Methods for identifying hotspot paths between regions in large-scale road networks

Abstract

This application relates to the field of intelligent transportation technology, and in particular to a method for identifying hotspot paths between regions in a large-scale road network. The method includes: dividing the target road network into multiple regions; acquiring license plate recognition data from the previous period on the target road network; extracting single-vehicle origin-destination (OD) trips from the license plate recognition data; and selecting cross-regional single-vehicle OD trips as data to be analyzed; grouping the data to be analyzed, with each group containing cross-regional single-vehicle OD trips within the same preset time interval and originating from the same origin and destination regions; simplifying and calculating the similarity of each group of data to obtain representative paths; aggregating the representative paths of the data to be analyzed to obtain a representative path list; obtaining the frequency of occurrence of each representative path based on the representative path list; and determining inter-regional hotspot paths in the target road network based on the frequency of occurrence. This application can efficiently, accurately, and intelligently identify inter-regional hotspot paths.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of urban intelligent transportation, and in particular to a hot path identification method between regions of large-scale road networks. BACKGROUND

[0002] With the acceleration of urbanization and the continuous rise of the number of motor vehicles, the urban transportation system is facing increasingly severe challenges, and problems such as traffic congestion, environmental pollution and low travel efficiency are becoming increasingly prominent. Accurately grasping the traffic flow dynamics on large-scale road networks, especially identifying hot paths with high frequency and high flow between different functional regions (such as residential areas and business districts, urban core areas and satellite cities, and large transportation hubs and urban hinterlands), is of great significance for optimizing traffic organization, improving road network efficiency, and supporting urban planning and management.

[0003] In recent years, as an important part of intelligent transportation infrastructure, license plate recognition systems have been widely deployed across the country. License plate recognition data, with its wide coverage, fine-grained data, and strong continuity of collection, provides an unprecedented amount of real and fine-grained vehicle trajectory information source for transportation research. Compared with traditional coil detection and floating car data, license plate recognition data can more comprehensively and objectively depict the spatio-temporal movement behavior of vehicles in the road network, especially good at capturing complete path information formed by vehicles crossing different nodes.

[0004] Currently, hot path identification is usually performed by vehicle trajectory association rule analysis or clustering method. However, there are a large number of license plate recognition devices in large-scale urban road networks, generating hundreds of millions or even billions of traffic records daily, and the computational overhead of directly performing full-path enumeration and frequency statistics is huge, which is difficult to meet the timeliness requirements.

[0005] Therefore, there is an urgent need for a hot path identification method between regions of large-scale road networks. SUMMARY

[0006] (I) Technical problems to be solved

[0007] In view of the above-mentioned shortcomings and deficiencies of the prior art, the present application provides a hot path identification method between regions of large-scale road networks, which solves the technical problem of huge computational overhead and difficulty in meeting timeliness requirements in the prior art.

[0008] (II) Technical solutions

[0009] In order to achieve the above-mentioned purposes, the main technical solutions adopted by the present application include:

[0010] The embodiment of the present application provides a hot path identification method between regions of large-scale road networks, comprising:

[0011] S100, divide a target road network into multiple regions, obtain license plate recognition data of a period on the target road network, extract single-vehicle OD trips according to the license plate recognition data, and screen out single-vehicle OD trips crossing regions as to-be-analyzed data;

[0012] S200, group the to-be-analyzed data, each group of data containing single-vehicle OD trips in the same start and end regions and in the same preset time interval, and simplify and calculate the similarity of each group of data to obtain representative paths of the to-be-analyzed data;

[0013] The simplification and similarity calculation of each group of data include: numerical coding processing of each group of data, obtaining similar paths in each group of data as a path candidate set through k-gram feature extraction and a local sensitive hashing algorithm, and processing the path candidate set through similarity calculation and clustering analysis.

[0014] S300, aggregate the representative paths of the to-be-analyzed data to obtain a representative path list, obtain the occurrence frequency of each representative path according to the representative path list, and determine inter-regional hot paths in the target road network according to the occurrence frequency.

[0015] Optionally, in some embodiments of the present application, the S100 includes:

[0016] S110, obtain license plate recognition data of a period on a target road network, the license plate recognition data including a license plate number, a passing intersection ID, and a passing time; group the license plate recognition data according to the license plate number, and sort each group of license plate recognition data according to the passing time to obtain processed each group of license plate recognition data;

[0017] S120, obtain the time difference between two adjacent pieces of license plate recognition data in the processed each group of license plate recognition data, perform trajectory concatenation and disconnection processing according to the time difference, and obtain single-vehicle OD trips;

[0018] The single-vehicle OD trip includes a start intersection ID, an end intersection ID, a passing intersection sequence, a trip start time, and a trip end time.

[0019] S130, associate and match the single-vehicle OD trip with a region-intersection information table to obtain the start region and the end region of each single-vehicle OD trip, and screen out single-vehicle OD trips with different start regions and end regions as to-be-analyzed data.

[0020] Optionally, in some embodiments of the present application, in the S200, the numerical coding processing of each group of data, the obtaining of similar paths in each group of data as a path candidate set through k-gram feature extraction and a local sensitive hashing algorithm include:

[0021] S210. Divide the previous period into N time periods, group the data to be analyzed according to the start and end regions and the time periods, and obtain the grouped data to be analyzed.

[0022] S220. Based on each group of data in the grouped data to be analyzed, construct an intersection mapping dictionary and convert each group of data into a numerical integer sequence.

[0023] S230. The k-gram method is used to extract local features from the numerical integer sequence to obtain the path segment features. Based on the path segment features, a fast retrieval structure is established using the locality-sensitive hashing algorithm to filter out potentially similar bicycle OD trips and add them to the path candidate set.

[0024] Optionally, in some embodiments of this application, S230 includes:

[0025] S231. Based on the numerical integer sequence, a k-gram sliding window segmentation process is performed using a preset k value to obtain a set of path segments consisting of k consecutive intersection nodes;

[0026] S232. Based on the set of path segments, the MinHash algorithm is used to perform signature calculation to obtain a fixed-length MinHash signature vector that can characterize the OD travel characteristics of a single vehicle as the segment feature of the path.

