A path-based method for tracing traffic congestion sources in urban expressway corridors

Abstract

The present invention relates to a path-based urban expressway corridor traffic congestion tracing method, comprising the following steps: training phase: obtaining the travel speed and flow detection data time series of each lane in the required analysis period, constructing the section-level travel speed and flow information spatiotemporal matrix of the required analysis period; constructing the path traffic state tensor of the required analysis period based on the section-level travel speed and flow information spatiotemporal matrix; inputting the path traffic state tensor into a pre-trained traffic congestion tracing model, obtaining the congestion correlation of each section in the required analysis period, and mining the congestion tracing according to the congestion correlation. Compared with the prior art, the present invention establishes a complete urban expressway corridor traffic congestion tracing method with the path as the basic unit and considering the propagation law of upstream and downstream traffic states, which has important practical significance.

Description

Technical Field

[0001] The present invention relates to the field of intelligent transportation technology, and in particular to a path-based urban expressway corridor traffic congestion tracing method. Background Art

[0002] With the rapid advancement of urbanization and motorization, the number of motor vehicles has increased rapidly, and the contradiction between urban traffic supply and demand has become increasingly serious. Expressways are the main skeleton and artery of urban roads. They have great transportation benefits and undertake key traffic functions, which have an important impact on the efficiency of urban road traffic operations. Among them, expressway corridors refer to the main channels of expressways with large traffic demand and traffic capacity. They are frequent locations of traffic congestion in expressway networks. Therefore, alleviating urban expressway traffic congestion is an important part of urban traffic congestion management. Tracing the source of expressway traffic congestion is the basis for formulating solutions to improve expressway traffic congestion, and it has always been one of the industry's difficult and painful issues. Accurate and effective tracing of traffic congestion can provide decision-making support for the formation of traffic control strategies by traffic management departments to avoid or alleviate urban traffic congestion.

[0003] At present, the research on traffic congestion tracing at home and abroad can be roughly divided into two categories: one is the path flow estimation method, and the other is the congestion correlation analysis method. The path flow estimation method is based on traditional section detection data, vehicle equipment identification data, mobile data sources and multi-source fusion data sources. It has a certain research foundation and can be further divided into two ideas: analytical models and machine learning models. Analytical methods include maximum entropy models, least squares models, two-level programming models, Bayesian inference, particle filtering methods, and hybrid methods of the above models. Under the analytical framework, the application of the model is relatively mature and extensive, but it is still limited by the detector coverage, model constraints and prior information requirements. Machine learning methods use artificial intelligence technology to improve the model's ability to optimize high-dimensional problems, including artificial neural networks, deep neural networks, convolutional neural networks, and stacked autoencoders. With sufficient data source support, machine learning methods are particularly suitable for multi-source fusion data source scenarios, which can bring information gain and improve the accuracy of path flow estimation. Congestion correlation research is to quantitatively or qualitatively analyze the factors related to congestion points and segments. Compared with the path flow estimation method, this type of research is less common. Domestic and foreign scholars usually use the perspective of static congestion segment correlation or dynamic congestion propagation characteristics to conduct research, and the main method is association rule mining theory. In recent years, research in this field has mainly focused on how to mine and analyze the dynamic traffic congestion propagation law, such as spatiotemporal congestion subgraphs, causal association analysis methods, and Bayesian inference. In the above studies, path flow estimation focuses on the backtracking of congested traffic flow, believing that it is an important factor causing congestion, while congestion correlation studies the traffic congestion propagation law and locates the congestion correlation through certain relationships. Path flow estimation has a certain research foundation for different research scenarios (urban expressways and ground roads, etc.) and different data conditions, while congestion correlation analysis mainly reflects the congestion propagation law under different road network structures or traffic volume conditions by building a more universal and accurate model.

[0004] However, the accuracy and effectiveness of the above two methods in tracing traffic congestion sources need to be improved. Summary of the invention

[0005] The purpose of the present invention is to provide a path-based urban expressway corridor traffic congestion tracing method that simultaneously considers the vehicle flow including the starting and ending points passing through the congestion point segment and the traffic status so as to more accurately and effectively obtain the traffic congestion tracing.

