A method for matching dangerous goods electronic waybill with vehicle trajectory

Abstract

The application discloses a dangerous goods electronic waybill and vehicle track matching method, and relates to the technical field of traffic data processing. The method comprises the following steps: collecting track information, waybill information and vehicle running state information of a dangerous goods transport vehicle, performing time alignment and multi-source fusion, and obtaining a basic data set; calculating vehicle speed and direction change rate based on the basic data set, dividing track sections, determining continuity indexes, and obtaining track section basic scores; calculating track coincidence degrees according to the basic scores, obtaining modified confidence scores, performing confidence correction on the track sections, and obtaining section-level modified scores; recursively updating average trust degrees based on the section-level modified scores, obtaining state labels, judging high-risk track sections, calculating severity indexes, and performing hierarchical disposal. Through the confidence correction mechanism based on track continuity, direction stability and template route similarity, the application realizes accurate identification of vehicle track deviation, stop driving and abnormal behaviors.

Description

Technical Field

[0001] This invention relates to the field of traffic data processing technology, and in particular to a method for matching electronic waybills for dangerous goods with vehicle trajectories. Background Technology

[0002] With the accelerating digitalization and intelligentization of hazardous goods transportation management, electronic waybill systems are gradually becoming an important infrastructure for the safety supervision of hazardous goods transportation. These systems typically integrate multiple functions such as waybill generation, transportation process tracking, reporting and filing, and risk warning. By interacting with multiple data sources such as vehicle positioning terminals (GPS / BeiDou), fuel consumption monitoring, and communication modules, they achieve visualized management of the entire hazardous goods transportation process.

[0003] However, existing methods still have two limitations: First, existing dangerous goods transportation trajectory and waybill matching technologies are mainly based on simple time overlap and spatial proximity judgments, which are difficult to distinguish complex behaviors such as temporary stoppage, detour, and cargo transfer; Second, multi-source data suffer from communication delays, inconsistent sampling frequencies, and coordinate offsets during the collection process, which leads to ambiguity in the correspondence between trajectory segments and waybill status, thereby affecting the reliability of anomaly identification. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a method for matching electronic waybills for dangerous goods with vehicle trajectories to solve the problems of weak trajectory deviation recognition capability and insufficient reliability of anomaly recognition in existing technologies.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] This invention provides a method for matching electronic waybills for hazardous goods with vehicle trajectories. The method includes: collecting trajectory information, waybill information, and vehicle operating status information of hazardous goods transport vehicles; performing time alignment and multi-source fusion to obtain a basic dataset; based on the basic dataset, calculating the vehicle speed and direction change rate to obtain trajectory segments; determining the continuity index of the trajectory segments to obtain a basic score for each trajectory segment; calculating the trajectory overlap index based on the basic score to obtain a corrected confidence score; performing confidence correction on each trajectory segment using the corrected confidence score to obtain a segment-level corrected score; recursively updating the segment-level corrected score using exponential smoothing to obtain an average comprehensive confidence level; obtaining a backtracking support value using the average comprehensive confidence level; updating the status to obtain a status label; identifying high-risk trajectory segments using the status label to obtain a severity index; and classifying and handling the situation based on the severity index to obtain new auxiliary parameters.

[0008] As a preferred embodiment of the method for matching electronic waybills and vehicle trajectories for hazardous goods as described in this invention, the steps of collecting trajectory information, waybill information, and vehicle operation status information of hazardous goods transport vehicles, performing time alignment and multi-source fusion to obtain a basic dataset are as follows: collecting trajectory information, waybill information, and vehicle operation status information of hazardous goods transport vehicles to obtain a preliminary multi-source data sequence; based on the preliminary multi-source data sequence, performing a unified time reference alignment operation, generating a fused time sequence according to a set sampling interval, and completing missing time point data based on the principle of time proximity; and based on the fused time sequence, identifying and smoothing abnormal records according to data priority rules and rationality verification logic to obtain a basic dataset.

[0009] As a preferred embodiment of the method for matching dangerous goods electronic waybills with vehicle trajectories according to the present invention, the steps of calculating the vehicle speed and direction change rate and obtaining trajectory segments based on the basic dataset are as follows: constructing a single vehicle time series through the basic dataset, using event time sorting and configuring a fixed water level strategy to remove outliers and obtain a cleaned single vehicle trajectory sequence; calculating the speed and direction change rate based on the cleaned single vehicle trajectory sequence, identifying the stopping state, and performing on-site segmentation according to the three rules of gap triggering, parking triggering, and quality triggering to obtain trajectory segments.

[0010] As a preferred embodiment of the method for matching dangerous goods electronic waybills with vehicle trajectories according to the present invention, the steps of determining the continuity index of the trajectory segment and obtaining the basic score of the trajectory segment are as follows: calculating the maximum time interval, directional stability, and velocity robust dispersion for each trajectory segment to obtain the trajectory segment continuity index; extracting the latitude and longitude sequence of the trajectory segment and the waybill planned route through the trajectory segment continuity index; calculating the average lateral deviation and directional consistency index through coordinate projection and directional angle analysis; and calculating the trajectory segment basic score through the average lateral deviation and directional consistency index.

[0011] As a preferred embodiment of the method for matching dangerous goods electronic waybills with vehicle trajectories according to the present invention, the steps of calculating the trajectory overlap index and obtaining the corrected confidence score based on the trajectory segment base score are as follows: For each trajectory segment, a corresponding template route is generated based on the high-confidence trajectory set in the historical transportation task database; the average shape distance between the current trajectory segment and the template route is calculated using a dynamic time warping algorithm to obtain the trajectory overlap index; based on the trajectory overlap index, acceleration sequences and direction change rate sequences are extracted from the original data of the trajectory segment, and their standard deviations are calculated to obtain stationarity and stability indices; smoothing correction is performed using the trajectory segment base score, trajectory overlap index, and stationarity and stability indices to calculate the corrected confidence score.

[0012] As a preferred embodiment of the method for matching electronic waybills for dangerous goods with vehicle trajectories according to the present invention, the step of performing confidence correction on each trajectory segment by correcting the confidence score to obtain a segment-level correction score specifically involves: calculating the waybill coverage within the trajectory segment; adjusting the corrected confidence score based on the waybill coverage to obtain a confidence enhancement score; obtaining the reporting type corresponding to the trajectory segment; mapping the reporting type to a reporting credibility weight; calculating the warehouse resale probability of the trajectory segment based on the warehouse proximity and loop strength of the trajectory segment; and calculating the segment-level correction score of the trajectory segment based on the confidence enhancement score, the reporting credibility weight, and the warehouse resale probability.