[0027] S233. Based on the MinHash signature vector, perform multi-round hash bucketing using the Locality Sensitive Hash algorithm to obtain a fast retrieval structure that supports fast proximity search; and based on the preset Jaccard similarity threshold and the fast retrieval structure, perform hash collision detection and similarity filtering to obtain single-vehicle OD trips with similarity exceeding the Jaccard similarity threshold and add them to the potential path similarity candidate set.

[0028] Optionally, in some embodiments of this application, in S232,

[0029] When using the MinHash algorithm for signature calculation, a dynamic hash seed strategy is adopted. A dedicated hash seed is assigned to path segments containing high-importance road segments, while a general hash seed is used for path segments that do not contain high-importance road segments to obtain the hash value of each path segment. The hash values ​​generated by different hash seeds are concatenated in the order of the path segments to obtain a fixed-length MinHash signature vector that can characterize the OD travel characteristics of a single vehicle.

[0030] Optionally, in some embodiments of this application, the high-importance road segment is obtained based on preset road network level parameters, specifically including:

[0031] Collect basic parameters of each road segment, standardize the basic parameters of each road segment to obtain standardized basic parameters, perform weighted fusion of the standardized basic parameters to obtain a comprehensive score of the road segment, and determine the road segment with a comprehensive score higher than a preset threshold as a high-importance road segment;

[0032] The basic parameters include road grade, average daily number of vehicles, intersection signal control complexity, and regional functional correlation.

[0033] Optionally, in some embodiments of this application, step S200, processing the path candidate set using similarity calculation and cluster analysis, includes:

[0034] S240. Based on the single-vehicle OD trips in the candidate set of potential similar paths, the longest common subsequence algorithm is used to calculate the path similarity between any two single-vehicle OD trips, and a sparse distance matrix is ​​constructed based on the path similarity.

[0035] S250. Based on the sparse distance matrix, the DBSCAN clustering algorithm is used to perform clustering analysis on the single-vehicle OD trips in the candidate set of potential similar paths, and the category label of each single-vehicle OD trip is obtained.

[0036] S260. Based on the category labels of each bicycle OD trip, count the number of occurrences of different paths in different bicycle OD trips in each category and the number of intersections contained in each path. Based on the number of occurrences of different bicycle OD trips in each category and the number of intersections contained in each trip, obtain the comprehensive path score of each bicycle OD trip in each category.

[0037] S270. Sort the comprehensive path scores of each bicycle OD trip in each category, and select the bicycle OD trip with the highest score in each category as the representative path.

[0038] Optionally, in some embodiments of this application, in step S260, obtaining the comprehensive path score for each vehicle OD trip in each category based on the number of occurrences of different single-vehicle OD trips in each category and the number of intersections included in each trip includes:

[0039] Input the frequency of occurrence of different single-vehicle OD trips in each category and the number of intersections included in each trip into the following formula to obtain the overall path score for each single-vehicle OD trip in each category:

[0040] R = P × lg(K);

[0041] Where R is the overall path score, P is the number of occurrences, and K is the number of intersections.

[0042] Optionally, in some embodiments of this application, in step S250, the neighborhood radius parameter and the minimum sample number parameter of the DBSCAN clustering algorithm are dynamically determined based on the size of the potential path similarity candidate set, specifically including:

[0043] Obtain the average and standard deviation of the path lengths of all single-vehicle origin-destination (OD) trips in the potential path similarity candidate set. Based on the average and standard deviation, dynamically adjust the values ​​of the neighborhood radius parameter and the minimum sample number parameter through a predefined mapping relationship.

[0044] Optionally, in some embodiments of this application, the method further includes:

[0045] S400. Based on the intersection reverse mapping dictionary, convert the inter-region hotspot paths, which are numerical integer sequences, into symbol sequences.

[0046] (III) Beneficial Effects

[0047] The large-scale road network hotspot path identification method provided in this application significantly reduces data storage and processing complexity by converting the intersection ID sequence in vehicle traffic data into a compact numerical sequence. Furthermore, it utilizes k-gram technology to extract path segment features and combines it with a locality-sensitive hashing algorithm to construct an approximate nearest neighbor index structure. This transforms the original global trajectory similarity comparison problem, which required O(n²) calculations, into a near-linear complexity problem requiring calculations only within the candidate pair set of hash collisions, generating a sparse similarity matrix. Simultaneously, a time-sharing processing strategy is employed, slicing the entire day's data hourly to compute local hotspot paths in parallel, and finally aggregating the results. This method effectively avoids the high-overhead processes of precise path inference and full-scale comparison, achieving efficient mining of inter-regional hotspot paths from massive license plate recognition data while ensuring recognition accuracy. This provides feasible technical support for real-time traffic status analysis and decision-making in urban road networks. Attached Figure Description

[0048] Figure 1 This is a flowchart illustrating a method for identifying hotspot paths between large-scale road network areas according to an embodiment of this application. Detailed Implementation

[0049] To better explain and facilitate understanding of this application, the following detailed description of the application is provided in conjunction with the accompanying drawings and specific embodiments.

[0050] In scenarios such as urban traffic management, road network planning, and travel service optimization, the accurate identification of hotspot routes between regions within a large-scale road network is crucial, providing core decision-making support for traffic congestion mitigation, traffic light timing optimization, and public transportation route adjustments. Currently, mainstream hotspot route identification methods in the industry are mostly based on license plate recognition data or GPS trajectory data, mining high-frequency routes from massive amounts of traffic data. However, existing technologies still have significant limitations when facing the specific scenario of "between regions within a large-scale road network," making it difficult to simultaneously meet the requirements of efficiency, accuracy, and practicality.

[0051] Specifically, existing technologies suffer from the following core technical problems: First, large-scale road networks involve massive amounts of data, and existing methods often use full data for direct calculation, resulting in high computational redundancy and low processing efficiency, making it difficult to adapt to real-time or near-real-time analysis needs. Second, hotspot paths have poor stability, and existing methods are mostly based on single-period data output results, without considering the fluctuations in path traffic at different times (such as morning rush hour and off-peak hours), which easily leads to misjudging "temporary high-frequency paths" as long-term hotspots. Third, the algorithms have weak adaptability and are difficult to dynamically adjust according to the characteristics of path data in different regions and at different times, resulting in insufficient robustness of the algorithms in complex road networks.