[0006] The purpose of the present invention can be achieved by the following technical solutions:

[0007] A path-based urban expressway corridor traffic congestion tracing method comprises the following steps:

[0008] Obtain the travel speed and flow detection data time series of each lane in the required analysis period, and construct the section-level travel speed and flow information spatiotemporal matrix of the required analysis period;

[0009] Based on the section-level travel speed and flow information spatiotemporal matrix, construct a path traffic state tensor for a required analysis period;

[0010] The path traffic state tensor is input into a pre-trained traffic congestion source tracing model to obtain the congestion correlation of each section in the required analysis period, and the congestion source is mined according to the congestion correlation.

[0011] Furthermore, the specific steps of constructing the spatiotemporal matrix of section-level travel speed and flow information for the required analysis period include:

[0012] Obtain the travel speed and flow detection data time series of each lane in the required analysis period;

[0013] Determine the time length of aggregation, perform double aggregation on the travel speed and flow detection data time series from the two dimensions of time and space, and obtain the travel speed and flow information at the spatial section where each detection point is located;

[0014] The travel speed and flow information at the spatial section where each detection point is located are organized into a matrix form to obtain the section-level travel speed and flow information spatiotemporal matrix.

[0015] Furthermore, the expressions for the travel speed and flow information at the spatial section where each detection point is located are obtained as follows:

[0016]

[0017]

[0018]

[0019]

[0020]

[0021] Among them, V i k , are the travel speed and flow rate of the kth aggregation time period of the detection section i after double aggregation, are the flow and travel speed of lane j in section i in the kth aggregation time period after time aggregation, T is the length of each aggregation time period, t is the sampling time interval of the detection point, and n is the expansion multiple of the time period length. They are the travel speed and flow information of lane j in detection section i at the mth sampling moment obtained by the fixed-point detector.

[0022] Furthermore, the expressions of the spatiotemporal matrix of the cross-section level travel speed and flow information are respectively:

[0023]

[0024]

[0025] Among them, S v , S q are the travel speed and flow space-time matrices, V i k is the travel speed of the kth aggregation time period of the detection section i after double aggregation, It is the flow rate of the kth aggregation time period of the detection section i after double aggregation, i = 1, 2, 3 ..., L, k = 1, 2, ..., H, L is the number of sections where the effective fixed-point detectors are located in the area, and H is the total number of aggregation time periods.

[0026] Furthermore, the specific steps of constructing the path traffic state tensor for the required analysis period include:

[0027] Obtain the connection relationship and distance length sequence between each section according to the physical space topology;

[0028] The forward and backward transmission process of the path is calculated from the two dimensions of time and space to obtain the path time and space point position at the spatial section where each detection point is located;

[0029] The path spatiotemporal point positions at the spatial sections where the respective detection points are located are indexed to the section-level travel speed and flow information spatiotemporal matrix, and the path traffic state tensors at the spatial sections where the respective detection points are located are obtained by combining them.

[0030] Furthermore, the expression of the path traffic state tensor is:

[0031]

[0032]

[0033] R=[R 1,1 ,...,R i,L ,R 2,1 ,...,R 2,L ,...,R H,1 ,...,R H,L ]

[0034] Among them, R h,l is the path time-space point at section l at time h, h = 1, 2, ..., H, l = 1, 2, 3 ..., L, L is the number of sections where effective fixed-point detectors are located in the area, H is the number of analysis time periods, Sv and S q are the space-time matrices of travel speed and flow, respectively.

[0035] Furthermore, the specific steps of obtaining a pre-trained traffic congestion tracing model include:

[0036] Based on the time series of travel speed and flow detection data of each lane, a historical section-level travel speed and flow information spatiotemporal matrix is ​​constructed;

[0037] Based on the historical section-level travel speed and flow information spatiotemporal matrix, the historical path traffic state tensor is constructed and training samples are formed;

[0038] Build deep neural network models;

[0039] Based on the training samples, the deep neural network model is trained to obtain optimal model parameters, and the training is completed to obtain a pre-trained traffic congestion tracing model.

