Road network traffic flow analysis method based on robust principal component tracking

Abstract

The invention relates to the technical field of traffic flow analysis, in particular to a road network traffic flow analysis method based on robust principal component tracking. The method comprises the steps of collecting traffic information of a target area road network; performing compensation processing and protocol processing on the traffic information to obtain an original traffic matrix of the traffic information; performing singular value decomposition on the original traffic matrix to obtain an original noise matrix; obtaining a random matrix constraint parameter of the relaxation principal component tracking method according to the original noise matrix; decomposing the original traffic matrix by using a relaxation principal component tracking method to obtain a low-rank matrix of traffic information; and decomposing the low-rank matrix by adopting a singular value decomposition method to obtain a time-varying feature component and a spatial feature component, and carrying out traffic flow analysis after clustering the spatial feature component. According to the method, sudden change traffic flow components contained in traffic information and noise components caused by various reasons can be effectively eliminated, and the accuracy of traffic flow analysis is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of traffic flow analysis, and in particular to a road network traffic flow analysis method based on robust principal component tracking. Background Art

[0002] With the rapid economic and social development of my country, the number of motor vehicles in cities has increased annually, and residents have become overly reliant on motorized transportation, leading to a significant imbalance between transportation supply and demand. Cities are facing increasingly severe traffic problems. With the rapid development of urbanization and the expansion of road infrastructure, collected traffic flow data is characterized by multi-source, multi-structure, massive, and large-scale continuity. The evolution of traffic flow states in road networks also exhibits high-dimensional, random, dynamic, and complex characteristics. Accurately understanding the evolution of traffic states in large-scale (i.e., encompassing a wide range and large scale) road networks is crucial for traffic flow control, traffic management decision-making, and alleviating urban traffic problems.

[0003] The periodicity and strong spatiotemporal correlations of urban traffic flow give it an internal low-rank characteristic. Therefore, this low-rank property can be exploited to decompose high-dimensional traffic flow data, achieving data dimensionality reduction and uncovering inherent intrinsic structures and spatiotemporal evolution patterns while retaining sufficient original information. Principal Component Analysis (PCA) in statistics is an effective method for feature extraction and data compression. It maps or transforms the original high-dimensional space of variables into a low-dimensional feature space, highlighting similarities and differences between variables and uncovering the implicit intrinsic structures and internal evolution patterns within the original variable system. In practical applications, PCA is an algorithm for dimensionality reduction and multivariate analysis, primarily used in data compression, image processing, neural networks, data mining, pattern recognition, and other fields. In the field of urban transportation, PCA-based methods are currently primarily used in traffic accident / anomaly detection, traffic flow state prediction, and data recovery. However, with the expanded research and widespread application of PCA, significant limitations have been identified. The effectiveness of a range of PCA-based methods is highly sensitive to the selection of certain parameters, such as the determination of the principal component structure of the network flow length pattern and the selection of anomaly thresholds. In addition, PCA requires that the original matrix has low noise and satisfies an independent and identically distributed Gaussian distribution. When the outliers in the network flow are too large or there is too much noise, the accuracy and efficiency of distinguishing the abnormal components from the long regular components will be reduced. Summary of the Invention

[0004] In order to solve the problems existing in the prior art, the present invention provides a road network traffic flow analysis method based on robust principal component tracking. The method collects traffic information from the road network in the target area; performs compensation and reduction processing on the traffic information to obtain the original traffic flow matrix of the traffic information; performs singular value decomposition on the original traffic flow matrix to obtain the original noise matrix; obtains the random matrix constraint parameters of the relaxed principal component tracking method based on the original noise matrix; decomposes the original traffic flow matrix using the relaxed principal component tracking method to obtain a low-rank matrix of the traffic information; decomposes the low-rank matrix using the singular value decomposition method to obtain time-varying characteristic components and spatial characteristic components, and clusters the spatial characteristic components to perform traffic flow analysis. The present invention can effectively eliminate the sudden change traffic flow components contained in the traffic information and the noise components caused by various reasons, thereby improving the accuracy of traffic flow analysis.

