Intelligent high-speed traffic monitoring and analysis method and system based on application fusion perception

By calculating the common evaluation coefficients and impact weights between high-flow sections, the problem of difficult to predict future traffic flow changes and neglecting mutual influence in the road sections in the prior art is solved, and more accurate high-flow forecasting and traffic scheduling support is achieved.

CN118865695BActive Publication Date: 2024-12-03SICHUAN WANWANWANZONG TECH CO LTD
View PDF 6 Cites 0 Cited by

Patent Information

Application Number
CN202411347442.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-26
Publication Date
2024-12-03
Estimated Expiration
2044-09-26

Smart Images

  • Figure CN118865695B_ABST
    Figure CN118865695B_ABST
Patent Text Reader

Abstract

The present invention provides an intelligent high-speed traffic monitoring and analysis method and system using fusion perception. The method and system are based on determining the first commonality evaluation coefficients between the last high-flow section in a high-flow section set and the remaining high-flow sections. For each remaining high-flow section, the high-flow section influence weight is determined according to the first commonality evaluation coefficients corresponding to the remaining high-flow sections and the next high-flow section of the remaining high-flow sections. Based on the above steps, the hidden correlations between adjacent high-flow sections in the high-flow section set are quantified, so that the high-flow section influence weight corresponding to the next high-flow section of the remaining high-flow sections with higher commonality with the last high-flow section is higher. This can fully reflect the hidden correlations between the high-flow sections in the high-flow section set, conform to the periodic pattern of the sections where high flow occurs in the target monitoring sections, increase the accuracy of high-flow section estimation, and provide a reference basis for traffic scheduling.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of data processing, and more specifically, to an intelligent high-speed traffic monitoring and analysis method and system using fusion perception. Background Art

[0002] With the acceleration of urbanization and the continuous growth of car ownership, highways, as an important part of the modern transportation system, face unprecedented challenges in traffic flow management. Traditional traffic monitoring methods often rely on single monitoring equipment (such as cameras, radars, etc.) and manual judgment, which makes it difficult to fully, real-time and accurately capture the dynamic changes of traffic on highways. Especially during peak hours and special weather conditions, traffic congestion and accidents occur frequently, which seriously affects road traffic efficiency and driving safety. In order to meet these challenges, smart highway traffic monitoring technology has emerged. This technology realizes all-round and multi-dimensional monitoring of highway traffic conditions by integrating data from multiple sensing devices, such as camera videos, radar speed measurement, and vehicle detector information. However, most of the existing smart highway traffic monitoring methods focus on the real-time processing and display of a single data source, lack of in-depth mining and correlation analysis of historical traffic data, and it is difficult to accurately predict the changing trend of future traffic flow, especially in dealing with complex traffic patterns and periodic congestion problems. In addition, traditional traffic monitoring methods often ignore the inherent connection and mutual influence between different high-flow sections. Traffic flow on highways does not exist in isolation. Congestion on road sections, whether in time or space, often has a close correlation, leading to a chain reaction. Therefore, how to quantify this correlation and make more accurate traffic predictions and scheduling decisions based on it has become a key issue that needs to be solved in smart highway traffic monitoring technology. Summary of the invention

[0003] In view of this, an embodiment of the present invention provides an intelligent high-speed traffic monitoring and analysis method and system using fusion perception.

[0004] According to one aspect of an embodiment of the present invention, there is provided an intelligent high-speed traffic monitoring and analysis method using fusion perception, the method comprising: obtaining a high-flow section set of a target monitoring section, wherein the high-flow section set comprises a plurality of high-flow sections, the plurality of high-flow sections are arranged in time sequence according to the order in which high-flow occurs, and each high-flow section comprises corresponding fusion perception information; determining a first commonality evaluation coefficient between the last high-flow section in the high-flow section set and the remaining high-flow sections, respectively, wherein the remaining high-flow sections are high-flow sections in the high-flow section set except the last high-flow section; for each of the remaining high-flow sections, determining a high-flow section influence weight according to the first commonality evaluation coefficient corresponding to the remaining high-flow section and the next high-flow section of the remaining high-flow section; determining a target high-flow section according to the high-flow section influence weight corresponding to each of the remaining high-flow sections, wherein the target high-flow section is a section after the last high-flow section where high-flow occurs in the target monitoring section.

[0005] According to another aspect of an embodiment of the present invention, a computer system is provided, comprising: a processor; and a memory, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the processor executes the method described above.

[0006] The beneficial effects of the present invention include at least: the intelligent high-speed traffic monitoring and analysis method and system using fusion perception provided in the embodiment of the present invention, based on determining the first commonality evaluation coefficient between the last high-flow section in the high-flow section set and the remaining high-flow sections, for each remaining high-flow section, the high-flow section influence weight is determined according to the first commonality evaluation coefficient corresponding to the remaining high-flow sections and the next high-flow section of the remaining high-flow sections; based on the above steps, the hidden correlation between adjacent high-flow sections in the high-flow section set is quantified, so that the high-flow section influence weight corresponding to the next high-flow section of the remaining high-flow sections with high commonality with the last high-flow section is higher; at the same time, the estimated target high-flow section can fully reflect the hidden correlation between the high-flow sections in the high-flow section set, conform to the periodic pattern of the sections where high flow occurs in the target monitoring sections, increase the accuracy of high-flow section estimation, and provide an important reference basis for traffic scheduling. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] Figure 1 A schematic diagram of an application scenario provided by an embodiment of the present invention.

[0008] Figure 2 A schematic diagram of the implementation flow of an intelligent high-speed traffic monitoring and analysis method using fusion perception provided in an embodiment of the present invention.

[0009] Figure 3 A hardware entity schematic diagram of a computer system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0010] In order to make the purpose, technical scheme and advantages of the present invention clearer, the technical scheme of the present invention is further elaborated in detail below in conjunction with the accompanying drawings and embodiments. The described embodiments should not be regarded as limiting the present invention. All other embodiments obtained by ordinary technicians in the field without creative work are within the scope of protection of the present invention. Unless otherwise defined, all technical and scientific terms used in this article have the same meaning as those commonly understood by technicians in the technical field of the present invention. The terms used in this article are only for describing the purpose of the present invention and are not intended to limit the present invention.

[0011] The intelligent high-speed traffic monitoring and analysis method based on application fusion perception provided by the embodiment of the present invention can be applied to Figure 1 In the application environment shown. Among them, the terminal 102 communicates with the computer system 104 through the network. The data storage system can store the data that the computer system 104 needs to process. The data storage system can be integrated on the computer system 104, or it can be placed on the cloud or other network servers. Among them, the fused perception data can be stored in the local storage of the terminal 102, or it can be stored in the data storage system or the cloud storage associated with the computer system 104. When data processing is required, the computer system 104 can obtain the fused perception data from the local storage of the terminal 102, or the data storage system, or the cloud storage. Among them, the terminal 102 can be but is not limited to a gateway, a radar, a vehicle detector, etc. The computer system 104 can be implemented with an independent server or a server cluster consisting of multiple servers.

[0012] The intelligent high-speed traffic monitoring and analysis method based on application fusion perception provided by the embodiment of the present invention is applied to a computer system, such as Figure 2 As shown, the specific steps include:

[0013] Step S10: Obtain a high-flow section set of the target monitoring section, wherein the high-flow section set includes multiple high-flow sections, the multiple high-flow sections are arranged in time sequence according to the order in which high flow occurs, and each high-flow section includes corresponding fusion perception information.

[0014] Specifically, the computer system can access and integrate real-time and historical data from different sensing devices (such as gateways, radars, cameras, etc.). These data cover multiple key dimensions of the target monitoring section, including but not limited to the section location, traffic volume, average vehicle speed, lane occupancy, accident records, weather conditions, and the impact of special events (such as holidays, large-scale events, etc.) on traffic flow. These diverse information together form the basis for fusion perception information.

[0015] Taking a busy highway as an example, the computer system can collect the following specific data: during a specific time period, the traffic volume on a certain section of the road increased sharply, exceeding the daily average; at the same time, the average speed of vehicles on the section of the road dropped significantly, indicating congestion; in addition, radar data also showed that the lane occupancy rate of the section of the road was close to saturation, and there were several minor accidents captured by the camera. These data, combined with the geographical location information of the section of the road and the weather conditions at the time (such as reduced visibility due to rainy days), together constitute the fusion perception information of the high-traffic section of the road.

[0016] Next, the computer system performs time-series analysis and processing on these fused perception information, identifies all sections where high traffic has occurred, and arranges them in chronological order to form a set of high-traffic sections. In this set, each high-traffic section contains the time of its occurrence, duration, specific fused perception information, and possible other metadata (such as the historical congestion record of the section, the status of surrounding traffic facilities, etc.). Through this processing method, the computer system not only captures the high-traffic events themselves, but also deeply explores the multi-dimensional characteristics behind these events, providing a solid data foundation for subsequent high-traffic prediction and flow control strategy formulation. For example, when analyzing continuous high-traffic sections, the computer system can find that certain sections are more prone to congestion during specific time periods (such as morning and evening peaks) or under specific weather conditions, thereby providing targeted warnings and diversion suggestions for traffic management departments.

[0017] Step S20: determining first commonality evaluation coefficients between the last high-flow segment in the high-flow segment set and the remaining high-flow segments, wherein the remaining high-flow segments are high-flow segments in the high-flow segment set except the last high-flow segment.

[0018] In the embodiment of the present invention, the task of step S20 is to quantify the similarity between the last high-flow segment in the high-flow segment set (denoted as segment L) and all other high-flow segments in the set (denoted as the segment set {R1, R2, ..., Rn}, where n is the number of remaining high-flow segments), that is, to calculate the first commonality evaluation coefficient.

[0019] To achieve this goal, the computer system uses the fused perception information collected in the previous steps (such as step S10) to convert each high-traffic road section into a multi-dimensional feature vector. These feature vectors capture multiple attributes of the road section, including but not limited to location information, traffic volume, average speed, lane occupancy, weather conditions, and special events. For example, the feature vector of road section L may be expressed as [location coordinates (x, y), traffic volume 500 vehicles / hour, average speed 60km / h, lane occupancy 80%, clear weather, no special events]; while the feature vector of road section R1 may be different, reflecting the traffic conditions of the road section at another time point.

[0020] The computer system then uses a suitable similarity measurement method to calculate the first commonality evaluation coefficient between the road segment L and each road segment Ri (i=1,2,...,n). This method can be a distance-based metric (such as Euclidean distance, Manhattan distance, or cosine similarity) or a more complex machine learning model output. Taking cosine similarity as an example, it evaluates the similarity of two feature vectors by calculating the cosine value of the angle between them in multidimensional space. The closer the cosine value is to 1, the more similar the two vectors are.

[0021] In actual operation, the computer system calculates the cosine similarity between the feature vector of the road segment L and the feature vector of each road segment Ri in the set, and obtains a series of similarity values, which are the first commonality evaluation coefficients. For example, if the cosine similarity between the road segment L and the road segment R1 is 0.9, and the cosine similarity between the road segment L and the road segment Rn is 0.5, then it can be considered that the road segment L is more similar to the road segment R1 in terms of traffic conditions, and may have similar congestion causes or periodic patterns.

[0022] Through this step, the computer system not only reveals the intrinsic connection between high-flow sections, but also provides an important reference for the calculation of high-flow section influence weights and the prediction of target high-flow sections in subsequent steps (such as steps S30 and S40). These first common evaluation coefficients help the system better understand the evolution of historical high-flow events, thereby improving the accuracy of predictions for future high-flow events and providing more accurate decision support for traffic management departments.

[0023] Step S30: for each remaining high-flow road section, determine the high-flow road section influence weight according to the first commonality evaluation coefficient corresponding to the remaining high-flow road section and the next high-flow road section of the remaining high-flow road section.

[0024] In the embodiment of the present invention, step S30 is a key link in building a prediction model, which involves evaluating the impact of each non-last high-flow section (referred to as section Ri, where i=1,2,...,n-1, and n is the total number of high-flow sections in the set) that may cause high-flow events in the future. This evaluation is achieved by comprehensively considering the similarity (i.e., the first commonality evaluation coefficient) between section Ri and the last high-flow section (referred to as section Ln) and the characteristics of the next high-flow section (referred to as section Ri+1) that occurs immediately after section Ri. The computer system plays a core role in this process and is responsible for calculating the high-flow section impact weight of each section Ri.

[0025] Specifically, the computer system reviews the first commonality evaluation coefficients calculated in step S20, which quantify the similarity between the road segment Ri and the road segment Ln. A higher similarity indicates that the road segment Ri and the road segment Ln may be closely related in terms of traffic conditions, congestion causes, or periodic patterns. Next, the computer system considers the characteristics of the road segment Ri+1, which are obtained by fusing the perception information, including but not limited to traffic flow changes, average vehicle speed, lane occupancy, etc., which reflect the specific manifestations of the actual high-traffic event that occurs immediately after the road segment Ri.

