Key road section identification method based on two-stage feature learning

Abstract

The invention provides a key road section recognition method based on two-stage feature learning, and the method comprises the steps: obtaining the static and dynamic embedding of a road section through a natural language processing technology based on mobile phone positioning data and road network data, and carrying out the recognition of a key road section through a two-stage feature learning method based on attention pooling and differential clustering. The road sections have obvious influence on traffic flow of a whole urban road network or a local area under a normal road condition; and obtaining a key road section identification result. Therefore, the road sections which have obvious influence on the traffic flow of the whole urban road network or the local area under the normal road condition can be identified. The method not only comprises road sections with relatively large influence in the global situation, but also covers relatively key road sections which have relatively small traffic flow but play an important connection role in a local area.

Description

Technical Field

[0001] The invention belongs to the field of spatial information technology, and in particular relates to a key section identification method based on two-stage feature learning. Background Art

[0002] In an urban road network, a few key sections are often crucial to the operating efficiency of the entire road network. If key sections can be identified and managed effectively, the traffic pressure on the entire road network can be alleviated, which is of great significance for improving the efficiency and reliability of the urban transportation system.

[0003] At present, existing technologies have proposed a variety of perspectives and methods for the study of critical road section identification. First, from the perspective of traffic flow, some studies identify road sections that play an important role in traffic state propagation by analyzing traffic flow data. For example, the paper "Identification of Critical Road Sections in Urban Road Networks Based on GPS Data" is based on floating vehicle GPS trajectory data and uses the Apriori algorithm to mine frequently used critical road sections in the road network; the paper "Identification of Critical Road Sections in Urban Traffic Road Networks Based on Spatiotemporal Correlation" analyzes the spatiotemporal correlation between road section traffic flow and evaluates the critical road sections that have a greater impact on traffic state propagation; in addition, the paper "Identifying Critical Congested Roads Based on Traffic Flow-AwareRoadNetwork Embedding" combines traffic flow distribution and road structure information to propose a road network embedding model for identifying congested road sections. Secondly, from the perspective of complex networks, researchers pay more attention to the relationship between the characteristics of road network structure and its criticality. The paper "A method of identifying critical road segment: A case study of peninsular Malaysia Road Network" identifies critical road segments by combining road segment length, betweenness centrality and road density; the paper "Identifying the Critical Links in Road Transportation Networks: Centrality-based approach utilizing structural properties" defines a criticality index of the relative importance of links by analyzing the similarities between road network structure and complex networks, and identifies vulnerable road segments; while the paper "A link criticality index embedded in the convex combinations solution of user equilibrium traffic assignment" proposes a new index that combines network topology and traffic flow characteristics to identify vulnerable nodes in the road network. In addition, methods such as cascading failure effect analysis and vulnerability analysis have also received much attention. These methods are usually used to evaluate the performance of road segments under emergencies or network failures, and can identify critical road segments that have a greater impact on the operation of the entire road network under extreme conditions.

[0004] However, most existing technologies fail to focus on identifying key sections in large-scale urban road networks under normal traffic conditions, which may be due to the lack of real-time traffic data for each section. However, the high cost of deploying monitoring equipment across the board makes it almost impossible to obtain traffic data for each section. In addition, most of these methods can only identify key sections with large traffic flow or good connectivity in the entire road network, while ignoring relatively key sections in local areas with small traffic flow. Summary of the invention

[0005] In view of the above limitations and shortcomings of the prior art, the present invention uses natural language processing technology to obtain static and dynamic embedding of road sections based on large-scale mobile phone positioning data, and proposes a two-stage feature learning method to identify road sections that have a significant impact on the entire urban road network or local area traffic flow under normal road conditions. This includes not only road sections with greater influence in the global context, but also relatively critical road sections that have a small traffic flow but play an important role in connecting local areas.

[0006] The technical solution specifically adopted by the present invention to solve the technical problem is:

[0007] A key road section identification method based on two-stage feature learning is based on mobile phone positioning data and road network data, natural language processing technology is used to obtain static and dynamic embeddings of road sections, and a two-stage feature learning method based on attention pooling and differentiable clustering is used to identify road sections that have a significant impact on the entire urban road network or local area traffic flow under normal road conditions; thereby obtaining key road section identification results.