[0005] The present invention adopts the following technical solution, a road network traffic flow analysis method based on robust principal component tracking, comprising: Collect traffic information of the target area road network; Performing compensation processing and regulation processing on the traffic information respectively to obtain an original flow matrix of the traffic information; Performing singular value decomposition on the original traffic matrix to obtain an original noise matrix after the singular value decomposition; Obtaining random matrix constraint parameters of a relaxed principal component tracking method according to the original noise matrix; Decomposing the original traffic matrix using a relaxed principal component tracking method according to the random matrix constraint parameters to obtain a low-rank matrix of traffic information; Performing singular value decomposition on the low-rank matrix of the traffic information, and obtaining a time-varying characteristic component and a spatial characteristic component of the traffic information according to the singular value matrix after the singular value decomposition; The k-means clustering algorithm is used to cluster the time-varying characteristic components and the spatial characteristic components, and the road network traffic flow analysis is performed based on the clustering results.

[0006] Furthermore, the compensation process is specifically as follows: The compensation process includes abnormal data inspection and data repair; The abnormal data inspection includes: traffic flow threshold inspection, location speed threshold inspection, occupancy threshold inspection and average effective vehicle length inspection; The data repair operation includes: when there is a lane with missing traffic information in the target area road network, repairing the missing traffic information by using the weighted moving average of the historical traffic information of the lane; When there is a lane with missing traffic information in the target area road network, and the traffic information of the adjacent lanes is not missing, the traffic information of the lane is repaired by using the regression estimation analysis method using the traffic information of the adjacent lanes; When the length of missing traffic information of the target lane reaches the set threshold and the traffic information of the adjacent lanes is also missing, the traffic information of the target lane is repaired using correlation regression estimation based on the traffic information of the two longitudinal adjacent lane sections.

[0007] Furthermore, the protocol processing is specifically as follows: A fixed-point detector is used to collect traffic information of the target road network. The traffic information collected from each lane of the target road network is merged according to the set time interval as the collection cycle, and the original traffic flow matrix of the traffic information is constructed based on the traffic information of all collection cycles within the collection time.

[0008] Furthermore, the original traffic matrix is ​​subjected to singular value decomposition to obtain the original noise matrix after singular value decomposition, which is specifically: Performing singular value decomposition on the original traffic matrix to obtain characteristic components under multiple singular value matrices; Perform fast Fourier decomposition on the characteristic component. If the frequency of the characteristic component after fast Fourier decomposition is a multiple of 12, the characteristic component is a periodic characteristic component. If there is data in the eigencomponent after fast Fourier decomposition that exceeds 5 times the standard deviation of the eigenmatrix, then the eigencomponent is a periodic eigencomponent; The characteristic components that do not belong to the periodic characteristic components and the time mutation characteristic components are regarded as the original noise components; An original noise matrix is ​​constructed according to the original noise components.

[0009] Furthermore, the random matrix constraint parameters of the relaxed principal component tracking method are obtained according to the original noise matrix, specifically: According to the F-norm of the original noise matrix, the random matrix constraint parameters of the relaxed principal component tracking method are obtained.

[0010] Furthermore, the original traffic matrix is ​​decomposed using the relaxed principal component tracking method according to the random matrix constraint parameters, which is expressed as: in, Represents a minimize operation, represents a low-rank matrix, represents the nuclear norm, represents a sparse matrix, represents the 1 norm, is the regularization parameter, represents the original traffic matrix, is the F norm, is the random matrix constraint parameter.

[0011] Furthermore, when the original flow matrix is ​​decomposed, the method further includes: transforming the optimization model established by the relaxed principal component tracking method through Lagrange multipliers; and iteratively solving the transformed optimization model using an accelerated proximal gradient algorithm.