[0026] In order to determine the high-flow section impact weight of section Ri, the computer system may use a weighted algorithm, in which the first commonality evaluation coefficient is used as one of the weight factors, which directly affects the predictive ability of section Ri for future high-flow events. At the same time, the characteristics of section Ri+1 may also be taken into consideration and converted into an indirect contribution to the impact weight of section Ri through some means (such as feature mapping, embedding vector calculation, etc.). For example, if the traffic volume of section Ri+1 surges and the average speed drops significantly, this may indicate that a certain state or condition of section Ri has an important impact on triggering subsequent high-flow events, and thus it will be evaluated higher when calculating the weight.

[0027] In actual operation, the computer system can perform some form of operation (such as dot product, weighted sum, etc.) on the first commonality evaluation coefficient and the feature vector of the road section Ri+1 to obtain a comprehensive score or weight value as the high-flow section influence weight of the road section Ri. This weight value not only reflects the similarity between the road section Ri and the road section Ln, but also incorporates the specific characteristics of the high-flow event actually occurring in the road section Ri+1, making the weight calculation more comprehensive and accurate.

[0028] Through this step, the computer system provides an important intermediate variable for predicting the sections that may generate high traffic in the future - the impact weight of the sections with high traffic. These weight values ​​will be used in the subsequent steps to comprehensively evaluate the impact of all relevant sections on future traffic conditions, thereby helping the traffic management department to take countermeasures in advance and optimize traffic scheduling and flow management strategies.

[0029] Step S40: determining a target high-flow section according to the high-flow section influence weight corresponding to each of the remaining high-flow sections, wherein the target high-flow section is a section where high flow occurs in the target monitoring section after the last high-flow section.

[0030] In the embodiment of the present invention, step S40 determines the section on the target monitoring section where high traffic is most likely to occur after the last high traffic section based on the high traffic section influence weight calculated in the previous step, i.e., the target high traffic section. The computer system plays a core role in this step, providing accurate prediction results for the traffic management department through comprehensive analysis and calculation.

[0031] Specifically, the computer system reviews the high-flow segment influence weights calculated for each remaining high-flow segment (referred to as segment Ri, where i=1,2,...,n-1, and n is the total number of high-flow segments in the set) in step S30. These weight values ​​reflect the influence of segment Ri on the high-flow event that may occur in the future, and are calculated based on the similarity between segment Ri and the last high-flow segment (referred to as segment Ln) and the characteristics of the next high-flow segment (referred to as segment Ri+1) that occurs immediately after segment Ri.

[0032] Next, the computer system uses a weighted aggregation or sorting selection method to determine the target high-flow section. One possible way is to accumulate or weighted average the high-flow section impact weights of all sections Ri to obtain a comprehensive evaluation value. This evaluation value can be regarded as a potential risk indicator of high flow on the target monitored section in the future. Then, the computer system combines the spatial location information and historical traffic patterns of the section to map this comprehensive evaluation value to a specific section, thereby determining the target high-flow section.

[0033] Another possible approach is to directly sort the road sections Ri according to the weights of the high-flow road sections, and select the road sections with the highest weights as the potential target high-flow road section candidate set. Subsequently, the computer system can further consider factors such as the geographical location, traffic facility conditions, and historical congestion records of these candidate road sections, and determine the final target high-flow road section through comprehensive evaluation.

[0034] In practical applications, the computer system can select the most appropriate determination method based on specific business needs and scenario characteristics. For example, if the target monitoring section is a long-distance highway, the computer system may be more inclined to use the weighted aggregation method, because this method can more comprehensively consider the mutual influence between sections and the traffic conditions of the entire road network. If the target monitoring section is a traffic node or hub area, the computer system can use a sorting selection method to quickly identify the sections most likely to be congested and take corresponding countermeasures.

[0035] Regardless of which method is used, the ultimate goal of step S40 is to provide accurate prediction results of target high-flow road sections for traffic management departments. These results will help traffic management departments formulate response strategies in advance, optimize traffic scheduling and flow management plans, thereby effectively alleviating traffic congestion and improving road traffic efficiency and safety.

[0036] As an implementation manner, in step S30, determining the high-flow section influence weight according to the first commonality evaluation coefficients corresponding to the remaining high-flow sections and the next high-flow section of the remaining high-flow sections may specifically include:

[0037] Step S31: Obtain the encoding representation information of the next high-flow section of the remaining high-flow sections.

[0038] In the embodiment of the present invention, step S31 is an important step in the process of determining the impact weight of high-flow road sections, which involves extracting and quantifying characteristic information of specific road sections from abundant traffic data so that subsequent steps can perform comprehensive evaluation based on this information. This step is automatically performed by the computer system, ensuring the accuracy and efficiency of data processing.

[0039] Specifically, the goal of step S31 is to obtain the encoded representation information of the "next high-traffic section" (denoted as section Ri+1) of the "rest of the high-traffic sections" (denoted as section Ri, where i=1,2,...,n-1), that is, to map the traffic characteristics of section Ri+1 into a low-dimensional vector space to form embedded features that are easy to calculate and compare.

[0040] In practical applications, the traffic characteristics of road section Ri+1 may include but are not limited to traffic volume, average speed, lane occupancy, accident rate, weather conditions, etc. These feature data usually come from a variety of sensing devices (such as radar, camera, weather station, etc.), and are preprocessed and fused to form fused sensing information. However, directly using these original features for calculation is often inefficient and difficult to capture the complex relationship between features. Therefore, the computer system adopts an effective method to convert these original features into coded representation information.

[0041] One method is to use embedding techniques. For example, you can use the autoencoder or Word2Vec model in deep learning to perform unsupervised learning on the features of the road segment Ri+1 and learn a low-dimensional vector representation, namely the embedded feature. This vector can capture the key feature information of the road segment Ri+1 while removing redundancy and noise, so that the similarities and differences between different road segments can be effectively represented in the vector space.

[0042] Taking the autoencoder as an example, the computer system can construct a neural network consisting of an input layer, a hidden layer, and an output layer. The input layer receives the original feature data of the road section Ri+1 (such as an array containing multiple feature values), and the hidden layer compresses the input data into a low-dimensional vector (i.e., embedded features) by learning the inherent structure in the data. The output layer attempts to reconstruct the input data. By training this neural network, the data reconstructed by the output layer is as close as possible to the data of the input layer, thereby learning effective embedded features.

[0043] In actual operation, the computer system can calculate an embedded feature vector for each road section Ri+1. This vector contains the key traffic feature information of the road section and can be used for subsequent calculations and analysis. For example, the embedded feature vector of road section Ri+1 may be an array containing 10 elements, each element corresponding to a specific traffic feature dimension. The specific values ​​learned through the embedding technology reflect the performance of the road section in these dimensions.

[0044] Through step S31, the computer system provides the necessary input data for determining the high-flow section influence weight, namely, the encoded representation information (embedded features) of the next high-flow section of the remaining high-flow sections. This information will be combined with the first commonality evaluation coefficient in subsequent steps to jointly calculate the high-flow section influence weight of each section Ri.

[0045] Step S32: determine the first multiplication result of the first commonality evaluation coefficient corresponding to the remaining high-flow sections and the encoded representation information of the next high-flow section of the remaining high-flow sections, and determine it as the high-flow section influence weight corresponding to the remaining high-flow sections.

[0046] In the embodiment of the present invention, step S32 is the core step of calculating the influence weight of the high-flow road section, which combines the first commonality evaluation coefficient between the road sections and the traffic characteristics of the specific road section (embodied by the coded representation information) to quantify the influence of each road section on the high-flow event that may occur in the future. This step is automatically performed by the computer system to ensure the accuracy and efficiency of the calculation.

[0047] Specifically, step S32 involves two key inputs: one is the first commonality evaluation coefficient (denoted as αi) between the "remaining high-flow sections" (denoted as section Ri) and the "last high-flow section" (denoted as section Ln), which reflects the similarity between section Ri and section Ln in terms of traffic conditions, congestion causes or periodic patterns; the other is the encoded representation information (denoted as Ei+1) of the "next high-flow section" (denoted as section Ri+1) of section Ri, which is a low-dimensional vector that captures the key traffic characteristics of section Ri+1.

[0048] During the calculation process, the computer system performs an element-wise multiplication operation to multiply the first commonality evaluation coefficient αi with each element in the encoded representation information Ei+1. Since αi is a scalar and Ei+1 is a vector, in actual operation, αi will be broadcasted to the same shape as Ei+1 for element-by-element multiplication. The result of this operation is a new vector, each element of which is the product of the corresponding element in the original Ei+1 vector and αi.

[0049] Next, the computer system can further process this product vector to obtain a single high-traffic section impact weight value (denoted as Wi). However, whether to perform such processing and how to process it depends on the specific application scenario and requirements. In some cases, in order to retain the detailed feature information of section Ri+1, the computer system can directly use the product vector as Wi; in other cases, in order to obtain a more concise and easy-to-compare weight value, the computer system can use some aggregation function (such as summation, average, etc.) to reduce the dimension of the product vector.

[0050] To give a specific calculation formula, assuming that Ei+1 is a vector containing m elements [e1, e2, ..., em], the calculation process of step S32 can be expressed as:

[0051] Product Vector ;

[0052] If the computer system chooses to use the summation function to obtain Wi, the calculation formula can be further simplified to:

[0053] ;

[0054] It can be understood that the above summation formula is only used as an example, and other aggregation functions may be used in actual applications or the form of the product vector may be kept unchanged.

[0055] Through the calculation of step S32, the computer system determines a high-flow section influence weight Wi for each section Ri. This weight value comprehensively considers the similarity between the section Ri and the section Ln and the specific traffic characteristics of the section Ri+1, and provides an important quantitative basis for the subsequent prediction of the target high-flow section. In the embodiment of the present invention, these weight values ​​will help the traffic management department to better understand the inherent laws of historical traffic data, so as to make more accurate and effective traffic scheduling and flow management decisions.

[0056] As an implementation manner, step S40, determining the target high-flow section according to the high-flow section influence weight corresponding to each of the remaining high-flow sections, may specifically include:

[0057] Step S41: Determine the sum of the high-flow section influence weights corresponding to the multiple high-flow section respectively, and determine it as the coding representation information of the target high-flow section.

[0058] In the embodiment of the present invention, step S41 is a key step to integrate the potential impact of multiple high-flow road sections on future traffic conditions. At this stage, the computer system will perform a series of calculations to convert the impact weights of the multiple high-flow road sections into a comprehensive index, which is then interpreted as the coded representation information of the target high-flow road section.

[0059] Specifically, the computer system reviews the high-flow segment impact weights (denoted as Wi) calculated for each "remaining high-flow segment" (denoted as segment Ri, where i=1,2,...,n-1) in step S30. These weight values ​​are determined based on the similarity (i.e., the first commonality evaluation coefficient) between segment Ri and the last high-flow segment (denoted as segment Ln) and the characteristics of the next high-flow segment (denoted as segment Ri+1) that occurs immediately after segment Ri, and they reflect the comprehensive influence of segment Ri on the possibility of causing high-flow events in the future.

[0060] Next, the computer system accumulates these weight values, that is, adds all Wi (i=1,2,...,n-1) to obtain a sum value. This sum value is a scalar that represents the overall potential impact of all considered high-flow sections on future traffic conditions. However, directly using this sum value as a prediction result is not intuitive and is not convenient for subsequent processing and interpretation.

[0061] Therefore, the computer system "maps" or "converts" this sum value into the coded representation information of the target high-traffic road segment. It should be noted here that the coded representation information is not a simple numerical value or identifier, but a multi-dimensional vector that can effectively represent the key features of the road segment in a low-dimensional space, so that the similarities and differences between the road segments can be quantified.

[0062] In this context, the encoding representation information of the target high-flow road section is a "synthetic" vector, which does not directly correspond to the existing encoding representation of a specific road section, but is indirectly obtained by accumulating the influence weights of multiple high-flow road sections. This synthetic vector may not exist directly in the pre-trained or constructed embedding space, but it can still find a relatively close position in the space, which reflects the comprehensive reflection of the impact of all the considered high-flow road sections on future traffic conditions.

[0063] To illustrate this point more specifically, it can be assumed that the encoded representation information of each high-traffic road section is a vector containing multiple feature dimensions (such as quantized values ​​of dimensions such as traffic volume, speed, lane occupancy, etc.). Although these original vectors are not directly used in step S41, some of their features may indirectly affect the calculation of the composite vector by affecting the weight Wi. For example, if a road section Ri obtains a higher Wi value due to its high traffic volume and low speed, then these features of this road section may be more significantly reflected in the composite vector.

[0064] Finally, the computer system outputs this synthetic vector as the coded representation information of the target high-traffic road section. Although this information does not directly correspond to any existing road section, it provides important input data for the subsequent nearest neighbor search (such as step S42), allowing the computer system to predict the target road section that may have high traffic in the future based on this comprehensive indicator.

[0065] Step S42: performing a nearest neighbor search operation on the coded representation information of the target high-flow road section to obtain the target high-flow road section.