[0040] Furthermore, the gradient descent method with adaptive learning rate is used to optimize the parameters of the deep neural network model.

[0041] Furthermore, the optimization function used in the training process is:

[0042]

[0043] Among them, Θ is the parameter to be optimized in the deep neural network model, is the label of the pth training sample, is the model output corresponding to the pth training sample, and P is the training sample size.

[0044] Furthermore, the deep neural network model includes a convolutional long short-term memory network, a soft attention module and a deformable convolutional neural network cascaded in sequence.

[0045] Compared with the prior art, the present invention has the following beneficial effects:

[0046] (1) The present invention obtains the travel speed and flow detection data time series of each lane through a fixed-point detector, and further constructs a path traffic state tensor, which fully considers the impact of the traffic state of each point on the path on the upstream and downstream of the current point, as well as the impact of the traffic state of the current point on the downstream point, thereby learning the impact mode and impact degree of the traffic state of each path from both the time dimension and the space dimension. It can more accurately and effectively trace the source of traffic congestion and provide a mitigation method for solving the traffic congestion problem on urban expressways.

[0047] (2) The present invention uses traffic flow detection parameters collected by fixed-point detectors to accurately describe traffic status characteristics and trace traffic congestion. The tracing results are highly reliable and can provide a solid foundation for engineering applications.

[0048] (3) The present invention fully considers the traffic state propagation law between points on the path when constructing the path traffic state tensor, thereby improving the accuracy of traffic congestion tracing, which has important practical significance for the traffic congestion tracing method of urban expressway corridors.

[0049] (4) The present invention captures the correlation changes and correlations between spatial and temporal series by adding a soft attention mechanism to the deep neural network, effectively improving the congestion tracing method's ability to extract the spatiotemporal correlation features of congestion, and outputs the degree of influence of different path flows on congestion through attention weights, thereby achieving congestion tracing. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 is a flow chart of the method of the present invention;

[0051] Figure 2 It is a structural block diagram of the traffic congestion tracing model of the present invention. DETAILED DESCRIPTION

[0052] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0053] This embodiment provides a path-based urban expressway corridor traffic congestion tracing method. Figure 1 As shown, the method comprises the following steps:

[0054] S1. Obtain the travel speed and flow detection data time series of each lane in the required analysis period, and construct the section-level travel speed and flow information spatiotemporal matrix of the required analysis period.

[0055] The build steps are:

[0056] (a) Obtain the travel speed and flow detection data time series of each lane during the required analysis period.

[0057] In this embodiment, the travel speed and flow detection data time series of each lane are obtained through a fixed-point detector.

[0058] (b) Determine the time length of aggregation, and perform double aggregation on the travel speed and flow detection data time series from both time and space dimensions to obtain the travel speed and flow information at the spatial section where each detection point is located.

[0059] The travel speed and flow information at the spatial section of each detection point is obtained by the following formula:

[0060]

[0061]

[0062]

[0063]

[0064]

[0065] Among them, V i k , are the travel speed and flow rate of the kth aggregation time period of the detection section i after double aggregation, are the flow and travel speed of lane j in section i in the kth aggregation time period after time aggregation, T is the length of each aggregation time period, t is the sampling time interval of the detection point, and n is the expansion multiple of the time period length. They are respectively the travel speed and flow rate of lane j in detection section i obtained by the fixed-point detector at the mth sampling moment.

[0066] (c) The travel speed and flow information at the spatial section where each detection point is located are organized into a matrix form to obtain the section-level travel speed and flow information spatiotemporal matrix.

[0067] The spatiotemporal matrix of the cross-sectional travel speed and flow information in step (c) is S v , S q Respectively expressed as:

[0068]

[0069]

[0070] Among them, S v , S q are the travel speed and flow space-time matrices, V i k is the travel speed of the kth aggregation time period of the detection section i after double aggregation, It is the flow rate of the kth aggregation time period of the detection section i after double aggregation, i = 1, 2, 3 ..., L, k = 1, 2, ..., H, L is the number of sections where the effective fixed-point detectors are located in the area, and H is the total number of aggregation time periods.

