Freight logistics track abnormity processing method and system

Abstract

The invention discloses a freight logistics track abnormity processing method and system, and belongs to the technical field of intelligent logistics monitoring, and the method comprises the steps: obtaining a track original data set of a target freight vehicle; constructing a behavior evolution model based on the historical trajectory data, wherein the behavior evolution model is used for describing conventional operation boundaries of the vehicle under different road types and time periods; performing multi-source fusion correction on the trajectory data to generate a trajectory point sequence; calculating credibility factors of the track points based on GPS stability, historical track goodness of fit and road matching degree; constructing an anomaly judgment graph, and generating an anomaly distribution function by adopting a recursive scoring mechanism; identifying a track abnormal fragment based on a dynamic adaptive threshold value and outputting an abnormal label; generating a scheduling suggestion according to the abnormal label and a vehicle scheduling rule, and pushing the scheduling suggestion to a vehicle-mounted system in real time; according to the invention, individuation, high precision and real-time performance of track abnormity identification are realized, and the method is suitable for an intelligent logistics management and dynamic scheduling system in a complex scene.

Description

Technical Field

[0001] This invention relates to the field of intelligent logistics monitoring technology, specifically to a method and system for handling abnormal freight logistics trajectories. Background Technology

[0002] In modern freight logistics systems, trajectory data is not only used for real-time positioning but also carries key functions such as anomaly warning, route planning, and scheduling optimization. However, most existing trajectory anomaly detection methods rely on simple rule templates or static threshold settings, which cannot accurately cope with the complexity of transportation routes and the diversity of anomaly situations. For example, in areas with severe signal interference, such as mountainous areas and urban fringe areas, GPS trajectories often exhibit "jump points" or "drifts," leading to misjudgments of anomalies. Furthermore, in multi-vehicle collaboration and cross-regional scheduling scenarios, the operating patterns of different vehicles vary significantly, making it difficult to adapt a unified model.

[0003] Some existing solutions attempt to introduce machine learning models for trajectory classification and anomaly identification, but still have the following limitations: (1) Most of them ignore the evolutionary relationship of behavioral patterns behind the trajectory and lack the ability to model the vehicle's own operating habits; (2) It is difficult to integrate signals from multiple sources such as GPS, vehicle sensors, and road basic data, resulting in insufficient stability of anomaly identification; (3) The trajectory points themselves lack a credibility measurement mechanism, making it difficult to handle the crosstalk of continuous anomaly points, further weakening the interpretability and scheduling linkage of the system.

[0004] Therefore, there is an urgent need for a method that can adaptively model freight vehicle historical behavior patterns, integrate multi-source location signals and construct location confidence weights, and realize real-time perception and classification of trajectory anomalies through a recursive strategy, thereby improving the overall stability, robustness and practicality of the system. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for handling abnormal freight logistics trajectories, in order to address the shortcomings in the prior art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for handling abnormal freight logistics trajectories, comprising: S100. Obtain the original dataset T of the target freight vehicle's trajectory within the current transportation cycle; S200. Construct a behavioral evolution model of the target vehicle based on historical trajectory data to characterize the normal operating mode and variation boundary of the target vehicle under different road types and time periods. S300. Perform multi-source fusion correction on the original trajectory dataset T to obtain the fused trajectory point sequence. ; S400. Calculate the confidence factor for each trajectory point in the trajectory point sequence T′. The confidence factor comprehensively considers GPS stability, historical trajectory consistency and road matching. S500, based on trajectory point sequence An anomaly determination graph G is constructed with a behavioral evolution model. In graph G, nodes represent trajectory segments, and edge weights represent the degree of behavioral deviation between points. A recursive scoring mechanism is used to traverse graph G to obtain the anomaly distribution function R(x). S600. When R(x) exceeds the dynamic adaptive threshold R, the corresponding trajectory segment is determined to be abnormal, and an abnormal label is output. S700 generates scheduling suggestions based on anomaly tags and vehicle scheduling rules and pushes them to the vehicle terminal system in real time.

[0007] Preferably, S400 includes: S401, Based on trajectory point sequence The original GPS signal quality index of each point is used to construct the GPS stability factor Q. The GPS stability factor Q integrates the number of satellites, the positioning accuracy factor and the time interval mutation rate to reflect the positioning stability of each point. S402. For each trajectory point, extract the density kernel function value corresponding to its location in the historical trajectory set, and construct the trajectory matching matrix H to measure the degree of matching between the current trajectory point and the historical behavior trajectory. S403. Combine digital map data to perform multi-scale road projection on the trajectory points, use a spatial consistency convolutional network to calculate the matching deviation value d between the point and the road centerline, and construct the road matching factor M. S404. Based on the collaborative scoring mechanism of Q, H and M, they are integrated into a credibility factor for trajectory points.