[0066] In an embodiment of the present invention, step S42 identifies and determines the specific road sections where high traffic may occur in the future by performing a nearest neighbor search operation on the coded representation information of the target high-traffic road section. This step is automatically completed by the computer system, taking advantage of the spatial distribution characteristics of the coded representation information and the efficiency of the nearest neighbor search algorithm. First, it is necessary to clarify the source of the coded representation information of the target high-traffic road section. In step S41, the computer system has calculated a comprehensive coded representation vector by integrating the influence weights of multiple high-traffic road sections. This vector indirectly reflects the comprehensive impact of all considered road sections on future traffic conditions. Although this vector does not directly correspond to the coded representation of any existing road section, it is located in an embedded space composed of the coded representations of all road sections. The space is pre-trained by a machine learning model (such as an autoencoder, Word2Vec, etc.) and can capture the similarities and differences between road sections.

[0067] Next, the computer system will perform a nearest neighbor search operation on the coded representation vector of the target high-flow section. The nearest neighbor search is an algorithm that finds the point (or vector) in the data set that is most similar to a given query point. In this scenario, the query point is the coded representation vector of the target high-flow section calculated in step S41, and the data set is a set of coded representation vectors of all possible sections (or at least high-flow sections that have appeared in historical records).

[0068] During the search, the computer system calculates the similarity between the target vector and each vector in the data set. The similarity metric can be cosine similarity, Euclidean distance, etc. The specific choice depends on the characteristics of the embedding space and the design of the search algorithm. Taking cosine similarity as an example, it measures the degree of proximity of two vectors in direction, and the larger the value, the more similar they are. The computer system traverses all the vectors in the data set and finds the vector with the highest similarity to the target vector, that is, the nearest neighbor vector.

[0069] Since each vector in the dataset corresponds to a specific road segment (this correspondence has been established when training the embedding space), the road segment corresponding to the nearest neighbor vector is regarded as the target high-traffic road segment for prediction. This process is similar to the inverse processing of word embedding in natural language processing, that is, given an embedding vector, the original vocabulary is restored by finding the most similar word vector.

[0070] For example, suppose the encoding representation vector of the target high-traffic road section is a two-dimensional vector [0.5, 0.8], which represents the position of the road section in the embedding space. The computer system searches the data set containing all possible road section encoding representations and finds the vector with the highest similarity, such as [0.51, 0.79], which corresponds to road section A. Therefore, the computer system predicts that road section A is the target road section where high traffic may occur in the future.

[0071] Through the nearest neighbor search operation in step S42, the computer system can accurately predict specific sections where high traffic may occur in the future based on the comprehensive impact weights of multiple high-traffic sections, providing powerful decision-making support for traffic management departments.

[0072] In one embodiment, multiple high-flow sections in the high-flow section set all have a first impact factor, the first impact factor characterizes the frequency of high-flow sections in the target monitoring section, and the first impact factor is positively correlated with the frequency. Based on this, step S32 determines the first multiplication result of the first commonality evaluation coefficient corresponding to the remaining high-flow sections and the coded characterization information of the next high-flow section of the remaining high-flow sections, which may specifically include: determining the multiplication result of the first commonality evaluation coefficient corresponding to the remaining high-flow sections, the coded characterization information of the next high-flow section of the remaining high-flow sections, and the first impact factor of the remaining high-flow sections, and determining it as the first multiplication result.

[0073] In the embodiment of the present invention, one implementation of step S32 takes into account the frequency of high traffic in each section in the high traffic section set, i.e., the first impact factor, which plays a key role in evaluating the impact of the section on future traffic conditions. In this step, the computer system not only calculates the product of the similarity between sections (the first commonality evaluation coefficient) and the encoded representation information of the section characteristics, but also introduces the first impact factor to more comprehensively quantify the high traffic section impact weight of each section.

[0074] Specifically, each "other high-flow section" (denoted as section Ri) has a corresponding "first impact factor" (denoted as βi) in the set, which characterizes the frequency of high-flow events on section Ri on the target monitoring section. For example, if section Ri frequently has high-flow events in past historical records, then its first impact factor βi will be relatively high; conversely, if section Ri rarely has high-flow events, then βi will be relatively low. The first impact factor βi is positively correlated with the frequency of high-flow events on section Ri, and is an important indicator for evaluating whether a section may cause high-flow events in the future.

[0075] In step S32, the computer system obtains the first commonality evaluation coefficient (denoted as αi) corresponding to the road section Ri and the coded representation information (denoted as Ei+1) of the next high-traffic road section of the road section Ri (denoted as road section Ri+1). The coded representation information Ei+1 is a multidimensional vector that captures the key traffic characteristics of the road section Ri+1, such as traffic volume, vehicle speed, lane occupancy, etc.

[0076] Next, the computer system performs an extended multiplication operation, which involves not only αi and Ei+1, but also the first influencing factor βi of the road section Ri. Specifically, the computer system multiplies each element in αi, βi, and Ei+1 element by element (note that since αi and βi are scalars, they will be broadcast to the same shape as Ei+1), and obtains a new vector, each element of which is the product of the corresponding element in the original Ei+1 vector with αi and βi. This new vector is the "first multiplication result" in step S32.

[0077] To give a specific calculation formula, assume that Ei+1 is a vector containing m elements [e1, e2, ...,em], then the first multiplication result can be expressed as:

[0078] First multiplication result = ;

[0079] In practical applications, the computer system may not directly output this vector as the weight of the high-flow section, but may further process this vector, such as summing, averaging, or applying some aggregation function, to obtain a single weight value (denoted as Wi). This weight value will comprehensively consider the similarity between the section Ri and the section Ln, the specific traffic characteristics of the section Ri+1, and the frequency of high flow on the section Ri itself, providing an important quantitative basis for subsequent prediction analysis.

[0080] By introducing the first influencing factor βi, the implementation of step S32 makes the calculation of the influence weight of the high-flow section more comprehensive and accurate. It not only takes into account the similarity between sections and the traffic characteristics of specific sections, but also fully considers the historical frequency of high flow in the section itself, thereby providing a more refined prediction capability for intelligent highway flow monitoring and analysis.

[0081] As an implementation manner, step S10, obtaining a set of high-flow sections of a target monitoring section, may specifically include:

[0082] Step S11: Obtain historical traffic monitoring data of the target monitoring section.

[0083] In step S11 of the embodiment of the present invention, the computer system collects and integrates historical traffic monitoring data about the target monitoring section from multiple data sources. This step is crucial for subsequent analysis because it provides the basic information required to construct a set of high-flow section, evaluate the similarity between sections, and predict future high-flow events. Specifically, the computer system performs the following operations to obtain historical traffic monitoring data:

[0084] First, the computer system is connected to a database or data center that stores historical traffic data. This data usually comes from a variety of sensing devices installed on the target monitoring section and its surrounding areas, including but not limited to traffic monitoring cameras, radar speed guns, vehicle detectors, and weather stations. Each device captures different types of traffic parameters, such as traffic flow, speed, lane occupancy, vehicle type distribution, and real-time weather conditions, based on its specific functions and accuracy.

[0085] Next, the computer system initiates a data retrieval request to filter and extract historical traffic monitoring data related to the target monitoring section according to predefined data query specifications. These specifications may include time range (such as the past year, a specific month, or a comparison between weekdays and weekends), geographic location (accurate to a specific road section or intersection), data type (such as traffic flow statistics, speed distribution, accident records, etc.) and data accuracy (such as real-time data every minute or aggregated data every hour).

[0086] When retrieving data, the completeness and accuracy of the data are considered. It can automatically perform data cleaning and preprocessing steps to remove noise, fill missing values, and correct erroneous records. For example, if the traffic volume data for a certain time period is missing due to equipment failure, the computer system can try to fill the gap using the average traffic volume in adjacent time periods or predicted values ​​based on historical trends.

[0087] Finally, the filtered, extracted and pre-processed historical traffic monitoring data will be loaded into the memory or temporary storage medium by the computer system for subsequent analysis steps. These data may be stored in a variety of formats, such as relational database tables, time series data sets, or custom data structures specifically used for traffic analysis. Regardless of the format used, they contain rich traffic information and can fully reflect the changes in traffic conditions of the target monitoring section at different time periods.

[0088] Taking a busy highway as an example, the computer system can obtain hourly traffic statistics and average speed records for the past year. These data will serve as the basis for analyzing the high-flow event pattern of the road section, evaluating the similarity of traffic conditions between different sections, and predicting the probability of high-flow events in the future. By deeply analyzing these data, the traffic management department can formulate more accurate and effective traffic management strategies to alleviate traffic congestion and improve road traffic efficiency and safety.

[0089] Step S12: extracting a plurality of initial high-flow sections where high-flow occurs sequentially from the historical flow monitoring data.

[0090] In an embodiment of the present invention, step S12 accurately identifies and extracts a series of initial high-flow sections where high-flow occurs successively from the historical flow monitoring data. Specifically, the computer system performs a comprehensive scan and analysis of the historical flow monitoring data. These data usually contain multi-dimensional information such as the traffic volume, vehicle speed, lane occupancy rate, etc. of the target monitoring section at different time points. In order to determine which sections have experienced high-flow events, the computer system sets one or more high-flow determination thresholds based on business needs and actual conditions. These thresholds may be based on the absolute value of the traffic volume (such as more than 500 vehicles per hour), the relative value (such as the traffic volume has increased by more than 20% compared to usual) or a comprehensive scoring system that integrates multiple indicators.

[0091] Taking the absolute value of traffic flow as an example, assume that the high-flow determination threshold set by the computer system is more than 500 vehicles per hour. When scanning historical data, the computer system checks the traffic flow records of the target monitoring section at each time point. When it is found that the traffic flow at a certain time point exceeds this threshold, it will mark the section corresponding to that time point as an initial high-flow section. Then, the computer system continues to scan backward to find the next section that meets the high-flow determination conditions. If the traffic flow at adjacent time points also exceeds the threshold, and these time points correspond to the same section or adjacent sections (depending on the specific meaning of "occurring in sequence" in the business definition), then these sections will be regarded as part of the initial high-flow section sequence where high flow occurs in sequence. In actual applications, the determination of high-flow events may not only rely on a single traffic flow indicator. The computer system can also comprehensively consider multiple factors such as vehicle speed, lane occupancy, accident records, etc. to build a more comprehensive and accurate high-flow determination model. In addition, in order to cope with the periodic changes in traffic flow (such as morning and evening peaks, holidays, etc.), the computer system can also use methods such as dynamically adjusting thresholds or introducing time series analysis to improve the accuracy of determination.

[0092] Finally, through the processing of step S12, the computer system can extract a series of initial high-flow sections where high-flow occurs successively from the historical traffic monitoring data. These sections will become important input data for subsequent analysis (such as the merging of sections and the assignment of the first influencing factor in step S13, the evaluation of the commonality of sections in step S20, etc.), providing a solid data foundation for intelligent highway traffic monitoring and analysis. For example, on a busy highway, the computer system can identify multiple initial high-flow sections where high-flow occurs successively in time periods such as 8:00-9:00 in the morning, 5:00-6:00 in the afternoon, and 9:00-10:00 in the evening through the processing of step S12. These sections will be further processed and used in subsequent analysis to reveal the inherent laws of highway traffic flow and predict possible high-flow events in the future.

[0093] Step S13: Merge the continuously repeated high-flow sections among the multiple initial high-flow sections into one high-flow section, and assign a first impact factor to the high-flow section obtained after the merger based on the number of continuously repeated high-flow sections, and assign a preset first impact factor to the high-flow section that has not been merged.

[0094] In the embodiment of the present invention, step S13 is a key step for further processing the initial high-flow road section, which aims to refine the data and improve the accuracy and efficiency of subsequent analysis by merging consecutively repeated high-flow road sections and assigning corresponding first impact factors. This process is automatically performed by the computer system without manual intervention, ensuring the consistency and objectivity of the processing.

[0095] Specifically, the computer system traverses the multiple initial high-flow sections extracted in step S12 to check whether they are continuously repeated. Continuous repetition refers to high-flow events that are adjacent in time series and represent the same section or a similar section (taking into account data accuracy and error). For example, if the traffic volume of section A is detected to exceed the high-flow threshold at three consecutive time points, then these three initial high-flow sections are considered to be continuously repeated.

[0096] For the identified continuous and repeated high-flow sections, the computer system merges them into one high-flow section. The merged high-flow section contains the time range and characteristic information of the original multiple sections, but is presented in a more concise form. The purpose of this is to reduce data redundancy and enable subsequent analysis to focus more on meaningful traffic change patterns.

[0097] Next, the computer system assigns corresponding first impact factors to the high-flow sections based on the number of high-flow sections after merging. The first impact factor is a quantitative indicator used to characterize the frequency or importance of high-flow sections on the target monitoring section. In this embodiment, the first impact factor is simply set to the number of high-flow sections after merging. For example, if three consecutive high-flow sections are merged into one, the first impact factor of the section is assigned to 3. The basis for this is that consecutive and repeated high-flow events often reflect more significant and persistent traffic pressure on the section, and therefore deserve a higher weight in subsequent analysis.

[0098] For the initial high-flow road segments that do not perform the merge operation (i.e., those high-flow events that are discontinuous in time series or represent different road segments), the computer system assigns them a preset first impact factor. This preset value is usually a benchmark value used to represent those relatively less frequent or significant high-flow events. In the example, high-flow road segments 3 and high-flow road segments D are assigned a first impact factor of 1 because they do not need to be merged. This value can be adjusted according to actual application scenarios and needs.