[0071] S2. Constructing a path traffic state tensor for the required analysis period based on the section-level travel speed and flow information spatiotemporal matrix for the required analysis period.

[0072] The path traffic status tensor constructed in step S2 is constructed in the following way:

[0073] (a) Obtain the connection relationship and distance length sequence between each section based on the physical space topology.

[0074] (b) The forward and backward transmission process of the path is calculated from the two dimensions of time and space to obtain the path time and space point positions at the spatial section where each detection point is located.

[0075] In step (b), the path time-space point position of each detection point at the spatial section is obtained by the following formula:

[0076] i. Initialize start = 1, path time and space point set

[0077] ii. If start mod c is 0, go to iii; otherwise go to iv.

[0078] iii. x = x + 1, .add((x,y)), convert to v.

[0079] iv. Calculate d i -Δt decision ×S v [x,n i ], if there is a value less than 0, then For the unselected n i Update its distance to d i =d i -Δt decision ×S v [x,n i ], turn to v.

[0080] v.start=start+1. If If |S route |=|S correlation [y]| then stop the iteration, otherwise go to i; otherwise, go to ii.

[0081] Among them, Δt collection is the detector detection frequency, Δt decision is the time length of the path propagation, Δt history is the analysis duration, S correlation is the connection relationship and distance length sequence between each section, d i is the distance to the i-th detection point, n i is the i-th detection point, S route is the path set of the current calculation section, S vis the space-time matrix of travel speed, (x, y) is the space-time point position, R h,l is the path time-space point at section l at time h, h = 1, 2, ..., H, l = 1, 2, 3 ..., L, L is the number of sections where effective fixed-point detectors are located in the area, and H is the number of analysis time periods.

[0082] (c) The spatiotemporal points of the path at the spatial section where each detection point is located are indexed into the spatiotemporal matrix of the section-level travel speed and flow information, and the path traffic state tensor at the spatial section where each detection point is located is obtained by combining them.

[0083] The path traffic state tensor R in step (c) is expressed as:

[0084]

[0085]

[0086] R=[R 1,1 ,...,R i,L ,R 2,1 ,...,R 2,L ,...,R H,1 ,...,R H,L ]

[0087] Among them, R h,l is the path time-space point at section l at time h, h = 1, 2, ..., H, l = 1, 2, 3 ..., L, L is the number of sections where effective fixed-point detectors are located in the area, H is the number of analysis time periods, S v , S q are the space-time matrices of travel speed and flow, respectively.

[0088] S3. Input the path traffic state tensor of the required analysis period into a pre-trained traffic congestion tracing model to obtain the path importance information matrix of each section of the required analysis period, and complete the traffic congestion tracing of the urban expressway corridor.

[0089] The specific steps to obtain a pre-trained traffic congestion tracing model include:

[0090] Based on the time series of travel speed and flow detection data of each lane, a historical section-level travel speed and flow information spatiotemporal matrix is ​​constructed;

[0091] Based on the historical section-level travel speed and flow information spatiotemporal matrix, the historical path traffic state tensor is constructed and training samples are formed;

[0092] Building a deep neural network model based on the historical path traffic state tensor;

[0093] Based on the training samples, the deep neural network model is trained to obtain optimal model parameters, and the training is completed to obtain a pre-trained traffic congestion tracing model.

[0094] In the step of pre-training the deep neural network model, constructing the historical section-level travel speed, flow information spatiotemporal matrix and the historical path traffic state tensor is the same as steps S1 and S2 and will not be repeated here.

[0095] The specific operations of forming training samples in this step are:

[0096] Determine the training sample size P, randomly select the starting time point of the i-th training sample, and obtain the path traffic state tensor in the h_in aggregate time periods before and after the starting time point: And the spatiotemporal matrix of section-level travel speed and flow information within the h_out aggregate time periods contained in the starting point is: Will As the input of the training sample, As the label of the training sample, p = 1, 2, ..., P.