[0008] Preferably, the calculation of Q, H, and M includes: The calculation of the GPS stability factor Q includes: extracting the corresponding raw GNSS data for each trajectory point and calculating the number of visible satellites. A stability scoring function is constructed using the horizontal precision factor HDOP and the time interval Δt between adjacent points: ,in This represents the average of the trajectory sampling times. This is an empirical adjustment coefficient; The calculation of the historical trajectory similarity factor H includes: using the current trajectory point Centered on a point, define a spatiotemporal neighborhood with radius r, and retrieve historical trajectory sets and... For trajectory segments whose trajectory feature vector similarity exceeds a threshold θ, a point density function D( ) is constructed using the kernel density estimation method. ),get This is used to reflect whether the current point is located in a densely distributed area of ​​historical normal trajectory distribution; The calculation of the road matching factor M includes: [the calculation of the trajectory points...] By mapping the point to the centerline of the road segment it is located using multi-scale projection, a spatially consistent convolutional network is used to extract the geometric difference features between the road boundary and the trajectory, and the centerline deviation distance of the point is output. And standardized to obtain the matching degree index: Where σ is the Sigmoid function, This is the maximum reasonable offset distance set.

[0009] Preferably, Q, H, and M are converted into a comprehensive feature vector, which is used as the input to the machine learning model. The machine learning model uses the prediction of the confidence factor label of each trajectory point by each set of comprehensive feature vectors as the prediction objective, and minimizes the sum of prediction errors of the confidence factor labels of all trajectory points as the training objective. The machine learning model is trained until the sum of prediction errors converges, at which point the model training stops. The confidence factor of each trajectory point is determined based on the model output. The machine learning model is a multinomial regression model.

[0010] Preferably, S500 includes: S501. Construct an anomaly detection graph G = (V, E), where each node... Represents trajectory points Its attributes include a credibility factor. behavioral deviation indicators ;side This represents the temporal coherence between trajectory point pairs, assigning association weights based on the rate of change of velocity and abrupt changes in direction. ; S502, Initialization Abnormal activation value of the starting trajectory point Define the recursive function A depth-first traversal strategy is used to expand the nodes of graph G, and the Φ function is applied sequentially to update the score of each node. S503, calculated for all nodes The values ​​are subjected to sliding normalization and abrupt change detection to form the final anomaly distribution function R(x), which is used to characterize the potential anomaly density changes in the entire trajectory.

[0011] Preferably, the step of outputting anomaly labels based on the anomaly distribution function R(x) includes: S601. Perform sliding window segmentation on R(x) and calculate the local mean within each window. with standard deviation A dynamic adaptive threshold function is constructed based on the global distribution of R(x) across the entire trajectory. , where k is the adjustment factor; S602, compare R(x) with the corresponding interval. For comparison, if there are M consecutive trajectory points x ∈ satisfy If so, then extract that segment as a candidate abnormal segment; S603, Behavioral Deviation Indicators Based on Candidate Abnormal Fragments Credibility factor Road matching degree Perform feature fusion, input the data into a multi-class decision model, determine the anomaly type of the segment, and output anomaly label E ∈{drift anomaly, path deviation, behavioral mutation, data distortion}.

[0012] Preferably, S200 includes: S201. Perform two-layer clustering on the trajectory data of the target vehicle in the last N transportation cycles according to road type and time period, and extract the trajectory feature set F, including average speed sequence, acceleration fluctuation value and trajectory offset rate. S202. Construct a multi-channel recurrent autoencoder network to encode and compress the feature set F, and introduce a residual mapping module between channels to retain fine-grained temporal behavior variation information; S203. Construct a behavior variation map in two-dimensional time-road coordinates based on multi-channel feature representation, and use a boundary learning function to fit the boundary of local mutation regions in the map to obtain the operational boundary set for each type of road and cycle. S204. The set of operating boundaries is used as the core constraint of the target vehicle behavior evolution model and stored in the model memory unit.

[0013] Preferably, in S100, the original trajectory dataset T includes: GPS positioning points, speed, acceleration, heading angle, road type, and timestamp.

[0014] Preferably, in S300, the correction includes joint optimization based on inertial navigation, geographic reference points, and signal strength data.

