Speed anomaly detection and fraud identification method and system based on trajectory data

Abstract

The invention aims to provide a speed anomaly detection and fraud identification method and system based on trajectory data, and belongs to the technical field of road traffic safety, the method realizes abnormal trajectory detection and fraud identification through multi-stage data processing: firstly, preprocessing original trajectory data, and removing invalid data; noise points are filtered based on a DBSCAN algorithm; track segments are divided according to vehicle speed changes; matching the moving track segment to a map road through a map matching algorithm; calculating an error ratio between the calculation speed and the equipment uploading speed, and identifying an abnormal track segment; and finally, analyzing the speed distribution of the abnormal track section, and identifying a counterfeit behavior. The system comprises a data preprocessing module, a trajectory noise filtering module, a trajectory division module, a map matching module, an abnormal speed detection module and a speed verification and forgery identification module. According to the method, the accuracy of speed anomaly detection is improved, the recognition capability of a hidden speed forgery behavior is enhanced, and the authenticity and credibility of trajectory data are improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent transportation data processing technology, and in particular to a method and system for speed anomaly detection and fraud identification based on trajectory data. Background Technology

[0002] With the widespread adoption of BeiDou / GPS positioning technology, trajectory data has been extensively used in vehicle management, transportation supervision, and driving behavior analysis. However, in practical applications, some drivers or transportation companies upload falsified trajectories or tamper with speed information to evade supervision or optimize records. Traditional trajectory data anomaly detection methods largely rely on empirical rules or simple statistical analysis, which are insufficient to address sophisticated forgery and data fraud, and their accuracy is limited. Therefore, an efficient detection method is needed to improve the credibility of trajectory data and the level of intelligent supervision. Summary of the Invention

[0003] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method and system for speed anomaly detection and fraud identification based on trajectory data. This method integrates multi-stage processing through preprocessing, DBSCAN noise reduction, trajectory segmentation, and HMM map matching, and combines dynamic threshold and speed distribution analysis to identify anomalies and forgeries. The system utilizes adaptive parameter clustering, Viterbi algorithm matching, and upper quartile threshold to achieve high-precision anomaly detection of trajectory data.

[0004] This solution improves the accuracy of speed anomaly detection and the ability to identify concealed forgery, providing reliable trajectory data authenticity assurance for fields such as intelligent transportation and enhancing the level of intelligent supervision. It effectively improves the detection rate and accuracy of speed fraud, providing highly credible data support for intelligent transportation supervision.

[0005] To achieve the above objectives, in a first aspect, the present invention provides a method for speed anomaly detection and fraud identification based on trajectory data, comprising the following steps:

[0006] Step S1: Obtain the original trajectory data of road vehicles. The original trajectory data includes multiple trajectory points. The original trajectory data is preprocessed to remove invalid data and refresh abnormal data to obtain a preprocessed trajectory point sequence. The original trajectory data includes the latitude and longitude coordinates of the trajectory points, the timestamp of the trajectory points, the device upload speed corresponding to the trajectory points, and the direction angle corresponding to the trajectory points, etc.

[0007] Step S2: Perform trajectory noise filtering on the preprocessed trajectory point sequence using the density-based spatial clustering (DBSCAN) algorithm and the spherical distance between trajectory points calculated using the spherical semi-versus formula. Identify and remove noise points that deviate from the trajectory sequence to obtain the denoised trajectory point sequence.

[0008] Further, in step S2, the spherical distance d is calculated using the following formula:

[0009]

[0010] Where r is the Earth's radius; φ1 and φ2 are the latitudes of the two trajectory points, respectively; λ1 and λ2 are the longitudes of the two trajectory points, respectively; and the second trajectory point is a trajectory point within the domain of the first trajectory point.

[0011] Step S3: Based on the vehicle speed change reflected by the device upload speed corresponding to the trajectory point, the denoised trajectory point sequence is divided into trajectories. The trajectory segments with continuous speed of 0 and dwell time exceeding the preset time threshold are determined as dwell trajectory segments, and the remaining trajectory segments are determined as running trajectory segments.

[0012] Step S4: For the running trajectory segment, a map matching algorithm is used to match the trajectory points in the running trajectory segment to the roads on the map to obtain the map matching path;

[0013] Step S5: Calculate the estimated speed based on the map matching path and the timestamp of the trajectory point, compare the estimated speed with the device upload speed corresponding to the trajectory point, calculate the speed error ratio, and if the speed error ratio exceeds the dynamic statistical threshold, mark it as an abnormal trajectory segment.