[0097] The deep neural network model includes a convolutional long short-term memory network, a soft attention module and a deformable convolutional neural network cascaded in sequence. The convolutional long short-term memory network includes a first convolutional long short-term memory network layer and a second convolutional long short-term memory network layer cascaded in sequence. The deformable convolutional neural network includes a first fully connected layer, a second fully connected layer, a deformable convolutional layer, a first convolutional layer, a second convolutional layer, a pooling layer, a third fully connected layer, a fourth fully connected layer and an output layer cascaded in sequence.

[0098] like Figure 2 As shown, this embodiment builds a traffic congestion tracing model based on a deep neural network model, and the specific steps are:

[0099] Step 1: As input to a deep neural network model.

[0100] Step 2: After the convolutional long short-term memory network, the specific operations are:

[0101] make where f ConvLSTM represents the activation function of the convolutional LSTM network layer, Indicates that ConvLSTM represents the matrix operation process of convolutional long short-term memory neural network;

[0102] make

[0103] Step 3: where f SoftAttentionrepresents the activation function of the soft attention module, and SoftAttention represents the matrix operation process of the soft attention module;

[0104] Step 4: After the deformable convolutional neural network, the specific operations are:

[0105] make Flatten represents the matrix operation and activation process of the fully connected layer;

[0106] make

[0107] make Where Reshape represents the dimension reconstruction process;

[0108] make where f DeConv represents the activation function of the deformable convolution layer, and DeConv represents the deformable convolution process of the deformable convolution layer;

[0109] make where f Conv Represents the activation function of the convolutional layer, and Conv represents the convolution process of the convolutional layer;

[0110] make

[0111] make Where Pool represents the pooling process of the deformable convolutional neural network;

[0112] make

[0113] make

[0114] make

[0115] Step 5: Use the gradient descent method with adaptive learning rate to optimize the parameters of the deep neural network model. The optimization function is:

[0116]

[0117] Among them, Θ is the parameter to be optimized in the model, is the label of the pth training sample, is the model output corresponding to the pth training sample, and P is the training sample size;

[0118] Step 6: Repeat steps 1 to 5 until the deep neural network model converges and the training is completed to obtain the traffic congestion tracing model.

[0119] According to the above-trained traffic congestion tracing model, the path traffic state tensor of the required analysis period is input into the model to obtain the congestion correlation of each section of the required analysis period, and the congestion source is mined according to the congestion correlation.

[0120] To verify the effectiveness of the above method, this embodiment uses a certain expressway as a scenario, constructs a simulation scenario of an urban expressway corridor, and analyzes the effect of traffic congestion tracing results on the application of relevant traffic control measures.

[0121] This embodiment is built based on the Tensorflow deep learning framework. After comparison and selection, the parameters of the traffic congestion tracing model are determined as follows: each sample contains data of size 319×60×1×2 (319 is the total number of path trajectories, for 25 main lines and 11 ramp sections respectively; 60 is the number of trajectory points generated within 60 minutes with a trajectory generation interval of 1 minute, where the value is 0 after reaching the boundary condition; 2 is two traffic state characteristic parameters, namely travel speed and flow), and a label of size 72 (72 is two traffic characteristic parameters of 36 sections). The time span involved in each sample is 55 minutes (first 30 minutes + current 5 minutes + next 30 minutes - overlapping 10 minutes). The model inputs 319 path states into a multi-path ConvLSTM module in groups, i.e., main line 1 to main line 25, ramp 1 to ramp 11, where the input step size is the number of paths in the corresponding section. In order to avoid the influence of multiple 0 values ​​when training ConvLSTM, a mask operation is performed (corresponding 0 values ​​are masked and do not participate in training). The output of the ConvLSTM module is the result of dimension increase of each path feature. Soft Attention is also calculated by section, and the results are weighted in each module to obtain the final state dimension increase result of each section (a vector with a length of 384). In the concatenation layer, the 36×1 section is reconstructed into 6×6, and the feature dimension is compressed to 64 dimensions, which is equivalent to the compression and reconstruction of the original feature. In the deformable convolutional neural network, 3×3 deformable convolution and normal convolution operators are used to extract high-dimensional features in turn. The model finally outputs a vector of size 72 as the travel speed and flow results of the section. The overall idea of ​​the model is to define the time range based on the original data to generate a path state set → use a multi-path ConvLSTM network to increase the dimension of the feature → use the Soft Attention module to comprehensively consider the influence of each path and output it weightedly → use the splicing layer to compress the feature dimension and reconstruct the image into a generalized structure → extract the topological result relationship of each section and the spatiotemporal correlation characteristics of the features through the deformable convolutional network → output the current travel speed and flow value of each section. The average absolute error of the model fitting reaches 0.623km / h, which lays a solid foundation for the accurate congestion tracing analysis task. The soft attention module in the model can trace the source of traffic congestion based on the path unit. When the model fits the travel speed and flow of all sections every 5 minutes, it can characterize the influence of different path states on traffic congestion, and the traffic congestion tracing results can be obtained according to the weight arrangement.