[0052] To address the aforementioned issues, the large-scale road network regional hotspot path identification method proposed in this application significantly improves processing efficiency and reduces algorithm implementation complexity through a full-process design encompassing targeted data filtering, efficient feature extraction, precise cluster analysis, and all-time aggregation optimization. Simultaneously, while ensuring recognition accuracy, it achieves efficient mining of regional hotspot paths from massive license plate recognition data, providing feasible technical support for real-time traffic status analysis and decision-making in urban road networks.

[0053] To better understand the above technical solutions, exemplary embodiments of this application will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of this application are shown in the drawings, it should be understood that this application can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this application can be understood more clearly and thoroughly, and that the scope of this application can be fully conveyed to those skilled in the art.

[0054] In this embodiment, to clarify the spatial logic and data association of hotspot path identification between regions, this application divides the road network system from macro to micro into the following levels: road network – region – road – road segment – ​​intersection. The definitions and core functions of each level are as follows:

[0055] Road network: A collection of all roads, road sections, intersections and traffic facilities within a predefined geographical area, used to define the overall identification scope;

[0056] Region: A road network sub-unit based on administrative or transportation function;

[0057] Road: A traffic route with its own name;

[0058] Road segment: The smallest continuous unit of a road divided by adjacent intersections;

[0059] Intersection: A node connecting road segments.

[0060] This hierarchical division achieves a closed loop of "scope limitation - regional screening - path planning - data standardization", which is compatible with the scheme of this application.

[0061] Figure 1 This is a flowchart illustrating a method for identifying hotspot paths between large-scale road network areas according to an embodiment of this application. Figure 1 As shown, this method for identifying hotspot paths across large-scale road networks includes:

[0062] Step S100: Divide the target road network into multiple areas, obtain the license plate recognition data of the previous period on the target road network, extract the single vehicle OD trip based on the license plate recognition data, and filter out the cross-regional single vehicle OD trip as the data to be analyzed.

[0063] A single-vehicle origin-destination (OD) trip refers to a complete journey based on a single vehicle.

[0064] Step S100 includes:

[0065] Step S110: Obtain the license plate recognition data of the target road network for the previous period. The license plate recognition data includes the license plate number, the intersection ID, and the time of passage. Group the license plate recognition data according to the license plate number, and sort each group of license plate recognition data according to the time of passage to obtain the processed license plate recognition data for each group.

[0066] The following table shows an example of license plate recognition data in this embodiment:

[0067] License plate recognition data example:

[0068]

[0069] In the specific implementation process, after obtaining the license plate recognition data of the target road network for the previous period, the license plate recognition data needs to be cleaned to remove data with empty license plate numbers or blurred recognition; outlier values ​​of vehicle passing times are filtered out, and duplicate records are deleted. Specifically, this includes:

[0070] For license plate number cleaning, the first step is to remove data with empty license plate numbers, including records with null values, empty strings, or only spaces. Such data cannot be associated with vehicle trajectories due to the lack of core identifiers. The second step is to focus on filtering and identifying fuzzy license plate information, specifically covering three categories: first, fuzzy recognition results containing placeholders, as this data cannot uniquely match vehicles due to missing key characters; second, records with character logic errors, such as ordinary small car license plates being 7 digits and new energy vehicle license plates being 8 digits, if the length does not match, it is judged as a recognition anomaly; third, suspected errors with character confusion, such as the simultaneous appearance of easily confused characters like the number "0" and the letter "O", the number "1" and the letter "I", the number "6" and the letter "b", which cannot pass the license plate rule verification. This type of data needs to be combined with the historical license plate database for fuzzy matching, and if the matching fails, it is removed.

[0071] For outlier filtering of vehicle passage times, checks are conducted from three aspects: time range, format specifications, and logical rationality. Regarding time range, the analysis period is clearly defined, and all records earlier than the start time of the period or later than the end time of the period are removed to avoid cross-period data interference. Regarding format specifications, the vehicle passage times are uniformly required to be in the string format "YYYY-MM-DD HH:MM:SS". Records that do not conform to this format are converted, and those that fail to convert are considered outliers. Regarding logical rationality, the continuity of time is verified in conjunction with the vehicle passage sequence. If, in two adjacent records of the same vehicle, the later passage time is earlier than the earlier one (e.g., "10:30:00" is recorded first, followed by "10:25:00"), this is considered a timestamp inversion anomaly, requiring investigation of equipment clock synchronization issues. Records that cannot be corrected are removed.

[0072] For deleting duplicate records, deduplication rules are established based on the definition of a unique passage event: the core criterion is that "the license plate number, intersection ID, and passage time are completely identical." Such records are usually caused by repeated captures by the device or redundant data transmission, and only one record needs to be retained. For approximately duplicate records with millisecond-level differences, the device deployment location needs to be considered. If they are from the same physical detection point, such as two loop detectors in the same lane, they are merged into one record and the earliest time is retained. If they are from different detection points, such as different lanes at the same intersection, they are considered valid passage records and are not deleted.

[0073] The license plate recognition data, after being cleaned as described above, can more accurately support the extraction of OD (Original Direction) trips for single vehicles across regions. This ensures the scientific validity of subsequent work such as road network traffic prediction based on OD data, cross-regional travel feature mining, and traffic resource allocation, providing real and reliable data for traffic management decisions.

[0074] Step S120: Obtain the time difference between two adjacent license plate recognition data in each group of processed license plate recognition data. Based on the time difference, perform concatenation and disconnection processing of the trajectory to obtain the single vehicle OD journey. The single vehicle OD journey includes: starting intersection ID, ending intersection ID, sequence of intersections passed, start time of journey, and end time of journey.