[0099] Through the processing of step S13, the computer system not only simplifies the data set of the initial high-flow section, but also provides an important quantitative basis for subsequent analysis by assigning the first impact factor. These merged and weighted high-flow sections will be used as elements of the high-flow section set to participate in subsequent commonality evaluation, impact weight calculation, and target high-flow section prediction steps. This process fully reflects the strict requirements for data quality and analysis accuracy in intelligent highway traffic monitoring and analysis.

[0100] Step S14: construct a high-flow section set by combining the multiple high-flow sections obtained after the merging and the corresponding first influencing factors.

[0101] In the embodiment of the present invention, the key information extracted from the original historical traffic monitoring data in step S14 is collated and organized into a structured high-traffic road section set. This step is automatically completed by the computer system, ensuring the accuracy and consistency of the data, and providing a solid foundation for subsequent analysis and prediction.

[0102] Specifically, in step S13, the computer system has merged the initial high-flow sections and assigned corresponding first impact factors according to the number of high-flow sections after merging. These merged and weighted high-flow sections, as well as their respective first impact factors, now need to be integrated into a unified data structure, i.e., a high-flow section set.

[0103] The high-flow segment set is an ordered data structure, in which each element represents a high-flow segment that has been merged. Each element not only contains the basic information of the high-flow segment (such as segment identification, location, occurrence time range, etc.), but also is associated with the corresponding first impact factor. This impact factor reflects the frequency or importance of high-flow on the segment in history and is a key parameter in subsequent analysis.

[0104] The process of constructing a high-flow segment set is relatively straightforward. The computer system creates an empty data structure to store the set elements. Then, it traverses all high-flow segments processed in step S13 (including those that have not been merged and those that have been merged) and adds them to the set as new elements. During the addition process, the computer system ensures that each element correctly contains the basic information of the high-flow segment and the corresponding first impact factor.

[0105] Taking a busy highway as an example, after processing in step S13, the computer system can obtain the following high-flow sections and their first impact factors: section A (time range: 8:00-10:00 a.m., first impact factor: 3), section B (time range: 3:00-5:00 p.m., first impact factor: 2), section C (time range: 7:00-8:00 p.m., first impact factor: 1). In step S14, these sections and their impact factors will be constructed into a high-flow section set, and the elements in the set will be sorted according to their order in the time series.

[0106] The resulting high-traffic road section set is a highly structured and easy-to-query data resource. The computer system can directly reference this set in subsequent analysis steps to obtain the required high-traffic road section information and its influencing factors. For example, when evaluating the commonality of high-traffic road sections (step S20), the computer system can traverse each element in the set to calculate the similarity between them; when calculating the influence weight of high-traffic road sections (step S30), the first influencing factor in the set can be combined to comprehensively evaluate the impact of different road sections on future traffic conditions. In this way, the high-traffic road section set plays a vital role in intelligent highway traffic monitoring and analysis.

[0107] As an implementation mode, step S12, extracting a plurality of initial high-flow sections where high-flow occurs sequentially from the historical flow monitoring data, may specifically include:

[0108] Step S12A: Obtain the average duration of high flow rate on the target monitoring section, perform wandering extraction on the historical flow monitoring data based on the screening controls that satisfy the average duration of high flow rate, and obtain multiple initial high flow sections where high flow rate occurs sequentially.

[0109] In an embodiment of the present invention, step S12A calculates the average duration of high flow on the target monitoring section and sets the screening conditions accordingly, thereby extracting the initial section sequence where high flow occurs in sequence from the historical traffic monitoring data. First, the computer system obtains records of multiple uninterrupted high flow events on the target monitoring section over a period of time in the past. These records usually contain the start time, end time and corresponding traffic flow data of each high flow event. By statistically analyzing these records, the computer system can calculate the average duration of high flow. This indicator reflects the average level of duration of high flow events on the target monitoring section and is an important basis for setting subsequent screening controls.

[0110] Take a highway as an example. Suppose that in the past week, there were three high-flow events on this section of road, each lasting 4 minutes, 5 minutes, and 6 minutes. The computer system records these time data and calculates the average duration of high-flow events: (4 minutes + 5 minutes + 6 minutes) / 3 = 5 minutes. This average value shows that on this section of road, high-flow events lasted an average of about 5 minutes.

[0111] Next, the computer system sets a filter control based on the calculated average duration of high traffic. The length of the filter control is equal to the average duration, which is used to perform a walk extraction operation in the historical traffic monitoring data. Walk extraction is a method of sequentially scanning data and extracting information based on specific conditions, which allows the computer system to examine the traffic flow data at each time point one by one along the time series.

[0112] During the walk extraction process, the computer system compares whether the traffic flow at consecutive time points exceeds the preset high-flow threshold, and whether the total number of these consecutive time points is equal to or close to the length of the filter control (taking into account the data sampling interval and error). If these conditions are met, the road segments corresponding to these time points will be regarded as the initial road segments where high-flow occurs successively, and will be extracted for subsequent analysis.

[0113] Continuing with the previous example, the computer system sets the length of the filter control to 5 minutes and starts to perform wandering extraction in the historical traffic monitoring data. Assuming that the data sampling interval is 1 minute, the computer system checks the traffic flow data at 5 consecutive time points. If the traffic flow at 5 consecutive time points exceeds the high flow threshold within a certain period of time, the road section corresponding to these 5 minutes will be extracted as an initial high flow section.

[0114] In this way, the computer system can accurately identify the initial road section sequence where high traffic volume has occurred successively in history based on the average maintenance time of high traffic volume in the target monitored road section. These road section sequences not only reflect the actual changes in traffic volume, but also provide important data support for subsequent commonality evaluation, impact weight calculation and target high traffic volume section prediction. In the embodiment of the present invention, the implementation of step S12A is of great significance for improving traffic management efficiency and alleviating traffic congestion.

[0115] Or; Step S12B: obtain the average of the high-flow sections where high-flow occurs continuously in the target monitoring section, perform wandering extraction on the historical flow monitoring data based on the filtering controls that meet the average, and obtain multiple initial high-flow sections where high-flow occurs sequentially.

[0116] In the embodiment of the present invention, step S12B provides an effective method for extracting initial high-flow sections based on the average of consecutive high-flow sections. This step analyzes historical traffic monitoring data to identify the frequency pattern of high-flow events on the target monitoring section, and sets screening conditions accordingly to extract an initial high-flow section sequence that is valuable for subsequent analysis.

[0117] First, the computer system obtains the number of high-flow sections where high-flow events have occurred continuously over a period of time on the target monitored section. These records reflect the continuity and frequency changes of high-flow events over time. By counting these records, the computer system can calculate the mean of the sections where high-flow events have occurred continuously, that is, the average number of sections involved in each continuous high-flow event. This mean serves as an important reference for the subsequent setting of screening controls, and helps to identify sections with periodic or regular high-flow characteristics.

[0118] Take a highway as an example. Assume that in the past period of time, the number of high-flow sections where high-flow events occurred continuously on this section is 3, 5, and 7 in chronological order. The computer system performs statistical analysis on these data and calculates the mean of the sections with continuous high-flow events: (3+5+7) / 3=5. This mean indicates that, on average, continuous high-flow events on this section will involve about 5 sections. Next, the computer system sets the length of the filter control according to the calculated mean. The length of the filter control is equal to the mean of the sections with continuous high-flow events, and is used for wandering extraction operations in historical traffic monitoring data. During the wandering extraction process, the computer system checks the traffic flow data at each time point one by one along the time series, looking for those continuous sequences of high-flow sections whose number is equal to or close to the length of the filter control.

[0119] Continuing with the previous example, the computer system sets the length of the filter control to 5 and starts to perform a walk-through extraction of historical traffic monitoring data. During the extraction process, the computer system focuses on situations where the traffic volume exceeds the high-flow threshold for 5 or nearly 5 consecutive sections (taking into account the data sampling interval and error). If such a sequence of sections is found, the computer will extract them as initial high-flow sections for subsequent analysis and prediction.

[0120] It should be noted that since the actual traffic flow may be affected by many factors (such as weather, holidays, special events, etc.), the number of sections with continuous high traffic flow may fluctuate. Therefore, in actual applications, the computer system can set a certain tolerance or adjust the length of the filter control to adapt to the data characteristics in different scenarios.

[0121] Through the implementation of step S12B, the computer system can effectively extract the initial high-flow section sequence based on the mean characteristics of the continuous high-flow sections on the target monitoring section. These sequences not only reflect the periodic or regular changes in traffic flow, but also provide an important data basis for subsequent commonality evaluation, influence weight calculation and target high-flow section prediction. In the embodiment of the present invention, the implementation of step S12B is of great significance for improving traffic management efficiency and optimizing traffic flow distribution.

[0122] As another implementation, step S32, determining a first multiplication result of the first commonality evaluation coefficient corresponding to the remaining high-flow sections and the coded representation information of the next high-flow section of the remaining high-flow sections, may specifically include:

[0123] Step S321: For each high-traffic section in the remaining high-traffic sections, determine the second commonality evaluation coefficient of the coded characterization information of the high-traffic section and the coded characterization information of the high-traffic sections in the high-traffic section set except the high-traffic section, perform a standardization operation on the second commonality evaluation coefficient to obtain a standardized commonality evaluation coefficient, determine the multiplication result of the standardized commonality evaluation coefficient and the coded characterization information of the high-traffic section, and determine it as the second influencing factor of the high-traffic section.

[0124] In the embodiment of the present invention, step S321 not only considers the characteristic information of the road section Ri itself (reflected by the coded representation information), but also deeply analyzes the similarity relationship between the road section Ri and other road sections in the set, so as to more comprehensively evaluate the influence of the road section Ri on high-traffic events in the future. First, the computer system obtains the coded representation information of each road section Ri. The coded representation information is usually a multidimensional vector that contains the position information of the road section Ri in a specific traffic feature space. These features may include traffic volume, vehicle speed, lane occupancy, accident rate, etc., depending on the richness of the fused perception data and the analysis requirements. For example, the coded representation information of the road section Ri can be a vector containing 5 elements [f1, f2, f3, f4, f5], where each element corresponds to a specific traffic feature value.

[0125] Next, the computer system compares the coded representation information of the road segment Ri with the coded representation information of all other road segments in the "high-traffic road segment set" except Ri. In order to quantify this similarity relationship, the computer system uses a suitable commonality evaluation algorithm to calculate the second commonality evaluation coefficient. This coefficient reflects the degree of proximity between the road segment Ri and other road segments in the feature space. Commonly used algorithms include cosine similarity, Pearson correlation coefficient, etc. Taking cosine similarity as an example, if the road segment Rj (j≠i) is another road segment in the set, then the second commonality evaluation coefficient between the road segment Ri and the road segment Rj can be calculated by the following formula:

[0126] ;

[0127] Among them, “·” represents the dot product operation of the vector, and “|| ||” represents the modulus (i.e. length) of the vector.

[0128] After calculating the second commonality evaluation coefficients between the road segment Ri and all other road segments in the set, the computer system performs a standardization operation on these coefficients. The purpose of the standardization operation is to eliminate the scale differences between different coefficients so that they can be compared and weighted on the same scale. Commonly used standardization methods include minimum-maximum standardization, Z-score standardization, etc. Here, taking Z-score standardization as an example, the standardized commonality evaluation coefficient can be calculated by the following formula:

[0129] ;

[0130] in, and are the mean and standard deviation of the second commonality evaluation coefficients of the road segment Rj and all other road segments (including Ri). Note that here, the evaluation coefficients of each road segment Rj are actually standardized, rather than directly standardizing the results of the road segment Ri; but in practical applications, since we are concerned about the similarity relationship between the road segment Ri and other road segments, this relationship can be indirectly reflected through the standardized coefficients. After obtaining the standardized commonality evaluation coefficients, the computer system multiplies them with the encoded representation information of the road segment Ri. The specific form of the multiplication operation here varies depending on the application scenario, but the basic idea is to use the standardized commonality evaluation coefficients as weights to weight each element in the encoded representation information of the road segment Ri. The multiplication result may be a new vector or scalar value that integrates the characteristic information of the road segment Ri itself and its similarity relationship with other road segments. Finally, this multiplication result (or the result after further processing) is determined as the second influencing factor of the road segment Ri. The second influencing factor is a quantitative indicator that reflects the degree to which the road segment Ri is jointly affected by its own characteristics and the similarity of other road segments when it may cause high traffic events in the future. Step S321 calculates the second commonality evaluation coefficient between the coded representation information of the road section Ri and the coded representation information of other road sections, performs a normalization operation, and multiplies the coded representation information of the road section Ri, thereby finally obtaining the second impact factor of the road section Ri. This process provides an important intermediate variable for the subsequent calculation of the impact weight of the high-flow road section.

[0131] Step S322: determine the multiplication result of the first commonality evaluation coefficient corresponding to the remaining high-flow sections, the encoding representation information of the next high-flow section of the remaining high-flow sections, and the second influencing factor of the remaining high-flow sections, and determine it as the first multiplication result.