[0015] The present invention also provides a freight logistics trajectory anomaly handling system, comprising: The data acquisition module obtains the raw dataset T of the target freight vehicle's trajectory within the current transportation cycle; The behavior modeling module constructs a behavior evolution model of the target vehicle based on historical trajectory data, which is used to characterize the normal operating mode and variation boundary of the target vehicle under different road types and time periods; The trajectory fusion and preprocessing module performs multi-source fusion correction on the original trajectory dataset T to obtain the fused trajectory point sequence. ; The trajectory reliability assessment module evaluates the trajectory point sequence. A confidence factor is calculated for each trajectory point, and the confidence factor comprehensively considers GPS stability, historical trajectory consistency and road matching. Anomaly scoring module, based on trajectory point sequence An anomaly determination graph G is constructed with a behavioral evolution model. In graph G, nodes represent trajectory segments, and edge weights represent the degree of behavioral deviation between points. A recursive scoring mechanism is used to traverse graph G to obtain the anomaly distribution function R(x). The anomaly detection module determines that there is an anomaly in the corresponding trajectory segment when R(x) exceeds the dynamic adaptive threshold R, and outputs an anomaly label. The scheduling response module generates scheduling suggestions based on anomaly tags and vehicle scheduling rules and pushes them to the vehicle-mounted system in real time.

[0016] The technical effects and advantages provided by the present invention in the above technical solution are as follows: 1. This invention provides a trajectory anomaly handling method for freight logistics scenarios. By introducing non-traditional techniques such as behavioral evolution modeling, multi-source data fusion, trajectory credibility assessment, and graph structure recursive scoring, it significantly improves the accuracy, response speed, and adaptability of trajectory anomaly identification. Specifically, the behavioral evolution model can characterize the operational boundary features of target vehicles under different road conditions and time periods, enabling individualized behavioral deviation detection; the trajectory point credibility factor integrates three dimensions: GPS signal stability, historical trajectory consistency, and road matching, effectively suppressing misjudgments or missed judgments caused by fluctuations in single data points.

[0017] 2. This invention is the first to introduce a recursive scoring mechanism into the trajectory anomaly graph structure, combining a depth-first traversal strategy with a dynamic threshold model to achieve high-precision anomaly segment localization at the trajectory level. Simultaneously, the linkage mechanism between anomaly labels and vehicle scheduling rules realizes a closed-loop processing path from anomaly identification to scheduling response, exhibiting good real-time performance and system compatibility. This method is applicable to complex transportation scenarios such as highways, urban roads, and mountain routes, and has broad application prospects, particularly suitable for large-scale fleet management, intelligent logistics systems, and urban freight safety supervision platforms. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0019] Figure 1 This is a mind map of the method of the present invention.

[0020] Figure 2 This is a mind map of the system modules of the present invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] Example 1, please refer to Figure 1 As shown in this embodiment, a method for handling abnormal freight logistics trajectories includes: S100. Obtain the original dataset T of the target freight vehicle's trajectory within the current transportation cycle; S200. Construct a behavioral evolution model of the target vehicle based on historical trajectory data to characterize the normal operating mode and variation boundary of the target vehicle under different road types and time periods. S300. Perform multi-source fusion correction on the original trajectory dataset T to obtain the fused trajectory point sequence. ; S400. Calculate the confidence factor for each trajectory point in the trajectory point sequence T′. The confidence factor comprehensively considers GPS stability, historical trajectory consistency and road matching. S500. Based on the trajectory point sequence T′ and the behavior evolution model, an anomaly judgment graph G is constructed. In the graph G, nodes represent trajectory segments, and edge weights represent the degree of behavioral deviation between points. A recursive scoring mechanism is used to traverse the graph G to obtain the anomaly distribution function R(x). S600. When R(x) exceeds the dynamic adaptive threshold R, the corresponding trajectory segment is determined to be abnormal, and an abnormal label is output. S700 generates scheduling suggestions based on anomaly tags and vehicle scheduling rules and pushes them to the vehicle terminal system in real time.

[0023] In one specific embodiment of the present invention, the original trajectory dataset T for trajectory anomaly detection is collected in real time by deploying a multi-source data acquisition module on the target freight vehicle. This module integrates a GPS receiver, an onboard inertial measurement unit (IMU), a vehicle control unit (VCU) interface, and a road information perception interface.

[0024] The original trajectory dataset T contains the following data fields: GPS location points: record the vehicle's current location coordinates at a frequency of 1 Hz; Speed: obtained synchronously from the GPS module and the vehicle's OBD interface, in km / h; Acceleration: The longitudinal and lateral accelerations of the vehicle are collected by the IMU module, and the unit is m / s². Heading angle: Represents the angle of the vehicle's direction of movement, ranging from [0°, 360°], and is obtained by fusing GPS heading and IMU data; Road type: Based on the matching results of the current coordinates point in the high-precision map, it is automatically labeled as highway, urban arterial road, rural road, etc. Timestamp: Records the time of data collection for each group, with millisecond precision, and is used for subsequent time series modeling.