[0014] Step S6: Perform speed verification and forgery identification on the abnormal trajectory segment. If the amount of data on the overspeed in the speed distribution of the abnormal trajectory segment is zero, and the speed shows a concentrated situation of exceeding 50% at the preset overspeed threshold, then it is determined to be a forgery.

[0015] Furthermore, the preprocessing includes: removing null data from the original trajectory point sequence, deleting periodically abnormal data, and refreshing data with physical limits. Periodically abnormal data refers to data whose sampling interval exceeds the set sampling frequency tolerance range. In actual trajectory acquisition, the device sampling frequency is generally once every 30 seconds. Considering network transmission jitter and device fluctuations, a tolerance margin of ±10 seconds is allowed. That is, if the time interval between two consecutive trajectory points exceeds 40 seconds or is less than 20 seconds, the point is determined to be refresh periodically abnormal data and is removed.

[0016] Further, in step S2, the neighborhood radius ∈ and the minimum number of neighborhood points minPts of the DBSCAN algorithm are adaptively adjusted according to the sampling frequency of the trajectory point sequence. In actual trajectory acquisition, the vehicle's GPS device typically uploads one trajectory point every 30 seconds, with a sampling frequency of 1 / 30Hz. In this invention, the theoretical maximum distance between trajectory points is calculated based on the trajectory point sampling interval and the vehicle's maximum driving speed, and the clustering neighborhood radius ∈ of DBSCAN is set accordingly. For example, if the vehicle travels at 60km / h within 30 seconds, the maximum theoretical distance is 500 meters, therefore the clustering radius ∈ is set to 500 meters, and minPts is set to 3 to ensure effective trajectory cluster density determination.

[0017] Furthermore, the preset time threshold used to determine the dwell trajectory segment is 0.5 hours, which can be dynamically adjusted according to actual application needs.

[0018] Furthermore, in step S4, the map matching algorithm is a map matching algorithm based on a Hidden Markov Model (HMM), which takes the GPS trajectory points as the explicit state and the road segment where the vehicle is located as the hidden state, defines the observation probability and the state transition probability, and obtains the optimal map matching path by solving the Viterbi algorithm.

[0019] In this invention, the map matching process is implemented using a Hidden Markov Model (HMM), which is an existing statistical modeling technique used to derive the most likely hidden state path from the observation sequence.

[0020] In this method, each GPS trajectory point is regarded as an observation in the HMM, referred to as the "manifest state"; the candidate road segments to which the trajectory point may be projected are regarded as the "hidden state".

[0021] The observation probability is defined as follows: the closer the spherical distance between a trajectory point and a candidate road segment, the greater the probability that the road segment is the true path. The state transition probability is defined as follows: the smaller the deviation between the path distance along the road network between two consecutive candidate road segments and the spherical distance between the trajectory points, the higher the transition probability.

[0022] The Viterbi algorithm (a commonly used shortest path search algorithm in Hidden Markov Models) is used as the solver. Its input is a sequence of trajectory points and corresponding candidate road segments, and its output is the sequence of optimal matching paths with the highest probability.

[0023] Furthermore, the observation probability is calculated based on the spherical distance from the trajectory point to the projection point of the road segment, and the observation probability P(z) t |r i The calculation formula for ) is as follows:

[0024]

[0025] Among them, z t Let r be the coordinate of the t-th time point. i For the i-th candidate road segment, x t,i For z t In r i The projection point on the z-axis, great_circle(z-axis) t ,x t,i ) represents z t With x t,i Great circle distance between them, σ z The initial state distribution is set to be based on the observation probability, representing the standard deviation of the observation error. When the distance between the observation point and the candidate road segment exceeds the distance threshold, the initial probability is set to zero to improve matching efficiency. An exemplary parameter is preferably 200 meters. This "exemplary parameter" refers to the recommended value selected in the experiments of this invention, which can balance efficiency and accuracy in actual road network matching. This invention does not limit this parameter to a fixed value; technicians can adjust it according to the actual road density and trajectory accuracy requirements. A typical value range is 150–300 meters.

[0026] In the Hidden Markov Model (HMM) map matching model, it is assumed that the actual driving path of the vehicle is a sequence of hidden states, and the GPS trajectory points are a sequence of observed states. At the initial moment of the model, the probability distribution of each candidate road segment as the initial hidden state needs to be set, which is called the initial state distribution.