[0012] The beneficial effects of the present invention are as follows: the present invention deconstructs the multi-scale and multi-dimensional road network traffic flow matrix based on the matrix decomposition theory and method, thereby filtering out the noise of the traffic flow matrix. First, the collection area and format of the road network traffic information are set, and the data quality is evaluated and analyzed, the erroneous data and missing data are compensated, and the original flow matrix of the traffic information is generated after the specification processing; secondly, the original flow matrix is ​​decomposed into a low-rank matrix representing the regularity demand through the relaxed principal component tracking method, which includes a sparse flow matrix and a noise matrix caused by large events or sudden accidents, and eliminates the traffic flow components containing mutations and noise components caused by various reasons, thereby effectively retaining the original scale of the long-term regular background traffic flow of the road network and the evolution of the traffic flow state, and eliminating the mutual influence and superimposed interference problems between different components under different analysis problems and research purposes. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0014] Figure 1 1. A flow chart of a method for analyzing road network traffic flow based on robust principal component tracking according to an embodiment of the present invention; Figure 2 This is a schematic diagram of a data collection format of an original traffic matrix according to an embodiment of the present invention; Figure 3 This is a schematic diagram of an algorithm for decomposition steps of a relaxed principal component tracking method according to an embodiment of the present invention; Figure 4 This is a schematic diagram of an algorithm for solving steps of an accelerated proximal gradient algorithm according to an embodiment of the present invention; Figure 5 This is a schematic diagram of an original traffic matrix according to an embodiment of the present invention; Figure 6 Schematic diagram of a decomposed low-rank matrix according to an embodiment of the present invention; Figure 7 This is a schematic diagram of a decomposed sparse matrix according to an embodiment of the present invention; Figure 8 Schematic diagram of a decomposed noise matrix according to an embodiment of the present invention. DETAILED DESCRIPTION

[0015] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0016] A flow chart of a road network traffic flow analysis method based on robust principal component tracking according to an embodiment of the present invention is shown in FIG. Figure 1 As shown, including: Collect traffic information of the target area road network; In the embodiment of the present invention, traffic information is collected from the traffic information collection system of fixed-point detectors on the expressway network in the central urban area of ​​Shanghai. The fixed-point detection system is a means of collecting traffic flow data in real time using detectors statically deployed at fixed positions. It mainly relies on the electromagnetic induction changes generated by magnetic induction coils buried under the road surface to detect the information of vehicles passing through. The collected traffic flow information is then transmitted to the traffic management center system or the information storage system for storage, calculation and analysis. It has 3,575 detectors deployed, covering almost all expressway sections in the Puxi area and some expressway sections in Pudong. The system collects raw data and records the number of vehicle types, occupancy time, passing speed, coil validity and other information of each type passing through the fixed-point coils at 20-second intervals, which constitutes the traffic information of the road network in the embodiment of the present invention.

[0017] Performing compensation processing and regulation processing on the traffic information respectively to obtain an original flow matrix of the traffic information; Generally, the data collected by the fixed-point detector system is of high accuracy. However, during the data collection, transmission and processing process, the system is affected by various factors such as abnormal weather or environment, the working status of the detection equipment, the system status of the information management center, and the communication network. The system cannot strictly complete the collection at the prescribed time interval, resulting in data information with main problems such as data accuracy deviation, data errors, irregular data time points and data loss. These faulty data cannot guarantee the effectiveness of subsequent analysis and application. Therefore, it is necessary to conduct targeted data quality inspection and evaluation on the original data, repair erroneous and invalid data, make up for lost data, improve data quality, and ensure the effectiveness of data use.