[0132] In the embodiment of the present invention, step S322 is an important step in the process of calculating the influence weight of the high-flow section, which combines multiple key factors to determine the first multiplication result of each "other high-flow section" (recorded as section Ri). This result not only takes into account the similarity between section Ri and the last high-flow section (recorded as section Ln) (reflected by the first commonality evaluation coefficient), but also incorporates the specific characteristics of the next high-flow section of section Ri (recorded as section Ri+1) (reflected by the coded representation information) and the relative importance of section Ri in the entire high-flow section set (reflected by the second influence factor).

[0133] First, the computer system has calculated the first commonality evaluation coefficient (denoted as αi) corresponding to the road section Ri through the previous steps. This coefficient reflects the similarity between the road section Ri and the road section Ln in the traffic feature space. At the same time, the computer system also obtains the coded representation information of the road section Ri+1 (denoted as Ei+1), which is a multidimensional vector containing the quantized value of the road section Ri+1 on a specific traffic feature. In addition, in step S321, the computer system also calculates the second influencing factor of the road section Ri (the specific vector or scalar form is not directly given here, but it is assumed that it has quantified the comprehensive influence of the road section Ri's own characteristics and the similarity of other roads in some way).

[0134] Next, the computer system will perform the core calculation in step S322. This step involves the multiplication of three key elements: the first commonality evaluation coefficient αi, the encoded representation information Ei+1 of the road segment Ri+1, and the second impact factor of the road segment Ri (assuming that it can be converted into a form compatible with Ei+1 for multiplication in some way, such as a vector with the same dimension as Ei+1 or a scalar that can weight each element in Ei+1).

[0135] However, it should be noted that since the specific form of the second impact factor is not explicitly given in step S321 (it may be a vector, scalar or more complex data structure), the specific form of the multiplication operation may be different in actual calculation. However, in order to simplify the description and give a specific calculation formula example, it can be assumed that the second impact factor is a vector (denoted as βi) of the same dimension as Ei+1, and each element of it weights the corresponding element in Ei+1.

[0136] Based on the above assumptions, the first multiplication result (denoted as Wi) can be calculated by the following formula:

[0137] ;

[0138] Among them, "⊙" represents the element-by-element multiplication operation, that is, each element in βi is multiplied by the element in the corresponding position in Ei+1 to obtain a new vector with the same dimension as Ei+1. Then, this new vector is multiplied by the first commonality evaluation coefficient αi (note that αi is a scalar here, so it will be multiplied by each element of the vector), and finally a weighted vector or scalar is obtained as the first multiplication result Wi.

[0139] In practical applications, since we may be more concerned with a single weight value rather than a vector, we can further process the above formula, such as summing or averaging the weighted vectors to obtain a scalar result. However, no matter what form is adopted, the first multiplication result Wi comprehensively reflects the degree to which the road segment Ri may cause high traffic events in the future, which is affected by its similarity with the road segment Ln, the specific characteristics of the road segment Ri+1, and the relative importance of the road segment Ri in the entire set.

[0140] In summary, step S322 calculates the first multiplication result of each remaining high-flow section by combining the first commonality evaluation coefficient, the encoding representation information of the next high-flow section, and the second impact factor of the section itself. This process provides an important quantitative basis for the subsequent determination of the impact weight of the high-flow section.

[0141] As an implementation manner, step S20, determining the first commonality evaluation coefficient between the last high-flow section in the high-flow section set and the remaining high-flow sections, may specifically include:

[0142] For each remaining high-traffic road segment, perform the following:

[0143] Step S21: determining a dot product result of the coded representation information of the last high-flow road section and the coded representation information of the remaining high-flow road sections;

[0144] Step S22: determining a product of the norm of the coded representation information of the last high-flow road section and the norm of the coded representation information of the remaining high-flow road sections;

[0145] Step S23: Calculate the ratio of the dot product result to the norm product result, and determine it as the first commonality evaluation coefficient.

[0146] In the embodiment of the present invention, the core task of step S20 is to quantify the similarity between the last high-flow segment (denoted as segment Ln) in the high-flow segment set and each of the remaining high-flow segments (denoted as segment Ri, where i=1,2,...,n-1), that is, to calculate the first commonality evaluation coefficient between them.

[0147] In step S21, the computer system obtains the coded representation information of the road section Ln and the road section Ri. The coded representation information is usually obtained by mapping the traffic characteristics of the road section into a low-dimensional vector space through some embedding technology (such as word embedding, graph embedding, etc.). These vectors capture the key characteristics of the road section in terms of traffic flow, vehicle speed, lane occupancy, etc., so that the similarities and differences between the road sections can be effectively represented in the vector space.

[0148] For example, the coded representation information of the road section Ln may be a vector containing 5 elements [ln1, ln2, ln3,ln4, ln5], while the coded representation information of the road section Ri may be another vector [ri1, ri2, ri3, ri4, ri5]. These two vectors represent the positions of the road section Ln and the road section Ri in the specific traffic feature space respectively. Next, the computer system calculates the dot product result of the two vectors. The dot product operation is the process of multiplying the corresponding elements of the two vectors and then summing them. The result reflects the degree of proximity of the two vectors in direction. For the coded representation information of the road section Ln and the road section Ri, the dot product result can be calculated by the following formula:

[0149] ;

[0150] The larger the value is, the more similar the traffic characteristics of road section Ln and road section Ri are.

[0151] After obtaining the dot product result, the computer system also needs to calculate the norm (i.e., the length or modulus of the vector) of the coded representation information of the road segment Ln and the road segment Ri, and multiply the two norms. The norm is a metric in the vector space, which represents the distance from the vector to the origin. For the coded representation information of the road segment Ln and the road segment Ri, their norms can be calculated by the following formulas:

[0152] ;

[0153] ;

[0154] Then, we multiply these two norms to get the norm multiplication result:

[0155] ;

[0156] This value reflects the product of the "absolute size" of segment Ln and segment Ri in the vector space.

[0157] Step S23: Calculate the first commonality evaluation coefficient

[0158] Finally, the computer system divides the dot product result obtained in step S21 by the norm product result obtained in step S22 to obtain the first commonality evaluation coefficient. This coefficient is actually a variant of cosine similarity, which measures the degree of proximity of two vectors in direction while removing the influence of vector length. For road segment Ln and road segment Ri, the first commonality evaluation coefficient can be calculated by the following formula:

[0159] ;

[0160] The value range of this coefficient is between [-1, 1]. The closer the value is to 1, the more similar the two road segments are. The closer the value is to -1, the less similar the two road segments are. A value of 0 means that the two road segments are orthogonal in direction.

[0161] In summary, step S20 successfully quantifies the similarity between the two road sections by calculating the dot product result, norm multiplication result and the ratio (i.e., the first commonality evaluation coefficient) of the encoded representation information of road section Ln and road section Ri. This result provides an important quantitative basis for the subsequent analysis of the mutual influence between road sections and the prediction of future high traffic events.

[0162] As an implementation method, the method may further include the following steps:

[0163] Step S50: Acquire multiple flow limiting strategies to be adopted for the target monitoring section.

[0164] Specifically, when executing step S50, the computer system accesses a database or knowledge base containing multiple flow limiting strategies. This database may be pre-defined and maintained by the traffic management department, which stores a variety of flow limiting measures that have been verified in practice, such as adjusting the timing of traffic lights, setting up temporary traffic control areas, guiding vehicles to detour, enabling variable lanes, etc. Each strategy is accompanied by a detailed description, applicable conditions, and expected effect evaluation, so that the computer system can screen and match according to the current traffic conditions.

[0165] In order to select flow control strategies more accurately, the computer system can also analyze the real-time traffic data and historical traffic monitoring data. These data come from various sensing devices (such as cameras, radars, vehicle detectors, etc.) installed on and around the target monitoring section. They can capture and transmit information about key traffic parameters such as traffic flow, speed, lane occupancy, accident records, etc. in real time. Through in-depth mining and analysis of these data, the computer system can identify key issues and potential risk points in the current traffic conditions, so as to select flow control strategies more targeted.

[0166] In some advanced application scenarios, computer systems may also use machine learning models to assist in the selection of flow-limiting strategies. These models can be trained and optimized based on a large amount of historical traffic data and the implementation effects of flow-limiting strategies, and learn the most effective flow-limiting measures under different traffic conditions. When faced with new traffic conditions, computer systems can input real-time traffic data into these models, and the models will automatically calculate the expected effect score or probability distribution of each flow-limiting strategy, thereby helping the computer system to quickly and accurately select the optimal flow-limiting strategy.

[0167] For example, on a busy highway, the computer system obtains a series of possible flow control strategies through step S50, including adjusting the entrance ramp signal light timing to reduce the traffic flow entering the main line, setting up temporary traffic control areas on specific sections to guide vehicles to detour, and enabling variable lanes to improve road capacity. After real-time analysis of current traffic data, the computer system finds that a certain section of the road is severely congested due to an accident, and it is expected that the traffic flow will continue to increase in the future. Based on this information, the computer system can choose to set up temporary traffic control areas and issue traffic guidance information as the current flow control strategy combination to quickly alleviate congestion and restore road traffic.

[0168] In short, step S50 is an important part of formulating effective traffic management strategies in intelligent highway traffic monitoring and analysis. By comprehensively utilizing database query, real-time data analysis, and machine learning model prediction, the computer system can quickly and accurately obtain a series of flow control strategies for the target monitoring section, providing strong support for subsequent strategy selection and implementation.

[0169] Step S60: determining a flow limiting strategy to be adopted for the target monitoring section from among multiple flow limiting strategies according to the target high-flow section.

[0170] In the embodiment of the present invention, step S60 is the core link of decision making, which requires the computer system to be able to intelligently select the most appropriate flow limiting measure from multiple alternative flow limiting strategies based on the predicted target high-flow road section to effectively cope with the upcoming traffic pressure.

[0171] When executing step S60, the computer system reviews the multiple flow limiting strategy sets obtained in step S50. These strategies may cover a variety of means from simple traffic signal adjustments to complex traffic induction schemes, aiming to optimize traffic flow, reduce congestion and improve road traffic efficiency in different ways. Next, the computer system analyzes the characteristics of the target high-flow section and the traffic environment in which it is located. These characteristics may include the geographical location of the section, historical traffic patterns, current traffic volume, vehicle speed, lane occupancy, and weather conditions. This information is crucial for understanding the causes of high-flow events, predicting their development trends, and evaluating the effects of different flow limiting strategies. In order to more accurately select flow limiting strategies, the computer system can use a variety of evaluation methods. A common method is a rule-based evaluation system that evaluates the degree of match between each flow limiting strategy and the target high-flow section based on a preset set of rules. These rules may be formulated based on the experience and knowledge of experts, taking into account best practices under different traffic conditions. For example, if the target high-flow section is temporarily congested due to an accident, the computer system may give priority to strategies for setting up temporary traffic control areas and issuing emergency traffic induction information.

[0172] Another smarter approach is to use machine learning models for policy selection. In this case, the computer system encodes the characteristics of the target high-traffic road section (such as traffic volume, speed, accident records, etc.) into a feature vector or feature array and passes it as input data to a pre-trained machine learning model. This model may be a classifier, regressor, or reinforcement learning agent, which predicts which flow restriction strategy is most effective under given traffic conditions by learning the relationship between historical traffic data and the effectiveness of flow restriction strategy implementation.

[0173] Taking the classifier as an example, it can output a probability distribution or confidence score, indicating the probability of each flow-limiting strategy being selected as the optimal strategy. Based on this output, the computer system can select the strategy with the highest probability or the highest score as the proposed flow-limiting measure. For example, if the model predicts that adjusting the traffic light timing strategy is most effective in alleviating congestion on the target high-flow section, then the computer system will select this strategy as the recommended solution.

[0174] Finally, the computer system submits the selected flow control strategy to the traffic management department for implementation. During the implementation process, the computer system can also continuously monitor changes in traffic conditions and evaluate and adjust the effectiveness of the flow control strategy based on real-time data. This closed-loop feedback mechanism helps ensure the effectiveness and adaptability of flow control measures, thereby maximizing the efficiency and effectiveness of traffic management.

[0175] In summary, step S60 is a key link in decision-making in intelligent highway traffic monitoring and analysis. By comprehensively using methods such as feature analysis, rule evaluation, and machine learning model prediction, the computer system can intelligently select the most appropriate flow limiting strategy to cope with the traffic pressure of the target high-flow section. This process not only reflects the advancement and intelligence of technology, but also provides strong support for improving traffic management efficiency and improving the public travel experience.

[0176] As an implementation mode, step S50, obtaining multiple flow limiting strategies to be adopted for the target monitoring section, may specifically include:

[0177] Step S51: obtaining the road section characterization information of the target monitoring road section;

[0178] Step S52: determining a third commonality evaluation coefficient between the road segment characterization information and each flow limiting strategy in the prior strategy set;

[0179] Step S53: Arrange the third commonality evaluation coefficients in order from large to small to obtain an arrangement result, and obtain K multiple current limiting strategies that are close to each other in the arrangement result, where K≥1.

