Path flow estimation method, computer system, and computer-readable storage medium

Abstract

The application relates to the technical field of data processing, in particular to a path flow estimation method, a computer system and a computer readable storage medium. A fusion strategy of K shortest path algorithm and A* search algorithm is introduced to screen and complete candidate paths, generate a candidate path set, and update particle weights based on three dimensions of travel time consistency, path consistency and detector measurability to guide the candidate path set to evolve in a more optimal and feasible direction. Dual prior information of tensor solution and particle reconstruction is introduced to construct a joint least square problem of path flow estimation, which can maintain a stable solution under the condition that neither turning flow nor path observation is complete. The application aims to solve the problem of how to realize accurate path flow estimation in a low mobile detector coverage scene.

Description

Path traffic estimation methods, computer systems and computer-readable storage media Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a path flow estimation method, a computer system, and a computer-readable storage medium. Background Technology

[0002] Route flow estimation, as a method of traffic state estimation, can provide a more refined description of the spatial flow and aggregation of traffic demand in the road network, serving as a crucial decision-making basis for traffic planning and control. Real-time, accurate, and holistic reconstruction of road traffic states is a prerequisite and key to achieving intelligent road traffic management. Road traffic state reconstruction provides data and service guarantees for traffic management and travel information services, and is a core element for providing precise and proactive control for intelligent transportation systems.

[0003] Traditional path flow estimation is based on road condition data collected by mobile detectors. The accuracy of the estimation results depends on the density of the mobile detectors. That is, the shorter the distance between the mobile detectors, the more accurate the flow estimation results. However, the density of mobile detectors is greatly constrained in complex urban traffic scenarios, resulting in path flow estimation results that cannot meet the accuracy required by intelligent transportation systems to reconstruct road traffic conditions.

[0004] In view of this, this application proposes a new path traffic estimation method, which aims to achieve accurate path traffic estimation in scenarios with low mobility detector coverage. Summary of the Invention

[0005] The main purpose of this application is to provide a path traffic estimation method that aims to solve the problem of how to achieve accurate path traffic estimation in scenarios with low mobility detector coverage.

[0006] To achieve the above objectives, this application provides a path traffic estimation method, the method comprising:

[0007] S10, determine or generate candidate trajectories from the collected vehicle condition data of the target road network;

[0008] S20, based on the path length and travel time corresponding to each candidate trajectory, select k shortest paths as target candidate paths, and generate a set of supplementary paths. The target candidate paths and the supplementary paths are used as a candidate path set, wherein the trajectory update probability in the candidate path set satisfies at least one of the following update rules:

[0009] The first trajectory update probability is determined based on the actual travel time collected by two adjacent detectors in the target road network and the average travel time between paths in the candidate path set.

[0010] The second trajectory update probability is determined based on the length of the preferred path in the candidate path set, wherein the preferred path refers to the path selected based on the vehicle travel time.

[0011] The third trajectory update probability is determined based on the actual number of detectors passed by in the vehicle driving trajectory collected in the target road network and the number of detectors contained in the preferred path in the candidate path set.

[0012] S30, acquire vehicle steering data corresponding to each turn in the target road network, decompose the vehicle steering data into multiple factor matrices and multiple core tensors, and determine the prior steering flow corresponding to the turn based on the factor matrices and the core tensors;

[0013] S40, the sum of the prior turning flows of all turns traversed by the preferred path is taken as the path flow corresponding to the preferred path.

[0014] Optionally, in step S20, the first trajectory update probability is determined based on the actual travel time collected by two adjacent detectors in the target road network and the average travel time between paths in the candidate path set, specifically including:

[0015]

[0016]

[0017] In the formula, It is the first one between two consecutive detectors The update probability of the preferred path, i.e. the update probability of the first trajectory. This represents the actual travel time between two consecutive adjacent detectors. This represents the average travel time between two consecutive detectors. It is the first The update probability of the preferred path, It is the number of detectors in the actual trajectory of the vehicle. For coefficients less than 1, The predicted time between the two detectors, A coefficient greater than 1.

[0018] Optionally, in step S20, determining the second trajectory update probability based on the length of the candidate paths in the candidate path set specifically includes:

[0019]

[0020] In the formula, Based on path length, the first The probability of updating the preferred path, i.e., the probability of updating the second trajectory. It is the first The length of the preferred path, It represents the total number of preferred paths.

[0021] Optionally, in step S20, the third trajectory update probability is determined based on the actual number of detectors passed by in the vehicle's trajectory collected in the target road network and the number of detectors contained in the paths in the candidate path set. This specifically includes:

[0022]

[0023] In the formula, Based on detector measurability The update probability of the preferred path, i.e., the update probability of the third trajectory. It is the first The number of detectors on the optimal path and the actual vehicle trajectory. It represents the number of detectors in the actual trajectory of the vehicle.

[0024] Optionally, the selection rules for the first trajectory update probability, the second trajectory update probability, and the third trajectory update probability satisfy the following:

[0025] The candidate trajectory is updated with weights based on the first trajectory update probability, the second trajectory update probability, and the third trajectory update probability;

[0026] The candidate trajectory with the largest sum of weights under the three selection rules after weight update is selected as the preferred trajectory to be added to the candidate path set.

