradient and the temporal gradient are added together to obtain the flow conservation constraint value. The flow conservation constraint value, the upper limit of the interval capacity, and the upper limit of the congestion density are used as constraints, and the opening sequence corresponding to maximizing the traffic capacity target value is taken as the optimal opening sequence.

[0133] The difference between the inlet flow and the outlet flow of the interval is calculated to obtain the flow difference value. The flow difference value is multiplied by the current interval opening sequence to obtain the corrected opening sequence. The corrected opening sequence is used as the interval opening control command for the next time period.

[0134] This invention provides a method for determining a traffic capacity target value based on density propagation speed and entropy weight method. The optimal opening time sequence is calculated by maximizing the traffic capacity target value, and the opening strategy is dynamically corrected based on the principle of traffic flow conservation. The specific implementation method is as follows:

[0135] In its implementation, the system first acquires real-time traffic density, average speed, and number of lanes for each section of the road. For example, for a highway consisting of five sections, sensors installed on the road can acquire the traffic density (vehicles / km), average speed (km / h), and number of lanes for each section. The raw data acquired are: Section 1 (20 vehicles / km, 80 km / h, 3 lanes), Section 2 (25 vehicles / km, 75 km / h, 3 lanes), Section 3 (30 vehicles / km, 70 km / h, 2 lanes), Section 4 (28 vehicles / km, 65 km / h, 2 lanes), and Section 5 (15 vehicles / km, 90 km / h, 3 lanes).

[0136] Next, these three indicators are normalized to make them comparable. The normalization method is to divide each indicator value by its maximum value. For example, for traffic density, the maximum value is 30 vehicles / km, so the normalized values ​​are: Interval 1 (0.67), Interval 2 (0.83), Interval 3 (1.0), Interval 4 (0.93), and Interval 5 (0.5). Similarly, average vehicle speed and number of lanes are normalized.

[0137] After normalization, the information entropy of each interval is calculated. For each interval, the normalized value of each indicator is multiplied by its natural logarithm and then summed. For example, the information entropy of interval 1 is calculated as: 0.67×ln(0.67)+0.89×ln(0.89)+1.0×ln(1.0)=-0.27-0.10+0=-0.37. The information entropy of other intervals is calculated in the same way, resulting in interval 2 (-0.33), interval 3 (-0.26), interval 4 (-0.31), and interval 5 (-0.42).

[0138] Simultaneously, the system measures the density propagation velocity, which is the speed at which changes in traffic density propagate along the road. The density propagation velocity can be calculated from the density differences at consecutive time points. For example, at two adjacent time points (5 minutes apart), a density propagation velocity of 12 km / h between interval 1 and interval 2 indicates that the density change propagates from interval 1 to interval 2 at this speed.

[0139] Based on the density propagation speed, the system calculates the difference in traffic flow density at the boundaries of each interval, which serves as a traffic flow conservation evaluation factor. For example, the difference in traffic flow density at the boundary between interval 1 and interval 2 is calculated as: (20×80×0.3+25×75×0.7)-(25×75×0.3+20×80×0.7)=1792.5-1682.5=110. Here, 0.3 and 0.7 are weighting coefficients determined based on the density propagation speed.

[0140] The information entropy and flow conservation evaluation factor of each interval are multiplied by the real-time flow of the interval, and then summed over time to obtain the target capacity value. For example, assuming the real-time flow of interval 1 is 1600 vehicles / hour, the capacity assessment value of this interval in the current time period is: -0.37×(-5)×1600=2960. The assessment values ​​of all intervals in all time periods are summed to obtain the overall capacity target value.

[0141] To determine the optimal opening sequence, the system calculates the rate of change of traffic flow between intervals as a function of distance to obtain the spatial gradient. For example, if the traffic flow difference between interval 1 and interval 2 is 100 vehicles / hour and the distance difference is 1 kilometer, then the spatial gradient is 100 vehicles / hour / km. Similarly, the system calculates the rate of change of traffic flow between intervals as a function of time to obtain the temporal gradient. For example, if the traffic flow change in interval 1 between two adjacent time periods is 50 vehicles / hour and the time difference is 10 minutes, then the temporal gradient is 300 vehicles / hour / hour.

[0142] The spatial gradient and temporal gradient are added together to obtain the flow conservation constraint value. For example, the flow conservation constraint value for interval 1 is 100 + 300 = 400 vehicles / hour / km / hour. The system also needs to consider the upper limit of interval capacity (e.g., 2000 vehicles / hour for interval 1) and the upper limit of congestion density (e.g., 50 vehicles / km for interval 1) as constraints.