[0025] The raw trajectory data is organized in a sequence structure and uploaded to the cloud processing module for subsequent behavior modeling and anomaly detection processes. To ensure data quality, a packet loss detection mechanism and data buffering and retransmission logic are implemented at the acquisition end to ensure the integrity and temporal consistency of the trajectory data.

[0026] In a preferred embodiment of the present invention, in order to construct a behavior evolution model that can reflect the individualized behavior pattern of the target vehicle, the system performs deep feature learning and boundary modeling on the trajectory data within its last N transportation cycles (N is generally set to 20 to 50 times), specifically including the following steps: The system first performs two-level clustering on historical trajectory data according to road type (e.g., highways, urban arterial roads, secondary roads) and time period (e.g., daytime, nighttime, peak hours) to enhance the homogeneity of the data structure. After clustering, a set of representative trajectory features F is extracted from each cluster, mainly including: Average speed sequence (reflecting driving rhythm on different road sections); Acceleration fluctuation values ​​(used to detect abnormal start-stop or driving style); Trajectory offset rate (quantifies the degree to which a trajectory point deviates from its historical center path).

[0027] For the feature set F, a multi-channel recurrent autoencoder network is used for high-dimensional sequence compression. Different channels handle velocity dynamics, acceleration changes, and offset features respectively, and a residual mapping module is introduced between channels. This module can preserve micro-level time-series variations during the encoding process, thereby avoiding feature degradation or ambiguity.

[0028] For the encoded multi-channel features, the system constructs a behavior variation map on a two-dimensional coordinate system (time period × road type), with each map unit reflecting the degree of feature deviation in that dimension. Subsequently, a boundary learning function (such as convex hull fitting combined with gradient-sensitive region detection) is used to extract local mutation boundaries in the map, forming a set of abnormal response intervals.

[0029] The aforementioned boundary intervals are summarized as the "operational boundary set," representing the range of typical behavioral fluctuations of the target vehicle under different time periods and road conditions, and stored in the model's memory unit. This behavioral evolution model will serve as a structural benchmark for anomaly identification during actual trajectory evaluation, used for dynamic comparison of the degree of deviation from the current behavior.

[0030] This embodiment combines multidimensional clustering, temporal compression, and graph boundary modeling to achieve high-precision boundary extraction and structural memory in personalized behavior modeling. This significantly improves the ability of subsequent anomaly detection to identify "behavioral mutations" in individual vehicles and has good generalization and transferability.

[0031] In embodiments of the present invention, to improve the spatial continuity and positioning accuracy of trajectory data and avoid the influence of GPS drift, breakpoints, or abnormal jumps on the trajectory determination results, the system performs multi-source fusion correction processing on the original trajectory dataset T to generate a more stable and reliable trajectory point sequence. Specifically, it includes the following steps: In areas with weak or interrupted GPS signals, acceleration and angular velocity data are acquired in real time through the inertial measurement unit (IMU) on the vehicle, and a Kalman filter is used for trajectory estimation and state update to achieve continuous compensation for missing or distorted GPS points and construct a preliminary corrected trajectory sequence T1.

[0032] The system uses a point cluster matching algorithm to perform geographic anchoring correction on T1 based on the distance deviation between the geographic reference points (such as intersection centers, speed limit signs, ETC gantries, bridges, etc.) in the high-precision map and the corresponding locations in the trajectory, so as to make the trajectory more accurately fit the road geometry and generate a structure-aligned trajectory sequence T2.

[0033] At some locations, the system introduces the GPS Received Signal Strength Index (RSSI), the number of satellites, and the HDOP index to construct a location confidence weight matrix W for each point. This W is then used as a fusion constraint factor to perform confidence-weighted correction on the trajectory point coordinates in T2, resulting in the final fused and corrected trajectory point sequence T′.

[0034] The above three types of data are integrated to form a multi-source joint optimization mechanism, ensuring that the trajectory data still has good continuity and accuracy even under unstable signal and complex road conditions.

[0035] This embodiment achieves highly robust correction of trajectory data without relying on a single data source, providing a more solid data foundation for subsequent credibility assessment and anomaly identification. It is particularly suitable for complex transportation scenarios such as densely populated urban high-rise areas, mountain tunnel sections, or high-speed transfers.