[0013] As a preferred embodiment of the method for matching dangerous goods electronic waybills with vehicle trajectories according to the present invention, the step of recursively updating the average comprehensive confidence level based on segment-level correction scores using an exponential smoothing method involves the following steps: expanding the segment-level correction scores on a unified time axis to obtain the scores for each second, and performing linear interpolation to complete the scores; setting an initial comprehensive confidence level, recursively updating the scores for each second using an exponential smoothing method to obtain candidate comprehensive confidence levels; and averaging the candidate comprehensive confidence levels for all valid moments within the trajectory segment to obtain the average comprehensive confidence level of the trajectory segment.

[0014] As a preferred embodiment of the method for matching dangerous goods electronic waybills with vehicle trajectories according to the present invention, the steps of obtaining backtracking support values ​​through average comprehensive trust level, updating the state, and obtaining state labels are as follows: Based on the consistency relationship between speed, throttle, fuel consumption, and acceleration, a physical consistency score is calculated; a confidence score is calculated through average comprehensive trust level and segment-level correction score; a continuity score is obtained by statistically analyzing sampling gaps and jump point ratios; the physical consistency score, confidence score, and continuity score are combined according to evidence weights to obtain backtracking support values; and the state is restored to matching, maintained in its original state, or upgraded to an anomaly based on the backtracking support values, and the state labels of the corresponding segments are updated accordingly.

[0015] As a preferred embodiment of the method for matching electronic waybills for dangerous goods with vehicle trajectories according to the present invention, the step of determining high-risk trajectory segments and obtaining severity indicators through status tags specifically involves: determining whether there are situations where there is a waybill but no trajectory or a trajectory but no waybill based on trajectory segments marked as abnormal in the status tags; for situations where there is a waybill but no trajectory or a trajectory but no waybill, it is determined to be a high-risk trajectory segment; for high-risk trajectory segments, based on the average comprehensive confidence level, warehouse reversal probability, and retrospective support value, the confidence gap, warehouse counter-evidence strength, and retrospective support insufficiency are calculated; the confidence gap, warehouse counter-evidence strength, and retrospective support insufficiency are arithmetically averaged to obtain the severity indicator.

[0016] As a preferred embodiment of the method for matching dangerous goods electronic waybills with vehicle trajectories described in this invention, the step of classifying and handling cases based on severity indicators to obtain new auxiliary parameters involves the following steps: classifying and handling severity indicators; generating a formal alarm when the severity indicator is greater than or equal to a high-risk threshold; entering a manual review queue when the severity indicator is between a low-risk threshold and a high-risk threshold; and recording without issuing an alarm when the severity indicator is below a low-risk threshold; extracting all alarm events and performing statistical analysis of trust levels and evidence distribution to obtain new auxiliary parameters.

[0017] The beneficial effects of this invention are as follows: By establishing a time-base alignment and multi-source fusion mechanism for trajectory information, waybill information, and operational status information, consistent collection and fusion of data throughout the entire hazardous goods transportation process are achieved; by introducing a confidence assessment and correction mechanism based on trajectory continuity, directional stability, and template route similarity, accurate identification of vehicle trajectory deviation, stopping, and abnormal behavior is achieved; and by constructing a severity index grading and audit closed-loop mechanism based on fusion confidence gap, warehouse counter-evidence strength, and backtracking support value, quantitative grading and traceability solidification of alarm results are achieved. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Fig. 1 A flowchart illustrating the method for matching electronic waybills for dangerous goods with vehicle trajectories.

[0020] Fig. 2 This is a flowchart for the fusion of multi-source data from trajectory and waybill.

[0021] Fig. 3 This is a flowchart of the baseline score and the corrected confidence score for the trajectory segment.

[0022] Fig. 4 This is a flowchart for trust level determination and anomaly alarm. Detailed Implementation

[0023] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0024] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0025] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0026] Reference Figs. 1-4 As one embodiment of the present invention, this embodiment provides a method for matching electronic waybills for dangerous goods with vehicle trajectories, comprising the following steps:

[0027] S1. Collect trajectory information, waybill information, and vehicle operation status information of hazardous goods transport vehicles, perform time alignment and multi-source fusion, and obtain basic datasets.

[0028] Vehicle location data, including time, latitude and longitude, speed, and acceleration, is recorded at a fixed sampling period (e.g., 1 second). The collected location points are stored as a trajectory sequence in chronological order to obtain trajectory information. Waybill information, including waybill number, task start and end time, origin, destination, and transportation status, is obtained. Vehicle operating status information, including engine fuel consumption, ignition switch status, and communication module online status, is obtained in real time via the vehicle's CAN bus.

[0029] A unified time base is used to align trajectory information, waybill information, and vehicle operation status information. A fused time series is generated at a unified time interval (e.g., 1 second). If trajectory information, waybill information, and vehicle operation status information all exist at the same time point, they are directly integrated into a complete record. If some data is missing, the nearest valid data is selected to fill in the gaps based on the principle of time proximity. If the sampling intervals of the multi-source data are different, trajectory information is retained first, followed by vehicle status information, and finally waybill information, to generate a fused record for the current moment. Outliers (e.g., zero speed but high fuel consumption, sudden changes in position, etc.) are subjected to rationality checks and smoothing. The fused data record includes a unified timestamp, location, speed, acceleration, fuel consumption, ignition status, communication status, waybill number, origin and destination, etc., ultimately forming a structured basic dataset.

[0030] S2. Based on the basic dataset, calculate the rate of change of vehicle speed and direction, obtain trajectory segments, determine the continuity index of trajectory segments, and obtain the basic score of trajectory segments.

[0031] Read the unified timestamp, plane coordinates, speed, acceleration, equipment quality indication and waybill related information from the basic dataset, construct a single vehicle time series according to vehicle ID and time series, sort by event time and configure water level (e.g. 5s). 5s can cover cellular backhaul jitter and avoid large buffer. When the time of a certain point is less than the currently observed maximum time −5s, it is regarded as late and discarded.

[0032] Outlier removal is performed on the time series of individual vehicles. If the speed is greater than the abnormally high speed value (e.g., 200 km / h) or less than the abnormally low speed value (e.g., 0 km / h), it is removed. If the straight-line distance between two adjacent points is greater than the abnormal distance (e.g., 5 km) and the time difference is less than the fixed time interval (e.g., 10 s), it is considered a jump point and removed. If the absolute value of acceleration is greater than the abnormal acceleration (e.g., 6 m / s²), it is marked as an anomaly and removed.