[0027] To improve matching efficiency, this invention sets the initial probability in the initial state distribution to a value based on the observation probability function. When the trajectory observation point z at time t... t With candidate road segment r i If the spherical distance between two points is greater than a set distance threshold, the road segment is considered unreasonable, and its initial probability is set to zero to avoid unnecessary computational overhead. This distance threshold is an exemplary parameter in this embodiment of the invention, with a value of 200 meters. t ,x t,i If π > 200m, then let π be... i =0, otherwise π i ∝P(z t |r i The "exemplary parameter" mentioned refers to the recommended value selected in the experiments of this invention, which can balance efficiency and accuracy in the actual road network matching process. This invention does not limit the parameter to a fixed value, and technicians can adjust it according to the actual road density and trajectory accuracy requirements. The typical value range is 150 to 300 meters.

[0028] Furthermore, the state transition probability is calculated using an exponential decay function based on the difference in path length between adjacent trajectory points and road segments. The state transition probability P(d) t The calculation formula for ) is as follows:

[0029]

[0030] Where, d t =|route(x) t,i ,x t+1,j )-great_circle(z t ,z t+1 )|,route(x t,i ,x t+1,j ) is the projection point x t,i To x t+1,j The path length between them, great_circle(z t ,z t+1 (z) represents the observation point. t To z t+1 The great circle distance between them, β is an adjustable parameter, with an example parameter of 0.5. The matching radius threshold and the state transition parameter β can all be dynamically adjusted according to the traffic environment.

[0031] Further, in step S5, the dynamic statistical threshold is set as the upper quartile of the historical speed error ratio set, and the speed error ratio... The calculation formula is:

[0032]

[0033] Among them, v 设备 For device upload speed, v 推算 To calculate the speed, v 推算 The speed is calculated by dividing the distance d between two adjacent trajectory points on the road network after map matching by the time interval Δt.

[0034] Secondly, the present invention also provides a speed anomaly detection and fraud identification system based on trajectory data, for performing the speed anomaly detection and fraud identification method based on trajectory data as described above, the system comprising:

[0035] The data preprocessing module is used to acquire raw trajectory data and preprocess the raw trajectory data to remove invalid data and refresh abnormal data, so as to obtain a preprocessed trajectory point sequence.

[0036] The trajectory noise filtering module is used to filter trajectory noise from the preprocessed trajectory data. It uses a density-based spatial clustering (DBSCAN) algorithm to identify and remove noise points that deviate from the trajectory sequence, thus obtaining a denoised trajectory point sequence.

[0037] The trajectory segmentation module is used to segment the denoised trajectory point sequence according to the vehicle speed change, and distinguish between stationary trajectory segments and running trajectory segments;

[0038] The map matching module is used to match the trajectory points in the running trajectory segment to the roads on the map using a map matching algorithm, thereby obtaining a map matching path.

[0039] An abnormal speed detection module is used to calculate the estimated speed based on the map matching path and the timestamp of the trajectory point, compare the estimated speed with the device upload speed corresponding to the trajectory point, calculate the speed error ratio, and identify abnormal trajectory segments based on whether the speed error ratio exceeds a dynamic statistical threshold.

[0040] The speed verification and forgery identification module is used to verify the speed and identify forgery of the abnormal trajectory segment. By analyzing the speed distribution of the abnormal trajectory segment, it determines whether there is a preset forgery speed pattern. If so, it is determined to be a forgery behavior.

[0041] Compared with existing technologies, this invention systematically processes raw trajectory data in multiple stages, including preprocessing, DBSCAN noise filtering based on adaptive parameters, trajectory segmentation based on speed and time, and map matching based on a Hidden Markov Model. It then compares the estimated speed based on the map-matched path with the device's upload speed and uses dynamic statistical thresholds to accurately identify abnormal trajectory segments. Finally, it analyzes the speed probability density distribution of abnormal trajectory segments to determine fraudulent behavior. This combined approach not only improves the accuracy and robustness of speed anomaly detection, effectively identifying and removing noisy data and distinguishing the true operating state of vehicles, but also significantly enhances the ability to identify sophisticated and covert speed fraud by introducing dynamic thresholds and targeted forgery pattern analysis. This comprehensively improves the authenticity, credibility, and intelligent level of trajectory data supervision, demonstrating greater adaptability and higher application value. Attached Figure Description

[0042] The accompanying drawings, which form part of this specification, illustrate embodiments of the invention and, together with the specification, serve to explain the principles of the invention.