[0036] In a preferred embodiment of the present invention, to improve the accuracy and robustness of anomaly identification in freight logistics trajectories, a multi-dimensional fusion trajectory credibility evaluation mechanism is designed. This mechanism constructs three core indicators based on GPS signal quality, historical trajectory distribution, and road structure matching information: GPS stability factor Q, historical trajectory consistency factor H, and road matching factor M. These three indicators are converted into a unified comprehensive feature vector and used as input to a machine learning model to train the model to predict the credibility factor label for each trajectory point.

[0037] This embodiment details the implementation process of this mechanism, specifically including the following steps: The system first collects a raw dataset T of the trajectory of the target freight vehicle during transportation. Each trajectory point... Includes the following fields: GPS location coordinates; speed With acceleration ; Heading angle ; GNSS signal indicators (including the number of satellites) Horizontal precision factor ); Sampling timestamp ; Road type (automatically marked by high-precision map); The calculation of the GPS stability factor Q includes: extracting the corresponding raw GNSS data for each trajectory point and calculating the number of visible satellites. A stability scoring function is constructed using the horizontal precision factor HDOP and the time interval Δt between adjacent points: ,in This represents the average of the trajectory sampling times. This is an empirical adjustment coefficient; The calculation of the historical trajectory similarity factor H includes: using the current trajectory point Centered on a point, define a spatiotemporal neighborhood with radius r, and retrieve historical trajectory sets and... For trajectory segments whose trajectory feature vector similarity exceeds a threshold θ, a point density function D( ) is constructed using the kernel density estimation method. ),get This is used to reflect whether the current point is located in a densely distributed area of ​​historical normal trajectory distribution; The calculation of the road matching factor M includes: [the calculation of the trajectory points...] By mapping the point to the centerline of the road segment it is located using multi-scale projection, a spatially consistent convolutional network is used to extract the geometric difference features between the road boundary and the trajectory, and the centerline deviation distance of the point is output. And standardized to obtain the matching degree index: Where σ is the Sigmoid function, This is the maximum reasonable offset distance set.

[0038] Q, H, and M are converted into a comprehensive feature vector. This comprehensive feature vector is used as the input to the machine learning model. The machine learning model uses the prediction of the confidence factor label of each trajectory point by each set of comprehensive feature vectors as the prediction objective. The training objective is to minimize the sum of the prediction errors of the confidence factor labels of all trajectory points. The machine learning model is trained until the sum of prediction errors converges, at which point the model training stops. The confidence factor of each trajectory point is determined based on the model output. The machine learning model is a multinomial regression model.

[0039] In embodiments of the present invention, to achieve high-precision and interpretable identification of potential anomalies in freight vehicle trajectories, the system introduces a graph-based recursive anomaly scoring mechanism, combined with an adaptive threshold model and a multi-class label recognition method, to construct a complete trajectory anomaly identification process. This process includes two core stages: constructing an anomaly determination graph and generating an anomaly distribution function R(x), and outputting anomaly labels based on R(x).

[0040] The system first performs graph structure modeling on the fused and corrected trajectory point sequence T′ to generate an anomaly determination graph G = (V, E).

[0041] Each trajectory point Corresponding to a node in the graph Its node attributes include two core metrics: credibility factor. (Calculated by machine learning model); behavioral deviation index This indicates the deviation of the current trajectory point relative to the historical running model, specifically including speed difference, acceleration fluctuation and sudden change in direction; For adjacent points in the trajectory Construct the edges in the graph Each edge is assigned a temporal coherence weight. This weight is calculated based on the following two items: rate of change of velocity. sudden change in heading angle ; Association weight It is obtained by weighted summation of the rate of change of velocity and the abrupt change in heading angle; The smaller the value, the more drastic the changes between trajectory points, and the more likely it is to be an abnormal transition.

[0042] To assess the propagation and accumulation trends of abnormal risks, the system introduces a recursive function Φ and performs scoring propagation on graph G based on a depth-first traversal strategy of the graph structure.

[0043] Start trajectory point corresponding nodes The initial abnormal activation value is set to (e.g., 0.01), that is: .

[0044] For any node in the graph Its anomaly score is defined as: ;in: The score for the previous trajectory point forms a dynamic accumulation process; The lower, The higher, the better The higher the value, the better; α, β, and γ are adjustable parameters that control the weight of the three factors in the score.

[0045] A depth-first traversal strategy is used to expand the nodes of graph G, and the Φ function is executed recursively in sequence to finally obtain the anomaly score value of each trajectory point. .

[0046] To enhance the distinguishability and stability of R(x) in actual trajectory sequences, the system... The sequence undergoes post-processing, including: Sliding window normalization: locally smooths R(x) to eliminate occasional jumps; Peak detection: The first derivative abrupt change detection method is used to identify continuous high value regions; An anomaly distribution function R(x) is formed as a quantitative expression of the anomaly risk distribution of trajectory segments.