[0143] The optimal opening sequence is determined by solving an optimization problem that maximizes the target capacity. For example, for a scheduling problem with 5 time intervals, the optimal opening sequence is: interval 1 (1,0,1,0,1), interval 2 (0,1,0,1,0), interval 3 (1,1,0,1,0), interval 4 (0,0,1,0,1), and interval 5 (1,0,1,1,0). Here, 1 represents open and 0 represents closed.

[0144] The system calculates the difference between the inlet and outlet flow rates of a section to obtain the flow difference value. This flow difference value is then multiplied by the current section opening sequence to obtain the corrected opening sequence. For example, if the inlet flow rate of section 1 is 1600 vehicles / hour and the outlet flow rate is 1500 vehicles / hour, the flow difference is 100 vehicles / hour, and the current opening sequence is 1, then the corrected opening sequence is 1 × (1 + 100 / 1600) = 1.0625, which, after rounding, is still 1, indicating that the section remains open. If the flow difference is large, it may cause the opening status to change.

[0145] The corrected opening sequence is sent to the traffic control system as the section opening control command for the next time period. For example, the control system will open the entrance ramp of section 1 according to command 1 and close the entrance ramp of section 2 according to command 0. The system continuously monitors traffic conditions, repeats the above calculation process periodically, and dynamically adjusts the opening strategy to adapt to changing traffic demands.

[0146] The above methods can be used to optimize road opening strategies based on the characteristics of traffic flow density propagation and the entropy weight method, thereby improving road capacity, reducing traffic congestion, and achieving smooth and efficient traffic flow.

[0147] The present invention provides a decision-making system for whether to open emergency lanes on highways under high traffic conditions, comprising:

[0148] The first unit is used to obtain vehicle driving data in the upstream, midstream and downstream of the target road segment during the target time period and to calculate the traffic flow index of each section.

[0149] The second unit is used to map vehicle driving data from the upstream, midstream and downstream sections to a high-dimensional feature space, calculate the feature correlation between the features of each section through tensor decomposition, resample the feature vectors of each section after mapping, and extract the frequency domain features and fluctuation period of traffic flow based on empirical mode decomposition and Fourier transform.

[0150] The third unit is used to normalize the feature correlation degree, the frequency domain feature and the fluctuation period and then perform a weighted summation. Based on the comparison between the weighted summation result and the preset opening threshold, it is determined whether to open the emergency lane.

[0151] The fourth unit is used to establish a traffic flow intelligent control system based on fluid dynamics if the emergency lane is opened. It calculates the traffic flow dynamics value based on the longitudinal spacing of vehicles and traffic flow data, and calculates the traffic pressure based on the traffic flow density. It calculates the density propagation speed based on the traffic pressure derivative. Based on the density propagation speed, it determines the traffic capacity target value of each section by combining the entropy weight method. It calculates the optimal opening sequence by maximizing the traffic capacity target value, and dynamically corrects the opening strategy based on the principle of traffic flow conservation.

[0152] The fifth unit is used to send an emergency lane opening instruction to the traffic management system based on the output results of the intelligent traffic flow control system; monitor the traffic flow indicators of each section of the open road segment in real time; and send an emergency lane closing instruction when the traffic flow indicators are detected to drop below the preset closing threshold.

[0153] A third aspect of the present invention provides an electronic device, comprising:

[0154] processor;

[0155] Memory used to store processor-executable instructions;