[0180] In step S51, the computer system collects multi-dimensional data of the target monitoring section, including but not limited to the traffic volume, speed, lane occupancy, accident records, weather conditions, etc. of the section. After processing, these data will be converted into section characterization information, which is usually a vector or array containing multiple features. For example, the section characterization information can be an array containing 5 elements [traffic volume, average speed, lane occupancy, number of accidents, weather condition code], each element quantifies the characteristics of the section in a certain aspect.

[0181] The prior strategy set is a database containing a variety of known flow limiting strategies, each of which has its specific application scenarios and effects. The computer system compares the segment characterization information of the target monitoring section with the description or characterization information of each flow limiting strategy in the prior strategy set, and calculates the third commonality evaluation coefficient between them. This coefficient reflects the similarity or degree of match between the segment characteristics and the flow limiting strategy. The calculation method may include cosine similarity, Pearson correlation coefficient, etc. For example, if a flow limiting strategy is particularly suitable for sections with high traffic volume and high lane occupancy, then when the segment characterization information of the target monitoring section shows high traffic volume and high lane occupancy, the third commonality evaluation coefficient between the strategy and the target section will be higher.

[0182] The computer system sorts all the calculated third commonality evaluation coefficients and then selects the top K flow limiting strategies as candidate strategies. The value of K can be determined based on actual needs and the richness of the strategy library. For example, if K is set to 3, the computer system will select the three flow limiting strategies that best match the target monitoring section as recommended solutions.

[0183] In another embodiment, step S50, obtaining multiple flow limiting strategies to be adopted for the target monitoring section, is implemented using a machine learning model, which may include a model training and reasoning process, specifically including:

[0184] Step S501: obtaining a deep neural network to be trained and a set of learning samples, wherein the set of learning samples includes a plurality of learning samples, each of which includes the segment representation information of the monitored segment sample, the segment representation information of the intercepted segment sample, and a supervision mark, wherein the supervision mark is used to indicate the confidence level of the intercepted segment sample in the monitored segment sample having high traffic;

[0185] Step S502: calling the deep neural network to be trained according to the road segment representation information of the monitored road segment sample and the road segment representation information of the intercepted road segment sample to obtain an estimation result;

[0186] Step S503: Determine the training error based on the supervision mark and the estimation result, and adjust the network parameters of the deep neural network to be trained based on the training error to obtain the debugged deep neural network.

[0187] Step S504: calling the debugged deep neural network according to the section representation information of the target monitoring section and the section representation information of the flow limiting strategy in the priori strategy set to obtain candidate estimation results corresponding to multiple flow limiting strategies respectively;

[0188] Step S505: Based on the candidate estimation results of the flow limiting strategy, multiple flow limiting strategies to be adopted for the target monitoring section are determined in the priori strategy set.

[0189] In step S501, the computer system first determines a deep neural network model to be trained, such as a randomly initialized recurrent neural network, and at the same time, obtains a learning sample set containing multiple learning samples. Each learning sample includes the segment characterization information of the monitored segment sample, the segment characterization information of the intercepted segment sample, and a supervisory mark. The supervisory mark is used to indicate the confidence level of the occurrence of high traffic in the intercepted segment sample, that is, the possibility of high traffic in the segment under given conditions. For example, the learning sample can be a record in historical traffic data, in which the segment characterization information describes the traffic conditions of the segment, the intercepted segment is the high traffic prediction result of the segment at a certain point in the future, and the supervisory mark is the confidence score of the prediction result.

[0190] In step S502, the computer system inputs the road segment representation information in the learning sample into the deep neural network to be trained, and allows the network to predict the possibility of high traffic in the intercepted road segment sample based on the input data. This prediction process is the forward propagation process of the neural network.

[0191] In step S503, the training error is calculated using the supervision mark and the prediction result, and the network parameters of the deep neural network are adjusted accordingly to optimize the model performance. When executing step S503, the computer system compares the prediction result of the deep neural network model for each learning sample with the supervision mark of the sample. The supervision mark is a known, real probability of high traffic or confidence score, which represents the actual situation of high traffic in the intercepted section sample in the learning sample at a certain point in the future. The prediction result is the probability of high traffic or confidence score predicted by the neural network model based on the section representation information of the monitored section sample.

[0192] In order to quantify the difference between the estimated result and the supervised label, the computer system uses a loss function to calculate the training error. The loss function is a mathematical expression that measures the degree of inconsistency between the model's predicted value and the true value. The smaller the value, the more accurate the model prediction. In deep learning, commonly used loss functions include Mean Squared Error (MSE), Cross-Entropy Loss, etc. For the binary classification problem in this scenario (i.e., determining whether a road section will have high traffic), the cross-entropy loss function is a suitable choice.

[0193] Taking the cross entropy loss function as an example, the computer system calculates the cross entropy loss between the estimated result and the supervisory label for each learning example, and then accumulates or averages the loss values ​​of all examples to obtain the total loss value on the entire training set, that is, the training error. This training error reflects the performance of the current neural network model in predicting high traffic events.

[0194] Next, the computer system uses the backpropagation algorithm to calculate the gradient of the training error with respect to the network parameters (such as weights and biases). The backpropagation algorithm is an effective neural network training algorithm that calculates the gradient of each network parameter layer by layer by propagating the training error from the output layer to the input layer, thereby guiding the update direction of the network parameters. After obtaining the gradient, the computer system uses an optimization algorithm (such as gradient descent, Adam optimizer, etc.) to update the network parameters. The optimization algorithm adjusts the value of the network parameters according to the size and direction of the gradient to minimize the training error. This process is iterated multiple times, and each iteration uses the current network parameters to learn the training set, calculate the new training error and gradient, and then update the network parameters. As the iterations proceed, the training error usually decreases, indicating that the prediction performance of the neural network model is constantly improving.

[0195] Finally, when the training error meets the preset stopping conditions (such as reaching the minimum error threshold, the training rounds reaching the upper limit, etc.), the training process will terminate, and the deep neural network model obtained at this time is the debugged model. This model has learned the ability to predict high-flow events from the road segment representation information, and can be applied to new target monitoring sections to screen and recommend flow control strategies.

[0196] Through the implementation of step S503, the computer system can automatically optimize the performance of the deep neural network model, so that it can more accurately predict high-traffic events and screen out appropriate flow limiting strategies, thereby providing more intelligent and efficient decision support for smart high-speed traffic monitoring and analysis.

[0197] In step S504, using the debugged deep neural network model, the computer system predicts the segment characterization information of the target monitored segment and the segment characterization information of each flow limiting strategy in the prior strategy set, and obtains the estimated results of the effect that each flow limiting strategy may bring after being implemented on the target segment. These estimated results can be the probability, confidence score or other quantitative indicators of the flow limiting strategy in alleviating high flow conditions.

[0198] In step S505, the computer system sorts or screens the current limiting strategies in the priori strategy set according to the candidate prediction results, and selects one or more current limiting strategies with the best prediction effect as the recommended solution. This selection process may be based on the absolute value of the prediction results, the ranking order, or other customized evaluation criteria.

[0199] In one embodiment, each high-flow section corresponds to a high-flow generating factor. Based on this, step S60 determines the flow limiting strategy to be adopted for the target monitoring section from among multiple flow limiting strategies according to the target high-flow section, which may specifically include:

[0200] Step S61: determine the current limiting strategy types corresponding to the multiple current limiting strategies respectively, encode the current limiting strategy types, and obtain the current limiting strategy type representation vector.

[0201] In the embodiment of the present invention, step S61 classifies, encodes and generates characterization vectors for multiple flow limiting strategies. This step is intended to provide a basis for subsequent commonality evaluation and strategy matching, ensuring that the computer system can accurately and efficiently determine the flow limiting measures that are most suitable for the target monitoring section.

[0202] First, the computer system conducts an in-depth analysis of the collected multiple flow-limiting strategies and identifies the flow-limiting strategy types to which they belong. These types may be divided based on a variety of factors, including but not limited to the scope of the strategy, the implementation method, the expected effect, and the mechanism of influence on traffic flow. For example, flow-limiting strategy types may include local signal light control, global lane closure, dynamic traffic induction, temporary speed limit zone setting, etc. Each type represents a specific idea and means to deal with traffic congestion or high flow conditions.

[0203] Next, the computer system encodes each flow-limiting strategy type. The purpose of encoding is to convert the abstract concept of flow-limiting strategy type into a form that can be processed by the computer, that is, the flow-limiting strategy type representation vector. This vector is multi-dimensional, and each dimension corresponds to a key feature or attribute of the flow-limiting strategy type. These features may include the difficulty of implementing the strategy, the cost input, the duration of the effect, the scope of impact on surrounding traffic, etc. Through encoding, the computer system can map each flow-limiting strategy type into a unique vector space, which is convenient for subsequent calculation and analysis. In actual operation, the encoding process may involve a series of complex algorithms and rules. For example, the computer system can use a method based on feature extraction and vectorization to convert the descriptive text or structured data of the flow-limiting strategy type into a numerical vector. This process may require the use of natural language processing (NLP) technology to parse text information, or the use of data mining technology to extract key features from structured data. Then, these features are quantified into specific values ​​and arranged in a vector form in a certain order.

[0204] Taking the local traffic light control strategy as an example, its flow control strategy type representation vector may contain the following eigenvalues: [implementation difficulty = 2, cost input = medium, effect duration = short time, impact range on surrounding traffic = local]. Here, the implementation difficulty and effect duration are expressed in numerical form, while the cost input and impact range on surrounding traffic are quantified by predefined levels or categories. Through this encoding method, the computer system can effectively distinguish and compare the local traffic light control strategy with other flow control strategy types in the vector space.

[0205] Finally, after processing in step S61, the computer system will obtain a set of multiple flow limiting strategy type characterization vectors. This set provides necessary data support for calculating the commonality evaluation coefficient between the target high-flow section and the flow limiting strategy type in the subsequent steps. By comparing the similarity or matching degree between different flow limiting strategy types and the characteristics of the target high-flow section, the computer system can intelligently recommend the most appropriate flow limiting measures, thereby effectively dealing with traffic congestion or high-flow conditions.

[0206] Step S62: Determine the fourth commonality evaluation coefficient between the encoding representation information of the target high-flow road section and the flow-limiting strategy type representation vector.

[0207] In the embodiment of the present invention, step S62 is a key step to connect the target high-flow section characteristics with the flow limiting strategy type, which measures the similarity or matching degree between the two by calculating the fourth commonality evaluation coefficient. The implementation of this step is crucial for the subsequent determination of the most appropriate flow limiting strategy.

[0208] First, the computer system reviews the set of flow limiting strategy type characterization vectors obtained in step S61, each of which represents the key features of different flow limiting strategy types. At the same time, the computer system also holds the coded characterization information of the target high-flow road section, which is the result of abstracting and quantifying the traffic conditions of the target road section by integrating the perception data. It is usually a multi-dimensional numerical vector, including multiple features such as traffic flow, speed, lane occupancy, accident records, etc.

[0209] Next, in order to determine the degree of commonality between the target high-flow road section and each flow restriction strategy type, the computer system uses a suitable similarity measurement method to calculate the fourth commonality evaluation coefficient. This measurement method should be able to capture the proximity between vectors in the feature space. Commonly used methods include cosine similarity, Euclidean distance, Pearson correlation coefficient, etc.

[0210] Taking cosine similarity as an example, the computer system calculates the cosine value of the angle between the encoded representation information of the target high-flow section and the representation vector of each flow-limiting strategy type. The closer the cosine value is to 1, the closer the two vectors are in direction, that is, the higher the degree of commonality between the target high-flow section and the corresponding flow-limiting strategy type. Specifically, the calculation process may involve multiplying the corresponding elements of the two vectors and then summing them, and then dividing them by the product of the two vector moduli to obtain a standardized similarity score.

[0211] For example, suppose the encoding representation information of the target high-flow road section is vector A=[traffic volume=800, average speed=40km / h, lane occupancy=90%], and the representation vector of a certain flow-limiting strategy type (such as local traffic light control) is B=[applicable traffic volume range=500-1000, average speed impact=significant reduction, lane occupancy sensitivity=high]. Although these two vectors do not directly correspond numerically, in actual calculations, the descriptive features of the strategy type can be converted into quantitative scores or weights for comparison with the actual feature vectors of the road section. Through cosine similarity calculation, if the result is close to 1, it can be considered that the local traffic light control strategy is highly matched with the traffic conditions of the target high-flow road section. It should be noted that in actual applications, since the representation vector of the flow-limiting strategy type may contain non-numerical features, additional preprocessing steps may be required, such as feature encoding, weight assignment, or the use of more complex similarity measurement models. In addition, if the technical solution involves machine learning models, these models can also be used to learn and optimize the calculation process of common evaluation coefficients, and continuously improve the accuracy and efficiency of the evaluation through a large amount of historical data and feedback.

[0212] Step S63: Determine the flow limiting strategy to be adopted for the target monitoring section from among multiple flow limiting strategies based on the fourth commonality evaluation coefficient.

[0213] In an embodiment of the present invention, step S63 is the final decision link of strategy selection and recommendation, which determines the flow limiting strategy to be adopted that is most suitable for the target monitoring section from multiple alternative flow limiting strategies based on the fourth commonality evaluation coefficient calculated in step S62.