[0033] It should be noted that speeds below abnormally low values ​​usually originate from abnormal field symbols or data filling, while speeds above abnormally high values ​​generally exceed the reasonable range of road traffic and are likely to correspond to instantaneous ultra-high-speed noise points, so they are removed. When two adjacent points experience a large-span displacement (greater than the abnormal distance) within a very short time (less than a fixed time interval), it is usually difficult to meet the vehicle's accessibility requirements, so it is identified as a jump point and the latter point is removed to maintain the continuity of the preceding trajectory. When the absolute value of acceleration calculated from adjacent points exceeds the abnormal acceleration, it is mostly due to noise causing a sudden change in speed. If it is retained, it will distort the speed or direction, so it is marked as abnormal and removed. If the value for removing abnormal values ​​is more stringent, it will increase the false removal of real boundary data and cause trajectory breaks. If the value for removing abnormal values ​​is more lenient, it will retain more abnormal points and reduce the accuracy of subsequent stop recognition and trajectory matching.

[0034] The speed and direction change rate are calculated directly from the speed in the single vehicle time series. The direction angle is calculated from the vectors of two adjacent points (planar coordinates), and the direction angle is taken as north-east. The direction change rate is obtained by the ratio of the difference in direction angle between two adjacent moments to the corresponding time difference. When the speed is less than the lower value of the stopping speed (e.g., 1.8 km / h) and the duration is greater than or equal to the stopping duration (e.g., 10 s), it is recorded as stopping.

[0035] It should be noted that, in order to suppress the amplification effect of low-speed jitter and positioning drift on direction calculation, when the speed is less than the low value of the stopping speed and the duration of this state is not less than the stopping duration, the time period is recorded as stopping. If the value is more stringent (e.g., increasing the speed threshold or shortening the duration), it is easy to misjudge short-term slow traffic and queuing as stopping; if the value is more lenient (e.g., decreasing the speed threshold or extending the duration), it may miss the detection of real stopping.

[0036] The trajectory is segmented using a deterministic three-trigger rule to segment the single-vehicle time series in situ. Specifically, the deterministic three-trigger rule is as follows: gap trigger, where the time difference between two adjacent points is greater than the gap time difference (e.g., 120s); stop trigger, where the cumulative continuous stop time is greater than or equal to the cumulative stop time (e.g., 30min); and quality trigger, where a point is judged as an outlier or missing a key field. When any of the trigger conditions occur, the previous valid sampling point is used as the end point of the current segment, and the set of continuous sampling points from the start index to the end point is output as a trajectory segment. The next valid sampling point after the trigger point is used as the starting index of the new segment to continue traversing and obtain the segment trajectory.

[0037] It should be noted that the time difference between intervals is used to identify time gaps caused by sampling interruptions or transmission delays. A small value will lead to frequent segmentation and fragmentation of the segment trajectory due to normal sampling jitter, while a large value will merge obvious gaps into the same segment, causing distortion of features within the segment. The cumulative stop time is used to distinguish between short stops and longer stops. A small value may incorrectly segment short stops such as congestion and red lights, while a large value will mix longer stops with driving in the same segment, causing segment boundary shifts. For each segment trajectory, the following parameters are obtained: mileage (cumulative spherical and planar distances between adjacent points), average segment speed (ratio of mileage to segment duration), stop ratio (proportion of stop time within the segment to the total segment duration), and directional stability (percentage of time with directional change rate falling within [-2, 2] degrees / second). The segment continuity index is then obtained, expressed as:

[0038] ;

[0039] in, Indicates the first The continuity index of the segment trajectory, Indicates the first The maximum time difference between adjacent sampling points within a segment trajectory. Indicates the first Directional stability within the segment trajectory Indicates the first Robust velocity dispersion within the trajectory segment. Represents the time smoothing constant. Represents the directional stability smoothing constant. This represents the velocity smoothing constant.

[0040] It should be noted that, It calculates the median absolute deviation of the velocity sequence after removing outliers. The velocity sequence is a subset of the velocity of a single vehicle within a certain trajectory segment time range. The empirical threshold was determined by statistically analyzing a large amount of historical data on the trajectories of normally operating vehicles and calculating the 95th percentile of the time interval between adjacent sampling points. It was determined by statistically analyzing a large amount of data on the rate of change of direction of a large number of normally driving vehicles, and taking the quantile value that covers about 95% of the normally driving samples in the distribution of the rate of change of direction. The threshold is determined by analyzing speed fluctuation data from a large number of normally driving vehicles and taking the quantile value corresponding to approximately 95% of the normal samples in the speed standard deviation distribution. Using approximately 95% quantile value as the basis for determining the empirical threshold is problematic. If the quantile value is lower, it will be too strict and may easily misjudge boundary fluctuations caused by road environment, sampling jitter, or individual driving behavior during normal operation as abnormal, thereby increasing false rejections and causing trajectory breaks. If the quantile value is higher, more abnormal fluctuations introduced by positioning drift, timestamp mismatch, or communication backfill will be retained, resulting in contamination of speed and direction change rates and deviation of segment trajectory boundaries. Therefore, selecting approximately 95% quantile value can achieve a better balance between false alarms and false negatives.

[0041] The latitude and longitude sequences and corresponding waybill planned route coordinate sequences within each trajectory segment are extracted from the basic dataset. The actual vehicle trajectory points are projected one by one onto the route centerline using coordinate projection, and the lateral distance from each sampling point to the route centerline is calculated. The average lateral deviation is obtained by averaging all distances. Simultaneously, the proportion of sampling points with a lateral deviation less than the lower limit (e.g., 150 meters) is calculated as the fitting ratio for that segment. The vehicle velocity direction angle is calculated based on the displacement direction of adjacent trajectory points, and the tangential direction angle of the waybill route centerline at the same location is obtained. The included angle is calculated. The proportion of sampling points with an included angle less than the fitting judgment angle (e.g., 30°) is calculated as the direction consistency ratio for that segment. The route fitting ratio and direction consistency ratio for each trajectory segment are calculated. If the fitting ratio is lower than the fitting threshold, the trajectory segment is determined to deviate from the task route and is not included in the calculation of the trajectory segment's base score. If the direction consistency ratio is lower than the lower limit (e.g., 0.5), the segment is marked as a direction abnormality segment.