[0122] According to the results of tracing the source of traffic congestion, the influence of different paths on traffic congestion can be obtained, and four different traffic control schemes are adopted: control the flow of the complete path, that is, control the traffic of the path with a large influence to pass through other ways; control the flow of entry, that is, control the flow of the on-ramp of the path with a large influence; divert to other exits, that is, divert the off-ramp flow of the path with a large influence to other ramps; control the flow of other complete paths, that is, control the traffic of other paths with a lower influence. In various flow control links, the flow of the path with the greatest influence is taken as the benchmark, and 10%, 20%, 30%, 40% and 50% of the traffic flow are regulated respectively. The results are shown in Table 1. When no traffic control is performed, the average travel speed is 29.2km / h, the average number of stops is 1.7 times, and the average parking delay per vehicle is 5.66 seconds.

[0123] Table 1 Traffic control effect verification based on traffic congestion tracing results

[0124]

[0125] As can be seen from Table 1, when the controlled flow rate increases from 10% to 50%, the benefits of each control method are improved, except that the effect of diverting to other exits is relatively unstable; since the processing of the flow rate to other intersections in the simulated diversion path is relatively simple, it may cause other traffic congestion problems, resulting in a decrease in travel speed (-1.47%), an increase in the number of stops (+8.24%), and an increase in delays (+6.71%); the effects of controlling the incoming flow and controlling the flow rate of other paths are similar, because when the incoming flow is uniformly controlled, the flow rate of each path is actually reduced, including paths with a large degree of influence and paths with a small degree of influence; the effect of regulating the path with the greatest degree of influence is the most obvious, and the travel speed can be increased by up to 23.36%, the number of stops can be reduced by up to 29.41%, and the delay can be reduced by up to 43.82%. It can be seen that based on the traffic congestion tracing results of this traffic congestion tracing model, a more effective and targeted traffic control strategy can be formulated, that is, the simulation verifies that the paths with the greatest impact are regulated, and the benefits of traffic control in alleviating traffic congestion are increased as much as possible under limited resources.

[0126] The urban expressway corridor traffic congestion tracing model proposed in the present invention can fully learn the impact pattern and degree of the complete path including the upstream and downstream starting and ending points passing through the congested point segment on traffic congestion. The results of the traffic congestion tracing model are highly reliable. The targeted traffic control strategy formulated based on this technology can increase the travel speed by up to 23.36% and reduce the number of stops by up to 29.41%, which can provide a solid and reliable foundation for engineering applications.

[0127] If the above functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program code.

[0128] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes. The schemes in the embodiments of the present invention may be implemented in various computer languages, for example, object-oriented programming language Java and literal scripting language JavaScript, etc.