[0027] Optionally, S30 specifically includes:

[0028] S31, the vehicle turning data is composed of three dimensions: traffic flow data at different times, the location of traffic nodes, and the driving direction of vehicles at the nodes;

[0029] S32, the three-dimensional tensor is decomposed into multiple factor matrices and multiple core tensors using the following formula, wherein the factor matrices are used to characterize the features of the three dimensions, and the core tensors are used to capture the relationships between the features:

[0030]

[0031] In the formula, Represents any element in a 3D tensor; i, j, and t represent indices in different dimensions, respectively; core tensor. The size is Its elements are denoted as ; The rank of the three-dimensional tensor; the first factor matrix is ​​located at the intersection. The size is Its elements are denoted as M is the size of the tensor Y in the first dimension; the second factor matrix represents the direction of travel. The size is Its elements are denoted as N is the size of the tensor Y in the second dimension; the third factor matrix is ​​time. The size is Its elements are denoted as T is the size of tensor Y in the third dimension;

[0032] S33 uses the conjugate gradient method to find the least squares approximate solution where the partial derivatives of the core tensor and each factor matrix are equal to zero.

[0033]

[0034]

[0035]

[0036]

[0037] In the formula, Represents the objective function, set A set of indices representing the observed elements; Indicates action on a set Orthogonal mappings on; is the regularization parameter; T represents transpose.

[0038] S34, determine the reconstructed tensor based on the factor matrix and the core tensor, and output the reconstructed tensor as the prior steering flow:

[0039]

[0040] In the formula, To reconstruct the tensor; To output the core tensor; , , This is the output factor matrix.

[0041] Optionally, S40 includes:

[0042] S41, Calculate the prior path flow corresponding to turn i based on the prior turning flow and the preset correlation coefficient. :

[0043]

[0044]

[0045] In the formula, To select the optimal path The actual data collection traffic; Let be the correlation coefficient between path k and turn i; This is the set of preferred paths within the candidate path set;

[0046] S42, the sum of the prior path flows of all turns traversed by the preferred path k is taken as the path flow of the preferred path k:

[0047]

[0048] in,

[0049]

[0050] In the formula, To optimize the path traffic of path k, It is the preferred path The prior path traffic corresponding to turn i; For turning Prior redirection flow; For the preferred path k, the turning points gather.

[0051] Optionally, S40 further includes:

[0052] The objective function is to minimize the weighted sum of the turning flow estimation error term and the path flow estimation error term, wherein the objective function... for:

[0053]

[0054] in:

[0055]

[0056] In the formula, For turning The actual measured flow rate; For turning Prior redirection flow; To select the optimal path The actual data collection traffic; To select the optimal path Prior path traffic; , These are the weighting coefficients for the two error terms, respectively. Let be the correlation coefficient between path k and turn i; For the turning set; It is a set of paths.

[0057] In addition, to achieve the above objectives, this application also provides a computer system comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps of the path traffic estimation method as described in any of the preceding claims.

[0058] In addition, to achieve the above objectives, this application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the path traffic estimation method as described in any of the preceding claims.

[0059] This application has at least the following beneficial effects:

[0060] (1) Traditional path flow estimation methods often use empirical allocation or static average value as the basis for turning flow estimation, which is difficult to reflect the spatiotemporal dependence characteristics between intersections. This application introduces the third-order tensor decomposition method into the intersection-time-direction three-dimensional structure of LPR data to construct a dynamic prior turning flow estimation framework, so as to achieve robust modeling of incomplete or noisy data;

[0061] (2) Unlike other traditional analysis methods whose accuracy is limited by data quality, the path traffic estimation method proposed in this application does not require assumptions about travel behavior and path selection;

[0062] (3) By establishing two types of error minimization objective functions, and introducing dual prior information of tensor solution and particle reconstruction, a joint least squares problem for path flow estimation is constructed, which can maintain a stable solution under the condition that neither turning flow nor path observation is complete. Attached Figure Description

[0063] Figure 1 is a flowchart illustrating the path traffic estimation method according to an embodiment of this application;

[0064] Figure 2 is a schematic diagram of the tensor decomposition principle involved in the embodiments of this application;

[0065] Figure 3 is a comparison chart of the combined error results of different ranks involved in the embodiments of this application;

[0066] Figure 4 is a comparison chart of error results for different weighting coefficients involved in the embodiments of this application;

[0067] Figure 5 shows a comparison of road traffic flow errors of license plate recognition data at different sampling rates in the embodiments of this application. Specifically, Figure 5(a) shows the comparison of road traffic flow errors of license plate recognition data at an 80% sampling rate, Figure 5(b) shows the comparison of road traffic flow errors of license plate recognition data at a 70% sampling rate, Figure 5(c) shows the comparison of road traffic flow errors of license plate recognition data at a 60% sampling rate, Figure 5(d) shows the comparison of road traffic flow errors of license plate recognition data at a 50% sampling rate, Figure 5(e) shows the comparison of road traffic flow errors of license plate recognition data at a 40% sampling rate, and Figure 5(f) shows the comparison of road traffic flow errors of license plate recognition data at a 30% sampling rate.