[0043] The invention will be more clearly understood with reference to the accompanying drawings and the following detailed description, wherein:

[0044] Figure 1 This is a flowchart of a speed anomaly detection and fraud identification method based on trajectory data provided in an embodiment of the present invention;

[0045] Figure 2 This is a schematic diagram of the map matching algorithm architecture of a speed anomaly detection and fraud identification method based on trajectory data provided in an embodiment of the present invention;

[0046] Figure 3 This is an exemplary code structure diagram of a map matching algorithm for a speed anomaly detection and fraud identification method based on trajectory data provided in an embodiment of the present invention;

[0047] Figure 4 This is an exemplary structure diagram of the pseudocode for the core steps of Hidden Markov Model (HMM) map matching provided in this embodiment of the invention;

[0048] Figure 5 This is an example diagram of noise data identification according to the present invention;

[0049] Figure 6 This is an example diagram of map matching in this invention. Detailed Implementation

[0050] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments and specific features in the embodiments are detailed descriptions of the technical solution of the present application, and not limitations thereof. Where there is no conflict, the embodiments and technical features in the embodiments can be combined with each other. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and should not be used to limit the scope of protection of the present invention.

[0051] In this article, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0052] like Figure 1 As shown, this embodiment provides a method for speed anomaly detection and fraud identification based on trajectory data, including the following steps:

[0053] Step S1: Obtain the original trajectory data of road vehicles and preprocess the original trajectory data to remove invalid data and refresh abnormal data, obtaining preprocessed trajectory data. The preprocessing includes: removing null data, data with abnormal refresh cycles, and data with abnormal physical limits from the original trajectory data. Data with abnormal refresh cycles refers to data whose acquisition interval exceeds the set sampling frequency tolerance range. The original input source of the trajectory data is real-time data collected by GPS devices, which includes multiple fields such as vehicle number, timestamp, latitude and longitude, speed, and direction angle. The original trajectory data is cleaned to remove invalid data with null values ​​and abnormal refresh cycles, as well as data with abnormal physical limits. The threshold for judging abnormal physical limits is 180 km / h, and the dynamic statistical threshold is set based on the upper quartile of the historical speed error ratio of the same vehicle.

[0054] Step S2: Perform trajectory noise filtering on the preprocessed trajectory data, use the density-based spatial clustering (DBSCAN) algorithm, and calculate the spherical distance between trajectory points using the spherical semi-versus formula; Figure 5 As shown, noise points deviating from the trajectory sequence are identified and removed to obtain denoised trajectory data. The neighborhood radius and minimum number of neighborhood points (minPts) of the DBSCAN algorithm are adaptively adjusted according to the sampling frequency of the trajectory data. Some trajectory points in the trajectory data are significantly deviated from the road network due to external factors interfering with GPS, and some trajectories even fall in mountains that are impossible to appear. Therefore, this part of the noise data also needs to be removed. The DBSCAN algorithm can extract sets of arbitrary shapes, so the DBSCAN algorithm is used to discover and remove noise points. In the clustering process of trajectory data, the spherical semi-versus formula is first used to calculate the distance between latitude and longitude. In the parameter setting of the clustering algorithm, the neighborhood radius ε is selected as 500 meters (considering the theoretical maximum displacement under a 30-second data interval), and the minimum number of points (minPts) is set to 3. Based on the maximum vehicle speed (120km / h) and the sampling interval (30 seconds), the maximum theoretical displacement is calculated to be 1km. Combined with GPS positioning error, the neighborhood radius ε is set to 500 meters, covering 95% of normal driving scenarios. minPts is set to 3 to ensure the minimum density of effective trajectory clusters. The spherical distance d is calculated using the following formula:

[0055]

[0056] Where d represents the spherical distance between adjacent trajectory points in kilometers; r is the Earth's radius, with a value of 6371 kilometers; φ1 and φ2 represent the latitude values ​​in radians between the two trajectory points; and λ1 and λ2 represent the longitude values ​​in radians between the two trajectory points. The main cluster data obtained from DBSCAN clustering is used as valid trajectory data, while noise points and other discrete small clusters are removed.

[0057] Step S3: Based on vehicle speed changes, the denoised trajectory data is divided into trajectory segments. Trajectory segments with continuous speeds of 0 and dwell times exceeding a preset threshold are identified as stationary trajectory segments, while the remaining trajectory segments are identified as running trajectory segments. The preset threshold for identifying stationary trajectory segments is 0.5 hours. When a vehicle is at an intersection in an urban road network, the dwell time is usually no more than 0.5 hours, hence the preset threshold is 0.5 hours. To adapt to highway service area scenarios, this can be adjusted to 1 hour. Because vehicles may stop for extended periods during transportation due to unloading, rest, or entertainment needs, the trajectory data remains stationary across multiple consecutive time intervals. This data dilutes speed features during speed feature analysis, making speed differences less significant and hindering speed feature extraction.