[0047] To determine which regions in R(x) constitute anomalies, the system introduces a dynamic adaptive threshold function. This is to enhance the model's adaptability to different trajectory environments.

[0048] Divide R(x) into multiple sub-segments according to a fixed window size (e.g., 20 points); Calculate the local mean of each segment. with standard deviation ; Constructing a dynamic threshold function: ; where k is an adjustment factor (empirically set such as k = 1.5 or selected based on model optimization).

[0049] Compare all points in R(x), and when M consecutive trajectory points (e.g., M=5) satisfy... > At that time, it was marked as a candidate abnormal trajectory segment. ; If this condition is met consecutively in multiple regions, multiple abnormal segments can be identified for further analysis by the classification model.

[0050] Each candidate anomaly fragment contains multiple trajectory points and their attributes. The system extracts the following features as input vectors: Average behavioral deviation index ; Average confidence factor ; Average road matching ; Constructing feature vectors The data is input into a pre-trained multi-class decision model (such as random forest, SVM, or lightweight neural network) to predict the anomaly type.

[0051] The model outputs anomaly labels E ∈ {drift anomaly (e.g., sudden GPS point jump); path deviation (e.g., vehicle deviating from the planned route); behavioral mutation (e.g., frequent acceleration / deceleration, sharp turns); data distortion (e.g., duplicate point sets, timestamp anomalies)}.

[0052] Finally, the system outputs the start and end points, anomaly label, and confidence score for each abnormal trajectory segment, which can be used by the subsequent scheduling system or manual review module.

[0053] In an embodiment of the present invention, to achieve intelligent linkage processing and real-time response to trajectory anomaly detection results, the system is designed based on anomaly label E and vehicle scheduling rules. The scheduling suggestion generation module, combined with the vehicle terminal communication system, enables real-time prompts and task adjustments for the driver. This process is automatically triggered after the anomaly detection module completes the identification and classification of abnormal segments.

[0054] The system first obtains the abnormal segment recognition results of the current vehicle from the trajectory analysis module, in the following format: ;in: Indicates the start and end points of the abnormal trajectory segment; ∈{drift anomaly, path deviation, behavioral mutation, data distortion}; Indicates the confidence score for anomalies; This represents the average confidence factor for abnormal segments.

[0055] The system will label the exception With predefined vehicle dispatch rule sets Perform mapping matching, where each anomaly label A corresponding set of pre-set scheduling response strategies Example rules are as follows: like =Drift abnormal and > 0.7 → Perform location resampling and suggest deceleration; like =Path deviation and currently outside the allowed offset area → Issue route correction command; like =Sudden behavioral changes and ≥ 2 consecutive triggers → Mark as abnormal driving behavior and trigger manual review process; like = Data distortion → Reset GNSS module and request redundant positioning signal activation.

[0056] Scheduling rule base It can be customized according to the transportation company's operating policies, road control rules and driving behavior restriction strategies, and supports hot updates.

[0057] The system automatically constructs a scheduling suggestion data structure D based on the matching results: D = {Vehicle ID, Timestamp, Anomaly Tag, Suggested Action, Priority, Additional Description}; where: Suggested actions are suggested action commands, such as "Please check positioning device", "Please return to the main path", "About to leave the specified area"; Priority is based on action priority (e.g., urgent, moderate, alert); Additional notes may include scheduling instructions or suggestions for manual intervention.

[0058] It is recommended that actions be encapsulated in a JSON structure or a standard CAN bus communication command format to facilitate adaptation to vehicle equipment from different manufacturers.

[0059] The generated scheduling suggestion D is pushed to the vehicle's onboard terminal system in real time via the communication module. The push method supports the following two types: Cellular communication (4G / 5G): Suitable for most connected vehicles, sending dispatch instructions to the in-vehicle smart screen or T-box in real time via TCP / HTTP channel; Edge caching + intermittent synchronization: In areas with unstable signals, the system can first store dispatch suggestions in the cache, and then send them out in batches when the vehicles regain communication.

[0060] At the IVT end, the system can prompt the driver or automatically trigger the onboard system to respond in the following ways: A pop-up message appears, such as "GPS is drifting abnormally. Please slow down and check your device." Navigation route adjustment: Automatically switch the navigation path back to the recommended route; Task status change: If the abnormality is severe, the current task will be paused and resumed after confirmation from the background. Fault upload or manual review trigger: The relevant anomaly report will be uploaded to the transportation management platform simultaneously.