[0008] Furthermore, a traffic corpus is constructed based on the preprocessed mobile phone positioning data and road network data: the starting and ending points of each motor vehicle trip are identified, and then the shortest path between each pair of starting and ending points is calculated, and the road sections sequentially passed by the shortest path are used as travel routes; the road sections in the road network are analogized to words, and the travel routes are analogized to sentences, and the construction of the traffic corpus is completed through the mapping of numbers to letters.

[0009] Furthermore, the method of preprocessing mobile phone positioning data and road network data includes: eliminating duplicate and invalid data, data with ping-pong or drift effects, and data outside the study area in the mobile phone positioning data; after data clipping and topology checking on the road network data, its NetworkX graphic object is generated.

[0010] Furthermore, the two-stage feature learning includes:

[0011] The first stage: Generate static embeddings for each road segment, perform clustering, and use the cluster centers as the initial cluster centers for differentiable clustering in the second stage;

[0012] Phase 2: After generating the dynamic embedding matrix for each road segment, use attention pooling and differentiable clustering to process it and calculate the loss function; then adjust the training parameters of the second phase according to the loss function value;

[0013] When the loss function value converges, the fusion features of each road section are obtained; finally, K-Means clustering is performed on each fusion feature, and the obtained cluster center is the key road section to be identified.

[0014] Furthermore, the generating of the static embedding of each road section is specifically generating the static embedding of each road section based on the Skip-gram mode:

[0015] In the traffic corpus, the context of each road segment corresponds to the nearby road segments in the travel route; the Skip-gram mode based on Word2Vec is used to simultaneously learn the co-occurrence relationship and order relationship of road segments in the travel route; the traffic corpus is used to train Skip-gram, and the formula of the loss function is defined as follows;

[0016]

[0017] Where T is the total number of road sections, w t represents the tth road segment, c is the given context window size, p(w t+j |w t ) is the given target segment w t In the case of t+j Probability of occurrence.

[0018] Furthermore, the dynamic embedding is specifically to generate a dynamic embedding of a road segment based on ELMo:

[0019] Use ELMo to extract the dynamic embedding of the i-th road segment i , the formula is as follows;

[0020]

[0021] Where L is the number of BiLSTM layers, γ is a learnable scaling parameter, and S k is the learnable weight of each layer, The k in is the k-th layer LSTM, and i refers to the i-th road section. represents the concatenation of the forward and backward hidden states;

[0022] The traffic corpus is used to train ELMo, and the cross entropy loss is used as the loss function to measure the difference between the probability distribution of the predicted road section at the current location of the model and the one-hot distribution of the actual road section. The specific calculation formula is as follows:

[0023]

[0024] Where N is the number of road segments in the current processing batch; C is the total number of road segments; y ij is a binary indicator. When the sample i, corresponding to a specific location segment in the current travel route sequence that needs to be predicted, has the highest probability of belonging to the segment j of the corresponding category, y ij =1, otherwise y ij =0; p ij is the probability that the model predicts that sample i belongs to category j;

[0025] ELMo generates different dynamic embeddings according to the number of times the same road segment appears in different travel routes, and forms a dynamic embedding matrix by vertical stacking to contain all dynamic embeddings of the road segment.

[0026] Furthermore, the second stage training process is specifically as follows:

[0027] First, ELMo is used to generate a dynamic embedding matrix for each road segment, and attention pooling is used to form a fixed-dimensional feature vector for each road segment. Then, differentiable clustering is performed on the feature vectors of each road segment, and the distance from each road segment feature vector to each initial cluster center obtained in the first stage is calculated and assigned to the cluster with the closest distance, thereby obtaining the clustering result. Finally, the loss function value is calculated based on the clustering result, which is used to optimize the parameters of attention pooling and differentiable clustering in the next round of training.