[0042] It should be noted that setting the lateral deviation to be less than the low value of the lateral deviation judgment is to absorb reasonable offsets caused by positioning errors, road width, parallel roads, and centerline abstraction. If the value is too small, it is easy to misjudge actual driving along the line as not fitting, while if the value is too large, it will relax the criteria to include points that deviate from the route, thus reducing the deviation recognition capability. Setting the included angle to be less than the fitting judgment angle is to accommodate errors in curves, lane changes, and direction estimation. If the value is too small, it is easy to misjudge on curved roads, while if the value is too large, it will weaken the distinction between reverse or yaw. Setting the direction consistency ratio to be lower than the low value of the ratio is to ensure that most points are in the same direction. If the value is too high, it will increase the misjudgment rate, while if the value is too low, it will be easy to miss direction anomalies.

[0043] The basic score for a trajectory segment is calculated using the following expression:

[0044] ;

[0045] in, Indicates the first The basic score of the segment trajectory, Indicates the first The average lateral deviation within the segment trajectory, Represents the spatial scale constant. Indicates the first The average direction cosine within the segment trajectory, Indicates the average speed of the segment. This indicates the average reference speed for the line. This indicates the elimination of zero items.

[0046] It should be noted that, It is achieved by statistically analyzing the average lateral deviation distribution of a large number of normal trajectory samples and taking the 95th percentile value as the spatial scale constant. The process involves filtering vehicle trajectories from the historical database that follow the same route and have no abnormal records, and calculating the median of the average speed distribution of these segments as the reference average speed for the route. The fitting threshold is determined by selecting a set of confirmed historical trajectory samples without abnormalities, calculating the fitting percentage (i.e., the percentage of sampling points with a lateral deviation less than the low value of the lateral deviation), and setting the percentage value below the 5th percentile of the distribution characteristics as the fitting threshold based on the distribution characteristics of these samples. If a higher percentile is used instead of the 5th percentile, more trajectory segments that are within the lower bound of normal fluctuations will be misjudged as deviations and removed, leading to an increase in false removals, a decrease in available trajectory segments, and a reduction in subsequent matching recall. If a lower percentile is used, some abnormal segments that actually deviate but whose fitting percentage is still not extremely low may be allowed to pass.

[0047] S3. Based on the baseline score of the trajectory segment, calculate the trajectory overlap index and obtain the corrected confidence score.

[0048] For each trajectory segment, a dynamic time warping algorithm is used to match the trajectory shape with the template route, calculating the average shape distance of the trajectory segment. Specifically, these trajectories are time-normalized and coordinate-aligned. A density clustering algorithm is used to extract the main path, forming a template route in the form of a latitude and longitude sequence. During matching, the dynamic time warping (DTW) algorithm is used to compare the sampling point sequence of the current trajectory segment with the coordinate sequence of the template route point by point, calculating the shortest matching path distance between the two curves in the time and spatial dimensions. The ratio of the cumulative shortest matching path distance to the number of matching points is used as the average shape distance. Based on the ratio of the average shape distance to the average matching distance of the template, the overlap between the trajectory and the template is calculated.

[0049] Acceleration and directional rate of change sequences are extracted from the raw data of the trajectory segment. The standard deviation of the acceleration sequence is calculated to characterize the vehicle's driving stability, and the standard deviation of the directional rate of change sequence is calculated to reflect directional stability. When the standard deviation of acceleration or directional rate of change is large, it indicates that the vehicle has significant acceleration / deceleration or directional fluctuations. In this case, a smoothing penalty is applied to the baseline score of the trajectory segment. Specifically, when the standard deviation of acceleration or directional rate of change exceeds the stability threshold, the ratio of the actual fluctuation value to the stability threshold is converted into a decay ratio, reducing the confidence score of the trajectory segment. The product of the baseline score of the trajectory segment and the degree of overlap is obtained, and smoothing correction is performed based on the acceleration and directional stability results to obtain the corrected confidence score.

[0050] Acceleration and directional change rate sequences are extracted from the original trajectory segment data. The standard deviations of the acceleration and directional change rate sequences are calculated to represent the degree of acceleration / deceleration and steering fluctuations of the vehicle within the trajectory segment. If the standard deviation of acceleration or directional change rate exceeds the corresponding stationarity threshold, the current trajectory segment is considered to have significant fluctuations. The ratio of the actual standard deviation to the stationarity threshold is used as the attenuation ratio to smooth and correct the base score of the current trajectory segment. Specifically, the product of the trajectory segment base score and the overlap between the trajectory and the template is used as the segment-level original confidence score. The attenuation ratio is then applied to the segment-level original confidence score to obtain the corrected confidence score of the current trajectory segment.

[0051] It should be noted that the template route is a standardized path template pre-generated in the historical transportation task database, derived from a set of high-confidence trajectories of multiple hazardous goods transport vehicles within the same transportation route, the same origin and destination warehouses, and similar time periods. The stability threshold is determined by selecting high-confidence trajectory segments from the historical database that have not generated any abnormal alarms, extracting their acceleration and direction change rate sequences, calculating the standard deviation for each segment, and obtaining a sample set of vehicle fluctuation characteristics under normal operating conditions. Subsequently, these standard deviation samples are plotted as probability distribution curves, and quantile statistical analysis is performed on them. The quantile value of approximately 95% of the cumulative probability is taken as the acceleration stability threshold and the direction stability threshold, respectively. If a higher quantile is used instead of approximately 95%, more abnormal segments (severe fluctuations caused by positioning jitter, timestamp mismatch, or abnormal driving) may be misjudged as stable, leading to an increase in missed detections. If a lower quantile is used, some normal but slightly volatile segments (such as those with dense curves, slopes, or congested following conditions) may be misjudged as unstable, resulting in an increase in misjudgments and a decrease in usable segments.

[0052] When the corrected confidence score is below the low confidence threshold, most of the trajectory corresponds to a serious deviation from the actual trajectory. This trajectory segment is marked as a low confidence segment and reported as an anomaly. When the confidence score is between the low confidence threshold and the high confidence threshold, there is a certain error but it has not exceeded the judgment range. This segment is marked as a segment that needs to be reviewed. When the confidence score is above the high confidence threshold, most of the trajectory is highly consistent with the task route. The trajectory is judged to be a complete match with the waybill task.