[0061] Example 2, please refer to Figure 2 As shown in this embodiment, a freight logistics trajectory anomaly handling system includes: The data acquisition module obtains the raw dataset T of the target freight vehicle's trajectory within the current transportation cycle; The behavior modeling module constructs a behavior evolution model of the target vehicle based on historical trajectory data, which is used to characterize the normal operating mode and variation boundary of the target vehicle under different road types and time periods; The trajectory fusion and preprocessing module performs multi-source fusion correction on the original trajectory dataset T to obtain the fused trajectory point sequence. ; The trajectory reliability assessment module evaluates the trajectory point sequence. A confidence factor is calculated for each trajectory point, and the confidence factor comprehensively considers GPS stability, historical trajectory consistency and road matching. Anomaly scoring module, based on trajectory point sequence An anomaly determination graph G is constructed with a behavioral evolution model. In graph G, nodes represent trajectory segments, and edge weights represent the degree of behavioral deviation between points. A recursive scoring mechanism is used to traverse graph G to obtain the anomaly distribution function R(x). The anomaly detection module determines that there is an anomaly in the corresponding trajectory segment when R(x) exceeds the dynamic adaptive threshold R, and outputs an anomaly label. The scheduling response module generates scheduling suggestions based on anomaly tags and vehicle scheduling rules and pushes them to the vehicle-mounted system in real time.

[0062] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A method for handling abnormal freight logistics trajectories, characterized in that: include: S100. Obtain the original dataset T of the target freight vehicle's trajectory within the current transportation cycle; S200. Construct a behavioral evolution model of the target vehicle based on historical trajectory data to characterize the normal operating mode and variation boundary of the target vehicle under different road types and time periods. S300. Perform multi-source fusion correction on the original trajectory dataset T to obtain the fused trajectory point sequence T′; S400. Calculate the confidence factor for each trajectory point in the trajectory point sequence T′. The confidence factor comprehensively considers GPS stability, historical trajectory consistency and road matching. S500. Based on the trajectory point sequence T′ and the behavior evolution model, an anomaly judgment graph G is constructed. In the graph G, nodes represent trajectory segments, and edge weights represent the degree of behavioral deviation between points. A recursive scoring mechanism is used to traverse the graph G to obtain the anomaly distribution function R(x). S600. When R(x) exceeds the dynamic adaptive threshold R, the corresponding trajectory segment is determined to be abnormal, and an abnormal label is output. S700 generates scheduling suggestions based on anomaly tags and vehicle scheduling rules and pushes them to the vehicle terminal system in real time.

2. The method for handling abnormal freight logistics trajectories according to claim 1, characterized in that: The S400 includes: S401. Based on the original GPS signal quality index of each point in the trajectory point sequence T′, construct a GPS stability factor Q. The GPS stability factor Q integrates the number of satellites, the positioning accuracy factor and the time interval mutation rate to reflect the positioning stability of each point. S402. For each trajectory point, extract the density kernel function value corresponding to its location in the historical trajectory set, and construct the trajectory matching matrix H to measure the degree of matching between the current trajectory point and the historical behavior trajectory. S403. Combine digital map data to perform multi-scale road projection on the trajectory points, use a spatial consistency convolutional network to calculate the matching deviation value d between the point and the road centerline, and construct the road matching factor M. S404. Based on the collaborative scoring mechanism of Q, H and M, they are integrated into a credibility factor for trajectory points.

3. The method for handling abnormal freight logistics trajectories according to claim 2, characterized in that: The calculations of Q, H, and M include: The calculation of the GPS stability factor Q includes: extracting the corresponding raw GNSS data for each trajectory point and calculating the number of visible satellites. A stability scoring function is constructed using the horizontal precision factor HDOP and the time interval Δt between adjacent points: ,in This represents the average of the trajectory sampling times. This is an empirical adjustment coefficient; The calculation of the historical trajectory similarity factor H includes: using the current trajectory point Centered on a point, define a spatiotemporal neighborhood with radius r, and retrieve historical trajectory sets and... For trajectory segments whose trajectory feature vector similarity exceeds a threshold θ, a point density function D( ) is constructed using the kernel density estimation method. ),get This is used to reflect whether the current point is located in a densely distributed area of ​​historical normal trajectory distribution; The calculation of the road matching factor M includes: [the calculation of the trajectory points...] By mapping the point to the centerline of the road segment it is located using multi-scale projection, a spatially consistent convolutional network is used to extract the geometric difference features between the road boundary and the trajectory, and the centerline deviation distance of the point is output. And standardized to obtain the matching degree index: Where σ is the Sigmoid function, This is the maximum reasonable offset distance set.