[0058] Step S4: For the aforementioned running trajectory segment, such as Figure 6 As shown, a map matching algorithm is used to match the trajectory points in the running trajectory segment to the roads on the map, obtaining the map matching path. The map matching algorithm is based on a Hidden Markov Model (HMM), using GPS trajectory points as explicit states and the road segment where the vehicle is located as implicit states. Observation probabilities and state transition probabilities are defined, and the optimal map matching path is obtained through the Viterbi algorithm. To adapt to the subsequent map matching module, the original data needs to be formatted. The trajectory data after trajectory division is saved as a standard GPX file (GPS Exchange Format), which is based on the XML standard.

[0059] In the GPX file structure, <trkpt>Each tag stores a single trajectory point, with latitude and longitude as attributes and timestamps as sub-tags; for example... Figure 3 As shown, each GPX file corresponds to one driving trajectory segment;

[0060] Electronic map data is acquired using the OSM.PBF open-source format (OpenStreetMap Protocol bufferBinaryFormat). Complete road network information within the study area is downloaded for subsequent map matching operations. After completing the above data preparation, the actual execution phase of the map matching algorithm can begin.

[0061] For each new trajectory data point, the map matching algorithm performs the following steps:

[0062] (1) Screening of adjacent road sections

[0063] Centered on the trajectory point, all road segments within a 200-meter radius are searched as candidate sets. If no candidate road is found, the search area is expanded to 500 meters and marked as a low-confidence region. Only road segments that meet the initial observation probability greater than zero are retained, i.e., the distance between the trajectory point and the road segment is within a set threshold. This step narrows the matching space and improves computational efficiency.

[0064] (2) Viterbi array update

[0065] After determining the set of candidate road segments, the following array information is updated using the basic derivation logic of Hidden Markov Models (HMM):

[0066] Each candidate road segment is taken as the current hidden state, and the state transition probability between it and the hidden state at the previous time step is calculated.

[0067] Update the maximum path probability to the current hidden state based on the observation probability and transition probability of trajectory points and road segments;

[0068] Record the forward pointers to each hidden state for subsequent optimal path backtracking.

[0069] (3) Implicit chain continuity check

[0070] If at a certain moment, the maximum path probability of all candidate hidden states is zero, it indicates that the hidden state chain has broken.

[0071] At this point, the break point and related status information are recorded as the starting point for the next trajectory matching iteration, ensuring the integrity and continuity of trajectory processing.

[0072] Through iterative updates using the above steps, the optimal road sequence corresponding to each trajectory segment can be derived, completing the map matching process and providing a precise location information basis for subsequent speed calculation and anomaly detection.

[0073] Step S5: Calculate the estimated speed based on the map matching path and the timestamps of the trajectory points. Compare the estimated speed with the device upload speed corresponding to the trajectory points to calculate the speed error ratio. If the speed error ratio exceeds a dynamic statistical threshold, it is marked as an abnormal trajectory segment. The dynamic statistical threshold is set as the upper quartile of the historical speed error ratio set. The calculation formula is:

[0074]

[0075] Among them, v 设备 For device upload speed, v 推算 To calculate the speed, v 推算 The velocity is calculated by dividing the distance d between two adjacent trajectory points on the road network after map matching by the time interval Δt. The upper quartile refers to the value located at the 75th percentile in the sorted dataset. It is calculated by the error ratio of all identified trace segments in the sample set and is updated as the data accumulates.

[0076] The error ratio of all trajectory segments is statistically analyzed. In statistics, quartiles are used to describe the characteristics of data distribution. In datasets that conform to a normal distribution, a value that differs from the dataset mean by one standard deviation is usually used as the outlier detection standard. However, in GPS fraud detection based on road matching, the distance error obtained from map matching is uncertain due to the loose nature of the data source. Therefore, quartiles are used as the anomaly detection threshold. If the trajectory segment error ratio exceeds the quartile threshold, it is marked as an abnormal trajectory segment. Step S6: Speed ​​verification and forgery identification are performed on the abnormal trajectory segments. If the amount of data on speed exceeding the speed limit in the speed distribution of the abnormal trajectory segment is zero, and the speed is concentrated at a preset speed limit threshold (e.g., 50 km / h), it is determined to be a batch forgery behavior. That is, the speed distribution of the abnormal trajectory segment is statistically analyzed to generate a probability density function (PDF). If the number of speed exceeding data points is zero, and the speed values ​​are concentrated near the speed limit threshold (e.g., 50 km / h) with a concentration exceeding 50%, it is determined to be a fixed-speed forged trajectory.

[0077] This invention enables low-cost trajectory anomaly detection and significantly improves the identification rate of forged speed data, ensuring the credibility and security of data analysis.

[0078] Further, in step S2, the spherical distance d is calculated using the following formula:

[0079]

[0080] Where r is the Earth's radius; φ1 and φ2 are the latitudes of the two trajectory points, respectively; and λ1 and λ2 are the longitudes of the two trajectory points, respectively.