[0075] The table below shows a sample of single-vehicle OD travel data in this embodiment:

[0076] Example of bicycle OD travel data:

[0077]

[0078] Specifically, step S120 includes:

[0079] Step S121: Calculate the time difference between two adjacent license plate recognition data in each group of license plate recognition data: Convert all vehicle passing times in each group of license plate recognition data from "YYYY-MM-DD HH:MM:SS" to Unix second-level timestamps, calculate the time difference between adjacent records one by one and convert them to minute units with one decimal place, while retaining the original second-level difference. If the sequence contains only one record, mark it as "invalid single-node sequence". If the time difference is less than 0, mark it as "time sequence abnormal" and temporarily store it.

[0080] Step S122: Conduct trajectory concatenation and disconnection based on time difference: Referencing urban travel patterns, the default trajectory concatenation threshold is set to 120 minutes. Starting from the first record of each sequence, the current trajectory starts at the first record. Subsequent records are evaluated sequentially. If a subsequent record is less than the trajectory concatenation threshold, the record is included in the current trajectory and the endpoint node is updated. If a subsequent record is greater than the trajectory concatenation threshold, the current trajectory is saved as a complete travel segment, and the record is used as the starting node of a new trajectory. After traversing all records, the last incomplete trajectory is saved as a complete travel segment.

[0081] Step S123: Perform trajectory spatial topology rationality verification: Load the pre-built "intersection-road network topology relationship table", verify the connectivity of adjacent intersections of each complete journey segment one by one. If the adjacent intersections are marked as "directly connected", the verification passes. If they are "not directly connected", the "spatial jump anomaly" is determined. The trajectory is split at the abnormal position and the type is marked. The split trajectory segments are temporarily stored.

[0082] Step S124: Conduct trajectory time matching rationality verification: Calculate the "theoretical shortest travel time" based on the "physical distance between adjacent intersections" and "maximum speed limit of the road" in the topology table. If the actual travel time is greater than or equal to the theoretical shortest travel time and less than three times the theoretical shortest travel time, the time verification is passed. Otherwise, it is judged as "time abnormal". Split the trajectory at the abnormal location and mark the type. Filter out the trajectory segments that have passed both spatial and temporal verification and contain at least 2 intersection nodes. Remove invalid segments with single nodes or double abnormalities.

[0083] Step S125: For valid trajectory segments, extract core fields such as "starting intersection ID, ending intersection ID, sequence of intersections passed, start time of trip, and end time of trip", and supplement derived fields such as "trip duration, number of intersections passed, and trajectory verification result", and finally output the structured valid single-vehicle OD trip.

[0084] Furthermore, a multi-device recognition verification is added at the same time. If the same license plate is recognized by two or more devices at non-adjacent intersections in the same second, it is determined to be a device mis-capture or a license plate clone abnormality, the record is marked and the abnormal segment is identified.

[0085] Step S120 above ensures the accuracy of single-vehicle trajectory segmentation through precise time difference calculation, avoiding both erroneous merging of different trips and erroneous segmentation of the same trip, thus guaranteeing the continuity and independence of the trajectory. Combined with dual verification of spatial topology and time matching, it effectively filters out false and abnormal trajectories such as intersection jumps and unreasonable timing, ensuring that the extracted trajectory truly reflects the actual travel situation of vehicles. The final output structured OD trip data not only includes core start and end information but also supplements derived fields such as trip duration and route nodes, providing standardized and high-quality basic data for subsequent cross-regional trip screening and road network travel characteristic analysis. This lays a reliable foundation for subsequent work such as traffic flow statistics and road network planning, improving the scientific nature of traffic analysis and decision-making.

[0086] Step S130: Match the single vehicle OD trip with the area-intersection information table to obtain the start and end areas of each single vehicle OD trip, and filter out single vehicle OD trips with different start and end areas as data to be analyzed.

[0087] The area-intersection information table mainly includes intersection ID, intersection coordinates (latitude and longitude), intersection name, and the code and name of the corresponding area, such as administrative division code and district name, functional zone identifier and zone name. Some additional information such as intersection level can be added. In terms of acquisition methods, official datasets can be downloaded from the public data open network, and map APIs such as Gaode and Baidu can be used to extract intersection and area association data in batches. Alternatively, official association tables can be obtained from local authorities, or data can be extracted through open geographic data such as OpenStreetMap. After acquisition, coordinate unification, duplicate record removal, and consistency verification of area code and name are required to ensure that the data is accurate and usable.

[0088] The following table shows a sample of the area-intersection information table data in this embodiment:

[0089] Example of intersection information table data:

[0090]

[0091] Step S200: Group the data to be analyzed, with each group containing the same start and end areas and cross-regional single-vehicle OD trips within the same preset time interval; and simplify and calculate the similarity of each group of data to obtain the representative path of the data to be analyzed.

[0092] The sample data to be analyzed in this embodiment is shown in the table below:

[0093] Example of data to be analyzed:

[0094]

[0095] The simplification and similarity calculation for each set of data includes: numerical encoding of each set of data; obtaining similar paths in each set of data as path candidate sets through k-gram feature extraction and locality-sensitive hashing algorithm; and processing the path candidate sets using similarity calculation and cluster analysis.

[0096] In step S200, each set of data undergoes numerical encoding processing. Similar paths in each set of data are obtained as a path candidate set through k-gram feature extraction and locality-sensitive hashing, including:

[0097] Step S210: Divide the previous period into N time periods, group the data to be analyzed according to the start and end regions and the time periods they belong to, and obtain the grouped data to be analyzed.

[0098] In this embodiment, the cycle duration is determined based on the analysis objective. If focusing on daily commutes, it can be set to one calendar day; if analyzing weekly patterns, it can be set to seven days; and if comparing holidays and weekdays, it can be flexibly combined. The time period division adopts a "basic granularity + peak-hour intensification" strategy. By default, it uses a 1-hour basic granularity to divide into N time periods, with peak traffic hours (morning and evening rush hours) refined to a 30-minute granularity. Each time period is assigned a unique identifier to ensure clear and identifiable time period boundaries. Subsequently, the cycle is finely segmented according to the set rules, with the start and end times of each time period accurate to the second. For cross-day cycles, date information is embedded in the identifier, and a "time period division table" containing the time period identifier, start and end times, and granularity type is generated as the basis for data attribution.