[0068] Figure 6 is a comparison chart of path flow error of license plate recognition data under different sampling rates according to the embodiments of this application. Among them, Figure 6(a) is a comparison chart of path flow error of license plate recognition data at 80% sampling rate, Figure 6(b) is a comparison chart of path flow error of license plate recognition data at 70% sampling rate, Figure 6(c) is a comparison chart of path flow error of license plate recognition data at 60% sampling rate, Figure 6(d) is a comparison chart of path flow error of license plate recognition data at 50% sampling rate, Figure 6(e) is a comparison chart of path flow error of license plate recognition data at 40% sampling rate, and Figure 6(f) is a comparison chart of path flow error of license plate recognition data at 30% sampling rate.

[0069] Figure 7 is a line graph showing the percentage error of different estimation methods under different indicators in the embodiments of this application. Figure 7(a) is a line graph showing the percentage error of different estimation methods under the RMSE indicator, Figure 7(b) is a line graph showing the percentage error of different estimation methods under the MAE indicator, and Figure 7(c) is a line graph showing the percentage error of different estimation methods under the MAPE indicator.

[0070] Figure 8 shows the fitting curve relationship between sampling rate and error under different indicators in the embodiments of this application. Among them, Figure 8(a) shows the fitting curve relationship between sampling rate and error under the RMSE indicator, Figure 8(b) shows the fitting curve relationship between sampling rate and error under the MAE indicator, and Figure 8(c) shows the fitting curve relationship between sampling rate and error under the MAPE indicator.

[0071] Figure 9 is a schematic diagram of the hardware operating environment of the computer system involved in the embodiments of this application.

[0072] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0073] To better understand the above technical solutions, exemplary embodiments of this disclosure will be described in more detail below with reference to the accompanying drawings. While exemplary embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of this disclosure to those skilled in the art.

[0074] First Embodiment

[0075] Referring to Figure 1, this embodiment provides a path traffic estimation method, which includes the following steps:

[0076] S10, determine or generate candidate trajectories from the collected vehicle condition data of the target road network;

[0077] In this embodiment, the road network refers to a network system composed of roads of various grades and functions interconnected through intersections, including road surfaces, lane markings, traffic signs, and traffic lights. This embodiment uses the target road network as a unit to estimate path flow and reconstruct urban road traffic conditions.

[0078] The vehicle condition data of the road network is collected by the mobile measurement system, and the candidate trajectory is the possible driving trajectory of the vehicle generated by the deep learning model based on the vehicle condition data, which is usually multiple.

[0079] In some alternative implementations, the preprocessing rules for vehicle condition data can satisfy the following aspects:

[0080] 1. If the detection equipment detects the same vehicle at the same time or within a short period of time, for duplicate data occurring at the same time, delete the redundant data and keep only one data. For duplicate data occurring within a short period of time, only the last data detected should be kept. Low-quality vehicle condition data caused by the performance of the detection equipment, environmental factors (light, weather), and vehicle factors (license plate obstruction, license plate damage) should be appropriately retained when calculating turning flow.

[0081] Second: Number each intersection in the road network, such as "Intersection A" corresponding to number 1, "Intersection B" corresponding to number 2, and so on.

[0082] Third, because the lane function fields in the original data are strings and inconsistent, it's impossible to calculate turning flow numerically. Therefore, it needs to be converted to a numerical type, with each direction represented by a single value. For example, "left turn lane" is represented by 1, "straight lane" by 2, and "right turn lane" by 3.

[0083] Fourth, only retain the hour, minute, and second of each vehicle passage time, and finally unify the timestamp to reduce the workload of subsequent data analysis;

[0084] V. The remaining data after normalization filtering.

[0085] S20, based on the path length and travel time corresponding to each candidate trajectory, select k shortest paths as target candidate paths, and generate a set of supplementary paths. The target candidate paths and the supplementary paths are used as a candidate path set, wherein the trajectory update probability in the candidate path set satisfies at least one of the following update rules:

[0086] The first trajectory update probability is determined based on the actual travel time collected by two adjacent detectors in the target road network and the average travel time between paths in the candidate path set.

[0087] The second trajectory update probability is determined based on the length of the preferred path in the candidate path set, wherein the preferred path refers to the path selected based on the vehicle travel time.

[0088] The third trajectory update probability is determined based on the actual number of detectors passed by in the vehicle driving trajectory collected in the target road network and the number of detectors contained in the preferred path in the candidate path set.

[0089] In this step, we propose a novel approach to constructing the candidate path set. To improve the coverage of candidate paths in trajectory reconstruction, we introduce a fusion strategy of the K-shortest path algorithm and the A* search algorithm to expand the range of initial particle (i.e., candidate path) generation, thereby enhancing particle diversity.

[0090] It should be noted that the preferred route refers to the route selected based on vehicle travel time. Specifically, the candidate routes in the candidate route set are sorted according to vehicle travel time, and the top N routes are selected as the preferred routes, where N is greater than 1.

[0091] Specifically, based on path length and vehicle travel time, the K-shortest path algorithm is used. First, paths that do not meet the conditions are eliminated. Second, the downstream intersections of the missing origin and destination and the upstream intersections of the missing destination are determined based on the names of the missing origin and destination intersections and lane functions. Finally, a basic trajectory set is filtered out based on the travel information of the missing origin and destination and the road network topology information.