[0081] Further, in step S4, the observation probability is calculated based on the spherical distance from the trajectory point to the projection point of the road segment, and the observation probability P(z) is... t |r i The calculation formula for ) is as follows:

[0082]

[0083] Among them, z t Let r be the coordinate of the t-th time point. i For the i-th candidate road segment, x t,i For z t In r i The projection point on the z-axis, great_circle(z-axis) t ,x t,i ) represents z t With x t,i Great circle distance between them, σ z This represents the standard deviation of the observation error. The initial state distribution is set based on the observation probability. When the distance between the observation point and the candidate road segment exceeds the distance threshold, the initial probability is set to zero to improve matching efficiency. An example parameter is 200 meters.

[0084] Furthermore, the state transition probability is calculated using an exponential decay function based on the difference in path length between adjacent trajectory points and road segments. The state transition probability P(d) t The calculation formula for ) is as follows:

[0085]

[0086] Where, d t =|route(x) t,i ,x t+1,j )-great_circle(z t ,z t+1 )|,route(x t,i ,x t+1,j ) is the projection point x t,i To x t+1,j The path length between them, great_circle(z t ,z t+1 (z) is the observation point. t To z t+1 The great circle distance between them, β, is an adjustable parameter. The "observation point" z t The term "projection point" refers to the GPS trajectory point collected by the vehicle at time t, i.e., the explicit state in the HMM model; the "projection point" x t,i This refers to the observation point z. t In candidate road segment r i The vertical projection points on the surface are used to calculate the actual possible location of the vehicle;

[0087] Distance from great_circle(z) t ,z t+1 The distance () represents the great circle distance between two observation points; the distance route(x) represents the distance between two observation points. t,i ,x t+1,j () represents the shortest travel path distance between two adjacent projection points on the road network.

[0088] As one embodiment, a speed anomaly detection and fraud identification system based on trajectory data is provided for performing the speed anomaly detection and fraud identification method based on trajectory data as described above, characterized in that the system comprises:

[0089] The data preprocessing module is used to acquire raw trajectory data and preprocess the raw trajectory data to remove invalid data and refresh abnormal data, so as to obtain preprocessed trajectory data.

[0090] The trajectory noise filtering module is used to filter trajectory noise from the preprocessed trajectory data. It uses a density-based spatial clustering (DBSCAN) algorithm to identify and remove noise points that deviate from the trajectory sequence, thus obtaining denoised trajectory data.

[0091] The trajectory segmentation module is used to segment the denoised trajectory data according to changes in vehicle speed, distinguishing between stationary trajectory segments and running trajectory segments;

[0092] The map matching module is used to match the trajectory points in the running trajectory segment to the roads on the map using a map matching algorithm, thereby obtaining a map matching path.

[0093] An abnormal speed detection module is used to calculate the estimated speed based on the map matching path and the timestamp of the trajectory point, compare the estimated speed with the device upload speed corresponding to the trajectory point, calculate the speed error ratio, and identify abnormal trajectory segments based on whether the speed error ratio exceeds a dynamic statistical threshold.

[0094] The speed verification and forgery identification module is used to verify the speed and identify forgery of the abnormal trajectory segment. By analyzing the speed distribution of the abnormal trajectory segment, it determines whether there is a preset forgery speed pattern. If so, it is determined to be a forgery behavior.

[0095] This invention provides a method and system for speed anomaly detection and forgery identification based on trajectory data. By performing multi-dimensional cleaning, noise identification, trajectory division, map matching, and speed estimation analysis on GPS-collected trajectories, it accurately identifies forgery behavior in uploaded data, thereby improving the authenticity and credibility of trajectory data.

[0096] First, this invention preprocesses the raw trajectory data to remove null values, refresh cycle anomalies, and physical limit anomalies (such as speeds exceeding a reasonable range of 180 km / h), ensuring the basic quality of the data source.

[0097] Secondly, for the offset points in the trajectory caused by GPS errors, this invention uses a density-based spatial clustering algorithm (DBSCAN) to identify and remove noise points. Specifically, the spherical distance between trajectory points is calculated using the spherical semi-versus formula, and a neighborhood radius ε (e.g., 500 meters) and a minimum number of points minPts (e.g., 3) are set as clustering discrimination parameters. Only the main clusters are retained, and outlier noise data is removed.

[0098] Subsequently, the trajectory is divided based on speed change characteristics. When multiple consecutive trajectory points have a speed of zero for a duration exceeding a preset threshold (e.g., 0.5 hours), it is determined to be a stationary segment, and the remaining parts are classified as running segments. This method effectively extracts continuous driving behavior characteristics and eliminates the interference of parking and stationary stops on speed statistics.