[0099] Next, the time period to which each OD trip belongs is determined based on its "trip start time" in the data to be analyzed: the trip start time is matched with the corresponding time period interval in the "time period division table", and boundary times are assigned to the current time period, while "time period abnormal data" outside the period range are filtered out. Then, double grouping is performed according to the combination of "starting region - ending region", i.e., "region pair" and "time period": the "starting region + ending region" is combined into a region pair identifier, and data aggregation tools are used to group according to "region pair identifier + time period identifier" to form a dataset of "same region pair + same time period", and "small sample groups" containing 3 or fewer trips and empty groups are removed. Finally, the completeness of the grouping results is verified—the total number of trips in all groups is counted, and the error with the total number of trips in the valid time period must be ≤5%. Each group is named "region pair identifier_time period identifier", and the group identifier, region pair, time period, list of included OD trips, and number of trips are stored in a structured manner, laying the foundation for subsequent path feature extraction.

[0100] Step S220: Based on each group of data in the grouped data to be analyzed, construct an intersection mapping dictionary and convert each group of data into a numerical integer sequence;

[0101] In the specific implementation process, for each group of data to be analyzed after grouping, the "intersection sequence passed" of each trip within the group is first traversed, extracting all appearing intersection IDs, and then deduplicating them to obtain a unique list of intersections, ensuring that each intersection appears only once within the group. Subsequently, a unique "intersection mapping dictionary" is constructed based on this intersection list: each intersection ID in the list is assigned a unique non-negative integer as an index, the dictionary keys are the original intersection IDs, and the values ​​are the corresponding integer indices. The mapping relationship within the same group remains absolutely consistent; that is, regardless of how many trips a certain intersection appears in, its corresponding integer index remains fixed, avoiding sequence distortion due to mapping fluctuations. After the dictionary construction is completed, the "intersection sequence passed" of each OD trip within the group is transformed one by one: according to the order of the intersections in the trip, each intersection ID is replaced with the corresponding integer index in the dictionary, forming an ordered sequence of integers.

[0102] In addition, verification is performed simultaneously during the conversion process: if the intersection ID of a certain trip does not appear in the intersection set within the group, it is marked as an "abnormal sequence" and will not be included in subsequent processing; if the length of the converted sequence is 0 or 1, that is, there are no valid intersection nodes, it will also be judged as an invalid sequence and removed.

[0103] By constructing a mapping dictionary independently for each group, we can avoid confusion caused by duplicate intersection IDs between different groups, compress data dimensions, and ensure that the converted integer sequence retains the temporal relationship and path structure of the original intersection sequence. This provides a unified and computable basic data format for operations such as k-gram local feature extraction and hash calculation in step S230.

[0104] Step S230: Use the k-gram method to extract local features from the numerical integer sequence to obtain the path segment features; based on the path segment features, use the locality-sensitive hashing algorithm to establish a fast retrieval structure and filter out potentially similar bicycle OD trips to add to the path candidate set.

[0105] Step S230 includes:

[0106] Step S231: Based on the numerical integer sequence, perform k-gram sliding window segmentation using a preset k value to obtain a set of path segments consisting of k consecutive intersection nodes.

[0107] Specifically, step S231 includes:

[0108] Determine the value of k, which represents the number of consecutive intersection nodes in each path segment. The setting of k needs to be combined with the average length of the paths in the group and the complexity of the road network topology. If the paths in the group are mostly short distances, the value of k is set to 3 to capture local turning features; if the paths are mostly long distances, the value of k is set to 5 to retain the continuity of longer road segments; for path groups of mixed lengths, a dynamic k value strategy is adopted to avoid the over-segmentation of short paths or the loss of details of long paths due to a fixed k value.

[0109] Specifically, the number of nodes in all paths within a group is counted. The 25th and 75th percentile values ​​are used as interval boundaries to classify paths into three categories: short paths, medium paths, and long paths. Short paths correspond to k=3, medium paths to k=4, and long paths to k=5. For extremely short paths with fewer than k (e.g., a 3-node path corresponding to k=3), the number of path nodes is directly used as the k value. For very long paths (e.g., more than 20 nodes), the upper limit of k is set to 6.

[0110] By using the above method of dynamically allocating k values, short paths can be prevented from being over-segmented by k=5, while long paths can retain continuous road segment features by using k=5. Ultimately, this makes the feature extraction of path segments of different lengths more suitable for their own structure.

[0111] After determining the value of k, a sliding window segmentation is performed on each numerical integer sequence: using k consecutive integers in the sequence as a window, starting from the beginning of the sequence, the window slides sequentially with a step size of 1 until the end of the window touches the end of the sequence. For example, for an integer sequence [0, 1, 2, 3, 4], when k=3, the set of path segments obtained after sliding window segmentation is [0, 1, 2], [1, 2, 3], and [2, 3, 4]. Each segment retains the continuous temporal relationship of the intersection nodes in the original sequence, and adjacent segments overlap through k-1 nodes (such as [0, 1, 2] and [1, 2, 3] sharing nodes 1 and 2 in the previous example), ensuring the continuity of local features.

[0112] Step S232: Based on the set of path segments, use the MinHash algorithm to perform signature calculation and obtain a fixed-length MinHash signature vector that can characterize the OD travel characteristics of a single vehicle as the segment feature of the path.

[0113] When using the MinHash algorithm for signature calculation, a dynamic hash seed strategy is adopted. A dedicated hash seed is assigned to path segments containing high-importance road segments, while a general hash seed is used for path segments that do not contain high-importance road segments to obtain the hash value of each path segment. The hash values ​​generated by different hash seeds are concatenated in the order of the path segments to obtain a fixed-length MinHash signature vector that can characterize the OD travel characteristics of a single vehicle.

[0114] High-importance road sections are obtained based on preset road network level parameters, specifically including:

[0115] Collect basic parameters for each road segment, standardize the basic parameters for each road segment, obtain the standardized basic parameters, perform weighted fusion of the standardized basic parameters, obtain the comprehensive score of the road segment, and determine the road segment with a comprehensive score higher than the preset threshold as a high-importance road segment;

[0116] The basic parameters include road grade, average daily number of vehicles, intersection signal control complexity, and regional functional correlation.