[0018] In the embodiment of the present invention, traffic information is processed mainly by means of compensation processing and regulation processing techniques. Among them, compensation processing includes abnormal data detection and data repair, specifically: Abnormal data detection: Before repairing the data, it is necessary to detect and determine whether the data record belongs to the category of abnormal data. First, a data missing test is performed. By analyzing the collected traffic information, it is found that in addition to the missing data at the collection time point in the usual sense, the data collection also includes continuous 0 values. That is, although the data contains time point collection records, all record values ​​are 0 for a long time. Secondly, a data anomaly test is performed, including traffic flow threshold test, location speed threshold test, occupancy threshold test and average effective vehicle length test. Finally, a traffic flow parameter consistency test is performed. The specific judgment conditions are shown in Table 1: Table 1 Schematic table of abnormal data detection and judgment conditions For the case where each section lane detector is not completely damaged, in an embodiment of the present invention, based on the linear correlation between the detection data of adjacent lanes, the historical detection data collected by the section detector in the lane where traffic information is missing and the data collected by the section detector of its adjacent lane are used to repair the traffic information data of that period; specifically, for a lane where the section lane detector data is missing for less than 1 hour, the weighted moving average estimation is performed by reading the historical detection data of the first n time periods collected by the section monitor in the lane to repair the missing detection data of the lane; for a lane where the section detector data is missing for more than 1 hour, a regression estimation method is used based on the detection data collected by the section detector in its adjacent lane to repair the missing traffic information data in the lane where the section detector is missing for more than 1 hour; if the section detector data is missing for more than 1 hour and the section detector data of its adjacent lane is also missing, the missing traffic information data is estimated based on the correlation between the traffic flow data of the two adjacent longitudinal road sections; the expression of the weighted moving average algorithm in the embodiment of the present invention is: in, represents the traffic information of the i-th road section at the k-th moment, Indicates that the i-th road segment is in Traffic information at all times, is the weighted coefficient corresponding to the traffic information at the j moments before the k moment, , It means there are a total of the first n moments, and its value is taken according to the principle that the closer to k, the larger the value, and the farther away from k, the smaller the value.

[0019] In the embodiment of the present invention, the data reduction operation is specifically as follows: In order to reduce the spatiotemporal scope of data operations, it is necessary to perform reduction processing on the traffic status data. Fixed-point detection coils are arranged on the road section by lane. Therefore, the traffic flow data of each lane on the expressway section must be merged into the uplink and downlink traffic flow data of the section. In the embodiment of the present invention, fixed-point detectors are used to collect traffic flow data. Therefore, the traffic flow data of each lane is reduced according to the following formula, which is expressed as: in, Indicates the upstream and downstream traffic flow data after protocol processing. Indicates the The detector in Traffic volume collected from lanes, is the total number of lanes, represents the traffic flow in m time periods, Indicates the In the time interval Traffic flow at all times.

[0020] A 20-second time interval will generate 4,320 data records a day, and the fluctuation pattern of the collected data is too detailed, which is too complicated for the analysis of traffic flow data. Therefore, in the embodiment of the present invention, the data sets with a time interval of 20 seconds are merged into traffic flow data with the required time span. Since the policy on restricting the use of foreign license plates has been adjusted to a time span of 30 minutes, the original data are also aggregated and merged according to the 30-minute time period in the embodiment of the present invention.

[0021] In the embodiment of the present invention, the traffic data collected by the fixed-point detectors on the expressway is processed and simplified. With a time period of 30 minutes, there are 1056 (24 days * 48 time periods) time periods, and the road section j is defined at time t i The traffic flow is x ij， Then we can get an original traffic matrix with a dimension of 1056×529. The jth column of the matrix represents the traffic flow of the jth road section in the selected data period, and the ith row represents the road traffic flow of all road sections in the same data collection period. The matrix format is as follows: Figure 2 As shown in the figure, each row is a collection cycle, for example Indicates that the acquisition period of the first row in the matrix is The road traffic flow between them; each section in the figure is each road segment.

[0022] Performing singular value decomposition on the original traffic matrix to obtain an original noise matrix after singular value decomposition; In an embodiment of the present invention, singular value decomposition is first performed on the original traffic matrix to obtain characteristic components under multiple singular value matrices; the characteristic components are further subjected to fast Fourier decomposition. If the frequency of the characteristic component after fast Fourier decomposition is a multiple of 12, the characteristic component is a periodic characteristic component; if there is data in the characteristic component after fast Fourier decomposition that exceeds 5 times the standard deviation of the characteristic matrix, the characteristic component is a periodic characteristic component; the characteristic components that are not periodic characteristic components and time mutation characteristic components are regarded as original noise components; and an original noise matrix is ​​constructed based on the original noise components.