[0099] For the trajectory segment, this invention employs a map matching method based on a Hidden Markov Model (HMM). GPS trajectory points are treated as explicit states, and candidate road segments as hidden states. Observation probabilities and state transition probabilities are defined, and the optimal road sequence is derived using the Viterbi algorithm. The observation probability measures the spatial matching degree between trajectory points and the projected points of road segments; the state transition probability is defined based on the distance difference between the matching projections of preceding and following trajectory points on the road network; the HMM model uses the Viterbi algorithm to solve for all possible road matching paths and outputs the optimal path sequence with the highest probability.

[0100] During the matching process, neighboring candidate road segments are first screened, then the optimal path probability array is dynamically updated, and the state is recorded when a hidden state chain break occurs to ensure the continuity of trajectory matching.

[0101] After obtaining accurate road matching, this invention calculates the vehicle's actual driving speed based on the trajectory point timestamps and the road segment matching distance, and calculates the error ratio between the calculated speed and the device's upload speed. The upper quartile of all error ratios is used as a dynamic anomaly detection threshold. If the error ratio of a single trajectory segment exceeds this threshold, it is marked as an abnormal trajectory segment.

[0102] Finally, the abnormal trajectory segments are further analyzed for speed distribution to generate a probability density function (PDF). If the abnormal segment shows zero speeding data in the speed distribution and forms a high concentration (more than 50%) at the speed limit threshold (e.g., 50 km / h), the trajectory segment can be determined to be data forged using fixed parameters and marked.

[0103] This invention enables efficient detection and classification of forgery in trajectory data, and is particularly suitable for trajectory authenticity review and risk supervision in fields such as freight vehicles and urban transportation. It has the advantages of high processing accuracy, strong adaptability and high degree of automation.

[0104] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.< / trkpt>

Claims

1. A method for speed anomaly detection and fraud identification based on trajectory data, characterized in that, Includes the following steps: Step S1: Obtain the original trajectory data of road vehicles, preprocess the original trajectory data to obtain a preprocessed trajectory point sequence; the original trajectory data includes the latitude and longitude coordinates of the trajectory points, the timestamp of the trajectory points, and the device upload speed corresponding to the trajectory points; Step S2: Perform trajectory noise filtering on the preprocessed trajectory point sequence. Based on the density-based spatial clustering algorithm, calculate the spherical distance between each trajectory point and other trajectory points in its neighborhood using the spherical semi-versus formula, determine its spatial distribution density, and the spherical distance is based on the latitude and longitude coordinates of the trajectory point. Remove all noise points to obtain the denoised trajectory point sequence. Step S3: Based on the vehicle speed change reflected by the device upload speed corresponding to the trajectory point, the denoised trajectory point sequence is divided into trajectories. The trajectory segments with continuous speed of 0 and dwell time exceeding the preset time threshold are determined as dwell trajectory segments, and the remaining trajectory segments are determined as running trajectory segments. Step S4: For the running trajectory segment, a map matching algorithm is used to match the trajectory points in the running trajectory segment to the roads on the map to obtain the map matching path; Step S5: Calculate the estimated speed based on the map matching path and the timestamp of the trajectory point, compare the estimated speed with the device upload speed corresponding to the trajectory point, calculate the speed error ratio, and if the speed error ratio exceeds the dynamic statistical threshold, mark it as an abnormal trajectory segment. Step S6: Perform speed verification and forgery identification on the abnormal trajectory segment. If the amount of data on the overspeed in the speed distribution of the abnormal trajectory segment is zero, and the speed shows a concentrated situation of exceeding 50% at the preset overspeed threshold, then it is determined to be a forgery.

2. The speed anomaly detection and fraud identification method based on trajectory data according to claim 1, characterized in that, The preprocessing includes: removing invalid trajectory points with null values ​​in the fields; removing abnormal points whose sampling time interval exceeds the tolerance range of the set sampling frequency, which is 30±10 seconds; and removing abnormal points whose speed exceeds the physical limit of the vehicle.

3. The speed anomaly detection and fraud identification method based on trajectory data according to claim 1 or 2, characterized in that, In step S2, the radius ∈ of the neighborhood of the DBSCAN algorithm and the minimum number of neighborhood points minPts are adaptively adjusted according to the trajectory point sampling frequency; the sampling frequency refers to the time interval between continuous data uploads by the trajectory device, in seconds, and is set to be sampled once every 30±10 seconds; ∈ is set to 400~1000 meters, and minPts is set to 3~5.