[0092]

[0093] In the formula: The weight of edge n is determined by the travel time or distance.

[0094] The A* search algorithm is used to search for potential paths that the K-shortest path algorithm cannot cover. The path evaluation function for A* is:

[0095]

[0096] In the formula: This represents the known cost (cumulative distance or time) from the starting point to the current node. This represents the heuristic estimate (shortest path estimate) from the current node to the destination.

[0097] The fusion strategy is as follows:

[0098] (1) Use the K-shortest path algorithm to obtain K shortest paths as the basic candidate set. ;

[0099] (2) Using the A* search algorithm, set a path length threshold and generate a set of supplementary paths. :

[0100]

[0101] In the formula: It is a redundancy coefficient used to limit the "deviation" of the A* expansion path.

[0102] The two are merged to form the final candidate path set:

[0103]

[0104] Furthermore, in order to determine the priority of each path among the preferred paths in the candidate path set, it is necessary to update the particle weights in the candidate path set. In this embodiment, three criteria are selected for consideration: travel time consistency, path consistency, and detector measurability.

[0105] In some alternative implementations, the quantification of travel time consistency, i.e., the first trajectory update probability, is calculated based on the similarity between the actual travel time and the average travel time of the vehicle by comparing the two consecutive detectors.

[0106] Specifically, it includes:

[0107]

[0108]

[0109] In the formula, It is the first one between two consecutive detectors The update probability of the preferred path, i.e. the update probability of the first trajectory. This represents the actual travel time between two consecutive adjacent detectors. This represents the average travel time between two consecutive detectors. It is the first The update probability of the preferred path, It is the number of detectors in the actual trajectory of the vehicle. For coefficients less than 1, The predicted time between the two detectors, A coefficient greater than 1.

[0110] In some alternative implementations, path consistency involves comparing the similarity between candidate paths and known trajectories, i.e., the second trajectory update probability. The length of the candidate trajectory is calculated to update the particle weights. Specifically, this includes:

[0111]

[0112] In the formula, Based on path length, the first The probability of updating the preferred path, i.e., the probability of updating the second trajectory. It is the first The length of the preferred path, It represents the total number of preferred paths.

[0113] In some alternative implementations, the quantification of detector measurability, i.e. the third trajectory update probability, is to compare the number of detectors that the vehicle passes through in its actual trajectory (i.e., the number of times the vehicle is captured) with the number of sensors contained in each trajectory candidate. The aim is to assign higher weights to candidate trajectories with the same number of detectors as the number of detectors that the vehicle's actual trajectory passes through.

[0114] Specifically, it includes:

[0115]

[0116] In the formula, Based on detector measurability The update probability of the preferred path, i.e., the update probability of the third trajectory. It is the first The number of detectors on the optimal path and the actual vehicle trajectory. It represents the number of detectors in the actual trajectory of the vehicle.

[0117] It should be noted that this embodiment does not limit the specific usage of the three trajectory update probabilities.

[0118] S30, acquire vehicle steering data corresponding to each turn in the target road network, decompose the vehicle steering data into multiple factor matrices and multiple core tensors, and determine the prior steering flow corresponding to the turn based on the factor matrices and the core tensors;

[0119] In this step, vehicle steering data at various steering nodes at different times are preprocessed. The factor matrices represent the latent features of time, location, and driving direction, respectively, while the core tensor captures the complex relationships between these features. By decomposing the obtained factor matrices and core tensors, the latent features and patterns of each dimension are calculated. These features and patterns are used as prior values ​​to obtain the prior steering flow at different times.

[0120] Further and optionally, referring to the schematic diagram of tensor decomposition principle shown in FIG2, S30 specifically includes:

[0121] S31, the vehicle turning data is composed of three dimensions: traffic flow data at different times, the location of traffic nodes, and the driving direction of vehicles at the nodes;

[0122] S32, the three-dimensional tensor is decomposed into multiple factor matrices and multiple core tensors using the following formula, wherein the factor matrices are used to characterize the features of the three dimensions, and the core tensors are used to capture the relationships between the features:

[0123]

[0124] In the formula, Represents any element in a 3D tensor; i, j, and t represent indices in different dimensions, respectively; core tensor. The size is Its elements are denoted as ; The rank of the three-dimensional tensor; the first factor matrix is ​​located at the intersection. The size is Its elements are denoted as M is the size of the tensor Y in the first dimension; the second factor matrix represents the direction of travel. The size is Its elements are denoted as N is the size of the tensor Y in the second dimension; the third factor matrix is ​​time. The size is Its elements are denoted as T is the size of tensor Y in the third dimension;

[0125] In addition, due to objective factors, there are some missing data collected by the detector. In order to obtain the core tensor and factor matrix, the approximation problem is described as the following optimization problem in this embodiment.

[0126]

[0127]

[0128] Where: set A set of indices representing the observed elements; Indicates action on a set Orthogonal mappings on; The Frobenius norm of the tensor; For regularization parameters.