[0023] Specifically: In the embodiment of the present invention, the original traffic matrix is ​​subjected to singular value decomposition, and the original traffic matrix X is decomposed into In the form of multiple singular value matrices, the characteristic components are obtained, where is the left singular value matrix, is the right singular value matrix, It is an orthogonal matrix composed of diagonal elements; further fast Fourier decomposition is performed on the eigenvalues ​​of each column in the matrix, and the periodicity of the eigenvalues ​​is identified by spectrum analysis: when the peak of the main frequency of the eigenvalue after fast Fourier decomposition is an integer multiple of 12, the eigenvalue is determined to be a periodic eigenvalue; at the same time, a sliding window standard deviation test is used. If there are three consecutive data points in the eigenvalue after fast Fourier decomposition that exceed the range of 5 times the standard deviation of the eigenvalue matrix, the eigenvalue is a time mutation eigenvalue; the components that meet the characteristics of low amplitude distribution and non-periodic fluctuation are classified as original noise components; finally, the filtered original noise components are reconstructed by singular value sorting through weighted linear combination to construct an original noise matrix that conforms to the dimensions of the original matrix.

[0024] Obtaining random matrix constraint parameters of a relaxed principal component tracking method according to the original noise matrix; In the embodiment of the present invention, the estimation algorithm of the random matrix constraint parameters is as follows Figure 3 As shown, after the original noise matrix is ​​obtained by decomposing the original traffic matrix, the random matrix constraint parameters for relaxing the principal component tracking method in the embodiment of the present invention can be obtained by taking the F norm of the original noise matrix.

[0025] The relaxed principal component tracking method is used to decompose the original traffic matrix according to the random matrix constraint parameters to obtain the low-rank matrix of traffic information.

[0026] The original traffic matrix is ​​composed of a low-rank matrix with a low-rank structure or a low-rank structure approximate to a low-rank structure and a randomly distributed sparse noise matrix. However, there are some traffic demands in the road network that cannot be simply classified into low-rank parts and sparse parts. In the embodiment of the present invention, the Relaxed Principal Component Pursuit (RPCP) method is used to establish an optimization model to decompose the original traffic matrix. The optimization model no longer requires the original matrix to be strictly and accurately decomposed into the sum of the low-rank matrix and the sparse matrix, but uses a new variable For the F-norm constraint problem of random demand, the decomposition is allowed to have a certain difference range to simulate random demand fluctuations, which can be expressed as: in, represents minimization, represents a low-rank matrix, represents the nuclear norm, represents a sparse matrix, represents the 1 norm, is the regularization parameter, represents the original traffic matrix, is the F norm, is the random matrix constraint parameter; specifically, represents the objective function to be minimized, whose purpose is to find a suitable low-rank matrix and sparse matrices , making The value of is the smallest, Represents a low-rank matrix The nuclear norm is the sum of the singular values ​​of the matrix, which is used to constrain the low rank of the matrix. Because most of the singular values ​​of the low rank matrix are zero or close to zero, minimizing the nuclear norm can promote the matrix Has low-rank properties. Represents a sparse matrix The 1-norm is the sum of the absolute values ​​of the matrix elements. Minimizing the 1-norm can make the matrix Most of the elements in are zero, that is, it is sparse; is a regularization parameter used to balance the low-rank matrix The nuclear norm and sparse matrix of The importance of the 1-norm, The larger the value, the more emphasis is placed on sparse matrices. Sparsity; The smaller the value, the more emphasis is placed on low-rank matrices. The low rank property of Represents the constraint condition, requiring that the error between the original traffic matrix and the sum of the decomposed low-rank matrix and sparse matrix be within a certain range. Represents the original traffic matrix, that is, the matrix that needs to be decomposed, : represents the F norm (Frobenius norm) of the matrix. The F norm is the square root of the sum of the squares of the matrix elements and is used to measure the error between matrices; is the random matrix constraint parameter, which limits the maximum allowable error between the original matrix and the decomposition result.