[0214] First, the computer system reviews the fourth commonality evaluation coefficients between all flow limiting strategy types and the target high-flow road sections calculated in step S62. These coefficients constitute an evaluation index system that reflects the degree of match between different flow limiting strategy types and the traffic conditions of the target road section. The higher the coefficient value, the more suitable the flow limiting strategy type is for application to the current target monitoring road section.

[0215] Next, the computer system sorts or filters the flow limiting strategies based on these evaluation coefficients. The sorting operation may be performed in descending order of coefficient values, so that the flow limiting strategies at the top are the options that best match the target road section. The screening operation may set a threshold and only retain the flow limiting strategies with coefficient values ​​exceeding the threshold as candidate solutions. Both methods can effectively narrow the selection range and improve decision-making efficiency.

[0216] However, in practical applications, relying solely on the fourth commonality evaluation coefficient for decision-making may not be comprehensive enough. The computer system also needs to consider other factors, such as the implementation cost of the flow control strategy, the impact on surrounding traffic, feasibility, and the specific needs of the traffic management department. Therefore, after determining the preliminary candidate strategies, the computer system further evaluates the comprehensive benefits of these strategies to ensure that the final selected flow control strategy not only meets the actual situation of the target monitoring section, but also can achieve the overall optimization of traffic management.

[0217] For example, suppose that after the calculation and ranking of the fourth commonality evaluation coefficient, the two strategies of local signal light control and lane closure are determined as preliminary candidate solutions. At this point, the computer system can further compare the implementation costs of the two strategies: local signal light control may be low-cost and easy to implement, but the effect on traffic flow regulation may be relatively limited; while lane closure, although more effective, is more expensive and may have a greater impact on surrounding traffic. After comprehensive consideration, if the traffic management department prefers a plan with a higher cost-effectiveness ratio, then local signal light control can become the final flow control strategy to be adopted.

[0218] In addition, if the technical solution involves machine learning models, these models can also play an important role in step S63. For example, a decision support model can be trained to comprehensively evaluate the effects and costs of different flow limiting strategies, and give optimal strategy recommendations based on historical data and real-time traffic conditions. This data-driven decision-making method can further improve the accuracy and efficiency of flow limiting strategy selection.

[0219] In one implementation, step S63, determining a flow limiting strategy to be adopted for the target monitoring section from a plurality of flow limiting strategies based on the fourth commonality evaluation coefficient, may specifically include:

[0220] Step S631: Obtain element values ​​of current limiting analysis elements at multiple evaluation levels of the current limiting strategy, wherein the current limiting analysis elements at multiple evaluation levels include a fourth common evaluation coefficient.

[0221] In step S631 of the embodiment of the present invention, the computer system comprehensively and accurately obtains the element values ​​of the flow limiting analysis elements of each flow limiting strategy at multiple evaluation levels. These element values ​​not only provide a data basis for subsequent screening and sorting, but also ensure that the final recommended flow limiting strategy can comprehensively consider various practical factors and achieve the optimization of traffic management.

[0222] First, the computer system retrieves detailed information of all candidate strategies from the flow control strategy database. These strategies may include but are not limited to signal light control, lane closure, traffic induction, speed limit zone setting, etc. Each strategy has its specific application scenario and effect. For each strategy, the computer system collects information on flow control analysis elements at multiple evaluation levels related to it.

[0223] These evaluation levels may cover multiple dimensions such as strategy effectiveness evaluation, cost investment, implementation difficulty, and impact on surrounding traffic. Specifically for each dimension, there will be specific flow-limiting analysis elements and their corresponding element values. For example, under the effect evaluation dimension, the fourth commonality evaluation coefficient is a key flow-limiting analysis element, which reflects the degree of match between the flow-limiting strategy and the traffic conditions of the target monitored section. In addition to the fourth commonality evaluation coefficient, there may be other elements such as the historical average congestion relief time, the proportion of traffic flow reduction, etc., which are used to more comprehensively evaluate the effectiveness of the strategy.

[0224] Under the cost input dimension, flow control analysis factors may include direct costs (such as human and material inputs), indirect costs (such as economic losses caused by traffic control), etc. These factor values ​​are usually expressed in numerical form to facilitate subsequent calculations and comparisons. The implementation difficulty dimension may involve evaluation factors such as technical difficulty, operational complexity, and coordination workload. The impact on surrounding traffic dimension focuses on the chain reactions that may occur after the implementation of the strategy, such as whether it will lead to increased congestion on adjacent roads and whether it will affect the normal operation of public transportation.

[0225] In order to obtain the element values ​​of these flow-limiting analysis elements, the computer system can use a variety of data sources and methods. For example, for element values ​​based on statistical data such as the historical average congestion relief time, the computer system can directly query and obtain them from the database of the traffic management department; for element values ​​with strong subjectivity such as implementation difficulty, the computer system may need to refer to the evaluation opinions of experts or collect them through questionnaires; for element values ​​obtained by calculation such as the fourth commonality evaluation coefficient, they need to be obtained after completing the relevant calculations in the previous steps.

[0226] Taking the traffic light control strategy as an example, its flow limiting analysis factors may include: the fourth commonality evaluation coefficient = 0.85 (indicating that the strategy is highly matched with the traffic conditions of the target road section), historical average congestion relief time = 20 minutes (indicating that the implementation of the strategy can relieve the road section congestion within 20 minutes on average), direct cost = medium (indicating that the implementation of the strategy requires a certain amount of capital and manpower investment), implementation difficulty = low (indicating that the strategy is easy to operate and implement), and impact on surrounding traffic = local (indicating that the strategy mainly affects the traffic conditions in the target road section and its surrounding area within a small range). These factor values ​​together constitute the comprehensive performance of the traffic light control strategy at multiple evaluation levels, providing strong support for subsequent strategy selection.

[0227] Step S632: Call the screening network to make an estimate based on the current limiting analysis elements at multiple evaluation levels to obtain the selection confidence of the current limiting strategy.

[0228] In the embodiment of the present invention, step S632 uses the screening network to comprehensively consider the current limiting analysis elements at multiple evaluation levels to estimate the selection confidence of each current limiting strategy. The implementation of this step depends not only on the detailed data collected in the early stage, but also on the precise calculation of the key technology of the screening network.

[0229] First, the flow limiting analysis elements of multiple evaluation levels obtained in step S631 are organized into an input format that can be recognized by the screening network. These elements may include numerical data (such as the fourth commonality evaluation coefficient, the historical average congestion relief time, etc.), categorical data (such as the implementation difficulty level, the scope of impact on surrounding traffic, etc.), and text or image data that may have been encoded (such as experts' text descriptions of the effectiveness of the strategy, real-time images of traffic monitoring cameras, etc.). In order to input these diverse data into the screening network, the computer system may need to perform data preprocessing operations, such as data cleaning, feature scaling, one-hot encoding (One-Hot Encoding) or embedding representation (Embedding), etc. Next, the computer system inputs the organized flow limiting analysis elements into the screening network. The screening network is a pre-trained machine learning model that may be built based on a deep learning algorithm, such as a convolutional neural network (CNN), a recurrent neural network (RNN), a fully connected neural network (FCN) or a combination thereof. The design goal of this model is to learn the complex mapping relationship between flow limiting analysis elements and the confidence of flow limiting strategy selection, so that when a new flow limiting analysis element is given, the selection confidence of the corresponding flow limiting strategy can be accurately estimated. Inside the screening network, the input data will be processed by multiple layers of neurons, and each layer will introduce nonlinear factors through the activation function, so that the model can learn the nonlinear relationship between input and output. As the data is transmitted layer by layer, the model will gradually abstract higher-level feature representations and finally output the selection confidence of each flow limiting strategy. These confidence values ​​are usually between 0 and 1, indicating the model's confidence in whether the flow limiting strategy is suitable for the current traffic conditions.

[0230] Taking the traffic light control strategy as an example, assume that the corresponding flow limiting analysis factors include the fourth commonality evaluation coefficient = 0.85, the historical average congestion relief time = 20 minutes, the implementation difficulty level = low, and the impact range on surrounding traffic = local. When these factors are input into the screening network, the network will calculate and reason based on the parameters and rules learned internally, and finally output a selection confidence value, such as 0.9. This value means that the screening network has a 90% confidence that the traffic light control strategy is the appropriate choice under the current traffic conditions.

[0231] It should be noted that the performance and accuracy of the screening network depends largely on its training process and the quality of the training data. During the training phase, the computer system uses a large amount of sample data containing flow control analysis elements and real selection results to train the model so that the model can learn the effective mapping relationship from input to output. In addition, as traffic conditions and flow control strategies continue to change, the screening network also needs to regularly update its parameters and rules to maintain the timeliness and accuracy of its prediction capabilities.

[0232] Step S633: The flow limiting strategy corresponding to the selected confidence level that is greater than the selected confidence level reference value is used as the flow limiting strategy to be adopted for the target monitoring section.

[0233] In the embodiment of the present invention, the implementation of step S633 depends on the selection confidence calculated by the screening network in step S632 and a preset selection confidence reference value.

[0234] First, the computer system obtains the selection confidence of each flow limiting strategy based on the output result of step S632. These confidence values ​​reflect the evaluation results of the screening network for the applicability of each strategy under the current traffic conditions. For example, for the signal light control strategy, its selection confidence may be 0.85; for the lane closure strategy, its selection confidence may be 0.7; and for the traffic induction strategy, its selection confidence may be 0.9. Next, the computer system presets a selection confidence reference value as a threshold for screening flow limiting strategies. This reference value is usually determined based on a combination of factors such as historical experience, the requirements of the traffic management department, and real-time traffic conditions. The purpose is to ensure that only those flow limiting strategies with a high enough selection confidence will be recommended to the traffic management department, thereby improving the effectiveness and feasibility of strategy implementation.

[0235] After obtaining the selected confidences of all flow-limiting strategies and selecting confidence reference values, the computer system compares these confidence values ​​one by one. For those flow-limiting strategies that are greater than the selected confidence reference values, the computer system regards them as the flow-limiting strategies to be adopted for the target monitoring section. These strategies are usually excellent choices at multiple evaluation levels, which can effectively cope with the traffic pressure of the target section and improve road traffic efficiency.

[0236] Taking the selection confidence reference value of 0.8 as an example, if the selection confidence of the signal light control strategy is 0.85, then the strategy will be used by the computer system as one of the proposed flow limiting strategies. On the contrary, if the selection confidence of the lane closure strategy is 0.7, which is lower than the reference value of 0.8, then the strategy will not be recommended. In this way, through comparison and screening, the computer system can ensure that the final recommended flow limiting strategy not only meets the requirements of the traffic management department, but also can play the best effect under the current traffic conditions.

[0237] It should be noted that in actual applications, there may be a situation where the selection confidence of multiple flow-limiting strategies is greater than the reference value. At this time, the computer system can further consider other factors (such as implementation cost, impact on surrounding traffic, etc.) to sort or weight these strategies, so as to provide more comprehensive and detailed decision support for the traffic management department. In any case, the core goal of step S633 is to ensure that only those flow-limiting strategies that have been strictly screened and evaluated will be finally adopted and implemented.

[0238] In another embodiment, step S63, determining a flow limiting strategy to be adopted for the target monitoring section from a plurality of flow limiting strategies according to the fourth commonality evaluation coefficient, includes:

[0239] Step S63a: taking the current limiting strategy type corresponding to the largest fourth commonality evaluation coefficient as the target current limiting strategy type.

[0240] In the embodiment of the present invention, step S63a is a key step in determining the flow limiting strategy to be adopted in the second implementation mode. The core of this step is to quickly lock the flow limiting strategy type that best matches the traffic conditions of the target monitored section, that is, the target flow limiting strategy type, by comparing the fourth commonality evaluation coefficients of different flow limiting strategy types.

[0241] First, the computer system reviews the fourth commonality evaluation coefficient of each flow limiting strategy type calculated in step S62. These coefficients are calculated based on the similarity between the coded representation information of the target high-flow road section and the representation vector of the flow limiting strategy type, reflecting the degree of matching between different flow limiting strategy types and the traffic characteristics of the target road section. For example, for the signal light control strategy type, its fourth commonality evaluation coefficient may be 0.85; for the lane closure strategy type, its coefficient may be 0.7; and for the traffic induction strategy type, its coefficient may be 0.8. Next, the computer system sorts these fourth commonality evaluation coefficients and finds the largest one. The flow limiting strategy type corresponding to this largest coefficient is regarded as the most appropriate flow limiting means under the current traffic conditions, that is, the target flow limiting strategy type. This process is automated, and the computer system can complete the comparison and sorting of a large number of coefficients in a very short time.

[0242] Taking the traffic light control strategy type as an example, if its fourth commonality evaluation coefficient is the largest among all flow limiting strategy types, then the computer system will use this strategy type as the target flow limiting strategy type. This means that under the current traffic conditions of the target monitoring section, the traffic light control strategy is considered to be the most likely means to effectively alleviate traffic pressure and improve road traffic efficiency.