[0053] It should be noted that the low-confidence threshold and high-confidence threshold are determined by the adjusted confidence score distribution of two categories of trajectory segments: valid matches and severe deviations, in historical manually verified samples. The low-confidence threshold is taken as the corresponding percentile value of the adjusted confidence score distribution of severe deviation samples, such as the 95th percentile. If a higher percentile is used, more boundary segments will be suppressed to low confidence, leading to an increase in false alarms. If a lower percentile is used, some severe deviation segments will enter the review process or even be allowed, leading to an increase in false negatives. The high-confidence threshold is taken as the corresponding percentile of the adjusted confidence score distribution of valid matches, such as the 5th percentile. If a lower percentile is used, more segments that are not perfectly matched will be directly judged as matches, increasing the number of false positives. If a higher percentile is used, more matching segments that could have passed directly will fall into the review process, increasing the amount of manual review.

[0054] S4. By correcting the confidence score, perform confidence correction on each trajectory segment to obtain the segment-level correction score.

[0055] The system obtains time-series markers indicating whether a valid waybill is held from the waybill management platform. It then calculates the percentage of time within a trajectory segment that a valid waybill is held, using this percentage as the waybill coverage. If the waybill coverage is high, the revised confidence score for that trajectory segment is conservatively increased slightly. The system also obtains the reporting types within a segment from the reporting management platform, mapping these types to reporting confidence weights. Only reporting types indicating no business operation, suspension, or pending scrapping are included in the calculation set and assigned lower weights. Other reporting types or normal status are assigned higher weights, while no reporting type is assigned a neutral to low weight.

[0056] Taking the first location point within the trajectory segment with a speed greater than or equal to 5 km / h as the center, calculate the minimum distance from it to the nearest warehouse, and combine this with the loop closure strength within the segment to obtain the probability of warehouse restocking. The expression is:

[0057] ;

[0058] in, Indicates the first The probability of warehouse restocking within the segment trajectory. Indicates the minimum warehouse distance. Represents the distance scale constant. Indicates the first Intra-segment lap intensity of the segment trajectory This represents the closure sensitivity coefficient. This represents the warehouse radius threshold function, which takes the value 1 when the vehicle is no more than 5 km from the center of the warehouse, and 0 otherwise.

[0059] It should be noted that, This was determined by fitting the distance distribution of a large number of normal relocation samples, so that 95% of the relocation trajectories are concentrated within 5km; The loop strength within a segment is obtained by calculating the cumulative change in direction of vehicles within the trajectory segment and the path compression ratio (i.e., the ratio of total mileage to the straight-line distance between the start and end points), and then normalizing and combining the two. The cumulative change in direction is obtained by calculating the difference in direction angle between adjacent trajectory points, taking the absolute value of all differences and summing them up. It involves collecting vehicle trajectories that have been loaded, unloaded, or transferred near the warehouse, calculating the number of loops and the cumulative amount of directional changes, and statistically analyzing the actual transfer rate when there are 2 to 3 obvious loops in these trajectories. The loop sensitivity coefficient that best reflects the actual situation at this rate is then selected.

[0060] A hard rule is set for business operations: a trajectory segment will be downgraded only if it simultaneously meets the following conditions: no waybill is held, the reporting type is "no business, stopped or awaiting scrapping," it exceeds 5km from the warehouse, and the segment mileage is >10km. Otherwise, no downgrading is applied, and a segment-level correction score is calculated using the following expression:

[0061] ;

[0062] in, Indicates the first Segment-level correction of segment trajectory, Indicates the first The credibility weight of segment trajectory reporting. Indicates the first Corrected confidence score of segment trajectory, Indicates the first The weighting coefficient of the segment trajectory.

[0063] It should be noted that, The weight reduction is triggered when a trajectory segment simultaneously meets four conditions: no waybill, reported as having no business and being out of service or awaiting scrapping, distance to the warehouse exceeding 5km, and mileage exceeding 10km. Take 0.6), otherwise take 1, if a weight reduction is triggered. Setting the score below 0.6 can over-penalize edge cases (e.g., order delays, untimely updates to reporting information but still related tasks), leading to a lower overall matching confidence score and increased false positives. If a weighting reduction is triggered... If the value is higher than 0.6, the highly suspected non-task trajectory segments can still contribute significantly to the segment-level correction score, which can easily raise the final confidence score, resulting in the release of abnormal segments and an increase in missed reports.

[0064] S5. Based on the segment-level correction score, an exponential smoothing method is used for recursive updates to obtain the average comprehensive trust level.

[0065] The segment-level correction score of each trajectory segment is expanded on a unified time axis with a fixed sampling interval (e.g., 1 second) as the sampling period. Within the time interval covered by the segment, the segment-level correction score of that segment is copied as a second-by-second time score. If the time gap between two adjacent segments does not exceed the gap time difference, linear interpolation is performed to fill the gap. If it exceeds the gap time difference, a blank window is maintained and no interpolation is performed.

[0066] The overall trust score of the first valid moment is directly set as the score of that moment. For each new moment, exponential smoothing is used to proportionally merge the overall trust score of the previous moment with the moment score of the segment-level correction score of the current moment according to the smoothing coefficient to obtain the candidate trust score. If the interval with the previous moment is less than or equal to the gap time difference and the current score is significantly low, a small deduction is made according to the anomaly suppression coefficient to accelerate risk exposure (deduction is only made for short windows and not for long gaps), limiting the overall trust score to the range of 0 to 1. If it is a long gap (greater than the gap time difference), the recursion is paused and no update is made, and the previous valid value is maintained.

[0067] It should be noted that the smoothing coefficient and anomaly suppression coefficient were obtained through offline calibration of historical manually verified samples. Different parameter combinations were enumerated on the sample sets of the effective matching window and the anomaly window. The comprehensive trust score was generated according to the established recursive rules and the pass, review and reporting judgment were completed accordingly. The overall classification accuracy corresponding to each group of parameters was calculated, and the parameter group with the highest accuracy was selected as the final value.

[0068] When the overall trust score is continuously higher than the upper threshold and maintained for at least a fixed time interval, the current state is set to "matched"; when the overall trust score falls between the lower and upper thresholds, it is set to "suspicious"; when the overall trust score is continuously lower than the lower threshold and maintained for at least a fixed time interval, it is set to "abnormal"; hysteresis is used, and the state is switched only if the continuous dwell time in each interval is greater than or equal to the fixed time interval, otherwise the original state is maintained to prevent jitter; the overall trust score of all valid moments in the trajectory segment is averaged to obtain the average overall trust score of the trajectory segment.