[0028] The loss function consists of two parts. The first part is to minimize the ratio of the intra-cluster distance to the inter-cluster distance. The second part is the moving distance between the current cluster center and the initial cluster center. The calculation formula is as follows:

[0029]

[0030] In the formula, K represents the number of clusters, and sample x i represents the feature vector of the i-th road segment, c k and are the cluster center of the kth cluster and its initial cluster center, D k is the set of samples in the kth cluster, α, β and γ are the weight coefficients of the three loss terms respectively.

[0031] Furthermore, the attention pooling first calculates the input attention score of each dynamic embedding in the dynamic embedding matrix of each road section through linear transformation and activation function, then normalizes these scores using the Softmax function, and finally uses the normalized weights to perform weighted summation on the dynamic embedding matrix to obtain the final representation;

[0032]

[0033] In formula (4), αi represents the attention score of the i-th embedding in the dynamic embedding matrix, x i represents the i-th embedding in the dynamic embedding matrix, w x For input x i Perform a linear transformation, b x is the bias vector, tanh is the activation function, w T Used to perform linear transformation on the result after tanh processing. Indicates the softmax normalization of the attention score; in formula (5), z is the road segment feature vector obtained after pooling;

[0034] The differentiable clustering uses the softmax function to achieve soft allocation and calculates the probability of each road segment feature vector after attention pooling to belong to each cluster. The formula is as follows:

[0035]

[0036] In the formula, r ik is the characteristic vector x of road segment i i The probability of belonging to cluster k, K represents the number of clusters, represents the kth cluster of clustering, and β is a parameter used to control the hardness or softness of the allocation;

[0037] After soft allocation, the cluster centers are updated according to the results. k_center It is the weighted average of all samples in the cluster. The specific formula is as follows;

[0038]

[0039] And, a key section identification system based on two-stage feature learning, comprising:

[0040] Traffic corpus construction module: Calculate the travel route set based on mobile phone positioning data and road network data, and complete the construction of the traffic corpus through letter mapping.

[0041] Road segment embedding extraction module: Natural language processing technology is used to extract the static embedding and dynamic embedding matrices of each road segment.

[0042] And the recognition module: using a two-stage feature learning method based on attention pooling and differentiable clustering to identify the road sections that have a significant impact on the entire urban road network or local area traffic flow under normal road conditions; thereby obtaining key road section identification results.

[0043] An electronic device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps of the key section identification method based on two-stage feature learning as described above are implemented.

[0044] Compared with the prior art, the present invention and its preferred solution have at least the following outstanding advantages: (1) The static and dynamic embedding of road sections are well integrated through two-stage feature learning. (2) It realizes the identification of relatively key road sections in local areas with small traffic flow in large-scale road networks. (3) The identification results show certain advantages in spatial distribution and various evaluation indicators. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments:

[0046] Figure 1 This is a technical framework diagram of key section identification based on two-stage feature learning in an embodiment of the present invention.

[0047] Figure 2 This is an example diagram of the Skip-gram mode of an embodiment of the present invention.

[0048] Figure 3 This is an example diagram of the ELMo model according to an embodiment of the present invention.

[0049] Figure 4 This is an example diagram of the attention pooling mechanism of an embodiment of the present invention.

[0050] Figure 5 This is a diagram showing the identification results of the first 3% of key road sections according to an embodiment of the present invention.

[0051] Figure 6 This is a diagram of the key road section identification results of each method in the embodiments of the present invention.

[0052] Figure 7 Schematic diagram of geographical location distribution of various regions in an embodiment of the present invention.

[0053] Figure 8 Schematic diagram of the distribution of key sections in various areas relative to the embodiment of the present invention. DETAILED DESCRIPTION

[0054] In order to make the features and advantages of this patent more obvious and easy to understand, the following embodiments are specifically described in detail as follows:

[0055] It should be noted that the following detailed descriptions are illustrative and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used in this specification have the same meanings as those commonly understood by those skilled in the art to which the present application belongs.