[0117] Step S232 uses the MinHash algorithm to compress the high-dimensional path fragment set into a fixed-length signature vector. This significantly reduces the data dimensionality to reduce subsequent computing power consumption and improve processing efficiency. It also uses a dynamic hash seed strategy to accurately strengthen the key path features and avoid diluting core details. At the same time, the fixed-length vector provides a unified comparison benchmark for different paths, solving the "incomparability" problem caused by different numbers of original fragments. It also allows the signature vectors of similar paths to maintain a high degree of overlap, laying a reliable foundation for subsequent local sensitive hash retrieval and similar path screening, and reducing the probability of false detection and false negative detection.

[0118] Step S233: Based on the MinHash signature vector, perform multi-round hash bucketing using the Locality Sensitive Hash algorithm to obtain a fast retrieval structure that supports fast proximity search; and based on the preset Jaccard similarity threshold and the fast retrieval structure, perform hash collision detection and similarity filtering to obtain bicycle OD trips with similarity exceeding the Jaccard similarity threshold and add them to the potential path similarity candidate set.

[0119] In the specific implementation process, firstly, a multi-round locality-sensitive hashing (LSH) bucketing strategy is designed to construct a fast retrieval structure. Based on the fixed-length MinHash signature vector generated in step S232, it is evenly divided into M continuous sub-vectors, with each round of bucketing using an independent set of hash functions. For the signature vector of each path, sub-vectors are extracted according to the round and the corresponding hash function is applied to calculate the hash value of each round. The path ID is then mapped to the corresponding bucket for that round. For example, if the signature vector of a path is [3, 5, 2, 7, 9, 1, ...], in the first round, the first 4 dimensions [3, 5, 2, 7] are taken and the hash value is calculated to be 12, which is assigned to bucket 12 in the first round; in the second round, the next 4 dimensions are taken and the hash value is calculated to be 8, which is assigned to bucket 8 in the second round, and so on. After multiple rounds of bucketing, a multi-layer retrieval structure of "round - bucket ID - path ID list" is constructed, where each bucket stores all path IDs mapped to that bucket, ensuring that potential similar paths can be quickly located through the bucket index later.

[0120] Secondly, candidate path pairs are identified through hash collision detection. For each path within a group, all buckets it belongs to in the M rounds of bucketing are traversed, and the IDs of other paths within these buckets are collected—that is, paths that have "collided" with the current path in at least one round of bucketing. After deduplication, the "initial candidate set" for that path is formed. For example, if path A is assigned to bucket 12, bucket 8, and bucket 5 in 3 rounds of bucketing, then all other paths in buckets 12, 8, and 5 are included in the initial candidate set of A. This process quickly narrows down the potential similarity range through the "shared bucket" feature, avoiding pairwise comparisons of all paths within the group and significantly improving retrieval efficiency.

[0121] Next, precise filtering is performed based on the Jaccard similarity threshold. For each path in the initial candidate set, its actual Jaccard similarity with the current path is calculated. This is achieved by calculating the ratio of the number of intersection fragments to the number of union fragments from the original path fragment sets of the two paths. The calculated result is compared with a preset threshold: if it is greater than or equal to the threshold, it is determined to be a "highly similar path," and both paths are added to each other's path candidate sets; if it is less than or equal to the threshold, it is removed from the candidate set to avoid interference from false similarity items.

[0122] Finally, the candidate path sets are optimized: duplicate similarity relationships are merged, such as A→B and B→A are considered as the same pair, candidate sets containing only themselves are filtered, and a "similarity score" field is added to each candidate set to facilitate prioritizing highly similar path pairs during subsequent cluster analysis.

[0123] This combined strategy of "rapidly narrowing the scope through multiple rounds of binning + accurate verification of actual similarity" ensures both a high recall rate for similar paths and the accuracy of the candidate set through Jaccard threshold control, providing a high-quality foundation for the clustering analysis in step S200.

[0124] In step S200, the path candidate set is processed using similarity calculation and cluster analysis, including:

[0125] Step S240: Based on the single-vehicle OD trips in the candidate set of potential similar paths, the longest common subsequence algorithm is used to calculate the path similarity between any two single-vehicle OD trips, and a sparse distance matrix is ​​constructed based on the path similarity.

[0126] Specifically, step S240 includes: first, calculating the length of the longest common subsequence of the two paths, that is, the length of the longest subsequence of nodes appearing in the same order in the two sequences. For example, if one path is [0, 1, 2, 3, 4] and the other is [0, 2, 3, 5], their longest common subsequence is [0, 2, 3], with a length of 3. Then, this length is converted into a similarity value between 0 and 1. The calculation method is to divide the length of the longest common subsequence by the length of the longer of the two paths, thus avoiding similarity distortion caused by excessive differences in path lengths. In the example above, the two paths are 5 and 4 in length, respectively, so the similarity is 3 divided by 5, resulting in 0.6.

[0127] The specific calculation formula is as follows:

[0128] S(i,j)=M ij / Max(M i M j );

[0129] Where S(i,j) is the similarity between single-vehicle OD trips i and j, and M ijM represents the number of intersections to be matched. i M represents the total number of intersections included in trip i. j Let M be the total number of intersections included in trip j, and Max(M) be the number of intersections included in trip j. i M j ) is M i and M j The maximum value in.

[0130] Based on this, a sparse distance matrix for clustering is constructed. The core principle is to retain key distance information and compress its storage. The distance value is obtained by subtracting the similarity from 1. The smaller the distance, the more similar the two paths are. For example, a similarity of 0.6 corresponds to a distance of 0.4. Then, sparsification is performed: a distance threshold is set, and only elements with a distance less than or equal to the threshold are retained; path pairs with a distance exceeding the threshold are considered "dissimilar paths" and their distance values ​​are not stored in the matrix, thereby reducing the matrix storage space.

[0131] The matrix is ​​stored in a triplet format of "path ID-path ID-distance value" or a sparse matrix data structure. The row and column indices correspond to the path IDs, and the non-zero elements are the distance values. Finally, the matrix needs to be verified: 10% of the path pairs are randomly selected and the original intersection sequence is manually compared with the calculated distance values ​​to verify the rationality of the algorithm. For example, two paths that obviously share multiple consecutive intersections should have a small distance. At the same time, the symmetry of the matrix is ​​checked to ensure that there are no calculation logic errors.