[0129] This optimization problem is solved using step S33:

[0130] S33 uses the conjugate gradient method to find the least squares approximate solution where the partial derivatives of the core tensor and each factor matrix are equal to zero.

[0131]

[0132]

[0133]

[0134]

[0135] In the formula, Represents the objective function, set A set of indices representing the observed elements; Indicates action on a set Orthogonal mappings on; The Frobenius norm of the tensor; is the regularization parameter; T represents transpose.

[0136] S34, determine the reconstructed tensor based on the factor matrix and the core tensor, and output the reconstructed tensor as the prior steering flow:

[0137]

[0138] In the formula, To reconstruct the tensor; To output the core tensor; , , This is the output factor matrix.

[0139] S40, the sum of the prior turning flows of all turns traversed by the preferred path is taken as the path flow corresponding to the preferred path.

[0140] In this embodiment, we equate the path traffic of the preferred path to the sum of the prior turning traffic that passes through the path and then turns.

[0141] Specifically, S40 includes:

[0142] S41, Calculate the prior path flow corresponding to turn i based on the prior turning flow and the preset correlation coefficient. :

[0143]

[0144]

[0145] In the formula, To select the optimal path The actual data collection traffic; Let be the correlation coefficient between path k and turn i; This is the set of preferred paths within the candidate path set; The correlation coefficient matrix between turning and the preferred path:

[0146]

[0147] S42, the sum of the prior path flows of all turns traversed by the preferred path k is taken as the path flow of the preferred path k:

[0148]

[0149] in,

[0150]

[0151] In the formula, To optimize the path traffic of path k, It is the preferred path The prior path traffic corresponding to turn i; For turning Prior redirection flow; For the preferred path k, the turning points gather.

[0152] In the technical solution provided in this embodiment, a fusion strategy of the K-shortest path algorithm and the A* search algorithm is introduced to screen and complete candidate paths, generating a candidate path set. The particle weights are updated based on three dimensions: travel time consistency, path consistency, and detector measurability, to guide the candidate path set to evolve in a better and more feasible direction. The dual prior information of tensor solution and particle reconstruction is introduced to construct a joint least squares problem for path flow estimation, which can maintain a stable solution under the condition that neither turning flow nor path observation is complete. In addition, compared with traditional path flow estimation methods, the method used in this embodiment is data-driven and does not require assumptions about travel behavior and path selection.

[0153] Second Embodiment

[0154] Based on the first embodiment, in this embodiment, the selection rules for the first trajectory update probability, the second trajectory update probability, and the third trajectory update probability satisfy the following:

[0155] The candidate trajectory is updated with weights based on the first trajectory update probability, the second trajectory update probability, and the third trajectory update probability;

[0156] The candidate trajectory with the largest sum of weights under the three selection rules after weight update is selected as the preferred trajectory to be added to the candidate path set.

[0157] It should be noted that, under normal circumstances, the weight sums of two candidate trajectories are not the same. However, this embodiment provides an additional explanation for the extreme case: if there are two candidate trajectories with the same weight sum, then both are added to the candidate dataset as preferred trajectories.

[0158] Third Embodiment

[0159] Based on any embodiment, in this embodiment, based on the least squares method, the objective function is the weighted sum of the path flow estimation error term and the road segment turning flow estimation error term. The objective function includes two error terms: the first term is the deviation between the road segment turning flow and the prior estimate of the road segment turning flow, and the second term is the deviation between the path flow and the prior estimate of the path flow. According to the correlation constraint between the road segment flow and the path flow, the weighted sum of the two error terms is minimized.

[0160]

[0161]

[0162] In the formula, For turning The actual measured flow rate; For turning Prior redirection flow; To select the optimal path The actual data collection traffic; To select the optimal path Prior path traffic; , These are the weighting coefficients for the two error terms, respectively. Let be the correlation coefficient between path k and turn i; For the turning set; It is a set of paths.

[0163] Verification Example 1

[0164] Based on any of the above embodiments, this embodiment verifies the effectiveness of the methods or models constituted by the methods involved.

[0165] In this embodiment, the road network within the first ring road of the study area is selected as the target road network. This network contains 22 major signal-controlled intersections, one of which is a T-junction, and the rest are cross intersections, totaling 64 road segments (by direction). The data selected is license plate recognition data from the corresponding road network area on August 24, 2024 (morning peak 7:00-9:00, evening peak 17:00-19:00), totaling 248,798 vehicle records. Data for three signal-controlled intersections (numbered 20, 21, and 22) is missing. To accurately estimate path flow, all missing vehicle trajectories need to be reconstructed. Vehicle travel trajectories are obtained from the license plate recognition data. First, based on vehicle dwell time and location, the complete vehicle trajectory is divided into different trips. In areas where congestion is infrequent, the dwell time threshold is set to 5 minutes, while in areas where congestion is frequent, the dwell time threshold is set to 10 minutes. This invention studies the morning and evening peak hours, which are prone to congestion; therefore, the dwell time threshold is set to 10 minutes in this embodiment. The vehicle reconstruction results are shown in Table 1.