[0056] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0057] Based on mobile phone positioning data and road network data, this paper introduces natural language processing technology, attention pooling and differentiable clustering to process the static and dynamic embedding of each road section, and uses this as a feature to identify key sections. Finally, the results are verified by three evaluation indicators. Its main contents include:

[0058] (1) Preprocessing mobile phone positioning data and road network data;

[0059] (2) Construction and validation of the traffic corpus;

[0060] (3) Two-stage feature learning;

[0061] (4) Obtain key road section identification results.

[0062] The embodiment of the present invention proposes a key section identification method based on two-stage feature learning. The basic architecture of the solution is as follows: Figure 1 As shown, the following steps are included:

[0063] The first step is to pre-process the mobile phone positioning data and road network data;

[0064] (1) Eliminate the duplicate and invalid data, data with ping-pong or drift effects, and data outside the study area from the mobile phone positioning data. (2) After data clipping and topology checking of the road network data, generate its NetworkX graphic object.

[0065] The second step is to build and validate the traffic corpus;

[0066] (1) Identify the starting and ending points of each motor vehicle trip, and then calculate the shortest path between each pair of starting and ending points. The road segments that the shortest path passes through in sequence are the travel routes of this trip. (2) Compare the road segments in the road network to words and the travel routes to sentences. Through the mapping of numbers to letters (i.e., 0-a, 1-b,..., 9-j), the construction of the traffic corpus is completed. (3) The rationality of the traffic corpus is verified by the "80 / 20 rule" of traffic flow distribution and average speed.

[0067] The third step is two-stage feature learning;

[0068] (1) Phase 1: Generate static embeddings for each road segment, cluster them, and use their cluster centers as the initial cluster centers for differentiable clustering in the second phase. (2) Phase 2: After generating the dynamic embedding matrix for each road segment, use attention pooling and differentiable clustering to process them and calculate the relevant loss function. (3) According to the loss function value, continuously adjust the training parameters of the second phase.

[0069] Step 4: Obtain key road section identification results;

[0070] (1) After the loss function converges, the fused features of each road section are obtained; (2) K-Means clustering is performed on each fused feature; (3) The cluster center is the key road section to be identified.

[0071] In the above steps, the generation of static and dynamic embeddings of road segments and the training process of the second stage are the focus of the present invention and are discussed in detail below.

[0072] (1) Static and dynamic embedding generation of road sections

[0073] (a) Generate static embeddings for each road segment based on the Skip-gram model

[0074] In the traffic corpus, the context of each road segment is the nearby road segments in the travel route. Based on the Skip-gram model of Word2Vec, the co-occurrence relationship and order relationship of road segments in the travel route can be learned at the same time. If the traffic corpus is used to train Skip-gram, the formula of its loss function is defined as follows.

[0075]

[0076] Where T is the total number of road sections, w t represents the tth road segment, c is the given context window size, p(w t+j |w t ) is the given target segment w t In the case of t+j Probability of occurrence.

[0077] (b) Generate dynamic embedding of road segments based on ELMo

[0078] ELMo (Embedded Language Model) can dynamically adjust the word embedding according to the word's usage in a specific context, thereby effectively dealing with the problem of polysemy. Similarly, the same road segment may also present different characteristics in different travel routes (different contexts). Therefore, the present invention uses ELMo to extract the dynamic embedding of the ith road segment. i , the formula is as follows.

[0079]

[0080] Where L is the number of BiLSTM layers, γ is a learnable scaling parameter, and S k is the learnable weight of each layer, The k in is the k-th layer LSTM, and i refers to the i-th road section. Represents the concatenation of the forward and backward hidden states.

[0081] The traffic corpus is used to train ELMo, and the cross-entropy loss is used as its loss function to measure the difference between the probability distribution of the predicted road section at the current location of the model and the one-hot distribution of the actual road section. The specific calculation formula is as follows:

[0082]

[0083] Where N is the number of road segments in the current processing batch; C is the total number of road segments; y ij is a binary indicator. When the category with the highest probability of sample i (i.e., the probability distribution of a specific location segment in the current travel route sequence that needs to be predicted) belongs to category j (i.e., segment j), y ij =1, otherwise y ij =0; p ij is the probability that the model predicts that sample i belongs to category j.