[0069] It should be noted that the upper and lower thresholds of the overall trust level are determined by the overall trust level distribution of severely deviated segments and valid matching segments in historical manual verification samples. The percentile corresponding to the severely deviated distribution is taken as the lower threshold of the overall trust level, such as the 95th percentile. If the lower threshold of the overall trust level is taken as a higher percentile, the scope of low-confidence reporting will expand, and false alarms and unnecessary handling will increase. If a lower percentile is taken, some severely deviated segments will fall into the review or even pass zone, and the number of missed reports will increase. The percentile corresponding to the valid matching distribution is taken as the upper threshold of the overall trust level, such as the 5th percentile. If a higher percentile is taken, more matching segments that could have been directly passed will enter the review, increasing the number of reviews and decreasing the pass rate. If a lower percentile is taken, some boundary segments are more likely to be directly judged as high-confidence passes, increasing the risk of false release.

[0070] S6. Obtain the backtracking support value through the average comprehensive trust level, update the status, and obtain the status label.

[0071] When any time interval is marked as suspicious or abnormal, the interval is immediately backtracked and verified using a symmetrical window of the first 10 minutes + the interval itself + the last 10 minutes. Within the symmetrical window, three types of objectively quantifiable and directly collectable evidence are statistically analyzed, including physical consistency score, confidence score, and continuity score.

[0072] Two types of manually verified window samples were selected from the historical database: Category 1 consisted of valid matching windows, and Category 2 consisted of abnormal windows. For physical consistency scores, several consistency rules that could be directly determined by fields were pre-listed. The success rate of each rule was calculated in both types of samples, and the difference between the success rate in valid matching windows and the success rate in abnormal windows was used as the discriminant. Only rules with positive discriminant values ​​were retained for scoring. The proportion of each rule in the total score was determined by the ratio of the rule's discriminant value to the sum of the discriminant values ​​of all retained rules, ensuring that the sum of all proportions was 1. The physical consistency score within a window was calculated by taking the success rate of each rule within the window, and then weighted and summed according to the proportion, with a range limited to 0 to 1. For confidence scores, the comprehensive confidence score sequence expanded by second within the window was aligned with the segment-level correction score of the trajectory segment to which the current second belongs. The comprehensive confidence score was enumerated on the historical verification samples. The fusion ratio of the segment-level correction score is calculated separately. For each fusion ratio, the classification accuracy for classifying a sample as valid or abnormal is calculated. The fusion ratio with the highest classification accuracy is taken as the final fusion ratio. The internal confidence score is calculated by averaging the data second by second according to the fusion ratio, and is limited to 0 to 1. For the continuous score, the distribution of the maximum sampling gap and the distribution of the proportion of outliers are statistically analyzed in the samples of the valid matching window. The acceptable upper bound of the maximum sampling gap and the proportion of outliers are taken as the corresponding percentile of the distribution, such as the 95th percentile. The maximum sampling gap and the proportion of outliers are calculated first within the window. If the maximum sampling gap and the proportion of outliers do not exceed their respective upper bounds, the continuous score is 1. If either of them exceeds the upper bound, the ratio of the excess to the upper bound is used as the deduction ratio, and the arithmetic mean is taken. The continuous score is obtained by subtracting the arithmetic mean from 1, and the result is limited to 0 to 1.

[0073] It should be noted that several consistency rules that can be directly determined by the fields include, but are not limited to, the following: if the speed is greater than zero, the switch state should be open, otherwise it is considered inconsistent; if the switch state is closed, the speed should be zero, otherwise it is considered inconsistent; calculate the speed difference between two adjacent points, if the speed at the later moment is greater than the speed at the earlier moment, the acceleration should be positive, otherwise it is considered inconsistent; if the speed at the later moment is less than the speed at the earlier moment, the acceleration should be negative, otherwise it is considered inconsistent, etc.

[0074] It should be noted that the acceptable upper bounds for the maximum sampling gap and the proportion of outliers are both set at the 95th percentile of the corresponding distribution. This is to ensure that the vast majority of normal windows are within this upper bound and thus will not be penalized. If a higher percentile is used, the upper bound will be raised, and more windows with large gaps or a high proportion of outliers will still be considered acceptable. This would reduce the continuous penalty, but outliers would be more likely to be allowed to pass, increasing the risk of missed detections. If a lower percentile is used, the upper bound will be lowered, and fluctuations at the tail end of the normal window would be more likely to trigger penalties, resulting in an overall lower continuous score, more misjudgments, and an increase in unnecessary reviews.

[0075] Using the probabilistic synthesis approach of gap complementation, the three types of evidence—physical consistency score, confidence score, and continuity score—are combined into a single retrospective support value, expressed as:

[0076] ;

[0077] in, Indicates the backtracking support value. Represents the physical consistency score, Indicates the confidence score. Indicates continuous division, Indicates the weight of evidence.

[0078] It should be noted that, The method involves selecting a representative set of trajectory samples from long-term operational records, including confirmed normal segments, false alarm segments, and genuine abnormal segments. For each type of sample, the contribution of the three types of evidence scores to the final judgment accuracy is calculated. Then, the normalized result of this contribution is used as the initial weight value. By continuously comparing the consistency between the backtracking support value and the manual review result, the weight is fine-tuned so that the weighted result of the three can most accurately distinguish between revocable and maintainable abnormal states, thus obtaining the evidence weight.

[0079] when When the evidence is greater than or equal to the upper judgment threshold, the evidence within the judgment window is sufficient, the original suspicious / abnormal label is removed, and the status of the interval is changed to matching; when When the threshold is less than or equal to the lower threshold, the evidence is deemed insufficient, and the original label is maintained. If the original label was suspicious, it is upgraded to abnormal. When the value is between the lower and upper thresholds, the original state is maintained and the state label is obtained.

[0080] It should be noted that the upper and lower judgment thresholds are determined by statistically analyzing the distribution quantiles of the backtesting support values ​​of matching windows and abnormal windows in historical manual verification samples. For example, if the lower judgment threshold is taken as the 95th percentile of the comprehensive judgment distribution of abnormal windows in historical manual verification, and a higher percentile is used instead of the 95th percentile, more windows will fall into the category of insufficient evidence and remain / upgrade abnormal, increasing the number of false alarms and actions. If a lower percentile is used, some genuine abnormal windows will enter the middle zone or even have their markings revoked, increasing the number of missed reports. Similarly, if the upper judgment threshold is taken as the 5th percentile of the comprehensive judgment distribution of matching windows in historical manual verification, and a higher percentile is used, more matching windows that could have been revoked will remain in their original state, increasing the number of reviews. If a lower percentile is used, boundary windows are more likely to be directly judged as matches and have their markings revoked, increasing the risk of false releases.

[0081] S7. By using status tags, identify high-risk trajectory segments, obtain severity indicators, classify and handle them according to severity indicators, and obtain new auxiliary parameters.