[0166] Table 1. Results of Vehicle Reconfiguration Experiment

[0167]

[0168] In this embodiment, vehicle passage data included in some trajectories are used as experimental data. The time window of a trajectory is determined according to the collection time of its starting point. By summarizing the O and D points of travel trajectories under different time windows, the OD matrix for all time windows in the study area can be obtained. O refers to the starting point, and D refers to the ending point. This matrix is ​​defined as the true OD of the area. Table 2 shows the estimation results of some OD pairs.

[0169] Table 2. OD Estimation Assessment Results

[0170]

[0171] As shown in Table 2, during the morning rush hour, the RMSE ranged from 5 to 87 vehicles, and the MAPE ranged from 8.42% to 26.07%. During the evening rush hour, the RMSE ranged from 7 to 127 vehicles, and the MAPE ranged from 8.73% to 31.71%. Furthermore, the table shows that the OD pair (5, 2) had a higher RMSE during the evening rush hour. This may be due to the large number of vehicles in that area and the limited number of LPRs, resulting in many vehicles not being detected.

[0172] Furthermore, referring to Figure 3, which compares the error results of different rank combinations, it can be seen that when the rank combination is (5, 7, 6), the RMSE (approximately 24 vehicles) and MAE (approximately 16 vehicles) both reach their lowest values, while the MAPE also remains at a low level of approximately 25%. This indicates that this combination performs optimally in terms of error control, effectively balancing the model's fitting and generalization abilities. In contrast, when the rank combination is (7, 6, 5), all three error indicators reach their highest values, with RMSE and MAE exceeding 30 and MAPE reaching approximately 38%. This indicates that this combination tends to overfit, and the excessive model complexity leads to a decrease in generalization ability. This suggests that setting the rank value too small may result in insufficient information extraction, while setting it too large may introduce redundant features. Therefore, combining the results shown in the figure with the actual error performance, the rank combination (5, 7, 6) was ultimately selected for the tensor decomposition process to ensure optimal estimation performance.

[0173] Furthermore, referring to Figure 4, which compares the error results with different weighting coefficients, it can be seen that as the ratio of the weighting coefficients of the path flow error term and the turning flow error term in the objective function gradually decreases, the error in path flow estimation also gradually increases. When the weighting coefficient of the path flow error term is greater than the weighting coefficient of the turning flow error term, i.e., the weighting ratio... At that time, the fluctuation increase of the path flow estimation error is small, but when the weight coefficient of the path flow error term is less than the weight coefficient of the turning flow error term, that is, the weight ratio is smaller. At that time, the fluctuation increase of the path flow estimation error was relatively large, from the weighting coefficient ratio arrive The path flow estimation error rose to 31.3%. It can be seen that when the path flow error term weights are not dominant, the path flow estimation accuracy is positively correlated with its weights. This is because each turning flow is an independent estimate, while the prior path flow estimate is obtained from the estimates of multiple turning flows along the path, which to some extent neutralizes the error. Therefore, for the ILS-PFE model, the weight coefficient combination should be set... This ensures the accuracy and robustness of path flow estimation.

[0174] Furthermore, referring to Figures 5(a) to 5(f), which compare the road traffic flow error of license plate recognition data at sampling rates of 80%, 70%, 60%, 50%, 40%, and 30%, respectively, it can be seen that the LPR sampling rate affects the accuracy of road traffic flow estimation. As the sampling rate decreases, the points gradually move away from the 45-degree dashed line. This indicates that the accuracy of road traffic flow estimation deteriorates as the sampling rate decreases, especially when the sampling rate is below 60%.

[0175] Furthermore, referring to Figures 6(a) to 6(f), which compare the path flow error of license plate recognition data at sampling rates of 80%, 70%, 60%, 50%, 40%, and 30%, it can be seen that the LPR sampling rate affects the accuracy of path flow estimation. As the sampling rate decreases, the points gradually move away from the 45-degree dashed line. This indicates that the accuracy of path flow estimation deteriorates as the sampling rate decreases, especially when the sampling rate is below 60%.

[0176] Furthermore, referring to Figures 7(a) to 7(c), which are line graphs showing the percentage error of different estimation methods under the RMSE, MAE, and MAPE metrics, it can be seen that ILS-PFE is lower than PFE and PFE-I at different sampling rates. However, at high sampling rates of 80% and 70%, LSTM-PFE has a lower error than ILS-PFE, while at sampling rates of 60% and below, LSTM-PFE has a higher error than ILS-PFE. Therefore, it can be concluded that LSTM-PFE has higher estimation accuracy than ILS-PFE at high sampling rates, and ILS-PFE has higher estimation accuracy than LSTM-PFE at low sampling rates. Overall, however, ILS-PFE has better overall robustness than LSTM-PFE. Additionally, at a sampling rate of 40%, the RMSE of ILS-PFE increases to 68.07, while the RMSE of PFE increases to 106.81. This is mainly because PFE does not use the reconstructed path flow data generated by particle filtering. Reconstructing path traffic data is very meaningful for improving the accuracy of path traffic estimation under low LPR coverage.

[0177] In addition, referring to Figures 8(a) to 8(c), which are the fitting curves of sampling rate and error under the RMSE, MAE and MAPE indices respectively, it can be seen that the error of path flow estimation decreases significantly with the increase of sampling rate, and the two show a good linear negative correlation. When the sampling rate reaches more than 60%, the error level is low, indicating that the sampling rate plays a decisive role in the accuracy of the model.