[0084] ELMo generates different dynamic embeddings according to the number of times the same road segment appears in different travel routes, and can be stacked vertically to form a dynamic embedding matrix, which contains all the dynamic embeddings of the road segment.

[0085] (2) Second stage training process

[0086] (a) Attention Pooling

[0087] Attention pooling is a method for extracting important information from sequence data. It assigns an attention weight to each time step and uses these weights to perform a weighted sum over the sequence, thereby generating a fixed-dimensional representation for each sequence of unequal length.

[0088] In the present invention, attention pooling first calculates the input attention score of each dynamic embedding in the dynamic embedding matrix of each road section through linear transformation and activation function, then normalizes these scores using the Softmax function, and finally uses the normalized weights to perform weighted summation on the dynamic embedding matrix to obtain the final representation. The above process is as follows.

[0089]

[0090] In formula (4), α irepresents the attention score of the i-th embedding in the dynamic embedding matrix, x i represents the i-th embedding in the dynamic embedding matrix, w x For input x i Perform a linear transformation, b x is the bias vector, tanh (hyperbolic tangent function) is the activation function, w T Used to perform linear transformation on the result after tanh processing. Indicates the softmax normalization of the attention score. In formula (5), z is the road segment feature vector obtained after pooling.

[0091] (b) Differentiable Clustering

[0092] Differentiable clustering is a method that combines clustering algorithms with deep learning frameworks, which allows the clustering process to be trained in an end-to-end differentiable neural network. Traditional clustering algorithms, such as K-means, are usually non-differentiable, which means they cannot be directly trained with neural networks, because the optimization of neural networks usually relies on gradient descent and backpropagation.

[0093] In the present invention, differentiable clustering uses the softmax function to implement soft allocation, and calculates the probability of each road segment feature vector after attention pooling to belong to each cluster. The formula is as follows:

[0094]

[0095] In the formula, r ik is the characteristic vector x of road segment i i The probability of belonging to cluster k, K represents the number of clusters, represents the kth cluster of clustering, and β is a parameter used to control the hardness or softness of the allocation.

[0096] After soft allocation, the cluster centers are updated according to the results. k_center It is the weighted average of all samples in the cluster. The specific formula is as follows.

[0097]

[0098] (c) The complete process of the second stage training

[0099] First, the dynamic embedding matrix of each road segment is generated by ELMo, and it is processed by attention pooling to form a fixed-dimensional feature vector for each road segment. Then, differentiable clustering is performed on the feature vectors of each road segment, that is, the distance from each road segment feature vector to each initial cluster center (obtained from the first stage) is calculated and assigned to the closest cluster to obtain the clustering result. Finally, the loss function value is calculated based on the clustering result to optimize the parameters of attention pooling and differentiable clustering in the next round of training.

[0100] The loss function consists of two parts. The first part is to minimize the ratio of the intra-cluster distance to the inter-cluster distance. The second part is the moving distance between the current cluster center and the initial cluster center. The calculation formula is as follows:

[0101]

[0102] In the formula, K represents the number of clusters, and sample x i represents the feature vector of the i-th road segment, c k and are the cluster center of the kth cluster and its initial cluster center, D k is the set of samples in the kth cluster, α, β and γ are the weight coefficients of the three loss terms respectively.

[0103] When the loss function value converges, the fused features of each road section after two-stage learning can be obtained, and K-Means clustering is performed on it. The cluster center is the key road section to be identified.

[0104] The efficacy of the solution provided by the present invention is further demonstrated and illustrated by means of specific test examples below:

[0105] Based on the mobile phone positioning data and POI data of a large city in China, this example uses the two-stage feature learning method provided by the present invention to identify key road sections within the urban road network level. The experimental results of identifying 3% key road sections are as follows: Figure 5 As shown in Figure 2, key sections are more likely to be distributed in areas with larger traffic volumes, while there are also a certain number of relatively key sections in local areas with smaller traffic volumes. Figure 6 As shown, (a)(b)(c)(d) correspond to 2% key road section recognition results, (e)(f)(g)(h) correspond to 3% key road section recognition results, and (i)(j)(k)(l) correspond to 4% key road section recognition results.