[0027] In the embodiment of the present invention, in the process of decomposing the original flow matrix by adopting the relaxed principal component tracking method, the accelerated proximal gradient algorithm is adopted to solve it.

[0028] The Accelerated Proximal Gradient (APG) algorithm is an iterative algorithm for solving convex optimization problems. It combines the ideas of the gradient descent method and the proximal operator, and accelerates convergence by introducing an acceleration mechanism (such as Nesterov acceleration). In this embodiment of the present invention, the Accelerated Proximal Gradient (APG) algorithm is used to iteratively update the values ​​of the low-rank matrix L and the sparse matrix S, so that the objective function gradually decreases and the optimal solution is eventually found. Before solving the algorithm, the problem needs to be expressed as shown in the following form using Lagrange multipliers: Where, represents minimizing low-rank matrix and coefficient matrix operations, Represents a low-rank matrix The kernel function, Represents a sparse matrix The 1-norm of represents the square of the F norm, X is the original traffic matrix, and are the regularization parameters respectively.

[0029] Then, by Figure 4 The algorithm steps shown are used to solve the problem, thereby eliminating the traffic flow components containing mutations and noise components caused by various reasons, and retaining the long-term regular background traffic flow of the road network and the original scale of the traffic flow state evolution. The embodiment of the present invention provides the following Figure 5 The decomposition results of the original traffic matrix shown are as follows: Figure 6 、 Figure 7 and Figure 8 As shown, Figure 6 is the low-rank matrix after decomposition; Figure 7 is the decomposed sparse matrix; Figure 8 is the noise matrix after decomposition.

[0030] The low-rank matrix of traffic information is subjected to singular value decomposition, and the time-varying characteristic components and spatial characteristic components of the traffic information are obtained according to the singular value matrix after singular value decomposition.

[0031] The embodiment of the present invention further decomposes the reconstructed long regular traffic flow matrix low rank matrix L using a singular value decomposition (SVD) method, as shown in the following formula: Where: U is the left singular matrix; S is the singular value diagonal matrix; V is the right singular matrix, and the superscript T is the transpose of the matrix. represents the column vector of the left singular matrix, represents the singular values ​​in the singular value diagonal matrix, represents the column vector of a right singular matrix, and i represents an element in the matrix.

[0032] The singular value diagonal matrix S obtained by decomposition can represent the information contribution weight of the corresponding eigenvector, and is arranged in descending order of value. The larger the singular value, the greater the information energy contained in each eigenvector. The cumulative sum of squares of the singular values ​​is calculated, that is, the singular value variance contribution; the left singular matrix U obtained by singular value decomposition is the time fluctuation characteristic of the road network traffic flow, and the time-varying eigenlaw of the traffic flow state of the entire road network is analyzed; the right singular matrix V obtained by SVD decomposition is the spatial distribution of the corresponding time variation characteristics. The fluctuation characteristics of the first two dimensions of time mainly explain the variation characteristics of the extended period.

[0033] After performing singular value decomposition on the low-rank matrix, time-varying feature components and spatial feature components that can characterize road network traffic information are obtained. For the time-varying feature component, it includes the characteristics of road network traffic information that change with time, such as changes in traffic flow during peak hours in the morning and evening. The spatial feature component includes the characteristics of different road sections and regions, such as the difference in traffic flow between the central business district and the suburbs of the city. Therefore, the embodiment of the present invention uses the k-means clustering algorithm to cluster the spatial feature components. The specific means adopted in the clustering process can refer to any implementation method in the existing technology, and the present invention does not impose any restrictions; thus, road network traffic flow analysis is performed based on the clustering results.