[0129] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0130] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0131] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0132] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0133] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A path-based urban expressway corridor traffic congestion tracing method, characterized in that: The following steps are involved: Obtain the travel speed and flow detection data time series of each lane in the required analysis period, and construct the section-level travel speed and flow information spatiotemporal matrix of the required analysis period; Based on the section-level travel speed and flow information spatiotemporal matrix, construct a path traffic state tensor for a required analysis period; Input the path traffic state tensor into a pre-trained traffic congestion source tracing model to obtain the congestion correlation of each section in the required analysis period, and mine the congestion source according to the congestion correlation; The specific steps to construct the spatiotemporal matrix of cross-section-level travel speed and flow information for the required analysis period include: Obtain the travel speed and flow detection data time series of each lane in the required analysis period; Determine the time length of aggregation, perform double aggregation on the travel speed and flow detection data time series from the two dimensions of time and space, and obtain the travel speed and flow information at the spatial section where each detection point is located; The travel speed and flow information at the spatial section where each detection point is located are combined into a matrix form to obtain a section-level travel speed and flow information spatiotemporal matrix; The expression for obtaining the travel speed and flow information at the spatial section where each detection point is located is: Among them, V i k , are the travel speed and flow rate of the kth aggregation time period of the detection section i after double aggregation, are the flow and travel speed of lane j in section i in the kth aggregation time period after time aggregation, T is the length of each aggregation time period, t is the sampling time interval of the detection point, and n is the expansion multiple of the time period length. They are the travel speed and flow information of lane j in detection section i at the mth sampling moment obtained by the fixed-point detector.

2. The path-based urban expressway corridor traffic congestion tracing method according to claim 1 is characterized in that: The expressions of the section-level travel speed and flow information space-time matrix are: Among them, S v , S q are the travel speed and flow space-time matrices, V i k is the travel speed of the kth aggregation time period of the detection section i after double aggregation, It is the flow rate of the kth aggregation time period of the detection section i after double aggregation, i = 1, 2, 3 ..., L, k = 1, 2, ..., H, L is the number of sections where the effective fixed-point detectors are located in the area, and H is the total number of aggregation time periods.

3. The path-based urban expressway corridor traffic congestion tracing method according to claim 1 is characterized in that: The specific steps to construct the path traffic status tensor for the required analysis period include: Obtain the connection relationship and distance length sequence between each section according to the physical space topology; The forward and backward transmission process of the path is calculated from the two dimensions of time and space to obtain the path time and space point position at the spatial section where each detection point is located; The path spatiotemporal point positions at the spatial sections where the respective detection points are located are indexed to the section-level travel speed and flow information spatiotemporal matrix, and the path traffic state tensors at the spatial sections where the respective detection points are located are obtained by combining them.

4. The path-based urban expressway corridor traffic congestion tracing method according to claim 3 is characterized in that: The expression of the path traffic state tensor is: R=[R 1,1 ,...,R i,L ,R 2,1 ,...,R 2,L ,...,R H,1 ,...,R H,L ] Among them, R h,l is the path time-space point at section l at time h, h = 1, 2, ..., H, l = 1, 2, 3 ..., L, L is the number of sections where effective fixed-point detectors are located in the area, H is the number of analysis time periods, S v and S q are the space-time matrices of travel speed and flow, respectively.

5. The path-based urban expressway corridor traffic congestion tracing method according to claim 1 is characterized in that: The specific steps to obtain a pre-trained traffic congestion tracing model include: Based on the time series of travel speed and flow detection data of each lane, a historical section-level travel speed and flow information spatiotemporal matrix is ​​constructed; Based on the historical section-level travel speed and flow information spatiotemporal matrix, the historical path traffic state tensor is constructed and training samples are formed; Build deep neural network models; Based on the training samples, the deep neural network model is trained to obtain optimal model parameters, and the training is completed to obtain a pre-trained traffic congestion tracing model.

6. The path-based urban expressway corridor traffic congestion tracing method according to claim 5 is characterized in that: The gradient descent method with adaptive learning rate is used to optimize the parameters of the deep neural network model.

7. The path-based urban expressway corridor traffic congestion tracing method according to claim 6 is characterized in that: The optimization function used in the training process is: Among them, Θ is the parameter to be optimized in the deep neural network model, is the label of the pth training sample, is the model output corresponding to the pth training sample, and P is the training sample size.

8. The path-based urban expressway corridor traffic congestion tracing method according to claim 5 is characterized in that: The deep neural network model includes a convolutional long short-term memory network, a soft attention module and a deformable convolutional neural network that are cascaded in sequence.

Images