[0106] Among them, ①Apriori: an algorithm for mining frequent item sets in a corpus; the test results correspond to (a)(e)(i)②Node2Vec: the Node2Vec vector of each road section can be obtained through depth-first search to identify key sections; the test results correspond to (b)(f)(j)③Word2Vec: directly use the Word2Vec static embedding of each road section to identify key sections; the test results correspond to (a)(e)(k); the test results of the present invention correspond to (d)(h)(l).

[0107] Next, three customized road segment evaluation indicators are calculated for the four methods. The comparative experimental results of 2%, 3%, and 4% key road segment identification are shown in Table 1.

[0108] Table 1 Comparison of key road section recognition results of four methods

[0109]

[0110] (1) Introduction to evaluation indicators

[0111] The present invention uses the designed k-th order traffic flow coverage, k-th order local traffic flow bearing rate, and k-th order local traffic flow correlation for evaluation. The introduction of these three road section evaluation indicators is as follows.

[0112] Before introducing the evaluation indicators, we first assume that the nearest neighbor and second nearest neighbor of a road segment are its first-order and second-order nearest neighbor segments. Based on this, we can define the k-th nearest neighbor segment and k-th nearest neighbor segment (all nearest neighbor segments from the first to the k-th order) of a road segment.

[0113] (a) k-th order traffic flow coverage

[0114] k-order traffic flow coverage CR k It refers to the proportion of the traffic flow of each key section and all its k-order neighboring sections to the traffic flow of the entire road network. The calculation formula is as follows:

[0115]

[0116] Where N k (S) is the set of key road segments S itself and its k-order neighboring road segments, and F(i) is the traffic flow of road segment i.

[0117] The higher the traffic flow coverage of a set of key road sections, the stronger the influence of these key road sections in the entire road network.

[0118] (b) k-th order local traffic flow load ratio

[0119] k-order local traffic flow rate LFR k It refers to the traffic flow of a road section divided by the sum of its own traffic flow and the traffic flow of its k-order neighboring road sections. The calculation formula is as follows:

[0120]

[0121] Where F(i) is the traffic flow of road section i, N k (i) is the set of kth nearest neighboring links of link i.

[0122] The higher the k-order local traffic flow bearing rate of a road section, the stronger its influence in the k-order local road network area. The average value of the k-order local traffic flow bearing rate of each road section in the key road section set is the k-order local traffic flow bearing rate of the key road section set.

[0123] (c) k-th order local traffic flow correlation

[0124] Using the traffic flow data of a single road segment and its k-th order neighboring road segments in different time intervals, a Spearman correlation coefficient value can be calculated, which can be regarded as the k-th order local traffic flow correlation of the road segment.

[0125] Assuming that there are n time intervals in total, the specific calculation formula for the k-th order local traffic flow correlation of a single road segment v is as follows:

[0126]

[0127] Where Nk(v) and |Nk(v)| are the set of k-th nearest neighboring road segments of road segment v and their number, respectively. and are the traffic flows of road segment v and its kth nearest neighboring road segment u (u∈Nk(v)) in the i-th time interval. and are the ranks of the traffic flow of road segment v and its kth nearest neighboring road segment u in the i-th time interval.

[0128] ρ of a road segment k The higher (v) is, the higher the correlation between the traffic flow changes of the road section and its k-th order neighboring road sections is, indicating that the influence of the road section in its k-th order local area is stronger.

[0129] The average value of the k-th order local traffic flow correlation of each link in the key link set is the k-th order local traffic flow correlation value of the key link set.

[0130] (2) Analysis of recognition results

[0131] The comparison results show that the key sections identified by this method are more distributed in areas with large traffic flow in the road network. At the same time, it can also identify relatively key sections in areas with small traffic flow, and its overall performance in various evaluation indicators is also better.

[0132] In addition, to further prove that the proposed method can identify a certain number of relatively critical sections in areas with less traffic flow, several relevant areas in the 4% critical section identification results are shown here.