[0156] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0157] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0158] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0159] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A decision-making method for whether to open emergency lanes on highways under high traffic conditions, characterized in that, include: Obtain vehicle driving data for the upstream, midstream, and downstream sections of the target road segment during the target time period, and calculate traffic flow indicators for each section; Vehicle driving data from the upstream, midstream, and downstream sections are mapped to a high-dimensional feature space. Tensor decomposition is used to calculate the feature correlation between features in each section, including: Obtain vehicle speed data for the upstream, midstream, and downstream sections, calculate the standard deviation of vehicle speed for each section, and divide it by the corresponding mean vehicle speed to obtain the traffic flow fluctuation intensity index for each section; when the traffic flow fluctuation intensity index is greater than the fluctuation threshold, multiply the preset time window length by the traffic flow fluctuation intensity index. The feature vectors of each interval are extracted within the adjusted time window. The feature vectors are then input into a random projection matrix for local sensitive hashing. The elements of the random projection matrix follow a standard normal distribution. Through multiple mapping iterations, the mapped feature vectors are obtained. The mapped feature vectors are organized into third-order tensors in the interval and time dimensions. The tensor reconstruction error is calculated. When the reconstruction error is less than the reconstruction error threshold, the current decomposition rank is determined to be the optimal decomposition rank. The optimal decomposition rank is used to perform kernel tensor decomposition to obtain the interval mode decomposition matrix and the time mode decomposition matrix. The static correlation degree between intervals is obtained by performing an inner product operation on the eigenvectors of adjacent intervals in the interval mode decomposition matrix and dividing them by the corresponding modulus product. The temporal mode decomposition matrix is ​​divided into current feature data and historical feature data according to the time window, and the dynamic correlation degree between the current feature data and the historical feature data is calculated. The static correlation degree and the dynamic correlation degree are weighted and summed to obtain the feature correlation degree between each interval feature. The feature vectors of each mapped interval are resampled, and the frequency domain features and fluctuation period of traffic flow fluctuations are extracted based on empirical mode decomposition and Fourier transform, including: The feature vectors of each mapped interval are discretized over time according to the sampling interval to obtain the traffic flow feature sampling sequence. Empirical mode decomposition (EMD) is performed on the traffic flow feature sampling sequence through iterative screening to obtain multiple intrinsic mode functions (IMFs) representing different frequency characteristics. The IMFs are then converted into analytical signals using Hilbert transform, and instantaneous frequency and instantaneous amplitude are extracted from the analytical signals. The discrete Fourier transform of the multiple intrinsic mode functions is performed to obtain the spectral components of each intrinsic mode function. The power spectral density is obtained by weighted summation of the spectral components. The dominant frequency component is extracted from the power spectral density, and the phase accumulation characteristics are calculated by combining the spectral phase of each intrinsic mode function to obtain the frequency domain characteristics of traffic flow fluctuation. The frequency deviation between the instantaneous frequency of each intrinsic mode function and the dominant frequency component is calculated. The frequency similarity is obtained by Gaussian weighting. The root mean square of the instantaneous amplitude of each intrinsic mode function is calculated as the energy value. The frequency similarity is multiplied by the energy value to obtain the contribution of each intrinsic mode function to the dominant frequency component. The contribution is used as a weighting coefficient and superimposed with the frequency of the corresponding intrinsic mode function to obtain the traffic flow fluctuation period considering non-stationary characteristics. The feature correlation degree, the frequency domain feature and the fluctuation period are normalized and then weighted and summed. Based on the comparison between the weighted summation result and the preset opening threshold, it is determined whether to open the emergency lane. If the emergency lane is opened, a traffic flow intelligent control system based on fluid dynamics is established. The system calculates the traffic flow dynamics value based on the longitudinal spacing of vehicles and traffic flow data, and calculates the traffic pressure based on the traffic flow density. The density propagation speed is calculated based on the traffic pressure derivative. Based on the density propagation speed, the target capacity value of each section is determined by combining the entropy weight method. The optimal opening sequence is calculated by maximizing the target capacity value, and the opening strategy is dynamically corrected based on the principle of traffic flow conservation. Based on the output of the intelligent traffic flow control system, an emergency lane opening command is sent to the traffic management system; the traffic flow index of each section of the open road segment is monitored in real time, and when the traffic flow index is detected to drop below the preset closing threshold, an emergency lane closing command is sent.

2. The method according to claim 1, characterized in that, Empirical mode decomposition (EMD) is performed on the traffic flow feature sampling sequence through iterative filtering to obtain multiple intrinsic mode functions (EMFs) representing different frequency characteristics. The EMFs are then converted into analytic signals using Hilbert transform, and instantaneous frequency and amplitude are extracted from these analytic signals, including: The upper and lower envelopes of the traffic flow sampling feature sequence are calculated by cubic spline interpolation, and the arithmetic mean of the upper and lower envelopes is taken as the mean envelope. The mean envelope is subtracted from the traffic flow sampling feature sequence to obtain the current iteration sequence. The upper envelope calculation, lower envelope calculation, and mean envelope calculation are repeatedly performed on the current iteration sequence to obtain the iterative component sequence. Calculate the sum of squares of the differences between the iterative component sequence and the current iterative sequence, divide the sum of squares by the sum of squares of the current iterative sequence to obtain the standard deviation, and when the standard deviation is less than the deviation threshold, determine the iterative component sequence as the first intrinsic mode function; Subtracting the first intrinsic mode function from the traffic flow feature time series sequence yields the first residual sequence. When the first residual sequence is a non-monotonic function, the upper envelope calculation, lower envelope calculation, mean envelope calculation, and iterative component sequence calculation are repeatedly performed on the first residual sequence to obtain the second intrinsic mode function. Perform Hilbert transforms on the first eigenmode function and the second eigenmode function respectively, and construct an analytic signal in complex form from each eigenmode function and its corresponding Hilbert transform result; The phase function is obtained by calculating the arctangent value based on the real and imaginary parts of the analytic signal; the phase derivative is obtained by numerical differentiation of the phase function, and the instantaneous frequency sequence is obtained by dividing the phase derivative by 2π; the instantaneous amplitude sequence is obtained by calculating the square root of the sum of the squares of the real and imaginary parts of the analytic signal.