[0132] This process not only allows for the precise capture of the ordered similarity features of the path in the node sequence using the longest common subsequence algorithm, which better reflects the correlation of the actual path direction compared to simple node intersection calculation, but also balances computational efficiency and data accuracy through sparsification. The final output sparse distance matrix provides a reliable distance metric basis for subsequent DBSCAN clustering.

[0133] Step S250: Based on the sparse distance matrix, the DBSCAN clustering algorithm is used to perform clustering analysis on the single vehicle OD trips in the candidate set of potential similar paths, and obtain the category label of each single vehicle OD trip.

[0134] The following table shows an example of the single-vehicle OD (Original Distance) travel classification label in this embodiment:

[0135] Example of bicycle OD (Original Distance) travel category label:

[0136]

[0137] For category 1, the paths 0;1;2;3;4;5 appear twice, with a comprehensive score of 2×lg6=1.56. The paths 2;3;4;5 appear once, with a comprehensive score of 1×lg4=0.6. Therefore, 0;1;2;3;4;5 are selected as representative paths.

[0138] In step S250, the neighborhood radius parameter and minimum sample number parameter of the DBSCAN clustering algorithm are dynamically determined based on the size of the candidate set of potential similar paths, specifically including:

[0139] The mean and standard deviation of the path lengths of all single-vehicle origin-destination (OD) trips in the candidate set of potential similar paths are obtained. Based on the mean and standard deviation, the values ​​of the neighborhood radius parameter and the minimum sample number parameter are dynamically adjusted through a predefined mapping relationship.

[0140] Step S260: Based on the category labels of each bicycle OD trip, count the number of occurrences of different paths in different bicycle OD trips in each category and the number of intersections contained in each path. Based on the number of occurrences of different bicycle OD trips in each category and the number of intersections contained in each trip, obtain the comprehensive path score of each bicycle OD trip in each category.

[0141] In step S260, based on the frequency of occurrence of different single-vehicle OD trips in each category and the number of intersections included in each trip, the comprehensive path score for each single-vehicle OD trip in each category is obtained, including:

[0142] Input the frequency of occurrence of different single-vehicle OD trips in each category and the number of intersections included in each trip into the following formula to obtain the overall path score for each single-vehicle OD trip in each category:

[0143] R = P × lg(K);

[0144] Where R is the overall path score, P is the number of occurrences, and K is the number of intersections.

[0145] Step S270: Sort the overall path scores of each bicycle OD trip in each category, and select the bicycle OD trip with the highest score in each category as the representative path.

[0146] Step S300: Aggregate representative paths of the data to be analyzed to obtain a list of representative paths; based on the list of representative paths, obtain the frequency of occurrence of each representative path, and determine the inter-regional hotspot paths in the target road network based on the frequency of occurrence.

[0147] The following table shows examples of hotspot paths between regions in this embodiment:

[0148] Example of hotspot path between regions:

[0149]

[0150] The large-scale road network inter-regional hotspot path identification method in this embodiment also includes:

[0151] Step S400: Based on the intersection reverse mapping dictionary, convert the inter-region hotspot paths, which are numerical integer sequences, into symbol sequences.

[0152] This embodiment presents a method for identifying hotspot paths between regions in a large-scale road network. It transforms symbolic sequences into numerical sequences and combines k-gram feature extraction to capture key path segments, avoiding the complex and time-consuming process of reconstructing complete vehicle paths in traditional methods, thus reducing data storage and processing complexity. Simultaneously, through locality-sensitive hash indexing and sparse similarity calculation, the computational complexity of OD trip similarity comparison is reduced from O(n²) quadratic level in traditional methods to near O(n) linear level, significantly improving the processing speed of massive data and solving the efficiency dilemma of full data enumeration. Furthermore, this implementation is naturally compatible with distributed frameworks such as Spark / Hadoop, supporting the horizontal expansion of ultra-large-scale road networks and meeting the needs of real-time analysis at the city level. At the same time, it ensures the global consistency of the final results through aggregation strategies. While ensuring recognition accuracy, it achieves efficient mining of hotspot paths between regions under massive license plate recognition data, providing feasible technical support for real-time traffic status analysis and decision-making in city-level road networks.

[0153] In the description of this application, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0154] In this application, unless otherwise expressly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this application according to the specific circumstances.

[0155] In this application, unless otherwise expressly specified and limited, "above" or "below" the second feature can mean that the first and second features are in direct contact, or that they are in indirect contact through an intermediate medium. Furthermore, "above," "on top of," and "over" the second feature can mean that the first feature is directly above or diagonally above the second feature, or simply that the first feature is at a higher horizontal level than the second feature. "Below," "below," and "under" the second feature can mean that the first feature is directly below or diagonally below the second feature, or simply that the first feature is at a lower horizontal level than the second feature.

[0156] In the description of this specification, the terms "one embodiment," "some embodiments," "embodiment," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0157] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make modifications, alterations, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A large-scale road network inter-regional hotspot path identification method, characterized in that, The method comprises the following steps: S100, dividing a target road network into multiple regions, obtaining license plate recognition data of the target road network in a period, extracting single-vehicle OD trips from the license plate recognition data, and screening single-vehicle OD trips across regions as to-be-analyzed data; The step of extracting single-vehicle OD trips from the license plate recognition data comprises the following steps: calculating a time difference between adjacent license plate recognition data, performing trajectory concatenation and disconnection processing based on a preset trajectory concatenation threshold, and performing spatial topology rationality and time matching rationality checking by using road network topology relationship to output structured single-vehicle OD trips; S200, grouping the to-be-analyzed data, each group of data containing single-vehicle OD trips across regions in the same start and end regions and in the same preset time interval; and simplifying and calculating the similarity of each group of data to obtain a representative path of the to-be-analyzed data; The step of simplifying and calculating the similarity of each group of data comprises the following steps: numerically encoding each group of data, obtaining similar paths in each group of data as a path candidate set by using a k-gram feature extraction and a local sensitive hashing algorithm, and processing the path candidate set by using a similarity calculation and clustering analysis; When the k-gram feature extraction and the local sensitive hashing algorithm are used to obtain similar paths in each group of data as a path candidate set, a k-gram method is used for local feature extraction, a MinHash algorithm is used for signature calculation, and a local sensitive hashing algorithm is used for multi-round hashing bucket processing to establish a fast retrieval structure, and potential similar single-vehicle OD trips are screened out to join the path candidate set; When the MinHash algorithm is used for signature calculation, a dynamic hashing seed strategy is used, a dedicated hashing seed is assigned to a path segment containing a high importance road segment, and a general hashing seed is used for a path segment not containing a high importance road segment; S300, aggregating the representative paths of the to-be-analyzed data to obtain a representative path list, obtaining the occurrence frequency of each representative path according to the representative path list, and determining an inter-regional hot path in the target road network according to the occurrence frequency.