[0082] For time periods marked as abnormal in the status label of the trajectory segment, an abnormality type is determined. Specifically, for each waybill in the business completion status, its return time and the first printing time are taken to form a backtracking interval. If there is any trajectory point in the interval with a speed greater than the abnormally low speed value, it is judged as having a trajectory and no warning is issued. If there is no trajectory or the speed of all trajectory points is 0, it is judged as having a waybill but no trajectory. For time periods with vehicle displacement, checks are performed hourly. If a valid waybill is held at that time, this type of warning is not calculated. If no waybill is held, the system checks for registration. If the registration type is "no business, vehicle stopped" or "pending scrapping," the system is included in the calculation. If the registration type is "accident stop," "repair and maintenance," "general freight," "other," or "no business, empty driving," the system is not calculated. For the trajectory segment included in the calculation, further judgment is made. The first point in the segment with a speed greater than or equal to a fixed starting speed (e.g., 5 km / h) is used as the center. If a warehouse exists within the abnormal distance, the system is not triggered. Conversely, if the abnormal distance is exceeded and the segment mileage is greater than a fixed driving distance (e.g., 10 km), the system is judged as having a trajectory but no waybill.

[0083] When instances of waybills without tracking information or tracking information without waybills are discovered, these are considered high-risk business anomalies requiring severity assessment. Specifically, for each marked anomaly interval, three directly quantifiable indicators are calculated: confidence gap, warehouse counter-evidence strength, and insufficient backtracking support. The confidence gap measures the degree of decline in the credibility of the tracking data within that interval. It compares the average overall confidence level with a preset high confidence threshold. When the average overall confidence level is lower than the high confidence threshold, the difference between the high confidence threshold and the average overall confidence level is taken as the confidence gap. The difference between 1 and the warehouse resale probability is taken as the warehouse counter-evidence strength. Insufficient backtracking support is determined by reading the backtracking support value. When the backtracking support value is lower than a preset backtracking confidence lower limit, the difference between 1 and the backtracking support value is taken as insufficient backtracking support. The severity index is calculated by taking the arithmetic mean of the confidence gap, warehouse counter-evidence strength, and insufficient backtracking support.

[0084] When the severity index is greater than or equal to the high-risk threshold, it is directly judged as high-risk and a formal alarm is generated; when the severity index is between the low-risk threshold and the high-risk threshold, it is marked as requiring manual review; when the severity index is less than or equal to the low-risk threshold, it is only recorded as low-risk and no alarm is triggered.

[0085] It should be noted that the high confidence threshold is determined by statistically analyzing the comprehensive confidence distribution of a large number of normal transportation trajectories, taking the mean or 95th percentile of its stable interval. The preset backtesting confidence lower limit is determined by statistically analyzing the backtesting support values ​​of a large number of manually verified historical anomaly samples, taking the boundary point that most effectively distinguishes between sufficient and insufficient evidence. The low-risk and high-risk thresholds are determined by the distribution of severity indicators for two categories—those requiring and not requiring alerts—in historical manually verified samples. The percentile corresponding to the distribution requiring no alerts is taken as the low-risk threshold, such as the 95th percentile. If a higher percentile is taken, more samples that originally did not require alerts will be included in the review or even alert interval, leading to false positives. The increased volume of reports and manual reviews means that if a lower percentile is used, some boundary samples that should have entered the review process will be directly classified as low-risk and will not trigger an alarm, increasing the risk of missed detections. If the corresponding percentile of the alarm distribution is used as the high-risk threshold, such as the 5th percentile, if a higher percentile is used, more real alarm-required samples will be pushed into the review area, reducing the triggering of formal alarms but increasing the risk of missed reports. If a lower percentile is used, the high-risk threshold will be lowered, and boundary samples will be more likely to be directly judged as high-risk and trigger formal alarms, increasing the cost of false alarms and handling.

[0086] All alarm events are extracted, and a manual review queue is established based on factors such as severity level, alarm type, time distribution, and whether it involves duplicate vehicles. Reviewers verify samples in the queue through a visual audit interface, retrieving corresponding trajectory playback, waybill status, warehouse geographical information, and reported snapshots. They manually determine whether the alarm is true and label it as true / abnormal, explainable / abnormal, or data error. The difference in trust distribution between manually confirmed normal and abnormal samples is statistically analyzed, and the upper and lower thresholds of the overall trust level are adjusted to the positions that most clearly distinguish the two types of samples, making the confidence interval and abnormal interval more consistent with reality. Contribution analysis is performed on three types of evidence: physical consistency, confidence score, and continuity. If a certain type of evidence has a higher proportion in manually confirmed correct samples, its weight is increased; conversely, it is decreased, ensuring that the fusion calculation is more in line with actual performance. For preset auxiliary parameters such as the lower confidence limit, multiple candidate values ​​are compared by offline playback of historical samples from the past month, and the combination with the smallest weighted error between false alarm rate and false negative rate is selected as the new auxiliary parameter.

[0087] This embodiment also provides a computer device applicable to the method of matching dangerous goods electronic waybills with vehicle trajectories, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the method of matching dangerous goods electronic waybills with vehicle trajectories as proposed in the above embodiment.

[0088] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0089] This embodiment also provides a storage medium storing a computer program, which, when executed by a processor, implements the method for matching electronic waybills for dangerous goods with vehicle trajectories as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0090] In summary, this invention achieves consistent data collection and fusion across the entire hazardous goods transportation process by establishing a time-base alignment and multi-source fusion mechanism for trajectory information, waybill information, and operational status information; it enables accurate identification of vehicle trajectory deviation, stoppage, and abnormal behavior by introducing a confidence assessment and correction mechanism based on trajectory continuity, directional stability, and template route similarity; and it achieves quantitative grading and traceability solidification of alarm results by constructing a severity classification and audit closed-loop mechanism that integrates confidence gaps, warehouse counter-evidence strength, and backtracking support values.

[0091] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for matching electronic waybills for dangerous goods with vehicle trajectories, characterized in that: include, Collect trajectory information, waybill information, and vehicle operation status information of hazardous goods transport vehicles, perform time alignment and multi-source fusion, and obtain basic datasets; Based on the basic dataset, calculate the rate of change of vehicle speed and direction, obtain trajectory segments, determine the continuity index of trajectory segments, and obtain the basic score of trajectory segments. Based on the baseline score of the trajectory segment, the trajectory overlap index is calculated to obtain the corrected confidence score. Then, confidence correction is performed on each trajectory segment using the corrected confidence score to obtain the segment-level corrected score. Based on the segment-level correction score, an exponential smoothing method is used for recursive updates to obtain the average comprehensive confidence level. The average comprehensive confidence level is then used to obtain the backtracking support value, perform state updates, and obtain the state label. By using status tags, high-risk trajectory segments are identified, severity indicators are obtained, and graded actions are taken based on these severity indicators, thereby obtaining new auxiliary parameters.