[0133] Since there is no unified standard for the division of large and small traffic flow areas, the definition of urban functional areas is used here to divide them, because different functional areas correspond to different traffic needs. For example, the commercial area has a large traffic volume during the day, so it can be regarded as an area with a large traffic flow; while the residential and industrial areas have a small traffic volume during the same period, so they can be regarded as areas with a small traffic flow.

[0134] The geographical distribution of each region is as follows Figure 7 The corresponding detailed distribution of relative key sections is shown in Figure 8 shown.

[0135] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0136] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes 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 generate 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.

[0137] 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.

[0138] 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 1A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0139] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in the present invention should be understood by people with ordinary skills in the field to which the present invention belongs. The "first", "second" and similar words used in the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. "Include" or "comprise" and similar words mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0140] The above is only a preferred embodiment of the present invention, and does not limit the present invention in other forms. Any technician familiar with the profession may use the above disclosed technical content to change or modify it into an equivalent embodiment with equivalent changes. However, any simple modification, equivalent change and modification made to the above embodiment according to the technical essence of the present invention without departing from the technical solution of the present invention still belongs to the protection scope of the technical solution of the present invention.

[0141] This patent is not limited to the above-mentioned optimal implementation mode. Anyone can derive various other forms of key section identification methods based on two-stage feature learning under the inspiration of this patent. All equal changes and modifications made according to the scope of the patent application of the present invention should be covered by this patent.

Claims

1. A key section identification method based on two-stage feature learning, characterized in that: Based on mobile phone positioning data and road network data, natural language processing technology is used to obtain static and dynamic embeddings of road sections, and a two-stage feature learning method based on attention pooling and differentiable clustering is used to identify road sections that have a significant impact on the entire urban road network or local area traffic flow under normal road conditions; thereby obtaining key road section identification results.

2. The key section identification method based on two-stage feature learning according to claim 1 is characterized in that: A traffic corpus is constructed based on preprocessed mobile phone positioning data and road network data: the starting and ending points of each motor vehicle trip are identified, and then the shortest path between each pair of starting and ending points is calculated. The road sections passed by the shortest path in sequence are used as travel routes; the road sections in the road network are analogized to words, and the travel routes are analogized to sentences. The construction of the traffic corpus is completed through the mapping from numbers to letters.

3. The key section identification method based on two-stage feature learning according to claim 2 is characterized in that: The method of preprocessing mobile phone positioning data and road network data includes: removing the duplicate and invalid data, data with ping-pong or drift effects, and data outside the study area in the mobile phone positioning data; after data clipping and topology checking on the road network data, generating its NetworkX graphic object.

4. The key section identification method based on two-stage feature learning according to claim 1 is characterized in that: The two-stage feature learning includes: The first stage: Generate static embeddings for each road segment, perform clustering, and use the cluster centers as the initial cluster centers for differentiable clustering in the second stage; Phase 2: After generating the dynamic embedding matrix for each road segment, use attention pooling and differentiable clustering to process it and calculate the loss function; then adjust the training parameters of the second phase according to the loss function value; When the loss function value converges, the fusion features of each road section are obtained; finally, K-Means clustering is performed on each fusion feature, and the obtained cluster center is the key road section to be identified.

5. The key section identification method based on two-stage feature learning according to claim 4 is characterized in that: The generating of static embedding of each road section is specifically generating static embedding of each road section based on Skip-gram mode: In the traffic corpus, the context of each road segment corresponds to the nearby road segments in the travel route; the Skip-gram mode based on Word2Vec is used to simultaneously learn the co-occurrence relationship and order relationship of road segments in the travel route; the traffic corpus is used to train Skip-gram, and the formula of the loss function is defined as follows; Where T is the total number of road sections, w t represents the tth road segment, c is the given context window size, p(w t+j |w t ) is the given target segment w t In the case of t+j Probability of occurrence.