2. The method of claim 1, wherein, The S100 comprises: S110, obtaining license plate recognition data of a target road network in a period, the license plate recognition data comprising a license plate number, a passing intersection ID, and a passing time; grouping the license plate recognition data according to the license plate number, and sorting each group of license plate recognition data according to the passing time to obtain processed each group of license plate recognition data; S120, obtaining a time difference between adjacent two pieces of license plate recognition data in the processed each group of license plate recognition data, performing trajectory concatenation and disconnection processing according to the time difference, and obtaining single-vehicle OD trips; The single-vehicle OD trips comprise a start intersection ID, an end intersection ID, a passing intersection sequence, a trip start time, and a trip end time; S130, associating and matching the single-vehicle OD trips with a region-intersection information table to obtain a start region and an end region of each single-vehicle OD trip, and screening single-vehicle OD trips with different start regions and end regions as to-be-analyzed data.

3. The method of claim 1, wherein, The S200 includes: S210, dividing the last period into N time periods, grouping the to-be-analyzed data according to start and end areas and time periods, and obtaining the grouped to-be-analyzed data; S220, constructing a junction mapping dictionary based on each group of data in the grouped to-be-analyzed data, and converting each group of data into a numerical integer sequence; S230, performing local feature extraction on the numerical integer sequence by using a k-gram method, obtaining a fragment feature of a path, and establishing a fast retrieval structure by using a local sensitive hashing algorithm based on the fragment feature of the path, and screening out potential similar bicycle OD trips to join a path candidate set.

4. The method of claim 3, wherein, The S230 includes: S231, performing k-gram sliding window segmentation processing on the numerical integer sequence by using a preset k value, and obtaining a path fragment set composed of consecutive k junction nodes; S232, performing signature calculation on the path fragment set by using a MinHash algorithm, and obtaining a fixed-length MinHash signature vector representing the characteristics of a bicycle OD trip as a fragment feature of a path; S233, performing multi-round hash bucket processing on the MinHash signature vector by using a local sensitive hashing algorithm, obtaining a fast retrieval structure supporting fast neighbor search, and performing hash collision detection and similarity screening processing according to a preset Jaccard similarity threshold and the fast retrieval structure, obtaining bicycle OD trips with a similarity exceeding the Jaccard similarity threshold and adding them to the path candidate set of potential path similarity.

5. The method of claim 4, wherein, In the S232, When performing signature calculation by using a MinHash algorithm, a dynamic hash seed strategy is used to assign a dedicated hash seed to a path fragment containing a high importance section, and a general hash seed is used to a path fragment not containing a high importance section, to obtain hash values of each path fragment; the hash values generated by different hash seeds are spliced in the order of path fragments to obtain a fixed-length MinHash signature vector representing the characteristics of a bicycle OD trip.

6. The method of claim 5, wherein, The high importance section is obtained based on a preset road network level parameter, and specifically includes: Collecting basic parameters of each section, performing standardization processing on the basic parameters of each section to obtain standardized basic parameters, performing weighted fusion on the standardized basic parameters to obtain a section comprehensive score, and determining a section with a section comprehensive score higher than a preset threshold as a high importance section; The basic parameters include road grade, daily average vehicle number, intersection signal control complexity, and regional function correlation.

7. The method of claim 3, wherein, In the S200, the path candidate set is processed by using similarity calculation and clustering analysis, including: S240, calculating the path similarity of any two bicycle OD trips by using a longest common subsequence algorithm based on the bicycle OD trips in the path candidate set of potential path similarity, and constructing a sparse distance matrix according to the path similarity; S250, according to the sparse distance matrix, using DBSCAN clustering algorithm to cluster analysis of the bicycle OD trip in the path candidate set similar to the potential path, obtain the class label of each bicycle OD trip; S260, according to the class label of each bicycle OD trip, the number of different paths in different bicycle OD trips in each class and the number of intersections contained in each path are counted, and the path comprehensive score of each bicycle OD trip in each class is obtained according to the number of different bicycle OD trips in each class and the number of intersections contained in each trip; S270, sort the path comprehensive score of each bicycle OD trip in each class, and select the highest score bicycle OD trip in each class as the representative path.

8. The method of claim 7, wherein, In the S260, according to the number of different bicycle OD trips in each class and the number of intersections contained in each trip, the path comprehensive score of each bicycle OD trip in each class is obtained, including: the number of different bicycle OD trips in each class and the number of intersections contained in each trip are input into the following formula to obtain the path comprehensive score of each bicycle OD trip in each class: R=P×lg(K); wherein R is the path comprehensive score, P is the number of occurrences, and K is the number of intersections.

9. The method of claim 7, wherein, In the S250, the neighborhood radius parameter and the minimum sample number parameter of the DBSCAN clustering algorithm are dynamically determined according to the size of the path candidate set similar to the potential path, and specifically include: obtaining the average value and standard deviation of the path length of all bicycle OD trips in the path candidate set similar to the potential path, and dynamically adjusting the value of the neighborhood radius parameter and the minimum sample number parameter through the pre-defined mapping relationship based on the average value and standard deviation.

10. The method of claim 1, wherein, The method further comprises: S400, according to the intersection reverse mapping dictionary, convert the inter-regional hot path as a numerical integer sequence into a symbol sequence.

Images