2. The method for matching electronic waybills for dangerous goods with vehicle trajectories as described in claim 1, characterized in that: The process involves collecting trajectory information, waybill information, and vehicle operation status information of hazardous goods transport vehicles, performing time alignment and multi-source fusion to obtain a basic dataset. The specific steps are as follows: Collect trajectory information, waybill information, and vehicle operation status information of hazardous goods transport vehicles to obtain preliminary multi-source data sequences; Based on the preliminary multi-source data sequence, a unified time reference alignment operation is performed, a fused time series is generated according to the set sampling interval, and data missing at time points is filled in based on the principle of time proximity. Based on the fused time series, abnormal records are identified and smoothed according to data priority rules and rationality verification logic to obtain the basic dataset.

3. The method for matching electronic waybills for dangerous goods with vehicle trajectories as described in claim 2, characterized in that: The process of calculating the vehicle's speed and direction change rate based on the basic dataset to obtain the trajectory segment involves the following steps: Using a basic dataset, a single vehicle time series is constructed. Event time sorting and a fixed water level strategy are configured to remove outliers and obtain the cleaned single vehicle trajectory sequence. Based on the cleaned single-vehicle trajectory sequence, the speed and direction change rate are calculated, the stopping state is identified, and local segmentation is performed according to three rules: gap trigger, parking trigger, and quality trigger to obtain trajectory segments.

4. The method for matching electronic waybills for dangerous goods with vehicle trajectories as described in claim 3, characterized in that: The steps for determining the continuity index of the trajectory segment and obtaining the basic score of the trajectory segment are as follows: For each trajectory segment, the maximum time interval, directional stability, and velocity robustness dispersion are calculated to obtain the trajectory segment continuity index. By using the trajectory segment continuity index, the latitude and longitude sequence of the trajectory segment and the waybill planning route are extracted. The average lateral deviation and directional consistency index are calculated by analyzing the coordinate projection and direction angle. The basic score of the trajectory segment is calculated by using the average lateral deviation and directional consistency index.

5. The method for matching electronic waybills for dangerous goods with vehicle trajectories as described in claim 4, characterized in that: The specific steps for calculating the trajectory overlap index and obtaining the corrected confidence score based on the trajectory segment baseline score are as follows: For each trajectory segment, a corresponding template route is generated based on the high-confidence trajectory set in the historical transportation task database. The average shape distance between the current trajectory segment and the template route is calculated using a dynamic time warping algorithm to obtain the trajectory overlap index. Based on the trajectory overlap index, acceleration sequence and direction change rate sequence are extracted from the original trajectory segment data, and their standard deviation is calculated to obtain stationarity and stability indexes. Smoothing corrections are performed using trajectory segment base scores, trajectory overlap indices, and stationarity and stability indices, and the corrected confidence score is calculated.

6. The method for matching electronic waybills for dangerous goods with vehicle trajectories as described in claim 5, characterized in that: The process involves performing confidence correction on each trajectory segment to obtain a segment-level correction score. The specific steps are as follows: Calculate the waybill coverage within the trajectory segment, adjust the corrected confidence score based on the waybill coverage, and obtain the confidence enhancement score; Obtain the reporting type corresponding to the trajectory segment, map the reporting type to the reporting credibility weight, and calculate the warehouse dumping probability of the trajectory segment through the warehouse proximity and loop strength. The segment-level correction score of the trajectory segment is calculated by using the confidence enhancement score, the reporting credibility weight, and the warehouse resale probability.

7. The method for matching electronic waybills for dangerous goods with vehicle trajectories as described in claim 6, characterized in that: The average comprehensive trust score is obtained by recursively updating the score based on segment-level correction using an exponential smoothing method. The specific steps are as follows: The segment-level corrections are expanded on a unified time axis to obtain the scores for each second, and then linear interpolation is performed to complete the scores. An initial overall confidence level is set, and the score is updated recursively using an exponential smoothing method to obtain candidate overall confidence levels. The average comprehensive confidence score of the trajectory segment is obtained by averaging the candidate comprehensive confidence scores of all valid moments within the trajectory segment.

8. The method for matching electronic waybills for dangerous goods with vehicle trajectories as described in claim 7, characterized in that: The steps involve obtaining a backtracking support value through average comprehensive trust level, updating the state, and obtaining a state label. Based on the consistency relationship between speed, throttle, fuel consumption and acceleration, the physical consistency score is calculated. The confidence score is calculated by the average comprehensive confidence score and the segment-level correction score. The continuity score is obtained by statistically analyzing the sampling gap and jump point ratio. The physical consistency score, confidence score, and continuity score are combined according to the evidence weights to obtain the backtesting support value; By backtracking the support values, operations can be performed to restore the state to a match, maintain the original state, or upgrade it to an anomaly, and the state label of the corresponding segment can be updated.

9. The method for matching electronic waybills for dangerous goods with vehicle trajectories as described in claim 8, characterized in that: The specific steps for identifying high-risk trajectory segments and obtaining severity indicators through status tags are as follows: By identifying track segments marked as abnormal in the status labels, determine whether there are situations where there is a waybill but no track or a track but no waybill. For situations where there is a waybill but no tracking information or a tracking information but no waybill, it is determined to be a high-risk tracking segment. For high-risk tracking segments, based on the average comprehensive confidence level, the probability of warehouse resale, and the retrospective support value, the confidence gap, the strength of warehouse counter-evidence, and the insufficiency of retrospective support are calculated. The severity index is obtained by arithmetically averaging the confidence gap, warehouse counter-evidence strength, and backtracking support insufficiency.

10. The method for matching electronic waybills for dangerous goods with vehicle trajectories as described in claim 9, characterized in that: The specific steps for grading treatment based on severity indicators and obtaining new auxiliary parameters are as follows: The severity index is classified and handled accordingly. When the severity index is greater than or equal to the high-risk threshold, a formal alarm is generated. When the severity index is between the low-risk threshold and the high-risk threshold, it enters the manual review queue. When the severity index is lower than the low-risk threshold, it is only recorded and no alarm is issued. Extract all alarm events, perform statistical analysis on trust level and evidence distribution, and obtain new auxiliary parameters.

Images