[0034] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A road network traffic flow analysis method based on robust principal component tracking, characterized in that: include: Collect traffic information of the target area road network; Performing compensation processing and regulation processing on the traffic information respectively to obtain an original flow matrix of the traffic information; Performing singular value decomposition on the original traffic matrix to obtain an original noise matrix after the singular value decomposition; Obtaining random matrix constraint parameters of a relaxed principal component tracking method according to the original noise matrix; Decomposing the original traffic matrix using a relaxed principal component tracking method according to the random matrix constraint parameters to obtain a low-rank matrix of traffic information; Performing singular value decomposition on the low-rank matrix of the traffic information, and obtaining a time-varying characteristic component and a spatial characteristic component of the traffic information according to the singular value matrix after the singular value decomposition; The time-varying characteristic components and the spatial characteristic components are clustered using a k-means clustering algorithm, and road network traffic flow analysis is performed based on the clustering results.

2. The method for analyzing road network traffic flow based on robust principal component tracking according to claim 1, characterized in that: The compensation process is specifically as follows: The compensation process includes abnormal data inspection and data repair; The abnormal data inspection includes: traffic flow threshold inspection, location speed threshold inspection, occupancy threshold inspection and average effective vehicle length inspection; The data repair operation includes: when there is a lane with missing traffic information in the target area road network, repairing the missing traffic information by using the weighted moving average of the historical traffic information of the lane; When there is a lane with missing traffic information in the target area road network, and the traffic information of the adjacent lanes is not missing, the traffic information of the lane is repaired by using the regression estimation analysis method using the traffic information of the adjacent lanes; When the length of missing traffic information of the target lane reaches the set threshold and the traffic information of the adjacent lanes is also missing, the traffic information of the target lane is repaired using correlation regression estimation based on the traffic information of the two longitudinal adjacent lane sections.

3. The method for analyzing road network traffic flow based on robust principal component tracking according to claim 1, characterized in that: The specific protocol processing is as follows: A fixed-point detector is used to collect traffic information of the target road network. The traffic information collected from each lane of the target road network is merged according to the set time interval as the collection cycle, and the original traffic flow matrix of the traffic information is constructed based on the traffic information of all collection cycles within the collection time.

4. The method for analyzing road network traffic flow based on robust principal component tracking according to claim 1, characterized in that: Perform singular value decomposition on the original traffic matrix to obtain the original noise matrix after singular value decomposition, specifically: Performing singular value decomposition on the original traffic matrix to obtain characteristic components under multiple singular value matrices; Perform fast Fourier decomposition on the characteristic component. If the frequency of the characteristic component after fast Fourier decomposition is a multiple of 12, the characteristic component is a periodic characteristic component. If there is data in the eigencomponent after fast Fourier decomposition that exceeds 5 times the standard deviation of the eigenmatrix, then the eigencomponent is a periodic eigencomponent; The characteristic components that do not belong to the periodic characteristic components and the time mutation characteristic components are regarded as the original noise components; An original noise matrix is ​​constructed according to the original noise components.

5. The method for analyzing road network traffic flow based on robust principal component tracking according to claim 1, characterized in that: The random matrix constraint parameters of the relaxed principal component tracking method are obtained according to the original noise matrix, specifically: According to the F-norm of the original noise matrix, the random matrix constraint parameters of the relaxed principal component tracking method are obtained.

6. The method for analyzing road network traffic flow based on robust principal component tracking according to claim 1, characterized in that: The original traffic matrix is ​​decomposed using the relaxed principal component tracking method according to the random matrix constraint parameters, which is expressed as: in, Represents a minimize operation, represents a low-rank matrix, represents the nuclear norm, represents a sparse matrix, represents the 1 norm, is the regularization parameter, represents the original traffic matrix, is the F norm, is the random matrix constraint parameter.

7. The method for analyzing road network traffic flow based on robust principal component tracking according to claim 1, characterized in that: When the original flow matrix is ​​decomposed, the method further includes: transforming the optimization model established by the relaxed principal component tracking method through Lagrange multipliers; and iteratively solving the transformed optimization model using an accelerated proximal gradient algorithm.

Images