[0178] As one implementation scheme, Figure 9 is a schematic diagram of the hardware operating environment of the computer system involved in the embodiment of this application.

[0179] As shown in Figure 9, the computer system may include: a processor 1001, such as a CPU; a memory 1005; a user interface 1003; a network interface 1004; and a communication bus 1002. The communication bus 1002 is used to enable communication between these components. The user interface 1003 may include a display screen or an input unit such as a keyboard. Optionally, the user interface 1003 may also include a standard wired interface or a wireless interface. The network interface 1004 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface). The memory 1005 may be high-speed RAM or non-volatile memory, such as a disk drive. Optionally, the memory 1005 may also be a storage device independent of the aforementioned processor 1001.

[0180] Those skilled in the art will understand that the computer system architecture shown in Figure 9 does not constitute a limitation on the computer system and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0181] As shown in Figure 9, the memory 1005, as a storage medium, may include an operating system, a network communication module, a user interface module, and computer programs. The operating system is a program that manages and controls the hardware and software resources of the computer system, as well as the operation of the computer programs and other software or programs.

[0182] In the computer system shown in Figure 9, the user interface 1003 is mainly used to connect to the terminal and communicate with the terminal; the network interface 1004 is mainly used to communicate with the backend server; and the processor 1001 can be used to call the computer program stored in the memory 1005.

[0183] In this embodiment, the computer system includes: a memory 1005, a processor 1001, and a computer program stored in the memory and executable on the processor, wherein:

[0184] When processor 1001 calls a computer program stored in memory 1005, it performs the following operations:

[0185] S10, determine or generate candidate trajectories from the collected vehicle condition data of the target road network;

[0186] S20, based on the path length and travel time corresponding to each candidate trajectory, select k shortest paths as target candidate paths, and generate a set of supplementary paths. The target candidate paths and the supplementary paths are used as a candidate path set, wherein the trajectory update probability in the candidate path set satisfies at least one of the following update rules:

[0187] The first trajectory update probability is determined based on the actual travel time collected by two adjacent detectors in the target road network and the average travel time between paths in the candidate path set.

[0188] The second trajectory update probability is determined based on the length of the preferred path in the candidate path set, wherein the preferred path refers to the path selected based on the vehicle travel time.

[0189] The third trajectory update probability is determined based on the actual number of detectors passed by in the vehicle driving trajectory collected in the target road network and the number of detectors contained in the preferred path in the candidate path set.

[0190] S30, acquire vehicle steering data corresponding to each turn in the target road network, decompose the vehicle steering data into multiple factor matrices and multiple core tensors, and determine the prior steering flow corresponding to the turn based on the factor matrices and the core tensors;

[0191] S40, the sum of the prior turning flows of all turns traversed by the preferred path is taken as the path flow corresponding to the preferred path.

[0192] Furthermore, those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program includes program instructions and can be stored in a storage medium, which is a computer-readable storage medium. The program instructions are executed by at least one processor in a computer system to implement the process steps of the embodiments of the above methods.

[0193] Therefore, this application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the various steps of the path traffic estimation method described in the above embodiments.

[0194] The computer-readable storage medium can be any computer-readable storage medium capable of storing program code, such as a USB flash drive, portable hard drive, read-only memory (ROM), magnetic disk, or optical disk.

[0195] It should be noted that, since the storage medium provided in the embodiments of this application is the storage medium used to implement the methods of the embodiments of this application, those skilled in the art can understand the specific structure and variations of the storage medium based on the methods described in the embodiments of this application, and therefore will not be repeated here. All storage media used in the methods of the embodiments of this application fall within the scope of protection of this application.

[0196] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0197] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in one or more blocks of the flowchart illustrations and / or one or more blocks of the block diagrams.

[0198] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means that implement the functions specified in one or more flowcharts and / or one or more block diagrams.

[0199] These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, such that the instructions, which execute on the computer or other programmable apparatus, provide steps for implementing the functions specified in one or more flowcharts and / or one or more block diagrams.

[0200] It should be noted that any reference signs placed between parentheses in the claims should not be construed as limiting the claims. The word "comprising" does not exclude the presence of components or steps not listed in the claims. The word "a" or "an" preceding a component does not exclude the presence of a plurality of such components. This application can be implemented by means of hardware comprising several different components and by means of a suitably programmed computer. In a unit claim enumerating several means, several of these means may be embodied by the same item of hardware. The use of the words first, second, and third, etc., does not indicate any order. These words can be interpreted as names.