6. The key section identification method based on two-stage feature learning according to claim 4 is characterized in that: The dynamic embedding is specifically based on ELMo to generate dynamic embedding of road segments: Use ELMo to extract the dynamic embedding of the i-th road segment i , the formula is as follows; Where L is the number of BiLSTM layers, γ is a learnable scaling parameter, and S k is the learnable weight of each layer, The k in is the k-th layer LSTM, and i refers to the i-th road section. represents the concatenation of the forward and backward hidden states; The traffic corpus is used to train ELMo, and the cross entropy loss is used as the loss function to measure the difference between the probability distribution of the predicted road section at the current location of the model and the one-hot distribution of the actual road section. The specific calculation formula is as follows: Where N is the number of road segments in the current processing batch; C is the total number of road segments; y ij is a binary indicator. When the sample i, corresponding to a specific location segment in the current travel route sequence that needs to be predicted, has the highest probability of belonging to the segment j of the corresponding category, y ij =1, otherwise y ij =0; p ij is the probability that the model predicts that sample i belongs to category j; ELMo generates different dynamic embeddings according to the number of times the same road segment appears in different travel routes, and forms a dynamic embedding matrix by vertical stacking to contain all dynamic embeddings of the road segment.

7. The key section identification method based on two-stage feature learning according to claim 4 is characterized in that: The second stage training process is as follows: First, ELMo is used to generate a dynamic embedding matrix for each road segment, and attention pooling is used to form a fixed-dimensional feature vector for each road segment. Then, differentiable clustering is performed on the feature vectors of each road segment, and the distance from each road segment feature vector to each initial cluster center obtained in the first stage is calculated and assigned to the cluster with the closest distance, thereby obtaining the clustering result. Finally, the loss function value is calculated based on the clustering result, which is used to optimize the parameters of attention pooling and differentiable clustering in the next round of training. The loss function consists of two parts. The first part is to minimize the ratio of the intra-cluster distance to the inter-cluster distance. The second part is the moving distance between the current cluster center and the initial cluster center. The calculation formula is as follows: In the formula, K represents the number of clusters, and sample x i represents the feature vector of the i-th road segment, c k and are the cluster center of the kth cluster and its initial cluster center, D k is the set of samples in the kth cluster, α, β and γ are the weight coefficients of the three loss terms respectively.

8. The key section identification method based on two-stage feature learning according to claim 1 is characterized in that: The attention pooling first calculates the input attention score of each dynamic embedding in the dynamic embedding matrix of each road section through linear transformation and activation function, then normalizes these scores using the Softmax function, and finally uses the normalized weights to perform weighted summation on the dynamic embedding matrix to obtain the final representation; In formula (4), α i represents the attention score of the i-th embedding in the dynamic embedding matrix, x i represents the i-th embedding in the dynamic embedding matrix, w x For input x i Perform a linear transformation, b x is the bias vector, tanh is the activation function, w T Used to perform linear transformation on the result after tanh processing. Indicates the softmax normalization of the attention score; in formula (5), z is the road segment feature vector obtained after pooling; The differentiable clustering uses the softmax function to achieve soft allocation and calculates the probability of each road segment feature vector after attention pooling to belong to each cluster. The formula is as follows: In the formula, r ik is the characteristic vector x of road segment i i The probability of belonging to cluster k, K represents the number of clusters, represents the kth cluster of clustering, and β is a parameter used to control the hardness or softness of the allocation; After soft allocation, the cluster centers are updated according to the results. k_center It is the weighted average of all samples in the cluster. The specific formula is as follows; 9. A key section identification system based on two-stage feature learning, characterized in that: include: Traffic corpus construction module: calculates the travel route set based on mobile phone positioning data and road network data, and completes the construction of the traffic corpus through letter mapping; Road segment embedding extraction module: uses natural language processing technology to extract the static embedding and dynamic embedding matrices of each road segment; And the recognition module: using a two-stage feature learning method based on attention pooling and differentiable clustering to identify the road sections that have a significant impact on the entire urban road network or local area traffic flow under normal road conditions; thereby obtaining key road section identification results.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps of the key section identification method based on two-stage feature learning as described in any one of claims 1-8 are implemented.

Images