[0243] It should be noted that although step S63a quickly determines the target flow limiting strategy type through a simple sorting operation, in actual applications, other factors may need to be further considered to comprehensively evaluate the feasibility of the flow limiting strategy. For example, even if the fourth commonality evaluation coefficient of the signal light control strategy is the highest, if its implementation is difficult or costly, the traffic management department may tend to choose other suboptimal but easier to implement strategies. Therefore, step S63a is usually only a preliminary screening link in the strategy selection and recommendation process, and subsequent steps will conduct more in-depth evaluation and analysis on this basis.

[0244] Step S63b: Arrange the flow limiting strategies of the target flow limiting strategy type in descending order according to the element values ​​of multiple flow limiting analysis elements to obtain an arrangement result, select the first M flow limiting strategies from the arrangement result, and determine them as the flow limiting strategies to be adopted.

[0245] In an embodiment of the present invention, after the target flow limiting strategy type is determined in step S63a, the computer system will further refine the selection process, conduct a comprehensive evaluation of all flow limiting strategies under this type, and sort them according to the element values ​​of multiple flow limiting analysis elements, and finally select the optimal flow limiting strategy combination. First, the computer system collects the element values ​​of multiple flow limiting analysis elements of all flow limiting strategies under the target flow limiting strategy type. These elements may include multiple dimensions such as implementation cost, impact on surrounding traffic, execution difficulty, historical effect evaluation, and each dimension has specific quantitative indicators or grade divisions. For example, for a traffic light control strategy, its implementation cost may be measured by capital investment and manpower requirements, the impact on surrounding traffic may be evaluated through simulation or historical data, and the execution difficulty may be scored based on technical complexity and ease of operation.

[0246] Next, the computer system comprehensively scores all the flow limiting strategies under the target flow limiting strategy type based on the element values ​​of these flow limiting analysis elements. The scoring process may involve a variety of mathematical methods such as weighted summation, analytic hierarchy process (AHP), fuzzy comprehensive evaluation, etc. The specific choice depends on the relative importance of the elements and the complexity of the evaluation. Weighted summation is a common and intuitive method, which first assigns a weight to each element, then multiplies the element value of each flow limiting strategy by the corresponding weight and sums them up to obtain the comprehensive score of the strategy.

[0247] After obtaining the comprehensive scores of all the current limiting strategies, the computer system arranges them in descending order according to the scores and generates a ranking result. This ranking result reflects the advantages and disadvantages of different current limiting strategies after comprehensively considering multiple evaluation factors.

[0248] Finally, the computer system selects the first M flow-limiting strategies from the arrangement results as the flow-limiting strategies to be adopted. The value of M can be set according to actual needs and resource conditions. It represents the number of flow-limiting strategies that the traffic management department is willing to implement or give priority to at the same time. These selected flow-limiting strategies are usually excellent choices at multiple evaluation levels. They can not only effectively cope with the traffic pressure of the target monitoring section, but also achieve a good balance in terms of cost, impact, and difficulty of implementation.

[0249] Taking the traffic light control strategy as an example, suppose that after comprehensive evaluation, three traffic light control schemes have high comprehensive scores and rank in the top three. If M is set to 2, the computer system will recommend these two schemes to the traffic management department as the proposed flow control strategies. These two schemes may be traffic light control strategies for different time periods, different road sections or different traffic conditions. Together, they constitute an effective combination of measures to deal with the current traffic conditions.

[0250] In one embodiment, a computer system is provided. The computer system may be a server, and its internal structure diagram may be as follows: Figure 3 As shown. The computer system includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. Among them, the processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer system is used to provide computing and control capabilities. The memory of the computer system includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer system is used to store fusion perception data. The input / output interface of the computer system is used to exchange information between the processor and an external device. The communication interface of the computer system is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, an intelligent high-speed traffic monitoring and analysis method using fusion perception is implemented.

[0251] Those skilled in the art will understand that Figure 3 The structure shown in the figure is only a block diagram of a part of the structure related to the scheme of the present invention, and does not constitute a limitation on the computer system to which the scheme of the present invention is applied. The specific computer system may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

Claims

1. A smart high-speed traffic monitoring and analysis method using fusion perception, characterized in that: The method comprises: Acquire a high-flow section set of a target monitoring section, wherein the high-flow section set includes a plurality of high-flow sections, the plurality of high-flow sections are arranged in time sequence according to the order in which high-flow occurs, and each high-flow section includes corresponding fusion perception information; Determine a first commonality evaluation coefficient between the last high-flow section in the high-flow section set and the remaining high-flow sections, wherein the first commonality evaluation coefficient is obtained by using a similarity measurement method, wherein the remaining high-flow sections are high-flow sections in the high-flow section set except the last high-flow section; For each of the remaining high-flow road sections, determining a high-flow road section influence weight according to the first commonality evaluation coefficient corresponding to the remaining high-flow road section and the next high-flow road section of the remaining high-flow road section; Determine a target high-flow section according to the high-flow section influence weight corresponding to each of the remaining high-flow sections, wherein the target high-flow section is a section after the last high-flow section where high flow occurs in the target monitoring section; The determining the high-flow section influence weight according to the first commonality evaluation coefficients corresponding to the remaining high-flow sections and the next high-flow section of the remaining high-flow sections comprises: Obtaining the encoding representation information of the next high-flow road section of the remaining high-flow road sections; Determine a first multiplication result of the first commonality evaluation coefficient corresponding to the remaining high-flow road sections and the encoded representation information of the next high-flow road section of the remaining high-flow road sections, and determine it as the high-flow road section influence weight corresponding to the remaining high-flow road sections; The step of determining the target high-flow section according to the high-flow section influence weight corresponding to each of the remaining high-flow sections includes: Determine the sum of the high-flow section influence weights corresponding to the multiple high-flow section respectively, and determine it as the coding representation information of the target high-flow section; A nearest neighbor search operation is performed on the coded representation information of the target high-flow road section to obtain the target high-flow road section.

2. The method according to claim 1, characterized in that: The plurality of high-flow sections in the high-flow section set all have a first impact factor, the first impact factor represents the frequency of high-flow occurrence of the high-flow section in the target monitoring section, and the first impact factor is positively correlated with the frequency; The determining of a first multiplication result of the first commonality evaluation coefficient corresponding to the remaining high-flow road sections and the encoded representation information of the next high-flow road section of the remaining high-flow road sections includes: Determine the multiplication result of the first commonality evaluation coefficient corresponding to the remaining high-flow sections, the coded representation information of the next high-flow section of the remaining high-flow sections and the first influencing factor of the remaining high-flow sections, and determine it as the first multiplication result.

3. The method according to claim 2, characterized in that The step of obtaining a high traffic section set includes: Acquire historical traffic monitoring data of the target monitoring section; Extracting a plurality of initial high-flow road sections where high-flow occurs sequentially from the historical flow monitoring data; Merging the continuously repeated high-flow sections among the multiple initial high-flow sections into one high-flow section, and assigning the first impact factor to the high-flow section obtained after the merger based on the number of the continuously repeated high-flow sections, and assigning the preset first impact factor to the high-flow section that has not been merged; The plurality of high-flow road sections obtained after the merging and the corresponding first influencing factors are used to construct a high-flow road section set.

4. The method according to claim 3, characterized in that The multiple initial high-flow sections where high-flow occurs sequentially are extracted from the historical flow monitoring data, including: Obtaining the average high flow maintenance time of the target monitoring section, performing wandering extraction on the historical flow monitoring data based on the screening controls that satisfy the average high flow maintenance time, and obtaining a plurality of initial high flow sections where high flow occurs sequentially; or; The average of the high-flow sections where high-flow occurs continuously in the target monitoring section is obtained, and the historical flow monitoring data is subjected to wandering extraction based on the filter controls that satisfy the average to obtain a plurality of initial high-flow sections where high-flow occurs sequentially.

5. The method according to claim 1, characterized in that The determining of a first multiplication result of the first commonality evaluation coefficient corresponding to the remaining high-flow road sections and the encoded representation information of the next high-flow road section of the remaining high-flow road sections includes: For each of the remaining high-flow road sections, determine a second commonality evaluation coefficient between the coded characterization information of each high-flow road section and the coded characterization information of the high-flow road sections in the high-flow road section set except for each high-flow road section, perform a standardization operation on the second commonality evaluation coefficient to obtain a standardized commonality evaluation coefficient, determine a multiplication result of the standardized commonality evaluation coefficient and the coded characterization information of the high-flow road section, and determine it as the second influencing factor of the high-flow road section; Determine the multiplication result of the first commonality evaluation coefficient corresponding to the remaining high-flow sections, the coded representation information of the next high-flow section of the remaining high-flow sections, and the second influencing factor of the remaining high-flow sections, and determine it as the first multiplication result.

6. The method according to any one of claims 1 to 5, characterized in that: The determining of the first commonality evaluation coefficient between the last high-flow road segment in the high-flow road segment set and the remaining high-flow road segments respectively comprises: The following process is performed for each of the remaining high-flow road segments: Determine a dot product result of the coded representation information of the last high-flow road section and the coded representation information of the remaining high-flow road sections; Determine a product of a norm of the coded representation information of the last high-flow road section and a norm of the coded representation information of the remaining high-flow road sections; Calculating the ratio of the dot product result to the norm product result, and determining the ratio as the first commonality evaluation coefficient; The method further comprises: Acquire multiple current limiting strategies to be adopted for the target monitoring section; A flow limiting strategy to be adopted for the target monitoring section is determined from among the multiple flow limiting strategies according to the target high-flow section.

7. The method according to claim 6, characterized in that The acquiring of multiple current limiting strategies to be adopted for the target monitoring section includes: Acquiring the road section representation information of the target monitoring road section; Determine a third commonality evaluation coefficient between the road segment characterization information and each of the current limiting strategies in the prior strategy set; Arrange the third commonality evaluation coefficients in order from large to small to obtain an arrangement result, and obtain K of the multiple current limiting strategies that are close to each other in the arrangement result, where K≥1; or; The acquiring of multiple current limiting strategies to be adopted for the target monitoring section includes: Acquire a deep neural network to be trained and a set of learning samples, wherein the set of learning samples includes a plurality of learning samples, each learning sample includes the segment representation information of a monitored segment sample, the segment representation information of an intercepted segment sample, and a supervision mark, wherein the supervision mark is used to indicate the confidence level of the intercepted segment sample in the monitored segment sample having high traffic; Calling the deep neural network to be trained according to the road section representation information of the monitored road section sample and the road section representation information of the intercepted road section sample to obtain an estimation result; Determine a training error according to the supervision mark and the estimation result, and adjust network parameters of the deep neural network to be trained according to the training error to obtain a debugged deep neural network; Calling the debugged deep neural network according to the section representation information of the target monitoring section and the section representation information of the current limiting strategy in the prior strategy set to obtain candidate estimation results corresponding to the multiple current limiting strategies respectively; Based on the candidate estimation results of the current limiting strategy, determining a plurality of current limiting strategies to be adopted for the target monitoring section in the priori strategy set; Each of the high-flow road sections corresponds to a high-flow generating factor; the method of determining the flow limiting strategy to be adopted for the target monitoring road section from the multiple flow limiting strategies according to the target high-flow road section includes: Determine the current limiting strategy types corresponding to the multiple current limiting strategies respectively, encode the current limiting strategy types, and obtain a current limiting strategy type representation vector; Determine a fourth commonality evaluation coefficient between the encoding representation information of the target high-flow road section and the flow-limiting strategy type representation vector; According to the fourth commonality evaluation coefficient, a proposed flow limiting strategy for the target monitoring section is determined from among the multiple flow limiting strategies.

8. The method according to claim 7, characterized in that Determining the proposed current limiting strategy for the target monitoring section from among the multiple current limiting strategies based on the fourth commonality evaluation coefficient includes: Obtaining element values ​​of current limiting analysis elements at multiple evaluation levels of the current limiting strategy, wherein the current limiting analysis elements at multiple evaluation levels include the fourth commonality evaluation coefficient; Calling a screening network to make an estimate based on the current limiting analysis elements at the multiple evaluation levels to obtain a selection confidence of the current limiting strategy; The flow limiting strategy corresponding to the selected confidence level greater than the selected confidence level reference value is used as the flow limiting strategy to be adopted for the target monitoring section; or; Determining the proposed current limiting strategy for the target monitoring section from among the multiple current limiting strategies based on the fourth commonality evaluation coefficient includes: Taking the current limiting strategy type corresponding to the largest fourth commonality evaluation coefficient as the target current limiting strategy type; The flow limiting strategies of the target flow limiting strategy type are arranged in descending order according to the element values ​​of multiple flow limiting analysis elements to obtain an arrangement result, and the first M flow limiting strategies are selected from the arrangement result and determined as the flow limiting strategies to be adopted.

9. A computer system, characterized in that: include: processor; and a memory, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the method as claimed in any one of claims 1 to 8.

Citation Information

Patent Citations

  • City short-time traffic flow prediction method based on traffic flow space-time similarity

    CN108564790A

  • Short-term traffic flow prediction model based on dynamic space-time analysis

    CN113420414A

  • Urban traffic flow reasoning method based on dynamic multi-view graph neural network

    CN114925836A

  • Channel operation state deduction prediction method based on multi-source information deep fusion

    CN117558124A

  • Vehicle traffic flow prediction method with missing data

    US20210407284A1