4. The method for handling abnormal freight logistics trajectories according to claim 3, characterized in that: Q, H, and M are converted into a comprehensive feature vector. This comprehensive feature vector is used as the input to the machine learning model. The machine learning model uses the prediction of the confidence factor label of each trajectory point by each set of comprehensive feature vectors as the prediction objective. The training objective is to minimize the sum of the prediction errors of the confidence factor labels of all trajectory points. The machine learning model is trained until the sum of prediction errors converges, at which point the model training stops. The confidence factor of each trajectory point is determined based on the model output. The machine learning model is a multinomial regression model.

5. The method for handling abnormal freight logistics trajectories according to claim 1, characterized in that: The S500 includes: S501. Construct an anomaly detection graph G = (V, E), where each node... Represents trajectory points Its attributes include a credibility factor. behavioral deviation indicators ;side This represents the temporal coherence between trajectory point pairs, assigning association weights based on the rate of change of velocity and abrupt changes in direction. ; S502, Initialization Abnormal activation value of the starting trajectory point Define a recursive function A depth-first traversal strategy is used to expand the nodes of graph G, and the Φ function is applied sequentially to update the score of each node. S503, calculated for all nodes The values ​​are subjected to sliding normalization and abrupt change detection to form the final anomaly distribution function R(x), which is used to characterize the potential anomaly density changes in the entire trajectory.

6. The method for handling abnormal freight logistics trajectories according to claim 5, characterized in that: The output of anomaly labels based on the anomaly distribution function R(x) includes: S601. Perform sliding window segmentation on R(x) and calculate the local mean within each window. with standard deviation A dynamic adaptive threshold function is constructed based on the global distribution of R(x) across the entire trajectory. , where k is the adjustment factor; S602, compare R(x) with the corresponding interval. For comparison, if there are M consecutive trajectory points x ∈ satisfy If so, then extract that segment as a candidate abnormal segment; S603, Behavioral Deviation Indicators Based on Candidate Abnormal Fragments Credibility factor Road matching degree Perform feature fusion, input the data into a multi-class decision model, determine the anomaly type of the segment, and output anomaly label E ∈{drift anomaly, path deviation, behavioral mutation, data distortion}.

7. The method for handling abnormal freight logistics trajectories according to claim 1, characterized in that: S200 includes: S201. Perform two-layer clustering on the trajectory data of the target vehicle in the last N transportation cycles according to road type and time period, and extract the trajectory feature set F, including average speed sequence, acceleration fluctuation value and trajectory offset rate. S202. Construct a multi-channel recurrent autoencoder network to encode and compress the feature set F, and introduce a residual mapping module between channels to retain fine-grained temporal behavior variation information; S203. Construct a behavior variation map in two-dimensional time-road coordinates based on multi-channel feature representation, and use a boundary learning function to fit the boundary of local mutation regions in the map to obtain the operational boundary set for each type of road and cycle. S204. The set of operating boundaries is used as the core constraint of the target vehicle behavior evolution model and stored in the model memory unit.

8. The method for handling abnormal freight logistics trajectories according to claim 1, characterized in that: In S100, the original trajectory dataset T includes: GPS positioning points, speed, acceleration, heading angle, road type, and timestamp.

9. A method for handling abnormal freight logistics trajectories according to claim 1, characterized in that: In S300, the correction includes joint optimization based on inertial navigation, geographic reference points, and signal strength data.

10. A freight logistics trajectory anomaly handling system, used to implement the freight logistics trajectory anomaly handling method according to any one of claims 1-9, characterized in that: include: The data acquisition module obtains the raw dataset T of the target freight vehicle's trajectory within the current transportation cycle; The behavior modeling module constructs a behavior evolution model of the target vehicle based on historical trajectory data, which is used to characterize the normal operating mode and variation boundary of the target vehicle under different road types and time periods; The trajectory fusion and preprocessing module performs multi-source fusion correction on the original trajectory dataset T to obtain the fused trajectory point sequence T′. The trajectory reliability assessment module evaluates the trajectory point sequence. A confidence factor is calculated for each trajectory point, and the confidence factor comprehensively considers GPS stability, historical trajectory consistency and road matching. Anomaly scoring module, based on trajectory point sequence An anomaly determination graph G is constructed with a behavioral evolution model. In graph G, nodes represent trajectory segments, and edge weights represent the degree of behavioral deviation between points. A recursive scoring mechanism is used to traverse graph G to obtain the anomaly distribution function R(x). The anomaly detection module determines that there is an anomaly in the corresponding trajectory segment when R(x) exceeds the dynamic adaptive threshold R, and outputs an anomaly label. The scheduling response module generates scheduling suggestions based on anomaly tags and vehicle scheduling rules and pushes them to the vehicle-mounted system in real time.

Images