4. The speed anomaly detection and fraud identification method based on trajectory data according to claim 3, characterized in that, In step S2, the spherical distance d is calculated using the following formula: Where r is the Earth's radius; φ1 is the latitude of the first trajectory point, φ2 is the latitude of the second trajectory point; λ1 is the longitude of the first trajectory point, λ2 is the longitude of the second trajectory point, and the second trajectory point is a trajectory point within the domain of the first trajectory point.

5. The speed anomaly detection and fraud identification method based on trajectory data according to claim 1, characterized in that, The preset time threshold used to determine the stationary trajectory segment is 0.5 hours.

6. The speed anomaly detection and fraud identification method based on trajectory data according to claim 1, characterized in that, In step S4, the map matching algorithm is a map matching algorithm based on a hidden Markov model. The preprocessed GPS trajectory points are taken as the model observation state, denoted as the explicit state, and the candidate road segments where the vehicle may be located are taken as the implicit state, denoted as the hidden state. The observation probability and state transition probability of the hidden Markov model are defined, and the optimal map matching path is obtained by solving the Viterbi algorithm.

7. The speed anomaly detection and fraud identification method based on trajectory data according to claim 6, characterized in that, The observation probability is calculated based on the spherical distance from the trajectory point to the projection point of the road segment, and the observation probability P(z) t |r i The calculation formula for ) is as follows: Among them, z t Let r be the coordinate of the t-th time point. i For the i-th candidate road segment, x t,i For z t In r i The projection point on the z-axis, great_circle(z-axis) t ,x t,i ) represents z t With x t,i Great circle distance between them, σ z This represents the standard deviation of the observation error.

8. The speed anomaly detection and fraud identification method based on trajectory data according to claim 7, characterized in that, The state transition probability is based on the difference in path length from adjacent trajectory points to road segments, and the state transition probability P(d) is... t Modeling is performed based on the path consistency between adjacent observation points projected onto candidate road segments; the formula for calculating the state transition probability is as follows: where d t = |route(x t,i , x t+1,j ) - great_circle(z t , z t+1 )|, z t ,z t+1 Let x be the GPS observation point at time t and time t+1. t,i ,x t+1,j For z t ,z t+1 In candidate road segment r i ,r j The projection point on the road; route() represents the road path length, route(x t,i ,x t+1,j ) is the projection point x t,i To x t+1,j The path length between them, great_circle(z t ,z t+1 ), representing the great circle distance on the sphere, and β is an adjustable parameter used to control the tolerance for path deviation.

9. The speed anomaly detection and fraud identification method based on trajectory data according to claim 1, characterized in that, In step S5, the dynamic statistical threshold is set to the upper quartile of the historical speed error ratio set; the speed error ratio The calculation formula is as follows: Among them, v 设备 The speed field recorded in the GPS track points; v 推算 The speed is calculated by dividing the distance d between two adjacent trajectory points on the road network after map matching by the time interval Δt; the upper quartile refers to the value located at the 75th percentile in the sorted dataset.

10. A system for performing the speed anomaly detection and fraud identification method based on trajectory data as described in any one of claims 1-9, characterized in that, The system includes: A data preprocessing module is used to acquire raw trajectory data and clean the raw trajectory data to obtain a preprocessed sequence of trajectory points. The cleaning process includes: Remove data points with empty fields as invalid data; Trajectory points whose collection time interval exceeds the preset tolerance range are removed and treated as abnormal data in the refresh cycle; Data points whose speed exceeds physical limits are removed as physical limit anomalies. This is how we obtain cleaned, effective trajectory data for subsequent analysis and processing. The trajectory noise filtering module is used to filter trajectory noise from the preprocessed trajectory point sequence. It uses a density-based spatial clustering algorithm to identify and remove noise points that deviate from the trajectory sequence, thus obtaining a denoised trajectory point sequence. The trajectory segmentation module is used to segment the denoised trajectory point sequence according to the vehicle speed change, and distinguish between stationary trajectory segments and running trajectory segments; The map matching module is used to match the trajectory points in the running trajectory segment to the roads on the map using a map matching algorithm, thereby obtaining a map matching path. An abnormal speed detection module is used to calculate the estimated speed based on the map matching path and the timestamp of the trajectory point, compare the estimated speed with the device upload speed corresponding to the trajectory point, calculate the speed error ratio, and identify abnormal trajectory segments based on whether the speed error ratio exceeds a dynamic statistical threshold. The speed verification and forgery identification module is used to verify the speed and identify forgery of the abnormal trajectory segment. By analyzing the probability density function PDF of the speed distribution of the abnormal trajectory segment, it determines whether there is a preset forgery speed pattern. If so, it is determined to be a forgery behavior.

Images