3. The method according to claim 1, characterized in that, Establish a traffic flow intelligent control system based on fluid dynamics. Calculate traffic flow dynamics values ​​based on vehicle longitudinal spacing and traffic flow data, and combine this with traffic flow density to calculate traffic pressure. Calculate density propagation velocity based on the traffic pressure derivative, including: Traffic flow density data and traffic flow speed data for three intervals are obtained. The partial derivatives of the traffic flow density data and the traffic flow speed data with respect to time are calculated to obtain the traffic flow density change rate and the traffic flow speed change rate. The longitudinal spacing data of adjacent vehicles within each interval is obtained. The difference between the longitudinal spacing and the safe following distance threshold is divided by the square of the longitudinal spacing to obtain the distance influence factor. The traffic flow density change rate, the traffic flow speed change rate and the distance influence factor are weighted and aggregated to obtain the traffic flow dynamics value. Calculate the traffic flow density difference between adjacent intervals, substitute the traffic flow density difference into the hyperbolic tangent function to obtain the interval coupling factor, multiply the interval coupling factor, the traffic flow dynamics value and the power function of the current interval traffic flow density to obtain the traffic pressure; calculate the square root of the derivative of the traffic pressure with respect to the traffic flow density, and multiply it by the ratio of the interval length and the road width to obtain the density propagation speed.

4. The method according to claim 1, characterized in that, Based on the density propagation speed, the target capacity value for each interval is determined using the entropy weight method. The optimal opening sequence is calculated by maximizing the target capacity value, and the opening strategy is dynamically corrected based on the principle of traffic flow conservation, including: The three indicators of traffic flow density, average speed and number of lanes in each interval are normalized to obtain the normalized value of each indicator. The normalized value of each indicator is multiplied by its natural logarithm and then added together to obtain the information entropy of each interval. Based on the density propagation speed, the product of traffic flow density and average speed in adjacent intervals is weighted and calculated to obtain the traffic flow density difference at the boundary of each interval. The traffic flow density difference is used as the flow conservation evaluation factor. The information entropy of each interval, the flow conservation evaluation factor, and the real-time flow of the interval are multiplied and accumulated in the time dimension to obtain the traffic capacity target value. The spatial gradient is obtained by calculating the rate of change of the interval flow with distance, and the temporal gradient is obtained by calculating the rate of change of the interval flow with time. The spatial gradient and the temporal gradient are added together to obtain the flow conservation constraint value. The flow conservation constraint value, the upper limit of the interval capacity, and the upper limit of the congestion density are used as constraints, and the opening sequence corresponding to maximizing the traffic capacity target value is taken as the optimal opening sequence. The difference between the inlet flow and the outlet flow of the interval is calculated to obtain the flow difference value. The flow difference value is multiplied by the current interval opening sequence to obtain the corrected opening sequence. The corrected opening sequence is used as the interval opening control command for the next time period.

5. A decision-making system for whether to open the emergency lane on a highway under high traffic conditions, used to implement the method as described in any one of claims 1-4, characterized in that, include: The first unit is used to obtain vehicle driving data in the upstream, midstream and downstream of the target road segment during the target time period and to calculate the traffic flow index of each section. The second unit is used to map vehicle driving data from the upstream, midstream and downstream sections to a high-dimensional feature space, calculate the feature correlation between the features of each section through tensor decomposition, resample the feature vectors of each section after mapping, and extract the frequency domain features and fluctuation period of traffic flow based on empirical mode decomposition and Fourier transform. The third unit is used to normalize the feature correlation degree, the frequency domain feature and the fluctuation period and then perform a weighted summation. Based on the comparison between the weighted summation result and the preset opening threshold, it is determined whether to open the emergency lane. The fourth unit is used to establish a traffic flow intelligent control system based on fluid dynamics if the emergency lane is opened. It calculates the traffic flow dynamics value based on the longitudinal spacing of vehicles and traffic flow data, and calculates the traffic pressure based on the traffic flow density. It calculates the density propagation speed based on the traffic pressure derivative. Based on the density propagation speed, it determines the traffic capacity target value of each section by combining the entropy weight method. It calculates the optimal opening sequence by maximizing the traffic capacity target value, and dynamically corrects the opening strategy based on the principle of traffic flow conservation. The fifth unit is used to send an emergency lane opening command to the traffic management system based on the output results of the intelligent traffic flow control system. The system monitors traffic flow indicators in each section of the open road in real time. When the traffic flow indicators are detected to drop below the preset closure threshold, an emergency lane closure command is sent.

6. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 4.

7. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 4.

Images