[0201] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0202] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A path flow estimation method, characterized in that, The method includes the following steps: S10, determining or generating candidate trajectories from the collected vehicle condition data of the target road network; S20, selecting k shortest paths as target candidate paths based on the path length and travel time corresponding to each candidate trajectory, and generating a set of supplementary paths, using the target candidate paths and the supplementary paths as a candidate path set, wherein the trajectory update probability in the candidate path set satisfies at least one of the following update rules: determining a first trajectory update probability based on the actual travel time collected by two adjacent detectors in the target road network and the average travel time between paths in the candidate path set; determining a second trajectory based on the length of the preferred path in the candidate path set. Update probability, wherein the preferred path refers to the path selected based on vehicle travel time; determine the third trajectory update probability based on the number of actual detectors passed by the vehicle's travel trajectory collected in the target road network and the number of detectors included in the preferred path in the candidate path set; S30, acquire the vehicle turning data corresponding to each turn in the collected target road network, decompose the vehicle turning data into multiple factor matrices and multiple core tensors, and determine the prior turning flow corresponding to the turn according to the factor matrices and the core tensors; S40, take the sum of the prior turning flows of all turns passed by the preferred path as the path flow corresponding to the preferred path.

2. The method as described in claim 1, characterized in that, In step S20, the first trajectory update probability is determined based on the actual travel time collected by two adjacent detectors in the target road network and the average travel time between paths in the candidate path set. Specifically, this includes: ； In the formula, It is the first one between two consecutive detectors The update probability of the preferred path, i.e. the update probability of the first trajectory. This represents the actual travel time between two consecutive adjacent detectors. This represents the average travel time between two consecutive detectors. It is the first The update probability of the preferred path, It is the number of detectors in the actual trajectory of the vehicle. For coefficients less than 1, The predicted time between the two detectors, A coefficient greater than 1.

3. The method as described in claim 1, characterized in that, In step S20, the second trajectory update probability is determined based on the length of the candidate paths in the candidate path set, specifically including: In the formula, Based on path length, the first The probability of updating the preferred path, i.e., the probability of updating the second trajectory. It is the first The length of the preferred path, It represents the total number of preferred paths.

4. The method as described in claim 1, characterized in that, In step S20, the third trajectory update probability is determined based on the actual number of detectors passed by in the vehicle trajectories collected in the target road network and the number of detectors contained in the paths in the candidate path set. Specifically, this includes: In the formula, Based on detector measurability The update probability of the preferred path, i.e., the update probability of the third trajectory. It is the first The number of detectors on the optimal path and the actual vehicle trajectory. It represents the number of detectors in the actual trajectory of the vehicle.

5. The method according to any one of claims 1 to 4, characterized in that, The selection rules for the first trajectory update probability, the second trajectory update probability, and the third trajectory update probability are as follows: the candidate trajectory is updated with weights based on the first trajectory update probability, the second trajectory update probability, and the third trajectory update probability; the candidate trajectory with the largest sum of weights under the three selection rules after the weight update is selected as the preferred trajectory to be added to the candidate path set.

6. The method as described in claim 1, characterized in that, S30 specifically includes: S31, constructing vehicle turning data using three dimensions: traffic flow data at different times, the location of traffic nodes, and the driving direction of vehicles at the nodes; S32, decomposing the three-dimensional tensor into multiple factor matrices and multiple core tensors using the following formula, wherein the factor matrices are used to characterize the features of the three dimensions, and the core tensors are used to capture the relationships between the features: In the formula, Represents any element in a 3D tensor; i, j, and t represent indices in different dimensions, respectively; core tensor. The size is Its elements are denoted as ； The rank of the three-dimensional tensor; the first factor matrix is ​​located at the intersection. The size is Its elements are denoted as M is the size of the tensor Y in the first dimension; the second factor matrix represents the direction of travel. The size is Its elements are denoted as N is the size of the tensor Y in the second dimension; the third factor matrix is ​​time. The size is Its elements are denoted as T is the magnitude of tensor Y in the third dimension; S33, using the conjugate gradient method to find the least squares approximate solution where the partial derivatives of the core tensor and each factor matrix are equal to zero: ； ； ； In the formula, Represents the objective function, set A set of indices representing the observed elements; Indicates action on a set Orthogonal mappings on; The regularization parameter is T; T represents transpose; S34, determine the reconstructed tensor based on the factor matrix and the core tensor, and output the reconstructed tensor as the prior steering flow: In the formula, To reconstruct the tensor; To output the core tensor; 、 、 This is the output factor matrix.

7. The method as described in claim 1, characterized in that, S40 includes: S41, calculating the prior path flow corresponding to turn i based on the prior turning flow and the preset correlation coefficient. ： ； In the formula, To select the optimal path The actual data collection traffic; Let be the correlation coefficient between path k and turn i; S42, the sum of the prior path flows of all turns traversed by the preferred path k is taken as the path flow of the preferred path k. ;in, In the formula, To optimize the path traffic of path k, It is the preferred path The prior path traffic corresponding to turn i; For turning Prior redirection flow; For the preferred path k, the turning points gather.

8. The method as described in claim 1, characterized in that, S40 further includes: using the weighted sum of the turning flow estimation error term and the path flow estimation error term as the objective function, wherein the objective function... for: ;in: In the formula, Turn The corresponding prior path traffic; For turning Prior redirection flow; To select the optimal path The actual data collection traffic; To select the optimal path Prior path traffic; 、 These are the weighting coefficients for the two error terms, respectively. Let be the correlation coefficient between path k and turn i; For the turning set; It is the set of preferred paths in the candidate path set.

9. A computer system, characterized in that, The computer system includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps of the path traffic estimation method as described in any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the path traffic estimation method as described in any one of claims 1 to 8.

Images